Untouched open Tcl bug tickets: Already fixed

Triage of 818 open and Pending Bug tickets of Tcl with no comment or check-in from me on 2026-10-16 (in Tcl's workflow Pending means "resolved, waiting for confirmation"). Reproduced on Linux (main 9.1 tip, 9.0 tip, 8.7b1, 8.6 tip, 8.5, 8.4 and old releases), on FreeBSD, NetBSD, OpenBSD and OpenIndiana VMs for platform tickets, and on a Windows 11 VM. Rows marked patch link to a prototype patch (287 in all, in tcl-triage/prototypes). Updated 2026-11-11: the emoji before each ticket id shows its progress.

Already fixed 188

⏳a127bf64152026auto_path is documented in two manpagespatchFIXED since 9.1 (main) (312151d236, 2026-06-16); 9.0 and 8.6 still have the duplicate (harmless, docs only)

Open ticket a127bf6415 · 38. Init - Library - Autoload · filed 2026-06-04 · Pending (resolved, waiting for confirmation)

⏳ Since triage (updated 2026-11-11)

Status now: Pending (Fixed)

    Full verdict

    FIXED since 9.1 (main) (312151d236, 2026-06-16); 9.0 and 8.6 still have the duplicate (harmless, docs only)

    • Platform docs
    • Tested doc/library.n and doc/tclvars.n on main, 9.0, 8.6. On main library.n now points to tclvars for auto_path. The removed library.n entry had a lappend auto_path [file dirname [info script]]/lib example and a "file normalize if the script uses cd" note; neither was moved to tclvars.n.
    • Cause n/a
    • Prototype prototypes/a127bf6415.patch (10 changed lines; optional: puts the lost example and note into the tclvars.n auto_path entry; checked with man -l)
    • Suggested ticket comment The resolution holds on main (312151d236); the ticket can be closed. The lappend example and the file normalize note from the old library.n entry could also go into tclvars.n (small patch available).
    ⬜6094de399d2026registry and dde commands are available in static builds without a package requireFIXED since 9.0.4 (624bac9741, 2026-05-16; also in 9.0 changes.md) and on main since 9.1b0 (TIP 755, 01705864c9, 2026-06-02). Not fixed in 8.6 …

    Open ticket 6094de399d · 38. Init - Library - Autoload · filed 2026-05-10

    Full verdict

    FIXED since 9.0.4 (624bac9741, 2026-05-16; also in 9.0 changes.md) and on main since 9.1b0 (TIP 755, 01705864c9, 2026-06-02). Not fixed in 8.6 (8.6.18 is the last release). Checked by reading the code only, no static Windows build here.

    • Platform Windows (static builds)
    • Tested code reading on main/9.0/8.6. 9.0: win/tclAppInit.c no longer calls Registry_Init/Dde_Init; static builds get package ifneeded ... load {} Registry/Dde from tclInterp.c. Main: TclInitStaticPackages (a TIP 755 post-init callback) registers Dde with package ifneeded dde 1.5.0. registry moved into the core as ::tcl::registry, and package require registry (1.4) adds the ::registry alias. Script to confirm on a static build: win/6094de399d.tcl
    • Cause (old) tclAppInit.c ran Registry_Init/Dde_Init directly in static builds.
    • Prototype none (already fixed)
    • Suggested ticket comment Fixed in 9.0.4 by 624bac9741 and in 9.1 by TIP 755 (01705864c9); 8.6 will not get it. Can be closed once checked on a static build.
    ⬜4f06d7ca082026package require in secondary interpreters fails for static packagesFIXED since 9.0.4 (624bac9741, 2026-05-16) and on main since 9.1b0 (TIP 755, 01705864c9, 2026-06-02); 8.6 not fixed (workaround `load {} Regist…

    Open ticket 4f06d7ca08 · 39. Package Manager · filed 2026-05-09

    Full verdict

    FIXED since 9.0.4 (624bac9741, 2026-05-16) and on main since 9.1b0 (TIP 755, 01705864c9, 2026-06-02); 8.6 not fixed (workaround load {} Registry). Checked by reading the code only.

    • Platform Windows (static builds)
    • Tested code reading. The same check-ins as 6094de399d. 9.0 adds the package ifneeded registry/dde ... load {} ... lines to the child-interp init script in tclInterp.c (static Windows builds only). Main registers them for every interp through the TIP 755 post-init callback. The test package-bug-4f06d7ca08 (constraint isStaticBuild) is in 9.0. The same script covers this one: win/4f06d7ca08.tcl
    • Cause (old) static packages were registered with Tcl_StaticLibrary on the main interp only, and no package ifneeded existed for child interps.
    • Prototype none (already fixed)
    • Suggested ticket comment Fixed in 9.0.4 by 624bac9741 (test package-bug-4f06d7ca08) and in 9.1 by TIP 755 (01705864c9). Can be closed; 8.6 keeps the load {} Registry workaround.
    ⏳8f5e7a26332025Tk build for 9.1 failing in safe-1.2 testcaseFIXED on the Tk side (safe-1.2 now builds its expected list from interp hidden of a plain Tcl safe interp; Jan re-enabled it 2026-03-11). Tcl sid…

    Open ticket 8f5e7a2633 · 06. Time Measurement · filed 2025-11-13 · Pending (resolved, waiting for confirmation)

    ⏳ Since triage (updated 2026-11-11)

    Status now: Pending (Fixed)

      Full verdict

      FIXED on the Tk side (safe-1.2 now builds its expected list from interp hidden of a plain Tcl safe interp; Jan re-enabled it 2026-03-11). Tcl side unchanged: main still hides clock, tcl:clock:add/format/scan in safe interps (9.0 does not); clock format/clock seconds work there via aliases.

      • Platform generic
      • Tested main c3bf6dd052 and 9.0.5 (interp hidden of a safe interp, work-a6/t2.tcl); read ~/tcltk/tk/tests/safe.test
      • Cause Tcl 9.1 clock init reorg made clock an unsafe ensemble with aliases; dkf's branch bug-8f5e7a2633 (last 58f8db9362, 2026-03-11) to mark clock safe is unmerged.
      • Prototype none (Tk test already fixed; the Tcl policy part is on dkf's branch)
      • Suggested ticket comment The Tk safe-1.2 failure is gone. The ticket can be closed unless the bug-8f5e7a2633 branch (making [clock] safe in 9.1 as in 9.0) is still wanted; it is not merged.
      ✅1577f1d3732025various fundamental problems with recursive mutexes and further consequencesFIXED since 9.0.3 (cbb5b13ed1 2025-09-22 Unix, d27472f8b2 2025-09-24 Windows, fc66aa236c 2025-09-27 tclEncoding.c; all via ticket 893f8cc5db, Clo…

      Open ticket 1577f1d373 · None · filed 2025-08-13

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Duplicate)

      • 🔧 2026-10-21 check-in bc65a512d2 on unix-monotonic-condwait, not merged: Fix 3328635fff: Use CLOCK_MONOTONIC for timed condition waits on Unix where pthread_condattr_setclock() is available, so changes of the system time do not aff
      • 💬 2026-10-21 serhiy.storchaka: The nested-mutex and condition-wait problems from this thread were handled in 893f8cc5db and fixed in 9.0.3 (cbb5b13ed1, d27472f8b2, fc66aa236c). Closing as a duplicate. The realtime-clock issue of timed condition waits deserves a ticket of its own if wanted.
      • 💬 2026-10-21 serhiy.storchaka: The remaining clock issue (timed waits on a wall-clock deadline) is tracked in 3328635fff. Proposed fix in branch unix-monotonic-condwait.
      Full verdict

      FIXED since 9.0.3 (cbb5b13ed1 2025-09-22 Unix, d27472f8b2 2025-09-24 Windows, fc66aa236c 2025-09-27 tclEncoding.c; all via ticket 893f8cc5db, Closed/Fixed)

      • Platform generic
      • Tested code reading only. The ticket only points at the tcl-core thread (no repro, and no network here). 8.6 has no TIP 509 recursive mutexes ("> 8.6").
      • Cause TIP 509 nested mutexes: Tcl_ConditionWait released only one level of a recursively held mutex (Unix and Win32), and tclEncoding.c depended on recursion. 893f8cc5db fixed the wait/unlock logic, tclEncoding.c no longer needs recursion, and mutex/condition tests went in (apn-mutex-tests, 2025-09-30). One point from the thread is still open: unix/tclUnixThrd.c Tcl_ConditionWait still computes the deadline from Tcl_GetDayTime() and calls pthread_cond_init(..., NULL), i.e. CLOCK_REALTIME, so timed waits follow wall-clock jumps. That is a separate RFE-sized change.
      • Prototype none (already fixed; the monotonic-clock part is a design change)
      • Suggested ticket comment The nested-mutex problems from this thread were fixed under 893f8cc5db (in 9.0.3 and 9.1). What is left is that Tcl_ConditionWait on Unix still uses the realtime clock (no CLOCK_MONOTONIC condattr). Close this as a duplicate of 893f8cc5db, or keep it open only for the monotonic-clock issue?
      ⬜248c105fcc2025Errors installing Tcl 9.1 on macOSpatchFIXED in main only (3bc7c91ea8 + 67924cd5df, 2026-04-14/15, PHILB/jan.nijtmans: SC_PROG_TCLSH rejects a tclsh older than 8.6); still present in 9.0…

      Open ticket 248c105fcc · 53. Configuration and Build Tools · filed 2025-05-29

      🧪 Prototype check (2026-10-24)

      248c105fcc.patch: VERIFIED (the configure half was already fixed on main by 3bc7c91ea8/67924cd5df, as triage said)

      • Base main (tip d6dd6d0f71)
      • Before ~/tcltk/8.5/bin/tclsh8.5 tools/tcltk-man2html.tcl -> ERROR: version conflict ... have 8.5.19, need 8.6- / If running this script from 'make html', set the NATIVE_TCLSH environment variable to point to an installed tclsh9.0 (or the equivalent tclsh90.exe on Windows). (misleading: unix/Makefile assigns NATIVE_TCLSH = /usr/bin/tclsh9.0, so only make NATIVE_TCLSH=... overrides it; 8.6 is enough).
      • Patch applied cleanly (6 changed lines)
      • After same command -> If running this script from 'make html', pass NATIVE_TCLSH=<path> to make (or re-run configure) to point to an installed tclsh 8.6 or later.; make -n html NATIVE_TCLSH=/x/tclsh8.6 uses /x/tclsh8.6; with a 9.1 tclsh the script still starts (--help prints usage).
      • Test none possible (build-tool message)
      Full verdict

      FIXED in main only (3bc7c91ea8 + 67924cd5df, 2026-04-14/15, PHILB/jan.nijtmans: SC_PROG_TCLSH rejects a tclsh older than 8.6); still present in 9.0 and 8.6 configure

      • Platform build (macOS framework install; cause generic unix)
      • Tested unix/tcl.m4 on main has the 8.6- version check, 9.0/8.6 tips do not. macOS itself CANNOT-TEST, but the install path (macosx/GNUmakefile adds html-tcl to INSTALL_TARGETS; NATIVE_TCLSH = @TCLSH_PROG@ found Apple's /usr/bin/tclsh 8.5.9) is generic. Setting NATIVE_TCLSH in the environment has no effect because the Makefile assigns it (only make NATIVE_TCLSH=... overrides) - which answers the reporter's second question.
      • Cause unix/tcl.m4:SC_PROG_TCLSH took the first tclsh on PATH regardless of version (fixed on main); tools/tcltk-man2html.tcl's error text says "NATIVE_TCLSH environment variable" and "tclsh9.0", both misleading.
      • Prototype prototypes/248c105fcc.patch (3 lines: error message now says to pass NATIVE_TCLSH=<path> to make and asks for tclsh 8.6 or later). Checked with tclsh8.5.
      • Suggested ticket comment Main now rejects a tclsh older than 8.6 in configure (&#91;3bc7c91ea8&#93;, &#91;67924cd5df&#93;); 9.0 does not have it yet. NATIVE_TCLSH is a make variable, so it has to be given as make NATIVE_TCLSH=..., not in the environment; proposed patch fixes the misleading message in tcltk-man2html.tcl.
      ✅f2b5f89c0d2024clock scan of leapsecond: wrong result in 8.6 and 9.0FIXED since 8.6.16 for the 8.6 day-off error (595fad24d7, 2024-11-15). The 9.0/main error for 23:59:60 is the documented behaviour (NOT-A-BUG: …

      Open ticket f2b5f89c0d · 16. Commands A-H · filed 2024-11-14

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-10-21 serhiy.storchaka: The 8.6 previous-day result is fixed in 8.6.16 by 595fad24d7. For 9.x it was decided not to accept leap seconds; leap-second acceptance was reverted and clock.n documents "leap seconds are not supported" since 9.0.1 (a5b130a5fa). Closing.
      Full verdict

      FIXED since 8.6.16 for the 8.6 day-off error (595fad24d7, 2024-11-15). The 9.0/main error for 23:59:60 is the documented behaviour (NOT-A-BUG: "leap seconds are not supported" was added to clock.n in a5b130a5fa, 2024-11-23).

      • Platform generic
      • Tested main, 9.0.5, 8.6.18, 8.6.16, 8.6.15. 8.6.15: free-form 2012-06-30 23:59:60 → 1341014399 (previous day; 24:00:00 also wrong). 8.6.16+: 1341100800, the same as the formatted scan. 9.0/main: "invalid time" with the default -validate 1. 24:00:00 → next day 00:00 on all of them.
      • Cause (8.6, fixed) ToSeconds -1 in the free-form scan; the remaining part is a design decision (POSIX time has no leap second).
      • Prototype none
      • Suggested ticket comment The 8.6 previous-day result is fixed since 8.6.16 (595fad24d7). In 9.x, rejecting 23:59:60 is now documented in clock.n (a5b130a5fa). Since the resolution is "Later" and the leap-second design question was dropped, this can be closed.
      ⬜c5825d32352024tip 430 install.tcl/SUPPORT_BUILTIN_ZIP_INSTALLFIXED since 9.0.3 (typo: 22ad86b21b, 2025-09-05); the C part is now explicitly unimplemented

      Open ticket c5825d3235 · - New Builtin Commands · filed 2024-10-11

      Full verdict

      FIXED since 9.0.3 (typo: 22ad86b21b, 2025-09-05); the C part is now explicitly unimplemented

      • Platform generic
      • Tested code reading of library/install.tcl and generic/tclZipfs.c on main and 9.0 tip (8.6 has no install.tcl)
      • Cause install.tcl had two mkzip switch arms; dkf's rework 22ad86b21b "rework install.tcl..." (9.0 branch, merged to main) changed the second to mkimg. In TclZipfs_AppHook the SUPPORT_BUILTIN_ZIP_INSTALL block now starts with #error "SUPPORT_BUILTIN_ZIP_INSTALL not implemented - TODO" (main and 9.0), so defining it fails on purpose instead of with the argvPtr error.
      • Prototype none (typo already fixed; implementing the install hook is a feature, not a small fix)
      • Suggested ticket comment The duplicate <tt>mkzip</tt> arm in install.tcl was fixed by 22ad86b21b (9.0.3). The SUPPORT_BUILTIN_ZIP_INSTALL hook in tclZipfs.c is now marked with <tt>#error ... not implemented - TODO</tt>; is anyone planning to finish it, or should the dead code be removed and this ticket closed?
      ⬜e38dce74e22024Command line built with list not properly quotedFIXED since 9.0.0 (8627aad9bd, 2024-09-25); left open only for sebres' consistency suggestion (design question)

      Open ticket e38dce74e2 · 17. Commands I-L · filed 2024-09-17

      Full verdict

      FIXED since 9.0.0 (8627aad9bd, 2024-09-25); left open only for sebres' consistency suggestion (design question)

      • Platform generic
      • Tested main build, 9.0.5, 8.6.18: list {*}$cmd {*}$args with cmd=##foo gives {##foo} everywhere; lrange/lreplace/linsert canonicalize too; lappend L (no values, compiled and not) keeps ##foo on all three, same as 8.6
      • Cause TIP 625 shortcut in TclListObjAppendElements for elemCount 0 (regression from d9b720c46d, fixed by 8627aad9bd)
      • Prototype none (fixed; the remaining question is whether no-op lappend should canonicalize — sebres' branch bug-e38dce74e2-214cc0eb22-v2 is unmerged, last activity 2024-11-14, its core-* tag cancelled)
      • Suggested ticket comment The reported regression was fixed by 8627aad9bd in 9.0.0. Behaviour now matches 8.6 for all list commands, including no-op <tt>lappend</tt> (keeps the string rep in both). Unless someone wants to pursue branch bug-e38dce74e2-214cc0eb22-v2, this can be closed.
      ⬜9c258a841a2024format - undocumented type specifiers + possible bugsFIXED since 9.0.0 (0c66ee7d75 merged by 9c0c9d1a4a, 2024-05-30): j/z/t now truncate like C and z/t/j/q are documented. Leftovers are by design:…

      Open ticket 9c258a841a · - New Builtin Commands · filed 2024-05-28

      Full verdict

      FIXED since 9.0.0 (0c66ee7d75 merged by 9c0c9d1a4a, 2024-05-30): j/z/t now truncate like C and z/t/j/q are documented. Leftovers are by design: I/I32/I64 still accepted on all platforms and undocumented (Jan's Windows-only variant 01e6a242c4 was closed unmerged); %a strips trailing zeros on glibc (accepted as OK)

      • Platform generic
      • Tested main build and 9.0.5 with the reporter's loop: jd/td/zd → -9223372036854775807 / 0 (truncated, same as ld); %a 16 → 0x1p+4 on Linux
      • Cause n/a (fixed)
      • Prototype none (fixed; the remaining items are design decisions)
      • Suggested ticket comment The j/z/t truncation and the documentation were fixed in 9.0.0 (9c0c9d1a4a). I/I32/I64 stay undocumented on purpose (they are meant for Tcl_ObjPrintf), and the %a trailing-zero difference was judged harmless. Can this be closed?
      💬e4d394cd242024mingw builds fail for pkgsFIXED for the Thread/minizip part (Thread 37b67651e1, 2025-01-07: zip built with zipfs mkzip); the 32-bit part CANNOT-TEST (no 32-bit MinGW toolc…

      Open ticket e4d394cd24 · 53. Configuration and Build Tools · filed 2024-05-02

      💬 Since triage (updated 2026-11-11)

      Status now: Open

      • 💬 2026-10-21 serhiy.storchaka: Rechecked with main and Thread 3.0.7 in pkgs/, on Windows 11 with MSYS2 and no zip or minizip on PATH. Both a 32-bit MINGW32 (i686) and a 64-bit UCRT64 build configure and build pkgs/thread during make; make packages gives tcl9thread307.dll, and package require thread, package require Ttrace (from the attached zip) and thread::send work in both. The Thread part was fixed by Thread itself, which now builds its zip …
      Full verdict

      FIXED for the Thread/minizip part (Thread 37b67651e1, 2025-01-07: zip built with zipfs mkzip); the 32-bit part CANNOT-TEST (no 32-bit MinGW toolchain on the VM)

      • Platform build (Windows MinGW)
      • Tested code reading of Thread trunk (thread.fossil): Makefile.in now runs echo "zipfs mkzip ..." | $(TCLSH_NATIVE), no ZIP_PROG/minizip use. The TEA tclconfig tcl.m4 trunk (tclconfig.fossil) still sets ZIP_PROG=./minizip relative to the package dir when no zip is on PATH, so other TEA packages relying on it can hit the same error; Tcl's own win/tcl.m4 has the same fallback but Tcl builds minizip itself.
      • Cause TEA tcl.m4 ZIP fallback points at a minizip the package never builds (Thread no longer uses it).
      • Prototype none (fix belongs to Thread/tclconfig; Thread already fixed)
      • Suggested ticket comment The Thread failure was fixed in Thread &#91;f36e5b333c&#93; (it now uses zipfs mkzip with the native tclsh). Does the 32-bit MinGW build still skip pkgs with current 9.0? If not, this can be closed.
      Windows VM, run 1
      • Platform build (Windows MinGW)
      • Verdict FIXED for the Thread/minizip part (Thread 37b67651e1, 2025-01-07: zip built with zipfs mkzip); the 32-bit part CANNOT-TEST (no 32-bit MinGW toolchain on the VM)
      • Tested VM MSYS2 UCRT64 build of main with Thread trunk 3.0.7 (39d9f08aa0) placed in pkgs/, no zip or minizip on PATH: make packages install-packages succeeds; tcl9thread307.dll has the zip attached (package require Ttrace works), thread::create/send work
      • Cause (historic) Thread's Makefile used the TEA ZIP_PROG=./minizip fallback; the fallback is still in tclconfig tcl.m4 for other packages
      • Prototype none
      • Suggested ticket comment With current Thread (3.0.7) the 64-bit MinGW build of pkgs works without zip/minizip on PATH (Thread now uses zipfs mkzip). Could someone with a 32-bit MinGW check whether pkgs are built there? Otherwise this can be closed.
      ✅f2f597cb782024-singleproc 1 testing defeats two validity mode coverageFIXED since 9.0.4 / 9.1b0 (a37536d4d7, 2026-03-07, merged to main in 7018ffb43e); clock-ivm.test now restores the mode after sourcing clock.tes…

      Open ticket f2f597cb78 · 06. Time Measurement · filed 2024-04-04

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-10-21 serhiy.storchaka: Fixed in 9.0.4 and 9.1.0 by a37536d4d7 (trunk: 7018ffb43e): clock-ivm.test restores the validity mode after running, so with -singleproc 1 clock.test runs in the default mode again.
      Full verdict

      FIXED since 9.0.4 / 9.1b0 (a37536d4d7, 2026-03-07, merged to main in 7018ffb43e); clock-ivm.test now restores the mode after sourcing clock.test. 8.6 has no clock-ivm.test.

      • Platform generic
      • Tested main make test TESTFLAGS="-singleproc 1 -file clock*.test" prints "off" for clock-ivm.test and "on" for clock.test, all pass
      • Cause tests/clock-ivm.test flipped the validity mode but did not flip it back.
      • Prototype none (fixed)
      • Suggested ticket comment Fixed by a37536d4d7 (9.0) / 7018ffb43e (main): clock-ivm.test switches the validity mode back, so with -singleproc 1 clock-ivm.test runs with "off" and clock.test with "on".
      ✅0dc653a2fb2024Undocumented feature: Tcl_InitStubs sets version/exact parameters to current for alpha/beta headersFIXED since 9.0.2 (1e24bb8799, 2025-02-01 "Since TCL_RELEASE_LEVEL == TCL_FINAL_RELEASE, those Tcl_InitStubs() variants are now obsolete")

      Open ticket 0dc653a2fb · 50. Embedding Support · filed 2024-04-03

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-10-21 serhiy.storchaka: The remaining alpha/beta special cases of the Tcl_InitStubs macro were removed by 1e24bb8799 (in 9.0.2), so there is nothing left to document.
      Full verdict

      FIXED since 9.0.2 (1e24bb8799, 2025-02-01 "Since TCL_RELEASE_LEVEL == TCL_FINAL_RELEASE, those Tcl_InitStubs() variants are now obsolete")

      • Platform generic
      • Tested read generic/tcl.h at core-9-0-b1, b2, b3, 9.0.0, 9.0.1, 9.0.2, 9.0 tip, main
      • Cause generic/tcl.h had a #elif TCL_RELEASE_LEVEL == TCL_FINAL_RELEASE variant of the Tcl_InitStubs macro; up to 9.0b1 it forced an exact match, from 9.0b2 to 9.0.1 non-final headers passed "9.0b2"/"9.0b3" as minimum version unless exact. This was the "last bits" Jan kept the ticket open for; 9.0.2 and main have only the plain macro.
      • Prototype none (nothing left to fix; no doc needed since the behaviour is gone)
      • Suggested ticket comment The alpha/beta special case of the Tcl_InitStubs macro is gone since 1e24bb8799 (in 9.0.2 and 9.1). Closing.
      ✅c9098c8bae2024make html fails in tclib when included in tcl/pkgsFIXED since 9.0.2 / 8.6.17 (ea8480993a, 3b95f5210d / 09d7e24d80, 2025-01-27, oehhar, ticket 27fa20249d - a duplicate)

      Open ticket c9098c8bae · 53. Configuration and Build Tools · filed 2024-03-30

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Duplicate)

      • 💬 2026-10-21 serhiy.storchaka: Fixed by 3b95f5210d for 27fa20249d (in 9.0.2 and 8.6.17): tcltk-man2html.tcl now skips package folders without configure.ac/configure.in ("package folder without package ignored"). Closing as a duplicate of 27fa20249d.
      Full verdict

      FIXED since 9.0.2 / 8.6.17 (ea8480993a, 3b95f5210d / 09d7e24d80, 2025-01-27, oehhar, ticket 27fa20249d - a duplicate)

      • Platform build
      • Tested main tools/tcltk-man2html.tcl with an extra pkgs/junk dir without configure.ac/.in, run with own main tclsh and with tclsh8.6: prints "package folder without package ignored: junk" and completes. Same code in 9.0 and 8.6 tips.
      • Cause (was) uncaught open of pkgs/<dir>/configure.ac in tcltk-man2html.tcl.
      • Prototype none (already fixed)
      • Suggested ticket comment Fixed by &#91;3b95f5210d&#93; (ticket &#91;27fa20249d&#93;, in 9.0.2 and 8.6.17): package folders without configure.in/ac are now skipped. Can be closed as a duplicate of 27fa20249d.
      ⬜1b8a893ded2024TCL_PACKAGE_PATH path is (wrongly) braced.FIXED since 8.6.15 (c2a94d79f6, 2024-03-04, jan.nijtmans; brace form introduced by 6acb955a47, present 8.6.11-8.6.14); reopened for a design disput…

      Open ticket 1b8a893ded · 53. Configuration and Build Tools · filed 2024-03-03

      Full verdict

      FIXED since 8.6.15 (c2a94d79f6, 2024-03-04, jan.nijtmans; brace form introduced by 6acb955a47, present 8.6.11-8.6.14); reopened for a design dispute, not a bug

      • Platform build (unix)
      • Tested installed tclConfig.sh: 8.6.10 '/…/lib ', 8.6.11-8.6.14 '{/…/lib} ', 8.6.15 / 8.6.18 / 9.0.5 / main '/…/lib' (colon-separated when several dirs). tcl_pkgPath on 9.0.5 is a proper list.
      • Cause (was) unix/configure.in brace-quoting of TCL_PACKAGE_PATH entries.
      • Prototype none (the remaining question - Tcl list vs. native path syntax, also RFE 3057b6261 - needs a decision/TIP, not a patch)
      • Suggested ticket comment The extra braces are gone since 8.6.15 (&#91;c2a94d79f6&#93;); TCL_PACKAGE_PATH is now colon-separated. Whether it should be a Tcl list again is the design question of RFE &#91;3057b6261&#93;; suggest closing this one and continuing there.
      ⬜f23022e07a2024Encoding table mismatches vis-a-vis ICUpatchkoi8-u part FIXED since 8.6.15 / 9.0.0 (5f43417351, 2024-03-15; 8.6.14 still gives U+0403); mac tables still REPRODUCED (main, 9.0, 8.6)

      Open ticket f23022e07a · 44. UTF-8 Strings · filed 2024-02-19

      🧪 Prototype check (2026-10-24)

      f23022e07a.patch: VERIFIED (after adding the tools/encoding sources). macUkraine FF and macDingbats were not touched, as in the triage.

      • Base main (tip d6dd6d0f71)
      • Before work-a3/vf23.tcl -> macGreek: U+00AD U+F8A0, macRomania: U+015E U+2126 U+015F U+00A4 U+0162 U+0163; Python's mac_greek/mac_romanian give U+20AC U+00AD and U+0218 U+03A9 U+0219 U+20AC U+021A U+021B
      • Patch applied (encoding.test hunk offset by 11 lines), then extended. The original changed only library/encoding/*.enc, but those files are generated from tools/encoding/macGreek.txt and macRomania.txt (tools/encoding/Makefile, txt2enc), so I updated the 8 entries there as well. txt2enc output from the new .txt files is identical to the patched .enc data. Now 30 changed lines.
      • After (zip rebuilt): vf23.tcl -> the Apple/Python values, round trip OK. encoding.test 231 passed/0 failed, both with TCL_LIBRARY and via make test
      • Test existing encoding-bug-f23022e07a (from the patch) fails on the unpatched tip / passes after
      Full verdict

      koi8-u part FIXED since 8.6.15 / 9.0.0 (5f43417351, 2024-03-15; 8.6.14 still gives U+0403); mac tables still REPRODUCED (main, 9.0, 8.6)

      • Platform generic
      • Tested 0x80–0xFF of each single-byte table on main vs Python's codecs (based on current Apple/Unicode mapping files; no ICU here). macGreek differs at 9C (Tcl U+00AD, Apple U+20AC) and FF (U+F8A0 vs U+00AD); macRomania at AF/BF/DE/DF (cedilla vs comma-below U+0218/0219/021A/021B), BD (U+2126 vs U+03A9), DB (U+00A4 vs U+20AC); macUkraine FF (U+00A4 vs U+20AC, Python compares against mac_cyrillic, so less certain). macCyrillic, macIceland, macTurkish, macCroatian, macRoman, koi8-r/u identical. macDingbats not checked (no reference codec available).
      • Cause library/encoding/mac*.enc are from the 1995-era Apple tables (pre-euro, pre-comma-below).
      • Prototype prototypes/f23022e07a.patch (61 lines, test encoding-bug-f23022e07a): updates the 8 macGreek/macRomania entries; encoding.test passes. Note: the main build embeds the library in libtcl9.1.1.zip and //zipfs: comes first in encoding dirs, so the zip must be deleted and rebuilt to test .enc changes.
      • Suggested ticket comment koi8-u was fixed in 8.6.15 (5f43417351). macGreek (9C, FF) and macRomania (AF, BD, BF, DB, DE, DF) still differ from the current Apple mappings; proposed update of those tables with test encoding-bug-f23022e07a.
      ✅d63061a1ac2024PRIVATE != CONTROL in UnicodeFIXED since 8.6.15 / 9.0.0 (0480bdc823, 2024-03-03)

      Open ticket d63061a1ac · 44. UTF-8 Strings · filed 2024-01-15

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-10-21 serhiy.storchaka: Fixed by 0480bdc823 (in 8.6.15 and 9.0.0): private-use characters (category Co) are no longer string is control. Cc and Cf are unchanged.
      Full verdict

      FIXED since 8.6.15 / 9.0.0 (0480bdc823, 2024-03-03)

      • Platform generic
      • Tested string is control on U+E000/U+F8FF/U+F0000/U+10FFFD: 1 on 8.6.13 and 8.6.14, 0 on 8.6.15, 8.6.18, 9.0.5 and main; Cc/Cf (U+000A, U+00AD, U+200B, U+FEFF) still 1.
      • Cause (was) the "control" table in generic/tclUniData.c included category Co.
      • Prototype none (already fixed).
      • Suggested ticket comment Fixed by 0480bdc823 (in 8.6.15 and 9.0.0): private-use characters are no longer "string is control". Can be closed.
      ✅6d2ef441cc2023Password protected zifps files cannot be opened in a+ modeFIXED since 9.0.0 (9c204498a8, 2023-09-29, core-8-branch); Pending → can be closed

      Open ticket 6d2ef441cc · - New Builtin Commands · filed 2023-09-28 · Pending (resolved, waiting for confirmation)

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-10-21 serhiy.storchaka: Fixed in 9.0.0 by 9c204498a8 (trunk: bea92dfdc5). The remaining macOS-only failures of the password tests are tracked in bbe7c6ff9e.
      Full verdict

      FIXED since 9.0.0 (9c204498a8, 2023-09-29, core-8-branch); Pending → can be closed

      • Platform generic
      • Tested main build and 9.0.5: mount tests/zipfiles/test-password.zip with password, open .../cipher.bin a+ succeeds; zipfs.test passes on main
      • Cause n/a (fixed)
      • Prototype none
      • Suggested ticket comment Verified on 9.0.5 and 9.1: <tt>a+</tt> on a password-protected member works. The macOS-only failures are tracked separately in bbe7c6ff9e (tests disabled on Darwin there), so this can be closed.
      ✅c315de9e442023Password protected zip files can be rewritten without the passwordFIXED since 9.0.0 (9c204498a8, 2023-09-29); Pending → can be closed

      Open ticket c315de9e44 · - New Builtin Commands · filed 2023-09-27 · Pending (resolved, waiting for confirmation)

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-10-21 serhiy.storchaka: Fixed in 9.0.0 by 9c204498a8 (trunk: bea92dfdc5): encrypted members can no longer be opened for writing without the password. The macOS-only test failures are tracked in bbe7c6ff9e.
      Full verdict

      FIXED since 9.0.0 (9c204498a8, 2023-09-29); Pending → can be closed

      • Platform generic
      • Tested main build and 9.0.5: without the password, open in r, w, a and r+ all fail with "decryption failed - no password provided"
      • Cause n/a (fixed)
      • Prototype none
      • Suggested ticket comment Verified on 9.0.5 and 9.1: write modes on an encrypted member now fail without the password. The macOS part is tracked in bbe7c6ff9e; can be closed.
      ⬜f9764cd7a92023tcl Tcl_CreateTimerHandler has flaws which may never invoker input functionFIXED in main (9.1) by TIP 723 (merge acb0a91c2c, 2026-01-08, oehhar: Tcl_CreateTimerHandler and after use Tcl_GetMonotonicTime); REPRODUCED (9.0…

      Open ticket f9764cd7a9 · 55. Other Tools · filed 2023-09-21

      Full verdict

      FIXED in main (9.1) by TIP 723 (merge acb0a91c2c, 2026-01-08, oehhar: Tcl_CreateTimerHandler and after use Tcl_GetMonotonicTime); REPRODUCED (9.0, 8.6)

      • Platform generic
      • Tested without touching the system clock: LD_PRELOAD shim (work-a2/ft/shim.c) that makes the process's wall clock jump back one day 1 s after start; script after 1500 {set done ok}; after 5000 {...}; vwait done. Main (own build): fires after 1.5 s. 9.0.5 and 8.6.18: neither timer fires (killed after 20 s).
      • Cause 9.0/8.6 generic/tclTimer.c computes timer deadlines from Tcl_GetTime (wall clock); fixed on main by TIP 723.
      • Prototype none (fixed on main by a TIP; backporting the monotonic timer rework to 9.0/8.6 is not a small patch)
      • Suggested ticket comment Fixed for 9.1 by TIP 723 (&#91;acb0a91c2c&#93;): timers now use the monotonic clock, so setting the system clock back no longer delays them. 9.0 and 8.6 still use wall-clock time; close as fixed in 9.1 unless a backport is wanted.
      ⬜29b19f5ef02023Inconsistencies in zipfs path separator on WindowsFIXED since 9.0.0 (zipfs path handling merged in 492fb812e9 / trunk 35dd9ec56b, 2023-10-15, ticket c3364db38a): only / is a zipfs separator…

      Open ticket 29b19f5ef0 · - New Builtin Commands · filed 2023-09-14

      Full verdict

      FIXED since 9.0.0 (zipfs path handling merged in 492fb812e9 / trunk 35dd9ec56b, 2023-10-15, ticket c3364db38a): only / is a zipfs separator, consistently

      • Platform Windows
      • Tested Linux main: only / works, consistently (all three backslash forms → 0; backslash is an ordinary character there)
      • Cause likely generic path normalization on Windows: //zipfs:\mt\... starts with two separators and is normalized as a UNC-like root (host "zipfs:"), converting all \ to /, while //zipfs:/mt\... keeps the backslashes in the tail that zipfs then looks up literally. Note TclIsZipfsPath() in generic/tclZipfs.c looks inverted: on _WIN32 it requires an exact "//zipfs:/" prefix, on other platforms it accepts \ for / (35dd9ec56b, 2023-10-15).
      • Prototype none (needs Windows to decide where to normalize)
      • Suggested ticket comment On Unix only / is a zipfs separator, consistently. On Windows the leading //zipfs:\ form is probably accepted only because the generic normalizer treats it like a UNC root; also TclIsZipfsPath() accepts \ on Unix but not on Windows, which looks reversed.
      Windows VM, run 4
      • Platform Windows
      • Verdict FIXED since 9.0.0 (zipfs path handling merged in 492fb812e9 / trunk 35dd9ec56b, 2023-10-15, ticket c3364db38a): only / is a zipfs separator, consistently
      • Tested VM win/29b19f5ef0.tcl (tests/zipfiles/test.zip mounted at //zipfs:/mt) on 9.1b1 and 9.0.5: //zipfs:/mt/testdir/test2 exists=1; //zipfs:/mt\testdir\test2, //zipfs:\mt\testdir\test2, //zipfs:\mt/testdir/test2 all exists=0 (the report had 1 for the last two). Same as Linux.
      • Cause TclIsZipfsPath() on _WIN32 now requires the exact //zipfs:/ prefix, so a backslash after the volume is no zipfs path at all. Cosmetic leftover: file normalize //zipfs:\mt\x gives C://zipfs:\mt\x (treated as a volume-relative native path).
      • Prototype none
      • Suggested ticket comment Since 9.0.0 (35dd9ec56b) zipfs paths on Windows accept only /, also right after //zipfs: (tested on Windows 11 with 9.0.5 and 9.1: all three backslash forms give &#91;file exists&#93; 0). Closing.
      ⬜27b3ce29972023Memory access violation in scripts > 2**31 bytesFIXED since 9.0b1 (20ed83cf25, 2023-04-30) for the crash; ticket was reopened by dgp (2025-02-26) as a reminder to lift the INT_MAX limit in 9.1, w…

      Open ticket 27b3ce2997 · 47. Bytecode Compiler · filed 2023-04-28

      Full verdict

      FIXED since 9.0b1 (20ed83cf25, 2023-04-30) for the crash; ticket was reopened by dgp (2025-02-26) as a reminder to lift the INT_MAX limit in 9.1, which is an RFE

      • Platform generic
      • Tested main build and 9.0.5: try, eval, uplevel, proc body, namespace eval, apply with a 2**31+9 byte script all give "Script length 2147483657 exceeds max permitted length 2147483646." without crash; subst and info complete on the same string work. 8.6 not applicable (strings are limited to INT_MAX there).
      • Cause compiler and TIP 280 line tracking use int offsets; 20ed83cf25 added a length check.
      • Prototype none (lifting the limit is a large redesign)
      • Suggested ticket comment No crash any more in 9.0/9.1 (length check from 20ed83cf25). What remains is the RFE to support scripts > INT_MAX; suggest changing the type to RFE or closing.
      ⬜d34442777f2023Tcl_GetIndexFromObjStruct should have size_t offset parameterpatchFIXED since 9.0 in the API (88334c6e8c, 2022-10-23, TIP 628: Tcl_Size offset) — DOCS: doc/GetIndex.3 still says int offset on main and 9.0

      Open ticket d34442777f · 13. Index Object · filed 2023-04-21

      🧪 Prototype check (2026-10-24)

      d34442777f.patch: VERIFIED

      • Base main (tip d6dd6d0f71); docs only
      • Before generic/tcl.decls on the tip declares Tcl_Size offset; doc/GetIndex.3:47 still says .AP int offset in
      • Patch applied cleanly (2 changed lines)
      • After man -l GetIndex.3 shows Tcl_Size offset (in)
      • Test none possible (docs only)
      Full verdict

      FIXED since 9.0 in the API (88334c6e8c, 2022-10-23, TIP 628: Tcl_Size offset) — DOCS: doc/GetIndex.3 still says int offset on main and 9.0

      • Platform docs
      • Tested tcl.decls on 8.6 (int), 9.0 and main (Tcl_Size); ticket text is only "More information will follow soon", never followed up.
      • Cause doc not updated when the prototype changed.
      • Prototype prototypes/d34442777f.patch (1-line doc change .AP Tcl_Size offset in).
      • Suggested ticket comment In 9.0 the offset parameter is already Tcl_Size (signed, same width as size_t); only GetIndex.3 still says int. Proposed doc fix; unless size_t (unsigned) was really meant, this can be closed.
      🟢7e3f26c7482023Memory corruption false positive with TCL_MEM_DEBUGpatchFIXED since 9.0.0 (false positive: f505b17748, 2023-04-02) — but that fix turned the check into dead code. FREEDREFCOUNTFILLER is now 0xe8e8…e8, …

      Open ticket 7e3f26c748 · - New Builtin Commands · filed 2023-04-02

      🟢 Since triage (updated 2026-11-11)

      Status now: Open

      • 🔧 2026-10-18 check-in 5495155636 on memdebug-freed-marker, in main: Fix \[7e3f26c748\]: with TCL_MEM_DEBUG, the "previously disposed object" check could no longer fire, because freed memory was filled with 0x61 while \[f505b1774
      • 💬 2026-10-18 serhiy.storchaka: f505b17748 changed the expected refCount of freed objects to 0xE8..., but Tcl_DbCkfree still fills freed memory with 0x61, so the "decrementing refCount of previously disposed object" panic can no longer fire; a use-after-free decrement is caught only by the threaded check ("Tcl_Obj allocated in another thread", without file and line). The 0x61616161 checks in tclDisassemble.c and tclPreserve.c never matched on …
      Full verdict

      FIXED since 9.0.0 (false positive: f505b17748, 2023-04-02) — but that fix turned the check into dead code. FREEDREFCOUNTFILLER is now 0xe8e8…e8, while Tcl_DbCkfree still fills freed memory with GUARD_VALUE 0x61, so a freed Tcl_Obj has refCount 0x6161616161616161 and the "decrementing refCount of previously disposed object" panic (with file/line) can never fire. A use-after-free is caught only by the threaded-build check ("Tcl_Obj allocated in another thread", no file/line), and not at all in non-threaded builds. tclDisassemble.c and tclPreserve.c (size_t refCount) still compare against the 32-bit 0x61616161. 8.6: refCount is int and 0x61616161 references cannot fit in a 32-bit address space, so in practice the false positive is 64-bit 9.x only.

      • Platform generic
      • Tested main --enable-symbols=mem build (out of tree, work-a6/memdbg), with a small C program that decrements a freed object (work-a6/uaf.c): before → "allocated in another thread"; after → "file = uaf.c, line = 10 / decrementing refCount of previously disposed object". I did not run the 0x61616162-element lrepeat (~13 GB).
      • Cause generic/tclObj.c FREEDREFCOUNTFILLER does not match tclCkalloc.c GUARD_VALUE (the comment says it should).
      • Prototype prototypes/7e3f26c748.patch (26 changed lines, no test: only MEM_DEBUG builds are affected) — new TCL_FREED_REFCOUNT in tclInt.h = (size_t)-1 / 0xFF * 0x61 (0x61 repeated over the width of size_t), used by tclObj.c, tclDisassemble.c and tclPreserve.c. On 64-bit no real refCount gets that high; on 32-bit 0x61616161 cannot be reached either. obj/lrepeat/preserve/compile tests pass on the mem-debug build.
      • Suggested ticket comment f505b17748 fixed the false positive, but the new filler 0xe8… no longer matches what Tcl_DbCkfree writes (0x61), so this panic can never fire now. Proposed fix: compare against 0x61 repeated over the width of size_t (0x6161616161616161 on 64-bit) in tclObj.c, tclDisassemble.c and tclPreserve.c.
      ✅6390566ecd2023Testcase cmdAH-4.3.13.00DC0000.tail.utf-32.tcl8.a fails sometimesFIXED since 9.0.0 (b27eacb7fc, 2023-04-04, trunk; the first attempt was 69a4b72885, 2023-03-17). Resolution is already "Fixed", only Status is …

      Open ticket 6390566ecd · 16. Commands A-H · filed 2023-03-17

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-10-21 serhiy.storchaka: Fixed by b27eacb7fc (in 9.0.0). The test passes repeatedly on main, and all utf-* convertfrom tests of cmdAH-4.3.13/14 are clean under valgrind. 8.6 has no utf-32 encoding with profiles and is not affected. Closing.
      Full verdict

      FIXED since 9.0.0 (b27eacb7fc, 2023-04-04, trunk; the first attempt was 69a4b72885, 2023-03-17). Resolution is already "Fixed", only Status is still Open. 8.6 has no utf-32 encoding and no profiles, so it is not affected.

      • Platform generic
      • Tested main own build — cmdAH-4.* run 10× (all pass); all cmdAH-4.3.1[34].*utf-* convertfrom tests (11744) under valgrind on a -DPURIFY build: no errors.
      • Cause generic/tclEncoding.c utf-32 → internal conversion tried to combine a lone low surrogate with an uninitialised previous value (relevant only for TCL_UTF_MAX < 4, which is gone in 9.x).
      • Prototype none (fixed)
      • Suggested ticket comment Fixed by b27eacb7fc (in 9.0.0). The test passes repeatedly and the utf-* convertfrom tests are clean under valgrind on main. Can be closed.
      🟢21dbc600ae2022exec hangs, iff stdin & stdout both were previously closed.FIXED on main and 9.0 (main 32b22bfeb5, core-9-0-branch d2c4f74bac, 2026-10-21; unreleased, will be in 9.1.1/9.0.5); still REPRODUCED on 8.6 (8.6.1…

      Open ticket 21dbc600ae · 16. Commands A-H · filed 2022-12-06

      🟢 Since triage (updated 2026-11-11)

      Status now: Open

      • 🔧 2026-10-21 check-in 32b22bfeb5 on trunk, in main, 9.0: Fix 21dbc600ae: exec hangs, iff stdin & stdout both were previously closed
      • 🔧 2026-10-21 check-in d2c4f74bac on core-9-0-branch, in main, 9.0: Fix 21dbc600ae: exec hangs, iff stdin & stdout both were previously closed
      • 🔧 2026-10-20 check-in 4af19e7744 on exec-closed-stdio, in main: Declare errPipeIn/errPipeOut volatile in TclpCreateProcess: fork() is vfork() here, which may clobber them (-Wclobbered).
      • 💬 2026-10-20 jan.nijtmans: I'm seeing this compiler warning, even though it appears to work fine: unix/tclUnixPipe.c:469:13: warning: variable ‘errPipeIn’ might be clobbered by ‘longjmp’ or ‘vfork’ [-Wclobbered] 469 | TclFile errPipeIn, errPipeOut; | ^~~~~~~~~ unix/tclUnixPipe.c:469:24: warning: variable ‘errPipeOut’ might be clobbered by ‘longjmp’ or ‘vfork’ [-Wclobbered] 469 | TclFile errPipeIn, errPipeOut; | ^~~~~~~~~~
      • 💬 2026-10-20 jan.nijtmans: This warning is caused by the line pid = fork(); in line 602
      • 💬 2026-10-20 serhiy.storchaka: The warning is not new in this branch: TclpCreateProcess() is unchanged, but unix/tclUnixPipe.c has #define fork vfork when HAVE_VFORK is set, so pid = fork(); is really vfork(), after which gcc warns about locals that may be clobbered. dsArray and newArgv there were already volatile for this reason, errPipeIn/errPipeOut were not. Fixed in 4af19e7744 on the branch: they are volatile now (TclpCreatePipe() fills p…
      Full verdict

      FIXED on main and 9.0 (main 32b22bfeb5, core-9-0-branch d2c4f74bac, 2026-10-21; unreleased, will be in 9.1.1/9.0.5); still REPRODUCED on 8.6 (8.6.18 and core-8-6-branch tip)

      • Platform unix
      • Tested close stdin; close stdout; exec true: main own build (c092dae4af) OK, exec-23.1 passes; installed 9.0.5 (pre-fix build) and 8.6.18 hang (timeout 5 s).
      • Cause unix/tclUnixPipe.c: with fds 0/1 closed the pipeline's pipe got descriptors 0 and 1, which TclpCloseFile() never closes, so the parent kept the write end open.
      • Existing work exec-closed-stdio (c6f3b26c02, 4af19e7744 volatile fix after Jan's -Wclobbered report) merged into main and 9.0 by jan.nijtmans, branch closed, core-exec-closed-stdio tag cancelled; not in core-8-6-branch (no F_DUPFD code there).
      • Prototype none (branch exists)
      • Suggested ticket comment none needed now; when backporting to 8.6 (or deciding not to), close the ticket (Jan merged it but left it Open).
      🟢697b1bbfe32022Incorrect argument handling by [regexp]FIXED on main only, unreleased (6e93204003, 2026-10-14); REPRODUCED (9.0, 8.6)

      Open ticket 697b1bbfe3 · 18. Commands M-Z · filed 2022-09-18

      🟢 Since triage (updated 2026-11-11)

      Status now: Open

      • 🔧 2026-10-10 check-in a64804ce70 on regexp-compile-dash-pattern, in main: Fix cases #2 and #4 of 697b1bbfe3: the bytecode compiler for [regexp] checked for a leading "-" only in simple words, so patterns like \- or -\\ were compiled
      • 🔧 2026-10-11 check-in 0b36adaf16 on regexp-compile-runtime-switch, in main: Fix [1902890]: compiled [regexp] did not treat an RE value starting with "-" as a switch, unlike the command. When the RE is not known at compile time and there
      Full verdict

      FIXED on main only, unreleased (6e93204003, 2026-10-14); REPRODUCED (9.0, 8.6)

      • Platform generic
      • Tested (in procs, i.e. compiled): #2 regexp \- $str, #6 set i -; regexp $i $str: main error (like the command), 9.0/8.6 return 1. #4 regexp -\\ $str: main bad option "-\", 9.0/8.6 "invalid escape \ sequence". Interpreted regexp - $str: 8.6 "bad option", main/9.0 "ambiguous option" (all options abbreviatable since 3069454 fix). #1/#3/#5 are correct behaviour.
      • Cause TclCompileRegexpCmd checked for a leading "-" only in backslash-free literal words, and not at all for variable REs.
      • Prototype none (merged)
      • Existing work regexp-compile-dash-pattern (tip a64804ce70) and regexp-compile-runtime-switch (tip 0b36adaf16), merged into main by 6e93204003; not in 9.0/8.6
      • Suggested ticket comment Fixed on main by 6e93204003; compiled and interpreted &#91;regexp&#93; now agree for all six cases. Not in 9.0/8.6.
      ⬜5294e6feca2022minizip for TIP #430 brokenFIXED since 9.0.0 (3481a747a0, 2022-04-15, core-8-branch, later merged up). The recursive-directory change (-r, addPathToZip) has been kept thr…

      Open ticket 5294e6feca · - New Builtin Commands · filed 2022-04-14

      Full verdict

      FIXED since 9.0.0 (3481a747a0, 2022-04-15, core-8-branch, later merged up). The recursive-directory change (-r, addPathToZip) has been kept through later zlib updates: main and 9.0 ship zlib 1.3.2 with it, and 8.6 has the tinydir-based version. Resolution is already Fixed, only Status is still Open.

      • Platform build
      • Tested main — make minizip, then minizip -o -r t.zip d stores d/a.txt and d/sub/b.txt
      • Cause zlib update dropped the local patch (2022)
      • Prototype none (fixed)
      • Suggested ticket comment Fixed by 3481a747a0, and the change has survived the later zlib updates (minizip -r works with zlib 1.3.2 on main/9.0). Can be closed.
      ⬜1bd796d9c22022multiple issues in Tcl_DString functionsFIXED since 9.0 (Tcl_Size DString fields, 88334c6e8c 2022-10-23 / 18d4bd49a6 2023-05-20, plus "max size for a Tcl value" checks in Tcl_DStringAppen…

      Open ticket 1bd796d9c2 · 11. Conversions from String · filed 2022-04-11

      Full verdict

      FIXED since 9.0 (Tcl_Size DString fields, 88334c6e8c 2022-10-23 / 18d4bd49a6 2023-05-20, plus "max size for a Tcl value" checks in Tcl_DStringAppend); REPRODUCED on 8.6 only

      • Platform generic
      • Tested reporter's script (1 GiB of NULs through encoding convertfrom ascii): main and 9.0 tip finish in 2–4 s with the correct length; 8.6.18 still loops (killed by a 300 s timeout).
      • Cause (8.6) int length/spaceAvl in Tcl_DString; the result of 2^31 bytes (NUL -> C0 80) cannot fit, so Tcl_ExternalToUtfDString keeps getting TCL_CONVERT_NOSPACE.
      • Prototype none (8.6 fix needs the API decision from the ticket: panic vs unsigned spaceAvl; 9.x is fine).
      • Suggested ticket comment In 9.0+ Tcl_DString uses Tcl_Size and the reporter's script works (2–4 s). 8.6.18 still loops forever on it. Suggest closing as fixed in 9.0, unless someone wants an 8.6 panic instead of the hang.
      🔧ce4483c2cb2022Audit GetFormatSpec for size_tpatchGetFormatSpec part FIXED since 9.0 (06ceffb2f3, 2023-04-24: strtoull, saturates at TCL_SIZE_MAX instead of INT_MAX); related overflow REPRODUCED (m…

      Open ticket ce4483c2cb · 12. ByteArray Object · filed 2022-03-11

      🔧 Since triage (updated 2026-11-11)

      Status now: Open

      • 🔧 2026-10-17 check-in dd5cd86e81 on binary-format-size-overflow, not merged: Fix 2788582fff and ce4483c2cb: [binary format] with huge counts or offsets panicked or returned a wrong result, because the first pass added the field sizes
      • 💬 2026-10-17 serhiy.storchaka: GetFormatSpec no longer truncates at INT_MAX since 9.0 (06ceffb2f3), but BinaryFormatCmd still adds the field sizes without checking for overflow: [binary format x4611686018427387904x4611686018427387904] returns an empty string and [binary format x99999999999999999999] panics (same as 2788582fff). Proposed fix in branch binary-format-size-overflow: report "max size for a Tcl value exceeded" instead. Tests bina…
      Full verdict

      GetFormatSpec part FIXED since 9.0 (06ceffb2f3, 2023-04-24: strtoull, saturates at TCL_SIZE_MAX instead of INT_MAX); related overflow REPRODUCED (main, 9.0, 8.6) in BinaryFormatCmd

      • Platform generic
      • Tested own main build, 9.0 tip, 8.6.18. binary format x4611686018427387904x4611686018427387904 (and the a equivalent) silently returns "" on main/9.0 (offset wraps); 8.6: binary format x2147483647x2147483647 returns "" too. binary format x99999999999999999999 and @99999999999999999999 panic ("negative length specified or max size of a Tcl value exceeded"). binary scan with huge counts behaves correctly.
      • Cause generic/tclBinary.c:BinaryFormatCmd sums the field sizes into offset with no overflow check (also (count + 7) / 8 overflows for count = TCL_SIZE_MAX), then Tcl_SetByteArrayLength panics or gets a wrapped length.
      • Prototype prototypes/ce4483c2cb.patch (119 lines, tests binary-81.1–81.4): checks every offset += and @ against BYTEARRAY_MAX_LEN and raises "max size for a Tcl value (N bytes) exceeded" (errorcode TCL MEMORY, as in binary encode); binary.test passes. Very large but representable sizes still end in the usual "unable to realloc" panic.
      • Suggested ticket comment GetFormatSpec no longer truncates at INT_MAX since 9.0 (06ceffb2f3), but BinaryFormatCmd can still overflow when adding field sizes: binary format x4611686018427387904x4611686018427387904 returns an empty string and binary format x99999999999999999999 panics. Proposed fix: overflow checks in BinaryFormatCmd raising "max size for a Tcl value exceeded", tests binary-81.*.
      ⏳24949cbd7c2021Many functions of tclPathObj are not thread safe (can segfault by changing of FS epoch)FIXED since 8.6.17 / 9.0.3 for the concrete crash (sebres' 0d48bb289c + 4aae9497e4 for 61c01e0edb, on core-8-5-branch, merged up)

      Open ticket 24949cbd7c · 36. Pathname Management · filed 2021-08-10 · Pending (resolved, waiting for confirmation)

      ⏳ Since triage (updated 2026-11-11)

      Status now: Pending

        Full verdict

        FIXED since 8.6.17 / 9.0.3 for the concrete crash (sebres' 0d48bb289c + 4aae9497e4 for 61c01e0edb, on core-8-5-branch, merged up)

        • Platform generic (reported crash is in win/tclWinFile.c)
        • Tested code check: TclpMatchInDirectory in win/tclWinFile.c holds a reference on fileNamePtr in core-8-6-17 and core-9-0-3 (not in 8.6.16 / 9.0.2), and in current 8.6, 9.0 and main. Release dates: 8.6.17 2025-08-13, 9.0.3 2025-11-11. Not stress-tested (Windows-only race).
        • Cause path object's intrep (and fileNamePtr) dropped by Tcl_FSConvertToPathType when another thread bumps the global FS epoch; fixed by temporary Tcl_IncrRefCount on the interim objects.
        • Existing work none open; the fixes are merged (main, core-9-0-branch, core-8-6-branch).
        • Prototype none
        • Suggested ticket comment none needed; Pending since 2026-10-14 with Serhiy's question unanswered. Can be closed as Fixed (in 8.6.17 and 9.0.3) if the reporter names no other concrete case.
        ✅df6a0e97f22021valgrind hates format-2.18FIXED since 9.0b1 (3bb6f6602c, 2022-07-04, "TIP #619: New TCL_COMBINE flag for Tcl_UniCharToUtf()")

        Open ticket df6a0e97f2 · 10. Objects · filed 2021-06-07

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Out of Date)

        • 💬 2026-10-21 serhiy.storchaka: No longer reproducible: since TIP #619 (3bb6f6602c, in 9.0.0) Tcl_UniCharToUtf only looks at the previous bytes when TCL_COMBINE is passed, and valgrind on format-2.18 is clean on 8.6, 9.0 and 9.1.
        Full verdict

        FIXED since 9.0b1 (3bb6f6602c, 2022-07-04, "TIP #619: New TCL_COMBINE flag for Tcl_UniCharToUtf()")

        • Platform generic
        • Tested valgrind on format-2.18/2.19 with a -DPURIFY main build: clean; 9.0 tip tcltest under valgrind: clean; 8.6 tip tcltest with the same format commands under valgrind: clean
        • Cause in 8.7a5 Tcl_UniCharToUtf (generic/tclUtf.c) always inspected buf[0]/buf[1] for a preceding high surrogate when given a low surrogate, reading uninitialised bytes when called from ExtendStringRepWithUnicode. Since TIP 619 it only does so when the caller passes TCL_COMBINE.
        • Prototype none (fixed)
        • Suggested ticket comment No longer reproducible: valgrind on format-2.18 is clean on 9.0 and 9.1 (also with -DPURIFY). Tcl_UniCharToUtf only looks at the previous bytes with TCL_COMBINE since 3bb6f6602c (TIP 619). Closing.
        ⬜e39cb3f4622021namespace is removed from other namespace paths before deletion is completeFIXED since 8.7a5 / 9.0a3 (949252950f on core-8-branch, 2021-05-19; merged to trunk c2f1c13b59, 2021-05-20); still REPRODUCED on 8.6 (not backported)

        Open ticket e39cb3f462 · 21. [namespace] · filed 2021-05-18

        Full verdict

        FIXED since 8.7a5 / 9.0a3 (949252950f on core-8-branch, 2021-05-19; merged to trunk c2f1c13b59, 2021-05-20); still REPRODUCED on 8.6 (not backported)

        • Platform generic
        • Tested reporter's script: main build, 9.0.5, 8.7b1 print t1 t2; 8.6.18 prints t1
        • Cause generic/tclNamesp.c: a dying namespace was unlinked from other namespaces' paths before its commands' delete traces ran; pooryorick's fix (branch bug-e39cb3f462631a99) keeps it in the path until teardown completes.
        • Prototype none (fixed in 9.x; 8.6 backport is a behaviour change in a corner case, not proposed)
        • Suggested ticket comment Fixed in 8.7/9.0 by 949252950f (in 9.0.0 and later); 8.6 keeps the old behaviour. Ticket is already Open/Fixed: suggest closing.
        ✅6f16de3d8d2020docs for cmdline misleading/inadequateFIXED since tcllib 1.21 (tcllib 6dd76779dd, 2021-03-02, tcllib ticket 2dce7b748a "Clarified handling of the argv variable for getoptions")

        Open ticket 6f16de3d8d · 69. Other · filed 2020-09-18

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: This is Tcllib's cmdline. It was fixed there in Tcllib 1.21 by 6dd76779dd (Tcllib ticket 2dce7b748a): the getoptions docs now say that the processed options are removed from argv.
        Full verdict

        FIXED since tcllib 1.21 (tcllib 6dd76779dd, 2021-03-02, tcllib ticket 2dce7b748a "Clarified handling of the argv variable for getoptions")

        • Platform docs (Tcllib, not Tcl)
        • Tested tcllib trunk modules/cmdline/cmdline.man now says for getoptions "If options are found the list is modified and the processed arguments are removed from the start of the list", and the example adds "argv is modified now...".
        • Cause n/a
        • Prototype none
        • Suggested ticket comment This belongs to Tcllib and was addressed there by [https://core.tcl-lang.org/tcllib/info/6dd76779dd] (tcllib ticket 2dce7b748a). The getoptions docs now say that processed options are removed from argv. Closing.
        ✅a16a1119e52020API changes introduced in TIP #445 are not documentedFIXED since 9.0.0 (4da790cbea, 2024-02-20, ticket 2f4b495427; Tcl_InitStringRep description added by 996b2264cd 2025-05-05 for 78f4421436)

        Open ticket a16a1119e5 · 10. Objects · filed 2020-09-01

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: Documented in doc/ObjectType.3 since 9.0.0 by 31e8b2c9eb (ticket 2f4b495427; the names are now Tcl_*InternalRep), and Tcl_InitStringRep got its own description in 996b2264cd (9.0.2).
        Full verdict

        FIXED since 9.0.0 (4da790cbea, 2024-02-20, ticket 2f4b495427; Tcl_InitStringRep description added by 996b2264cd 2025-05-05 for 78f4421436)

        • Platform docs
        • Tested doc/ObjectType.3 on main and 9.0 documents Tcl_FreeInternalRep, Tcl_InitStringRep, Tcl_HasStringRep, Tcl_ObjInternalRep, Tcl_StoreInternalRep, Tcl_FetchInternalRep (the TIP's IntRep names were renamed to InternalRep). The API does not exist in 8.6; 8.7b1 docs lack it (8.7 is not maintained).
        • Cause n/a
        • Prototype none (already done)
        • Suggested ticket comment Documented in doc/ObjectType.3 since 9.0.0 by 4da790cbea (ticket 2f4b495427); Tcl_InitStringRep text added in 996b2264cd (78f4421436). Duplicate, can be closed.
        ✅b5777d3d322020Crash corner-case in TIP #502 implementationFIXED since 8.7a5/9.0 (dgp-nested-execution merged into core-8-branch in a7c1c1b5bb, 2020-09-15); Pending -> can be closed

        Open ticket b5777d3d32 · 13. Index Object · filed 2020-08-25 · Pending (resolved, waiting for confirmation)

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 🔧 2020-09-15 check-in f6c7bd54e5 on dgp-nested-execution, not merged: Add test for [string replace] troubles.
        • 💬 2026-10-21 serhiy.storchaka: The crash was fixed in 9.0.0 by fca1a2a297, with test util-9.59 (passes on trunk and 9.0). 8.6 is not affected, since TIP #502 is 8.7+. The broader question of protecting other re-entry points of the bytecode engine (see branch dgp-nested-execution) can get a ticket of its own if someone wants to pursue it.
        Full verdict

        FIXED since 8.7a5/9.0 (dgp-nested-execution merged into core-8-branch in a7c1c1b5bb, 2020-09-15); Pending -> can be closed

        • Platform generic
        • Tested string index abcd 0-0x10000000000000000 (also interactive with two Enters, inside a proc, 0+..., lindex) on main, 9.0 tip, 8.7b1: returns "" without a crash; 8.6.18 gives the old "bad index" error (8.6 never had TIP 502). string-12.24/12.25 (bignum index arithmetic) in tests/string.test cover this on main. The alternative branch bug-b5777d3d32 (separate interp, tip 43eeb75b62) was never merged and is not needed.
        • Cause (was) GetEndOffsetFromObj called Tcl_ExprObj from TEBC without DECACHE_STACK_INFO, corrupting the execution stack.
        • Prototype none (fixed).
        • Suggested ticket comment The fix from dgp-nested-execution is in core-8-branch since a7c1c1b5bb (2020-09-15), so in 8.7 and 9.x; no crash on 9.0 or main. Can be closed; branch bug-b5777d3d32 can be closed as unmerged.
        ⬜255c1161102020freebsd: test failuresFIXED on main since 9.1.0 (14e87701df, 2026-09-21); REPRODUCED (9.0, 8.6) for unixFCmd-20.1. http-4.14 and the socket test: NOT-REPRODUCED (main,…

        Open ticket 255c116110 · 29. http Package · filed 2020-07-23

        Full verdict

        FIXED on main since 9.1.0 (14e87701df, 2026-09-21); REPRODUCED (9.0, 8.6) for unixFCmd-20.1. http-4.14 and the socket test: NOT-REPRODUCED (main, 9.0, 8.6).

        • Platform unix (FreeBSD)
        • Tested code reading only. The unixFCmd-20.x part (ZFS returns ENOTSUP for UF_IMMUTABLE) is handled on main by 14e87701df (gahr, 2026-09-21, "Fix the test for the readonlyAttr constraint on FreeBSD": the readonlyAttr probe uses chflags uchg/nouchg on FreeBSD); not merged to core-9-0-branch/core-8-6-branch (their unixFCmd.test has no chflags probe). The socket failure was a local /etc/hosts IPv6 issue (reporter). http-4.14 ("connect failed connection refused" expected, empty result seen) touched recently by de7efe2cc2 (2026-10-12, badurl uses a just-released port).
        • Cause tests/unixFCmd.test readonlyAttr constraint (fixed on main); http-4.14 likely environment (IPv6 localhost resolution).
        • Prototype none (needs FreeBSD; the main fix exists)
        • Suggested ticket comment The unixFCmd part is addressed on main by 14e87701df (readonlyAttr probe uses chflags on FreeBSD); it still needs backporting to 9.0/8.6. Needs a re-run of unixFCmd.test and http.test on FreeBSD/ZFS to see if http-4.14 still fails.
        • To run on the FreeBSD VM main and 9.0 make test TESTFLAGS="-file 'unixFCmd.test http.test'" on a ZFS filesystem.
        Unix VMs, run 1
        • Platform unix (FreeBSD, ZFS)
        • Verdict FIXED on main since 9.1.0 (14e87701df, 2026-09-21); REPRODUCED (9.0, 8.6) for unixFCmd-20.1. http-4.14 and the socket test: NOT-REPRODUCED (main, 9.0, 8.6).
        • Tested FreeBSD 15.1, home and /tmp on ZFS. unixFCmd.test: main 0 failed (20.x skipped: readonlyAttr off, chflags uchg gives "Operation not supported"); 9.0 and 8.6 fail unixFCmd-20.1 with could not set flags for file "foo.test": operation not supported. Applying the 14e87701df diff to 9.0 and 8.6 applies cleanly and makes unixFCmd.test pass (0 failed). http.test: main 154/0 failed, 9.0 356/0, 8.6 75/0 (http-4.14 passes). socket.test: main 191/0 failed, 8.6 175/0 (stock /etc/hosts with ::1 and 127.0.0.1 localhost).
        • Cause tests/unixFCmd.test readonlyAttr probe only queried the flag, which works on ZFS while setting UF_IMMUTABLE fails with ENOTSUP. http-4.14/socket were the reporter's IPv6 /etc/hosts setup.
        • Prototype none (fix exists; backport = cherry-pick of 14e87701df to core-9-0-branch and core-8-6-branch, verified on FreeBSD)
        • Suggested ticket comment On FreeBSD 15.1/ZFS, unixFCmd-20.1 still fails on 9.0 and 8.6 and is fixed on main by 14e87701df; cherry-picking it to 9.0 and 8.6 makes unixFCmd.test pass there. http.test and socket.test pass on all three. After the backport this can be closed.
        ✅56ebd829532020::tcl::tm::UnknownHandler allows stale module data to persist in safe interpFIXED since 8.6.11 (8927bbee05, 2020-07-13 on safe-bugfixes-8-6; merged into core-8-6-branch by a638b9a73b, 2020-08-28; 8.6.11 tagged 2020-12-30)

        Open ticket 56ebd82953 · 39. Package Manager · filed 2020-07-15

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.6.11 by 8927bbee05, merged with a638b9a73b, and present in 9.0; test safe-9.23 covers it.
        Full verdict

        FIXED since 8.6.11 (8927bbee05, 2020-07-13 on safe-bugfixes-8-6; merged into core-8-6-branch by a638b9a73b, 2020-08-28; 8.6.11 tagged 2020-12-30)

        • Platform generic
        • Tested library/tm.tcl in main, 9.0, 8.6 tips has the ![interp issafe] condition; safe-9.2* (incl. safe-9.23) pass on main build and 8.6 tip tcltest.
        • Cause library/tm.tcl ::tcl::tm::UnknownHandler kept an older package ifneeded script; in safe interps the token→dir mapping can change.
        • Prototype none (already fixed)
        • Suggested ticket comment The fix from safe-bugfixes-8-6 (8927bbee05) was merged into 8.6 (a638b9a73b) and is in 8.6.11 and later, 9.0 and main; test safe-9.23 passes. Can be closed.
        ✅a1bd37b7192020clock (free)scan of ISO 8601 timestamp with literal T behaves strangeFIXED since 8.6.11 (858e370f1a, 2020-07-09, merge of bug-a1bd37b719 into core-8-6-branch; 8.6.11 tagged 2020-12-30)

        Open ticket a1bd37b719 · - New Builtin Commands · filed 2020-05-20

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.6.11 by 858e370f1a: free-form clock scan now accepts and documents CCyy-mm-ddThh:mm:ss, so the T is no longer taken as a military time zone.
        Full verdict

        FIXED since 8.6.11 (858e370f1a, 2020-07-09, merge of bug-a1bd37b719 into core-8-6-branch; 8.6.11 tagged 2020-12-30)

        • Platform generic
        • Tested main build, 9.0.5, 8.6.18, 8.6.11: clock scan "2020-05-20T14:40:00" -gmt 1 → 14:40:00 GMT; 8.6.10 gives 21:40:00 (the bug). clock.n documents the 4th format CCyy-mm-ddThh:mm:ss. A standalone trailing " T" word is still military zone T (UTC-7), which is correct.
        • Cause generic/tclGetDate.y had no CCyy-mm-ddThh:mm:ss rule, so the T was parsed as a military time zone.
        • Prototype none (already fixed)
        • Suggested ticket comment Fixed by 858e370f1a (in 8.6.11 and later, 9.0, main): "CCyy-mm-ddThh:mm:ss" is now a documented free-form ISO format. Can be closed.
        ✅25a42264db2020TIP 445 docs missingFIXED since 9.0.0 (4da790cbea, 2024-02-20, ticket 2f4b495427)

        Open ticket 25a42264db · 10. Objects · filed 2020-03-13

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        Full verdict

        FIXED since 9.0.0 (4da790cbea, 2024-02-20, ticket 2f4b495427)

        • Platform docs
        • Tested same as a16a1119e5: all TIP 445 routines are in doc/ObjectType.3 on main and 9.0.
        • Cause n/a
        • Prototype none (already done)
        • Suggested ticket comment Documented in doc/ObjectType.3 since 9.0.0 (4da790cbea, ticket 2f4b495427). Same as a16a1119e5; can be closed.
        ✅e83a97c7fe2019windows: paths starting with tilde (~) are "partially" absoluteFIXED since 9.0 (TIP 602, cff0348cb6, 2022-09-14); REPRODUCED on 8.6 by design

        Open ticket e83a97c7fe · 36. Pathname Management · filed 2019-11-08

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: Since Tcl 9.0 (TIP 602, cff0348cb6) a path starting with ~ is an ordinary relative name, so file join and glob treat ~abc.txt and ~* as files in the directory. 8.6 keeps its documented tilde expansion. Closing as fixed in 9.0.
        Full verdict

        FIXED since 9.0 (TIP 602, cff0348cb6, 2022-09-14); REPRODUCED on 8.6 by design

        • Platform generic (Windows behaviour, tested with testsetplatform windows)
        • Tested tcltest main (30d8ccce48), 9.0 tip, 8.6 tip; unix and testsetplatform windows. 9.x: file join e:/temp ~abc.txt -> e:/temp/~abc.txt, file pathtype ~abc relative, glob ~* matches files in cwd. 8.6: -> ~abc.txt, absolute, "user "*" doesn't exist".
        • Cause tilde expansion of ~user paths (8.x), removed in Tcl 9 by TIP 602.
        • Prototype none (8.6 behaviour is the documented tilde semantics; changing it there would be incompatible)
        • Suggested ticket comment Since Tcl 9.0 (TIP 602, cff0348cb6) paths starting with ~ are ordinary relative names, so file join and glob behave as asked. 8.6 keeps tilde expansion by design; proposing to close as fixed in 9.0.
        🟢524b35a8b52019can't force regex subexpression greediness as stated by man pageFIXED on main only, unreleased (b0b515295b, 2026-10-14); REPRODUCED (9.0, 8.6)

        Open ticket 524b35a8b5 · 43. Regexp · filed 2019-05-12

        🟢 Since triage (updated 2026-11-11)

        Status now: Open

        • 🔧 2026-10-10 check-in 98ef5206be on regexp-1-1-quantifier-preference, in main: Fix 524b35a8b5: a {1,1} quantifier was always dropped as vacuous, so {1,1}? could not make a greedy atom non-greedy (and {1,1} a non-greedy one greedy). Keep
        • 💬 2026-11-10 oehhar: Jan, is there any additional action? Serhiy proposes to also merge the doc fix in fe4fca32d. Thanks for all, Harald
        Full verdict

        FIXED on main only, unreleased (b0b515295b, 2026-10-14); REPRODUCED (9.0, 8.6)

        • Platform generic
        • Tested ^(l*){1,1}?(.*)(f*)$ on llmmmfff: main llmmmfff {} llmmmfff {} (expected), 9.0/8.6 llmmmfff ll mmmfff {}. ^(.*?){1,1}(.*)$ on abc: main abc abc {}, 9.0/8.6 abc {} abc.
        • Cause generic/regcomp.c parseqatom() dropped every {1,1} as vacuous.
        • Prototype none (merged)
        • Existing work regexp-1-1-quantifier-preference (tip 98ef5206be) merged into main by b0b515295b; bug-524b35a8b5 (Jan, 2019, tip 0040d90962, based on core-8-5-branch, no tests) unmerged and superseded, can be closed
        • Suggested ticket comment Fixed on main by b0b515295b; not in 9.0/8.6. Branch bug-524b35a8b5 is superseded.
        ✅ece07271f42019Bug in Tcl_ParseArgsObjv() TCL_ARGV_GENFUNCFIXED since 8.6.15 / 9.0.0 (c28a455139, 2024-04-04, under duplicate ticket 7cb7409e05, with a test)

        Open ticket ece07271f4 · 45. Parsing and Eval · filed 2019-03-25

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Duplicate)

        • 💬 2026-10-21 serhiy.storchaka: Duplicate of 7cb7409e05. Fixed by 8cfa5d1ee9 (8.6.15) and c28a455139 (9.0.0): the handler's return value is now treated as the number of consumed arguments.
        Full verdict

        FIXED since 8.6.15 / 9.0.0 (c28a455139, 2024-04-04, under duplicate ticket 7cb7409e05, with a test)

        • Platform generic
        • Tested by code reading on main, 9.0 and 8.6: all three now do srcIndex += gf_ret; objc -= gf_ret;, as the reporter suggested.
        • Cause generic/tclIndexObj.c:Tcl_ParseArgsObjv treated the handler's return value as the remaining count instead of the consumed count.
        • Prototype none (already fixed)
        • Suggested ticket comment Fixed by c28a455139 (ticket 7cb7409e05), in 8.6.15 and 9.0.0. Closing as a duplicate.
        ⬜b9ce8b73e42018msgcat: documentation for mcloadedlocales is incorrectpatchFIXED since 9.0 (fa310f2735, 2022-05-27 "msgcat documentation: reflect code", main and core-8-branch); still wrong on core-8-6-branch

        Open ticket b9ce8b73e4 · 30. msgcat Package · filed 2018-12-23

        🧪 Prototype check (2026-10-24)

        b9ce8b73e4.patch: VERIFIED

        • Base core-8-6-branch (tip 326e1bf923); main/9.0 docs already fixed (fa310f2735)
        • Before work-a4/v4msgcat.tcl on the 8.6 tip (msgcat 1.6.2): mcloadedlocales get -> unknown subcommand "get": must be clear, or loaded; mcloadedlocales present en -> wrong # args: should be "msgcat::mcloadedlocales subcommand"; loaded -> uk_ua uk {}; doc documents get/present ?locale?.
        • Patch applied cleanly on core-8-6-branch (10 changed lines); the mcloadedlocales text now has the same content as main's msgcat.n (only line wrapping and the .VS TIP499 markers differ)
        • After groff 8 warnings (same 8 before)
        • Test none possible (docs only)
        Full verdict

        FIXED since 9.0 (fa310f2735, 2022-05-27 "msgcat documentation: reflect code", main and core-8-branch); still wrong on core-8-6-branch

