pads being reused in coderefs called from DESTROY #13170
Comments
From @doyCreated by @doyGiven this script: sub foo { my $got; $got = $other; $block->(); { sub DESTROY { # "exit(0) unless fork" also triggers the bug, $got is keeping its previous value when foo() is called from the DESTROY commit 840a7b7 Re: [PATCH] The largest hoax of all times? Subject: Re: [PATCH] The largest hoax of all times? Subject: Re: [PATCH] The largest hoax of all times? Fix the unpredictable order of DESTROYs. p4raw-id: //depot/perl@7991 Perl Info |
From @cpansproutOn Thu Aug 15 16:17:42 2013, doy@tozt.net wrote:
I cannot reproduce this on darwin or linux. -- Father Chrysostomos |
|
The RT System itself - Status changed from 'new' to 'open' |
From @iabynOn Fri, Aug 16, 2013 at 05:17:33AM -0700, Father Chrysostomos via RT wrote:
I can't reproduce it on linux either. Tried lots of versions, threaded / -- |
From victor@vsespb.ruReproduced under perl 5.10.1 (linux), but only after adding 'use got the following message: (in cleanup) this is clearly a bug at 2.pl line 7. On Thu Aug 15 16:17:42 2013, doy@tozt.net wrote:
5.18/bin:/home/doy/.bin/nethack:/home/doy/.bin:/usr/local/sbin:/usr/local/bin:/usr/lib/ccache/bin:/usr/local/sbin:/usr/local/bin:/usr/bin:/usr/bin/vendor_perl:/usr/bin/core_perl
PERLBREW_PATH=/home/doy/perl5/perlbrew/bin:/home/doy/perl5/perlbrew/perls/perl-
|
From @cpansproutOn Fri Aug 16 08:03:36 2013, vsespb wrote:
I get it with warnings, too. It is hard to debug this sort of thing $ ./perl -ILib -w foo (gdb) run Program exited normally. Program exited normally. It exhibits the reported buggy behaviour consistently with no breakpoint. With a breakpoint it runs to completion with no error, fails an So, whose bug *is* this? -- Father Chrysostomos |
From @cpansproutOn Fri Aug 16 08:42:07 2013, sprout wrote:
... debugging stuff ...
I think we can disregard that. It is probably a compiler/debugger If I add ‘warn $$’ to the top of the script at change the fork statement $ ./perl -ILib foo It appears to have something to do with fork’s return value being on the -- Father Chrysostomos |
From @cpansproutOn Fri Aug 16 08:51:24 2013, sprout wrote:
Fatal warnings give me a segmentation violation. That may be another 11623 0x000000010030cec2 in Perl_die_unwind (msv=0x1057f5970) at -- Father Chrysostomos |
From victor@vsespb.ruadding $SIG{__WARN__}=sub{die}; change error message to (in cleanup) (in cleanup) this is clearly a bug 3 9258 at i.e. there is a warning, indeed. On Fri Aug 16 08:55:11 2013, sprout wrote:
|
From @cpansproutOn Fri Aug 16 08:55:11 2013, sprout wrote:
No, actually it has nothing to do with fork. Just call exit(0) and you exit() results in a call to die_unwind in pp_ctl.c, which calls POPSUB, Unwinding the savestack causes the anon sub’s lexicals to be freed
Yes. it is: $ ./perl -Ilib -e 'use warnings FATAL=>"all"; DESTROY{die} bless []; exit' I don’t see an obvious way to fix either of these bugs. -- Father Chrysostomos |
From victor@vsespb.rualso happens even if program is not terminated at all: ... On Sat Aug 17 07:04:37 2013, sprout wrote: |
From @cpansproutOn Sat Aug 17 07:04:37 2013, sprout wrote:
Well I have an idea about the latter. The error in the destructor is We should simply bypass fatal warnings here. In fact, I wonder why this One reason I ask is that changing it only to ignore fatal warnings will (I’m still thinking about the original issue in the ticket. That’s a -- Father Chrysostomos |
From @doyOn Sat, Aug 24, 2013 at 07:38:55PM -0700, Father Chrysostomos via RT wrote:
It's really pretty clear that fatal warnings were never thought through
I think this is a good idea (but yeah, also a separate bug). -doy |
From @cpansproutOn Sat Aug 24 19:58:41 2013, doy@tozt.net wrote:
Implemented on the sprout/fataldestroy branch. I had to change quite a
(I can’t fix, or even diagnose, bugs without finding more. :-) -- Father Chrysostomos |
From @cpansproutOn Sat Aug 17 07:04:37 2013, sprout wrote:
The code in question is spread out all over the place. This piece in /* pop unwanted frames */ if (ix < cxstack_ix) { if (ix < 0) dounwind(ix) causes CvDEPTH to be decreased. LEAVE_SCOPE(oldsave) frees What makes this really tricky is that we cannot simply call LEAVE_SCOPE Trying to disentangle that (and differently for exit/die/die in Maybe what we need to do is put the clearing of lexicals inside POPSUB, This is a mess.... -- Father Chrysostomos |
From @cpansproutOn Sun Aug 25 08:18:03 2013, sprout wrote:
Fixed in commit 2537512. :-) -- Father Chrysostomos |
|
@mauke - Status changed from 'open' to 'resolved' |
Capture or suppress 'exiting' warnings Former name referred to RT ticket 119311. In the Oct 2019 transfer of tickets to GitHub issues, the old ticket became Perl/perl5#13170. Update MANIFEST
Migrated from rt.perl.org#119311 (status was 'resolved')
Searchable as RT119311$
The text was updated successfully, but these errors were encountered: