Tcl Source Code

View Ticket
Login
2026-11-01
20:37 • Closed ticket [28de60cef5]: Segfault in destructor using tsv::lock plus 7 ot... artifact: 56956b0708 user: dkf
20:36
Fix [28de60cef5] and [b4637c4312] check-in: f3f08274f7 user: dkf tags: trunk, main
20:30
Fix [28de60cef5] and [b4637c4312] check-in: 298d1d97d5 user: dkf tags: core-9-0-branch
17:57
Backport of Fix [28de60cef5]: reading an object variable from code compiled at run time (e.g. [eval]... check-in: db7f34af59 user: dkf tags: oo-destructor-deleted-namespace-90
2026-10-13
19:08 • Ticket [b4637c4312] upvar segfault for destroyed object status still Open wi... artifact: 42f3dc584f user: serhiy.storchaka
19:06
Fix [b4637c4312]: [upvar] to an object variable after the object was destroyed in its method crashed... closed check-in: de6527681f user: serhiy.storchaka tags: oo-destructor-deleted-namespace
15:27 • Ticket [28de60cef5] Segfault in destructor using tsv::lock status still Open... artifact: 42d6509f0a user: serhiy.storchaka
15:25
Fix [28de60cef5]: reading an object variable from code compiled at run time (e.g. [eval]) in a destr... check-in: f38a74a978 user: serhiy.storchaka tags: oo-destructor-deleted-namespace
2020-02-16
10:44 • New ticket [28de60cef5] Segfault in destructor using tsv::lock. artifact: d76b6e5673 user: cyan

Ticket UUID: 28de60cef5133f1bd5c07dadf4d1a6e1989878c4
Title: Segfault in destructor using tsv::lock
Type: Bug Created on: 2020-02-16 10:44:05
Submitter: cyan Assigned to: nobody
Subsystem: 35. TclOO Package Severity: Severe
Priority: 5 Medium Last modified: 2026-11-01 20:37:55
Status: Closed Closed by: dkf
Resolution: Fixed Closed on: 2026-11-01 20:37:55
Version: 8.6.18, 9.0.5, 9.1.0
Description:
I'm not sure this is the right place for this (maybe Thread?), but I've hit a
segfault that involves the timing of an object's namespace deletion.  I'm
filing it against TclOO because I've hit a similar issue with a destructor
implemented in c that doesn't involve tsv::lock, but the this version uses only
Tcl script and standard packages distributed with Tcl releases:

package require Thread

oo::class create foo {
    variable thing

    constructor {} {
        set thing something
    }

    destructor {
        set local $thing
        rename [self] {}
        tsv::lock foo {
            puts "local: $local"   ;# works
            puts "thing: $thing"   ;# segfault
        }
    }
}

[foo new] destroy


I admit that "rename [self] {}" in the destructor is silly, but it works if the
tsv::lock isn't involved, and it's there to simulate the situation that occurs
in the real code where I found this problem, where it is less silly.

The crash happens on a call to
Tcl_CreateHashEntry(TclVarTable(contextPtr->oPtr->namespacePtr) in
ProcedureMethodCompiledVarConnect where contextPtr->oPtr->namespacePtr is NULL.

In the possibly related case with a C method and destructor, the C method does
a Tcl_DeleteCommandFromToken(interp, thisObjectCommand) on itself for the side
effect of destroying the instance (if a more direct way of doing this exists
please point me to it).  Sometimes this does the expected thing and calls the
destructor before returning from the Tcl_DeleteCommandFromToken, but  under
other circumstances (haven't found the relevant trigger yet, maybe NR context?)
it chains to the destructor after the c method that called
Tcl_DeleteCommandFromToken returns.  But in that case (without calling the
destructor yet) all the object metadata has been removed and the namespace
deleted when Tcl_DeleteCommandFromToken returns, causing some difficulties for
the destructor when it runs later.
User Comments:
serhiy.storchaka added on 2026-10-13 15:27:14:

Still crashes with 8.6.17, 9.0 and 9.1. It does not need Thread: eval or uplevel 0 instead of tsv::lock crash the same way. rename [self] {} deletes the namespace of the object, and the variable resolver of TclOO then looks up the variable in the deleted namespace.

Proposed fix in branch oo-destructor-deleted-namespace: do not resolve object variables when the namespace is already deleted; reading such a variable is then an error. Test oo-3.13.


dkf added on 2026-11-01 20:37:55:

It was a missing NULL check. Ought to be fixed now (9.0 onwards).