        • Platform docs
        • Tested msgcat::mcloadedlocales get fails ("must be clear, or loaded") on main, 9.0.5, 8.6.18 as the code intends. main/9.0 msgcat.n documents loaded/clear; 8.6 tip msgcat.n still documents get, present and a ?locale? argument.
        • Cause doc/msgcat.n (8.6) describes subcommands that were renamed/dropped before TIP 499 was merged.
        • Prototype prototypes/b9ce8b73e4.patch (10 changed lines, docs only, against core-8-6-branch: synopsis and description as in 9.0)
        • Suggested ticket comment Fixed in 9.0 docs by fa310f2735; 8.6 msgcat.n still describes get/present. Proposed backport of the doc text to 8.6.
        ✅1df76ec7512018tcltest: software termination does not result in a failure of the testFIXED since 8.6.9 (TIP 525 a4ea2b2387, 2018-10-29; repaired by 0ef0b36b31, 2018-11-08)

        Open ticket 1df76ec751 · 34. tcltest Package · filed 2018-10-24

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: Since Tcl 8.6.9 (tcltest 2.5, TIP 525: a4ea2b2387, repaired by 0ef0b36b31) a test file killed by a signal is listed under "Test files exiting with errors" and tcltest::runAllTests returns 1, so all.tcl can exit with a failure status. Checked with main, 9.0 and 8.6. Closing.
        Full verdict

        FIXED since 8.6.9 (TIP 525 a4ea2b2387, 2018-10-29; repaired by 0ef0b36b31, 2018-11-08)

        • Platform generic
        • Tested main (own build c3bf6dd052), 9.0 tip, 8.6 tip, 8.6.8 — a child test file killed with SIGTERM under runAllTests
        • Cause in 8.6.8 (tcltest 2.4.1) runAllTests printed "Test file error: child killed ..." and listed the file under "Test files exiting with errors" but returned "" so all.tcl could not fail. Since tcltest 2.5.0 (8.6.9) it returns 1 in that case. The "Failed" counter stays 0 (no test result was received from the killed child), which is reasonable: the file is reported as a file error, and the exit status is non-zero.
        • Prototype none (already fixed)
        • Suggested ticket comment Since Tcl 8.6.9 (tcltest 2.5, TIP 525) a test file killed by a signal is listed under "Test files exiting with errors" and &#91;tcltest::runAllTests&#93; returns 1, so all.tcl can exit with a failure status. Verified with main, 9.0 and 8.6 tips. Proposing to close.
        ⬜baf43873722018file join trashes valid network pathsFIXED since 9.0.0 (TIP 402 "General Platform UNC Support", merged c1632c3763, 2022-10-04); still present in 8.6 (8.6.7–8.6 tip), by design there

        Open ticket baf4387372 · 16. Commands A-H · filed 2018-07-27

        Full verdict

        FIXED since 9.0.0 (TIP 402 "General Platform UNC Support", merged c1632c3763, 2022-10-04); still present in 8.6 (8.6.7–8.6 tip), by design there

        • Platform unix (the Windows part is a different case)
        • Tested main, 9.0 tip, 8.6 tip, 8.6.6, 8.6.8. On main/9.0 file join //a/b -> //a/b, file join //a/b //c/d -> //c/d, file join //host share file -> //host/share/file, file normalize //a/b/x -> //a/b/x; 8.6.8+ give /a/b etc. (8.6.6 kept //a/b but not //host share file). With testsetplatform windows all versions give file join //myprinter myqueue myfile -> /myprinter/myqueue/myfile, while file join //myprinter/myqueue myfile keeps //: a bare //server without a share is not a UNC root on Windows.
        • Cause 8.6 collapses a leading // on Unix since the f34cf83dd0 fix (e015bfe72f, 2017-04-28); TIP 402 made // a UNC prefix on all platforms in 9.0 (test filename-7.19 now expects //bar)
        • Prototype none (fixed by TIP 402 in 9.0; backporting a TIP behaviour change to 8.6 isn't appropriate)
        • Suggested ticket comment Fixed in Tcl 9.0 by TIP 402 (General Platform UNC Support, c1632c3763): on Unix a leading // is now kept by &#91;file join&#93; and &#91;file normalize&#93;. 8.6 keeps the old behaviour. On Windows, //server alone (without a share) is not a UNC root, so use &#91;file join //server/share ...&#93;. Proposing to close.
        ⬜83116f5e052018Tcl_Async* documentation wrongpatchFIXED since 9.0.0 / 8.7 (TIP #511, 1a6b389968, 2021-08-20) — still DOCS on 8.6

        Open ticket 83116f5e05 · 69. Other · filed 2018-06-08

        🧪 Prototype check (2026-10-24)

        83116f5e05.patch: VERIFIED

        • Base core-8-6-branch (tip 326e1bf923)
        • Before 8.6 generic/tclAsync.c Tcl_AsyncMark still does Tcl_MutexLock / Tcl_ThreadAlert / Tcl_MutexUnlock (l.165-171), and doc/Async.3 still recommends calling it from signal handlers without a caveat
        • Patch applied cleanly (7 changed lines, docs)
        • After man -l doc/Async.3 renders the new paragraph correctly
        • Test none possible (documentation only)
        Full verdict

        FIXED since 9.0.0 / 8.7 (TIP #511, 1a6b389968, 2021-08-20) — still DOCS on 8.6

        • Platform docs
        • Tested read doc/Async.3 and generic/tclAsync.c on main, 9.0, 8.7b1, 8.6
        • Cause 8.6 Async.3 says a signal handler should call Tcl_AsyncMark, but in a threaded core Tcl_AsyncMark (tclAsync.c) does Tcl_MutexLock + Tcl_ThreadAlert, which are not async-signal-safe. 8.7/9.x added Tcl_AsyncMarkFromSignal (TIP #511, proposed by the reporter) and the man page now tells signal handlers to use it.
        • Prototype prototypes/83116f5e05.patch (8.6 only, against core-8-6-branch, 7 changed lines, doc only: a paragraph saying Tcl_AsyncMark is not async-signal-safe in threaded builds; checked with groff)
        • Suggested ticket comment Resolved in 8.7/9.0 by TIP #511 (Tcl_AsyncMarkFromSignal, 1a6b389968); the 9.x Async.3 documents it. For 8.6 only a doc note is possible: proposed text in a patch (Tcl_AsyncMark is not async-signal-safe in a threaded build).
        ✅dc69e5ecf02018[clock scan] does not recognize AKST or AKDTFIXED since 8.6.15 / 9.0.0 (5e7ea80f7b, 2024-03-04, clock.tcl LegacyTimeZone; bb22bd59d0, 2024-03-11, tclGetDate.y free-form parser)

        Open ticket dc69e5ecf0 · 16. Commands A-H · filed 2018-05-21

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.6.15 and 9.0.0: AKST/AKDT were added to both scanners by 5e7ea80f7b and bb22bd59d0. Formatted Anchorage times now scan back correctly. Other historical abbreviations such as AHDT are ambiguous; use numeric offsets or -timezone for them.
        Full verdict

        FIXED since 8.6.15 / 9.0.0 (5e7ea80f7b, 2024-03-04, clock.tcl LegacyTimeZone; bb22bd59d0, 2024-03-11, tclGetDate.y free-form parser)

        • Platform generic
        • Tested main, 9.0 tip, 8.6 tip, 8.6.8–8.6.18: clock format ... -timezone :America/Anchorage output scans back exactly (free-form and -format "... %Z ...") from 8.6.15 on; 8.6.14 and older fail with "syntax error" / "time zone akdt not found"
        • Cause akst/akdt were missing from the zone tables in tclGetDate.y and clock.tcl. Older Anchorage abbreviations (e.g. AHDT, 1970) and zp4..zp6 in the free-form parser still don't scan on any version; that is the wider "all abbreviations" point in the comments, not this ticket
        • Prototype none (fixed)
        • Suggested ticket comment AKST/AKDT were added to both clock scan parsers in 5e7ea80f7b and bb22bd59d0 (in 8.6.15 and 9.0.0); formatted Anchorage times now scan back correctly on 8.6 and 9.x. Proposing to close; other historical abbreviations (AHDT etc.) are ambiguous, and numeric offsets or -timezone should be used for them.
        ✅6676bc25482018process-7.3 failsFIXED since 8.7a3 / 9.0.0 (71618fd407, 2018-05-26 "Fixed test process-7.3": test limited to the win constraint)

        Open ticket 6676bc2548 · - New Builtin Commands · filed 2018-05-11

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: Both problems are fixed for 9.0.0: 743007c3cc added the missing cleanupTests, and 71618fd407 limits process-7.3 to Windows (on Unix exit -1 is just exit status 255, not a kill).
        Full verdict

        FIXED since 8.7a3 / 9.0.0 (71618fd407, 2018-05-26 "Fixed test process-7.3": test limited to the win constraint)

        • Platform unix
        • Tested main make test -file process.test (18 total, 17 passed, 1 skipped (7.3, win)); 9.0 tip tcltest: same skip
        • Cause the test expected exit -1 to look like a killed child. On Unix it is exit status 255 (CHILDSTATUS), which is correct. Only on Windows does exit code 0xFFFFFFFF map to CHILDKILLED. The second problem (missing cleanupTests) was fixed by dgp in 743007c3cc on 2018-05-11.
        • Prototype none (already fixed)
        • Suggested ticket comment Both problems are fixed: the missing cleanupTests in 743007c3cc and the test in 71618fd407 (process-7.3 is now Windows-only, because on Unix exit -1 is just status 255). Can be closed.
        ✅e80478c91d2018configure fails on win-64 under msys bashFIXED since 8.6.11 (4454ae265c, 2019-11-27, "Don't bother paths of 64-bit SDK any more", core-8-5-branch; merged to 8.6 by ca372c7c7e the same day;…

        Open ticket e80478c91d · 53. Configuration and Build Tools · filed 2018-04-30

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: The guessing of the Microsoft Platform SDK paths for 64-bit builds was removed in 4454ae265c (8.6.11), so with the vcvars64 environment the plain link and rc are used; the test ... == line that gave "unexpected operator" is also gone from win/configure. Closing.
        Full verdict

        FIXED since 8.6.11 (4454ae265c, 2019-11-27, "Don't bother paths of 64-bit SDK any more", core-8-5-branch; merged to 8.6 by ca372c7c7e the same day; 8.6.11 tagged 2020-12-31)

        • Platform build (Windows, MSVC + MSYS configure)
        • Tested win/tcl.m4 at core-8-6-8/9/10 contains the "Microsoft Platform SDK" paths, 8.6.11+ and main do not; the test ... == ... (line 4399 "unexpected operator") is also gone from win/configure on main
        • Cause old win/tcl.m4 hard-coded the VS2008 Platform SDK paths for 64-bit MSVC builds
        • Prototype none (already fixed)
        • Suggested ticket comment The Platform SDK path guessing was removed in 4454ae265c (in 8.6.11), so with vcvars64 the plain link/rc are used. Can be closed as fixed.
        ✅e87bcf819f2018[array for] fails to detect array change made through upvarFIXED since 9.0.0 (TIP #634 a3a43c9363, 2022-10-05: TclVarHashTable got an arrayPtr back-pointer, so unsetting an element through a link now reac…

        Open ticket e87bcf819f · 07. Variables · filed 2018-04-22

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: Fixed in 9.0.0 as a side effect of TIP #634 (a3a43c9363): element hash tables now point to their array, so unsetting an element through an upvar link ends active searches and [array for] reports the change. Branch bug-e87bcf819f was not needed.
        Full verdict

        FIXED since 9.0.0 (TIP #634 a3a43c9363, 2022-10-05: TclVarHashTable got an arrayPtr back-pointer, so unsetting an element through a link now reaches the parent array and calls DeleteSearches)

        • Platform generic
        • Tested main and 9.0 (tip and installed 9.0.5): the upvar variant gives "array changed during iteration", and array startsearch + unset l invalidates the search. 8.7b1 still has the bug. 8.6.18 has no &#91;array for&#93;; there the startsearch variant keeps the search alive (nextelement returns the next key). That is harmless, because the element stays in the hash table while the link refers to it. Branch bug-e87bcf819f (2018) was never merged.
        • Cause before TIP 634, unset through a link variable did not know the containing array (generic/tclVar.c:UnsetVarStruct got arrayPtr == NULL).
        • Prototype none (fixed in 9.0; no &#91;array for&#93; in 8.6)
        • Suggested ticket comment Fixed in 9.0 as a side effect of TIP #634 (a3a43c9363), which gave element hash tables a pointer to their array. Both &#91;array for&#93; and &#91;array startsearch&#93; now see the unset made through upvar. 8.6 has no &#91;array for&#93;. Branch bug-e87bcf819f can be closed unmerged.
        ✅63e15dab992018clock scan ignores certain abbreviated optionsFIXED since 8.6.14 / 9.0.0 (72f78b6085, 2023-06-03, for af3ebc5faf); branch avl-fix-clock-scan-option-prefixes (c41ad33cb4, 1f26e5f3f1, 201…

        Open ticket 63e15dab99 · 16. Commands A-H · filed 2018-04-08

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.6.14 by 72f78b6085 (ticket af3ebc5faf): abbreviated options such as -form, -gm and -t are now honoured by clock scan; 9.0 handles them as well. Branch avl-fix-clock-scan-option-prefixes is superseded.
        Full verdict

        FIXED since 8.6.14 / 9.0.0 (72f78b6085, 2023-06-03, for af3ebc5faf); branch avl-fix-clock-scan-option-prefixes (c41ad33cb4, 1f26e5f3f1, 2018-04-08) was never merged, the same fix was done separately

        • Platform generic
        • Tested main, 9.0 tip, 8.6 tip, 8.5.19, 8.6.8–8.6.15: clock scan 01/02/2003 -form %d/%m/%Y gives 2003-01-02 (the format is ignored, free-form scan) on 8.5–8.6.13 and 2003-02-01 from 8.6.14 on; -gm/-t/-l prefixes also work on current versions
        • Cause library/clock.tcl ::tcl::clock::scan stored saw($flag) with the abbreviated flag but tested saw(-format) etc.
        • Prototype none (fixed)
        • Suggested ticket comment Fixed for 8.6.14 by 72f78b6085 (same approach as branch avl-fix-clock-scan-option-prefixes); 9.0 has the C clock implementation, which handles prefixes. Proposing to close; the avl branch can be closed as superseded.
        ⬜56ab601b7d2018Canonical sign becomes wrong after list duplication resp. botched after build of string representation of the copy.patchFIXED since 9.0b1 (fe628d36ce, 2023-07-16, for a366c6efee; on top of the new list store c3c75e63cd); still REPRODUCED on 8.6

        Open ticket 56ab601b7d · 14. List Object · filed 2018-03-14

        🧪 Prototype check (2026-10-24)

        56ab601b7d.patch: VERIFIED

        • Base core-8-6-branch (tip 326e1bf923); main/9.0 already fixed by fe628d36ce
        • Before on the unpatched 8.6 tip, the patch's listobj-bug-56ab601b7d (teststringobj "list a ; list b" -> list, testobj duplicate, regenerate the copy's string, eval the original) -> a {;} list b (the original is wrongly treated as canonical)
        • Patch applied cleanly (23 changed lines)
        • After -> b; tests listObj, list, lindex, lset, lrange, lreplace, concat, eval, basic, lsearch, cmdIL: 0 failed
        • Test existing listobj-bug-56ab601b7d (from the patch) fails before / passes after
        Full verdict

        FIXED since 9.0b1 (fe628d36ce, 2023-07-16, for a366c6efee; on top of the new list store c3c75e63cd); still REPRODUCED on 8.6

        • Platform generic
        • Tested work-a1/56ab.tcl with tcltest (teststringobj "list a ; list b" → convert to list → testobj duplicate → regenerate the copy's string → eval the original): 8.6 tip returns a {;} list b (original evaluated as a canonical list, ; ignored), main and 9.0 tip return b.
        • Cause 8.6 generic/tclListObj.c:UpdateStringOfList sets canonicalFlag = 1 on the List struct, which is shared with the original by DupListInternalRep/TclListObjCopy. 9.0 only sets LISTSTORE_CANONICAL when storePtr->refCount < 2.
        • Prototype prototypes/56ab601b7d.patch (8.6, core-8-6-branch; 23 changed lines, test listobj-bug-56ab601b7d): the same refCount < 2 guard, as in 9.0; the test fails on unpatched 8.6 tip; listObj, list, lindex, lset, lrange, lreplace, concat, eval, basic tests pass
        • Suggested ticket comment Fixed in 9.0 by fe628d36ce (shared list store is not marked canonical). 8.6 still has it (test with testobj duplicate + eval). Proposed fix for 8.6: the same guard in UpdateStringOfList, test listobj-bug-56ab601b7d.
        ✅95a8293a292018Build failure with Aqua support on OS X 10.8 and 10.9FIXED since Tk 8.6.9 (Tk e972802e92, 2018-10-30, culler, "Many updates for macOS")

        Open ticket 95a8293a29 · 52. Portability Support · filed 2018-01-07

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: Fixed in Tk 8.6.9 by Tk check-in e972802e92: Gestalt is used when building for macOS older than 10.10, as in the MacPorts patch.
        Full verdict

        FIXED since Tk 8.6.9 (Tk e972802e92, 2018-10-30, culler, "Many updates for macOS")

        • Platform macOS (Tk, filed in the Tcl tracker)
        • Tested Tk macosx/tkMacOSXXStubs.c at core-8-6-8 uses MIN_REQUIRED < 1080, at core-8-6-9 < 101000, since 8.6.10 MAC_OS_X_VERSION_MAX_ALLOWED < 101000; tkMacOSXInit.c uses the same guard (Tk 8.6 and main)
        • Cause NSOperatingSystemVersion (10.10+) used when building on 10.8/10.9
        • Prototype none (already fixed)
        • Suggested ticket comment Tk 8.6.9 uses Gestalt for SDKs older than 10.10 (Tk e972802e92), as the MacPorts patch did. Can be closed as fixed.
        ✅eb71701c362017unixNotfy-1.* tests hang in --disable-threads buildFIXED since 8.7a3 (test-side: e0ed0ab7c7, 2018-05-27, unixNotfy-1.1/1.2 constrained nonPortable)

        Open ticket eb71701c36 · 01. Notifier · filed 2017-11-15

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: Since e0ed0ab7c7 (8.7a3, 9.0.0) unixNotfy-1.1/1.2 are constrained nonPortable, and Tcl 9 has no unthreaded build. In 8.6 unthreaded builds they run and pass.
        Full verdict

        FIXED since 8.7a3 (test-side: e0ed0ab7c7, 2018-05-27, unixNotfy-1.1/1.2 constrained nonPortable)

        • Platform unix
        • Tested 8.6 tip built --disable-threads (select notifier): unixNotfy.test passes, 1.1/1.2 run and pass. 8.7b1 built --disable-threads (epoll notifier): 1.x skipped (nonPortable); forced with -constraints nonPortable they still hang (killed after 30 s). main/9.0 have no unthreaded build any more and also mark 1.x nonPortable.
        • Cause the epoll notifier (TIP 458, 8.7) does not report "would wait forever" after the last file handler is deleted in an unthreaded build, which unixNotfy-1.x expect; Jan disabled the tests rather than changing the notifier.
        • Prototype none (only relevant to unthreaded 8.7, which is not maintained; Tcl 9 is always threaded)
        • Suggested ticket comment The tests have been constrained nonPortable since e0ed0ab7c7 (8.7a3), and Tcl 9 no longer supports --disable-threads; 8.6 unthreaded builds pass them. Can be closed.
        ⬜d3cf54139d2017static package Tcltest as provided in app tcltest lacks child interp supportFIXED since 9.1.0 (TIP 755, Tcl_RegisterPostInitProc; merged 01705864c9, 2026-06-02); still REPRODUCED on 9.0 and 8.6

        Open ticket d3cf54139d · - New Builtin Commands · filed 2017-11-14

        Full verdict

        FIXED since 9.1.0 (TIP 755, Tcl_RegisterPostInitProc; merged 01705864c9, 2026-06-02); still REPRODUCED on 9.0 and 8.6

        • Platform generic (test build)
        • Tested main (own build): interp create child; child eval package require tcl::test gives 9.1.1 and testpreferstable exists in the child. 9.0 tip and 8.6 tip: package require of Tcltest/tcl::test in a child fails, while load {} Tcltest child works (the static library is registered with Tcl_StaticLibrary). package.test passes on main and 9.0, with no skipped testpreferstable tests (dgp's 2017 workaround in package.test).
        • Cause unix/tclAppInit.c only did Tcltest_Init in the main interp; TIP 755 registers TcltestStaticInit as a post-init proc, so every new interp gets the package
        • Prototype none (fixed on main; for 9.0/8.6 load {} Tcltest $child is the workaround and the test suite already uses it)
        • Suggested ticket comment In 9.1 (TIP 755, 01705864c9) the tcltest app registers its test package for every interp, so &#91;child eval package require tcl::test&#93; works. On 8.6/9.0 &#91;load {} Tcltest child&#93; is the workaround, and package.test no longer depends on it. Proposing to close as fixed in 9.1.0.
        ⬜6aea6a8e252017dict size overflowFIXED since 9.0 (Tcl_HashTable.numEntries is 64-bit Tcl_Size, earlier size_t); 8.6: CANNOT-TEST (needs >2^31 entries, well over 100 GB RAM), still …

        Open ticket 6aea6a8e25 · 15. Dict Object · filed 2017-11-08

        Full verdict

        FIXED since 9.0 (Tcl_HashTable.numEntries is 64-bit Tcl_Size, earlier size_t); 8.6: CANNOT-TEST (needs >2^31 entries, well over 100 GB RAM), still int by code reading

        • Platform generic
        • Tested code reading only (main/9.0 tcl.h, tclDictObj.c; 8.6 tcl.h int numEntries, tclDictObj.c int numElems = numEntries * 2 in UpdateStringOfDict overflows already at 2^30 entries)
        • Cause 8.6 Tcl_HashTable counters and Tcl_DictObjSize are int; changing them is an ABI change.
        • Prototype none (8.6 int sizes are ABI; 9.0 already fixed)
        • Suggested ticket comment In 9.0 the hash table and dict sizes are Tcl_Size (64-bit), so this cannot overflow there. 8.6 keeps int sizes for ABI reasons. Suggest closing as fixed in 9.0.
        ✅7f2928e3a92017Test Failure - macOSXFCmd-2.7 + macOSXFCmd-4.1 - On MacOSFIXED since 8.6.9 (test-only fixes: ad4bdf3aa8 2018-10-09 sorts the glob results in macOSXFCmd-4.1, backported in a5f5207521; 9e48ee9cb3 2018…

        Open ticket 7f2928e3a9 · 37. File System · filed 2017-11-07

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: Both tests were adjusted in October 2018, in 8.6.9 and later: ad4bdf3aa8 sorts the glob results in macOSXFCmd-4.1 (APFS returns them unsorted), and 9e48ee9cb3 marks macOSXFCmd-2.7, which only works with HFS+ resource forks, nonPortable. Closing.
        Full verdict

        FIXED since 8.6.9 (test-only fixes: ad4bdf3aa8 2018-10-09 sorts the glob results in macOSXFCmd-4.1, backported in a5f5207521; 9e48ee9cb3 2018-10-11, on 8.6 51f1b3295b 2018-10-17, marks the rsrclength test nonPortable because it only passes on HFS+, not APFS)

        • Platform macOS
        • Tested read tests/macOSXFCmd.test on main and 8.6 tip: macOSXFCmd-2.7 has constraint nonPortable, macOSXFCmd-4.1 wraps every glob in lsort. Not run (no macOS).
        • Cause APFS returns glob results unsorted and doesn't support resource-fork length changes the way HFS+ did.
        • Prototype none (already fixed)
        • Suggested ticket comment Both tests were adjusted in Oct 2018 (ad4bdf3aa8 sorts the glob results; 9e48ee9cb3 marks 2.7, the HFS+-only rsrclength test, nonPortable), in 8.6.9 and later. Proposing to close.
        ✅e7d30944752017Test Failure - io-60.1 - On MSYS builds of core-8-branchFIXED since 8.6.10 (1688d8f73e, 2019-01-08: "missed load of tcltest-library in sub-process" for io-60.1 and chan-io-60.1)

        Open ticket e7d3094475 · 37. File System · filed 2017-11-07

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: Fixed by 1688d8f73e (8.6.10, and in 9.0): the child process of io-60.1 now loads the Tcltest library before using testbytestring, the same fix as for chan-io-60.1 in ab951a7503. Closing.
        Full verdict

        FIXED since 8.6.10 (1688d8f73e, 2019-01-08: "missed load of tcltest-library in sub-process" for io-60.1 and chan-io-60.1)

        • Platform Windows (test suite)
        • Tested read tests/io.test and chanio.test on main, 9.0, 8.6: the child script now starts with catch {load $::tcltestlib Tcltest}. Not run on MSYS.
        • Cause win/Makefile.in runs the tests with tclsh plus -load of the Tcltest DLL; the child [interpreter] is plain tclsh without testbytestring.
        • Prototype none (already fixed)
        • Suggested ticket comment Fixed by 1688d8f73e (2019-01-08, 8.6.10+): the sub-process now loads the Tcltest library before using testbytestring. Proposing to close.
        ✅75ae1eca5a2017Test Failure - fCmd-9.4.a - On MSYS builds of core-8-branchFIXED (test constraint): 38eca27069 2019-01-14 restored the XP-only constraint on fCmd-9.4.a in core-8-branch. The test was later dropped altoget…

        Open ticket 75ae1eca5a · 37. File System · filed 2017-11-07

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: The XP-only constraint of fCmd-9.4.a was restored in 38eca27069 (2019), and the Windows variant of the test no longer exists in 9.0 and later (8.6 always had the win2000orXP constraint on it). Closing.
        Full verdict

        FIXED (test constraint): 38eca27069 2019-01-14 restored the XP-only constraint on fCmd-9.4.a in core-8-branch. The test was later dropped altogether: main, 9.0 and 8.7b1 only have the unix fCmd-9.4. On 8.6 it always kept win2000orXP (since 2010), so it is skipped on Win7+.

        • Platform Windows (test suite)
        • Tested read tests/fCmd.test on main, 9.0, 8.7b1, 8.6. Not run on Windows.
        • Cause on Vista+ a read-only attribute on a directory does not prevent renaming it the same way as on XP; the 8.7 copy of the test had lost its XP constraint.
        • Prototype none (already fixed)
        • Suggested ticket comment The XP constraint was restored in 38eca27069 (2019-01-14) and the Windows variant no longer exists in 8.7/9.x (8.6 always had win2000orXP). Proposing to close.
        ✅52f7abe82f2017Test Failure - cmdAH-20.7.1 + cmdAH-24.14.1 + cmdAH-28.13.1 + cmdAH-29.6.1 - On MSYS builds of core-8-branchFIXED since 8.6.9 (tests relaxed in 59a4bcc685, 2018-07-20)

        Open ticket 52f7abe82f · 16. Commands A-H · filed 2017-11-07

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: Since 59a4bcc685 (2018, in 8.6.9) these tests accept the ENOENT result. On Windows 11 file atime/mtime/size/stat/type of <dir>/CON.txt give ENOENT, and cmdAH-20.7.1, 24.14.1, 27.4.1, 28.13.1 and 29.6.1 pass with 8.6.18 and 9.1. Closing.
        Full verdict

        FIXED since 8.6.9 (tests relaxed in 59a4bcc685, 2018-07-20)

        • Platform Windows
        • Tested code reading; Linux run of win/52f7abe82f.tcl (all ok, but this is not meaningful for CON). The tests now accept "could not read ..." for atime/mtime and trap POSIX ENOENT for size/stat/type, which is exactly the MSYS result in the report
        • Cause on some Windows versions or setups, "<dir>/CON.txt" is not mapped to the console device, so stat gives ENOENT
        • Prototype none
        • Repro win/52f7abe82f.tcl checks the five commands against the current test expectations and prints a VERDICT line
        • Suggested ticket comment Since mid-2018 these tests accept the ENOENT result (see 59a4bcc685), in 8.6, 9.0 and main. Proposing to close unless they still fail on a current MSYS2 build.
        Windows VM, run 4
        • Platform Windows (test suite)
        • Verdict FIXED since 8.6.9 (tests relaxed in 59a4bcc685, 2018-07-20)
        • Tested VM win/52f7abe82f.tcl on 9.1b1, 9.0.5, 8.6.18: on Windows 11 <dir>/CON.txt is no longer the console device, every command fails with POSIX ENOENT, which the current tests accept. cmdAH.test -match of the five tests: 5/5 pass with tcltest91 and tcltest86 (MSVC builds; the code path is the same for MSYS2).
        • Cause Windows versions differ in whether CON.txt inside a directory maps to the console; Windows 11 gives ENOENT, like the MSYS report
        • Prototype none
        • Suggested ticket comment The tests accept the ENOENT result since 59a4bcc685 (2018). On Windows 11 &#91;file atime/mtime/size/stat/type&#93; of <dir>/CON.txt give ENOENT and cmdAH-20.7.1, 24.14.1, 27.4.1, 28.13.1, 29.6.1 pass with 8.6.18 and 9.1. Closing.
        ✅ab951a75032017Test Failure - chan-io-60.1 - On MSYS builds of core-8-branchFIXED since 8.6.10 (1688d8f73e, sebres, 2019-01-08; merged to core-8-branch/trunk)

        Open ticket ab951a7503 · 25. Channel System · filed 2017-11-07

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: Fixed by 1688d8f73e (8.6.10, and in 9.0): the child process of chan-io-60.1 now loads the Tcltest library before using testbytestring (the constraint was checked in the parent, but the child had no such command). Closing.
        Full verdict

        FIXED since 8.6.10 (1688d8f73e, sebres, 2019-01-08; merged to core-8-branch/trunk)

        • Platform Windows (MSYS), cause generic
        • Tested Linux main and 8.6 tip tcltest: chan-io-60.1 passes. Code reading of the test on main/9.0/8.6.
        • Cause the child script run through openpipe used testbytestring, but the child is a plain interpreter without the Tcltest package, so it failed with "invalid command name testbytestring" (dgp's question: the constraint is evaluated in the parent, not the child). 1688d8f73e makes the child load $::tcltestlib Tcltest; ad420b8994 (2025-02-24) hardened it further.
        • Prototype none (fixed)
        • Suggested ticket comment Fixed by 1688d8f73e (8.6.10+, and on 8.7/9.0): the child process of *io-60.1 now loads the Tcltest library before using testbytestring. Closing.
        ✅17902c18822017env(PATHEXT) not presentFIXED since 8.6.10 (e8d0d31e48, 2019-03-07, for ticket 9471e6e304; also in 9.0/main)

        Open ticket 17902c1882 · 08. Environment Variables · filed 2017-11-01

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: The approach suggested here, upper-casing PATH and COMSPEC in tclEnv.c, has been in Tcl since 8.6.10 (e8d0d31e48, ticket 9471e6e304), and Tcl 9 no longer has the init.tcl env traces. The reporter could not reproduce the original problem, so closing.
        Full verdict

        FIXED since 8.6.10 (e8d0d31e48, 2019-03-07, for ticket 9471e6e304; also in 9.0/main)

        • Platform Windows (ReactOS)
        • Tested by code reading. The reporter withdrew the PATHEXT report and blamed the init.tcl PATH/COMSPEC case-fixing traces interacting with tclEnv.c, and proposed upper-casing PATH/COMSPEC in generic/tclEnv.c instead. That is what e8d0d31e48 did ("Enforce PATH and COMSPEC to be all uppercase. This eliminates additional trace logic otherwise required in init.tcl"). The traces are gone from library/init.tcl on main/9.0/8.6. ReactOS itself cannot be tested here.
        • Cause library/init.tcl env traces (removed).
        • Prototype none (already done)
        • Suggested ticket comment The approach suggested here (upper-case PATH/COMSPEC in tclEnv.c and drop the init.tcl traces) went into 8.6.10 with e8d0d31e48 (ticket 9471e6e304). Closing as fixed; please reopen if ReactOS still has a problem with 8.6.18 or 9.0.
        ✅0e15cc83672017unixFCmd-20.1 and unixInit-3.1 fail in Tcl8.6.7 / FreeBSD 11.1 + gcc5unixInit-3.1 FIXED since 8.6.9 (a1207d9a2f, 2018-11-13); unixFCmd-20.1 FIXED on main since 9.1.0 (14e87701df, 2026-09-21), REPRODUCED (9.0, 8.6)

        Open ticket 0e15cc8367 · 34. tcltest Package · filed 2017-08-17

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        Full verdict

        unixInit-3.1 FIXED since 8.6.9 (a1207d9a2f, 2018-11-13); unixFCmd-20.1 FIXED on main since 9.1.0 (14e87701df, 2026-09-21), REPRODUCED (9.0, 8.6)

        • Platform unix (FreeBSD)
        • Tested read tests/unixFCmd.test and unixInit.test on main, 9.1.0 tag, 9.0, 8.6.
        • Cause test-only: readonlyAttr constraint was set because querying/unsetting the flag works on ZFS while setting it fails; unixInit-3.1 expected iso8859-1 for LANG=C while FreeBSD reports utf-8.
        • Prototype none (fix exists on main; a backport is a cherry-pick of 14e87701df to core-9-0-branch / core-8-6-branch)
        • Suggested ticket comment unixInit-3.1 was relaxed in a1207d9a2f (8.6.9). unixFCmd-20.1 is fixed on main by 14e87701df (9.1.0); 9.0 and 8.6 still need the backport. To confirm, run unixFCmd.test on the FreeBSD VM with core-8-6-branch / core-9-0-branch.
        Unix VMs, run 1
        • Platform unix (FreeBSD)
        • Verdict unixInit-3.1 FIXED since 8.6.9 (a1207d9a2f, 2018-11-13); unixFCmd-20.1 FIXED on main since 9.1.0 (14e87701df, 2026-09-21), REPRODUCED (9.0, 8.6)
        • Tested FreeBSD 15.1/ZFS, clang 19: unixInit.test passes on main, 9.0, 8.6. unixFCmd-20.1 fails on 9.0 and 8.6 (ENOTSUP from chflags), skipped on main; with 14e87701df applied to 9.0/8.6 unixFCmd.test passes.
        • Cause test-only, see 255c116110.
        • Prototype none (cherry-pick 14e87701df to core-9-0-branch / core-8-6-branch; verified)
        • Suggested ticket comment unixInit-3.1 was fixed in a1207d9a2f (8.6.9). unixFCmd-20.1 still fails on 9.0/8.6 on FreeBSD/ZFS and is fixed on main by 14e87701df; the cherry-pick works on both branches (tested on FreeBSD 15.1). Duplicate of 255c116110 for the unixFCmd part.
        ✅fed5069f8a2017NetBSD: support LDFLAGSFIXED in tclconfig (f636db8084, 2020-02-11, for ticket f20bee173b); TEA packages bundled with 8.6.11+ / 9.x get it with their refreshed tclconfig

        Open ticket fed5069f8a · 81. Bundled Packages · filed 2017-07-16

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: Fixed in TEA/tclconfig by f636db8084 (2020, ticket f20bee173b): MAKE_SHARED_LIB now passes ${LDFLAGS} on all Unix platforms, NetBSD included, so bundled packages built with a current tclconfig honour LDFLAGS.
        Full verdict

        FIXED in tclconfig (f636db8084, 2020-02-11, for ticket f20bee173b); TEA packages bundled with 8.6.11+ / 9.x get it with their refreshed tclconfig

        • Platform build (TEA/tclconfig, NetBSD and other non-Linux Unixes)
        • Tested read tclconfig trunk tcl.m4 (bed05771cd) and its history; no NetBSD build done.
        • Cause before f636db8084, TEA's Unix MAKE_SHARED_LIB was ${SHLIB_LD} -o $@ ... and LDFLAGS only got in through SHLIB_LD, which only Linux (and Darwin) set with ${LDFLAGS_DEFAULT}. Now MAKE_SHARED_LIB is ${SHLIB_LD} ${LDFLAGS} ${LDFLAGS_DEFAULT} -o $@ ... for every non-Windows platform, NetBSD included.
        • Prototype none (already fixed)
        • Suggested ticket comment Since tclconfig f636db8084 (2020, ticket f20bee173b) MAKE_SHARED_LIB passes ${LDFLAGS} on every Unix platform, NetBSD included, so bundled packages built with a current tclconfig honour LDFLAGS. Can be closed; reopen if a current package still drops them on NetBSD.
        ⬜46f801ea5a2017Tcl bug: unexpected stack trace in "unknown"FIXED since 8.6.7 (432f9304f2, 2017-06-26), main symptom; ticket intentionally left open by dgp for the remaining question of how &#91;tailcall&#…

        Open ticket 46f801ea5a · 18. Commands M-Z · filed 2017-06-06

        Full verdict

        FIXED since 8.6.7 (432f9304f2, 2017-06-26), main symptom; ticket intentionally left open by dgp for the remaining question of how &#91;tailcall&#93; should add to -errorinfo

        • Platform generic
        • Tested dgp's demo (work-a2/46f801ea5a.tcl: auto_index entry proc demo {} {tailcall error foo}): 8.6.6 gives "Tcl bug: unexpected stack trace in unknown"; 8.6.7, 8.6.8, 8.6.18, 9.0.5, main all give foo with errorInfo foo / while executing "error foo" / invoked from within "demo" (no "(procedure demo ...)" line, since tailcall removed the frame).
        • Cause library/init.tcl:unknown checked that the autoloaded command's -errorinfo ended with the expected trace and errored otherwise; tailcall breaks that assumption. Since 432f9304f2 it falls back to a less exact trace. Regression test init-4.N is in tests/init.test, labelled with the misspelled id "[Bug 46f801ed5a]".
        • Prototype none (fixed; open part is a design question)
        • Suggested ticket comment The BADTRACE error is gone since 8.6.7 (432f9304f2); the demo now gives "foo" on 8.6/9.0/9.1. What is left is how &#91;tailcall&#93; should append to -errorinfo. Close this, or keep it open only for that question? (The test in init.test cites the id as 46f801ed5a.)
        ✅fbc88fb6452017[variable] doesn't follow variable resolution rulesFIXED since 9.0.0 (TIP 278, 580ef6f047, 2017-11-03 on trunk; 9.0.0 released 2024-09-25)

        Open ticket fbc88fb645 · 07. Variables · filed 2017-05-30

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: Fixed in Tcl 9.0 by TIP #278 (580ef6f047): relative qualified variable names no longer fall back to the global namespace, so $a::a and variable a::a agree. 8.6 keeps the old rules for compatibility.
        Full verdict

        FIXED since 9.0.0 (TIP 278, 580ef6f047, 2017-11-03 on trunk; 9.0.0 released 2024-09-25)

        • Platform generic
        • Tested main, 9.0 tip: in a proc in ::foo, $a::a and variable a::a now both refuse to fall back to ::a::a. 8.6 tip, 8.5.19, 8.4.20: $a::a falls back to ::a::a while variable a::a errors, as reported. (upvar a::a resolves in the global caller frame, so it is not a counter-example.)
        • Cause before TIP 278, relative qualified names fell back to the global namespace in normal variable lookup, but [variable] always resolves relative to the current namespace. TIP 278 removed the fallback, so the two now agree.
        • Prototype none (8.6 keeps the old documented resolution rules; changing them there would be incompatible)
        • Suggested ticket comment Since Tcl 9.0 (TIP 278, 580ef6f047), relative qualified names no longer fall back to the global namespace, so $a::a and variable a::a agree. 8.6 keeps the old behaviour on purpose. Closing as fixed in 9.0.
        ⬜39f6304c2e2016Tcl_LinkVar is not tolerant to minus, plus, dotpatchFIXED since 8.6.7 (2017-08-09; also on core-8-5-branch after 8.5.19, unreleased) (jan.nijtmans, branch bug-39f6304c2e merged into core-8-5-branch, …

        Open ticket 39f6304c2e · 09. Linked C Variables · filed 2016-12-17

        🧪 Prototype check (2026-10-24)

        39f6304c2e.patch: VERIFIED

        • Base main (tip d6dd6d0f71)
        • Before work-a4/v4link.tcl (testlink): set int -0x -> can't set "int": variable must have integer value; same for int +0b, wide -0o; set real -. / +. / -0x -> variable must have real value
        • Patch applied cleanly (35 changed lines)
        • After all six assignments accepted (0 -0x, 0 -., ...); link.test 78 passed / 0 failed; var.test 217 passed, 4 skipped, 0 failed
        • Test existing link-2.11 fails before (link.test 77 passed, 1 failed) / passes after
        Full verdict

        FIXED since 8.6.7 (2017-08-09; also on core-8-5-branch after 8.5.19, unreleased) (jan.nijtmans, branch bug-39f6304c2e merged into core-8-5-branch, core-8-6-branch and trunk 2017-01; the reporter confirmed on 2017-01-25). A remaining gap (avl42, 2017-02-15) is REPRODUCED (main, 9.0, 8.6).

        • Platform generic
        • Tested tcltest testlink on main, 9.0 tip, 8.6 tip. -, +, empty, 0x, . and 1e- are accepted everywhere. Still rejected: -0x, +0x, -0b (int and double) and -., +. (double).
        • Cause generic/tclLink.c:GetInvalidIntFromObj only accepts an unsigned 2-char radix prefix, and SetInvalidRealFromAny only accepts a bare ., so a sign before them is refused
        • Prototype prototypes/39f6304c2e.patch (35 changed lines, test link-2.11). It skips an optional sign before the radix prefix and accepts [+-]. for doubles. link.test passes.
        • Suggested ticket comment The main problem was fixed in 2017 in all branches. The prefixes avl42 mentioned (-0x, +0b, ...) and -. / +. for doubles are still rejected on 9.1. Proposed fix with test link-2.11 (prototype). After that, the ticket can be closed.
        ✅e8004edfe52016tip 458 test failuresFIXED since 8.7a1/9.0 (TIP 458 merged to trunk as 858569fb04, 2017-05-18, after the branch fixes the reporter mentions)

        Open ticket e8004edfe5 · - New Builtin Commands · filed 2016-12-12

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: TIP 458 was merged in 858569fb04 (2017) after these branch problems were fixed. With the epoll notifier on current main, aaa_exit.test, basic.test, chan.test and event.test all pass. Closing.
        Full verdict

        FIXED since 8.7a1/9.0 (TIP 458 merged to trunk as 858569fb04, 2017-05-18, after the branch fixes the reporter mentions)

        • Platform unix (Linux epoll notifier)
        • Tested main build with NOTIFIER_EPOLL=1: aaa_exit.test, basic.test, chan.test, event.test all pass (no hang, no epoll_ctl error).
        • Cause early tip-458 branch bugs in the epoll notifier, fixed on the branch before the merge.
        • Prototype none (fixed)
        • Suggested ticket comment TIP 458 was merged in 858569fb04 (2017), and with the epoll notifier on current main aaa_exit, basic, chan and event tests all pass. Closing.
        ✅f4f44174e42016protect notifier against signalsFIXED since 9.0 for extensions that use the new API (TIP 511 Tcl_AsyncMarkFromSignal, 1a6b389968, 2021-08-20, in 8.7/9.0); REPRODUCED (main, 9.0,…

        Open ticket f4f44174e4 · 01. Notifier · filed 2016-12-05

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: Addressed in 9.0 by TIP 511 (1a6b389968): signal handlers should call Tcl_AsyncMarkFromSignal(), which is async-signal safe, instead of Tcl_AsyncMark(). Extensions such as TclX need to switch to it; nothing more is planned for 8.6.
        • 💬 2026-10-21 serhiy.storchaka: What remains is in TclX: Tcl_AsyncMark() is still not async-signal safe (it locks a mutex), and TclX still calls it from its signal handler (generic/tclXsignal.c), so the hang in the script above can still happen. With Tcl 9.0 TclX should call Tcl_AsyncMarkFromSignal(asyncHandler, signalNum) instead, which takes no mutex. 8.6 has no such API; there TclX would need its own self-pipe or sigwait() thread to call Tcl_…
        Full verdict

        FIXED since 9.0 for extensions that use the new API (TIP 511 Tcl_AsyncMarkFromSignal, 1a6b389968, 2021-08-20, in 8.7/9.0); REPRODUCED (main, 9.0, 8.6) with plain Tcl_AsyncMark called from a signal handler, which is what the ticket (TclX) does — that usage stays unsafe by design and 8.6 has no alternative.

        • Platform unix
        • Tested work-a4/sig: small extension installing a SIGHUP handler + a child shell sending SIGHUP in a tight loop while the event loop runs pipe/timer events. Tcl_AsyncMark: deadlock (timeout) 5/5 on main and 5/5 on 8.6.18. Tcl_AsyncMarkFromSignal: 0 hangs in 40×3 s runs on main and 40×3 s on 9.0.5 (~4M signals per 8 s run). TclX itself not available here.
        • Cause generic/tclAsync.c:Tcl_AsyncMark takes asyncMutex (and Tcl_ThreadAlert takes the notifier mutex); a signal arriving in the thread that holds it (e.g. inside Tcl_AsyncInvoke) self-deadlocks. TIP 511 routes signals through the notifier thread with async-signal-safe calls. Branches bug-f4f44174e and aspect-async-pipe (2016/2017) were never merged and are superseded.
        • Prototype none (fixed by TIP 511 in 9.0; 8.6 would need new API)
        • Suggested ticket comment Fixed in 8.7/9.0 by TIP 511: signal handlers must call Tcl_AsyncMarkFromSignal() instead of Tcl_AsyncMark() (verified: no hang in 40 stress runs on 9.0/main, while Tcl_AsyncMark still deadlocks). TclX needs to switch to the new call; nothing more planned for 8.6. Branches bug-f4f44174e/aspect-async-pipe are superseded.
        ⬜1907b57bc42016RFE: Support for system libtommath?FIXED since 9.0 (TIP 538 "Externalize libtommath", d796a3955d 2019-12-13 on core-8-branch; unix configure --with-system-libtommath, default yes w…

        Open ticket 1907b57bc4 · 56. LibTomMath · filed 2016-07-25

        Full verdict

        FIXED since 9.0 (TIP 538 "Externalize libtommath", d796a3955d 2019-12-13 on core-8-branch; unix configure --with-system-libtommath, default yes when tommath.h is found; also in 8.7). 8.6 keeps the bundled copy.

        • Platform build
        • Tested main/9.0/8.7 unix/configure.ac have --with-system-libtommath; 8.6 configure.in has none
        • Cause n/a
        • Prototype none
        • Suggested ticket comment Implemented by TIP 538 in 9.0: configure --with-system-libtommath (used by default when an installed libtommath is found). 8.6 will keep the bundled copy; suggest closing.
        🟢7e4d5a0a2b2016regexp behaviour differs based on passing option as literal v/s variableFIXED on main only, unreleased (6e93204003, 2026-10-14); REPRODUCED (9.0, 8.6)

        Open ticket 7e4d5a0a2b · 47. Bytecode Compiler · filed 2016-05-20

        🟢 Since triage (updated 2026-11-11)

        Status now: Open

        • 🔧 2026-10-11 check-in 0b36adaf16 on regexp-compile-runtime-switch, in main: Fix [1902890]: compiled [regexp] did not treat an RE value starting with "-" as a switch, unlike the command. When the RE is not known at compile time and there
        Full verdict

        FIXED on main only, unreleased (6e93204003, 2026-10-14); REPRODUCED (9.0, 8.6)

        • Platform generic
        • Tested proc c3 {} {set switch -nocase; regexp $switch "Looking for -nocase"}: main "wrong # args" (same as $cmd $switch ...), 9.0/8.6 return 1.
        • Cause same as 1902890fff.
        • Prototype none (merged)
        • Existing work regexp-compile-runtime-switch merged into main by 6e93204003; not in 9.0/8.6
        • Suggested ticket comment Fixed on main by 6e93204003 (see 1902890fff); not in 9.0/8.6.
        ✅7393be0be62016Tcl Thread library doesn't work in Debug build on Windows_x64FIXED since 8.6.6 (point 2: 5ed3db2343 on core-8-6-branch, 5aad54f649 on trunk, 2016-06-14: _MCW_PC is not passed to _controlfp on _WIN64). Point 1…

        Open ticket 7393be0be6 · 80. Thread Package · filed 2016-04-18

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: Point 2 was fixed in 8.6.6 by 5ed3db2343: _MCW_PC is no longer passed to _controlfp() on Win64. Point 1 is also resolved: the Thread nmake build writes pkgIndex.tcl with the DLL name of the build being made, including the debug suffix.
        Full verdict

        FIXED since 8.6.6 (point 2: 5ed3db2343 on core-8-6-branch, 5aad54f649 on trunk, 2016-06-14: _MCW_PC is not passed to _controlfp on _WIN64). Point 1 is also resolved: the Thread nmake build now generates pkgIndex.tcl from pkgIndex.tcl.in with @PKG_LIB_FILE@ = $(PRJLIBNAME), which carries the build's suffix (g for debug), so a debug build's index loads its own debug DLL.

        • Platform Windows
        • Tested source inspection (win/tclWinThrd.c main/8.6; thread win/rules.vc default-pkgindex-tea); no Windows run
        • Cause n/a
        • Prototype none
        • Suggested ticket comment Both points are resolved. _MCW_PC was fixed in &#91;5ed3db2343&#93; (8.6.6), and the Thread nmake build now writes pkgIndex.tcl with the DLL name of the build being made, debug suffix included. Can be closed.
        ✅9b470294672016testing existence of env(some_thing) destroys tracesFIXED (jan.nijtmans, 5e24c6a9db on bug-9b47029467631832, merged to trunk 2016-02-22; test env-7.3). The fix holds on main, 9.0, 8.6 and 8.5.

        Open ticket 9b47029467 · 08. Environment Variables · filed 2016-02-15 · Pending (resolved, waiting for confirmation)

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.6.5 by d644df2cad, with test env-7.3 (passes on trunk, 9.0 and 8.6). No review concerns came up since, so closing.
        Full verdict

        FIXED (jan.nijtmans, 5e24c6a9db on bug-9b47029467631832, merged to trunk 2016-02-22; test env-7.3). The fix holds on main, 9.0, 8.6 and 8.5.

        • Platform generic
        • Tested the reporter's script on main, 9.0 tip, 8.6 tip, 8.5.19: the write trace fires after info exists env(...). env-7.3 is in tests/env.test on main.
        • Cause n/a (fixed)
        • Prototype none
        • Suggested ticket comment Still fixed on 8.5 through 9.1 (env-7.3), and no better solution has come up since 2016. Can be closed.
        ✅05382a53172015extra pipes opened with execFIXED since 8.5.13 (662a69f571, 2012-09-07, backport of 5a5e87fae7 from 2008 that was in 8.6.0) "Fix missing CLOEXEC on internal pipes [2417695]"

        Open ticket 05382a5317 · 16. Commands A-H · filed 2015-12-09

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.5.13 by 662a69f571 (backport of 5a5e87fae7, in 8.6.0): the notifier's internal pipes are now close-on-exec, so exec'd children no longer inherit them.
        Full verdict

        FIXED since 8.5.13 (662a69f571, 2012-09-07, backport of 5a5e87fae7 from 2008 that was in 8.6.0) "Fix missing CLOEXEC on internal pipes [2417695]"

        • Platform unix
        • Tested work-a2/05382a5317.tcl (chain of N exec ... & children, each entering the event loop first; the leaf counts its pipe fds) with N=0/1/5: the count stays constant on main (0, epoll/eventfd), 9.0.5 (0), 8.6.18 and 8.6.0 (2, its own notifier pipe), 8.5.19 and 8.5.9 (0). The local 8.5 builds are non-threaded and have no notifier pipe, so the 8.5.9 leak itself could not be shown; the verdict rests on the commit.
        • Cause the threaded notifier's trigger pipe lacked FD_CLOEXEC before 8.5.13, so every exec'd child inherited its parent's notifier pipes.
        • Prototype none (fixed)
        • Suggested ticket comment The notifier pipes got FD_CLOEXEC in 5a5e87fae7 (8.6.0) and 662a69f571 (8.5.13). A chain of exec'd children keeps a constant pipe count on 8.6/9.0/9.1. Closing as fixed.
        ✅7d47ba7d942015Stack overruns in regexp codeFIXED on main only, unreleased (4a966b9df4, 2026-10-19); REPRODUCED (9.0, 8.6: SIGSEGV)

        Open ticket 7d47ba7d94 · 43. Regexp · filed 2015-10-17 · Pending (resolved, waiting for confirmation)

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 🔧 2026-10-19 check-in 4a966b9df4 on trunk, in main: Fix 7d47ba7d94: the regexp engine could overflow the C stack on deeply nested REs
        • 🔧 2026-10-28 check-in 0b7238035e on musl-stack-limit, in main: Fix 7d47ba7d94 on musl: TclpGetCStackLimit() returned NULL on Linux without glibc, so the regexp engine assumed a 1 MB stack and stack-4.1 segfaulted. Use pth
        • 🔧 2026-10-28 check-in e67bd0fb30 on trunk, in main: 7d47ba7d94 follow-up: on musl: TclpGetCStackLimit() returned NULL
        • 🔧 2026-10-16 check-in 3de02ad5e8 on regexp-stack-depth, in main: Fix 7d47ba7d94: the regexp engine could overflow the C stack on deeply nested REs. Rebase to trunk
        • 💬 2026-10-19 jan.nijtmans: Now merged to 4a966b9df4 In addition, tested on MacOS. CI build is clean. I'm reluctant to merge further to core-9-0-branch for now, since Tcl 9.0.5 is coming soon.
        • 💬 2026-10-28 serhiy.storchaka: On Alpine Linux (musl), stack-4.1 segfaults: TclpGetCStackLimit() returns NULL there, so the 1 MB default is used. Fixed on branch musl-stack-limit.
        • 💬 2026-10-28 jan.nijtmans: > On Alpine Linux (musl), stack-4.1 segfaults Thanks! This part now merged to trunk too.
        Full verdict

        FIXED on main only, unreleased (4a966b9df4, 2026-10-19); REPRODUCED (9.0, 8.6: SIGSEGV)

        • Platform generic
        • Tested (^$)+ returns 0 everywhere. 40000 nested parens: main "regular expression is too complex", 9.0/8.6 segfault. 600 nested (?=: main returns 1, 9.0/8.6 segfault.
        • Cause unbounded recursion in parse()/checkLAConstraint() etc. with STACK_TOO_DEEP defined as 0; fixed by real stack-depth checks (TclpGetCStackLimit) and per-execution LA DFA cache.
        • Prototype none (merged)
        • Existing work regexp-stack-depth (tip 3de02ad5e8) merged into main by 4a966b9df4; not in core-9-0-branch (Jan: not before 9.0.5) or core-8-6-branch
        • Pending resolution holds on main. Can be closed if main-only is accepted; keep Pending if a 9.0 backport after 9.0.5 is planned.
        • Suggested ticket comment Verified on main: deep nesting now fails with "regular expression is too complex"; 9.0/8.6 still crash. Close, or keep Pending until the post-9.0.5 backport.
        ✅8e53aea5b92015Safe interpreter paths differ from non-safe (test safe-14.1 fails)FIXED since 8.6.11 (081a8b9b50, 2020-07-13, branch safe-bugfixes-8-6, merged before 8.6.11 of 2020-12-30)

        Open ticket 8e53aea5b9 · 20. [interp] · filed 2015-07-15

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.6.11 by 081a8b9b50 (safe-bugfixes-8-6): subdirectories of tm roots now go only into the access path, not into the child's tm path. With an extra tm directory that has subdirectories, the child's and the parent's tm lists are the same with 8.6.11, 8.6.18, 9.0.5 and main (they differ with 8.6.10). Closing.
        Full verdict

        FIXED since 8.6.11 (081a8b9b50, 2020-07-13, branch safe-bugfixes-8-6, merged before 8.6.11 of 2020-12-30)

        • Platform generic
        • Tested repro adding a tm path with a subdirectory and comparing the child's tm list with the parent's (the safe-14.1 check): 8.6.4 and 8.6.10 DIFF (subdirs like .../8.4/platform added to the child tm path); 8.6.11, 8.6.12, 8.6.18, 9.0.5, main SAME.
        • Cause safe.tcl added the recursively found subdirectories ([2854929]) to the child tm path too; kjnash's fix adds them only to the access path (firstpass).
        • Prototype none (fixed)
        • Suggested ticket comment Fixed in 8.6.11 by &#91;081a8b9b50&#93; (safe-bugfixes-8-6): subdirectories of tm roots go only into the access path, not the child's tm path. safe-14.1 passes with an extra tm dir that has subdirectories; can be closed.
        ⬜f63b3ab13d2015Possible bug in clock scanFIXED since 9.0.0 (TIP 688 clock rewrite 7137ea11e9, 2024-04-01, plus TIP 690 "-valid 1 by default" 49e8257d9a/f330be6e51, 2024-03/04); still…

        Open ticket f63b3ab13d · 16. Commands A-H · filed 2015-02-06

        Full verdict

        FIXED since 9.0.0 (TIP 688 clock rewrite 7137ea11e9, 2024-04-01, plus TIP 690 "-valid 1 by default" 49e8257d9a/f330be6e51, 2024-03/04); still present in 8.6 and 8.7b1

        • Platform generic
        • Tested main (own build), 9.0.5: clock scan "2015 02 06 10 00 00 2" -format "%Y %m %d %H %M %S %u" raises "unable to convert input string: invalid day of week", and 2015-02-31 gives "invalid day". 8.6.18 and 8.7b1 accept both silently (8.6 has no -valid option).
        • Cause 8.6's Tcl-coded clock scan never checks scanned fields against each other. 9.0 checks them (tclClock.c, "invalid day of week"), with later fixes for Sunday (5fe15e460a, 2024-07-16).
        • Prototype none (fixed in 9.0; backporting TIP 688/690 validation to 8.6 is a behaviour change)
        • Suggested ticket comment Since 9.0.0 (TIP 688/690) clock scan validates by default and rejects this input with "invalid day of week". 8.6 keeps the old non-validating behaviour. Close as fixed in 9.0 (same as 3475995fff).
        ⏳9d75181ee72014command exec seems to start processes as suspended and then sleeps forever (win2k8)FIXED since 8.6.7 (likely; ad65e576ee, 2017-05-01) — REPRODUCED once on the pre-fix baseline 8.6.6, never on 8.6 tip, 9.0, 9.1 (8 runs each)

        Open ticket 9d75181ee7 · 24. Channel Commands · filed 2014-12-09 · Pending (resolved, waiting for confirmation)

        ⏳ Since triage (updated 2026-11-11)

        Status now: Pending (Remind)

          Full verdict

          FIXED since 8.6.7 (likely; ad65e576ee, 2017-05-01) — REPRODUCED once on the pre-fix baseline 8.6.6, never on 8.6 tip, 9.0, 9.1 (8 runs each)

          • Platform Windows
          • Tested Linux not affected (Windows pipe threads). Stress repro win/9d75181ee7.tcl (N threads exec'ing a child tclsh in a loop, watchdog reports a thread stuck > 30 s; single-threaded fallback without Thread). Checked on Linux 9.0.5 + Thread 3.0.7 (1200 execs, no hang) only to validate the script.
          • Cause (per sebres) pipe reader/writer threads of win/tclWinPipe.c blocked thread start-up; reworked by ad65e576ee (TclPipeThreadInfo workers), present in 8.6.7+, 9.0, main.
          • Existing work fix-1997007 merged into core-8-6-branch (ad65e576ee, 2017-05-01, closed); core-8-5-branch-fix-1997007 never merged (8.5 EOL).
          • Prototype none (fixed / not reproducible here)
          • Suggested ticket comment none new; Serhiy asked sebres on 2026-10-14 whether it happens with 8.6.7+. If win/9d75181ee7.tcl with 24+ threads for a few minutes shows no hang on Win11 and there is no answer, close as Fixed (8.6.7).
          Windows VM, run 6
          • Platform Windows
          • Verdict FIXED since 8.6.7 (likely; ad65e576ee, 2017-05-01) — REPRODUCED once on the pre-fix baseline 8.6.6, never on 8.6 tip, 9.0, 9.1 (8 runs each)
          • Tested win/9d75181ee7.tcl, 120 s x 10 threads per run (the "24" thread argument was dropped by my run wrapper, so all versions ran with the default 10 threads; same for every version), 8 runs each, interleaved, ~25 000 execs per run: 8.6.6 (%USERPROFILE%\rel\8.6.6, Thread 2.8.13): 1 of 8 runs BUG PRESENT (run 3: threads 0,1,5,7,9 stuck at ~1440 execs; one exec'd child tclsh86t stayed alive 13+ min with 0 CPU time, two threads in Wait/UserRequest, i.e. it never ran, matching the report; the parent finished but the child kept the pipe open; I killed that child by PID). 8.6 tip (8.6.18), 9.0.5, 9.1.1: 0 of 8 runs hung, no exec errors.
          • Cause (per sebres) pipe reader/writer thread start-up in win/tclWinPipe.c, reworked by ad65e576ee (8bd13f07bd, branch fix-1997007), in 8.6.7+, 9.0 and main.
          • Prototype none (fixed)
          • Suggested ticket comment Reproduced on Windows 11 with 8.6.6 (1 of 8 two-minute runs of a 10-thread exec stress, a child stuck before running any code), not with 8.6 tip, 9.0 or 9.1 (0 of 8 each). Fixed in 8.6.7 by ad65e576ee; can be closed. (1/8 vs 0/8 is weak evidence by itself; it agrees with sebres' analysis.)
          ⬜9969cf8ca62014memleak in execute-10.3FIXED since 8.6.2 (f076f44f1d, 2014-07-17)

          Open ticket 9969cf8ca6 · 47. Bytecode Compiler · filed 2014-07-11

          Full verdict

          FIXED since 8.6.2 (f076f44f1d, 2014-07-17)

          • Platform generic
          • Tested main with a -DPURIFY build under valgrind: execute-10.3 and all execute-10.* show 0 bytes definitely/indirectly lost.
          • Cause TEBC bailed out on ->rewind before tearing down the frame's trace data; dgp moved the bailout after that housekeeping. The ticket was left open only until a follow-up ticket on execution traces vs. [yield] was filed (none found by title search).
          • Prototype none (fixed)
          • Suggested ticket comment The leak was plugged by &#91;f076f44f1d&#93; (in 8.6.2); valgrind on a PURIFY build of 9.1 shows no leak in execute-10.*. The remaining questions about execution traces and yield are a separate issue; this ticket can be closed.
          ⬜1211aceef22014temp file refcount breakageFIXED since 8.6.2 (44648f6210, 2014-07-11, dgp) — resolution is already "Fixed"; left Open on purpose ("I welcome better solutions, but for now I…

          Open ticket 1211aceef2 · 27. Channel Types · filed 2014-07-11

          Full verdict

          FIXED since 8.6.2 (44648f6210, 2014-07-11, dgp) — resolution is already "Fixed"; left Open on purpose ("I welcome better solutions, but for now I just want to plug the leak")

          • Platform unix
          • Tested code reading on main, 9.0 and 8.6: the leak fix is present.
          • Cause (historical) 2f3da59db1 returned the refcount-1 nameObj.
          • Prototype none (fixed; the duplicate costs one small object per temp-library load — not worth changing)
          • Suggested ticket comment The leak fix 44648f6210 is in all maintained branches; proposing to close.
          ✅3608957fff2013Events posted by Tcl_QueueEvent linger in queueFIXED since 9.0 in practice (TIP 458 epoll/kqueue notifier waits in the calling thread, so the signal interrupts epoll_wait); REPRODUCED on 8.6 (th…

          Open ticket 3608957fff · 01. Notifier · filed 2013-03-24

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Out of Date)

          • 💬 2026-10-21 serhiy.storchaka: Tcl_QueueEvent() does not wake the notifier and must not be called from a signal handler. Since 9.0 the right way is Tcl_AsyncMarkFromSignal() (TIP 511, 1a6b389968) with the event queued from the async handler, and with the epoll/kqueue notifier the event is handled at once. Closing.
          Full verdict

          FIXED since 9.0 in practice (TIP 458 epoll/kqueue notifier waits in the calling thread, so the signal interrupts epoll_wait); REPRODUCED on 8.6 (threaded select notifier). NOT-A-BUG by contract: Tcl_QueueEvent does not alert the notifier, and calling it from a signal handler is not async-signal safe — TIP 511 Tcl_AsyncMarkFromSignal (8.7/9.0) is the supported way.

          • Platform unix
          • Tested sighup.zip attachment is truncated in the repository (1650 of ~231000 bytes); own extension work-a4/sig/qev.c (SIGHUP handler → Tcl_QueueEvent, [after 3000] "Boo!"), SIGHUP at 500 ms, 10 runs each: main and 9.0.5 handle the event at ~502 ms 10/10; 8.6.18 only at ~3000 ms together with the timer 10/10.
          • Cause 8.6 unix/tclUnixNotfy.c threaded build waits on a condition variable while the notifier thread does select(); the signal does not wake the waiting thread and Tcl_QueueEvent does not call Tcl_AlertNotifier.
          • Prototype none
          • Suggested ticket comment On 9.0 the event is handled at once (epoll/kqueue notifier, TIP 458); on 8.6 threaded it still waits for the next event. Tcl_QueueEvent/Tcl_Alloc are not async-signal safe anyway; signal handlers should use Tcl_AsyncMarkFromSignal() (TIP 511, 9.0) and queue the event from the async handler. Proposing to close.
          ⬜3606119fff2013load.test test independencepatchFIXED for all tests named in the ticket (load-4.2, 7.4, 8.1, 8.3a, 8.4, unload-2.3-2.6, 3.3-3.7, 4.1-4.6 pass alone on main and 8.6 tip); the newer…

          Open ticket 3606119fff · 40. Dynamic Loading · filed 2013-02-26

          🧪 Prototype check (2026-10-24)

          3606119fff.patch: VERIFIED

          • Base main (tip d6dd6d0f71); tests only
          • Before every test of load.test and unload.test run alone (make test TESTFLAGS="-file F -match T", script work-a1/v1/3606.sh): only load-8.3b fails (Total 30 Passed 0 Skipped 29 Failed 1); the ones named in the ticket pass alone
          • Patch applied cleanly (3 changed lines)
          • After no test fails alone; load.test 30/30 passed, unload.test 27/27 passed
          • Test the patch is itself the test fix; load-8.3b alone fails before / passes after
          Full verdict

          FIXED for all tests named in the ticket (load-4.2, 7.4, 8.1, 8.3a, 8.4, unload-2.3-2.6, 3.3-3.7, 4.1-4.6 pass alone on main and 8.6 tip); the newer load-8.3b (split from 8.3 in 2021, 828291fc64) still fails alone on main and 8.6

          • Platform generic (tests)
          • Tested every test of load.test and unload.test run alone with -match on main (own build) and the 8.6 tip tcltest; 9.0 tip has no dltest build (same test code)
          • Cause tests/load.test load-8.3b expects [info loaded child] to contain Test and Pkgb, which are loaded into child only by load-2.2 and load-7.1.
          • Prototype prototypes/3606119fff.patch (3 changed lines: load Test and Pkgb into child in the setup block before load-8.x; every load/unload test passes alone and in the full files)
          • Suggested ticket comment The tests listed here now pass when run alone; only the newer load-8.3b still depends on load-2.2/7.1. Proposed fix: load Test and Pkgb into child in the setup block before load-8.x (patch available).
          ✅3606118fff2013interp.test test independenceFIXED since 8.6.7 (dgp's test-suite backlog merged in 98f348e049, 2016-07-18)

          Open ticket 3606118fff · 20. [interp] · filed 2013-02-26

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.6.6 by 27e05ac49a (ticket 77d58e3a7a): all the listed interp tests pass when run alone.
          Full verdict

          FIXED since 8.6.7 (dgp's test-suite backlog merged in 98f348e049, 2016-07-18)

          • Platform generic (test suite)
          • Tested each of interp-2.4 2.10 3.7 3.9 3.10 6.4 6.6 7.3 7.4 7.5 14.1 run alone with -match: pass on main, 9.0 tip, 8.6 tip; 10 of 11 fail with the 8.5 test files on 8.5.19.
          • Cause tests relied on interps/aliases created by earlier tests; now created in -setup or by catch {...} lines between tests.
          • Prototype none (fixed)
          • Suggested ticket comment All listed interp tests pass in isolation on 8.6, 9.0 and 9.1 (setups added in &#91;98f348e049&#93;, first in 8.6.7); 8.5 is no longer maintained. Can be closed.
          ⬜3606117fff2013info.test test independenceFIXED since 8.6.0 (df9c4cb142 "More variable cleansing", dkf, 2009-10-29 on trunk); only 8.5 (EOL) still has it

          Open ticket 3606117fff · 17. Commands I-L · filed 2013-02-26

          Full verdict

          FIXED since 8.6.0 (df9c4cb142 "More variable cleansing", dkf, 2009-10-29 on trunk); only 8.5 (EOL) still has it

          • Platform generic
          • Tested -match info-30.2 and -match info-30.3 alone: pass on main, 9.0 tip, 8.6 tip tcltest builds. They fail with 8.5.13 and 8.5.19 on the 8.5 tree ("can't read res": the 8.5 test uses set res in namespace eval xxx, which writes ::res only if a global res is left over from earlier tests).
          • Cause tests/info.test (8.5): info-30.2/30.3 depended on a global res left by earlier tests. 8.6+ uses variable res / $xxx::res.
          • Prototype none (8.5 is EOL)
          • Suggested ticket comment info-30.2/30.3 pass in isolation on 8.6, 9.0 and main since df9c4cb142 (they now use variable res). Only the EOL 8.5 branch still has the old tests; closing.
          ✅3606116fff2013history-8.3 fails in isolationFIXED since 8.6.7 (98f348e049, 2016-07-18)

          Open ticket 3606116fff · 19. [history] · filed 2013-02-26

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.6.6 by 0a7e505919 (ticket 77d58e3a7a): history-8.3 now has its own setup and passes when run alone.
          Full verdict

          FIXED since 8.6.7 (98f348e049, 2016-07-18)

          • Platform generic (test suite)
          • Tested history-8.3 alone passes on main, 9.0 tip, 8.6 tip; fails with the 8.5 test file on 8.5.19.
          • Cause the test relied on history-8.2's history clear; a setup with history clear; history add "Testing" was added.
          • Prototype none (fixed)
          • Suggested ticket comment history-8.3 passes in isolation on 8.6, 9.0 and 9.1 (setup added in &#91;98f348e049&#93;, first in 8.6.7). Can be closed.
          ✅3606108fff2013env.test test independenceFIXED since 8.6.9 (9e55a28229 "Remove dependencies between tests in env.test", 2018-06-17)

          Open ticket 3606108fff · 08. Environment Variables · filed 2013-02-26

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.6.6 by 0a7e505919 (ticket 77d58e3a7a; reworked further in 9e55a28229): env-2.*, env-3.1 and env-4.* pass when run alone.
          Full verdict

          FIXED since 8.6.9 (9e55a28229 "Remove dependencies between tests in env.test", 2018-06-17)

          • Platform generic (test suite)
          • Tested env-2.3 2.4 3.1 4.1 4.3 4.4 4.5 alone pass on main, 9.0 tip, 8.6 tip (env-4.2 was deleted); env-4.1..4.5 fail with the 8.5 test file on 8.5.19.
          • Cause tests relied on variables set by earlier tests; now use setup1/setup2/cleanup1 procs.
          • Prototype none (fixed)
          • Suggested ticket comment Fixed by &#91;9e55a28229&#93; (first in 8.6.9): env tests now have their own setup/cleanup and pass in isolation on 8.6, 9.0 and 9.1. Can be closed.
          ✅3605722fff2013chanio.test has 3 tests, that didn't run independend.FIXED since 8.6.6 (98f348e049, 2016-07-18, dgp "merge backlog of test suite & doc improvements": chan-io-33.5 / chan-io-52.11 / io-52.11 got thei…

          Open ticket 3605722fff · 25. Channel System · filed 2013-02-23

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          Full verdict

          FIXED since 8.6.6 (98f348e049, 2016-07-18, dgp "merge backlog of test suite & doc improvements": chan-io-33.5 / chan-io-52.11 / io-52.11 got their own -setup; io-33.5's file is now created at file level)

          • Platform generic (test suite)
          • Tested chan-io-28.7, chan-io-33.5, chan-io-52.11, io-28.7, io-33.5, io-52.11 each run alone with -match on main, 9.0 tip and 8.6 tip: all pass, none hangs (8.6 io-28.7 is skipped by a constraint).
          • Cause n/a
          • Prototype none
          • Suggested ticket comment All six tests pass standalone on 8.6, 9.0 and main since the test rework in 98f348e049. Proposing to close.
          ⬜3598502fff2012Tcl_ListObj docs - shared objectspatchFIXED since 9.0 docs: doc/ListObj.3 documents both points on main and 9.0 — the NULL-objv preallocation (33a6f75e9c, 2020-04-24, from 8.7) and "req…

          Open ticket 3598502fff · 14. List Object · filed 2012-12-26

          🧪 Prototype check (2026-10-24)

          3598502fff.patch: VERIFIED

          • Base core-8-6-branch (tip 326e1bf923)
          • Before 8.6 doc/ListObj.3 still has no REFERENCE COUNT MANAGEMENT section, while generic/tclListObj.c panics "%s called with shared object" in Tcl_SetListObj (l.328), Tcl_ListObjAppendList (500), Tcl_ListObjAppendElement (559), Tcl_ListObjReplace (855)
          • Patch applied cleanly (16 changed lines, docs)
          • After man -l doc/ListObj.3 renders the new section correctly; its statements match the 8.6 code
          • Test none possible (documentation only)
          Full verdict

          FIXED since 9.0 docs: doc/ListObj.3 documents both points on main and 9.0 — the NULL-objv preallocation (33a6f75e9c, 2020-04-24, from 8.7) and "require an unshared listPtr" in REFERENCE COUNT MANAGEMENT (5b3ba7d74a, 2021-04-26). 8.6 ListObj.3 has the NULL-objv text but no unshared-listPtr rule (the code panics "called with shared object").

          • Platform docs
          • Tested read doc/ListObj.3 on main, 9.0 and 8.6
          • Cause the reference-count section was never backported to 8.6.
          • Prototype prototypes/3598502fff.patch (8.6 docs, core-8-6-branch; 16 changed lines: REFERENCE COUNT MANAGEMENT section for Tcl_NewListObj/Tcl_SetListObj/AppendList/AppendElement/Replace, taken from 9.0 and limited to what 8.6 does; checked with man -l)
          • Suggested ticket comment Both points are documented in ListObj.3 since 8.7/9.0. Proposed: backport the "unshared listPtr" paragraph to the 8.6 man page, then close.
          ✅3598300fff2012unix: tcl.h does not include sys/stat.hFIXED since 8.5.14 / 8.6.1 (590472b443 trunk / db771d0a61 core-8-5-branch, 2013-01-19)

          Open ticket 3598300fff · 52. Portability Support · filed 2012-12-24

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.5.14 and 8.6.1 by 992ef8a601 and 590472b443: tclPort.h now includes tclUnixPort.h (and so <sys/stat.h>) before tcl.h, also on macOS. In Tcl 9, tcl.h always uses struct stat on Unix.
          Full verdict

          FIXED since 8.5.14 / 8.6.1 (590472b443 trunk / db771d0a61 core-8-5-branch, 2013-01-19)

          • Platform unix (musl libc)
          • Tested code reading on main, 9.0 and 8.6 (no musl toolchain here). The 8.6 generic/tclPort.h includes tclUnixPort.h (which includes <sys/stat.h>) before tcl.h, and the Apple stat64 case is handled in tclUnixPort.h. On 9.0/main, tcl.h always uses typedef struct stat Tcl_StatBuf on Unix, so the musl #define stat64 stat problem cannot occur.
          • Cause tcl.h picked struct stat64 before <sys/stat.h> had been seen. The final fix (after the 8abba84224 revert because of OSX Leopard) put the platform port header before tcl.h. The leftover branch bug-3598300 (rebased 2018/2019, never merged) only drops HAVE_STRUCT_STAT64 on Apple.
          • Prototype none (fixed)
          • Suggested ticket comment Fixed in 8.5.14/8.6.1 by 590472b443: tclPort.h now includes tclUnixPort.h (and so <sys/stat.h>) before tcl.h. Tcl 9 always uses struct stat. Closing.
          ✅3588687fff2012Tcl_InitStubs() Broken As DesignedFIXED since 9.0.0 (mostly; dac224dbfc by jan.nijtmans, 2016-06-27, plus later magic handling in tclStubLib.c, e.g. 7704f91f2e, 2025-01-07). Par…

          Open ticket 3588687fff · 39. Package Manager · filed 2012-11-20

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 9.0.0: Tcl_InitStubs is now a macro that passes the extension header's TCL_STUB_MAGIC and major/minor version (dac224dbfc), so an extension built with Tcl 8 headers cannot pass the checks against a Tcl 9 stub table. It gets "this extension is compiled for Tcl 8.6" (7704f91f2e).
          Full verdict

          FIXED since 9.0.0 (mostly; dac224dbfc by jan.nijtmans, 2016-06-27, plus later magic handling in tclStubLib.c, e.g. 7704f91f2e, 2025-01-07). Part of dgp's branch bug-3588687 was merged to trunk for 8.6.0 (e738bbbfd9, 2012-11-29).

