Untouched open Tcl bug tickets: Docs

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.

Docs 66

⬜e4c24e86b52026timer documentation typospatchDOCS (mostly fixed on main; one item left)

Open ticket e4c24e86b5 · 18. Commands M-Z · filed 2026-08-10

🧪 Prototype check (2026-10-24)

e4c24e86b5.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before doc/timer.n:123 still has \fB monotonic\fR (stray space) and line 133 the unindented .LP Note that the output differs from after info. (the other typos from the ticket are already fixed on main)
  • Patch applied cleanly (doc only)
  • After man -l timer.n shows the note indented under the timer info item; troff warnings unchanged (1, pre-existing)
  • Test none possible (doc)
Full verdict

DOCS (mostly fixed on main; one item left)

  • Platform docs
  • Tested doc/timer.n on main (2026-10-16). All the typos were fixed in a14a16c336. The indentation of the after ms script and "Wall clock events" paragraphs was fixed by the .RS/.PP blocks cherry-picked in 151d53761e (2026-09-07). The only item left: "Note that the output differs from after info." still sits after the event-loop paragraph at the end of DESCRIPTION, not under timer info. (timer exists only in 9.1.)
  • Cause doc/timer.n
  • Prototype prototypes/e4c24e86b5.patch (8 changed lines: moves the note into the timer info item as an .RS/.PP block; also fixes the stray space in \fB monotonic). Checked with man -l.
  • Suggested ticket comment Indentation was fixed by 151d53761e. Proposed patch moves "Note that the output differs from after info" under the timer info item; after that the ticket can be closed.
⬜07000bd2372025Bignum documentation needs to be clarified w.r.t mp_int handlingpatchDOCS

Open ticket 07000bd237 · 56. LibTomMath · filed 2025-11-10

🧪 Prototype check (2026-10-24)

07000bd237.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before docs only; doc/IntObj.3 on tip still says nothing about who owns/clears the mp_int. Checked the claims against generic/tclObj.c: TclSetBignumInternalRep zeroes the mp_int (dp=NULL, so a later mp_clear is a no-op), GetBignumFromObj does mp_init_copy/mp_init_i64 or moves digits (Take, unshared) and returns TCL_ERROR without initializing on failure.
  • Patch applied cleanly (14 added lines)
  • After groff -man -Tutf8 doc/IntObj.3 renders both new paragraphs; no new warnings (only the usual man.macros ones).
  • Test none possible (docs)
Full verdict

DOCS

  • Platform docs
  • Tested read generic/tclObj.c (main): Tcl_SetBignumObj either mp_clear()s the value (fits a wide) or moves its digits into the object and zeroes the mp_int (TclSetBignumInternalRep); GetBignumFromObj does mp_init_copy / mp_init_i64 or moves the digits out (Take on unshared), so the caller always owns an initialized mp_int on success; Tcl_InitBignumFromDouble returns TCL_ERROR before mp_init for Inf/NaN. doc/IntObj.3 says none of this; only the source comment for Tcl_GetBignumFromObj does.
  • Cause doc/IntObj.3 lacks the ownership rules.
  • Prototype prototypes/07000bd237.patch (14 added lines in doc/IntObj.3: two paragraphs, for New/Set and for Get/Take/InitBignumFromDouble)
  • Suggested ticket comment Proposed doc patch for IntObj.3: New/SetBignumObj take over and clear the mp_int; Get/TakeBignumFromObj and Tcl_InitBignumFromDouble initialize it themselves and on success the caller must mp_clear it (on failure it is left uninitialized).
⬜858a6151942025oo::define supports classvariable but this is not really documentedpatchDOCS (partly NOT-A-BUG: there is no oo::define classvariable; classvariable is a method-body command documented in classvariable(n), which the …

Open ticket 858a615194 · 10. Objects · filed 2025-10-08

🧪 Prototype check (2026-10-24)

858a615194.patch: VERIFIED

  • Base main (tip d6dd6d0f71); docs only
  • Before oo::define oo::object classvariable x -> invalid command name "classvariable"; classvariable exists only as ::oo::Helpers::classvariable (documented in doc/classvariable.n), and define.n uses it in the TIP 478 example without saying so or listing it in SEE ALSO
  • Patch applied cleanly (7 changed lines; the "35 lines" in the triage counted context)
  • After man -l define.n renders the new sentence before the example and classvariable(n) in SEE ALSO
  • Test none possible (docs only)
Full verdict

DOCS (partly NOT-A-BUG: there is no oo::define classvariable; classvariable is a method-body command documented in classvariable(n), which the define.n example uses without saying so)

  • Platform docs
  • Tested main: oo::define oo::object classvariable x -> invalid command name; ::oo::define::* has no classvariable.
  • Cause doc/define.n example (TIP 478) uses classvariable without a pointer to classvariable(n); SEE ALSO lacks it. The earlier comment proposing to add it to the definitions list is mistaken.
  • Prototype prototypes/858a615194.patch (35 lines, doc only): one explanatory sentence before the example + classvariable(n) in SEE ALSO. Class constants: no mechanism, left as is.
  • Suggested ticket comment [classvariable] is not an oo::define subcommand but a command for use inside methods, documented in classvariable(n). Proposed doc patch: explain this in the define.n example and add classvariable(n) to SEE ALSO.
⬜81ebe684b42025TclFindCompiledLocal() header comments are falsepatchDOCS (confirmed in main, 9.0, 8.6 sources)

Open ticket 81ebe684b4 · 47. Bytecode Compiler · filed 2025-05-30

🧪 Prototype check (2026-10-24)

81ebe684b4.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before code reading of TclFindCompiledLocal on the tip: with envPtr->procPtr == NULL it returns TCL_INDEX_NONE for a NULL name and only looks up the local cache for named vars (never creates); the header says a temporary is "always created" and that create==1 always yields a new index.
  • Patch applied cleanly (27 changed lines, comment only); every new statement matches the code (procPtr NULL path, create/name NULL path, AnonymousLocal callers)
  • After tclCompile.c compiles without warnings; build OK
  • Test none possible (comment only)
Full verdict

DOCS (confirmed in main, 9.0, 8.6 sources)

  • Platform docs (internal comment)
  • Tested source reading of main generic/tclCompile.c; the 8.6/9.0 code has the same non-proc branch.
  • Cause generic/tclCompile.c:TclFindCompiledLocal: when envPtr->procPtr is NULL (non-body script) it only looks up the frame's LocalCache and returns TCL_INDEX_NONE for a NULL name (and never creates); this branch came with 302b0120aa (2008-06-08, 8.6 dev), after the 8.5 comment was written.
  • Prototype prototypes/81ebe684b4.patch (47 lines, comment only, no test): rewrites description/Results/Side effects to cover the non-proc case and the TCL_INDEX_NONE result of AnonymousLocal().
  • Suggested ticket comment Since 302b0120aa (2008) the non-proc branch never creates locals, so NULL names give TCL_INDEX_NONE there. Proposed corrected header comment in prototype patch.
⬜78f44214362025Tcl_InitStringRep has no description in doc/ObjectType.3patchDOCS (partly addressed: 8f9c7f5fd8, 2025-05-02, on main only; 9.0 still lacks it; 8.6 has no such API)

Open ticket 78f4421436 · 10. Objects · filed 2025-05-02

🧪 Prototype check (2026-10-24)

78f4421436.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before doc/ObjectType.3:119 still has the mislabelled "\fBTcl_FreeInternalRep\fR performs the function of ... TclInitStringRep" paragraph plus the duplicate from 8f9c7f5fd8, and "uninitialzed" twice
  • Patch applied cleanly (doc only)
  • After one real Tcl_FreeInternalRep paragraph, one Tcl_InitStringRep paragraph; checked against generic/tclObj.c (assert objPtr->bytes == NULL || bytes == NULL, Tcl_AttemptAlloc/Realloc, always NUL-terminated, returns NULL on failure, Tcl_FreeInternalRep = TclFreeInternalRep): accurate. troff warnings unchanged (1, pre-existing)
  • Test none possible (doc)
Full verdict

DOCS (partly addressed: 8f9c7f5fd8, 2025-05-02, on main only; 9.0 still lacks it; 8.6 has no such API)

  • Platform docs
  • Tested doc/ObjectType.3 in main, tcl9.0, tcl8.7.
  • Cause since 4da790cbea (2024-02-20) the Tcl_InitStringRep paragraph was mislabelled "Tcl_FreeInternalRep performs ..." (so Tcl_FreeInternalRep had no real description); 8f9c7f5fd8 then added a second copy with the right name, leaving a duplicate. Code check: Tcl_InitStringRep uses Tcl_AttemptAlloc/Realloc and does not panic; it returns NULL on failure.
  • Prototype prototypes/78f4421436.patch (56 lines, doc only): real Tcl_FreeInternalRep paragraph; one Tcl_InitStringRep paragraph (bytes must be NULL if a string rep exists, null termination, no panic, NULL on allocation failure); "uninitialzed" typo fixed. Applies to 9.0 too.
  • Suggested ticket comment The paragraph headed Tcl_FreeInternalRep actually described Tcl_InitStringRep, so main now has it twice and 9.0 has it under the wrong name. Proposed doc patch fixes both; should go to 9.0 as well.
⬜31fedb885e2025Tcl_ExprLong documentation omissionpatchDOCS (behaviour confirmed in main, 9.0 and 8.6 by reading the code)

Open ticket 31fedb885e · 69. Other · filed 2025-03-25

🧪 Prototype check (2026-10-24)

31fedb885e.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before tcltest testexprlong {} -> This is a result: 0 (TCL_OK), testexprdouble {} -> 0.0, testexprstring {} -> 0, while testexprlongobj {} and expr {} -> empty expression; doc/ExprLong.3 does not mention it
  • Patch applied cleanly (6 changed lines, docs)
  • After man -l doc/ExprLong.3 renders the new paragraph correctly and it matches the observed behaviour
  • Test none possible (documentation only)
Full verdict

DOCS (behaviour confirmed in main, 9.0 and 8.6 by reading the code)

  • Platform docs
  • Tested generic/tclBasic.c on main, 9.0 and 8.6: Tcl_ExprLong/Tcl_ExprDouble/Tcl_ExprBoolean store 0 and Tcl_ExprString sets the result to 0 for "" ("Legacy compatibility"). The *Obj variants raise "empty expression". doc/ExprLong.3 does not mention this.
  • Cause doc/ExprLong.3 omission (the code comment says this is deliberate legacy behaviour)
  • Prototype prototypes/31fedb885e.patch (6 changed lines, docs only; checked the rendering with man -l)
  • Suggested ticket comment Proposed doc fix (patch attached) documents the legacy behaviour: an empty string is not an error for Tcl_ExprLong/Double/Boolean/String, unlike expr and Tcl_ExprObj. Deprecating these functions (jan's suggestion) can be done separately.
⬜236b4d8e752024fcopy / chan copy documentationDOCS (still open: doc/fcopy.n is standalone, doc/chan.n copy is a shorter, partly different text)

Open ticket 236b4d8e75 · None · filed 2024-04-23

Full verdict

DOCS (still open: doc/fcopy.n is standalone, doc/chan.n copy is a shorter, partly different text)

