Triage of all 509 open and Pending tickets of tcllib (bugs, patches and feature requests) on 2026-11-15. Each ticket was checked against tcllib trunk on Linux with Tcl 9.1, 9.0, 8.6 and 8.5, using local servers for the network modules and a Windows 11 VM for Windows-only tickets. Outdated, superseded, duplicate and misfiled tickets are listed separately. Rows marked patch link to a prototype fix with a regression test (147 in all, in tcllib-triage/prototypes). Updated 2026-11-16.
NOT-REPRODUCED (all 5 tests pass on 9.1 and 8.6 against a local websocket server); the real remaining problem is that the testsuite depends on the external server ws.ifelse.io
wsUrl pointed at a local tcllib websocket::server that answers "Request served by local". 9.1.1 (http 2.10.3) and 8.6.18 (http 2.9.9) both gave 5/5. Socket reuse (tests 1.3/1.4) no longer happens: http ≥ 2.9.8 opens a fresh socket for upgrade requests (websocket.tcl ~1203).
sak test run does not contact external hosts.
Status now: Pending
NOT-REPRODUCED (round trip of PDF-like binary data, with CR/LF, ^Z and high bytes, small (deflate in one chunk) and >2 MB (stream), is byte-identical)
cmp, plus unzip -t, on 9.1 and 8.6. Both files were identical and the archive was OK.
trailer keyword. That is normal for PDF 1.5+ with cross-reference streams, so the "trailer not found" message comes from the PDF reader, not from tcllib. aku asked for a reproducer in 2021 and got no reply.
-keepalive 1 and the peer either sending a close frame or closing abruptly, waiting 3 s. Th…NOT-REPRODUCED. I tried a server-side connection with -keepalive 1 and the peer either sending a close frame or closing abruptly, waiting 3 s. There was no bgerror and no pending after afterwards, on 9.1 and 8.6. Disconnect cancels Connection(liveness).
::websocket::New resets Connection(liveness) to "" without cancelling an already scheduled ping, e.g. when takeover runs twice on the same socket. That would leave an orphan timer, and since Windows reuses socket names, it could fire later. Unconfirmed.
NOT-REPRODUCED (9.1, 8.6; plain-http proxy path)
geturl+cleanup calls through autoproxy::init 127.0.0.1:port with a local proxy: RSS and the number of ::http::* vars stay flat on 9.1 and 8.6. The HTTPS/CONNECT path could not be tested with a real TLS proxy (no tls package here). aku could not reproduce it in 2019 either.
state(sock) missing in http::Finish/reset) come from the Tcl 8.5/8.6 http package. That package was rewritten in Tcl 9.
NOT-REPRODUCED (20 concurrent runs, 10 on 9.1 and 10 on 8.6, all 48/48 passed). The suite was rewritten for cron 2.0 (d9b48225d8, 2016-07) and the 10/11-second "elapsed" tests the ticket describes are commented out ("Test removed - Was too unstable on a busy computer").
sak test run modules/cron, 20 runs in parallel
NOT-REPRODUCED (the attached s21SOIImVi parses correctly on 9.1 and 8.6, same as PyYAML); the other file (s21BwGvY4m) is lost with the paste site
yaml2dict and compared key sections with PyYAML; they match.
NOT-REPRODUCED (CONNECT already sends basic auth)
init 127.0.0.1:port + configure -basic -username me -password secret, the CONNECT request carries Proxy-Authorization: Basic bWU6c2VjcmV0. On 9.x it fails earlier because of 29ff5c3abd.
options(basic) since 4d9dbdb9fd (2008). Only Basic is supported, no Digest/NTLM, as for plain http.
{ "a" : 0 } gives a 0). The C lexer was rewritten in 2014 (11390a7baa, bison parser). It now u…Status now: Pending (Invalid)
NOT-REPRODUCED (Tcl implementation, 9.1 and 8.6: { "a" : 0 } gives a 0). The C lexer was rewritten in 2014 (11390a7baa, bison parser). It now uses strtod's end pointer, so no whitespace can end up in a number.