          • Platform generic (C API / build)
          • Tested read generic/tcl.h and generic/tclStubLib.c on main, 9.0, 8.6
          • Cause the ticket's problem was that the stub library's TCL_STUB_MAGIC, not the extension's header, was checked. In 9.x Tcl_InitStubs is a macro that passes the extension header's TCL_STUB_MAGIC and TCL_MAJOR/MINOR_VERSION (in exact) to the stub library. The library checks them against the interp's table and gives clear errors ("this extension is compiled for Tcl 8.6"). Tcl 8 headers (no version bits) are checked against the Tcl 8 magic, so 8-header + 9-libtclstub + 9 interp is refused. 8.6 keeps the 3-argument function, but all 8.x tables share one magic and layout, so the mismatch cannot arise there.
          • Prototype none (design change already done in 9.0)
          • Suggested ticket comment Since 9.0 Tcl_InitStubs() is a macro passing the header's TCL_STUB_MAGIC and major/minor version (dac224dbfc), so a header/stub-library/interp mismatch is detected at load time with a clear error. Proposing to close as fixed in 9.0.
          ⬜3562482fff2012info frame gives wrong line number in catchFIXED since 8.6.0 (still present in 8.5.19, the last 8.5 release); exact check-in not identified

          Open ticket 3562482fff · None · filed 2012-08-28

          Full verdict

          FIXED since 8.6.0 (still present in 8.5.19, the last 8.5 release); exact check-in not identified

          • Platform generic
          • Tested attached script plus proc/eval variants on main, 9.0.5, 8.6.0–8.6.3, 8.6.18 (correct: line 5) and 8.5.19 (wrong: line 4, and off by one in a proc body after ;# lines)
          • Cause in 8.5 the line counter skipped the newline of an empty command consisting only of ;# comment; the 8.6 compiler (TclCompileScript line tracking via envPtr->line) counts it correctly.
          • Prototype none (fixed in 8.6; 8.5 is EOL)
          • Suggested ticket comment Works correctly since 8.6.0 (line 5 reported); only 8.5 is affected and it is no longer maintained. Suggest closing as Fixed (out of date).
          ✅3547593fff2012fcmd test failures on WindowsFIXED since 8.6.14 / 9.0 (tests rewritten: apnadkarni's testchmod fixes, merged to trunk in 24072d92fc, 2023-03-07; 2.8.1/tmpspace fixed in 2012 …

          Open ticket 3547593fff · 37. File System · filed 2012-07-23

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: All the listed tests were reworked: winFCmd-2.8.1 and 6.9 were removed, 6.1, 6.13 and 9.3 were rewritten with Ashok's testchmod fixes (24072d92fc, 2023), and 7.15 accepts EEXIST. Fixed in 8.6.14 and 9.0. Closing.
          Full verdict

          FIXED since 8.6.14 / 9.0 (tests rewritten: apnadkarni's testchmod fixes, merged to trunk in 24072d92fc, 2023-03-07; 2.8.1/tmpspace fixed in 2012 by jan.nijtmans)

          • Platform Windows
          • Tested code reading of tests/winFCmd.test on main and 8.6 tip: winFCmd-2.8.1 gone, 6.9 removed (dup of 6.1), 6.1/6.13/9.3 now use an intermediate td0 directory with FILE_DELETE_CHILD cleared and expect EACCES (constraint notInCIenv), 7.15 accepts EACCES or EEXIST
          • Cause the old tests assumed XP semantics of chmod 000 directories; the parent's FILE_DELETE_CHILD right (Full Control) decided the result.
          • Prototype none (already fixed)
          • Suggested ticket comment All the listed tests were reworked (2.8.1 removed, 6.1/6.13/9.3 rewritten with the testchmod fixes in 2023, 6.9 removed, 7.15 accepts EEXIST). Fixed in 8.6.14 and 9.0; closing.
          💬3545876fff2012tclLoadDyld.c deprecationsFIXED as far as the warnings go (deprecation warnings silenced by #pragma GCC diagnostic ignored "-Wdeprecated-declarations", trunk 4760483bbd …

          Open ticket 3545876fff · 40. Dynamic Loading · filed 2012-07-19

          💬 Since triage (updated 2026-11-11)

          Status now: Open

          • 💬 2026-10-29 chrstphrchvz: Usage of these functions is still present, but their deprecation warnings have been suppressed since 4760483b (8.5.19 and 8.6.5). My impression is that usage of these functions is not strictly necessary and possibly no longer beneficial. NSCreateObjectFileImageFromMemory() was intentionally deprecated without replacement due to security concerns. The remaining deprecated functions were replaced by macOS 10.5's i…
          Full verdict

          FIXED as far as the warnings go (deprecation warnings silenced by #pragma GCC diagnostic ignored "-Wdeprecated-declarations", trunk 4760483bbd 2015-04-09; the pragma is in 8.6 too); the deprecated NSModule APIs are still used on main, 9.0, 8.6, now only under TCL_LOAD_FROM_MEMORY (load from memory, e.g. from zipfs)

          • Platform macOS / build
          • Tested code reading only (CANNOT-TEST on macOS here); normal loading uses dlopen/dlsym
          • Cause TclpLoadMemory and the symbol lookup for memory-loaded bundles need NSCreateObjectFileImageFromMemory/NSLinkModule/NSLookupSymbolInImage, which have no non-deprecated replacement.
          • Prototype none (nothing to change unless load-from-memory is dropped; that is a feature decision)
          • Suggested ticket comment The warnings have been silenced since 2015 (4760483bbd); the deprecated NSModule calls remain only for TCL_LOAD_FROM_MEMORY, which has no dlopen-based replacement. Suggest closing; reopen if Apple removes the API.
          ⬜3541726fff2012Clock Scan with year/ year/monthFIXED since 9.0 (TIP 688 clock engine, dd916eacf3, 2024-04-01): 2009 -format %Y gives 2009-07-09 and 2009/09 -format %Y/%m gives 2009-09-09 w…

          Open ticket 3541726fff · 06. Time Measurement · filed 2012-07-09

          Full verdict

          FIXED since 9.0 (TIP 688 clock engine, dd916eacf3, 2024-04-01): 2009 -format %Y gives 2009-07-09 and 2009/09 -format %Y/%m gives 2009-09-09 with base 2012-07-09, so the scanned year/month are kept; the missing day still comes from the base. 8.6.18 and 8.7b1 still return the base date.

          • Platform generic
          • Tested main c3bf6dd052, 9.0.5, 8.7b1, 8.6.18 (work-a6/t9.tcl)
          • Cause 8.6 library/clock.tcl: "year" or "year + month" is not a complete field set, so rule &#91;7&#93; (base date) applies.
          • Prototype none (fixed in 9.x; 8.6 behaviour is documented)
          • Suggested ticket comment Since 9.0 (TIP 688, dd916eacf3) the scanned year and month are kept; the missing parts come from -base. 8.6 keeps the documented behaviour (base date). Suggest closing.
          ⬜3540794fff2012Clock adds days without unitFIXED since 9.0 (TIP 688, dd916eacf3, 2024-04-01, validation on by default): clock scan "Mar. 1980" now fails with "invalid day"; with `-valid …

          Open ticket 3540794fff · 06. Time Measurement · filed 2012-07-06

          Full verdict

          FIXED since 9.0 (TIP 688, dd916eacf3, 2024-04-01, validation on by default): clock scan "Mar. 1980" now fails with "invalid day"; with -valid 0, and in 8.6.18, it still gives 2017-08-01 for base 2012-07-09.

          • Platform generic
          • Tested main c3bf6dd052, 9.0.5, 8.6.18 (work-a6/t9.tcl, t10.tcl)
          • Cause free-form grammar reads "Mar 1980" as month + day of month 1980 (not as relative days), normalised to 2017-08-01; 9.x validation rejects day 1980.
          • Prototype none (fixed in 9.x by default; making "Mon YYYY" a year would change the free-form grammar)
          • Suggested ticket comment In 9.0 "Mar. 1980" is rejected ("invalid day"): the free-form scanner reads 1980 as the day of month, and validation is now on by default. 8.6 (and 9.x with -valid 0) still normalise it. Suggest closing.
          ✅3534607fff2012tpool::get sets errorInfo incorrectlyFIXED since Thread 2.7.1/2.8.0 (Thread check-in 6e420680e0, 2014-08-14, "Fix [3534607]: tpool::get sets errorInfo incorrectly")

          Open ticket 3534607fff · 80. Thread Package · filed 2012-06-12

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in Thread 2.7.1 by 6e420680e0: tpool::get now sets errorInfo before the result, so the first line is no longer doubled. Checked with Thread 2.8.13 and 3.0.7.
          Full verdict

          FIXED since Thread 2.7.1/2.8.0 (Thread check-in 6e420680e0, 2014-08-14, "Fix [3534607]: tpool::get sets errorInfo incorrectly")

          • Platform generic (Thread package)
          • Tested job proc foo {} {error "my error"}; foo then catch tpool::get. errorInfo and -errorinfo are correct, with no doubled first line, on main + Thread 3.0.7 and 8.6.18 + Thread 2.8.13.
          • Cause n/a
          • Prototype none
          • Suggested ticket comment Fixed in Thread &#91;6e420680e0&#93; (2014). errorInfo from tpool::get is correct with Thread 2.8.13 and 3.0.7. Can be closed.
          ✅3522894fff2012Solaris build fails: TclBN_mp_init_set_int not foundFIXED since 8.6.12 / 8.7a3 (8619c767c0 core-8-branch, 2019-12-24; 8.6: 70529bfafa, 2021-02-08)

          Open ticket 3522894fff · 53. Configuration and Build Tools · filed 2012-05-02

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: The cause was the user's CPPFLAGS (-I/usr/local/include with older installed Tcl headers) coming before Tcl's own include directories. Since 8619c767c0 (8.6.12, and in 9.0) Tcl's own -I directories come first. Checked on Linux with a directory of decoy tcl.h/tclInt.h/tclPort.h passed in CPPFLAGS: 8.6.11 picks them up and fails, 8.6.12, core-8-6-branch and main build fine. Closing as fixed.
          Full verdict

          FIXED since 8.6.12 / 8.7a3 (8619c767c0 core-8-branch, 2019-12-24; 8.6: 70529bfafa, 2021-02-08)

          • Platform build
          • Tested unix/Makefile.in at the release tags: core-8-6-10 and core-8-6-11 still have ${CFLAGS} ... -I${GENERIC_DIR}, core-8-6-12 and later have -I"${BUILD_DIR}" -I${UNIX_DIR} -I${GENERIC_DIR} -I${TOMMATH_DIR} ${CFLAGS}. The main build log confirms that Tcl's own -I come before the user CFLAGS/CPPFLAGS.
          • Cause configure merges the user's CPPFLAGS (here -I/usr/local/include) into CFLAGS, and STUB_CC_SWITCHES put CFLAGS before Tcl's own -I dirs. An older installed tclTomMath.h/tclTomMathDecls.h therefore shadowed the source tree's copy and produced a reference to a missing TclBN_ symbol (pooryorick saw the same thing on Linux). Commit 8619c767c0 "Put our own included (-I) directories before directories coming in though CFLAGS" fixed it.
          • Prototype none (fixed)
          • Suggested ticket comment The cause was the user CPPFLAGS -I/usr/local/include (an older installed tclTomMath*.h) coming before Tcl's own include dirs. Since 8619c767c0 (8.6.12, 8.7/9.0) Tcl's -I directories come first. Closing as fixed.
          ✅3518301fff2012Tests assume nothing is already listening on port 9999FIXED since 8.6.14 (903c407a86, 2023-06-19, merged to 8.7 6e88d47b21 and main 2bedc17b74)

          Open ticket 3518301fff · 25. Channel System · filed 2012-04-16

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: The io/chan-io tests use no fixed port since 903c407a86 (8.6.14, 9.0.0). The httpold-4.12 timing problem was fixed in 8.6.13 by aaa3de736b (2641672fff), and httpold.test is gone in 9.0.
          Full verdict

          FIXED since 8.6.14 (903c407a86, 2023-06-19, merged to 8.7 6e88d47b21 and main 2bedc17b74)

          • Platform generic
          • Tested main (own build) and 8.6 tip with a foreign server holding port 9999: io-53.10 and chan-io-53.10 pass. Port 9999 is still hard-coded in core-8-6-13 and earlier, gone in core-8-6-14 (2024-02-21). httpold.test no longer exists.
          • Cause tests/io.test and chanio.test used socket -server ... 9999; now they use port 0.
          • Prototype none (fixed)
          • Suggested ticket comment Fixed by 903c407a86 (in 8.6.14, 9.0 and later): the io/chan-io tests now use port 0. httpold.test has been removed since. Closing.
          ✅3486740fff2012tcl crash dumpFIXED since 8.6.7 (7a3c8a4b0d on core-8-6-branch, 2016-10-11, fix for 3cc1d91345; on the 8.5 branch 509d964dcf, after 8.5.19, so in no 8.5 release)

          Open ticket 3486740fff · 41. Memory Allocation · filed 2012-02-11

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: The backtrace (TclFreeAllocCache called from the pthread TSD destructor at thread exit) matches 3cc1d91345, which was fixed in 8.6.7 by 7a3c8a4b0d. The allocator cache is no longer freed by a pthread key destructor. Closing; please reopen if it still happens with 8.6.7 or later.
          Full verdict

          FIXED since 8.6.7 (7a3c8a4b0d on core-8-6-branch, 2016-10-11, fix for 3cc1d91345; on the 8.5 branch 509d964dcf, after 8.5.19, so in no 8.5 release)

          • Platform unix (Linux, pthreads)
          • Tested decoded the attached crash.JPG (stored UTF-8 mangled; re-encoded to latin-1 in work-a6/att/crash2.jpg). Backtrace: clone -> start_thread -> __nptl_deallocate_tsd -> TclFreeAllocCache -> PutBlocks -> LockBucket -> jump to a bad address. Main and 8.6 now call pthread_key_create(&key, NULL) for the alloc cache key (unix/tclUnixThrd.c), so pthreads no longer calls TclFreeAllocCache when a thread exits.
          • Cause in 8.5.9, TclFreeAllocCache was registered as a pthread TSD destructor. Run automatically at thread exit (in unspecified order, after or besides Tcl_FinalizeThread), it touched freed or finalized allocator state. That is exactly 3cc1d91345.
          • Prototype none
          • Suggested ticket comment The backtrace (TclFreeAllocCache called from __nptl_deallocate_tsd) matches &#91;3cc1d91345&#93;, fixed in 8.6.7 (&#91;7a3c8a4b0d&#93;): the allocator cache is no longer freed by a pthread key destructor. Closing as fixed; please reopen if it still happens with 8.6.7+.
          ✅3475995fff2012[solved/opt] clock scan success by logical invalid date/timeFIXED since 9.0 (TIP 688, dd916eacf3, 2024-04-01): clock scan 2008-02-30 -> "invalid day", 24:62:80 -> "invalid time", -valid 0 gives the o…

          Open ticket 3475995fff · 06. Time Measurement · filed 2012-01-19

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Implemented in 9.0.0: TIP 688 (dd916eacf3) added clock scan -validate, and TIP 690 (f330be6e51) made validation the default, so 2008-02-30 and 24:62:80 are rejected. -valid 0 gives the old behaviour. 8.6 is unchanged.
          Full verdict

          FIXED since 9.0 (TIP 688, dd916eacf3, 2024-04-01): clock scan 2008-02-30 -> "invalid day", 24:62:80 -> "invalid time", -valid 0 gives the old result. 8.6.18 unchanged (no -valid option).

          • Platform generic
          • Tested main c3bf6dd052, 9.0.5, 8.6.18 (work-a6/t9.tcl)
          • Cause n/a (the reporter's own validation proposal landed as part of the 9.0 clock engine)
          • Prototype none (fixed)
          • Suggested ticket comment Implemented in 9.0 by TIP 688 (dd916eacf3): scan validates by default, -valid 0 restores the old behaviour. Suggest closing.
          ✅3432269fff2011load failure under windows, with unix pathsFIXED since 8.6.2 (Linux analysis points to 578237c4c6, 2014-04-10; not bisected)

          Open ticket 3432269fff · 40. Dynamic Loading · filed 2011-11-02

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.6.2 by 578237c4c6: native Windows paths now always use backslashes, so LoadLibraryExW() also finds dependent DLLs next to an extension that was loaded with a forward-slash path. Tested on Windows 11: fails with 8.6.0/8.6.1, works with 8.6.2, 8.6.18, 9.0.5 and 9.1.
          Full verdict

          FIXED since 8.6.2 (Linux analysis points to 578237c4c6, 2014-04-10; not bisected)

          • Platform Windows
          • Tested code reading of win/tclWinLoad.c and win/tclWinFile.c on main, 9.0, 8.6: TclpDlopen passes Tcl_FSGetNativePath() (now always backslashes) to LoadLibraryExW(..., LOAD_WITH_ALTERED_SEARCH_PATH)
          • Cause LOAD_WITH_ALTERED_SEARCH_PATH only searches the DLL's own directory for dependent DLLs (libmysql.dll here) when the path uses backslashes; in 8.6b2 the native path kept the forward slashes of the Tcl path, while [file native] gave backslashes.
          • Prototype none (already fixed)
          • Suggested ticket comment Since 8.6.2 the native path passed to LoadLibraryExW() always uses backslashes, so LOAD_WITH_ALTERED_SEARCH_PATH finds dependent DLLs next to the loaded one even for paths written with "/". Please reopen if it still happens with a current release.
          Windows VM, run 5
          • Platform Windows
          • Verdict FIXED since 8.6.2 (Linux analysis points to 578237c4c6, 2014-04-10; not bisected)
          • Tested VM work-win5/ld.tcl with two MSVC-built DLLs in ld/sub (ext.dll imports dep.dll, cwd elsewhere): load C:/.../ld/sub/ext.dll Ext fails on 8.6.0 and 8.6.1 ("this library or a dependent library could not be found"), works with backslashes; works with both forms on 8.6.2, 8.6.18, 9.0.5, 9.1b1 (and 8.5.19).
          • Cause LOAD_WITH_ALTERED_SEARCH_PATH needs a backslash path to search the DLL's directory for dependents; 8.6.0/8.6.1 passed the forward-slash form.
          • Prototype none (fixed)
          • Suggested ticket comment Fixed since 8.6.2: on Windows 11 loading an extension with a forward-slash path whose dependent DLL sits next to it fails with 8.6.0 and 8.6.1 but works with 8.6.2, 8.6.18, 9.0.5 and 9.1. Closing.
          ✅3407860fff2011thread::exit flawsFIXED since Thread 2.8.0 (0839286864, 2016-05-31; branch bug-3407860fff 679580a1b4)

          Open ticket 3407860fff · 80. Thread Package · filed 2011-09-11

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in Thread 2.8.0 by 0839286864: thread::exit now accepts an optional exit status (default 666), which thread::join returns, and rejects extra arguments.
          Full verdict

          FIXED since Thread 2.8.0 (0839286864, 2016-05-31; branch bug-3407860fff 679580a1b4)

          • Platform generic
          • Tested Thread 3.0.7 (trunk) on Tcl main 9.1.1: thread::exit 42 -> thread::join returns 42; thread::exit 1 2 -> "wrong # args: should be "thread::exit ?status?""; no arg -> 666. Thread 2.8.0 released 2016-07-15.
          • Cause ThreadExitObjCmd ignored its arguments; now accepts optional integer status and checks objc.
          • Prototype none (already fixed)
          • Suggested ticket comment Fixed in Thread 2.8.0 by 0839286864: &#91;thread::exit ?status?&#93; now takes an optional exit code (default 666) and rejects extra arguments. Can be closed.
          ⬜3389684fff2011clock-33.11* timing sensitiveFIXED (unreleased; 9f7ba66066, 2026-09-13, core-8-6-branch, merged to 9.0 cdf41ba7d7 and main 0ba0bea858)

          Open ticket 3389684fff · 06. Time Measurement · filed 2011-08-10

          Full verdict

          FIXED (unreleased; 9f7ba66066, 2026-09-13, core-8-6-branch, merged to 9.0 cdf41ba7d7 and main 0ba0bea858)

          • Platform generic
          • Tested read tests/clock.test on main, 9.0, 8.6 tips and core-9-0-4 / core-8-6-18 tags
          • Cause the old clock-33.11/33.11a used an unbounded while {1} loop waiting for two reads in the same millisecond. The new ::testClock::aligned-times helper runs the loop inside timerate ... 1000 and raises "unable to retrieve stable value ... in 1s", so a very slow build gets a failure instead of a hang. Released 9.0.4 and 8.6.18 still have the old loop.
          • Prototype none (already fixed on all branches)
          • Suggested ticket comment The tests were rewritten in 9f7ba66066 (merged to 9.0 and main): the alignment loop is now limited to 1s via &#91;timerate&#93;, so a slow build gets a failure instead of a hang. Can be closed once 8.6.19/9.0.5 are out.
          ⬜3387109fff2011clock -timezone bug with tzdata filespatchpart 1 (names with "+"/digits) FIXED since 8.6.16 / 9.0.1 (67e7522bd1, 2024-10-29, ticket 2c237beffb); part 2 (:GMT0 after GMT0) REPRODUCED (…

          Open ticket 3387109fff · 06. Time Measurement · filed 2011-08-06

          🧪 Prototype check (2026-10-24)

          3387109fff.patch: VERIFIED

          • Base main (tip d6dd6d0f71)
          • Before work-a4/v4clock.tcl (clock format 0 -timezone GMT0, then -timezone :GMT0) -> 0 GMT 1 {can't read "::tcl::clock::TZData(:GMT0)": no such element in array} (8.6 tip: same, can't read "TZData(:GMT0)")
          • Patch ported: the test hunk no longer applies and clock-71.x is now taken on main (he locale era tests), so the test is renamed clock-72.1 and placed after clock-71.2; the library/clock.tcl one-line change is identical. 12 changed lines.
          • After -> 0 GMT 0 GMT; clock.test (TCL_LIBRARY=<wt>/library ./tcltest): 8764 total, 8757 passed, 7 skipped, 0 failed
          • Test clock-72.1 (was clock-71.1 in the prototype), fails before / passes after
          Full verdict

          part 1 (names with "+"/digits) FIXED since 8.6.16 / 9.0.1 (67e7522bd1, 2024-10-29, ticket 2c237beffb); part 2 (:GMT0 after GMT0) REPRODUCED (main, 9.0, 8.6, 8.5)

          • Platform generic
          • Tested main build, 9.0.5, 8.6.18, 8.5.19 tip build, releases 8.5.10–8.5.19, 8.6.0–8.6.18 (:Etc/GMT+5 fails up to 8.6.15, works from 8.6.16; regexp gone in core-9-0-1)
          • Cause library/clock.tcl LoadTimeZoneFile tests for TZData($fileName), but a tzdata file sets TZData(:$fileName). After -timezone GMT0 (parsed as a POSIX zone, which sets TZData(GMT0)), -timezone :GMT0 skips sourcing the file, and SetupTimeZone then fails with can't read "::tcl::clock::TZData(:GMT0)".
          • Prototype prototypes/3387109fff.patch (12 changed lines, test clock-71.1; against main, and the same one-line change applies to 8.6)
          • Suggested ticket comment Names like Etc/GMT+5 work since 8.6.16/9.0.1 (67e7522bd1, 2c237beffb). The second issue remains: LoadTimeZoneFile checks TZData(name) instead of TZData(:name), so clock format 0 -timezone :GMT0 fails after -timezone GMT0. Proposed fix: check TZData(:$fileName), test clock-71.1.
          ✅3378439fff2011Cannot load tdbc::odbcFIXED since 8.6.0 (not reproducible with releases)

          Open ticket 3378439fff · 69. Other · filed 2011-07-26

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.6b2 by b90f247299: Tcl now registers a dummy package ifneeded TclOO, so the built-in TclOO is found by every package mechanism and tdbc::odbc no longer fails with "Error loading TclOO package".
          Full verdict

          FIXED since 8.6.0 (not reproducible with releases)

          • Platform generic (8.6 beta packaging)
          • Tested package require TclOO works on 8.6.18 (1.1.0) and 9.0.5 (1.3.1); tdbc itself is not installed in these builds, so tdbc::odbc was not loaded
          • Cause the 8.6b1.2 build used did not provide the TclOO package (as dkf noted, 8.6 always provides it, built in).
          • Prototype none
          • Suggested ticket comment Tcl 8.6.0 and later always provide the TclOO package (package require TclOO works), so the error from the 8.6b1 build no longer happens. Closing as out of date; tdbc problems belong in the tdbc tracker.
          ⬜3363556fff2011Tcl_SetObjResult can't handle NULL, despite documentationpatchFIXED since 9.0.0 (f9a80c2c9e, 2023-04-13, "Clean up Tcl_SetObjResult()", which dropped the misleading inline comment). Still present on 8.6 (comme…

          Open ticket 3363556fff · 10. Objects · filed 2011-07-11

          🧪 Prototype check (2026-10-24)

          3363556fff.patch: VERIFIED

          • Base core-8-6-branch (tip 326e1bf923)
          • Before core-8-6-branch generic/tclResult.c:523 still says "If NULL, the obj result is made an empty string object" while the code does Tcl_IncrRefCount(objPtr); main has the corrected comment ("The value to set as the result.", f9a80c2c9e)
          • Patch applied cleanly (4 changed lines, comment only, unchanged)
          • After 8.6 builds without warnings; no behaviour change, so no tests run
          • Test none possible (comment only)
          Full verdict

          FIXED since 9.0.0 (f9a80c2c9e, 2023-04-13, "Clean up Tcl_SetObjResult()", which dropped the misleading inline comment). Still present on 8.6 (comment only). The man page SetResult.3 never promised NULL support; the quoted text came from 8.0a1 docs.

          • Platform generic (C API comment)
          • Tested code reading of main, 9.0 and 8.6 generic/tclResult.c, and doc/SetResult.3
          • Cause core-8-6-branch generic/tclResult.c:Tcl_SetObjResult parameter comment still says "If NULL, the obj result is made an empty string object", but the code dereferences objPtr.
          • Prototype prototypes/3363556fff.patch (against core-8-6-branch, 4 changed lines, comment only)
          • Suggested ticket comment The man page never allowed NULL, and the wrong inline comment was removed on 9.0/main in f9a80c2c9e. Proposed comment fix for core-8-6-branch is available; after that this can be closed.
          ✅3181906fff2011Compile error with --disable-load on Mac OS XFIXED since 8.5.14 / 8.6.1 (bf36815647 trunk, f8694d903a core-8-5-branch, 2013-02-05, ticket [3433012])

          Open ticket 3181906fff · 40. Dynamic Loading · filed 2011-02-15

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.5.14 and 8.6.1 by bf36815647 (ticket [3433012]): tclLoadNone.c now provides dummy TclpLoadMemory functions, so a --disable-load build links on macOS.
          Full verdict

          FIXED since 8.5.14 / 8.6.1 (bf36815647 trunk, f8694d903a core-8-5-branch, 2013-02-05, ticket [3433012])

          • Platform build (macOS)
          • Tested Linux simulation on main: out-of-tree configure --disable-load CFLAGS=-DTCL_LOAD_FROM_MEMORY=1 links fine and load reports "dynamic loading is not currently available". 8.5/8.6/9.0/main sources all have the fallbacks. Not built on macOS.
          • Cause unix/configure defined TCL_LOAD_FROM_MEMORY on Darwin even with tclLoadNone.o; generic/tclLoadNone.c now has dummy TclpLoadMemoryGetBuffer/TclpLoadMemory.
          • Prototype none (already fixed)
          • Suggested ticket comment Fixed for 8.5.14/8.6.1 by bf36815647 (ticket [3433012]): tclLoadNone.c now has fallback TclpLoadMemory* functions, so --disable-load links on macOS. Closing.
          ⬜3156499fff2011more info frame incorrect source type resultspatchMostly FIXED, one case REPRODUCED (main, 9.0, 8.6, 8.5). The if "$$y != 1" case reporting line 14 was fixed by 8.5.9. Dynamic command words insid…

          Open ticket 3156499fff · 46. Traces · filed 2011-01-12

          🧪 Prototype check (2026-10-24)

          3156499fff.patch: VERIFIED

          • Base main (tip d6dd6d0f71)
          • Before work-a3/315649e.tcl (testevalex with $cmd42, [string cat info42], {*}[list info42], info42 on lines 2-5) -> -1 -1 -1 5; work-a3/315649.tcl (main script) -> source line -1 cmd $p1, source line -1 cmd [set p1]
          • Patch applied cleanly (29 changed lines). Its testConstraint testevalex line repeats info.test:1412 and can be dropped. It conflicts textually with 1695203fff.patch, since both append info-42.x at the end of info.test; that is trivial to merge.
          • After 315649e.tcl -> 2 3 4 5; 315649.tcl -> lines 6, 8, 10; the ticket attachment info_frame_3.tcl reports the expected line in every "should be" case (60, 14, 27, 35, 64, 68). Tests info, basic, trace, proc, source, parse, eval, apply, coroutine, nre: 0 failed
          • Test existing info-42.1 (from the patch) fails on the unpatched tip / passes after
          Full verdict

          Mostly FIXED, one case REPRODUCED (main, 9.0, 8.6, 8.5). The if "$$y != 1" case reporting line 14 was fixed by 8.5.9. Dynamic command words inside procs (p$p2, [namespace current]::test_proc) were fixed in 8.6.6 by 4b61afd660 (c94c7c4f4c on core-8-6-branch, 2016-06-16). Still wrong: a dynamic command word ($cmd, [..], {*}) in a script evaluated by TclEvalEx (the tclsh main script, Tcl_EvalFile/Tcl_EvalEx from C) reports line -1. That is the reporter's "64 should be source line" case.

          • Platform generic
          • Tested the attachment info_frame_3.tcl (work-a3/info_frame_3.tcl) on main, 9.0.5, 8.6.0-8.6.18 and 8.5.6-8.5.19. Minimal repros: work-a3/315649.tcl (main script) and 315649e.tcl (testevalex: gives -1 -1 -1 5, expected 2 3 4 5). sourced files take the compiled path and are correct.
          • Cause generic/tclBasic.c:TclEvalEx sets lines[word] = -1 for words not known at compile time, including word 0, and [info frame] reports line[0] as the command's line. TclArgumentEnter/Release skip word 0, so word 0 can safely always carry the command's line.
          • Prototype prototypes/3156499fff.patch (29 changed lines, test info-42.1, added at the end of info.test, testevalex constraint). It sets lines[0] to the line of the first word after word substitution and expansion. info/basic/trace/proc/source/parse tests pass, and info-42.1 fails without the fix.
          • Suggested ticket comment Most cases were fixed long ago (8.5.9; 4b61afd660 in 8.6.6). Remaining: in TclEvalEx (main script, Tcl_EvalFile/Tcl_EvalEx) a dynamic command word gives &#91;info frame&#93; line -1. Proposed fix: keep the command's line in line&#91;0&#93; (TclArgumentEnter ignores word 0); patch with test info-42.1 is available.
          🔧3099612fff2010false, shortcut string /../ simplification everywhere, all fpatchFIXED since 8.5.0 (8.4-only behaviour; no single check-in, came with the 8.5 path normalization rewrite)

          Open ticket 3099612fff · 37. File System · filed 2010-10-31

          🔧 Since triage (updated 2026-11-11)

          Status now: Open

          • 🔧 2026-10-19 check-in 854be12c0a on normalize-dotdot-after-link, not merged: Fix 3577801fff, 3570244fff, 3183016fff and 3099612fff: on Unix, [file normalize] and [cd ..] used the target of a link literally when ".." followed it,
          • 💬 2026-10-19 serhiy.storchaka: This works correctly since 8.5 (the link is resolved before ".."), and [file normalize] is covered by filesystem-1.26 to 1.28. As asked, branch normalize-dotdot-after-link adds test filesystem-1.58 for [file stat] and [file exists] through a relative link and "..". The ticket can be closed when it is merged.
          Full verdict

          FIXED since 8.5.0 (8.4-only behaviour; no single check-in, came with the 8.5 path normalization rewrite)

          • Platform unix (generic)
          • Tested Linux main, 9.0.5, 8.6.18, 8.5.0: file stat/exists/isdirectory etc/httpd/run/../tmp/. with run -> ../../var/run resolves through the link (same inode as var/tmp). 8.4.20 fails (lexical).
          • Cause 8.4's normalization collapsed x/.. lexically; 8.5+ TclFSNormalizeAbsolutePath resolves the link before ... Normalize is covered by filesystem-1.26..1.28, but no test checks access through link/...
          • Prototype prototypes/3099612fff.patch (14 changed lines, test filesystem-1.56; test only, passes on main and 8.6)
          • Suggested ticket comment Works correctly since 8.5 (link resolved before ..); filesystem-1.26..1.28 cover file normalize. Proposed test filesystem-1.56 covers file stat/file exists through a relative link and ... Can be closed once the test is added.
          ✅3075155fff2010oo depends on bytecoding?FIXED since 8.6.0 (0b5e337729, 2012-04-09)

          Open ticket 3075155fff · 07. Variables · filed 2010-09-24

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Since 8.6.0 (0b5e337729, bug [2712377]), [info vars] in a method lists the declared object variables itself, so the result no longer depends on bytecompilation. The show.tcl cases give the same result on 8.6 and 9.x.
          Full verdict

          FIXED since 8.6.0 (0b5e337729, 2012-04-09)

          • Platform generic
          • Tested attached show.tcl (incr vs. non-compiled myIncr / $::incr) on main (9.1.1 build), 9.0.5, 8.6.18, 8.6.0–8.6.6: all three variants now give {3 2 y! {}} {x! y!} {x! y!}.
          • Cause [info vars] (tclVar.c AppendLocals) did not know about TclOO declared variables, so the result depended on whether [incr] put y! into the LVT. Fixed by "Fix &#91;Bug 2712377&#93;: &#91;info vars&#93; and object declared variables", which appends the method's declared variables (objectVars section).
          • Prototype none (fixed)
          • Suggested ticket comment The result of &#91;info vars&#93; in methods no longer depends on bytecompilation: declared object variables are listed explicitly since 0b5e337729 (Tcl 8.6.0). The attached show.tcl gives identical results for all three cases in 8.6.0 through 9.1. Can be closed.
          ✅3075027fff20108.5 head testsuite failures on windowsFIXED since 8.6.6 (c96de4ed64, 2016-07-08, ticket a47641a031; merged to trunk 4589a9d17c)

          Open ticket 3075027fff · 36. Pathname Management · filed 2010-09-24

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: The last failures (filesystem-1.3/1.4) were fixed for 8.6.6 by c96de4ed64 (ticket a47641a031); the others were fixed already in 2011. 8.5 is no longer maintained. Closing.
          Full verdict

          FIXED since 8.6.6 (c96de4ed64, 2016-07-08, ticket a47641a031; merged to trunk 4589a9d17c)

          • Platform Windows
          • Tested code check only. The async/stack/tcltest failures were reported fixed in 2011; the remaining filesystem-1.3/1.4 were fixed by c96de4ed64 (TclJoinPath passed a non-absolute path to TclNewFSPathObj), whose check is still in main generic/tclPathObj.c:TclJoinPath. Not on 8.5 (EOL). Windows CI runs fileSystem.test.
          • Cause generic/tclPathObj.c:TclJoinPath cached a relative path as an FsPath "normalized" relative to the wrong base.
          • Prototype none (already fixed)
          • Suggested ticket comment The last failures (filesystem-1.3/1.4) were fixed for 8.6.6 by c96de4ed64 (ticket a47641a031); the others were already fixed in 2011. 8.5 is no longer maintained. Closing.
          ✅3069454fff2010difference between bcc'ed and eval'ed regexpFIXED on main and core-9-0-branch, unreleased (main 43345b959d 2026-10-15, 9.0 a478067f7a 2026-10-15; will be in 9.0.5); REPRODUCED (8.6 tip, 8.6.18)

          Open ticket 3069454fff · 43. Regexp · filed 2010-09-17 · Pending (resolved, waiting for confirmation)

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 🔧 2026-10-11 check-in 4d0c5cd25d on regexp-nocase-abbrev, in main: Rebase to trunk
          • 💬 2026-10-18 jan.nijtmans: Merged to 43345b959d and core-9-0-branch. Merge to 8.6 still to be done
          Full verdict

          FIXED on main and core-9-0-branch, unreleased (main 43345b959d 2026-10-15, 9.0 a478067f7a 2026-10-15; will be in 9.0.5); REPRODUCED (8.6 tip, 8.6.18)

          • Platform generic
          • Tested interpreted set c regexp; $c {*}$a with a={-n FoO dogfood}: main and 9.0 tip return 1, 8.6 tip and 8.6.18 give "bad option -n"; compiled form returns 1 everywhere. (Literal regexp {*}{...} is compiled, so it is not a valid interpreted test.)
          • Cause Tcl_RegexpObjCmd used TCL_EXACT; compiler accepted prefixes of -nocase. Jan's variant makes all regexp/regsub options abbreviatable.
          • Prototype none (fix merged)
          • Existing work regexp-nocase-abbrev (tip 4d0c5cd25d, Serhiy; superseded) -> jn-regexp-nocase-abbrev (e93dfd305e, Jan) merged into main 43345b959d and core-9-0-branch a478067f7a; not in core-8-6-branch
          • Pending resolution holds on main and 9.0. Can be closed once the 8.6 merge is done (Jan's 2026-10-18 comment: "Merge to 8.6 still to be done"), or now if 8.6 is to stay as is.
          • Side effect noticed on main/9.0 regexp - $str now says ambiguous option "-" instead of bad option "-" (consequence of making all options abbreviatable; harmless).
          • Suggested ticket comment Verified on main and 9.0 tip; 8.6 still rejects -n. Close after the 8.6 merge (or decide 8.6 keeps exact options).
          ✅3048353fff2010[interp limit ...] can dump coreFIXED since 8.5.11 (241cdd7be1, 2011-08-30)

          Open ticket 3048353fff · 20. [interp] · filed 2010-08-19

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed by b3333814a0 (8.5.11) and 241cdd7be1 (8.6.0), ticket [3398794]: interp limit {} ... now fails with "limits on current interpreter inaccessible" instead of panicking.
          Full verdict

          FIXED since 8.5.11 (241cdd7be1, 2011-08-30)

          • Platform generic
          • Tested interp limit {} commands -command pust -value 3 on main, 9.0.5, 8.6.18, 8.5.11, 8.5.19: error "limits on current interpreter inaccessible"; 8.5.9/8.5.10 panic "installing limit callback to the limited interpreter".
          • Cause tclInterp.c: the limit subcommand passed the current interp to SetScriptLimitCallback, which panics; fixed by "&#91;Bug 3398794&#93;: Use Tcl errors in scripts, not panics."
          • Prototype none (fixed)
          • Suggested ticket comment Fixed by 241cdd7be1 (Tcl 8.5.11): &#91;interp limit {} ...&#93; now raises "limits on current interpreter inaccessible" instead of panicking. Can be closed.
          ✅3023017fff2010win32 file isdirectory <dir>/* broken.FIXED since 8.6.1 (5567ed9fa5, 2013-03-19)

          Open ticket 3023017fff · 37. File System · filed 2010-06-29

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.6.1 by 5567ed9fa5 (ticket [3608360]): the FindFirstFile fallback, which expanded the *, is now used only for sharing violations. On Windows 11, 8.6.0 still returns 1 for file isdirectory dir/*; 8.6.1, 8.6.18, 9.0.5 and 9.1 return 0.
          Full verdict

          FIXED since 8.6.1 (5567ed9fa5, 2013-03-19)

          • Platform Windows
          • Tested code reading of main/9.0/8.6 win/tclWinFile.c. Linux is not affected (0 for dir/*). Windows repro: win/3023017fff.tcl.
          • Cause win/tclWinFile.c:NativeStat (and NativeAccess) fell back to FindFirstFileW, which expands wildcards, whenever GetFileAttributesEx failed (from the [2893771] change, 2009). Since 5567ed9fa5 the fallback is used only for ERROR_SHARING_VIOLATION, so dir/* (ERROR_INVALID_NAME) now fails as it should.
          • Prototype none (already fixed)
          • Suggested ticket comment Should be fixed since 8.6.1 by 5567ed9fa5 (ticket [3608360]): the FindFirstFile fallback is now used only for sharing violations. Closing; please reopen if file isdirectory dir/* still returns 1 on 8.6.1 or later.
          Windows VM, run 5
          • Platform Windows
          • Verdict FIXED since 8.6.1 (5567ed9fa5, 2013-03-19)
          • Tested VM win/3023017fff.tcl: 8.6.0 gives isdirectory(dir/*)=1, exists(dir/*)=1, exists(dir/C*)=1; 8.6.1, 8.6.18, 9.0.5, 9.1b1 (and 8.5.9, 8.5.19) all 0 0 0 0 0 1.
          • Cause win/tclWinFile.c NativeStat fell back to FindFirstFileW (wildcards) whenever GetFileAttributesEx failed; since 5567ed9fa5 only for sharing violations.
          • Prototype none (fixed)
          • Suggested ticket comment Fixed in 8.6.1 by 5567ed9fa5: on Windows 11 8.6.0 still returns 1 for &#91;file isdirectory dir/*&#93;, 8.6.1, 8.6.18, 9.0.5 and 9.1 return 0. Closing.
          ✅3009227fff2010PWD make variable in packages targetFIXED since 8.6.1 (0d1bc53480 "mSys doesn't have $PWD" + d433d96843 "$builddir is a local variable", both 2013-01-08, trunk; 8.6.1 tagged 2013-…

          Open ticket 3009227fff · 81. Bundled Packages · filed 2010-05-30

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.6.1 by 0d1bc53480 and d433d96843: the packages targets in win/Makefile.in pass --with-tcl=$builddir instead of the unset $(PWD).
          Full verdict

          FIXED since 8.6.1 (0d1bc53480 "mSys doesn't have $PWD" + d433d96843 "$builddir is a local variable", both 2013-01-08, trunk; 8.6.1 tagged 2013-09-19)

          • Platform build (win/Makefile.in, MSYS/mingw)
          • Tested history of win/Makefile.in: core-8-6-0 still has --with-tcl=$(PWD), core-8-6-1 and later (8.6 tip, 9.0, main) use --with-tcl=$$builddir, exactly the fix proposed in the ticket.
          • Cause win/Makefile.in packages target used $(PWD), unset under MSYS/CMD.
          • Prototype none (already fixed)
          • Suggested ticket comment Fixed in 8.6.1 by 0d1bc53480/d433d96843: the packages target now passes --with-tcl=$builddir. Can be closed.
          ✅2976509fff2010Bogus field reference in TtyInit() in unix/tclUnixChan.cFIXED since 8.6.1 (993c21721a, 2013-03-03, "retire obsolete termio and sgtty code", jenglish/joe)

          Open ticket 2976509fff · 27. Channel Types · filed 2010-03-25

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: The sgtty code, including this fragment, was removed by 993c21721a in 8.6.1.
          Full verdict

          FIXED since 8.6.1 (993c21721a, 2013-03-03, "retire obsolete termio and sgtty code", jenglish/joe)

          • Platform unix
          • Tested code reading: USE_SGTTY appears 9 times in core-8-6-0 unix/tclUnixChan.c, 0 in core-8-6-1 (2013-09-19), 8.6 tip and main. Still present in 8.5 (not maintained).
          • Cause the USE_SGTTY block referenced a nonexistent ttyPtr->stateUpdated; the whole sgtty branch was removed.
          • Prototype none (fixed)
          • Suggested ticket comment The sgtty code, including this fragment, was removed in 993c21721a (8.6.1). Closing.
          ⬜2976497fff2010dangerous assumptions in TclpCreateTempFileFIXED since 8.6.0 (2f3da59db1, 2012-11-14, ticket [2933003]; L_tmpnam definition removed by 9a008e7622, 2015-07-15). 8.5 still has the old code (EOL).

          Open ticket 2976497fff · 37. File System · filed 2010-03-25

          Full verdict

          FIXED since 8.6.0 (2f3da59db1, 2012-11-14, ticket [2933003]; L_tmpnam definition removed by 9a008e7622, 2015-07-15). 8.5 still has the old code (EOL).

          • Platform unix
          • Tested code reading: 8.6.0 tag, 8.6/9.0/main unix/tclUnixPipe.c + tclUnixFCmd.c.
          • Cause old unix/tclUnixPipe.c used char fileName[L_tmpnam + 9] filled with P_tmpdir. Now TclpCreateTempFile calls TclUnixOpenTemporaryFile, which builds the template in a Tcl_DString from DefaultTempDir() ($TMPDIR, then P_tmpdir only #ifdef P_tmpdir, then TCL_TEMPORARY_FILE_DIRECTORY).
          • Prototype none (already fixed)
          • Suggested ticket comment Fixed since 8.6.0 by 2f3da59db1: temp file creation moved to TclUnixOpenTemporaryFile, which uses a Tcl_DString and only uses P_tmpdir when it is defined. Closing.
          ✅2974481fff2010FreeBSD build failure -ltcl86FIXED since 8.6.12 (75f8d23bb7, 2021-11-01); earlier releases already matched file name and -l flag

          Open ticket 2974481fff · 53. Configuration and Build Tools · filed 2010-03-22

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Not a problem any more: on FreeBSD 15.1, 8.6 (core-8-6-branch) and 9.1 (main) configure, build with gmake and install without changes, producing libtcl8.6.so / libtcl9.1.so with a matching TCL_LIB_FLAG (-ltcl8.6 / -ltcl9.1), and the installed tclsh runs. The FreeBSD Ports-specific configuration that produced libtcl86.so.1 has been removed since (75f8d23bb7). Closing.
          Full verdict

          FIXED since 8.6.12 (75f8d23bb7, 2021-11-01); earlier releases already matched file name and -l flag

          • Platform build (FreeBSD)
          • Tested tcl.m4 at release tags: core-8-6-11 still has the "FreeBSD Ports" branch with SHARED_LIB_SUFFIX='${TCL_TRIM_DOTS}.so' (libtcl86.so, matching -ltcl86, no .so.1 mismatch). From core-8-6-12 on, and on 9.0/main, FreeBSD-* uses the common Linux branch (libtcl8.6.so / libtcl9.x.so, -ltcl8.6). Not built on a FreeBSD VM.
          • Cause the 2010 CVS snapshot's FreeBSD-ports config produced libtcl86.so.1 while linking with -ltcl86 and no unversioned symlink. That configuration was dropped in 75f8d23bb7 "Update DragonFly/FreeBSD configuration".
          • Prototype none (fixed)
          • Suggested ticket comment The FreeBSD Ports-style library naming was removed in 75f8d23bb7 (8.6.12). FreeBSD now builds libtcl8.6.so like Linux. Closing as out of date.

          (Optional check on the FreeBSD VM: ./configure && make && ./tclsh in an 8.6/9.0 tree.)

          ✅2957998fff2010Packages tdbc and itcl not built whenFIXED since 8.6.1 (same fix as 3009227fff: 0d1bc53480 + d433d96843, 2013-01-08)

          Open ticket 2957998fff · 81. Bundled Packages · filed 2010-02-24

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Same cause as 3009227fff: $(PWD) is unset under MSYS, so the packages picked up the system Tcl 8.4. Fixed in 8.6.1 by 0d1bc53480; the packages target now passes --with-tcl=$builddir.
          Full verdict

          FIXED since 8.6.1 (same fix as 3009227fff: 0d1bc53480 + d433d96843, 2013-01-08)

          • Platform build (win/Makefile.in, MSYS)
          • Tested history of win/Makefile.in. The log shows configure of the packages falling back to a system C:/Tcl (8.4) tclConfig.sh because --with-tcl=$(PWD) expanded to an empty/wrong dir under MSYS; since 8.6.1 the packages target passes --with-tcl=$$builddir and --with-tclinclude=$(GENERIC_DIR_NATIVE).
          • Cause win/Makefile.in packages target, $(PWD) unset in MSYS.
          • Prototype none (already fixed)
          • Suggested ticket comment Same cause as 3009227fff ($(PWD) unset under MSYS, so the packages found the system Tcl 8.4); fixed in 8.6.1 by 0d1bc53480. Can be closed.
          ✅2953593fff2010[gets $fd data] returns 0 in nonblocking modepatchFIXED since 8.6.13 / 9.0.0 (a3ed092aab, 2022-07-27, the fix for duplicate b3977d199b): torn CRLF in auto mode gave an extra empty line (gets re…

          Open ticket 2953593fff · 24. Channel Commands · filed 2010-02-17

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-16 serhiy.storchaka: Not reproduced: CRLF, LF and CR input split at random points gives no unexpected empty lines (LF CR input gives an empty line for each CR in auto mode, as expected). Testing this found a regression in 8.6.13, reported as 4c5a98f5a0.
          • 💬 2026-10-16 serhiy.storchaka: Correction: this is the torn CRLF case of b3977d199b (CR and LF in different reads give an extra empty line in auto mode). It reproduces in 8.4.20 to 8.6.12 and was fixed in 8.6.13 by a3ed092aab. That check-in also caused 4c5a98f5a0.
          Full verdict

          FIXED since 8.6.13 / 9.0.0 (a3ed092aab, 2022-07-27, the fix for duplicate b3977d199b): torn CRLF in auto mode gave an extra empty line (gets returns 0) in 8.4.20–8.6.12 (28 of 328 lines in a random-split test). The same check-in caused a new regression, filed as 4c5a98f5a0 (gets merges lines). Update 10-17: first triaged as NOT-REPRODUCED because only current versions were tested.

          • Platform generic
          • Tested (10-17): CRLF lines split at random points over a non-blocking pipe, -translation auto: 8.4.20, 8.5.0, 8.5.8, 8.5.9, 8.5.19, 8.6.0, 8.6.12 give 28 empty lines in 328; 8.6.13, 8.6.18, 9.0.5, main give none. -translation crlf, CR-only and LF-only input are clean everywhere.
          • Tested nonblocking pipe + fileevent with 10-char lines ended by \r\n, \n, \r, split at random points (300 lines) with -translation auto/lf/crlf/cr: no empty reads apart from the expected ones (\n\r input in auto mode gives an empty line per \r, which would be one explanation for the report if the device sent LF CR). While doing that: input "a\r" | "b" | "\nc\n" in separate reads gives {a bc} instead of {a b c} (blocking and nonblocking); with LF CR input some empty lines vanish. Old releases: 8.5.0–8.5.19 and 8.6.0–8.6.12 correct; 8.6.13, 8.6.14, 8.6.18, 9.0.5, main wrong.
          • Cause a3ed092aab (sebres, 2022-07-27, fix for b3977d199b, in 8.6.13) moved the reset of INPUT_SAW_CR in Tcl_GetsObj (generic/tclIO.c) so it happens only when an end of line is found. If the read after a trailing "\r" has no end of line, the flag stays set, and a later "\n" that starts a new buffer is dropped as if it were the LF of a CRLF pair.
          • Prototype prototypes/2953593fff.patch (against main; 39 changed lines, test io-29.36.3). Clears the flag as soon as any data follows the "\r", and sets it again in the restore: path when a skipped "\n" is pushed back (so the b3977d199b case, io-29.36.1/2, keeps working). Fixes all repros; io, chanio, ioCmd, ioTrans, encoding, zlib tests pass; io-29.36.3 fails without the fix. For 8.6 the bool needs to become int.
          • Suggested ticket comment The CRLF case does not give empty lines here. But a related regression from a3ed092aab (8.6.13): with -translation auto, "a\r", then "b", then "\nc\n" arriving in separate reads gives the lines "a" and "bc", because INPUT_SAW_CR is not cleared when the next read has no end of line. Proposed fix in Tcl_GetsObj, with test io-29.36.3.
          ✅2952904fff2010"floating point assist fault" on IA64 during Tcl startupFIXED since 8.6b2 (0ff9783605 on trunk, 2010-04-02, plus the follow-up for [2996549]; also in the 8.5 branch source, merge release not checked); IA…

          Open ticket 2952904fff · 48. Number Handling · filed 2010-02-16

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed by 0ff9783605 and the follow-up 0958a5f61a (8.6.0; backported to 8.5.10): the smallest denormal is no longer computed at startup. Present in all current branches.
          Full verdict

          FIXED since 8.6b2 (0ff9783605 on trunk, 2010-04-02, plus the follow-up for [2996549]; also in the 8.5 branch source, merge release not checked); IA64 itself CANNOT-TEST

          • Platform unix (Linux/IA64)
          • Tested code reading only: main, 9.0, 8.6 and 8.5 tclStrToD.c all have static double tiny = 0.0 with lazy initialization in MakeLowPrecisionDouble (the ticket's 2952904_2.patch), so nothing runs SafeLdExp(denormal) at startup any more.
          • Cause TclInitDoubleConversion computed the smallest denormal at startup, which IA64 does via a software assist trap.
          • Prototype none (already fixed)
          • Suggested ticket comment The deferred computation of tiny (0ff9783605, 2010-04-02) is in all branches (8.5, 8.6, 9.0, 9.1). IA64 hardware is gone; closing as fixed.
          ⬜2924753fff2010documentation joinable threadspatchFIXED since 9.0.3 (9183ace241, 2025-09-26, merged to main 15f6cb484d); still wrong on core-8-6-branch

          Open ticket 2924753fff · 49. Threading · filed 2010-01-02

          🧪 Prototype check (2026-10-24)

          2924753fff.patch: VERIFIED

          • Base core-8-6-branch (tip 326e1bf923)
          • Before doc/Thread.3:108 on the 8.6 tip still says "Windows currently does not support joinable threads", although win/tclWinThrd.c:240 honours TCL_THREAD_JOINABLE. main/9.0 lost the sentence in 9183ace241.
          • Patch applied cleanly (2 removed lines, same as 9183ace241)
          • After groff renders, no new warnings
          • Test none possible (docs)
          Full verdict

          FIXED since 9.0.3 (9183ace241, 2025-09-26, merged to main 15f6cb484d); still wrong on core-8-6-branch

          • Platform docs
          • Tested doc/Thread.3 on main/9.0 no longer has the sentence; 8.6 tip still says "Windows currently does not support joinable threads", although win/tclWinThrd.c (8.6) honours TCL_THREAD_JOINABLE.
          • Cause stale restriction in doc/Thread.3.
          • Prototype prototypes/2924753fff.patch (core-8-6-branch, 2 changed lines: cherry-pick of 9183ace241)
          • Suggested ticket comment The obsolete restriction was removed from Thread.3 in 9183ace241 (9.0.3 and main). Proposed: cherry-pick the same 2-line removal to core-8-6-branch, then close.
          ⬜2912880fff2009Win Debug Builds Fails if stack-check disabled.FIXED since 8.6.0 (the C stack check code, TCL_NO_STACK_CHECK, CheckCStack and the localInt variable in TclInterpReady, no longer exist in 8.6/9.…

          Open ticket 2912880fff · 41. Memory Allocation · filed 2009-12-11

          Full verdict

          FIXED since 8.6.0 (the C stack check code, TCL_NO_STACK_CHECK, CheckCStack and the localInt variable in TclInterpReady, no longer exist in 8.6/9.0/main); still present on the 8.5 branch, which is EOL

          • Platform build (8.5 only)
          • Tested grep of generic/ and win/ in 8.6, 9.0, main: no TCL_NO_STACK_CHECK/localInt; 8.5 tclBasic.c:3544 still has the unused localInt when TCL_NO_STACK_CHECK is defined (CheckCStack expands to 1). Not compiled on Windows.
          • Cause generic/tclBasic.c TclInterpReady (8.5): localInt only used through CheckCStack, which is a no-op with TCL_NO_STACK_CHECK.
          • Prototype none (attached patch fits 8.5; 8.6+ not affected)
          • Suggested ticket comment Stack checking was removed in 8.6 (NRE), so 8.6/9.x are not affected; 8.5 is no longer maintained. Closing as out of date.
          ✅2910867fff2009[tclPkgUnknown] and Safe BaseFIXED since 8.6.11 (a638b9a73b "Merge safe-bugfixes-8-6", kjnash, 2020-08-28; 8.6.11 tagged 2020-12-30)

          Open ticket 2910867fff · 38. Init - Library - Autoload · filed 2009-12-08

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.6.11 by a638b9a73b: the Safe Base source alias reports missing or unreadable files with -errorcode {POSIX EACCES}, which tclPkgUnknown ignores, so the spurious "no such file" log messages are gone. Index files that fail for other reasons are still logged, as intended.
          Full verdict

          FIXED since 8.6.11 (a638b9a73b "Merge safe-bugfixes-8-6", kjnash, 2020-08-28; 8.6.11 tagged 2020-12-30)

          • Platform generic
          • Tested a Safe Base interp (safe::interpCreate) with tclLog redefined to collect messages, then [package require nosuch]: 8.5.19 and 8.6.10 log 13-18 "error reading package index file $p(:N:)/pkgIndex.tcl: no such file or directory / permission denied"; 8.6.11, 8.6.18, 9.0.5 and main log none of those. Since that merge safe.tcl AliasSource returns -errorcode {POSIX EACCES} for missing/unreadable files, which tclPkgUnknown silently skips.
          • Remaining (different issues): real index files that are not safe-compatible are still logged: 8.6/9.0 Tk's pkgIndex.tcl ("not allowed to invoke subcommand normalize of file"), main the zipfs registry/pkgIndex.tcl ("can't read ::tcl_platform(os)"). Also in 9.x a custom -accessPath without -autoPath gives 'invalid command name "::tclPkgUnknown"', but safe.n documents that -autoPath must be given then.
          • Cause library/package.tcl tclPkgUnknown sources $dir/pkgIndex.tcl without [file exists] in safe interps; fixed on the Safe Base side (library/safe.tcl AliasSource).
          • Prototype none (already fixed)
          • Suggested ticket comment Since 8.6.11 (a638b9a73b) the Safe Base's source alias reports missing/unreadable files with -errorcode {POSIX EACCES}, which tclPkgUnknown ignores silently, so the spurious "no such file" tclLog messages are gone (checked 8.6.18, 9.0, main). Can be closed.
          ⬜2895741fff2009expr min and max don't work in a safe interpFIXED since 8.5.9 / 8.6b2 (e6c8bcd32e on core-8-5-branch and 40561c3f87 on trunk, 2009-12-29)

          Open ticket 2895741fff · 20. [interp] · filed 2009-11-11

          Full verdict

          FIXED since 8.5.9 / 8.6b2 (e6c8bcd32e on core-8-5-branch and 40561c3f87 on trunk, 2009-12-29)

          • Platform generic
          • Tested main (own build), 9.0.5, 8.6.18, 8.5.19, 8.5.9 OK (expr {min(3,1,2)} in interp create -safe gives 1); 8.5.7, 8.5.8 fail with invalid command name "tcl::mathfunc::min"
          • Cause min/max were procs defined in init.tcl, which safe interps don't source; dkf made them C commands in 2009.
          • Prototype none (already fixed)
          • Suggested ticket comment min() and max() work in safe interps since 8.5.9 (e6c8bcd32e/40561c3f87). dkf kept the ticket open (priority lowered) for the wider question of what safe interp creation should set up; that is RFE material, so this ticket can be closed.
          ✅2889257fff2009[array startsearch] and friends do not work on envFIXED since 8.6.2 (afaff33553, 2013-09-27, "[219226]: Rewrote how ::env is synchronized ...")

          Open ticket 2889257fff · 08. Environment Variables · filed 2009-10-30

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Duplicate)

          • 💬 2026-10-21 serhiy.storchaka: Duplicate of 1787911fff. Array searches on env work since 8.6.2, where afaff33553 stopped env from being recreated on array access.
          Full verdict

          FIXED since 8.6.2 (afaff33553, 2013-09-27, "[219226]: Rewrote how ::env is synchronized ...")

          • Platform generic
          • Tested main, 9.0.5, 8.6.2-8.6.18 OK (array nextelement env $s returns a name); 8.6.0, 8.6.1, 8.5.19 fail with couldn't find search "s-1-env"
          • Cause EnvTraceProc used to rebuild the whole env array on every array trace, which removed the searches; the 2013 rewrite syncs it in place.
          • Prototype none (already fixed; 8.5 is EOL)
          • Suggested ticket comment Array searches on env work since 8.6.2 (rewrite of env syncing in afaff33553, ticket [219226]). Can be closed.
          ⬜2869384fff2009Tcl_AsyncMark does not work with --enable-threadsFIXED since 8.7b1/9.0 via a new API (TIP 511 Tcl_AsyncMarkFromSignal, 1a6b389968, 2021-08-20); still present in 8.6 (no fix possible there withou…

          Open ticket 2869384fff · 04. Async Events · filed 2009-09-29

          Full verdict

          FIXED since 8.7b1/9.0 via a new API (TIP 511 Tcl_AsyncMarkFromSignal, 1a6b389968, 2021-08-20); still present in 8.6 (no fix possible there without the new API)

          • Platform unix
          • Tested code reading main/9.0/8.6 (repro needs a signal-driven wish loop; not run)
          • Cause generic/tclAsync.c:Tcl_AsyncMark takes asyncMutex and calls Tcl_ThreadAlert, which takes the notifier mutex: not async-signal-safe. Tcl_AsyncMark is unchanged on main, but Async.3 now tells signal handlers to call Tcl_AsyncMarkFromSignal, which goes through TclAsyncNotifier (unix/tcl{Epoll,Kqueue,Select}Notfy.c: no mutex, only a flag plus async-signal-safe wakeup calls).
          • Prototype none (resolved by TIP 511)
          • Suggested ticket comment Signal handlers should call Tcl_AsyncMarkFromSignal() (TIP 511, Tcl 8.7/9.0), which is async-signal-safe; Tcl_AsyncMark() remains for non-signal contexts, as Async.3 documents. Can be closed as fixed in 9.0.
          🔧2788582fff2009[binary] ugly formatSpec parsingpatchFIXED since 8.6.7 for [binary scan] (56c454c814 core-8-6-branch / 5c07ae32cf trunk, 2017-05-05, bug 6015221f59); REPRODUCED (main, 9.0) as a pani…

          Open ticket 2788582fff · 12. ByteArray Object · filed 2009-05-07

          🔧 Since triage (updated 2026-11-11)

          Status now: Open

          • 🔧 2026-10-17 check-in dd5cd86e81 on binary-format-size-overflow, not merged: Fix 2788582fff and ce4483c2cb: [binary format] with huge counts or offsets panicked or returned a wrong result, because the first pass added the field sizes
          • 💬 2026-10-17 serhiy.storchaka: The [binary scan] part is fixed since 8.6.7 (56c454c814). [binary format] still adds the field sizes without checking for overflow: [binary format x99999999999999999999] and [binary format @9223372036854775807] panic in 9.0 and 9.1, and in 8.6 [binary format x2147483647x2147483647] returns an empty string. Proposed fix in branch binary-format-size-overflow, with ce4483c2cb: report "max size for a Tcl value exc…
          Full verdict

          FIXED since 8.6.7 for [binary scan] (56c454c814 core-8-6-branch / 5c07ae32cf trunk, 2017-05-05, bug 6015221f59); REPRODUCED (main, 9.0) as a panic in [binary format], 8.6 gives silently wrong results

          • Platform generic
          • Tested main build, 9.0.5, 8.6.18, 8.6.0–8.6.14, 8.5.7/8.5.8/8.5.19. The ticket's scan cases: 8.5.x and 8.6.0–8.6.6 show the wrap-around (a0x80000000 matches an empty string), 8.5.7/8.5.8 segfault on one of them, 8.6.7+ and 9.x return 0 for all huge counts. But on main and 9.0.5 binary format a99999999999999999999 A, x9223372036854775807, x9223372036854775000x9000, @9223372036854775807 abort with Tcl_Panic "negative length specified or max size of a Tcl value exceeded". On 8.6.18 x2147483647x2147483647 and @2147483647x1 overflow int and silently return an empty string.
          • Cause generic/tclBinary.c BinaryFormatCmd: GetFormatSpec clamps the count to TCL_SIZE_MAX, but the first pass adds counts to offset without checking for overflow or BYTEARRAY_MAX_LEN, so Tcl_SetByteArrayLength gets a negative/oversized length and panics (BYTEARRAY_SIZE).
          • Prototype prototypes/2788582fff.patch (48 changed lines, tests binary-81.1..81.5). ADD_OFFSET macro checks each addition against BYTEARRAY_MAX_LEN, plus a final length check, and returns "max size for a Tcl value (N bytes) exceeded" with errorCode TCL MEMORY (same text as [binary encode base64]). binary.test passes; the tests panic the 9.0 tip. Counts that fit but cannot be allocated still panic in the allocator, as usual in Tcl.
          • Suggested ticket comment The [binary scan] side was fixed in 8.6.7 (56c454c814). [binary format] with a huge count or @offset still panics in 9.0/9.1 (offset overflow before Tcl_SetByteArrayLength). Proposed fix: check for overflow while computing the length, tests binary-81.*.
          ⬜2613599fff2009Tcl_DStrings lack length overflow protectionsFIXED since 9.0.0 (18d4bd49a6, apnadkarni, 2023-05-20, "Fix c9663296fd. Also refactor memory reallocation."). 8.6 still has no checks (by code …

          Open ticket 2613599fff · 10. Objects · filed 2009-02-18

          Full verdict

          FIXED since 9.0.0 (18d4bd49a6, apnadkarni, 2023-05-20, "Fix c9663296fd. Also refactor memory reallocation."). 8.6 still has no checks (by code reading).

          • Platform generic
          • Tested code reading. main/9.0 Tcl_DStringAppend panics before TCL_SIZE_MAX overflow, and Tcl_DStringSetLength panics at TCL_SIZE_MAX; growth uses TclAllocEx/TclReallocEx/TclUpsizeAlloc. 8.6 uses int and newSize * 2 with no check. A 1.7 GB encoding convertfrom iso8859-1 on 8.6.18 worked, so no script-level trigger was found. Tcl_DStringAppendElement on main has no explicit sum check, but it cannot be reached with a 64-bit Tcl_Size.
          • Cause 8.6 generic/tclUtil.c Tcl_DStringAppend/AppendElement/SetLength double the int size without an overflow check.
          • Prototype none (fixed in 9.x; for 8.6, dgp worried about changing the negative-length behaviour, which 9.x still has)
          • Suggested ticket comment 9.0 has overflow checks in the DString functions (18d4bd49a6) and a 64-bit Tcl_Size. 8.6 still doubles an int without checks, but no script-level trigger is known. Suggest closing as fixed in 9.0.
          ✅2564363fff2009string equality testing does not cover encoding variantsFIXED since 8.6.7 (164b69bdfd / 8e28fedaa0, 2017-06-05, follow-up to 67aa9a2070); main/9.x consistent (the identity encoding is gone in 9.0)

          Open ticket 2564363fff · 10. Objects · filed 2009-02-04

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.6.7 by 164b69bdfd (9.0: 8e28fedaa0) for 67aa9a2070: an overlong sequence no longer decodes to the same character, so eq, string equal and the non-compiled path agree before and after shimmering. 9.0 also dropped the identity encoding.
          Full verdict

          FIXED since 8.6.7 (164b69bdfd / 8e28fedaa0, 2017-06-05, follow-up to 67aa9a2070); main/9.x consistent (the identity encoding is gone in 9.0)

          • Platform generic
          • Tested \xc0\xa1 built with testbytestring (9.x/8.6 tcltest) or encoding convertfrom identity (8.x). main, 9.0 tip, 8.6 tip, 8.6.7, 8.6.18: eq / string equal / non-compiled $s equal all return 0, before and after string length (the overlong sequence now counts as 2 characters). 8.6.6: 0 before and 1 after shimmering. 8.5.18/8.5.19 releases: inconsistent (expr eq 0, string equal 1).
          • Cause (historical) Tcl_UtfToUniChar decoded the overlong \xc0\xa1 to "!", so the String intrep compared equal while the bytes did not. Since 67aa9a2070 invalid bytes decode as single bytes.
          • Prototype none (already fixed)
          • Suggested ticket comment Fixed by the 67aa9a2070 work in 8.6.7: an overlong sequence no longer decodes to the same character, so all comparison paths agree (and 9.0 removed the identity encoding). Can be closed.
          ✅2544740fff2009[interp create] O(N^2) search for unused cmd nameFIXED since 9.0.0 (7a665fa1b6, 2024-08-15, dkf "Minor optimisation in the [interp create] implementation")

          Open ticket 2544740fff · 20. [interp] · filed 2009-01-29

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 9.0.0 by 7a665fa1b6: a per-parent counter replaces the search from interp0. 8.6 keeps the old loop.
          Full verdict

          FIXED since 9.0.0 (7a665fa1b6, 2024-08-15, dkf "Minor optimisation in the [interp create] implementation")

          • Platform generic
          • Tested code reading main/9.0/8.6; timing 1000 vs 4000 [interp create] on main, 9.0.5, 8.6.18 (creation cost of ~0.7 ms per interp dominates; quadratic part not measurable at these sizes)
          • Cause tclInterp.c:Tcl_InterpObjCmd OPT_CREATE probed "interp0", "interp1", ... from 0 every time; main/9.0 now keep a per-parent counter idIssuer. 8.6 still has the loop from 0 (names are reused after delete there; 9.0 no longer reuses them).
          • Prototype none (already fixed in 9.0; 8.6 backport would change name reuse, not worth it)
          • Suggested ticket comment Fixed in 9.0.0 by 7a665fa1b6 (a per-parent counter <tt>idIssuer</tt> replaces the search from 0). 8.6 is unchanged. Can be closed.
          ⬜2502037fff2009NRE-enable [namespace unknown] dispatchpatchFIXED since 8.6.0 (f45dcb7151, 2009-03-22, msofer "NR-enable the handling of unknown commands [Bug 2502037]")

          Open ticket 2502037fff · 45. Parsing and Eval · filed 2009-01-12

          🧪 Prototype check (2026-10-24)

          2502037fff.patch: VERIFIED (test-only patch; the bug was already FIXED in 8.6.0)

          • Base main (tip d6dd6d0f71)
          • Before does not reproduce on any version with coroutines: the fix f45dcb7151 (2009-03-22) predates 8.6.0, which the ticket only lacked a test for
          • Patch applied cleanly (13 changed lines, test only)
          • After coroutine.test 81/0 failed, coroutine-7.20 passes
          • Test existing coroutine-7.20, regression test only; it cannot be made to fail on a current branch (the fix predates coroutines in any release)
          Full verdict

          FIXED since 8.6.0 (f45dcb7151, 2009-03-22, msofer "NR-enable the handling of unknown commands [Bug 2502037]")

          • Platform generic
          • Tested main, 8.6.18, 8.6.0: [yield] inside a [namespace unknown] handler works; 20000-deep recursion through namespace unknown works on main and 8.6.1+ (8.6.0 hits "too many nested evaluations")
          • Cause tclBasic.c:TEOV_NotFound now dispatches the handler via TclNRAddCallback + TclNREvalObjv. No test was ever added (dgp asked for one).
          • Prototype prototypes/2502037fff.patch (13 changed lines, test coroutine-7.20; test only, passes on main, 9.0 and 8.6)
          • Suggested ticket comment NR-enabled by f45dcb7151 in 2009; &#91;yield&#93; inside a &#91;namespace unknown&#93; handler works in 8.6 and 9.x. Proposed regression test coroutine-7.20; the ticket can be closed after that.
          ⬜2490456fff2009stringObj-10.2 misguided?FIXED since 8.6.0 for the test itself (79358a4faf, 2009-01-06, dgp: values built with [testdstring], so the parser is no longer involved); the st…

          Open ticket 2490456fff · 10. Objects · filed 2009-01-06

          Full verdict

          FIXED since 8.6.0 for the test itself (79358a4faf, 2009-01-06, dgp: values built with [testdstring], so the parser is no longer involved); the stale comment was dropped in 9.0 (gone in 9.0 and main, still present in core-8-6-branch)

          • Platform generic (tests)
          • Tested code reading of tests/stringObj.test on main, 9.0, 8.6
          • Cause n/a
          • Prototype none (only a stale 6-line comment remains in 8.6)
          • Suggested ticket comment stringObj-10.2 builds its value with &#91;testdstring&#93; since 79358a4faf, so it no longer tests the parser; the misleading comment is gone in 9.0. Can be closed.
          ⬜2461756fff2008[chan truncate] cannot work on reflected channelsFIXED since 8.7b1/9.0 (4aa6559799, 2021-03-30, RFE 4b4830eb54 (no TIP): refchan truncate method); 8.6 still rejects the method (feature, not …

          Open ticket 2461756fff · 27. Channel Types · filed 2008-12-23

          Full verdict

          FIXED since 8.7b1/9.0 (4aa6559799, 2021-03-30, RFE 4b4830eb54 (no TIP): refchan truncate method); 8.6 still rejects the method (feature, not backportable)

          • Platform generic
          • Tested main (own build), 9.0.5: refchan with a truncate method, [chan truncate $c 3] calls it; 8.6.18: "bad method truncate"
          • Cause tclIORChan.c had no truncateProc before that change.
          • Prototype none (already done)
          • Suggested ticket comment Since Tcl 8.7/9.0 reflected channels support an optional truncate method (RFE 4b4830eb54, check-in 4aa6559799), documented in refchan.n. Can be closed as fixed in 9.0.
          ✅2406596fff2008file delete reports an error on non-existent fileFIXED since 8.6.9 (5acb57c7ae, 2018-04-06, ticket 27b6822849; merged to trunk as 6fe50c20de)

          Open ticket 2406596fff · 37. File System · filed 2008-12-08

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.6.9 by 5acb57c7ae (ticket 27b6822849): file delete now ignores ENOENT when the file vanishes between lstat() and unlink(), in the generic layer as suggested here. The two-process script fails with 8.6.8 and passes with 8.6.9 and later.
          Full verdict

          FIXED since 8.6.9 (5acb57c7ae, 2018-04-06, ticket 27b6822849; merged to trunk as 6fe50c20de)

          • Platform generic
          • Tested Linux, reporter's two-process race, 3 s x 3 pairs per version: 8.6.8 fails every time ("no such file or directory"); 8.6.9, 8.6.18, 9.0.5 and main never fail.
          • Cause generic/tclFCmd.c:TclFileDeleteCmd ran lstat, then unlink, and reported ENOENT when the file vanished in between. Now ENOENT after a failed delete is ignored in the generic layer (as ferrieux suggested).
          • Prototype none (already fixed)
          • Suggested ticket comment Fixed since 8.6.9 by 5acb57c7ae (ticket 27b6822849): TclFileDeleteCmd now ignores ENOENT when the file disappears after the lstat. The reporter's two-process script fails on 8.6.8 and passes on 8.6.9+. Closing.
          ✅2166445fff2008filename-14.22 depends on cwdFIXED since 8.6.0 (03878f7b0c, 2012-10-05, ticket [3574819]); not on 8.5 (EOL)

          Open ticket 2166445fff · 36. Pathname Management · filed 2008-10-14

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.6.0 by 03878f7b0c (ticket [3574819]): the glob tests in fileName.test now run in a freshly created directory, so files in the current directory no longer matter.
          Full verdict

          FIXED since 8.6.0 (03878f7b0c, 2012-10-05, ticket [3574819]); not on 8.5 (EOL)

          • Platform generic (test suite)
          • Tested Linux main: fileName.test passes, and filename-14.22 also passes when tcltest is started in a directory with xyz, fooaq and goo/a.
          • Cause the glob tests ran in the caller's cwd. Since 03878f7b0c tests/fileName.test does cd [makeDirectory tcl[pid]] before the glob tests, so only files the test creates are present.
          • Prototype none (already fixed)
          • Suggested ticket comment Fixed for 8.6.0 by 03878f7b0c (ticket [3574819]): the glob tests now run in a fresh temporary directory. Closing.
          ⬜2114165fff2008eval after script cancellation does not workpatchFIXED since 8.6.0 (msofer in HEAD, 2008-09-17; resolution already "Fixed", left open only as a reminder to add tests)

          Open ticket 2114165fff · 45. Parsing and Eval · filed 2008-09-16

          🧪 Prototype check (2026-10-24)

          2114165fff.patch: FIXED-MEANWHILE (2008-09-17, msofer, before 8.6.0; test-only patch verified)

          • Base main (tip d6dd6d0f71)
          • Before does not reproduce on the tip: interp eval $i {interp cancel; while 1 {}} -> eval canceled, then interp eval $i {set x 1} -> 1 (and the same with -unwind). Fixed by msofer in HEAD on 2008-09-17 (before 8.6.0); the ticket was left open only as a reminder to add tests, and the patch is test-only.
          • Patch applied with offset +20 (re-saved, 17 changed lines, content unchanged)
          • After interp.test: 393 total, 386 passed, 7 skipped, 0 failed
          • Test existing interp-41.1/41.2 pass on the tip. Mutation check: with the six TclUnsetCancelFlags(iPtr) calls in tclBasic.c disabled, interp-41.2 fails (eval unwound) and interp-41.1 still passes, so the tests do catch a regression. NOTE: fb9acd64a5.patch also adds an interp-41.1; renumbered there (see below).
          Full verdict

          FIXED since 8.6.0 (msofer in HEAD, 2008-09-17; resolution already "Fixed", left open only as a reminder to add tests)

          • Platform generic
          • Tested main (own build), 9.0.5, 8.6.18: after interp cancel (and -unwind) in a child, the next interp eval works; same in the current interp and through Tcl_EvalEx ([testevalex]). The tclParse.c ordering issue mistachkin mentioned is gone (TclEvalEx lives in tclBasic.c and resets cancellation when numLevels returns to 0).
          • Cause n/a
          • Prototype prototypes/2114165fff.patch (17 changed lines, tests interp-41.1 and interp-41.2; test only, pass on main, 9.0 and 8.6)
          • Suggested ticket comment Works in 8.6, 9.0 and 9.1. The only existing cancellation tests are thread-7.* (Thread package); proposed tests interp-41.1/41.2 check that eval works again after &#91;interp cancel&#93; and &#91;interp cancel -unwind&#93;. Can be closed when they are merged.
          ⬜2030670fff2008Tcl_Panic in Tcl_ScanObjCmd after TclStackReallocpatchFIXED since 8.5.4 (63fdc405dd trunk / c547a0a022 core-8-5-branch, 2008-07-29; 8.5.4 tagged 2008-08-14); resolution already Fixed, left open as …

