Xavier de Gaye <[email protected]> added the comment:
In both cases the destructor cannot be traced because it is invoked from
functions called from [1] call_trace() where tracing is disabled:
Case local variable 'a':
On line 12, just before executing this line with a pdb step command, the
object referenced by 'a' lives in frame->f_locals. The step command causes the
ceval loop to execute the bytecodes corresponding to the statement on line 12
and to eventualy call [2] call_trampoline() with a 'return' trace event and to
call PyFrame_FastToLocalsWithError() here. This last call causes the last
reference to the previous 'a' object in frame->f_locals to be decremented and
the object to be deallocated but without its destructor being traced as tracing
has been disabled in call_trace().
Case local variable 'b':
Upon exiting the frame of the 'main' function, pdb keeps a reference to
frame->f_locals through its own attribute curframe_locals. Next, after
returning to the 'main' caller, this reference is decremented since
pdb.curframe_locals references now another object and the dictionary referenced
by the previous frame f_locals (the frame of 'main') is deallocated, but this
happens within pdb where tracing is disabled and the destructor of the object
that had been referenced by 'b' is therefore not traced.
PR 6730 proposes a fix for both cases.
[1] https://github.com/python/cpython/blob/master/Python/ceval.c#L4247
[2] https://github.com/python/cpython/blob/master/Python/sysmodule.c#L462
----------
_______________________________________
Python tracker <[email protected]>
<https://bugs.python.org/issue33446>
_______________________________________
_______________________________________________
Python-bugs-list mailing list
Unsubscribe:
https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