  • Platform docs
  • Tested compared doc/chan.n and doc/fcopy.n on main; checked the -size unit on main, 9.0.5, 8.6.18, 8.5.19 (4 x "ä" in a utf-8 file, both channels utf-8, fcopy -size 2).
  • Differences (1) chan.n always says "characters", while fcopy.n says bytes for binary input and characters otherwise. (2) fcopy.n has a "Depreciated feature" paragraph: with the tcl8 profile and equal encodings the copy is binary and -size counts bytes. That is false on 9.0/main, where the copy gives 2 characters/4 bytes with both strict and tcl8; only 8.6/8.5 copy 2 bytes. (3) chan.n lacks fcopy.n's translation, encoding/profile and EILSEQ error-recovery sections. (4) chan.n has typos ("result in the error", "provides message"). fcopy.n also uses inchan/outchan where the synopsis says inputChan/outputChan, and spells it "Depreciated".
  • Prototype none (docs restructuring; needs a decision on which text wins). The facts above should make the merge mechanical: move the translation/encoding/error sections to chan.n, drop the tcl8 paragraph for 9.x, and make fcopy.n a pointer like puts.n.
  • Suggested ticket comment fcopy.n's "deprecated feature" paragraph (tcl8 profile + same encoding = binary copy, -size in bytes) does not match 9.0/main: fcopy -size 2 copies 2 characters with both strict and tcl8; only 8.6 copied bytes. Apart from that, fcopy.n is a superset of chan.n (translation, encoding/profile, EILSEQ recovery), so its text can move into chan.n and fcopy.n can reference it.
⬜a97680cbed2023Tcl_FSJoinToPath(): outdated commentpatchDOCS (still outdated in main, 9.0, 8.6)

Open ticket a97680cbed · 37. File System · filed 2023-07-25

🧪 Prototype check (2026-10-24)

a97680cbed.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before tip still has the stale "temporarily have their refCount increased by one ... will be freed" comments in generic/tclFileName.c (Tcl_FSJoinToPath, TclFSJoinPathHelper) and the "will be freed when this function returns" sentence in doc/FileSystem.3; the code just forwards to TclJoinPath (tclPathObj.c), which only Incr/Decrs its own temporaries
  • Patch applied with offset +5 (re-saved, 19 changed lines, content unchanged)
  • After libtcl compiles without warnings; groff -man -Tutf8 doc/FileSystem.3 renders the new text
  • Test none possible (comments/docs only)
Full verdict

DOCS (still outdated in main, 9.0, 8.6)

  • Platform docs
  • Tested source inspection: generic/tclFileName.c comments of Tcl_FSJoinToPath and TclFSJoinPathHelper (main), same comment in 9.0 and 8.6; the code just forwards to TclJoinPath (since 55c8c6c155), which does not touch the inputs' refcounts. doc/FileSystem.3 on main still says zero-refcount inputs "will be freed when this function returns", contradicting the general rule in the same page that zero-refcount values must not be passed.
  • Cause comments/docs not updated after 55c8c6c155 (dgp, TclJoinPath refactoring)
  • Prototype prototypes/a97680cbed.patch (62 lines, comments in tclFileName.c ×2 + FileSystem.3 paragraph; no test)
  • Suggested ticket comment Still outdated on all branches, and the Tcl_FSJoinToPath paragraph in FileSystem.3 has the same stale statement. Proposed comment/doc fix: reference counts of the arguments are not changed and must not be zero.
⬜5c2f6c2ff42021info procs/body/args only return already auto_loaded procspatchDOCS (behaviour by design; REPRODUCED on main, 8.6)

Open ticket 5c2f6c2ff4 · 38. Init - Library - Autoload · filed 2021-01-06

🧪 Prototype check (2026-10-24)

5c2f6c2ff4.patch: VERIFIED

  • Base main (tip d6dd6d0f71); docs only
  • Before tcltest 5c2f.tcl -> info body tcl_endOfWord and info default ... -> "tcl_endOfWord" isn't a procedure, info procs tcl_end* empty; after auto_load tcl_endOfWord -> tcl_endOfWord, args str start. Behaviour as the new text describes; info.n says nothing about it
  • Patch applied cleanly (4 changed lines)
  • After man -l info.n renders the new sentences under info procs
  • Test none possible (docs only)
Full verdict

DOCS (behaviour by design; REPRODUCED on main, 8.6)

  • Platform docs
  • Tested main build, 8.6.18: info body tcl_endOfWord errors "isn't a procedure" and info procs tcl_end* is empty until the proc is called.
  • Cause auto-loadable procs only exist in auto_index until unknown/auto_load sources them; info (generic/tclCmdIL.c) looks only at existing commands. Intended.
  • Prototype prototypes/5c2f6c2ff4.patch (4 changed lines, docs only: info.n info procs paragraph mentions not-yet-loaded procs and auto_load)
  • Suggested ticket comment This is by design: auto-loadable procedures do not exist until they are loaded, so info procs/args/body/default don't see them. Proposed doc note in info.n pointing to auto_load.
⬜21e3c44ac92020Table encoding limitations?patchDOCS

Open ticket 21e3c44ac9 · 10. Objects · filed 2020-04-23

🧪 Prototype check (2026-10-24)

21e3c44ac9.patch: VERIFIED

  • Base main (tip d6dd6d0f71); docs only
  • Before doc/Encoding.3 does not state the limit. On the tip, generic/tclEncoding.c TableEncodingData still has unsigned short toUnicode/fromUnicode (256-entry pages keyed by the lead byte), and encoding convertto shiftjis \U1F600 -> unexpected character at index 0: 'U+01F600' (with -profile replace -> ?), so the documented limit is real
  • Patch applied cleanly (6 changed lines)
  • After man -l doc/Encoding.3 renders the new paragraph under ENCODING FILES
  • Test none possible (docs only)
Full verdict

DOCS

  • Platform docs
  • Tested reading generic/tclEncoding.c on main: TableEncodingData toUnicode/fromUnicode are still unsigned short pages indexed by the high byte, so .enc table encodings are limited to 1–2 byte sequences and BMP characters (anything above U+FFFF goes to the fallback). doc/Encoding.3 only implies it ("2-byte Unicode", "double-byte numbers").
  • Cause by design of the .enc table format.
  • Prototype prototypes/21e3c44ac9.patch (6 changed lines, doc/Encoding.3 only): one paragraph stating the 2-byte/BMP limit and pointing to Tcl_CreateEncoding.
  • Suggested ticket comment Confirmed: table-based .enc files can only map up to 2-byte sequences to U+0000–U+FFFF; anything else needs Tcl_CreateEncoding(). Proposed doc clarification for doc/Encoding.3 (ENCODING FILES).
⬜6e156333bc2019Documentation: Contradiction in Tcl_GetSring/Tcl_GetStringFromObj vs Tcl_UpdateStringProcpatchDOCS

Open ticket 6e156333bc · 10. Objects · filed 2019-07-22

🧪 Prototype check (2026-10-24)

6e156333bc.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before doc/StringObj.3:188-191 still says the writable pointer lets a Tcl_UpdateStringProc modify the string rep (contradicting ObjectType.3: bytes is NULL when updateStringProc is called)
  • Patch applied cleanly (doc only)
  • After text says writable for historical reasons, read-only in practice, updateStringProc builds the rep itself (Tcl_InitStringRep); troff warnings unchanged (1, pre-existing)
  • Test none possible (doc)
Full verdict

DOCS

  • Platform docs
  • Tested doc/StringObj.3 on main, 9.0 and 8.6 still says the pointer from Tcl_GetString is writable "so that extension author ... will be able to modify the string representation within the Tcl_UpdateStringProc", while ObjectType.3 says the updateStringProc is only called when bytes is NULL (calling Tcl_GetString from there would recurse). The sentence comes from b1c464dbca (msofer).
  • Cause misleading sentence in doc/StringObj.3.
  • Prototype prototypes/6e156333bc.patch (10 changed lines, doc only, for 9.x since it names Tcl_InitStringRep): says the pointer is writable for historical reasons but read-only in practice, and that an updateStringProc builds the rep itself (e.g. with Tcl_InitStringRep) and does not get it through Tcl_GetString. For 8.6 the same text without the Tcl_InitStringRep mention.
  • Suggested ticket comment The reporter is right: an updateStringProc runs only when bytes is NULL, so it cannot use Tcl_GetString(). Proposed doc fix for StringObj.3: drop that rationale and say the returned pointer must be treated as read-only.
⬜e7dea09f582018Doc glitch in NRE(3)patchDOCS (still present on main, 9.0, 8.6)

Open ticket e7dea09f58 · 60. NRE and coroutines · filed 2018-03-26

🧪 Prototype check (2026-10-24)

e7dea09f58.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before doc/NRE.3 EXAMPLE on the tip still has TheCmdPostProc vs TheCmdNRPostProc, \fITheCmdNRObjProc\fR without (, "a a wrapper", Tcl_NRCallObjProc in the text vs Tcl_NRCallObjProc2 in the code, "possible implement"
  • Patch applied cleanly (12 changed lines, unchanged)
  • After all names in the EXAMPLE section consistent (TheCmdOldObjProc / TheCmdNRObjProc / TheCmdNRPostProc, Tcl_NRCallObjProc2 in text and code); groff -man -ww gives only the usual man.macros warnings. chw's androwish diff (linked in the ticket) could not be compared (no network).
  • Test none possible (docs only)
Full verdict

DOCS (still present on main, 9.0, 8.6)

  • Platform docs
  • Tested by reading doc/NRE.3 on main, 9.0 and 8.6. The androwish diffs linked in the ticket cannot be fetched (no network), so the fix is reconstructed from the EXAMPLE section: the post-processing callback is called TheCmdPostProc in the text and the Tcl_NRAddCallback call but defined as TheCmdNRPostProc; TheCmdNRObjProc is missing its (; "a a wrapper"; the text says Tcl_NRCallObjProc while the code calls Tcl_NRCallObjProc2; Tcl_EvalObjEx/Tcl_NREvalObj are italic instead of bold; "possible implement".
  • Cause doc/NRE.3 EXAMPLE section.
  • Prototype prototypes/e7dea09f58.patch (12 changed lines, docs only, against main; 8.6 needs the same edits except the ...2 names)
  • Suggested ticket comment The EXAMPLE section of NRE.3 is still inconsistent (TheCmdPostProc vs TheCmdNRPostProc, a missing parenthesis, "a a"). Proposed doc fix. (Please check it against chw's androwish diff 0d82e5310a.)
⬜f4c118f67c2018Improve documentation on timezones for the clock commandpatchDOCS (still valid: clock.n TIME ZONES lists only the legacy abbreviations and gives no guidance on which location names to use; the list is also ou…

Open ticket f4c118f67c · 16. Commands A-H · filed 2018-01-25

🧪 Prototype check (2026-10-24)

f4c118f67c.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before doc/clock.n:820-832 lists only the old abbreviations (no akst, akdt, fwt, fst, kst, jdt, kdt, military letters) and no guidance; rf4c.tcl on the tip: -timezone akst=-0900 akdt=-0800 fwt=+0100 fst=+0200 kst=+0900 jdt/kdt=+1000 cst=-0600 ist=+0530 a=+0100 z=+0000, j -> time zone ":j" not found (free-form scan accepts the same ones)
  • Patch applied cleanly (doc only, 20 changed lines)
  • After list and the new paragraph (cst -0600, ist +0530, military a-z except j) match that behaviour; troff warnings unchanged (1, pre-existing)
  • Test none possible (doc)
Full verdict

DOCS (still valid: clock.n TIME ZONES lists only the legacy abbreviations and gives no guidance on which location names to use; the list is also out of date)