          Open ticket 2030670fff · 47. Bytecode Compiler · filed 2008-07-28

          🧪 Prototype check (2026-10-24)

          2030670fff.patch: VERIFIED (test-only patch; the bug itself was already FIXED in 8.5.4)

          • Base main (tip d6dd6d0f71)
          • Before scan 1 {%100000$d} -> 8.5.3: TclStackFree: incorrect freePtr. Call out of sequence? (panic); 8.5.4 and main tip: 100000 1. The bug was fixed in 2008 (63fdc405dd); the ticket only asks for a test
          • Patch applied cleanly (5 changed lines, test only)
          • After scan.test 185/0 failed (scan-16.1 passes)
          • Test existing scan-16.1, regression test only; it cannot fail on any current branch (fails/panics only on <= 8.5.3)
          Full verdict

          FIXED since 8.5.4 (63fdc405dd trunk / c547a0a022 core-8-5-branch, 2008-07-29; 8.5.4 tagged 2008-08-14); resolution already Fixed, left open as a reminder to add a test

          • Platform generic
          • Tested scan 1 {%100000$d} panics with "TclStackFree: incorrect freePtr. Call out of sequence?" on 8.5.0 and 8.5.3; works on 8.5.4, 8.6.18, 9.0.5 and main
          • Cause tclExecute.c:GrowEvaluationStack reused the move argument as a word count, so a zero-word block that needed to move to a new stack segment kept its old marker. Fixed by the separate moveWords variable. tclScan.c:Tcl_ScanObjCmd triggers it: with no variables nassign is a 0-byte TclStackAlloc, and a large XPG index makes TclStackRealloc grow it beyond the segment.
          • Prototype prototypes/2030670fff.patch (5 changed lines, test scan-16.1; test only, fails (panic) on 8.5.3, passes on 8.5.4, 8.6, 9.0 and main)
          • Suggested ticket comment Here is a simple script that triggers it: &#91;scan 1 {%100000$d}&#93; panics on 8.5.3 and works since 8.5.4 (63fdc405dd). Proposed regression test scan-16.1; the ticket can be closed after that.
          ✅1997860fff2008Tcl_GetInterpPath : minor bug in result affectationFIXED since 8.6.0 (b074e043df, 2012-07-31)

          Open ticket 1997860fff · 20. [interp] · filed 2008-06-19

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.6.0 by b074e043df: Tcl_GetInterpPath now resets the result before building the path.
          Full verdict

          FIXED since 8.6.0 (b074e043df, 2012-07-31)

          • Platform generic
          • Tested C harness (dlopen) on 8.5.0, 8.5.19 (repeated call gives "kid kid"), 8.6.0, 8.6.10, 8.6.18 (gives "kid"); main/9.0 by reading the code (same Tcl_SetObjResult(interp, Tcl_NewObj()) as 8.6)
          • Cause generic/tclInterp.c:Tcl_GetInterpPath did not reset the result when target==interp, so the elements were appended to the old result. dkf's "Purge use of Tcl_AppendElement" check-in made it set an empty result first. 8.5 still has the bug (EOL).
          • Prototype none (already fixed)
          • Suggested ticket comment Fixed in 8.6.0 by b074e043df (Tcl_GetInterpPath now resets the result first); 8.5 is not maintained. Can be closed.
          ✅1905560fff2008--enable-symbols 'make test' fails on MSVC2005FIXED since 8.5.3 (8d5cc700eb core-8-5-branch / 68b47f3256 trunk, 2008-04-06)

          Open ticket 1905560fff · 53. Configuration and Build Tools · filed 2008-03-02

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.5.3 by 8d5cc700eb: win/tcl.m4 defines _CRT_SECURE_NO_DEPRECATE and _CRT_NONSTDC_NO_DEPRECATE for MSVC (still so in 8.6 and 9.x), and win/cat.c now uses _read/_write. Closing.
          Full verdict

          FIXED since 8.5.3 (8d5cc700eb core-8-5-branch / 68b47f3256 trunk, 2008-04-06)

          • Platform Windows (build)
          • Tested code reading. win/tcl.m4 on main/9.0/8.6 adds -D_CRT_SECURE_NO_DEPRECATE -D_CRT_NONSTDC_NO_DEPRECATE to CFLAGS_DEBUG and CFLAGS_OPTIMIZE for MSVC (absent at core-8-5-2, present from core-8-5-3). win/cat.c on main also uses _read/_write now.
          • Cause MSVC 2005 POSIX-name deprecation warnings (C4996) combined with -WX in debug builds. kennykb (the reporter) fixed it two months later.
          • Prototype none (fixed)
          • Suggested ticket comment Fixed in 8.5.3 by 8d5cc700eb: win/tcl.m4 defines _CRT_NONSTDC_NO_DEPRECATE for MSVC, and cat.c now uses _read/_write. Closing.
          🟢1905562fff2008test reg-33.14 gets pointer smashFIXED (crash) since 8.5.3/8.6.0 (duptraverse depth limit, dkf 2008-07-01, raised to 15000 in 2013); real stack check on main only, unreleased (4a96…

          Open ticket 1905562fff · 43. Regexp · filed 2008-03-02

          🟢 Since triage (updated 2026-11-11)

          Status now: Open

          • 🔧 2026-10-16 check-in 3de02ad5e8 on regexp-stack-depth, in main: Fix 7d47ba7d94: the regexp engine could overflow the C stack on deeply nested REs. Rebase to trunk
          Full verdict

          FIXED (crash) since 8.5.3/8.6.0 (duptraverse depth limit, dkf 2008-07-01, raised to 15000 in 2013); real stack check on main only, unreleased (4a966b9df4, 2026-10-19)

          • Platform generic (stack-size dependent)
          • Tested regexp {(x{200}){200}$y} x and (x{222}){222}: no crash on main/9.0/8.6, all give "out of memory". On main reg-33.14 is a normal test again (accepts "too complex|out of memory"); reg/regexp/regexpComp/stack.test pass on main (1649 tests, 0 failed). 9.0/8.6 still mark reg-33.14 nonPortable.
          • Cause recursive duptraverse() in generic/regc_nfa.c. Note: on main with the default 8 MB stack the message is still the misleading "out of memory", because the old DUPTRAVERSE_MAX_DEPTH counter (NERR(REG_ESPACE)) is still there and fires before STACK_TOO_DEEP; "too complex" only appears with small stacks. Serhiy's ticket comment says such REs now fail with "regular expression is too complex" -- that is not what main does on Linux by default.
          • Prototype none (Serhiy's branch exists and is merged; changing duptraverse's NERR(REG_ESPACE) to REG_ETOOBIG, or dropping the counter now that STACK_TOO_DEEP exists, would be a 1-5 line follow-up)
          • Existing work regexp-stack-depth (tip 3de02ad5e8) merged into main by 4a966b9df4; Jan does not want it in 9.0 before 9.0.5; not in 8.6
          • Suggested ticket comment The crash is gone; 4a966b9df4 on main adds a real stack check and makes reg-33.14 portable. The duptraverse() depth counter still reports "out of memory" for this RE; could become REG_ETOOBIG or be removed now.
          🟢1902890fff2008compiled regexp behaves differentlyFIXED on main only, unreleased (6e93204003, 2026-10-14); REPRODUCED (9.0, 8.6)

          Open ticket 1902890fff · 47. Bytecode Compiler · filed 2008-02-27

          🟢 Since triage (updated 2026-11-11)

          Status now: Open

          • 🔧 2026-10-11 check-in 0b36adaf16 on regexp-compile-runtime-switch, in main: Fix [1902890]: compiled [regexp] did not treat an RE value starting with "-" as a switch, unlike the command. When the RE is not known at compile time and there
          Full verdict

          FIXED on main only, unreleased (6e93204003, 2026-10-14); REPRODUCED (9.0, 8.6)

          • Platform generic
          • Tested proc c1 {} {set pattern --; set text ----; regexp $pattern $text} -> main "wrong # args" (same as command), 9.0/8.6 return 1. regexp $e $s with e=-about: main 0 {} (like the command), 9.0/8.6 0.
          • Cause generic/tclCompCmds.c TclCompileRegexpCmd emitted INST_REGEXP for a non-literal RE without a runtime "-" check.
          • Prototype none (Serhiy's branch exists and is merged)
          • Existing work regexp-compile-runtime-switch (tip 0b36adaf16) and regexp-compile-dash-pattern (tip a64804ce70), both merged into main by 6e93204003; not in 9.0/8.6
          • Suggested ticket comment Merged into main 6e93204003 (together with 697b1bbfe3). Not in 9.0/8.6.
          ✅1889754fff200832 bit Tcl not able to access 64 bit registry hiveFIXED since 8.6.0 (TIP #362, registry 1.3, "[registry -32bit|-64bit]", 2010-03-30 per changes; docs a9f31bb325 2010-03-31)

          Open ticket 1889754fff · 32. registry Package · filed 2008-02-08

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Implemented in 8.6.0 by TIP #362 (3606e1e39c, registry 1.3): registry -64bit get ... from a 32-bit Tcl accesses the 64-bit registry view, and -32bit does the reverse.
          Full verdict

          FIXED since 8.6.0 (TIP #362, registry 1.3, "[registry -32bit|-64bit]", 2010-03-30 per changes; docs a9f31bb325 2010-03-31)

          • Platform Windows
          • Tested code/doc reading: registry.n (8.6, 9.0, main) documents registry ?-32bit|-64bit? option keyName ..., tclWinReg.c uses KEY_WOW64_32KEY/KEY_WOW64_64KEY. Not in 8.5.
          • Cause missing feature, implemented by TIP 362 (different syntax from the attached HP patch).
          • Prototype none (already implemented)
          • Suggested ticket comment Implemented by TIP #362 in Tcl 8.6 (registry 1.3): registry -64bit get ... from a 32-bit Tcl accesses the 64-bit view. Can be closed.
          ✅1879220fff2008Inconsistent naming of debug buildsFIXED since 8.7a5/9.0 (TIP #578 "Death to TCL_DBGX", 834ef99124, 2020-07-03); 8.6 unchanged by design

          Open ticket 1879220fff · 53. Configuration and Build Tools · filed 2008-01-24

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Since TIP #578 (834ef99124, Tcl 9.0) neither Unix nor Windows adds the "g" suffix for debug builds, so the naming is consistent. 8.6 keeps its naming. Closing.
          Full verdict

          FIXED since 8.7a5/9.0 (TIP #578 "Death to TCL_DBGX", 834ef99124, 2020-07-03); 8.6 unchanged by design

          • Platform build
          • Tested code reading. DBGX is gone from win/tcl.m4 on 8.7, 9.0 and main, so Windows debug builds now get the same names as release builds, as on Unix. 8.6 win/tcl.m4 still sets DBGX=g for --enable-symbols, while Unix dropped it in 8.5 (patch 1081595).
          • Cause Unix removed the "g" suffix in 8.5 but Windows kept it until TIP 578.
          • Prototype none
          • Suggested ticket comment Since TIP #578 (834ef99124, Tcl 8.7/9.0) neither Unix nor Windows adds the "g" debug suffix, so naming is consistent. 8.6 stays as is. Closing.
          ✅1845108fff2007FilterInputBytes/GetInput do not handle EINTRFIXED since 8.6.12 (346094351f, 2021-03-09, "Fix 8419c55e2c: Tclsh read loop does not handle EINTR"; in 9.0.0+ and main)

          Open ticket 1845108fff · 25. Channel System · filed 2007-12-05

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.6.12 and 9.0 by 346094351f (ticket 8419c55e2c): the Unix file channel now retries read() on EINTR.
          Full verdict

          FIXED since 8.6.12 (346094351f, 2021-03-09, "Fix 8419c55e2c: Tclsh read loop does not handle EINTR"; in 9.0.0+ and main)

          • Platform unix
          • Tested LD_PRELOAD shim installing a SIGUSR1 handler without SA_RESTART, tclsh reading commands from a fifo, SIGUSR1 sent while it blocks in read(): 8.4.20 and 8.6.11 exit silently; 8.6.12, 8.6.18, 9.0.5, main keep reading and run the next command (work-a5/eintr/run.sh)
          • Cause unix/tclUnixChan.c:FileInputProc returned EINTR as a hard error; now it retries read() in a loop.
          • Prototype none (already fixed)
          • Suggested ticket comment Fixed by 346094351f for 8419c55e2c (Tcl 8.6.12 and later, 9.0): FileInputProc now retries read() on EINTR. Verified with a non-SA_RESTART signal handler. Closing as duplicate/fixed.
          ⬜1826826fff2007sysnative path name not foundFIXED (file size/exists/open work in 8.6.18 and 9.1); remaining cosmetic bug REPRODUCED (8.6): file normalize C:/Windows/Sysnative/arp.exe gives …

          Open ticket 1826826fff · 37. File System · filed 2007-11-06

          Full verdict

          FIXED (file size/exists/open work in 8.6.18 and 9.1); remaining cosmetic bug REPRODUCED (8.6): file normalize C:/Windows/Sysnative/arp.exe gives C:/Windows//ARP.EXE; FIXED on main (9.1.0) by 0ac314cb96 (2026-07-16, also on core-9-0-branch, 32-bit 9.0 not built this run)

          • Platform Windows
          • Tested not testable on Linux. Repro script win/1826826fff.tcl (file size/exists/normalize/open of %SystemRoot%/Sysnative/arp.exe, verdict INCONCLUSIVE on a 64-bit tclsh).
          • Cause likely win/tclWinFile.c normalization (FindFirstFile per component): Sysnative is a WOW64 alias that doesn't appear in directory listings. Same family as 3463315 (nijtmans, 2013).
          • Prototype none
          • Suggested ticket comment Needs a 32-bit tclsh on 64-bit Windows. Probably the same cause as 3463315fff: normalization looks up each component, and Sysnative is not listed. Repro script attached.
          Windows VM, run 5
          • Platform Windows (32-bit tclsh on 64-bit Windows)
          • Verdict FIXED (file size/exists/open work in 8.6.18 and 9.1); remaining cosmetic bug REPRODUCED (8.6): file normalize C:/Windows/Sysnative/arp.exe gives C:/Windows//ARP.EXE; FIXED on main (9.1.0) by 0ac314cb96 (2026-07-16, also on core-9-0-branch, 32-bit 9.0 not built this run)
          • Tested VM win/1826826fff.tcl with 32-bit builds of tcl-wt91 (main) and tcl-wt86 (core-8-6-branch): file size 49152, exists 1, open OK on both; normalize C:/Windows/Sysnative/ARP.EXE on main, C:/Windows//ARP.EXE on 8.6. With prototypes/3463315fff.patch (the 8.6 backport) applied, 32-bit 8.6 also gives C:/Windows/Sysnative/ARP.EXE.
          • Cause win/tclWinFile.c TclpObjNormalizePath: FindFirstFileW of the WOW64 alias Sysnative fails (it is not listed), and 8.6 drops the component; same as 3463315fff/d40d8db3fb.
          • Prototype none new; prototypes/3463315fff.patch (8.6 backport of 0ac314cb96) fixes it, tested 32-bit on Windows.
          • Suggested ticket comment On Windows 11 a 32-bit 8.6.18 and 9.1 can stat and open C:/Windows/Sysnative/arp.exe. &#91;file normalize&#93; still drops the Sysnative component in 8.6 (fixed in 9.1.0 by 0ac314cb96, same cause as 3463315fff); the proposed 8.6 backport there fixes it too.
          ⬜1805794fff2007changing unix system timeFIXED since 9.1a1 (TIP 723 acb0a91c2c, 2026-01-08: timers use Tcl_GetMonotonicTime()); still REPRODUCED on 9.0 and 8.6

          Open ticket 1805794fff · 03. Timer Events · filed 2007-10-01

          Full verdict

          FIXED since 9.1a1 (TIP 723 acb0a91c2c, 2026-01-08: timers use Tcl_GetMonotonicTime()); still REPRODUCED on 9.0 and 8.6

          • Platform unix (generic timer code)
          • Tested LD_PRELOAD shim (work-a5/tj/shim.so) that moves CLOCK_REALTIME/gettimeofday back one day 1 s after start, script with a 100 ms periodic [after]: 8.6.18 and 9.0.5 stop after 5 ticks (timer waits a day); main runs all 25 ticks
          • Cause generic/tclTimer.c in 8.6/9.0 keeps timer deadlines as absolute wall-clock Tcl_Time (Tcl_GetTime), so a backward jump delays every pending timer by the jump.
          • Prototype none (fixed on main by TIP 723; a backport to 9.0/8.6 is a behaviour/API change, not a small fix)
          • Suggested ticket comment Fixed in Tcl 9.1 by TIP 723 (acb0a91c2c): &#91;after&#93; and timer handlers now use the monotonic clock, so wall-clock jumps no longer delay them. 8.6/9.0 keep the old wall-clock behaviour. Same issue as 1464890fff.
          ✅1799833fff2007commands evaluated via TclCompEvalObj() evade tracingFIXED since 8.6.0 (NRE rewrite, a646e4c362, 2008-07-13, removed TclCompEvalObj; every non-inlined command now goes through tclBasic.c:Dispatch wher…

          Open ticket 1799833fff · 45. Parsing and Eval · filed 2007-09-21

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed by the NRE rewrite a646e4c362 (8.6.0): TclCompEvalObj() is gone, and every command dispatch passes through the code with the DTrace cmd probes and compile tracing.
          Full verdict

          FIXED since 8.6.0 (NRE rewrite, a646e4c362, 2008-07-13, removed TclCompEvalObj; every non-inlined command now goes through tclBasic.c:Dispatch where the DTrace cmd-* probes are)

          • Platform generic
          • Tested main with a TCL_COMPILE_DEBUG build: tcl_traceCompile 2 prints the bytecode for scripts run by [eval], [uplevel #0] and [namespace eval]. DTrace probes not tested (no sys/sdt.h here); judged by code: TclNREvalObjv → EvalObjvCore → Dispatch is the single entry with TCL_DTRACE_CMD_ENTRY. 8.5 (EOL) still has TclCompEvalObj.
          • Cause in 8.5 TclCompEvalObj ran bytecode outside the paths with probes/trace output.
          • Prototype none (already fixed)
          • Suggested ticket comment TclCompEvalObj() is gone since NRE (8.6.0); all command dispatch passes through Dispatch() with the DTrace probes, and tcl_traceCompile 2 shows bytecode of eval'ed scripts in 9.1. Can be closed.
          ✅1787911fff2007array search of envFIXED since 8.6.2 (afaff33553, 2013-09-27, dkf: "Rewrote how ::env is synchronized ... no longer smashes the array", for [219226])

          Open ticket 1787911fff · 07. Variables · filed 2007-09-04

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.6.2 by afaff33553 (ticket [219226]): env is no longer unset and recreated on array access, so [array startsearch env], anymore and nextelement work.
          Full verdict

          FIXED since 8.6.2 (afaff33553, 2013-09-27, dkf: "Rewrote how ::env is synchronized ... no longer smashes the array", for [219226])

          • Platform generic
          • Tested main, 9.0, 8.6 tip, 8.6.2–8.6.11 OK; 8.6.0, 8.6.1, 8.5.19, 8.4.20 fail with couldn't find search "s-1-env"
          • Cause old EnvTraceProc called TclSetupEnv on array traces, which unset and recreated ::env, killing searches.
          • Prototype none (already fixed)
          • Suggested ticket comment Fixed in 8.6.2 by afaff33553 (env no longer recreated on array access); [array startsearch env]/anymore/nextelement work in 8.6, 9.0, 9.1. Can be closed.
          ✅1786647fff2007wrong variables in compat/memcmp.cFIXED since 8.4.16 (ec5ad54498, 2007-09-06)

          Open ticket 1786647fff · 52. Portability Support · filed 2007-09-02

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed by ec5ad54498 (8.4.16); compat/memcmp.c no longer exists in Tcl 9.
          Full verdict

          FIXED since 8.4.16 (ec5ad54498, 2007-09-06)

          • Platform build
          • Tested code reading; 8.6 compat/memcmp.c uses ptr1/ptr2 as in the attached patch. compat/memcmp.c was removed from 9.0/main by e275c72695 (2023-06-27).
          • Cause compat/memcmp.c dereferenced the void* args s1/s2 instead of ptr1/ptr2 (dkf fixed it, ticket left Open).
          • Prototype none (already fixed)
          • Suggested ticket comment Fixed by ec5ad54498 (8.4.16, 2007), compat/memcmp.c no longer exists in 9.0 (e275c72695). Closing.
          ✅1758527fff2007env bug on platform windowspatch 86FIXED since 9.0.4 / 9.1.0 (8e4061a167, 2026-05-21, under b492c33154); REPRODUCED (8.6)

          Open ticket 1758527fff · 08. Environment Variables · filed 2007-07-22

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Duplicate)

          • 💬 2026-10-21 serhiy.storchaka: Duplicate of b492c33154, fixed in 9.0.4 and 9.1.0 by 8e0656d976 / 2ae307305b: on Windows, setting an env element to "" now removes it from the array too. 8.6 is unchanged.
          Full verdict

          FIXED since 9.0.4 / 9.1.0 (8e4061a167, 2026-05-21, under b492c33154); REPRODUCED (8.6)

          • Platform Windows
          • Tested Linux not affected (empty env values are kept). Code: generic/tclEnv.c EnvTraceProc on _WIN32 now unsets env(x) when it is set to "" (present at tags core-9-0-4 and core-9-1-0, absent on core-8-6-branch); test env-11.0 (win constraint). Repro win/1758527fff.tcl.
          • Cause Windows putenv("X=") deletes the variable, while the Tcl array element stayed, so [info exists] said 1 and the read trace failed.
          • Prototype none (fixed in 9.0/9.1 under duplicate b492c33154, which is Closed/Fixed; backport to 8.6 would be the same ~10 lines if wanted)
          • Suggested ticket comment Duplicate of b492c33154, fixed in 9.0.4 and 9.1.0 (env element is now unset when set to "" on Windows). 8.6 still has the old behaviour. Close as duplicate unless an 8.6 backport is wanted.
          Windows VM, run 2
          • Platform Windows
          • Verdict FIXED since 9.0.4 / 9.1.0 (8e4061a167, 2026-05-21, under b492c33154); REPRODUCED (8.6)
          • Tested VM win/1758527fff.tcl: 9.1b1 and 9.0.5 info exists=0 read-error=1 (consistent); 8.6.18 info exists=1 read-error=1 (inconsistent)
          • Cause on Windows putenv("X=") deletes the variable while the env() element stays; 9.x EnvTraceProc unsets the element when it is set to ""
          • Prototype prototypes/1758527fff-86.patch (against core-8-6-branch, 25 changed lines incl. test env-11.0 copied from 9.0): straight backport of the 9.0 EnvTraceProc change. env.test passes on Windows with it (27/27); env-11.0 fails with unpatched 8.6.18. Tested on Windows.
          • Suggested ticket comment Duplicate of b492c33154, fixed in 9.0.4 and 9.1.0. Confirmed that 8.6.18 on Windows still has it; if a backport is wanted, the 9.0 change applies to 8.6 as is (tested with env-11.0 on Windows).
          ✅1752589fff2007file normalize misses two leading / on first level catalogFIXED in 8.6.7 (probably e015bfe72f, 2017-04-28, f34cf83dd0 "failing to collapse multiple /// into /"; not confirmed by building it. 8.5.19, rele…

          Open ticket 1752589fff · 37. File System · filed 2007-07-12

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.6.7 by e015bfe72f (ticket f34cf83dd0): file normalize //bin now gives /bin. In 9.0, a leading // on Unix is a network (UNC) root by design (TIP 402), so there //bin is left as is.
          Full verdict

          FIXED in 8.6.7 (probably e015bfe72f, 2017-04-28, f34cf83dd0 "failing to collapse multiple /// into /"; not confirmed by building it. 8.5.19, released 2016-02-08, also gives /bin, so 8.5 got a separate, earlier fix that I did not identify), then intentionally reintroduced in 8.7/9.0 by TIP 402 (c1632c3763, 2022-10-04): on Unix //name is now a UNC-style root, so file normalize //bin is //bin and file split //bin is //bin

          • Platform unix
          • Tested 8.4.2 /bin; 8.5.0–8.5.18, 8.6.0–8.6.6 //bin; 8.5.19, 8.6.7–8.6.18 /bin; 8.7b1, 9.0.5, main //bin
          • Cause generic/tclFileName.c TclpGetNativePathType/SplitUnixPath: the // network-prefix code is no longer limited to Cygwin/QNX since TIP 402. One oddity remains: file normalize //bin/ gives /usr/bin/ (symlink resolved, trailing slash kept) while //bin stays //bin.
          • Prototype none (behaviour defined by TIP 402)
          • Suggested ticket comment Fixed in 8.6.7. In 9.0 //bin is a network root by design (TIP 402), so it's left alone. The remaining inconsistency is file normalize //bin/, which gives /usr/bin/ with a trailing slash. Close, or keep open for that?
          ⬜1617864fff2006failing test socket -11.8 on port 4003FIXED since 8.6b2/8.6.0 (3f8b48b6b6, 2010-09-28, TIP 162 commit: "make more use of randomized ports"); 8.5 (EOL) still has the fixed ports

          Open ticket 1617864fff · 24. Channel Commands · filed 2006-12-18

          Full verdict

          FIXED since 8.6b2/8.6.0 (3f8b48b6b6, 2010-09-28, TIP 162 commit: "make more use of randomized ports"); 8.5 (EOL) still has the fixed ports

          • Platform generic (test suite)
          • Tested code reading: socket_*-11.8 on main/8.6 opens its servers on port 0 (getPort); no fixed listening ports left in tests/*.test (socket.test 327/333 only test argument errors). tests/socket.test at core-8-6-b1 and in tcl8.5 still uses 4003-4005.
          • Cause test used hard-coded ports 4003-4005 that can collide with local services (rquotad).
          • Prototype none (already fixed)
          • Suggested ticket comment Fixed since 8.6 by 3f8b48b6b6: the socket tests use ephemeral ports (port 0) now. Closing.
          ⬜1591407fff2006anomaly in [file tail] under ~ expansionFIXED since 9.0.0 (TIP 602 merge cff0348cb6, 2022-09-14); still present in 8.6 (by design there)

          Open ticket 1591407fff · 37. File System · filed 2006-11-06

          Full verdict

          FIXED since 9.0.0 (TIP 602 merge cff0348cb6, 2022-09-14); still present in 8.6 (by design there)

          • Platform generic
          • Tested main (own build c3bf6dd052), 9.0.5, 8.6.18, 8.4.20
          • Cause in 8.x a path consisting of a single ~user element is tilde-expanded by [file tail]/[file dirname] (generic/tclPathObj.c, TclPathPart → tilde handling), so a non-existent user errors. Tcl 9 dropped tilde expansion (TIP 602): [file tail ~nosuchuser] → "~nosuchuser", [file tail ~] → "~".
          • Prototype none (fixed on 9.x by design change; changing 8.6 would be a behaviour change)
          • Suggested ticket comment Since Tcl 9.0 (TIP 602, cff0348cb6) ~ is no longer expanded, and &#91;file tail ~fred&#93; returns "~fred" whether or not the user exists. 8.6 keeps the old behaviour. Closing as fixed in 9.0.
          ⬜1569064fff2006different ~ expansions in [file] causes fCmd 6.26 to failFIXED since 8.4.15/8.5a6 for the test failure (22b6ae1be7, das, 2006-11-10: expected result uses $::env(HOME)); the underlying inconsistency is g…

          Open ticket 1569064fff · 37. File System · filed 2006-10-02

          Full verdict

          FIXED since 8.4.15/8.5a6 for the test failure (22b6ae1be7, das, 2006-11-10: expected result uses $::env(HOME)); the underlying inconsistency is gone in 9.0 (TIP 602, cff0348cb6)

          • Platform unix
          • Tested 8.6.18, 8.4.20, 9.0.5 with HOME pointing through a symlink
          • Cause in 8.x [file dirname ~] normalizes the expanded home (resolving symlinks: gives .../real/home) while [file nativename ~] and error messages use $HOME literally. In 9.x ~ is literal and the test uses [file home], which returns $HOME unnormalized. 8.6 still shows the dirname/nativename discrepancy.
          • Prototype none (test already fixed; 8.6 ~ semantics not worth changing now)
          • Suggested ticket comment The fCmd-6.26 failure was fixed in 22b6ae1be7; in Tcl 9 tilde expansion was removed (TIP 602), so the inconsistent ~ normalization no longer exists. Closing.
          ✅1483802fff2006lsort inconsistant behavior for negative value of -indexFIXED since 8.6.9 (c07b8490b6 on branch bug-db36fa5122, 2018-03-09)

          Open ticket 1483802fff · 10. Objects · filed 2006-05-08

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.6.9 by 71c0a83836 (ticket db36fa5122): negative -index values now raise an error instead of being taken as end-relative.
          Full verdict

          FIXED since 8.6.9 (c07b8490b6 on branch bug-db36fa5122, 2018-03-09)

          • Platform generic
          • Tested 8.4.20, 8.5.x and 8.6.0–8.6.8 sort with negative -index values (as end-relative); 8.6.9+ error index "-2" cannot select an element from any list; 9.0 and main error index "-2" out of range. end, end-1 still work.
          • Cause TclGetIntFromIndex accepted negative numbers as encoded end-relative indices; fixed by the TclIndexEncode machinery ("Use new machinery to repair index value flaws in [lsort]"), released in 8.6.9 (2018-11-16). Tests lsort-* covering negative indices were added in 4bf1bd4bcc.
          • Prototype none (fixed)
          • Suggested ticket comment Fixed in 8.6.9 by c07b8490b6: negative -index values now raise an error (8.6: "cannot select an element from any list", 9.0: "out of range"). Can be closed.
          ⬜1470152fff2006Unix Notifier thread causes a number of problems.FIXED since 9.0 on Linux and the BSDs (TIP 458 858569fb04, 2017-05-18: per-thread epoll/kqueue notifier, no notifier thread; first in 8.7a1/9.0);…

          Open ticket 1470152fff · 01. Notifier · filed 2006-04-13

          Full verdict

          FIXED since 9.0 on Linux and the BSDs (TIP 458 858569fb04, 2017-05-18: per-thread epoll/kqueue notifier, no notifier thread; first in 8.7a1/9.0); 8.6 and other Unixes (select notifier) still use the notifier thread, with fork handled by pthread_atfork since 8.6.1 (94b207274c, 2013-07-22)

          • Platform unix
          • Tested code reading: main/9.0 builds on Linux define NOTIFIER_EPOLL (unix/tclEpollNotfy.c, no thread); unix/tclSelectNotfy.c keeps NotifierThreadProc plus AtForkChild (test unixforkevent-1.1). No runtime test (the ticket gives no repro, only an alternative notifier).
          • Cause design of the select-based threaded notifier (one global helper thread).
          • Prototype none (RFE/redesign; largely done by TIP 458)
          • Suggested ticket comment Since Tcl 8.7/9.0 (TIP 458) Linux and the BSDs use a per-thread epoll/kqueue notifier without the helper thread, and the select notifier re-creates the thread after fork via pthread_atfork (8.6.1+). Suggest closing; please open a new ticket with a repro if a specific problem remains.
          ⬜1464890fff2006after command hangs when system time changesFIXED since 9.1a1 (TIP 723 acb0a91c2c, 2026-01-08, monotonic timers); still REPRODUCED on 9.0 and 8.6 (same bug as 1805794fff)

          Open ticket 1464890fff · 03. Timer Events · filed 2006-04-05

          Full verdict

          FIXED since 9.1a1 (TIP 723 acb0a91c2c, 2026-01-08, monotonic timers); still REPRODUCED on 9.0 and 8.6 (same bug as 1805794fff)

          • Platform generic (tested on unix)
          • Tested same LD_PRELOAD clock-jump test as 1805794fff (work-a5/tj): with the wall clock moved back one day, an [after 100] loop stops on 8.6.18 and 9.0.5 and keeps running on main. Windows not tested (main's win/ timers also go through Tcl_GetMonotonicTime).
          • Cause generic/tclTimer.c in 8.6/9.0 schedules timers by wall-clock Tcl_GetTime.
          • Prototype none (fixed on main by TIP 723; not a small backport)
          • Suggested ticket comment Fixed in Tcl 9.1 by TIP 723 (acb0a91c2c): timers use the monotonic clock. 8.6/9.0 still have the old behaviour. Duplicate of 1805794fff.
          ✅1459837fff2006missing close error messageFIXED since 8.5.16 / 8.6.2 (7aa743c0ee on trunk(8.6) and 45d1b34648 on 8.5, 2014-04-04, for 97069ea11a)

          Open ticket 1459837fff · 24. Channel Commands · filed 2006-03-28

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.5.16 / 8.6.2 by 7aa743c0ee / 45d1b34648 (ticket 97069ea11a): when the driver fails to close, [close] now returns the POSIX error message.
          Full verdict

          FIXED since 8.5.16 / 8.6.2 (7aa743c0ee on trunk(8.6) and 45d1b34648 on 8.5, 2014-04-04, for 97069ea11a)

          • Platform generic
          • Tested LD_PRELOAD shim making close(2) fail with EIO on the channel's fd (work-a5/cl): 8.4.2, 8.4.20, 8.5.0-8.5.15, 8.6.0, 8.6.1: [close] fails with an empty message and errorCode NONE; 8.5.16+, 8.6.2+, 8.6.18, 9.0.5, main: "I/O error"/"input/output error", errorCode {POSIX EIO ...}.
          • Cause generic/tclIO.c:Tcl_Close returned TCL_ERROR for a close-proc (via FlushChannel/CloseChannel) error without setting the interp result.
          • Prototype none (already fixed)
          • Suggested ticket comment Fixed since 8.5.16/8.6.2 by 7aa743c0ee (ticket 97069ea11a): Tcl_Close now sets the POSIX error message when the driver's close fails. Verified with a simulated EIO from close(2). Closing.
          ✅1386377fff2005clock scan [clock format [clock seconds] -gmt 1]FIXED since 8.5.0 (kennykb restored zone names in the default format, 2006-07-31; check-in not identified)

          Open ticket 1386377fff · 06. Time Measurement · filed 2005-12-20

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: The default clock format uses the zone name again since 8.5.0 (7d3492d3c1), and free-form clock scan accepts positive numeric offsets such as +0000/+0100 since 8.6.11 (f041312885, ticket 5019748c73). Both round trips work on 8.6.11+ and 9.x.
          Full verdict

          FIXED since 8.5.0 (kennykb restored zone names in the default format, 2006-07-31; check-in not identified)

          • Platform generic
          • Tested 8.5.0, 8.6.18, 9.0.5, main build, 8.6 tip tcltest: the GMT, EST, EET and numeric (-0500) round trips give 0. Only ambiguous abbreviations stay wrong (IST for Asia/Kolkata comes back 30 min off). Released 8.6.18 also misreads "-05" (bug 7bad415e3e, fixed on the 8.6 tip).
          • Cause the original report (the "+0000" default format) has been gone since 8.5.0
          • Prototype none
          • Suggested ticket comment The -gmt round trip works since 8.5.0. Ambiguous abbreviations such as IST cannot round-trip by nature, and numeric "-05" is 7bad415e3e. Suggest closing.
          ✅1306694fff2005Sockets are not working in the child processesFIXED since 8.5.15/8.6.1 (rfe-notifier-fork: pthread_atfork() restarts the notifier thread in the child, 94b207274c; merged to core-8-5-branch [0…

          Open ticket 1306694fff · 24. Channel Commands · filed 2005-09-28

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Works since 8.5.15 / 8.6.1: after a fork the notifier thread is restarted in the child (e733a91cdb, 02909e227f, ticket a0bc856dcd), so fileevents on sockets fire there.
          Full verdict

          FIXED since 8.5.15/8.6.1 (rfe-notifier-fork: pthread_atfork() restarts the notifier thread in the child, 94b207274c; merged to core-8-5-branch 02909e227f, 2013-08-01)

          • Platform unix
          • Tested main (epoll notifier), 9.0, 8.6 (threaded, select notifier thread) tcltest builds: testfork, child opens socket -server, parent connects and gets the reply on all three. (Old releases have no testfork/TclX here; the fork date comes from the timeline.)
          • Cause in a threaded build the forked child had no notifier thread, so fileevents on the listening socket never fired; AtForkChild() in unix/tclUnixNotfy.c now reinitialises it.
          • Prototype none (fixed)
          • Suggested ticket comment Works since 8.5.15/8.6.1: the notifier now uses pthread_atfork() to restart its thread in a forked child (RFE a0bc856dcd, branch rfe-notifier-fork). Verified with testfork on 8.6, 9.0 and 9.1. Closing.
          ⬜1274916fff2005upvar and qualified varNamesFIXED since 9.0.0 (TIP 278, 580ef6f047, 2017-11-03; released 2024-09-25); still REPRODUCED in 8.6 (and 8.5, 8.4)

          Open ticket 1274916fff · 07. Variables · filed 2005-08-28

          Full verdict

          FIXED since 9.0.0 (TIP 278, 580ef6f047, 2017-11-03; released 2024-09-25); still REPRODUCED in 8.6 (and 8.5, 8.4)

          • Platform generic
          • Tested main, 9.0 tip, 8.6.18, 8.5.19, 8.4.20. On 8.x namespace eval a {upvar x b::x} fails with "parent namespace doesn't exist" while namespace eval a {set b::x} finds ::b::x. On 9.x both set b::x and upvar x b::x fail the same way (relative names no longer fall back to the global namespace), and both work once ::a::b exists, so upvar and set are consistent.
          • Cause 8.x: TclLookupVar with AVOID_RESOLVERS/LOOKUP_FOR_UPVAR (fix for 696893) skips the global fallback that ordinary variable lookup uses. TIP 278 removed that fallback for all lookups in 9.0.
          • Prototype none (inconsistency gone in 9.x; changing 8.6 name resolution now is not worth the risk)
          • Suggested ticket comment In 9.0 TIP 278 made relative qualified names resolve only against the current namespace, so &#91;upvar&#93; and &#91;set&#93; now agree. Only 8.6 keeps the old inconsistency; suggest closing as fixed in 9.0.
          ✅1227201fff2005does not translate computer namesFIXED since 8.6.0 (by code reading; 3f8b48b6b6, 2010-09-28, IPv6 rewrite: TclCreateSocketAddress() converts the host with Tcl_UtfToExternalDStrin…

          Open ticket 1227201fff · 27. Channel Types · filed 2005-06-24

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Since 8.6.0 (IPv6 rewrite, 3f8b48b6b6) the host name is converted to the system encoding before getaddrinfo() is called. Please reopen if a non-ASCII host name still fails with 8.6.18 or 9.0.
          Full verdict

          FIXED since 8.6.0 (by code reading; 3f8b48b6b6, 2010-09-28, IPv6 rewrite: TclCreateSocketAddress() converts the host with Tcl_UtfToExternalDString before getaddrinfo(); 8.5 got the same in ebf745562a, 2020-08-10). Windows confirmation optional: win/1227201fff.tcl

