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 Outdated, superseded, duplicate and misfiled tickets are listed separately.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-A-BUG (spam / unrelated request)
NOT-A-BUG (local socket error, not a server code. Only the diagnostic is confusing)
binary format h* (the code dumps it nibble-swapped). It reads error reading "sockeb61d42490": software caused connection abort.
ldap::ReceiveBytes returns the code error when read throws, here because the connection was aborted (ECONNABORTED). MessageReceiver sends that to the default branch, which labels it "Protocol error ... Error reading SEQUENCE" and hex-dumps the error text. The caller does get an error, which is correct. The wording is misleading and the socket is not closed.
error branch that reports the read error text in plain words and closes the socket.
error reading "sock...": software caused connection abort: the TCP connection was aborted (server or network), and ldap reported it. The message could be clearer, but no server reply is mishandled.
NOT-A-BUG (documented design). json.man now states that null becomes the string "null" and links this ticket (2b2bcd7ecf, 2026-01-27).
{"a":null,"b":"null"} gives a null b null
::httpd::server constructor already calls my start. The wiki snippet then calls HTTPD start a second time, which binds the sam…NOT-A-BUG (the ::httpd::server constructor already calls my start. The wiki snippet then calls HTTPD start a second time, which binds the same port again. With port auto the second start just allocates another port, which is why that "works")
httpd::server create HTTPD port 18000; HTTPD start gives "couldn't open socket: address already in use" on 9.1. Without the extra start the server runs (the man page's minimal example has no start call).
my start. httpd.man documents start as "Open the socket listener" but does not say that the constructor already calls it.
start could close existing listeners first)
HTTPD start from the wiki snippet tries to bind the same port a second time. Drop that line, as in the man page's minimal example.
NOT-A-BUG. Same issue as 6362dbd12b; it could also be closed as a DUPLICATE of 6362dbd12b. join and split are complementary: uri::join scheme file path /C:/src gives file:///C:/src, which splits back to path /C:/src. The leading / belongs to the URI path.
NOT-A-BUG (by design; DOCS)
option -a -default $::a gives can't read "::a" on 9.1/8.6. snit::type C [list option -a -default $::a] works.
list/string map, or set the option in the constructor) next to snit::type. The tests pin the error and the workaround.
list/string map or set the option in the constructor. A doc note is proposed.
file:///T:/dir1/file1.txt splits to path /T:/dir1/file1.txt, which is the RFC 3986/8089 path-absolute form. Since uri 1.2.7, uri.man…NOT-A-BUG. file:///T:/dir1/file1.txt splits to path /T:/dir1/file1.txt, which is the RFC 3986/8089 path-absolute form. Since uri 1.2.7, uri.man documents that on Windows the path of a file URI is not a valid local filename, and its EXAMPLES section shows how to convert. T:/dir1/file1.txt is not a URI, so unknown scheme 'T' is correct.
uri::split file:///T:/dir1/file1.txt and uri::split T:/dir1/file1.txt file.
uri::geturl accept such paths on Windows.
{}. Only the first entity is returned, which is documented.) A related real weakness was found: the…NOT-A-BUG (the input is not a JSON object: no {}. Only the first entity is returned, which is documented.) A related real weakness was found: the Tcl parser silently skips characters that are not tokens, e.g. {"a": @@ 1} gives a 1, and trailing garbage is ignored.
message, {"a":1} garbage gives a 1, {"a": @@ 1} gives a 1
regexp -all -inline $tokenRE, which drops unmatched text
{...}.
{$::x} is just the list string representation. lindex returns $::x exactly. The fact that an array gives a list is now documented.)NOT-A-BUG ({$::x} is just the list string representation. lindex returns $::x exactly. The fact that an array gives a list is now documented.)
json2dict {["length","$::x"]} gives length {$::x}
$::x, as lindex shows. Use list commands, or join, rather than the string form.
connect is the documented event and request was never documented). Re-adding request after 11 years …NOT-A-BUG (intentional change from 2015; connect is the documented event and request was never documented). Re-adding request after 11 years would surprise current users more than it helps.
NOT-A-BUG (support question, answered by aku in 2013)
NOT-A-BUG (expand pushes a new empty context, so cset values from the outer context are not visible. Resolution is already Invalid. A note was added to the documentation in e53cad73ec 2013-12-06 and is present in expander.man.)
Op_cpush $name expand)
%0D%0A), browsers send textarea content that way…NOT-A-BUG (by spec). application/x-www-form-urlencoded requires line breaks to be sent as CR LF (%0D%0A), browsers send textarea content that way, and Tcl's own http::formatQuery does the same. decode faithfully returns what was sent. Mapping CRLF to LF in decode would silently alter data.
ncgi::decode [ncgi::encode "a\nb"] gives "a\r\nb" on 9.1 and 8.6
map(\n) = %0D%0A)
NOT-A-BUG (support questions, not an RFE)
install called from a method after construction replaces the component, and delegated methods follow the new component.
install (or plain assignment to the component variable) works at any time, not only in the constructor. (3) delegate is a definition-time statement only.
delegate is only available in the type definition. Closing.
[fun] and [type] markup exists and the reporter got good results with it, see the 2007-09-23 comment. Creating man-page aliases/…NOT-A-BUG (the [fun] and [type] markup exists and the reporter got good results with it, see the 2007-09-23 comment. Creating man-page aliases/symlinks for installation is outside doctools' scope, as AK said.)
fun, type)
NOT-A-BUG (param is the raw attribute string, not a Tcl list)
> unchanged. A string with quoted values containing quotes or spaces is not a valid list.
NOT-A-BUG (NList passes server output through unchanged; the reporter agreed in 2006 that it is IIS behaviour)
NOT-A-BUG (documented restriction)
option -thisOrThat is rejected ("badly named option"). -thisÄtsch is accepted, because the check is [A-Z ] only.
option {-name resourceName ClassName} gives camel-case resource names.
option {-thisorthat thisOrThat ThisOrThat} for a camel-case resource name. Closing.
NOT-A-BUG (by design and documented)
-?/-help throw the usage message with errorcode CMDLINE USAGE on 9.1 and 8.6. ?/help without a dash are plain arguments.
try ... trap {CMDLINE USAGE} example to print it and exit.
{CMDLINE USAGE} as in the man page example. See also 472259ffff.
NOT-A-BUG (IIS answers LIST of a non-matching name with 550; Solaris ftpd answers 226 with an empty list. The package logs the 550 as an error and returns {} in both cases)
NOT-A-BUG (the testsuite confusion was fixed in 2005, ChangeLog 2005-08-26). Pure Tcl and critcl pad with backtick, like GNU/perl. Only the optional Trf accelerator uses ~, and Trf is not available for current Tcl here. uudecode accepts ` `, ~, =` and space padding equally, because the length byte decides.
NOT-A-BUG (deliberate since 2003-11-25, "better handling of multiple lines". log::logsubst does the same)
::log::log and ::log::logsubst. Both call the command once per line.
foreach line [split $text \n] {...} in log::log. Changing it now would alter the output of every existing user.
NOT-A-BUG (the server answers NLST on an empty directory with 450/550 "No files found", as wu-ftpd/ProFTPD do. The package then logs an error and still returns {}, the same value as for an empty listing)