Status now: Pending (Invalid)
NOT-REPRODUCED (aku could not confirm in 2011 with a memory-debug build; ticket is Pending/Invalid since then)
Status now: Pending
NOT-REPRODUCED (Arjen's 2015 analysis holds)
after handlers during md5's internal vwait, 20+20 runs on 9.1/8.6: correct hashes, no…NOT-REPRODUCED (normal use, and nested md5 -file calls from after handlers during md5's internal vwait, 20+20 runs on 9.1/8.6: correct hashes, no leaked or missing channels)
sak test run modules/doctools2idx: 955/955 pass on 9.1 and 8.6; test doctools-idx-50.* exists and passes)NOT-REPRODUCED (sak test run modules/doctools2idx: 955/955 pass on 9.1 and 8.6; test doctools-idx-50.* exists and passes)
can not find channel named "stdout" in the plugin's safe interp. This came from early 8.6 std-channel changes and no longer happens.
sak test run modules/profiler on 9.1: all progress output stays on one CR-overwritten line and the output is 7 lines in total; 18…NOT-REPRODUCED (sak test run modules/profiler on 9.1: all progress output stays on one CR-overwritten line and the output is 7 lines in total; 18 passed, 1 skipped)
NOT-REPRODUCED (9.1, 8.6; local fake SMTP server, bodies with and without a trailing newline)
NOT-REPRODUCED on Windows (9.1): fileutil.test install-2.1 and install-2.2 pass (fileutil 345 passed, 107 skipped, 0 failed).
glob -directory D [file join . / *] = glob -directory D /*. On Windows /* is volume-relative, which is likely why it failed there.
*. It passes on Linux 9.1/8.6. I could not show the failure without Windows.
/* ([file join . / *]), which is suspect on Windows. Proposed changing it to *.
sak test run of trunk with tclsh91 on the Windows VM. The odd glob pattern in the tests is harmless now, so the test-only prototype is cosmetic.
NOT-REPRODUCED (pop3 suite passes on 9.1 and 8.6, 44/45 with 1 TLS skip. dgp could not reproduce it either in 2006, after aku's dialog/test fixes)
nodir/x.txt against ftpd, active and passive, 9.1 and 8.6, returns 0 at once; the dialog server replying 553 after an earl…NOT-REPRODUCED (Put to nodir/x.txt against ftpd, active and passive, 9.1 and 8.6, returns 0 at once; the dialog server replying 553 after an early data connection also returns 0)
state.data as 744276. The 744276 prototype also covers put_sent.
NOT-REPRODUCED (9.1, 8.6)
mymethod callback to a method that does return -code error -errorcode {MY CODE}, and one that returns break. Error code, info and break code are preserved.
NOT-REPRODUCED (9.1, 8.6, 8.5; the ticket's trace-on-unset scenario)
incr on an unset variable creates it, so the catch {incr ...} in enter/leaveHandler nev…NOT-REPRODUCED (fixed by Tcl 8.5 semantics). Since 8.5, incr on an unset variable creates it, so the catch {incr ...} in enter/leaveHandler never throws and errorInfo is left alone.
catch {error orig}, then the first call of a profiled proc chain, then a check of $::errorInfo. It stays intact on 9.1, 8.6 and 8.5.
catch wrappers could be dropped as cleanup)
incr auto-creates the array element, so the catch never fires and errorInfo is untouched. tcllib now requires 8.5, so closing.
NOT-REPRODUCED (5 MB random binary encodes and decodes correctly in 7 ms / 80 ms on 9.1 and 8.6, and 1.6 s / 1.1 s with the pure-Tcl path on 8.5). The original report was the 8.4-era pure-Tcl encoder hitting a memory limit on Windows 98.
binary encode base64)
binary encode base64) and in about 1.5 s with the pure-Tcl code on 8.5. Closing as out of date.
a b, x{y}, d$[1], żółw all work)NOT-REPRODUCED (ftpd, 9.1 and 8.6: Cd/Pwd into a b, x{y}, d$[1], żółw all work)
$, brackets and non-ASCII letters work on current trunk. Without an example of a failing name this can't be investigated further.
NOT-REPRODUCED (List / Get / List sequences against ftpd, active and passive, 9.1 and 8.6)