From @zoffixznet

To get code that segfaults​:

  cd $(mktemp -d); git clone https://github.com/zoffixznet/perl6-Temp-Path . ; git checkout ca5ab2816705cf36911b91e4029224b4d4644a2d;

Now run this several times and it'll segfault ~50% of the time​:
 
  perl6 -Ilib t/03-DESTROY.t

The test file already does `use lib <lib>` and if you remove `-Ilib` switch, the segfault seems to disappear​:

  perl6 t/03-DESTROY.t

I tried making a small test case that just had a dummy module and was doing `use lib <lib>` but that didn't segfault.

Perhaps it's got to do with loading some installed modules?

From @zoffixznet

Also noticed another failure mode in a more recent version[^1] of the test file​:

fails (likely due to segv)​: prove -e 'perl6 -Ilib' -r t/03-DESTROY.t
succeeds​: prove -e 'perl6 -Ilib' -vlr t/03-DESTROY.t

And removing the `-Ilib` from prove line didn't help it, but removing the `use lib` from the
test file and using solely -Ilib avoided the problem.

[1] https://github.com/zoffixznet/perl6-Temp-Path/blob/0c2d855ffd4e602aafc34297f17d2a428c080cdd/t/03-DESTROY.t

From @zoffixznet

FWIW, it segfaults like below, not during module loading​:

cpan@​toaster​:/tmp/tmp.KcLTWw3PYk$ perl6 -Ilib t/03-DESTROY.t
1..4
  1..2
  ok 1 - exists before DESTROY
  ok 2 - does not exist after DESTROY
ok 1 - file
  1..2
  ok 1 - derivative exists before DESTROY
  ok 2 - derivative still exists after DESTROY
ok 2 - file, after changing path
  1..2
  ok 1 - exists before DESTROY
  ok 2 - does not exist after DESTROY
ok 3 - dir
  1..2
Segmentation fault
cpan@​toaster​:/tmp/tmp.KcLTWw3PYk$

And I tried running with RAKUDO_MODULE_DEBUG enabled but then it doesn't segfault.

From @zoffixznet

Some discussion​: https://irclog.perlgeek.de/perl6-dev/2017-06-11#i_14715962

02​:00 ugexe well, i have a good idea what it is then
02​:01 its the `run` that does precomp, and its the bug that has to do with stderr/stdout buffer, which one you read first, and which one you close first
02​:01 eveo It segfaults after running some tests tho. Like this is one of the outputs​: https://rt-archive.perl.org/perl6/Ticket/Display.html?id=131510#txn-1465483
02​:01 ugexe the prove thing would be related because -v probably adds a bunch of extra stuff to stderr
02​:02 eveo actually, seems to consistently segfault like that, after starting 3rd test grup
02​:03 ugexe yea, and does it change if you rm -rf lib/.precomp ?
02​:03 eveo It didn't segfault after removing it, but next run segfaulted at the same place
02​:04 ugexe yep
02​:04 so there should be a difference in RAKUDO_MODULE_DEBUG loading

From @ugexe

This appears to be resolved - possibly in rakudo/rakudo@37250ed

The RT System itself - Status changed from 'new' to 'open'