  • Platform docs
  • Tested doc/clock.n on main, 9.0, 8.6 (same text) against the LegacyTimeZone table in library/clock.tcl
  • Cause the abbreviation list in clock.n lacks akst, akdt, fwt, fst, kst, jdt, kdt and the military single letters, which clock.tcl accepts; nothing says that the abbreviations are ambiguous fixed offsets
  • Prototype prototypes/f4c118f67c.patch (main, doc only, 28 changed lines): completes the list, mentions the military letters, and adds a paragraph that the abbreviations are ambiguous fixed offsets (cst = -0600, ist = +0530) with examples of :America/Chicago, :Asia/Shanghai, :Asia/Kolkata, :Europe/London. A full legacy-to-location table as the reporter suggests was not added (many abbreviations have several meanings, see the comments of dc69e5ecf0). Checked with man -l. Side note: the tzdata path in the text is the odd "/no_backup/tools/lib/tcl9.0/clock/tzdata" (tcl8.5 in 8.6)
  • Suggested ticket comment Proposed doc update for clock.n in branch (name tbd): the abbreviation list now matches what clock.tcl accepts, and a new paragraph says that these are ambiguous fixed offsets and recommends location names such as :America/Chicago or :Asia/Shanghai. A complete mapping table is not practical because many abbreviations have several meanings.
⬜3c05b4568a2017Documentation for Tcl_EvalObj() is missingpatchDOCS (still missing on main, 9.0, 8.6)

Open ticket 3c05b4568a · 45. Parsing and Eval · filed 2017-11-21

🧪 Prototype check (2026-10-24)

3c05b4568a.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before generic/tclDecls.h:4260 still defines Tcl_EvalObj(interp, objPtr) (deprecated block); doc/Eval.3 does not mention it at all
  • Patch applied cleanly (8 changed lines, docs)
  • After man -l doc/Eval.3 shows Tcl_EvalObj in NAME and SYNOPSIS and the new "older procedure that is now deprecated ... equivalent to Tcl_EvalObjEx with flags set to 0" paragraph
  • Test none possible (documentation only)
Full verdict

DOCS (still missing on main, 9.0, 8.6)

  • Platform docs
  • Tested by reading. Tcl_EvalObj is still a macro for Tcl_EvalObjEx(interp, objPtr, 0) in generic/tclDecls.h (main: "Deprecated Tcl procedures" block; also in 9.0 and 8.6), but doc/Eval.3 lists only its sibling Tcl_GlobalEvalObj (NAME, SYNOPSIS and the "older procedures that are now deprecated" paragraph). It was dropped from the docs in f3b32fb71c (1999).
  • Cause doc/Eval.3.
  • Prototype prototypes/3c05b4568a.patch (8 changed lines, docs only: adds Tcl_EvalObj to NAME and SYNOPSIS, plus a sentence saying it is deprecated and equivalent to Tcl_EvalObjEx with flags 0)
  • Suggested ticket comment Tcl_EvalObj still exists as a compatibility macro (Tcl_EvalObjEx with flags 0) but is missing from Eval.3, while Tcl_GlobalEvalObj is documented. Proposed doc fix that documents it as deprecated next to Tcl_GlobalEvalObj.
⬜3121e790602017Create manpages for {tcl,tk}::pkgconfigpatchDOCS (still missing in main/9.0/8.6; only RegConfig.3 describes the generic list/get interface, tm.n mentions tcl::pkgconfig get libdir,runtime)

Open ticket 3121e79060 · 54. Configuration Reporting · filed 2017-11-01

🧪 Prototype check (2026-10-24)

3121e79060.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before no doc/pkgconfig.n on the tip (only RegConfig.3 and a tm.n mention)
  • Patch applied cleanly (new file); one wording fix: 64bit described as "a 64-bit build; on Unix only when configured with --enable-64bit" (TCL_CFG_DO64BIT is also set by the Windows makefiles for x64; on 64-bit Linux the key is 0) (52 lines now)
  • After tcl::pkgconfig list on the tip build gives exactly the keys the page lists (8 booleans, 5 dirs x install/runtime, dllfile,runtime); get nokey -> error key not known; groff -man -Tutf8 -ww doc/pkgconfig.n renders without warnings beyond the missing man.macros
  • Test none possible (docs only)
Full verdict

DOCS (still missing in main/9.0/8.6; only RegConfig.3 describes the generic list/get interface, tm.n mentions tcl::pkgconfig get libdir,runtime)

  • Platform docs
  • Tested doc/ of main, Tk main
  • Cause n/a
  • Prototype prototypes/3121e79060.patch (main, new doc/pkgconfig.n, 52 lines; renders with nroff): synopsis, list/get, all keys of tclPkgConfig.c (booleans, *,install / *,runtime paths, dllfile,runtime), example, SEE ALSO Tcl_RegisterConfig. tk::pkgconfig would need its own page in Tk (keys differ: fontsystem etc.).
  • Suggested ticket comment Proposed doc/pkgconfig.n describing tcl::pkgconfig and its keys; Tk needs a matching page for tk::pkgconfig.
⬜9b3ca8dcad2017auto_load docs are liespatchDOCS (confirmed on main, 9.0, 8.6, 8.4)

Open ticket 9b3ca8dcad · 38. Init - Library - Autoload · filed 2017-06-22

🧪 Prototype check (2026-10-24)

9b3ca8dcad.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before in a child interp with unset auto_path; set auto_index(zz) {proc zz {} {return ok}}: auto_load zz -> 1 (zz -> ok), auto_load parray -> 0 (no TCLLIBPATH/library fallback, as the doc claims); TCLLIBPATH=/nonexist/x -> auto_path starts with /nonexist/x <wt>/library ... (initialized at startup). main and 8.6.
  • Patch applied cleanly (10 changed lines)
  • After groff 4 warnings (same 4 before)
  • Test none possible (docs only)
Full verdict

DOCS (confirmed on main, 9.0, 8.6, 8.4)

  • Platform docs
  • Tested work-a2/9b3ca8dcad.tcl: with auto_path unset and env(TCLLIBPATH) pointing to a dir with a tclIndex, auto_load returns 0 on all versions. It does not fall back to TCLLIBPATH or the library dir.
  • Cause library/init.tcl:auto_load returns 0 when auto_path does not exist (if {![info exists auto_path]} {return 0}); TCLLIBPATH is only used by tcl::InitAutoPath at startup (as tclvars.n and library.n's env(TCLLIBPATH) entry already say).
  • Prototype prototypes/9b3ca8dcad.patch (10 changed lines, doc/library.n only): auto_path is initialized from TCLLIBPATH and the library dirs; with no auto_path only existing auto_index entries are used. No other misstatements found in the auto_load paragraph.
  • Suggested ticket comment Confirmed: auto_load never reads env(TCLLIBPATH); that variable only seeds auto_path at startup, and without auto_path auto_load just returns 0. Proposed doc fix for library.n attached.
⬜994f9d7a202017Tcl_ParseVarpatchDOCS (behaviour is intentional, the same on main and 8.6)

Open ticket 994f9d7a20 · - New Builtin Commands · filed 2017-06-17

🧪 Prototype check (2026-10-24)

994f9d7a20.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before testparsevar on the tip: {$} -> {$} {}, {$ x} -> {$} { x}, {$.} -> {$} ., {$x} -> 5 {}; doc/ParseCmd.3 does not mention the lone-$ result
  • Patch applied cleanly (5 changed lines, unchanged)
  • After groff -man -Tutf8 doc/ParseCmd.3 renders the new sentence correctly; it matches the observed behaviour
  • Test none possible (docs only)
Full verdict

DOCS (behaviour is intentional, the same on main and 8.6)

  • Platform docs
  • Tested testparsevar {$}, {$ x}, {$.} on the main build and the 8.6 tip tcltest: each returns $ with termPtr just past the $.
  • Cause generic/tclParse.c:Tcl_ParseVar returns the static string "$" when Tcl_ParseVarName produces a single TCL_TOKEN_TEXT token (no variable name follows), matching the script-level rule that a lone $ is literal. doc/ParseCmd.3 only mentions "value of the variable" and NULL.
  • Prototype prototypes/994f9d7a20.patch (5 changed lines, doc/ParseCmd.3): documents the lone-$ case.
  • Suggested ticket comment Returning "$" is deliberate: a $ without a following variable name is no substitution, same as in scripts. Proposed doc addition to ParseCmd.3 attached.
⬜f44329748b2016lsort: obsolete parts in the documentationpatchDOCS (confirmed: the paragraph is wrong since 8.5 and still in main, 9.0, 8.6 lsort.n)

Open ticket f44329748b · 17. Commands I-L · filed 2016-08-25

🧪 Prototype check (2026-10-24)

f44329748b.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before lsort -integer {abc} -> error expected integer but got "abc" on main and 8.6, so the NOTES paragraph (options do not constrain values for lists of fewer than two elements) is wrong.
  • Patch applied cleanly (5 changed lines, paragraph removed)
  • After groff 0 warnings (0 before)
  • Test none possible (docs only)
Full verdict

DOCS (confirmed: the paragraph is wrong since 8.5 and still in main, 9.0, 8.6 lsort.n)

  • Platform docs
  • Tested lsort -integer 0.0 errors on main and 8.6.18 ("expected integer but got 0.0") and returns 0.0 on 8.4.20; lsort -real abc errors on 8.6+ and is accepted on 8.4.
  • Cause since 8.5, lsort validates the elements of a one-element list; the paragraph in doc/lsort.n describes the 8.4 behaviour.
  • Prototype prototypes/f44329748b.patch (5 changed lines, doc/lsort.n): removes the paragraph.
  • Suggested ticket comment Agreed: since 8.5 a single element is validated too (lsort -integer 0.0 errors). Proposed fix: drop the obsolete paragraph from lsort.n.
⬜f8bc8efdf52016Tutorial lesson 15 uses listName inconsistently with lappend/linsert/lreplaceDOCS (claim valid; current tutorial text not checkable offline)

Open ticket f8bc8efdf5 · 75. Tutorial · filed 2016-08-23

Full verdict

DOCS (claim valid; current tutorial text not checkable offline)

  • Platform docs
  • Tested main, 8.6: linsert mylist 1 x -> mylist x (name used as a literal list); linsert $mylist 1 oneandahalf gives the expected result. doc/linsert.n and lreplace.n correctly use list.
  • Cause Lesson 15 (Tcl15.html) signatures/examples pass a list name to linsert/lreplace, which take a list value.
  • Prototype none (tutorial source not in tcl.fossil)
  • Suggested ticket comment linsert and lreplace take a list value, not a variable name; Lesson 15 should use list in the synopses and $mylist in the examples.
⬜9a829af10f2015Source encoding for modules (tcl::tm) not correctly documentedpatchDOCS (8.6 only; on 9.x the man page is already accurate)

Open ticket 9a829af10f · 39. Package Manager · filed 2015-12-09

🧪 Prototype check (2026-10-24)

9a829af10f.patch: VERIFIED

  • Base core-8-6-branch (tip 326e1bf923)
  • Before docs only. On the 8.6 tip library/tm.tcl:270 registers source -encoding utf-8 $file; under LANG=C a module holding UTF-8 é loads as U+00E9 (scan -> 233), while doc/tm.n shows plain source module_file
  • Patch applied cleanly (9 changed lines)
  • After groff renders, no new warnings. (The ifneeded script really is package provide X V;source -encoding utf-8 F; the doc's [list source -encoding utf-8 MF] keeps the page's existing simplification.)
  • Test none possible (docs)
Full verdict

DOCS (8.6 only; on 9.x the man page is already accurate)

