Tcl Source Code

Running valgrind on the Tcl test suite reveals numerous
memory leaks in unload.test.  They appear to be the
loaded module failing to clean up after itself;
since programmers are likely to use its code as an
example of writing an unloadable module, it should
do a better job.

valgrind log is attached.

Logged In: YES 
user_id=148712
Originator: NO

That was only a partial fix - TclPtrSetVar is a hairy monster wrt to the new value, and it remains one after this change. Tcl_SetVar2 needs to manage the refCount properly. 

Reverting the old fix, replacing with new fix in HEAD, 8.4 already correct.

Logged In: YES 
user_id=80530
Originator: NO


I suspect this needs a backport
for 8.4.15 ?

Logged In: YES 
user_id=99768
Originator: YES

Verified that the leaks are plugged; leaving this open in case the maintainer wants to add tclTest.c code to exercise this path (without dltest, we'd never have seen it).

Logged In: YES 
user_id=148712
Originator: NO

There was indeed a refCount bug as mistachkin described; fix committed.

Logged In: YES 
user_id=113501
Originator: NO


There does not seem to be any other place in the core that uses Tcl_SetVar2 with TCL_APPEND_VALUE.  The only place I see using TCL_APPEND_VALUE is in tclExecute.c with TclPtrSetVar.  Perhaps the string object allocated by Tcl_SetVar2 is always leaked if TCL_APPEND_VALUE is used?

Logged In: YES 
user_id=113501
Originator: NO


The pkgua library is leaking the interpTokenMap hash table itself (by not calling Tcl_DeleteHashTable in PkguaFreeTokensHashTable); however, the Tcl_SetVar leaks appear to be false alarms.  The memory being leaked is not owned by the package and the variables in question are not referred to by the package except to append values to them.

File Added - 227376: leakunload.txt