Dear Yves / perl5-porters, On Sun, 13 Nov 2022, demerphq wrote: > On Sun, 13 Nov 2022 at 02:22, Uwe Düffert <dueffert@uwe-dueffert.de> wrote: >> >> remember July? I did have a problem with perl-5.37.1 and perl-5.37.2 not >> being able to compile texinfo-6.8 and it took quite a few iterations to >> get that fixed... >> >> Well, TROUBLE IS BACK! >> >> While with perl-5.37.4 it was still possible to build texinfo-7.0, that >> fails with perl-5.37.5 - in a very similar way as in July: >> >> The perl command from texinfo-7.0 doc build as in >> https://uwe-dueffert.de/code/perl5.37.5/texinfo-7.0_doc_problem.sh >> has memory access issues again: >> https://uwe-dueffert.de/code/perl5.37.5/texinfo-7.0_valgrind.log >> >> Let me know if I shall try or provide anything, e.g. in case you have >> trouble reproducing that. > > When you run valgrind and things like that you should set PERL_DESTRUCT_LEVEL. > > B<NOTE 2>: To minimize the number of memory leak false alarms (see > L</PERL_DESTRUCT_LEVEL> for more information), you have to set the > environment variable PERL_DESTRUCT_LEVEL to 2. For example, like this: > > env PERL_DESTRUCT_LEVEL=2 valgrind ./perl -Ilib ... > > Looking at your valgrind output it looks to me like you ran it with a > normal destruct level, so it deliberately does not bother freeing > certain data structures at tear down. Sorry, I could have been more verbose. I was referring to the discusion in July were the valgrind output turned out to be the best indicator I could provide. The problem is *not* the long list of memory leaks after perl bails out, but the 3 topmost entries (invalid free/delete, address is 0 bytes, was alloced at) showing invalid memory access which cause perl5.37.5 to bail out with "malloc_consolidate(): unaligned fastbin chunk detected" That message is new [and more helpful than the perl5.37.1 and 5.37.2 behavior] and valgrind only confirms that with its words/details. Adding env PERL_DESTRUCT_LEVEL=2 as suggested does not seem to make any difference. In July for the problem in perl5.37.1 and perl5.37.2 there was the idea that a debug build of perl would provide additional insight, but it did not, with debug output the testcase from texinfo ran through successfully without bailing out, the memory access issue only appeared without debug stuff, perfectly reproducible for me. In perl5.37.3 and perl5.37.4 the problem was fixed, texinfo-7.0 docu builds successfully with those, but perl5.37.5 has a similar problem again... Regards UweThread Previous | Thread Next