  • Platform docs
  • Tested read library/tm.tcl and doc/tm.n on main, 9.0, 8.6
  • Cause 8.6 tm.tcl registers source -encoding utf-8 $file, but doc/tm.n says modules are loaded with plain source module_file, which in 8.6 means the system encoding. In 9.0/9.1 tm.tcl uses plain source $file, and source defaults to utf-8 there, so the man page is right.
  • Prototype prototypes/9a829af10f.patch (against core-8-6-branch, doc only, 9 changed lines: source -encoding utf-8 in the three examples plus one sentence saying modules are always read as utf-8)
  • Suggested ticket comment In 9.0+ source defaults to utf-8, so doc/tm.n is correct there. 8.6 tm.n still shows plain source. Proposed doc fix for core-8-6-branch: use source -encoding utf-8 in the examples and state that modules are always read as utf-8.
🔧fe4fca32d92015mixed-greediness regular expressions need better documentationDOCS (behaviour as described on main, 9.0, 8.6)

Open ticket fe4fca32d9 · - New Builtin Commands · filed 2015-03-24

🔧 Since triage (updated 2026-11-11)

Status now: Open

  • 🔧 2026-10-11 check-in 1512c7ee75 on doc-re-syntax-preference, not merged: Document how the first quantifier sets the preference of the whole RE (\[fe4fca32d9\]).
  • 💬 2026-11-10 oehhar: Serhiy proposed to merge this with ticket 524b35a8b5.
Full verdict

DOCS (behaviour as described on main, 9.0, 8.6)

  • Platform docs
  • Tested reporter's examples give 12 1 2 and 1234 1 234 on all three. All examples in the branch's new re_syntax.n text checked on main and 9.0: all give the documented results (incl. (?:(.+?), (.+)){1,1} -> {foo, bar} foo bar).
  • Cause re_syntax.n MATCHING NOTE was vague about "first quantifier decides the preference of the whole RE".
  • Prototype none (Serhiy's docs branch exists)
  • Existing work doc-re-syntax-preference (tip 1512c7ee75, single check-in) unmerged; doc/re_syntax.n is unchanged on main since the branch base, so it merges cleanly (dry run)
  • Suggested ticket comment none needed beyond Serhiy's; ready to merge (docs only). The examples still hold after the {1,1} fix b0b515295b for 524b35a8b5.
🔧7b0e690b702014chan push returns channel IDDOCS (duplicate of 2920003fff; same behaviour on main, 9.0.5, 8.6.18)

Open ticket 7b0e690b70 · 26. Channel Transforms · filed 2014-10-05

🔧 Since triage (updated 2026-11-11)

Status now: Open

  • 🔧 2026-10-14 check-in 766139d4e1 on doc-chan-push-result, not merged: Fix 7b0e690b70 and [2920003]: document that [chan push] returns the name of the channel, and that the handle passed to the transformation handler is not that
Full verdict

DOCS (duplicate of 2920003fff; same behaviour on main, 9.0.5, 8.6.18)

  • Platform docs
  • Tested same script as 2920003fff.
  • Cause see 2920003fff.
  • Existing work doc-chan-push-result (766139d4e1), unmerged, fixes both tickets.
  • Prototype none (branch exists)
  • Suggested ticket comment none needed; close together with 2920003fff when the branch is merged.
⬜3ce04c18732013Return command documentation RETRUN OPTION section "misleading"patchDOCS (main, 9.0, 8.6: same text)

Open ticket 3ce04c1873 · 18. Commands M-Z · filed 2013-08-26

🧪 Prototype check (2026-10-24)

3ce04c1873.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before doc/return.n:123/132 unchanged ("will also be stored in the global variable errorCode/errorInfo"); r3ce.tcl: catch {return -code error ... -errorcode {foo bar} foobar} -> code 2, errorCode stays NONE; via a proc -> code 1, errorCode foo bar
  • Patch applied cleanly (doc only)
  • After the new text (set only when the error is raised, i.e. -level reaches 0; catch/try directly around return see TCL_RETURN) matches that; troff warnings unchanged (1, pre-existing), renders fine
  • Test none possible (doc)
Full verdict

DOCS (main, 9.0, 8.6: same text)

  • Platform docs
  • Tested 9.0.5, 8.6.18: catch {return -code error -errorinfo {Oh no!} -errorcode {foo bar} foobar} gives 2 (TCL_RETURN) and leaves ::errorInfo/::errorCode unchanged. The same return inside a proc that is caught sets both.
  • Cause doc/return.n says -errorcode/-errorinfo are "also stored in the global variable" without saying this happens only when the error is actually raised, i.e. when the -level count reaches 0.
  • Prototype prototypes/3ce04c1873.patch (doc only, 10 changed lines, doc/return.n on main; applies to 8.6 as well since the text is identical)
  • Suggested ticket comment The globals are set only when the error is raised, i.e. when the -level count reaches 0. Evaluated directly in catch, return produces TCL_RETURN and no error. Proposed wording in doc/return.n.
⬜197e2bf85c2013info frame 'level' does not match documentationpatchDOCS (8.6 text is wrong; on 9.0/9.1 the entry now just says "(to be determined)")

Open ticket 197e2bf85c · - New Builtin Commands · filed 2013-07-14

🧪 Prototype check (2026-10-24)

197e2bf85c.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before doc/info.n:210 still says For a frame that corresponds to a level, (to be determined).; behaviour (rdoc1.tcl): from proc b called by proc a, info frame -1/-2 give level 1 / 2 (uplevel-style, positive), and uplevel 1 {info frame 0} has no level entry
  • Patch applied cleanly (doc only)
  • After the new text matches that behaviour; tbl info.n | troff -man -ww -z: 0 warnings before and after; man -l renders the paragraph correctly
  • Test none possible (doc)
Full verdict

DOCS (8.6 text is wrong; on 9.0/9.1 the entry now just says "(to be determined)")

  • Platform docs
  • Tested 9.0.5, 8.6.18 with the reporter's script: same output as reported (frame 4 -> level 1, frame 1 -> level 2, and level present for the eval frames too). Inside uplevel #0, the frames of the uplevel'd script have no level.
  • Cause generic/tclCmdIL.c:TclInfoFrame computes level as varFramePtr->level - framePtr->framePtr->level, i.e. an uplevel-style relative level (>= 0) of the variable frame the command ran in. It is added only if that CallFrame is on the current callerVarPtr chain.
  • Prototype prototypes/197e2bf85c.patch (doc only, 8 changed lines, doc/info.n on main; 8.6's info.n needs the same wording in its own paragraph)
  • Suggested ticket comment level is the uplevel-style relative level (0 or positive) of the variable frame the command ran in, so uplevel $level gives its context. eval/source frames share their caller's level. Proposed doc wording for info.n (9.x currently says "to be determined").
⬜3588809fff2012[namespace unknown] man page entry is in errorpatchDOCS (still wrong on main; behaviour is as the reporter says on main, 9.0, 8.6, 8.5)

Open ticket 3588809fff · 21. [namespace] · filed 2012-11-20

🧪 Prototype check (2026-10-24)

3588809fff.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before v4docs3.tcl: namespace unknown -> ::unknown in the global namespace, {} in a new namespace; after namespace unknown foo; namespace unknown {} -> {} again (main and 8.6). Doc says the default is ::unknown for all namespaces.
  • Patch applied cleanly (5 changed lines)
  • After groff 0 warnings (0 before)
  • Test none possible (docs only)
Full verdict

DOCS (still wrong on main; behaviour is as the reporter says on main, 9.0, 8.6, 8.5)

  • Platform docs
  • Tested main, 9.0.5, 8.6.18, 8.5.19: [namespace eval foo {namespace unknown}] returns "" and an unknown command in foo goes to the global handler (also a custom one set with [namespace unknown ::myunk]).
  • Cause doc/namespace.n says "The default handler for all namespaces is ::unknown".
  • Prototype prototypes/3588809fff.patch (5 changed lines, docs only). It says ::unknown is the default for the global namespace, other namespaces default to empty and so use the global handler, and an empty script restores the default.
  • Suggested ticket comment Confirmed, the man page is still wrong. Non-global namespaces default to an empty handler, which defers to the global namespace's handler. Proposed doc fix for namespace.n is available.
⬜3547034fff2012Tcl_InvalidateStringRep documentation on shared objectspatchDOCS (main, 9.0, 8.6 man pages say nothing about shared values)

Open ticket 3547034fff · None · filed 2012-07-22

🧪 Prototype check (2026-10-24)

3547034fff.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before doc/Object.3 on the tip says nothing about shared values; Tcl_InvalidateStringRep (tclObj.c) still does no shared check
  • Patch applied cleanly (9 changed lines, unchanged)
  • After groff -man -Tutf8 doc/Object.3 renders the new paragraph correctly
  • Test none possible (docs only)
Full verdict

DOCS (main, 9.0, 8.6 man pages say nothing about shared values)

  • Platform docs
  • Tested read doc/Object.3 and doc/ObjectType.3 on main, 9.0, 8.6
  • Cause doc/Object.3 describes Tcl_InvalidateStringRep without saying it must not be used on a shared value.
  • Prototype prototypes/3547034fff.patch (9 changed lines, doc only, doc/Object.3 after the Tcl_InvalidateStringRep paragraph; the same text applies to 9.0 and 8.6; checked with groff)
  • Suggested ticket comment Proposed doc addition to Object.3: Tcl_InvalidateStringRep must only be called on an unshared value, and a shared value that has a string representation must keep it unchanged, because a regenerated string is not guaranteed to be identical.
⬜3388256fff2011insufficient clarity in Tutorial 31DOCS (RFE-like improvement request; not checkable offline)

Open ticket 3388256fff · 75. Tutorial · filed 2011-08-08

Full verdict

DOCS (RFE-like improvement request; not checkable offline)

  • Platform docs
  • Tested none (text only)
  • Cause Lesson 31 (packages) mentions pkgIndex.tcl only in a bullet list; the tutstack example does not show it.
  • Prototype none (tutorial source not in tcl.fossil)
  • Suggested ticket comment Lesson 31's example should include the pkgIndex.tcl (or a Tcl Module file name) and the auto_path step explicitly.
⬜3376407fff2011Tcl_ParseCommand documentation bytes vs. charspatchDOCS (still says "characters" on main)

Open ticket 3376407fff · 45. Parsing and Eval · filed 2011-07-23

🧪 Prototype check (2026-10-24)

3376407fff.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before docs only; doc/ParseCmd.3:248 on the tip says "number of characters". generic/tcl.h says /* Number of bytes in token. */, and tclParse.c sets sizes as src - tokenPtr->start (byte pointers).
  • Patch applied cleanly (1 word)
  • After groff renders, no new warnings
  • Test none possible (docs)
Full verdict

DOCS (still says "characters" on main)

  • Platform docs
  • Tested doc/ParseCmd.3 on main; generic/tcl.h says Tcl_Size size; /* Number of bytes in token. */ and the parser computes sizes as byte offsets
  • Cause doc/ParseCmd.3 line 248: "the size field gives the total number of characters in the token"
  • Prototype prototypes/3376407fff.patch (2 changed lines, docs only: "characters" -> "bytes")
  • Suggested ticket comment Confirmed: Tcl_Token.size is a byte count (tcl.h says so, like commentSize/commandSize). Proposed one-word fix to ParseCmd.3 is available.
⬜3082225fff2010Clock doc says error to use -timezone with free-form scanpatchDOCS (the doc is wrong on main, 9.0, 8.6, 8.5; the code accepts and uses -timezone)

Open ticket 3082225fff · 06. Time Measurement · filed 2010-10-06

🧪 Prototype check (2026-10-24)

