overload stringify 5.18-5.22

[atoomic]

Env:
perl 5.20.2
B::C using bc522 with HEAD @734e776 ( cannot use 23016e1 with 5.20 )

Steps to reproduce

> cat test.pl
package OverloadTest;
use overload '""' => sub { ${$_[0]} };

package main;
my $foo = bless \(my $bar = "ok"), "OverloadTest";
print $foo."\n";

> perl test.pl
ok
> perlcc -r test.pl
pccOxMXJ.c:8483: warning: initialization makes pointer from integer without a cast
....

Seems a regression as this is working on 5.14

[rurban]

This started failing with 5.18.
testcase: t/issue172.t also with t/testc.sh 172 and now t/testc.sh 219
See old ticket GH #174

Maybe the root-cause is my workaround for the broken %version:: SvSTASH. See https://rt.cpan.org/Ticket/Display.html?id=81635

Update: It is not. Work in the branch overload-219

[toddr]

Right. You reported the bug against version except Devel::Peek and version are both perl as upstream. https://rt.cpan.org/Public/Bug/Display.html?id=107459

[toddr]

Bisected in blead.

c07f9fb2c73bebcf70fabbf464c0c3452af4fcbf is the first bad commit
commit c07f9fb2c73bebcf70fabbf464c0c3452af4fcbf
Author: Father Chrysostomos [email protected]
Date: Tue Nov 20 13:47:27 2012 -0800

Revert "Revert "8c34e50dc slowed down detruction with no DESTROY""

This reverts commit 95f9781bc2fad025553db0160ef9c2c5363312a1.

Now that the crash has been fixed by the preceding commit, we can
reinstate 7cc6787e9db.

:100644 100644 2d1d887fe8b4256b2b5682dc6d26065f12ab5a6d be2038f4e7cdaf2d918c5df419348fca09290ab0 M mro.c
:100644 100644 9f5c157a87b3f41f96566337e612a3674c129bbd a7914988e7861ead362d718960a408fd5bd0a5e3 M sv.c
bisect run success

[toddr]

Output at offending commit:

cd t && (rm -f perl; /bin/ln -s ../perl perl)
SV = IV(0x8c1e234) at 0x8c1e238
REFCNT = 1
FLAGS = (TEMP,ROK)
RV = 0x8c1e418
SV = PVHV(0x8c24010) at 0x8c1e418
REFCNT = 4
FLAGS = (OOK,OVERLOAD,SHAREKEYS)
STASH = 0x10/root/bisect.sh: line 10: 26914 Segmentation fault ./perl -Ilib -MDevel::Peek=Dump -E'version->new("v5.22.0");Dump(%version::)'

[toddr]

Reported to upstream. https://rt.perl.org/Public/Bug/Display.html?id=126410

[toddr]

Even with the patch applied to upstream, the original problem continues to be an issue so I think the upstream Devel::Peek regression is probably a distraction from the primary issue.

[toddr]

Test results with fix sent to upstream compiled in to perl 5.22.0

$>perlcc -e'package OverloadTest; use overload q{""} => sub { ${$_[0]} }; package main; my $foo = bless \(my $bar = "ok"), "OverloadTest"; print $foo."\n";' ; ./a.out
./a.out
/usr/local/cpanel/3rdparty/perl/522-debug/bin/perlcc: Unexpected compiler output
[WARNING] skip SvSTASH for overloaded HV %Errno:: flags=0x3280000c
 [WARNING] unknown (HV*)&sv_list[455] found in @static_free
OverloadTest=SCALAR(0x8112fb0)

[toddr]

Todd to bisect this today.

[rurban]

Patched my_curse for -fno-destruct (-O3) so we don't even need the core patch on older perls.
See bd78fe8

[toddr]

$>git bisect bad
bee7c5743fabd49c584c5edbb637c66c5c25ca9a is the first bad commit
commit bee7c5743fabd49c584c5edbb637c66c5c25ca9a
Author: Father Chrysostomos <[email protected]>
Date:   Fri May 18 17:02:39 2012 -0700

    sv.c: Don’t fiddle with AMAGIC in sv_bless

    Since overloading itself now checks whether caches are up to date, and
    since changes to the stash (@ISA, methods) turn the flag on and over-
    loading itself turns the flag off when it can, sv_bless no longer
    needs to deal with it at all.

:040000 040000 4a9679f7a036de75ae61b9dda7ebecc3bf5335ba 634b1a9b9c56600fac5f9d9e4335474d106cd95e M  lib
:100644 100644 0c940cb434f461f262ab8263d2f9f1552a04bcf2 1e7f4d2e341d79f5306194808fd65530a1f628f6 M  sv.c

diff --git a/sv.c b/sv.c
index 0c940cb..1e7f4d2 100644
--- a/sv.c
+++ b/sv.c
@@ -9518,11 +9518,6 @@ Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
     SvUPGRADE(tmpRef, SVt_PVMG);
     SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));

-    if (Gv_AMG(stash))
-       SvAMAGIC_on(sv);
-    else
-       (void)SvAMAGIC_off(sv);
-
     if(SvSMAGICAL(tmpRef))
         if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
             mg_set(tmpRef);

[rurban]

Uh, now that's interesting and should be pretty simple to fix, thanks!
It says that we have to trigger Gv_AMG, which initializes and updated the overload cache by ourselves now.

[atoomic]

a workaround for it is to adjust ISA at run time (using the overload branch)

#!perl 

package OverloadTest;
use overload '""' => sub { ${$_[0]} };

package main;
my $foo = bless \(my $bar = "ok" ), "OverloadTest";

INIT {
    push @OverloadTest::ISA, 'overload';
}
print "$foo\n";

> perlcc -S -r test.pl
ok

[toddr]

This is because the HvMROMETA (xpvhv_aux) is updated for the stash when @isa is changed. We need to start capturing and saving xpvhv_aux for all stashes for now on. This also would probably take away the need to set 'c3' on classes during XS init since this is simply populating that data.

[atoomic]

this dirty commit fixes the overload
cpanel@356e9e1

[toddr]

@rurban While that does fix things, the correct solution is to save and properly attach xpvhv_aux where necessary. From the looks of it, the data is static?

[rurban]

No, it's dynamic. I only save it for RITER/EITER, see there. However the code in mro_isa_changed_in or gv_stashpvnalready calls HvMROMETA() which initializes the xpvhv_aux struct.
Testing in branch debug-stashcv-gh219 dbfa262

I tested a static xpvhv_aux and it was not possible.

[rurban]

Fixed with 1.52_18, thanks @atoomic