Hello porters,
Witness this test program:
$ cat errsig.pl
$@ = 'test';
$SIG{'INT'} = sub {
$in = $@;
};
kill('INT', $$);
$after = $@;
if ($in eq 'test' && $after eq 'test') {
print("\$\@ preserved by sighandler\n");
} elsif ($after eq 'test') {
print("\$\@ localized in sighandler\n");
} else {
print("\$\@ clobbered by sighandler!\n");
}
Run with various version of Perl:
$ perl5.004 errsig.pl
$@ preserved by sighandler
$ perl5.005 errsig.pl
$@ preserved by sighandler
$ perl5.6 errsig.pl
$@ preserved by sighandler
$ perl5.8.4 errsig.pl
$@ clobbered by sighandler!
$ perl5.8.8 errsig.pl
$@ clobbered by sighandler!
I can see the C code that causes the behavior in mg.c:Perl_sighandler.
The G_EVAL flag was added in the call to call_sv(). The function should
enter scope and localize ERRSV before call_sv, and copy ERRSV and leave
scope thereafter. Then it should test and (if true) die with the copy,
rather than ERRSV, after the special processing that motivated the use
of G_EVAL.
Is this a bug? is it a feature?
Cheers,

Quoting Rafael Garcia-Suarez (rgarciasuarez (AT) gmail (DOT) com):
09/02/07, Merijn Broeren <merijnb (AT) iloquent (DOT) comwrote:
>I can see the C code that causes the behavior in mg.c:Perl_sighandler.
>The G_EVAL flag was added in the call to call_sv(). The function should
>enter scope and localize ERRSV before call_sv, and copy ERRSV and leave
>scope thereafter. Then it should test and (if true) die with the copy,
>rather than ERRSV, after the special processing that motivated the use
>of G_EVAL.
>
>Is this a bug? is it a feature?

What if there is a die() from within the signal handler ? Shouldn't $@
be propagated ? (not tested with older perls)

Yes, dieing with the copy would propagate the die message from the
signal handler. I just don't want the G_EVAL to clobber the results of
an eval. If I get a signal between the end of the eval block and my
testing $@ I'll get a false positive.

BTW, this was added by this change:

Mind if I supply a patch that preserves it?

Cheers,

Quoting Rafael Garcia-Suarez (rgarciasuarez (AT) gmail (DOT) com):
09/02/07, Merijn Broeren <merijnb (AT) iloquent (DOT) comwrote:
>What if there is a die() from within the signal handler ? Shouldn't $@
>be propagated ? (not tested with older perls)
>>
>Yes, dieing with the copy would propagate the die message from the
>signal handler. I just don't want the G_EVAL to clobber the results of
>an eval. If I get a signal between the end of the eval block and my
>testing $@ I'll get a false positive.

You confuse me, since there's no eval in your first example.

My first example was just a simple test case to see what happens to $@
when we call a signal handler. Normally I only care about $@ after an
eval.

K (even better with tests :)

Tests should always be there ofcourse.

Cheers,