3082225fff.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before doc/clock.n:900 still says -timezone gives an error with free-form scan; clock scan 10/12/2010 -timezone :UTC -> 1286841600 (no error), clock scan 10/12/2010 -locale fr -> legacy [clock scan] does not support -locale
  • Patch applied cleanly (doc only)
  • After rendered text "only the -base, -gmt and -timezone options are accepted. The -locale option will result in an error if -format is not supplied." matches the behaviour; troff warnings unchanged (1, pre-existing)
  • Test none possible (doc)
Full verdict

DOCS (the doc is wrong on main, 9.0, 8.6, 8.5; the code accepts and uses -timezone)

  • Platform docs
  • Tested main build, 9.0.5, 8.6.18, 8.5.19: clock scan 10/12/2010 -timezone CST5CDT gives 1286856000 and -timezone :UTC gives 1286841600; -locale de gives "legacy [clock scan] does not support -locale"
  • Cause doc/clock.n FREE FORM SCAN says only -base and -gmt are accepted and that -timezone and -locale give an error. Only -locale does.
  • Prototype prototypes/3082225fff.patch (6 changed lines, doc only, against main; the same paragraph is in 8.6 clock.n)
  • Suggested ticket comment Free-form scan accepts -timezone and uses it for the default zone; only -locale is rejected. Proposed doc fix: list -timezone among the accepted options.
⬜3056056fff2010tailcall's testsuite and docs incompletepatchDOCS

Open ticket 3056056fff · 60. NRE and coroutines · filed 2010-08-30

🧪 Prototype check (2026-10-24)

3056056fff.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before docs only; tailcall.n on the tip still says tailcall may not be used inside catch and requires command. work-a3/v3056.tcl on main and 8.6.18 confirms every statement in the new text: a caught tailcall gives code 2 and the proc still returns x; a second caught tailcall wins (y); catch {tailcall} cancels the pending one (the proc returns end); try ... on return catches it too; a bare tailcall returns "".
  • Patch applied cleanly (13 changed lines)
  • After groff renders, no new warnings
  • Test none possible (docs; tailcall.test already covers the catch cases, tailcall-12.x)
Full verdict

DOCS

  • Platform docs
  • Tested main, 8.6.18, 8.6.0: catch {tailcall list x} returns 2 and the proc still returns x at its end; a second caught tailcall replaces the first; tailcall with no args is accepted (synopsis says command is required) and cancels a pending tailcall. tailcall.test now has 40 tests incl. catch cases (tailcall-12.x), and the related tickets are about behaviour already covered.
  • Cause doc/tailcall.n says nothing about TCL_RETURN, the no-argument form or catching, and wrongly says it may not be used inside [catch].
  • Prototype prototypes/3056056fff.patch (13 changed lines, docs only)
  • Suggested ticket comment The test suite has grown (tailcall.test, 40 tests incl. catch), but tailcall.n still lacks the TCL_RETURN, caught-tailcall and no-argument semantics. Proposed doc update attached to the triage (synopsis ?command?, one paragraph).
⬜2987629fff2010== operator precedence?DOCS (not checkable offline)

Open ticket 2987629fff · 75. Tutorial · filed 2010-04-15

Full verdict

DOCS (not checkable offline)

  • Platform docs
  • Tested doc/expr.n on main/9.0/8.6 lists == != in the precedence table (OK there).
  • Cause Lesson 6 (Tcl6.html) operator list omits == and !=.
  • Prototype none (tutorial source not in tcl.fossil)
  • Suggested ticket comment expr.n lists ==/!=; the tutorial's operator list in Lesson 6 should too.
⬜2987620fff2010tutorial: typo in regexps (chap More Examples Of Regular ExpDOCS (claim confirmed; current tutorial text not checkable offline)

Open ticket 2987620fff · 75. Tutorial · filed 2010-04-15

Full verdict

DOCS (claim confirmed; current tutorial text not checkable offline)

  • Platform docs
  • Tested main, 8.6: {([0-9])*\.?([0-9]*)} on 87.65 gives first=7; {([0-9]*)...} gives 87. {"([^"])*"} gives u, {"([^"]*)"} gives tavu.
  • Cause Tcl20a.html has the quantifier outside the capture group (([0-9])*, ([^"])*).
  • Prototype none (tutorial source not in tcl.fossil)
  • Suggested ticket comment Confirmed: ([0-9])* captures only the last digit; Tcl20a should use ([0-9]*) and "([^"]*)".
⬜2963637fff2010Lesson 39 text/example mismatchDOCS (not checkable offline)

Open ticket 2963637fff · 75. Tutorial · filed 2010-03-04

Full verdict

DOCS (not checkable offline)

  • Platform docs
  • Tested none (text only)
  • Cause Lesson 39 (time) text describes a loop with variable k and multiplication; the example compares lsearch with array lookup.
  • Prototype none (tutorial source not in tcl.fossil)
  • Suggested ticket comment Lesson 39's text should be rewritten to describe the lsearch-vs-array example it actually shows.
⬜2962834fff2010Bugfixes in tutorial examplesDOCS (patch attached; not checkable offline)

Open ticket 2962834fff · 75. Tutorial · filed 2010-03-03

Full verdict

DOCS (patch attached; not checkable offline)

  • Platform docs
  • Tested attachment 4a0c77ea97 read: fixes unterminated puts "Directories: / "Files: in Tcl25.html (real syntax errors) and replaces chan copy/chan close by fcopy/close in Tcl26a.html (8.4 compat; unnecessary for 8.5+). Lesson 40 echo server issues (no -blocking 0, gets<0 treated as EOF) are real Tcl pitfalls.
  • Cause example code errors in lessons 25, 26a, 40.
  • Prototype none (attachment is against tcltutorial; Lesson 25 hunk applicable as is, 26a hunk should be dropped since the tutorial targets 8.5+)
  • Suggested ticket comment The Lesson 25 hunk fixes real syntax errors; the 26a hunk is not needed for 8.5+. Lesson 40 should check &#91;eof&#93; and &#91;fblocked&#93; instead of treating gets &lt; 0 as EOF.
⬜2932053fff2010Doc for Itcl C APIDOCS (still valid upstream)

Open ticket 2932053fff · 81. Bundled Packages · filed 2010-01-14

Full verdict

DOCS (still valid upstream)

  • Platform docs (Itcl, upstream)
  • Tested itcl.fossil trunk doc/*.3: Object.3 still has NAME "Itcl_CreateObject, Itcl_DeleteObject, Itcl_FindObject, Itcl_IsObject, Itcl_IsObjectIsa" but its SYNOPSIS/ARGUMENTS describe Itcl_PreserveData/ReleaseData/EventuallyFree (copy of Preserve.3); the other pages are minimal.
  • Cause Itcl doc/Object.3 copy-paste; C API never documented.
  • Prototype none (documentation writing in the Itcl repo, not a small fix)
  • Suggested ticket comment Still true in Itcl trunk (Object.3 synopsis is a copy of Preserve.3). This belongs to the Itcl tracker; suggest moving it there and closing here.
🔧2920003fff2009chan push does not return the documented valueDOCS (behaviour confirmed: [chan push] returns the channel name, the handler gets a separate rtN handle, on main, 9.0.5, 8.6.18)

Open ticket 2920003fff · 24. Channel Commands · filed 2009-12-23

🔧 Since triage (updated 2026-11-11)

Status now: Open

  • 🔧 2026-10-14 check-in 766139d4e1 on doc-chan-push-result, not merged: Fix 7b0e690b70 and [2920003]: document that [chan push] returns the name of the channel, and that the handle passed to the transformation handler is not that
Full verdict

DOCS (behaviour confirmed: [chan push] returns the channel name, the handler gets a separate rtN handle, on main, 9.0.5, 8.6.18)

  • Platform docs
  • Tested work-a3/cp62.tcl on main (own build), 9.0.5, 8.6.18: init handle=rt0, push returns=file5.
  • Cause doc/chan.n and doc/transchan.n copied the "returns a handle" wording from [chan create]; the code (and tests such as io-53.15) rely on the channel name being returned.
  • Existing work doc-chan-push-result (766139d4e1, 2026-10-14, one check-in on main), unmerged. Applies cleanly to main and 9.0, with fuzz to 8.6.
  • Prototype none (branch exists)
  • Suggested ticket comment none needed (Serhiy's 2026-10-14 comment points to the branch); the branch is ready to merge (docs only).
⬜2837073fff2009Retitle a few lessons, new lessonsDOCS (RFE: retitle introspection lessons, add event loop lesson)

Open ticket 2837073fff · 75. Tutorial · filed 2009-08-13

Full verdict

DOCS (RFE: retitle introspection lessons, add event loop lesson)

  • Platform docs
  • Tested none
  • Cause n/a
  • Prototype none (RFE; tutorial source not in tcl.fossil)
  • Suggested ticket comment Feature request for the tutorial; should be retyped as RFE.
⬜2779797fff2009inconsistent markupDOCS (reported fixed in the tutorial source by tallniel 2009-12-23; website update status not checkable offline)

Open ticket 2779797fff · 75. Tutorial · filed 2009-04-23

Full verdict

DOCS (reported fixed in the tutorial source by tallniel 2009-12-23; website update status not checkable offline)

  • Platform docs
  • Tested none
  • Cause some pages used old big-bold command markup.
  • Prototype none
  • Suggested ticket comment tallniel fixed the markup in 2009; if the published tutorial now shows consistent markup (e.g. Tcl34.html), this can be closed.
⬜2624762fff2009Tutorial updates for 8.5/8.6 new featuresDOCS (RFE: lessons on TclOO, TDBC, coroutines, lambdas, ...)

Open ticket 2624762fff · 75. Tutorial · filed 2009-02-21

Full verdict

DOCS (RFE: lessons on TclOO, TDBC, coroutines, lambdas, ...)

  • Platform docs
  • Tested none
  • Cause n/a
  • Prototype none (RFE)
  • Suggested ticket comment Feature request; should be retyped as RFE.
⬜2610482fff2009typo in section nine of tutorialDOCS (not checkable offline)

Open ticket 2610482fff · 75. Tutorial · filed 2009-02-17

Full verdict

DOCS (not checkable offline)

  • Platform docs
  • Tested none
  • Cause Tcl6a.html: "a floating-point number can be as a 1 followed by 300 zeros" lacks "large as".
  • Prototype none (tutorial source not in tcl.fossil)
  • Suggested ticket comment Tcl6a.html should read "can be as large as a 1 followed by 300 zeros".
⬜2561267fff2009Tcl_SetObjLength can produce invalid stringspatchDOCS (still true on main: Tcl_SetObjLength works on bytes, and doc/StringObj.3 does not warn about it)

Open ticket 2561267fff · 10. Objects · filed 2009-02-03

🧪 Prototype check (2026-10-24)

2561267fff.patch: VERIFIED (after the amendment; the original text was inaccurate for unicode-only values)

  • Base main (tip d6dd6d0f71)
  • Before work-a3/v2561.tcl: teststringobj set 1 "中文字"; teststringobj setlength 1 2 -> a 2-byte invalid string (ä¸ when shown as utf-8), so mid-character truncation happens as the ticket says. But teststringobj newunicode 1 0x4e2d 0x6587 0x5b57; teststringobj setlength 1 2 -> 中文: for a value that has only the internal Unicode rep (bytes == NULL), Tcl_SetObjLength counts characters, not bytes (tclStringObj.c, the else branch).
  • Patch applied cleanly, then amended. The original said flatly "NewLength is a length in bytes", which is wrong in the unicode-only case, so I added one parenthetical sentence about it. Now 7 changed lines.
  • After groff renders, no new warnings
  • Test none possible (docs)
Full verdict

DOCS (still true on main: Tcl_SetObjLength works on bytes, and doc/StringObj.3 does not warn about it)

  • Platform docs
  • Tested code/doc reading on main (generic/tclStringObj.c Tcl_SetObjLength; doc/StringObj.3)
  • Cause newLength is a byte count, with no check for character boundaries. Enlarging leaves arbitrary bytes in the gap, which the docs already mention.
  • Prototype prototypes/2561267fff.patch (4 changed lines, doc only: newLength is in bytes, mid-character truncation or uninitialized bytes after enlarging give an invalid string)
  • Suggested ticket comment Still true. Proposed doc addition to StringObj.3: newLength is a byte length, and truncating inside a multi-byte character (or not filling the bytes after enlarging) leaves an invalid string.
⬜2544760fff2009[interp exists] accepts 0 argumentspatchDOCS (behaviour is intended and documented. "If path is omitted, the invoking interpreter is used", and the wrong-args message says `interp exists …

Open ticket 2544760fff · 20. [interp] · filed 2009-01-29

🧪 Prototype check (2026-10-24)

2544760fff.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before interp exists (no argument) -> 1, while doc/interp.n:281 shows the synopsis interp exists path
  • Patch applied cleanly (doc only)
  • After man -l interp.n shows interp exists ?path?; troff warnings unchanged (1, pre-existing ^v register)
  • Test none possible (doc)
Full verdict

DOCS (behaviour is intended and documented. "If path is omitted, the invoking interpreter is used", and the wrong-args message says interp exists ?path?; only the man page synopsis says path is required)

  • Platform docs
  • Tested main, 9.0.5, 8.6.18: interp exists returns 1; two arguments give wrong # args: should be "interp exists ?path?"
  • Cause doc/interp.n synopsis interp exists path does not match the description or the implementation.
  • Prototype prototypes/2544760fff.patch (2 changed lines, doc: synopsis interp exists ?path?). Removing the 0-argument form would be an incompatible change, which is TIP territory.
  • Suggested ticket comment The 0-argument form is documented ("If path is omitted ...") and matches the wrong # args message. Proposed doc fix: show the path as optional in the synopsis. Removing the form would need a TIP.
⬜2464922fff2008Lesson 21: Written before 8.1?DOCS (claim confirmed; not checkable offline)

Open ticket 2464922fff · 75. Tutorial · filed 2008-12-24

Full verdict

DOCS (claim confirmed; not checkable offline)

  • Platform docs
  • Tested main: regexp {a\tb} "a\tb" -> 1, regexp {\x41} A -> 1 (the RE engine handles backslash escapes itself since 8.1, see doc/re_syntax.n).
  • Cause Tcl21.html claims backslash sequences mean nothing to the regexp parser.
  • Prototype none (tutorial source not in tcl.fossil)
  • Suggested ticket comment Since 8.1 the RE engine understands \t, \n, \xhh (re_syntax.n), so Lesson 21's bullet is wrong and the hex example should use a braced RE.
⬜2424616fff2008Undocumented flags accepted by Tcl_NREval* functionpatchDOCS (still undocumented; doc/Eval.3 has .\" TODO: document TCL_EVAL_INVOKE and TCL_EVAL_NOERR., and NRE.3 refers to Tcl_EvalObjv for flags)

Open ticket 2424616fff · 45. Parsing and Eval · filed 2008-12-13

🧪 Prototype check (2026-10-24)

2424616fff.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before doc/Eval.3 on the tip still has .\" TODO: document TCL_EVAL_INVOKE and TCL_EVAL_NOERR.
  • Patch applied cleanly; one wording fix in both entries: "it is ignored by other procedures" -> "other procedures do not support it", because Tcl_EvalObjEx passes its flags through to TclNREvalObjv for a pure list (tclBasic.c TclNREvalObjEx), so INVOKE is not simply ignored there (26 changed lines now)
  • After checked the text against tclBasic.c on the tip (TclNREvalObjv: INVOKE -> global-namespace lookup, no TEOV_SwitchVarFrame, no TclResetRewriteEnsemble; TEOV_PushExceptionHandlers: INVOKE skips TEOV_Error, NOERR skips all handlers incl. TEOV_Exception at level 1); groff -man -Tutf8 doc/Eval.3 renders both entries
  • Test none possible (docs only)
Full verdict

DOCS (still undocumented; doc/Eval.3 has .\" TODO: document TCL_EVAL_INVOKE and TCL_EVAL_NOERR., and NRE.3 refers to Tcl_EvalObjv for flags)

  • Platform docs
  • Tested main doc/Eval.3, doc/NRE.3, generic/tcl.h, tclBasic.c:TclNREvalObjv / TEOV_PushExceptionHandlers
  • Cause n/a
  • Prototype prototypes/2424616fff.patch (26 changed lines, doc only: TCL_EVAL_INVOKE and TCL_EVAL_NOERR entries in the FLAG BITS section of Eval.3, written from the tcl.h comment and the code; replaces the TODO)
  • Suggested ticket comment Proposed documentation of TCL_EVAL_INVOKE and TCL_EVAL_NOERR in Eval.3 (NRE.3 refers there for flags), replacing the TODO comment.
⬜2032614fff2008Missleading "switch" exampleDOCS (behaviour is correct per switch.n; whether the Tcl8.html example misleads not checkable offline)

Open ticket 2032614fff · 75. Tutorial · filed 2008-07-30

Full verdict

DOCS (behaviour is correct per switch.n; whether the Tcl8.html example misleads not checkable offline)

  • Platform docs
  • Tested none needed: in the braced form of switch, patterns are not substituted (doc/switch.n), so $center is a literal pattern; reporter's result is expected.
  • Cause Lesson 8 shows variable patterns inside the braced form without stressing that no substitution happens there.
  • Prototype none (tutorial source not in tcl.fossil)
  • Suggested ticket comment Patterns in the braced form of &#91;switch&#93; are not substituted; Lesson 8 should say so next to the first example.
⬜2028793fff2008Namespace dependence of interp aliaspatchDOCS (behaviour unchanged in main and 8.6, still undocumented in interp.n)

Open ticket 2028793fff · 20. [interp] · filed 2008-07-26

🧪 Prototype check (2026-10-24)

2028793fff.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before namespace eval ns {interp alias {} foo {} list A; interp alias {} q::bar {} list B} -> ::foo created (not ::ns::foo), q::bar -> ::ns::q::bar, namespace ::ns::q auto-created (v4docs1.tcl). interp.n says nothing about it.
  • Patch applied cleanly (7 changed lines); all three claims of the new text match the observed behaviour
  • After groff 2 warnings (same 2 before)
  • Test none possible (docs only)
Full verdict

DOCS (behaviour unchanged in main and 8.6, still undocumented in interp.n)

  • Platform docs
  • Tested main (own build), 8.6.18: the ticket's script gives the same results (foo goes to ::, test::foo goes to ::test::test::foo and creates ::test::test); also checked that a relative qualified name is resolved in the *source* interp's current namespace (namespace eval ns {interp alias c a::b {} list} creates ::a::b in child c). [proc] instead fails with "unknown namespace".
  • Cause tclInterp.c:AliasCreate uses Tcl_CreateObjCommand/Tcl_NRCreateCommand rules (CrtObjCmd.3: unqualified name goes to the global namespace).
  • Prototype prototypes/2028793fff.patch (7 changed lines, doc/interp.n only: describes how srcCmd is resolved)
  • Suggested ticket comment Behaviour is unchanged in 9.1 and comes from Tcl_CreateObjCommand. Proposed doc addition to interp.n: an unqualified srcCmd goes into the global namespace, a relative qualified one is resolved in the source interp's current namespace, and missing namespaces are created.
⬜2013699fff2008integer formats undocumentedpatchDOCS (still true on main: IntObj.3 says nothing about which strings are integers; the syntax is documented only in GetInt.3 and expr.n)

Open ticket 2013699fff · 48. Number Handling · filed 2008-07-08

🧪 Prototype check (2026-10-24)

2013699fff.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before doc/IntObj.3 on the tip still says nothing about which strings are integers. Checked the syntax the new sentence describes with incr s 0 on the tip build: " 12 "→12, +0x1_0→16, -0d1_000→-1000, 0b101→5, 0o17→15, _1→error; matches GetInt.3
  • Patch applied cleanly (8 changed lines, unchanged)
  • After groff -man -Tutf8 doc/IntObj.3 renders the new sentence and SEE ALSO entry correctly
  • Test none possible (docs only)
Full verdict

DOCS (still true on main: IntObj.3 says nothing about which strings are integers; the syntax is documented only in GetInt.3 and expr.n)

  • Platform docs
  • Tested doc reading (main); checked with [incr] on main and 8.6.18 which forms are accepted (whitespace, sign, 0b/0o/0d/0x, TIP 551 underscores; 0d and underscores are 9.0+ only, 017 is octal in 8.6, decimal in 9.x).
  • Cause doc/IntObj.3 lacks a syntax paragraph.
  • Prototype prototypes/2013699fff.patch (8 changed lines, docs only): one sentence in IntObj.3 pointing to Tcl_GetInt's syntax (whitespace, sign, radix prefixes, underscores), plus Tcl_GetInt in SEE ALSO.
  • Suggested ticket comment GetInt.3 documents the integer syntax (incl. 0d and TIP 551 underscores) since 9.0; proposed doc patch makes IntObj.3 refer to it.
⬜1801314fff2007Numeric Comparisons 101 - if: Error in example codeDOCS (claim confirmed; not checkable offline)

Open ticket 1801314fff · 75. Tutorial · filed 2007-09-24

Full verdict

DOCS (claim confirmed; not checkable offline)

  • Platform docs
  • Tested main, 8.6 (exit aliased in a child interp): if "$$y != 1" with y={[exit]} fails with invalid character "$" and never calls exit; if "$y != 1" does call exit.
  • Cause the "dangerous example" in the if lesson (Tcl4.html?) uses $$y, which does not demonstrate the double substitution.
  • Prototype none (tutorial source not in tcl.fossil)
  • Suggested ticket comment Confirmed: "$$y != 1" only raises a syntax error; the example should use "$y != 1" to show &#91;exit&#93; being run.
⬜1689385fff2007Tcl_SetObjLength documentation incompletepatchDOCS (behaviour unchanged in main, 9.0, 8.6; doc/StringObj.3 still describes only the byte case)

Open ticket 1689385fff · 10. Objects · filed 2007-03-27

🧪 Prototype check (2026-10-24)

1689385fff.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before teststringobj newunicode 1 0x3b1 0x3b2 0x3b3; teststringobj setlength 1 2; teststringobj get 1 -> αβ (length counted in characters; generic/tclStringObj.c Tcl_SetObjLength still has the bytes==NULL Unicode branch), while doc/StringObj.3 describes only the byte case
  • Patch applied cleanly (13 changed lines, unchanged)
  • After groff -man -Tutf8 doc/StringObj.3 renders the new paragraph correctly; text matches the code (SetStringFromAny first, so only pure-Unicode values take that branch)
  • Test none possible (docs only)
Full verdict

DOCS (behaviour unchanged in main, 9.0, 8.6; doc/StringObj.3 still describes only the byte case)

