panic: sv_len_utf8 cache with -d and -DDEBUGGING #13604
Comments
From @kentfredricThis is a bug report for perl from kentnl@cpan.org, This problem is hard for me to replicate as a reduced test case, because it seems linked to the stack depth of the problem. I have 2 scripts that trigger it, but they both require CPAN modules installed, and both require running under -d and -DDEBUGGING to trigger the failure. The crux of it is this snippet in perl5debug.pl causes the error for some reason, as it for some reason does not like calling length($sub) or substr( $sub, ) # If the last ten characters are '::AUTOLOAD', note we've traced Simply wrapping both these calls as : length("$sub") and substr( "$sub", ) Suppresses the problem entirely. Augmenting debug.pl like such prior to the length call is interesting: if ("$sub" =~ /^Git::Wrapper::Plus::Branches::/) { Because oddly, it yields SV = PV(0x2325470) at 0x1c7a170 Take note: current_branch has no magic, _current_branch_name has magic. And when invoking _current_branch_name directly, it does this: SV = PV(0x170a3c0) at 0x105f170 Noting that the previous dump showed _current_branch_name with magic, and this dump not, and instead, the magic is moved to ->git The other way to trigger it is: perl -d -E' use sanity;' , which causes the failure somewhere in Import::Into , but that loads way too many XS Magic things for me to try digging there. ( I ruled out it being a common failure point with autoviv and friends by slicing Moo and Moo::_Utils to not import strictures and checked the strictures modules were not being loaded with -d:TraceUse ) Flags: Site configuration information for perl 5.19.8: Configured by kent at Wed Jan 29 21:15:13 NZDT 2014. Summary of my perl5 (revision 5 version 19 subversion 8) configuration: @INC for perl 5.19.8: Environment for perl 5.19.8: |
From @kentfredric |
From perl5-porters@perl.orgKent Fredric wrote:
Here is a smaller case: $ perl5.19.9 -Mblib -de 'use utf8; require Git::Wrapper::Plus::Branches;Git::Wrapper::Plus::Branches::current_branch()' And blib/lib/Git/Wrapper/Plus/Branches.pm has been gutted and reduced to: package Git::Wrapper::Plus::Branches; The 'use utf8' and 'require' are necessary, so it seems to have some- |
|
The RT System itself - Status changed from 'new' to 'open' |
From @kentfredricMy other institutions were that it might be related to being called with a utf8 name but not being defined with a utf8 name, or vice versa, maybe causing problem. Or perhaps the problem flares up due to sharing a compile context somehow. |
From @kentfredricIf you rig @INC right, you can get this happening in a single file. Though its strange, seems entirely dependent on the `use utf8` value of the top context, and the value of utf8 in the compiled context is ignored. :/ |
From @kentfredric |
From perl5-porters@perl.orgKent Fredric wrote:
%INC is enough, as I have found out.
require Bareword causes the Bareword package to be vivified. But for $ ./perl -Ilib -e 'use utf8; ++$INC{"Branches.pm"};require Branches; use Devel::Peek; Dump ref bless [], Branches' Look, a 'one'-liner! $ PERL5DB='${^UTF8CACHE}=-1; sub DB::DB{} sub DB::sub{length($DB::sub); goto &$DB::sub}' ./miniperl -Ilib -de 'use utf8; ++$INC{"Branches.pm"};require Branches;Branches::cb();package Branches;sub _cbn {}sub cb{ _cbn();}' (BTW, ${^UTF8CACHE} is -1 by default under -DDEBUGGING, but you can Bisect points to me: commit a7999c0 Make sure $DB::sub is callable I suspect I just uncovered an existing utf8 cache bug, but I am |
From perl5-porters@perl.orgI wrote:
I think adding SvSETMAGIC to util.c:Perl_get_db_sub will fix it. |
From perl5-porters@perl.orgI wrote:
Hence, the workaround (untested) is to do this at the top of your mod- BEGIN { delete $Git::Wrapper::Plus::{"Branches::"} } |
From @ikegamiOn Sat, Feb 15, 2014 at 1:29 PM, Father Chrysostomos <sprout@cpan.org>wrote:
Well, the length cache is only used for UTF8-1 strings, so that's not |
From @cpansproutFixed in 795eb8c. -- Father Chrysostomos |
|
@cpansprout - Status changed from 'open' to 'resolved' |
From @demerphqOn 16 February 2014 06:00, Eric Brine <ikegami@adaelis.com> wrote:
UTF8-1? Yves -- |
From perl5-porters@perl.orgYves Orton asked:
Typo. UTF-8. |
From @ikegamiOn Mon, Feb 17, 2014 at 7:00 AM, demerphq <demerphq@gmail.com> wrote:
UTF8=1, as in the UTF8 flag set. |
Migrated from rt.perl.org#121255 (status was 'resolved')
Searchable as RT121255$
The text was updated successfully, but these errors were encountered: