The PyMem_Malloc(size) function has a well defined behaviour: if size is 0, a pointer different than NULL is returned. It looks like ckalloc(size) returns NULL if the size is NULL: see issue #21951 (bug on AIX).

Moreover, memory allocated by ckalloc() is not traced by the new tracemalloc module!

Attached patch replaces calls to ckalloc() and ckfree() with PyMem_Malloc() and PyMem_Free(). It may fix the issue #21951 on AIX.

There is still a call to ckfree() in Tkapp_SplitList(). This memory block is allocated internally by Tcl, not directly by _tkinter.c.

>        PyMem_Free(FREECAST argv);

FREECAST can be dropped here, PyMem_Free() always takes a void* pointer.

Event structs (Tkapp_CallEvent, VarEvent, CommandEvent) must have been allocated by the caller using Tcl_Alloc or ckalloc (see man Tcl_ThreadQueueEvent).

Here is updated patch which is synchronized with the tip after changes made in issue21951 and addresses my comments.

I read tkinter_pymem_2.patch.

Remaining calls to ckalloc():

* they are only used to allocate events passed later to Tcl_ThreadQueueEvent(). Tcl_ThreadQueueEvent doc explicitly says that the memory must be allocated by Tcl_Alloc or ckalloc, so it's correct (PyMem cannot be used).

Remaining calls to ckfree():

* Tkapp_SplitList() calls ckfree() on memory allocated by Tcl_SplitList(), it's correct.

* Tkapp_CallDeallocArgs() ckfree() on memory allocated by PyMem_Malloc() => wrong (see my review on Rietveld).

> * Tkapp_CallDeallocArgs() ckfree() on memory allocated by PyMem_Malloc() =>
> wrong 

Oh, you are right, thanks.

> (see my review on Rietveld).

Perhaps you forgot to press the "Publish" button.

tkinter_pymem_3.patch looks good to me.

And to me too. Please commit it, this is mainly your patch.

New changeset 9f1d3e6e6ce6 by Victor Stinner in branch 'default':
Closes #22336: attemptckalloc() with PyMem_Malloc() in _tkinter
http://hg.python.org/cpython/rev/9f1d3e6e6ce6

> And to me too. Please commit it, this is mainly your patch.

Ok, thanks, done.