  • Platform docs
  • Tested main, 9.0 with teststringobj newunicode + setlength: on a 3-char pure-Unicode value setlength 2 gives 2 chars (4 UTF-8 bytes); setlength 4 on 1 char gives 4 chars. 8.6 has the same branch (code reading; its teststringobj lacks newunicode).
  • Cause generic/tclStringObj.c Tcl_SetObjLength: when objPtr->bytes == NULL it resizes the Unicode rep, treating the length as characters, and leaves bytes NULL.
  • Prototype prototypes/1689385fff.patch (13 changed lines, doc only: new paragraph in doc/StringObj.3 describing the pure-Unicode case, and the NUL guarantee limited to values with a string rep)
  • Suggested ticket comment Still undocumented in 8.6/9.0/9.1. Proposed doc fix: say that for a value without string rep (pure Unicode) the length counts characters and objPtr->bytes stays NULL; call Tcl_GetString first when a byte buffer is needed.
⬜1670783fff2007binary man page missing Inf/NaN refpatchDOCS — and the man page is now wrong on main/9.0: binary.n still says a value that overflows single precision becomes FLT_MAX, but since [85ce4bf92…

Open ticket 1670783fff · 48. Number Handling · filed 2007-02-28

🧪 Prototype check (2026-10-24)

1670783fff.patch: VERIFIED

  • Base main (tip d6dd6d0f71); docs only
  • Before doc/binary.n says an overflowing f value becomes FLT_MAX; tcltest 1670.tcl on the tip -> binary format f 1e39 scans back as Inf, -1e39 -> -Inf, Inf/-Inf/NaN round-trip, payload NaN -> NaN(20000000), -NaN, and expr {$x+1} -> cannot use non-numeric floating-point value "-NaN" ... (so the man page is wrong on main)
  • Patch applied cleanly; one word changed (scan section: special values "are returned as", not "are stored as"); 11 changed lines
  • After man -l doc/binary.n renders both paragraphs correctly and matches the tip's behaviour above
  • Test none possible (docs only)
Full verdict

DOCS — and the man page is now wrong on main/9.0: binary.n still says a value that overflows single precision becomes FLT_MAX, but since 85ce4bf928 (48706a2425, 2022-11-05, 9.0 only) it becomes ±Inf. 8.6 behaves as documented (FLT_MAX, also for Inf).

  • Platform docs
  • Tested main build, 9.0.5, 8.6.18: f/d/q/Q/r/R round-trip of Inf, -Inf, NaN, 1e400, 1e39. 9.x: all give ±Inf/NaN; 8.6: f/r/R give ±3.4028234663852886e+38 for Inf and 1e39. [binary scan] gives NaN(1), -NaN for non-default NaNs.
  • Cause doc/binary.n not updated with the 9.0 change and never mentions special values.
  • Prototype prototypes/1670783fff.patch (11 changed lines, docs only, for main/9.0): format f: overflow -> infinity of the same sign, Inf/-Inf/NaN stored as IEEE special values; scan f: special values come back as Inf, -Inf, NaN, NaN(hex), possibly with a minus sign, and [expr] rejects NaN.
  • Suggested ticket comment binary.n still describes the 8.6 FLT_MAX clamp, which 85ce4bf928 changed in 9.0 to storing ±Inf, and never mentions Inf/NaN. Proposed doc patch for 9.0/9.1.
⬜1657421fff2007default translation of stdout on unix == crlf?patchDOCS (behaviour unchanged since 8.4 and deliberate: REPRODUCED on 8.4.20, 8.6.18, 9.0.5, main)

Open ticket 1657421fff · 27. Channel Types · filed 2007-02-11

🧪 Prototype check (2026-10-24)

1657421fff.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before script -qc "tcltest tr.tcl" with puts [fconfigure stdout -translation] -> bytes crlf\r\r\n on a pty; to a pipe lf. Behaviour as the triage described, still undocumented in chan.n
  • Patch applied cleanly (5 changed lines, doc/chan.n)
  • After man -l doc/chan.n renders the new sentence ("... except for terminals and serial ports on Unix ... the default is auto crlf ...") correctly
  • Test none possible (documentation only)
Full verdict

DOCS (behaviour unchanged since 8.4 and deliberate: REPRODUCED on 8.4.20, 8.6.18, 9.0.5, main)

  • Platform unix
  • Tested tclsh on a pty (python pty.fork): fconfigure stdout -translation is crlf on all versions and puts sends "\r\r\n" to the terminal (Tcl's CR plus the tty driver's ONLCR); stdin is auto. To a pipe/file stdout is lf.
  • Cause unix/tclUnixChan.c: TclpGetDefaultStdChannel (and TclpOpenFileChannel for serial devices) set -translation {auto crlf} for every channel of ttyChannelType. CRLF keeps output readable when a script puts the terminal into raw mode (exec stty raw), at the cost of a stray CR in cooked mode, which only non-terminal consumers (acme, script logs) show. doc/chan.n says Unix output defaults to lf and doesn't mention terminals.
  • Prototype prototypes/1657421fff.patch (doc/chan.n only, 5 changed lines, documents the terminal/serial default). Changing the default to lf would break the common stty raw idiom.
  • Suggested ticket comment This is deliberate: terminals and serial ports default to -translation {auto crlf} so output stays correct in raw mode; in cooked mode the extra CR is harmless on real terminals. Proposed: document it in chan.n (patch available) and close.
⬜1649077fff2007minor spelling correctionsDOCS (patch attached 43416cfd48; not checkable offline whether already applied)

Open ticket 1649077fff · 75. Tutorial · filed 2007-01-31

Full verdict

DOCS (patch attached 43416cfd48; not checkable offline whether already applied)

  • Platform docs
  • Tested none
  • Cause typos in tutorial html.
  • Prototype none (patch is against tcltutorial)
  • Suggested ticket comment Spelling patch against the tutorial source; apply what is still relevant and close.
⬜1615319fff2006special treatment for "DOCS (not checkable offline)

Open ticket 1615319fff · 75. Tutorial · filed 2006-12-13

Full verdict

DOCS (not checkable offline)

  • Platform docs
  • Tested none (rule is doc/Tcl.n rule [4]: " is special only at the start of a word)
  • Cause Lesson on grouping with {} does not mention that " is also only special at word start.
  • Prototype none (tutorial source not in tcl.fossil)
  • Suggested ticket comment Tcl.n rule &#91;4&#93; covers this; the grouping lessons should mention that " inside a word is literal.
⬜1610478fff2006TitlesDOCS (RFE: drop "101" from lesson titles)

Open ticket 1610478fff · 75. Tutorial · filed 2006-12-06

Full verdict

DOCS (RFE: drop "101" from lesson titles)

  • Platform docs
  • Tested none
  • Cause n/a
  • Prototype none (RFE)
  • Suggested ticket comment Cosmetic request for the tutorial; RFE.
⬜1525677fff2006TCL_INTERP in example codeDOCS (not checkable offline)

Open ticket 1525677fff · 75. Tutorial · filed 2006-07-20

Full verdict

DOCS (not checkable offline)

  • Platform docs
  • Tested none
  • Cause Tcl26.html examples use a placeholder TCL_INTERP; [info nameofexecutable] would make them runnable.
  • Prototype none (tutorial source not in tcl.fossil)
  • Suggested ticket comment Replace the TCL_INTERP placeholder in Lesson 26 with &#91;info nameofexecutable&#93;.
⬜1482667fff2006More detailed information on [if ... else] and spacesDOCS (RFE-like improvement request)

Open ticket 1482667fff · 75. Tutorial · filed 2006-05-05

Full verdict

DOCS (RFE-like improvement request)

  • Platform docs
  • Tested none
  • Cause n/a (}else{ and a brace on the next line are common beginner errors)
  • Prototype none (tutorial source not in tcl.fossil)
  • Suggested ticket comment Lesson 4/7 could add a short "common mistakes" note on }else{ and newline before {.
✅1329568fff2005[socket -async] and [puts]: Inadequate documentation?patchDOCS question answered by current docs (socket.n -async section since 8.6.1, chan.n puts buffering); but testing found a REPRODUCED regression (9.0…

Open ticket 1329568fff · 24. Channel Commands · filed 2005-10-18

✅ Since triage (updated 2026-11-11)

Status now: Closed (Fixed)

  • 💬 2026-10-16 serhiy.storchaka: The documentation now covers this: socket.n says that a blocking flush waits for the connection and a non-blocking one buffers the data, and chan.n says that non-blocking puts buffers without limit. Testing this found a 9.0 regression, reported as 251f46655c.
Full verdict

DOCS question answered by current docs (socket.n -async section since 8.6.1, chan.n puts buffering); but testing found a REPRODUCED regression (9.0, main; not 8.6, 8.7b1): closing a blocking socket -async while the connect is still pending loses the buffered data

  • Platform generic (docs) + unix (regression found while testing)
  • Tested Python server with a full listen queue (accepts after 1-2 s) so the connect stays pending (work-a5/as): non-blocking puts of 1 MB before the connect finishes is buffered (no error, -connecting 1, chan pending output = 1000000) and a blocking puts+flush waits for the connect, on 8.6.18 and main. But set s [socket -async ...]; puts $s hello; close $s (or switching to blocking before close): 8.6.18 and 8.7b1 deliver the data; 9.0.5, 9.0 tip and main fail with "transport endpoint is not connected" and the server receives nothing (3/3 runs each).
  • Cause generic/tclIO.c:Tcl_Close does a read half-close (close2Proc TCL_CLOSE_READ) before flushing. In 8.6/8.7 that was only done for channel types without a closeProc, so TCP sockets skipped it; since 9.0 (d5bc305ad2, 2020-03-04, removal of the old closeProc) it always runs, and unix/tclUnixSock.c:TcpClose2Proc calls shutdown(SHUT_RD) on the still-connecting socket, which on Linux aborts the connect (strace: shutdown, then poll -> SO_ERROR ECONNRESET). Windows' shutdown on an unconnected socket just fails with ENOTCONN (ignored), so probably Unix only.
  • Prototype prototypes/1329568fff.patch (against main, also applies to core-9-0-branch; 47 changed lines, ~10 of them code: TcpClose2Proc skips a read-only shutdown while TCP_ASYNC_PENDING; test socket-14.20 (child server with -backlog 1 that accepts late) fails without the fix with "transport endpoint is not connected"). socket.test passes (180 passed, 12 skipped).
  • Suggested ticket comment The docs now cover this (socket.n: blocking flush waits for the connect, non-blocking returns and buffers; chan.n: non-blocking puts buffers without limit). While testing I found a 9.0 regression: closing an async socket whose connect is still pending loses the buffered data, because Tcl_Close's read half-close does shutdown(SHUT_RD) on the connecting socket. Proposed fix: skip that shutdown while the connect is pending, test socket-14.20. Better tracked in its own ticket.
⬜1274918fff2005doc: upvar and qualified varNamespatchDOCS

Open ticket 1274918fff · 07. Variables · filed 2005-08-28

🧪 Prototype check (2026-10-24)

1274918fff.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before docs only; doc/upvar.n on tip still describes only procedure-local links. Checked every statement on the tip with work-a3/v1274918.tcl: namespace eval a {upvar #0 x y} makes ::a::y; unset a::y unsets ::x while the link stays (setting x again shows through a::y); qualified myVar in a proc (c::b::r, and b::s from a proc in ::c) lands in ::c::b; a missing parent namespace -> can't create "nosuch::z": parent namespace doesn't exist.
  • Patch applied cleanly (14 added lines)
  • After groff renders; no new warnings. A minor point the text leaves out: a qualified myVar cannot link to a procedure-local otherVar (upvar 0 loc ::c::b::q -> "can't create namespace variable that refers to procedure variable"). It is optional to mention.
  • Test none possible (docs)
Full verdict