          • Platform Windows
          • Tested code reading of generic/tclIOSock.c (main, 9.0, 8.6) and win/tclWinSock.c (8.5, 8.4); win/1227201fff.tcl sanity-run on Linux (it needs a non-ASCII computer name or a hosts-file entry passed as argument to be conclusive).
          • Cause 8.4 passed the UTF-8 host string straight to gethostbyname(). Since 8.6 it goes through the system encoding to the ANSI getaddrinfo(); names outside the ANSI code page still cannot be resolved (GetAddrInfoW would handle them), but with Tcl 9's UTF-8 manifest the system encoding is utf-8, so that case is gone too.
          • Prototype none (fixed)
          • Suggested ticket comment Since 8.6.0 (IPv6 rewrite, 3f8b48b6b6) the host name is converted to the system encoding before getaddrinfo() is called, and 8.5 got the same in ebf745562a. Closing as fixed; reopen if a non-ASCII host name still fails with 8.6.18 or 9.0.
          ⬜1201589fff2005bareword madness in expr parserpatchFIXED since 8.4.10 (ced7ccb24b trunk / eb0b85b5d7 core-8-4-branch, 2005-05-20)

          Open ticket 1201589fff · 45. Parsing and Eval · filed 2005-05-13

          🧪 Prototype check (2026-10-24)

          1201589fff.patch: VERIFIED (tests-only patch; ticket can be closed with it)

          • Base main (tip d6dd6d0f71)
          • Before bug fixed in 2005 (ced7ccb24b, 8.4.10), as the triage says; on the tip expr !n -> 1. On 8.4.2 the test's expressions give syntax error in expression "!n": variable references require preceding $ (also y, n, t, f, tr, !fa, ye, of, !of); 8.4.11 -> 1
          • Patch applied cleanly (tests only)
          • After parseExpr.test 288 total, 0 failed (main); parseExpr-24.* also pass on installed 8.6.18
          • Test existing parseExpr-24.1/24.2 (regression tests for the old fix); cannot fail on any current tip, the behaviour they check fails on 8.4.2
          Full verdict

          FIXED since 8.4.10 (ced7ccb24b trunk / eb0b85b5d7 core-8-4-branch, 2005-05-20)

          • Platform generic
          • Tested expr !n, !f, !fa, !of, tr, on work on main, 9.0, 8.6.18, 8.5.19, 8.4.20; ambiguous o is rejected. Unique-prefix booleans are documented indirectly (expr.n -> string is boolean -> Tcl_GetBoolean "may be abbreviated"). The other part of dgp's second patch, accepting eq/ne/in/ni as function names (expr {eq(1)} with a ::tcl::mathfunc::eq proc), fails with "missing operand" in 8.5 through main (not tested in 8.4, which has no ::tcl::mathfunc); a corner case nobody reported.
          • Cause fixed in tclParseExpr.c (8.4); the 8.5+ parser in tclCompExpr.c keeps the behaviour.
          • Prototype prototypes/1201589fff.patch (6 changed lines, tests only: parseExpr-24.1, parseExpr-24.2; also pass on 8.6.18), which provides the tests dgp kept the ticket open for
          • Suggested ticket comment The fix from 2005 still holds in 9.1/9.0/8.6. Proposed tests parseExpr-24.1/24.2 cover boolean-prefix barewords; with them the ticket can be closed.
          ⬜1199331fff2005per-volume working directoryFIXED on main (9.1.0) by 38f0ce657c (2026-03-25, fix for bca391ab51, "Track current directory for each drive"); REPRODUCED (9.0, 8.6)

          Open ticket 1199331fff · 37. File System · filed 2005-05-10

          Full verdict

          FIXED on main (9.1.0) by 38f0ce657c (2026-03-25, fix for bca391ab51, "Track current directory for each drive"); REPRODUCED (9.0, 8.6)

          • Platform Windows
          • Tested not on Linux (no drive letters); repro win/1199331fff.tcl (uses subst on a free drive letter, prints VERDICT)
          • Cause likely win/tclWinFile.c: TclpObjChdir uses SetCurrentDirectoryW, which does not update the per-drive "=X:" environment variables that GetFullPathName consults for "X:" / "X:foo", so a drive-relative cd always lands at the root (confirmed by patthoyts for 8.4.16/8.5.2).
          • Prototype none (Windows only; would need setting the =X: env var on cd)
          • Suggested ticket comment (after running win/1199331fff.tcl) Still present in 9.x: SetCurrentDirectory() doesn't record the per-drive cwd (the =X: environment variables cmd.exe uses), so "cd X:" always goes to X:/.
          Windows VM, run 5
          • Platform Windows
          • Verdict FIXED on main (9.1.0) by 38f0ce657c (2026-03-25, fix for bca391ab51, "Track current directory for each drive"); REPRODUCED (9.0, 8.6)
          • Tested VM win/1199331fff.tcl (temporary subst Q:, removed in the same session): 9.1b1 cd Q:; cd sub; cd C:; cd Q: gives Q:/sub (and cd C: returns to the C: cwd, Q:x → Q:/sub/x); 9.0.5 and 8.6.18 give Q:/ (and C:/), Q:x → Q:/x. (8.5.19 cannot run the script: no try.)
          • Cause before 38f0ce657c TclpObjChdir used SetCurrentDirectoryW only, which does not maintain the per-drive =X: environment variables; main now tracks them (TclWinUpdateDriveCwd).
          • Prototype none (the main fix is 235 changed lines in win/tclWinFile.c + tests; not in core-9-0-branch)
          • Suggested ticket comment Fixed in 9.1.0 by 38f0ce657c (ticket bca391ab51): on Windows 11 cd Q: now returns to the last directory on Q:. 9.0.5 and 8.6.18 still go to Q:/; the fix has not been backported.
          ✅1197447fff2005fCmd-6.26 error when /home is a linkFIXED since 8.4.15/8.5a6 (22b6ae1be7, 2006-11-10; duplicate of 1569064fff)

          Open ticket 1197447fff · 36. Pathname Management · filed 2005-05-07

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.5.0 by 22b6ae1be7: fCmd-6.26 now builds its expected result from $::env(HOME), so a symlinked /home no longer makes it fail (see also 1569064fff).
          Full verdict

          FIXED since 8.4.15/8.5a6 (22b6ae1be7, 2006-11-10; duplicate of 1569064fff)

          • Platform unix
          • Tested fCmd-6.26 with HOME reached through a symlink: passes on main (own build) and 8.6 tip tcltest
          • Cause the expected result was built from [file dirname ~], which resolved symlinks; the test now uses $::env(HOME) (and [file home] in 9.x).
          • Prototype none (fixed)
          • Suggested ticket comment Fixed by 22b6ae1be7 (fCmd-6.26 now compares with $::env(HOME)), same as 1569064. Verified with a symlinked HOME on 9.1 and 8.6. Closing.
          ⬜1122203fff2005env-4.3(4,5) fail. Locale problem ?FIXED since 8.6.9 (9e55a28229, 2018-06-17, tests only)

          Open ticket 1122203fff · 08. Environment Variables · filed 2005-02-14

          Full verdict

          FIXED since 8.6.9 (9e55a28229, 2018-06-17, tests only)

          • Platform unix
          • Tested env.test with LC_ALL=ru_RU.koi8r and ru_RU.utf8: passes on main, 9.0 tip, 8.6 tip; the 8.5.19 and 8.4.20 env.test still fails env-4.3/4.4/4.5 under KOI8-R.
          • Cause the tests set env names/values containing § and ¶, which KOI8-R cannot encode. 9e55a28229 ("Remove dependencies between tests in env.test") added encodingswitch iso8859-1 to the setup, so the tests no longer depend on the locale. set x \ub6 showing ? in a KOI8-R terminal is expected.
          • Prototype none (fixed)
          • Suggested ticket comment env.test switches the system encoding to iso8859-1 since 9e55a28229 (8.6.9); env-4.3..4.5 pass under ru_RU.KOI8-R in 9.1, 9.0 and 8.6. Can be closed.
          ✅1106853fff2005On Linux, exec >@channel may give incorrect errorFIXED since 8.5.8 (9039eeb584, 2009-07-23, for [Bug 2826248])

          Open ticket 1106853fff · 25. Channel System · filed 2005-01-21

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.5.8 and 8.6.0 by 9039eeb584 (8.5: 62c8e5320a): for a channel without OS handles, exec redirection now reports channel "x" does not support OS handles instead of "wasn't opened for writing".
          Full verdict

          FIXED since 8.5.8 (9039eeb584, 2009-07-23, for [Bug 2826248])

          • Platform generic
          • Tested exec echo hi >@$c / 2>@$c with a write-only reflected channel: main, 9.0, 8.6.18, 8.6.0, 8.5.19, 8.5.8 -> channel "rc0" does not support OS handles; 8.5.7 segfaults (the crash fixed by the same check-in).
          • Cause FileForRedirect() (generic/tclPipe.c) now uses the error set by Tcl_GetChannelHandle() when the driver has no getHandleProc. "wasn't opened for writing" is now shown only when a driver has a getHandleProc that fails without setting an error.
          • Prototype none (fixed)
          • Suggested ticket comment Since 8.5.8 (9039eeb584) exec reports channel "x" does not support OS handles for channels without an OS handle. Closing as fixed.
          ⬜1094682fff2005Odd constraint: unknownFailurepatchFIXED since 8.5.16/8.6.2 (67b870b07c / 0d300e1793, 2014-05-09: the constraint became knownBug). The tests still fail when knownBug is enabled (…

          Open ticket 1094682fff · 26. Channel Transforms · filed 2005-01-02

          🧪 Prototype check (2026-10-24)

          1094682fff.patch: PATCH-INCOMPLETE (the harness fix works, but iogt-3.0/4.0/5.0 still fail with knownBug enabled: new expected trails are needed, and someone has to judge whether the new trails are right)

          • Base main (tip d6dd6d0f71)
          • Before tcltest tests/iogt.test -match 'iogt-[345].0' -constraints knownBug -> 3 failed; echo server dies with invalid command name "tcltests::localIPv4Address" (3.0/4.0), iogt-5.0 returns "" because its script is in -result, not -body.
          • Patch applied cleanly (19 changed lines, tests/iogt.test only)
          • After same command -> the server starts and the body runs, but still 3 failed on stale expected trails (e.g. iogt-3.0 gives 1 {create/write write write ... flush/write delete/write}, expected 1 {create/write create/read write flush/write flush/read delete/write delete/read}). Default run: iogt.test Total 17 Passed 13 Skipped 4 Failed 0 (same as before).
          • Test none possible (the patch is a harness fix for knownBug tests; their expected results still need rework)
          Full verdict

          FIXED since 8.5.16/8.6.2 (67b870b07c / 0d300e1793, 2014-05-09: the constraint became knownBug). The tests still fail when knownBug is enabled (main, 9.0, 8.6), partly because the test harness itself is broken.

          • Platform generic (test suite)
          • Tested iogt.test -match iogt-[345].0 -constraints knownBug on main, 9.0 and 8.6 tcltest: 3 failures.
          • Cause in tests/iogt.test the echo server is started with exec tclsh (whatever tclsh is on PATH) and calls tcltests::localIPv4Address, which isn't defined in that subprocess, so 3.0/4.0 never connect. iogt-5.0 has -result where -body belongs, so its body never runs. With those fixed, the data comes through correctly (iogt-3.0 copy result 1), but the expected audit trails (read/write block sizes) are stale.
          • Prototype prototypes/1094682fff.patch (19 changed lines, test-harness only: echo server gets its address from argv and runs under [interpreter], iogt-5.0 -body). iogt.test still passes by default (4 skipped knownBug). The knownBug tests now run but still need new expected trails, so this is not a full fix.
          • Suggested ticket comment The constraint was renamed knownBug in 2014 (67b870b07c). The tests themselves can't pass as written: the echo server subprocess lacks tcltests::localIPv4Address, and iogt-5.0 uses -result instead of -body. A harness fix is ready, but the expected traces still need updating.
          ✅1048733fff2004possible segfaults with NULL interpsFIXED since 8.4.8/8.5a2 (088b6e6a3c, dkf, 2004-09-03, "Stop NULL interp args from causing crashes", Bug 1020538)

          Open ticket 1048733fff · 37. File System · filed 2004-10-17

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: The reported crash (file rename -force calling Tcl_FSFileAttrsSet() with a NULL interp) was fixed in 8.5.0 by 088b6e6a3c ([1020538]): the Unix attribute setters no longer touch a NULL interp on error. No crash with the other listed functions has been reported, so closing.
          Full verdict

          FIXED since 8.4.8/8.5a2 (088b6e6a3c, dkf, 2004-09-03, "Stop NULL interp args from causing crashes", Bug 1020538)

          • Platform generic
          • Tested main, 8.6.18, 8.4.20: file rename -force A B with B mode 0444 and chmod() failing (LD_PRELOAD shim) — no crash. Code: SetPermissionsAttribute and the other attribute setters guard interp != NULL, and Tcl_PosixError() accepts NULL.
          • Cause was Tcl_PosixError(NULL) inside the attribute setter called by FileCopyRename with a NULL interp.
          • Prototype none (fixed; the remaining list in the comments is an open-ended audit, most of those functions document NULL interp as allowed)
          • Suggested ticket comment The reported crash was fixed by 088b6e6a3c in 2004 (Vince confirmed it); the setters check for a NULL interp and Tcl_PosixError() accepts NULL. No other concrete crash reported in 20 years. Closing.
          ⬜1028264fff2004socket: failed write prevents readFIXED since 8.4.20/8.5.7 (late exit handlers + WSACleanup moved to them, ferrieux 2009-01-27; backported 2009-04-28; companion fix [Bug 2446662]). …

          Open ticket 1028264fff · 27. Channel Types · filed 2004-09-14

          Full verdict

          FIXED since 8.4.20/8.5.7 (late exit handlers + WSACleanup moved to them, ferrieux 2009-01-27; backported 2009-04-28; companion fix [Bug 2446662]). The ticket was reopened only so dgp could review the 8.4 backport before 8.4.20, which has since been released (2013) with the change.

          • Platform Windows
          • Tested TclCreateLateExitHandler(SocketExitHandler) is present in win/tclWinSock.c on 8.4, 8.6 and main. Linux main and 8.6 with ferrieux's srv.tcl/cli.tcl (pause 0 and 100 ms): server reads the message, its write succeeds (NOLUCK), no data loss. Windows not re-run.
          • Cause WSACleanup ran in an early exit handler before sockets were closed, so the peer got an RST and its queued input was dropped.
          • Prototype none (fixed)
          • Suggested ticket comment The fix (late exit handlers, plus [Bug 2446662]) is in 8.4.20, 8.5, 8.6 and 9.x, and the 8.4.20 review this was reopened for is moot now. Closing.
          ✅962519ffff2004nativeFilesystemRecord comparisions invalid with threadsFIXED since 8.4.13/8.5a4 (b8ef0f05a3 vincentdarley 2006-03-28 on trunk, bfa4344c2c das on core-8-4-branch, "fix to nativeFilesystemRecord compa…

          Open ticket 962519ffff · 37. File System · filed 2004-05-28

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.4.13/8.5.0 by b8ef0f05a3 (bfa4344c2c on 8.4): the loops over the thread-local filesystem list compare fsPtr with &tclNativeFilesystem instead of the static record. No problems were reported in this area since then, so closing.
          Full verdict

          FIXED since 8.4.13/8.5a4 (b8ef0f05a3 vincentdarley 2006-03-28 on trunk, bfa4344c2c das on core-8-4-branch, "fix to nativeFilesystemRecord comparisons", Bug 1064247)

          • Platform generic
          • Tested code review of main generic/tclIOUtil.c: loops over the thread-local list now compare fsRecPtr->fsPtr with &tclNativeFilesystem; the remaining &nativeFilesystemRecord comparisons (lines 780, 928) walk the global filesystemList, where that is correct.
          • Cause was comparing thread-local record copies against the static global record.
          • Prototype none (fixed)
          • Suggested ticket comment Fixed in 2006 by b8ef0f05a3: thread-local loops compare fsPtr with &tclNativeFilesystem; &nativeFilesystemRecord is only compared on the global list. Left open as a reminder since then; closing.
          ⬜961646ffff2004[file normalize] changes file referenceFIXED since 8.5.0 (the [open]/[file exists] inconsistency; check-in not pinned down, 8.4.20 still has it, 8.5.0 does not); the remaining [file norm…

          Open ticket 961646ffff · 37. File System · filed 2004-05-27

          Full verdict

          FIXED since 8.5.0 (the [open]/[file exists] inconsistency; check-in not pinned down, 8.4.20 still has it, 8.5.0 does not); the remaining [file normalize] behaviour is as documented

          • Platform generic
          • Tested main (own build), 9.0.5, 8.6.18, 8.5.19, 8.5.0, 8.4.20 on Linux
          • Cause in 8.4 [open noSuchDir/../file] opened ./file through the lexically normalized path. Since 8.5.0 [open] and [file exists] both pass the path to the OS unchanged and fail (0 / ENOENT). [file normalize noSuchDir/../file] still collapses ".." lexically when the directory does not exist, and file.n says it does that ("all ../ and ./ removed").
          • Prototype none (what is left is documented behaviour; leaving ".." to the native FS, as dgp's 2018 comment wants, would be a redesign)
          • Suggested ticket comment Since 8.5.0 &#91;open noSuchDir/../file&#93; and &#91;file exists noSuchDir/../file&#93; both fail, as the OS does; only &#91;file normalize&#93; still collapses ".." lexically for missing directories, and file.n documents that. Propose to close; open a TIP if ".." handling should move into each filesystem.
          ⬜959052ffff2004variable name resolution and unset varsFIXED since 9.0.0 (TIP 278, merged to trunk in 580ef6f047, 2017-11-03; [info vars] follow-up 5d2e5c3e82, 2024-07-14); still REPRODUCED on 8.6 a…

          Open ticket 959052ffff · 07. Variables · filed 2004-05-23

          Full verdict

          FIXED since 9.0.0 (TIP 278, merged to trunk in 580ef6f047, 2017-11-03; [info vars] follow-up 5d2e5c3e82, 2024-07-14); still REPRODUCED on 8.6 and 8.4 (by design there, the 2004 fix was reverted for compatibility)

          • Platform generic
          • Tested all reporter scriptlets on main, 9.0, 8.6, 8.4 — on 9.x relative names in a namespace never resolve to global (undefined/traced/upvar'd) variables, so [set x] in ::b or ::a creates ::b::x / ::a::x and [info vars] agrees; on 8.6 the old surprises remain
          • Cause 8.x Tcl_FindNamespaceVar global fallback ("creative writing") counts undefined-but-referenced global vars; removed by TIP 278 in 9.0
          • Prototype none (fixed in 9.0; 8.6 change was reverted on purpose because of compatibility, e.g. AlphaTk [959786])
          • Suggested ticket comment Resolved in Tcl 9.0 by TIP 278 (no global fallback for namespace variable resolution, 580ef6f047); all scriptlets now behave consistently. 8.6 keeps the old semantics for compatibility, so this can be closed as Fixed in 9.0.
          ⬜849514ffff2003Absolute pathnames with ../ and linksFIXED since 8.5.0 (5b1ae40484, 2003-12-14, "rewrite of file normalization"); the 8.4 backport f4ec15959a was reverted, and 8.4.20 still has the bug

          Open ticket 849514ffff · 36. Pathname Management · filed 2003-11-26

          Full verdict

          FIXED since 8.5.0 (5b1ae40484, 2003-12-14, "rewrite of file normalization"); the 8.4 backport f4ec15959a was reverted, and 8.4.20 still has the bug

          • Platform generic
          • Tested main (own build), 9.0.5, 8.6.18, 8.5.19, 8.5.0 correct; 8.4.20 wrong: link/../foo normalizes to ./foo and [open] reads the wrong file
          • Cause the old string-based ".." removal in TclFSNormalizeAbsolutePath ignored symlinks; it was rewritten for 8.5
          • Prototype none
          • Suggested ticket comment Fixed for 8.5.0 by 5b1ae40484; the 8.4 backport was reverted, and 8.4 is no longer maintained. Resolution is already Fixed, so propose to close.
          ⬜833713ffff2003Creating relative links with "file link .." commanFIXED since 8.5.0 (60f9fa8b4d, 2003-12-12, "allow creation of relative links"); never backported to 8.4 (8.4.20 still writes an absolute target)

          Open ticket 833713ffff · 37. File System · filed 2003-10-31

          Full verdict

          FIXED since 8.5.0 (60f9fa8b4d, 2003-12-12, "allow creation of relative links"); never backported to 8.4 (8.4.20 still writes an absolute target)

          • Platform unix
          • Tested main (own build), 9.0.5, 8.6.18, 8.5.19, 8.5.0 write the relative target "one"; 8.4.20 writes the absolute path
          • Cause TclpObjLink used Tcl_FSGetNativePath(toPtr), which gives the absolute path; since 8.5 the target string is used as given
          • Prototype none
          • Suggested ticket comment Fixed in 8.5.0 by 60f9fa8b4d: &#91;file link -symbolic rel one&#93; now stores the relative target. 8.4 is no longer maintained, so propose to close.
          ✅772581ffff2003children inherit 2 open pipes (security?)FIXED since 8.6.0 / 8.5.13 (5a5e87fae7, 2008-12-12, "Fix missing CLOEXEC on internal pipes [2417695]"; 8.5 backport 662a69f571, 2012-09-07)

          Open ticket 772581ffff · 49. Threading · filed 2003-07-16

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.6.0 by 5a5e87fae7 and in 8.5.13 by 662a69f571 ([2417695]): the notifier's internal pipe is close-on-exec, so exec'd children no longer inherit it.
          Full verdict

          FIXED since 8.6.0 / 8.5.13 (5a5e87fae7, 2008-12-12, "Fix missing CLOEXEC on internal pipes [2417695]"; 8.5 backport 662a69f571, 2012-09-07)

          • Platform unix
          • Tested threaded main, 9.0.5, 8.6.18 after starting the notifier (after+vwait) and opening a file: the parent has fds 3–6, an [exec sh] child sees only 0/1/2. Main/9.0 epoll notifier uses eventfd(EFD_CLOEXEC)/pipe2(O_CLOEXEC)/epoll_create1(EPOLL_CLOEXEC); select notifier sets FD_CLOEXEC on the trigger pipe (tclSelectNotfy.c:1011).
          • Cause the notifier thread's trigger pipe lacked FD_CLOEXEC (duplicate of [2417695]).
          • Prototype none
          • Suggested ticket comment Fixed long ago by 5a5e87fae7 (8.6) and 662a69f571 (8.5.13): the notifier's internal pipes are close-on-exec, and on Linux 9.x the epoll notifier uses EFD_CLOEXEC/O_CLOEXEC. Verified that exec'd children see only fds 0–2. Closing.
          ✅671624ffff2003enable-threads on FreeBSD5.0FIXED since 8.4.2 for the configure part (6fef89efec, 2003-02-19: THREADS_LIBS in LIBS while checking pthread_attr_setstacksize); the stack.test …

          Open ticket 671624ffff · None · filed 2003-01-21

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: The pthread_attr_setstacksize detection was fixed in 8.4.2 by 6fef89efec: configure now links the check with the thread libraries. Since 8.6, proc recursion no longer uses the C stack (NRE), so the stack.test crash on small default stacks is out of date. Closing.
          Full verdict

          FIXED since 8.4.2 for the configure part (6fef89efec, 2003-02-19: THREADS_LIBS in LIBS while checking pthread_attr_setstacksize); the stack.test recursion crash is moot since 8.6 (NRE, no C-stack recursion for proc calls)

          • Platform build (FreeBSD)
          • Tested read unix/tcl.m4 SC_ENABLE_THREADS on 8.4, 8.5, 8.6 and main (all check with LIBS="$LIBS $THREADS_LIBS"); not run on FreeBSD (a FreeBSD VM run of --enable-threads + stack.test would confirm, but FreeBSD 5 is long gone)
          • Cause configure tested pthread_attr_setstacksize without -pthread; small default thread stack made deep recursion crash.
          • Prototype none
          • Suggested ticket comment The configure check was fixed in 8.4.2 (6fef89efec) and since 8.6 proc recursion no longer uses the C stack (NRE). Closing as out of date.
          ✅627552ffff2002bad pipe error messageFIXED since 8.5.8 (9039eeb584, 2009-07-23)

          Open ticket 627552ffff · 25. Channel System · filed 2002-10-23

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.5.8 and 8.6.0 by 9039eeb584 (8.5: 62c8e5320a): a redirection from or to a channel without an OS handle now reports channel "x" does not support OS handles. See also 1106853fff.
          Full verdict

          FIXED since 8.5.8 (9039eeb584, 2009-07-23)

          • Platform unix
          • Tested main (own build), 9.0.5, 8.6.18, 8.5.0/8.5.7/8.5.8/8.5.19, 8.4.20; exec cat <@ $refchan, exec echo >@ $refchan, exec echo >@ $readOnlyFile
          • Cause generic/tclPipe.c:FileForRedirect now takes the error that Tcl_GetChannelHandle leaves via Tcl_SetChannelError: a channel without OS handles (e.g. chan create) gives channel "rc0" does not support OS handles, while a read-only file used for output still gives wasn't opened for writing. Before 8.5.8, 8.5.0 and 8.5.7 printed nothing for a reflected channel (they died in the getHandleProc==NULL path, Bug 2826248).
          • Prototype none (already fixed)
          • Suggested ticket comment Fixed by 9039eeb584 (Tcl 8.5.8): exec redirection from a channel without an OS handle now reports "channel ... does not support OS handles" instead of "wasn't opened for reading". Can be closed.
          ⬜582506ffff2002imported cmds do not fire exec tracespatchFIXED since 8.6.0 (281b7ede58, 2008-07-21, msofer: [namespace import] reimplemented on NRE, imported commands now dispatch through the normal eva…

          Open ticket 582506ffff · 21. [namespace] · filed 2002-07-16

          🧪 Prototype check (2026-10-24)

          582506ffff.patch: VERIFIED (test-only patch; bug already FIXED in 8.6.0)

          • Base main (tip d6dd6d0f71)
          • Before work-a5/imp.tcl (enter/leave traces on ::export::foo, call ::import::foo then ::export::foo) -> 8.5.19: enter leave (bug); main tip: enter leave enter leave (fixed since 8.6.0, 281b7ede58; ticket only lacked a test)
          • Patch applied cleanly (14 changed lines, test only)
          • After trace.test 287/0 failed, trace-41.1 passes
          • Test existing trace-41.1, regression test only; cannot fail on any current branch (the behaviour differs only on 8.5)
          Full verdict

          FIXED since 8.6.0 (281b7ede58, 2008-07-21, msofer: [namespace import] reimplemented on NRE, imported commands now dispatch through the normal eval path); still REPRODUCED on 8.5. Ticket left open only for the missing test (dgp 2008)

          • Platform generic
          • Tested reporter case on main, 9.0, 8.6 (trace on ::export::foo fires for import::foo), 8.5.19 (does not). The "inverse" case (trace added on the import name does not fire when the origin is called directly) is unchanged everywhere; traces on the import command stay attached to that name — not covered by the fix
          • Cause 8.5 InvokeImportedCmd called the command proc directly, bypassing TclEvalObjvInternal
          • Prototype prototypes/582506ffff.patch (14 changed lines, test trace-41.1 only, no code change) — the missing test; trace.test passes
          • Suggested ticket comment Fixed since 8.6 by 281b7ede58 (import on NRE); the open item was a test. Proposed test trace-41.1. The inverse case (a trace on the import name not firing for direct calls of the origin) still exists, if anyone wants that handled separately.
          ✅550646ffff2002exec (fork) from huge process failsFIXED since 8.6.14 (98e93f779e, 2023-06-08, ticket c54e4a1aeb; posix_spawnp autodetection 9ed0f4a330, 2023-06-30)

          Open ticket 550646ffff · 24. Channel Commands · filed 2002-04-30

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Since 8.6.14 and 9.0.0, exec on Unix creates children with vfork() (or posix_spawnp()) instead of fork(), so the parent's address space is no longer duplicated: b5b3c28401 and 42d3f0cdb9, ticket c54e4a1aeb.
          Full verdict

          FIXED since 8.6.14 (98e93f779e, 2023-06-08, ticket c54e4a1aeb; posix_spawnp autodetection 9ed0f4a330, 2023-06-30)

          • Platform unix
          • Tested strace of exec true on Linux: 8.6.13 calls plain fork (clone without CLONE_VM); 8.6.14, 8.6.18, 9.0.5 and main use posix_spawnp/vfork. The out-of-memory itself was not reproduced (Linux overcommit hides it; Solaris not tested).
          • Cause unix/tclUnixPipe.c:TclpCreateProcess used fork(), which needs swap reserved for a copy of the whole address space on systems without overcommit (Solaris). It now uses posix_spawnp() (glibc >= 2.24, other systems with it) and falls back to vfork() (configure checks HAVE_VFORK and HAVE_POSIX_SPAWNP unconditionally).
          • Prototype none (already fixed)
          • Suggested ticket comment Since Tcl 8.6.14 (98e93f779e, ticket c54e4a1aeb) exec on Unix uses posix_spawnp() or vfork() instead of fork(), so the large address space no longer has to be duplicated. Can be closed; reopen if it still fails on Solaris.
          ✅519868ffff2002[fblocked] on cmd pipe always returns 0FIXED since 8.6.2 / 8.5.16 (most likely 2a40293a49 on trunk and f6b7d496c4 on 8.5, 2014-05-20, "Rework the management of the CHANNEL_BLOCKED and CH…

          Open ticket 519868ffff · 25. Channel System · filed 2002-02-19

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Fixed in 8.5.16 and 8.6.2 by the CHANNEL_BLOCKED/EOF rework 2a40293a49 (8.5: f6b7d496c4): after a short non-blocking read from a command pipe, fblocked now returns 1.
          Full verdict

          FIXED since 8.6.2 / 8.5.16 (most likely 2a40293a49 on trunk and f6b7d496c4 on 8.5, 2014-05-20, "Rework the management of the CHANNEL_BLOCKED and CHANNEL_EOF flags")

          • Platform generic
          • Tested reporter's script on main, 9.0.5, 8.6.18 (correct: "0 bytes was blocked"); 8.4.20, 8.5.0–8.5.15, 8.6.0–8.6.1 print "non blocked"; it changes between 8.5.15 and 8.5.16 and between 8.6.1 and 8.6.2 (both released 2014-08-25). The exact check-in was not bisected by build.
          • Cause the old ReadChars path did not keep CHANNEL_BLOCKED after a short non-blocking [read]; dgp's 2014 flag rework fixed this.
          • Prototype none (already fixed)
          • Suggested ticket comment Works since 8.5.16/8.6.2 (CHANNEL_BLOCKED rework, 2a40293a49): the reporter's script now prints "read of 0 bytes was blocked". Can be closed.
          ⬜488882ffff2001problem with [info exists env(*)]FIXED since 9.0.4 / 9.1.0 (8e0656d976 on core-9-0-branch, 2ae307305b on main, 2026-05-21, ticket b492c33154); not fixed in 8.6 (8.6 tclEnv.c …

          Open ticket 488882ffff · 08. Environment Variables · filed 2001-12-04

          Full verdict

          FIXED since 9.0.4 / 9.1.0 (8e0656d976 on core-9-0-branch, 2ae307305b on main, 2026-05-21, ticket b492c33154); not fixed in 8.6 (8.6 tclEnv.c has no Windows empty-value handling)

          • Platform Windows
          • Tested code reading plus the existing Windows-only test env-11.0 (added by that fix): setting env(X) to "" on Windows now unsets the element, so [info exists] is 0 and [set] fails consistently. Unix keeps empty values (no issue)
          • Cause Windows removes a variable set to an empty value; before the fix the Tcl element stayed while the read trace failed
          • Prototype none (already fixed; 8.6 backport of 8e0656d976 would be the option)
          • Suggested ticket comment Duplicate of b492c33154, fixed in 9.0.4 and 9.1.0 (8e0656d976): on Windows, setting an env element to "" now unsets it, so &#91;info exists&#93; and &#91;set&#93; agree (test env-11.0). Not backported to 8.6.
          🟢446565ffff2001Capturing parens are very slowFIXED on main only, unreleased (9b818c9f14, 2026-10-16); REPRODUCED (9.0, 8.6)

          Open ticket 446565ffff · 43. Regexp · filed 2001-07-31

          🟢 Since triage (updated 2026-11-11)

          Status now: Open

          • 🔧 2026-10-11 check-in 5580370a8f on regsub-needed-submatches, in main: Fix [446565]: regsub always asked the regexp engine for 10 submatches, which can be far more expensive than finding the match. Ask only for the submatches that
          Full verdict

          FIXED on main only, unreleased (9b818c9f14, 2026-10-16); REPRODUCED (9.0, 8.6)

          • Platform generic
          • Tested attached regbug.zip (double-UTF-8 encoded in the repo, decoded): regsub -all on main 5.5 ms, 9.0 1.32 s, 8.6 1.33 s. With \1 in the replacement main still takes 1.16 s (known, stated in Serhiy's comment).
          • Cause generic/tclCmdMZ.c Tcl_RegsubObjCmd always asked for 10 submatches; computing submatch boundaries for this RE is far more expensive than finding the match.
          • Prototype none (Serhiy's branch exists and is merged)
          • Existing work regsub-needed-submatches (tip 5580370a8f) merged into main by 9b818c9f14; not in core-9-0-branch / core-8-6-branch
          • Suggested ticket comment Merged into main 9b818c9f14. Backport to 9.0/8.6 still open; the slow case with submatch references remains (engine limitation).
          ⬜219372ffff2000Cannot set env array element via upvar #0FIXED since 9.0.0 (TIP #634 merge a3a43c9363, 2022-10-05); still REPRODUCED on 8.6 (and 8.7b1, 8.4)

          Open ticket 219372ffff · 08. Environment Variables · filed 2000-10-26

          Full verdict

          FIXED since 9.0.0 (TIP #634 merge a3a43c9363, 2022-10-05); still REPRODUCED on 8.6 (and 8.7b1, 8.4)

          • Platform generic
          • Tested reporter's script and dgp's bar env(TEST) demo on main build, 9.0.5 (env set, child process sees it), 8.7b1/8.6.18/8.4.20 (value lost / "no such variable").
          • Cause before TIP 634, traces on the whole array (the env array's trace in generic/tclEnv.c) weren't fired for an upvar link to a single element. TIP 634 fires them; tests env-7.4..7.7 cover it.
          • Prototype none (TIP 634 is a 9.0 behaviour change; not backported to 8.6 on purpose)
          • Suggested ticket comment Fixed in 9.0 by TIP 634 (a3a43c9363, tests env-7.4..7.7). 8.6 keeps the old behaviour (workaround: upvar the whole env array). Suggest closing as Fixed with 9.0.0.
          ✅219237ffff2000Inconsistent socket error messages between NT and UNIXFIXED since 8.6.0 (getaddrinfo-based name resolution in generic/tclIOSock.c)

          Open ticket 219237ffff · 27. Channel Types · filed 2000-10-26

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: Since 8.6.0 (TIP #162, 3f8b48b6b6) both platforms resolve host names with getaddrinfo() and report the resolver's own message. On Windows an unknown host now gives "couldn't open socket: No such host is known." instead of "invalid argument" (checked with 8.6.18, 9.0.5 and 9.1).
          Full verdict

          FIXED since 8.6.0 (getaddrinfo-based name resolution in generic/tclIOSock.c)

          • Platform Windows
          • Tested main, 9.0.5, 8.6.18 on Linux with win/219237ffff.tcl (socket nonexistent-host-219237.invalid 80): couldn't open socket: Name or service not known.
          • Cause name resolution is now shared code, generic/tclIOSock.c:TclCreateSocketAddress (getaddrinfo + gai_strerror; on Windows gai_strerrorW), used by both unix/tclUnixSock.c and win/tclWinSock.c. So an unknown host should give "No such host is known." on Windows, not the old EINVAL from the WSAHOST_NOT_FOUND translation. The wording is the OS's own on each side, so it will never be the same text. Side note: errorCode stays NONE for resolver failures on all platforms.
          • Prototype none
          • Repro win/219237ffff.tcl (prints a VERDICT line)
          • Suggested ticket comment Since 8.6 both platforms resolve names with getaddrinfo() in generic/tclIOSock.c and report gai_strerror(), so the "invalid argument" message should be gone. To be confirmed on Windows, then this can be closed.
          Windows VM, run 3
          • Platform Windows
          • Verdict FIXED since 8.6.0 (getaddrinfo-based name resolution in generic/tclIOSock.c)
          • Tested VM win/219237ffff.tcl: socket nonexistent-host-219237.invalid 80 → couldn't open socket: No such host is known. on 9.1b1, 9.0.5 and 8.6.18 (Linux: Name or service not known); errorCode is NONE on both
          • Cause the old WSAHOST_NOT_FOUND → EINVAL translation is gone; both platforms report gai_strerror() text from TclCreateSocketAddress
          • Prototype none
          • Suggested ticket comment Fixed: on Windows 8.6.18, 9.0.5 and 9.1 an unknown host gives "couldn't open socket: No such host is known." (the system's own wording, as on Unix), no longer "invalid argument". Closing.
          ✅218871ffff2000FIX for auto_mkindex/auto_load bugs on funny procedure namesFIXED since 8.4.2 or earlier (no failing version found; the [namespace which] part was dgp's 2003-09-23 commit for 8.4.5)

          Open ticket 218871ffff · 38. Init - Library - Autoload · filed 2000-10-26

          ✅ Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • 💬 2026-10-21 serhiy.storchaka: The [info commands] problem was fixed by 56990efadd (8.3.0) and dbc5f2a73f/2ce276bdca (8.4.5). The auto_mkindex part kept open here works too: the reported procedure names are indexed and autoloaded correctly from 8.4.2 through 9.1. Closing.
          Full verdict

          FIXED since 8.4.2 or earlier (no failing version found; the [namespace which] part was dgp's 2003-09-23 commit for 8.4.5)

          • Platform generic
          • Tested reporter's procs plus {\\$foobar} and {[exit]} (work-a1/p218871/test.tcl, work-a1/218871ffff.tcl), each autoloaded in a fresh interp: all four are indexed correctly and autoload on main, 9.0.5, 8.6.18, 8.5.19, 8.4.20, 8.4.11 and 8.4.2; no GOTCHA error, no exit.
          • Cause n/a. The ticket stayed open only as dgp's reminder about auto_mkindex, and that now works. The remaining knownBug autoMkindex-3.3 (still fails on main) is a different issue: custom auto_mkindex_parser::command handlers and namespaces, not odd names.
          • Prototype none (fixed)
          • Suggested ticket comment The reported names (including \\$foobar and &#91;exit&#93;) are indexed and autoloaded correctly in 8.4.2 through 9.1, so the auto_mkindex issue kept open here is resolved. Closing.