From @exodist

This issue effects multiple versions of perl, but in different ways. I have
attached a test script that demonstrates the issue.

Here is the output in 5.20.1 and 5.10.1

5.20.1

  ok 1 - outer eval 1 completed

ok 2 \- We can escape
ok 3 \- outer eval 2 completed
not ok 4 \- We can escape when there is a BEGIN between us and the

escape point
# Failed test 'We can escape when there is a BEGIN between us and
the escape point'
# at test.pl line 37.
1..4
# Looks like you failed 1 test of 4.

5.10.1

  ok 1 - outer eval 1 completed

ok 2 \- We can escape
ok 3 \- outer eval 2 completed
ok 4 \- We can escape when there is a BEGIN between us and the escape

point
1..4
segmentation fault

From @exodist

test.pl

From @cpansprout

On Sun Nov 23 16​:34​:34 2014, exodist7@​gmail.com wrote​:

This issue effects multiple versions of perl, but in different ways. I have
attached a test script that demonstrates the issue.

Here is the output in 5.20.1 and 5.10.1

5.20.1

ok 1 \- outer eval 1 completed

ok 2 \- We can escape
ok 3 \- outer eval 2 completed
not ok 4 \- We can escape when there is a BEGIN between us and the

escape point
# Failed test 'We can escape when there is a BEGIN between us and
the escape point'
# at test.pl line 37.
1..4
# Looks like you failed 1 test of 4.

5.10.1

ok 1 \- outer eval 1 completed

ok 2 \- We can escape
ok 3 \- outer eval 2 completed
ok 4 \- We can escape when there is a BEGIN between us and the escape

point
1..4
segmentation fault

Here is a simplified case​:

$ ./perl -Ilib -e 'FOO​: { eval "BEGIN { last FOO }" or die $@​ }'
Label not found for "last FOO" at (eval 1) line 1.
BEGIN failed--compilation aborted at (eval 1) line 1.

$ perl5.14.4 -e 'FOO​: { eval "BEGIN { last FOO }" or die $@​ }'
Assertion failed​: (SvTYPE(av) == SVt_PVAV), function Perl_av_len, file av.c, line 730.
Abort trap​: 6

$ perl5.18.1 -e 'FOO​: { eval "BEGIN { last FOO }" or die $@​ }'
Can't use an undefined value as a subroutine reference.
END failed--call queue aborted at -e line 1.

Fixing this is very hard. ‘last’ expects to be able to find the target op and return that to the runloop. But with BEGIN we have multiple nested runloopss. I think the only way to fix it is to use the equivalent of ‘die’ internally to get to the right runloop. But that involves longjmping, which is something I stay away from. :-)

--

Father Chrysostomos

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