DOCS

  • Platform docs
  • Tested doc/upvar.n on main still only talks about "local variables in the current procedure"; behaviour checked on main, 9.0, 8.6: outside a proc namespace eval a {upvar #0 x y} creates namespace var ::a::y, unset a::y unsets ::x but the link stays; qualified myVar inside a proc creates a namespace variable; missing parent namespace is an error. Since 9.0 (TIP 278) relative qualified names resolve only relative to the current namespace, so the "madness" in the 2006-08-26 comment is gone in 9.x (both vars land in ::c::b), still present in 8.6.
  • Cause missing documentation
  • Prototype prototypes/1274918fff.patch (14 changed lines, doc only: two paragraphs in doc/upvar.n describing upvar outside procs and qualified otherVar/myVar; text matches 9.x semantics)
  • Suggested ticket comment Proposed doc text for upvar.n: two paragraphs that document upvar outside procedures (namespace variable, link persists until the namespace is deleted) and qualified names, following TIP 278 rules in 9.x.
⬜1237009fff2005Lesson 41: clock lesson does not deal with 8.5 featuresDOCS (RFE: mention clock scan -format)

Open ticket 1237009fff · 75. Tutorial · filed 2005-07-12

Full verdict

DOCS (RFE: mention clock scan -format)

  • Platform docs
  • Tested none
  • Cause n/a
  • Prototype none (RFE)
  • Suggested ticket comment Feature request; a sentence on clock scan ... -format in Lesson 41 would suffice.
⬜1072603fff2004tcl_pkgPath cross-platform documentationpatchDOCS (partly addressed)

Open ticket 1072603fff · 38. Init - Library - Autoload · filed 2004-11-24

🧪 Prototype check (2026-10-24)

1072603fff.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before grep -rn tcl_pkgPath win/*.c unix/*.c -> still set only in unix/tclUnixInit.c (TclpSetVariables); doc/msgcat.n and doc/pkgMkIndex.n still tell readers to install under $tcl_pkgPath without a Windows caveat (tclvars.n and tm.n are already correct)
  • Patch applied cleanly; two over-long source lines rewrapped (11 changed lines now, text unchanged)
  • After groff -man -Tutf8 doc/pkgMkIndex.n / doc/msgcat.n render the new sentences correctly (only the usual man.macros warnings); no code touched, no test run
  • Test none possible (docs only)
Full verdict

DOCS (partly addressed)

  • Platform docs
  • Tested tcl_pkgPath is still set only by unix/tclUnixInit.c (not in win/). tclvars.n (8.4 and later) already says "It is not used on Windows", but pkgMkIndex.n and msgcat.n still tell readers to install into $tcl_pkgPath without a caveat.
  • Cause docs.
  • Prototype prototypes/1072603fff.patch (9 changed lines, doc only): pkgMkIndex.n and msgcat.n now say auto_path, with tcl_pkgPath as the Unix case.
  • Suggested ticket comment tclvars.n already says tcl_pkgPath is not used on Windows. Proposed patch adjusts pkgMkIndex.n and msgcat.n to refer to auto_path, mentioning tcl_pkgPath only for Unix.
⬜967565ffff2004[info level] reports current interp's CallFrames onlypatchDOCS (behaviour by design: levels are per-interp, the same ones [uplevel]/[upvar] use)

Open ticket 967565ffff · 20. [interp] · filed 2004-06-06

🧪 Prototype check (2026-10-24)

967565ffff.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before r967.tcl (child proc p -> q -> alias show to the parent) -> info level in parent: 1 (child frames not counted; by design); doc/info.n:271-276 still calls the argument number and says nothing about interps
  • Patch applied cleanly (doc only)
  • After text says only the current interp's call frames count (alias callers invisible), number -> level; troff 0 warnings before/after
  • Test none possible (doc)
Full verdict

DOCS (behaviour by design: levels are per-interp, the same ones [uplevel]/[upvar] use)

  • Platform docs
  • Tested attached test.tcl on main, 9.0, 8.6 — the alias target in the parent sees ::test::procA (parent interp) as level 1, not procC in the child
  • Cause each interp has its own CallFrame stack (iPtr->varFramePtr); doc/info.n does not say so (and still calls the argument "number" while the synopsis says "level")
  • Prototype prototypes/967565ffff.patch (11 changed lines, docs only, no test) — doc/info.n: one sentence that only the current interp's frames are counted, plus number→level
  • Suggested ticket comment By design: &#91;info level&#93; counts the call frames of the current interp only, the same levels &#91;uplevel&#93; and &#91;upvar&#93; use. Proposed doc clarification for info.n; a cross-interp stack walk would need a TIP.
⬜945570ffff2004public routines need private #define'sDOCS (still true: tcl.h includes only <stdio.h>; FileSystem.3 names R_OK/W_OK/X_OK/F_OK and O_RDONLY etc. but not their headers). The Tcl_StatBuf p…

Open ticket 945570ffff · 37. File System · filed 2004-04-30

Full verdict

DOCS (still true: tcl.h includes only <stdio.h>; FileSystem.3 names R_OK/W_OK/X_OK/F_OK and O_RDONLY etc. but not their headers). The Tcl_StatBuf part is mostly solved by the TIP 316 accessors (Tcl_GetModeFromStat etc.).

  • Platform generic (API/docs)
  • Tested code reading of main generic/tcl.h, doc/FileSystem.3
  • Cause design. Defining these in tcl.h would clash with system headers (nijtmans 2012); new TCL_* names would need a TIP.
  • Prototype none (needs a decision: document the headers, or add TCL_* constants via a TIP)
  • Suggested ticket comment Still applies to Tcl_FSAccess modes and the openFileChannelProc flags; the stat part is covered by the TIP 316 accessors. Either document the needed system headers in FileSystem.3 or propose TCL_* constants in a TIP.
⬜624919ffff2002Tcl_AppendToObj docs confusingpatchDOCS — dkf's 2002 doc fix is in (bytes are TUTF-8, nulls as 0xC0 0x80), but the argument text in doc/StringObj.3 still says the bytes "may contain …

Open ticket 624919ffff · 10. Objects · filed 2002-10-17

🧪 Prototype check (2026-10-24)

624919ffff.patch: VERIFIED (ported)

  • Base main (tip d6dd6d0f71)
  • Before docs only; doc/StringObj.3:93-94 on the tip still says the bytes "may contain embedded null characters unless \fInumChars\fR is negative" (wrong argument, and it contradicts the next sentence)
  • Patch ported. The hunk fixing the \fIfIobjPtr\fR typo in the Tcl_IsEmpty synopsis was dropped: 50013b575a (2026-10-18, merged to main via f813fc7729) already fixed it. The ARGUMENTS hunk is unchanged; now 8 changed lines.
  • After groff renders, no new warnings
  • Test none possible (docs)
Full verdict

DOCS — dkf's 2002 doc fix is in (bytes are TUTF-8, nulls as 0xC0 0x80), but the argument text in doc/StringObj.3 still says the bytes "may contain embedded null characters unless numChars is negative" (wrong argument name, and contradicts the next sentence)

  • Platform docs
  • Tested raw NUL key via testbytestring on main/9.0 tcltest: $r($b) works but [info exists r($b)] gives 0 (non-compiled path truncates at the NUL); on 8.6 even $r($b) fails. Invalid input, so this backs up the doc rule rather than being a bug of its own
  • Cause doc wording only
  • Prototype prototypes/624919ffff.patch (10 changed lines, docs only) — say plainly that the first length bytes must not contain a raw null byte; also fixes the \fIfIobjPtr\fR typo in the Tcl_IsEmpty synopsis
  • Suggested ticket comment Still confusing: StringObj.3 says the bytes "may contain embedded null characters unless numChars is negative", yet nulls must be 0xC0 0x80 in TUTF-8. Proposed doc fix stating that raw null bytes are not allowed.
⬜414930ffff2001Undocumented public APIs, namespaceDOCS

Open ticket 414930ffff · 21. [namespace] · filed 2001-04-09

Full verdict

DOCS

  • Platform docs
  • Tested grep of doc/*.3 and generic/*.decls in main: Tcl_AppendExportList, Tcl_CreateNamespace etc. are in tcl.decls and doc/Namespace.3 (done 2007). The remaining Tcl_FindNamespaceVar, Tcl_PushCallFrame, Tcl_PopCallFrame, Tcl_AddInterpResolvers, Tcl_GetInterpResolvers, Tcl_RemoveInterpResolvers, Tcl_SetNamespaceResolvers, Tcl_GetNamespaceResolvers are still only in tclInt.decls (internal stubs) and undocumented.
  • Cause these are internal (tclInt.h) APIs despite the Tcl_ prefix.
  • Prototype none (decision whether to make them public is a TIP matter)
  • Suggested ticket comment All remaining names are internal stubs (tclInt.decls), not public API; documenting them means promoting them via a TIP. Suggest closing as "Won't Fix" (internal) or merging with 218101ffff.
⬜219149ffff2000bad interaction of input redirect from named pipe with backgpatchDOCS (REPRODUCED on main, 9.0, 8.6; by design according to ferrieux, 2013)

Open ticket 219149ffff · 27. Channel Types · filed 2000-10-26

🧪 Prototype check (2026-10-24)

219149ffff.patch: VERIFIED

  • Base main (tip d6dd6d0f71)
  • Before work-a4/v4fifo.tcl (writer opens the FIFO after 1.5 s): exec cat <v4fifo & -> returned after 1504 ms; exec sh -c {cat <v4fifo} & -> returned after 0 ms (the documented workaround works).
  • Patch applied cleanly (7 changed lines)
  • After groff 0 warnings (0 before); renders "... exec blocks too, even with “&”. Let a shell open the file in that case, as in exec sh -c {cat <fifo} &."
  • Test none possible (docs only)
Full verdict

DOCS (REPRODUCED on main, 9.0, 8.6; by design according to ferrieux, 2013)

  • Platform unix
  • Tested exec mkfifo fifo; exec cat <fifo >@stdout & on main, 9.0.5, 8.6.18: exec never returns (an after 2000 timer cannot fire, killed by timeout). exec sh -c {cat <fifo} >@stdout & returns at once and works.
  • Cause generic/tclPipe.c:TclCreatePipeline opens redirection files in the parent (FileForRedirect -> TclpOpenFile) before forking, so that errors can be reported synchronously. open() of a FIFO with no writer blocks.
  • Prototype prototypes/219149ffff.patch (7 changed lines, doc/exec.n only: documents the blocking open and the sh -c workaround; no test, docs only)
  • Suggested ticket comment Still the case: exec opens redirection files itself before it returns (so it can report errors), and opening a FIFO with no writer blocks. Proposed doc change for exec.n, with exec sh -c {cat <fifo} & as the workaround.
⬜218101ffff2000no man page for library procedures Tcl_AddInterpResolver TclDOCS

Open ticket 218101ffff · 21. [namespace] · filed 2000-10-26

Full verdict

DOCS

  • Platform docs
  • Tested main doc/*.3 has no page for Tcl_AddInterpResolvers / Tcl_GetInterpResolvers / Tcl_RemoveInterpResolvers / Tcl_SetNamespaceResolvers / Tcl_GetNamespaceResolvers; they're still only in tclInt.decls (internal stubs 111ff). Attachments Resolver.3 and resolver.c are present in the repo and complete.
  • Cause internal API with a Tcl_ prefix; never made public.
  • Prototype none (doc for an internal API; promoting it needs a TIP, and dkf noted the types would need documenting too)
  • Suggested ticket comment Still internal-only (tclInt.decls) in 9.1. George's Resolver.3 is attached and could be a base if a TIP makes the resolver API public; otherwise suggest closing together with 414930ffff.