eval: DESTROY may overwrite $@ - exceptions are not handled #5965
Comments
From burkhard.meier@gmx.deCreated by burkhard.e.f.meier@t-online.deThis is a bug report for perl from burkhard.e.f.meier@t-online.de, ----------------------------------------------------------------- eval { package SomeClass; sub new sub DESTROY ......results in the output: G:\tmp>perl -w eval.pl It seems that "die" first sets $@, then the local objects' destructors are Perl Info |
From @eserteBurkhard Meier (via RT) <perlbug@perl.org> writes:
I think the DESTROY method is responsible for preserving the value of Regards, -- Berlin Perl Mongers - http://berliner.pm.org |
From burkhard.meier@gmx.de
Come on, that's just a workaround... And who knows about this? BTW, does anybody if this specific problem will also occur in Perl6? Regards, |
From alex@chmrr.netHeya, -- |
From alex@chmrr.netperl-destroy.patchdiff -ru perl-current/pp_ctl.c perl-patched/pp_ctl.c
--- perl-current/pp_ctl.c 2005-07-08 13:03:02.000000000 -0400
+++ perl-patched/pp_ctl.c 2005-07-08 15:39:59.000000000 -0400
@@ -1432,6 +1432,7 @@
else {
sv_setpvn(ERRSV, message, msglen);
}
+ message = SvPV_const(ERRSV, msglen);
}
while ((cxix = dopoptoeval(cxstack_ix)) < 0
@@ -1471,6 +1472,12 @@
* minimal fix --GSAR */
PL_curcop = cx->blk_oldcop;
+ /* Leaving the block could run DESTROY blocks which might
+ * have altered $@; reset it here just in case */
+ if (message) {
+ sv_setpvn(ERRSV, message, msglen);
+ }
+
if (optype == OP_REQUIRE) {
const char* msg = SvPVx_nolen_const(ERRSV);
SV * const nsv = cx->blk_eval.old_namesv;
diff -ru perl-current/t/op/die.t perl-patched/t/op/die.t
--- perl-current/t/op/die.t 2005-07-06 16:38:38.000000000 -0400
+++ perl-patched/t/op/die.t 2005-07-08 15:42:12.000000000 -0400
@@ -1,6 +1,6 @@
#!./perl
-print "1..15\n";
+print "1..16\n";
$SIG{__DIE__} = sub { print ref($_[0]) ? ("ok ",$_[0]->[0]++,"\n") : @_ } ;
@@ -72,3 +72,27 @@
print "not " unless $ok;
print "ok 15\n";
}
+
+# [perl #17650] DESTROY can unset $@
+{
+ package SomeClass;
+
+ sub new {
+ my $self = {};
+ bless $self;
+ }
+
+ sub DESTROY {
+ eval { 1; };
+ }
+}
+
+{
+ local $SIG{__DIE__};
+ eval {
+ my $x = new SomeClass;
+ die;
+ };
+ print($@ ? "ok 16\n" : "not ok 16\n");
+}
+
|
From @iabynOn Fri, Jul 08, 2005 at 03:59:34PM -0400, Alex Vandiver wrote:
I don't think this fix is robust. You save the current PV value of ERRSV, -- |
From alex@chmrr.netOn Sun, 2005-07-10 at 14:29 +0100, Dave Mitchell wrote:
-- |
From alex@chmrr.netperl-destroy.patchdiff -ru perl-current/sv.c perl-patched/sv.c
--- perl-current/sv.c 2005-07-08 13:03:02.000000000 -0400
+++ perl-patched/sv.c 2005-07-18 00:12:01.000000000 -0400
@@ -5519,6 +5519,10 @@
PUSHMARK(SP);
PUSHs(tmpref);
PUTBACK;
+
+ /* Localise $@ so DESTROY caused by die can't reset $@ */
+ save_item(ERRSV);
+
call_sv((SV*)destructor, G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
diff -ru perl-current/t/op/die.t perl-patched/t/op/die.t
--- perl-current/t/op/die.t 2005-07-06 16:38:38.000000000 -0400
+++ perl-patched/t/op/die.t 2005-07-16 02:25:44.000000000 -0400
@@ -1,6 +1,6 @@
#!./perl
-print "1..15\n";
+print "1..17\n";
$SIG{__DIE__} = sub { print ref($_[0]) ? ("ok ",$_[0]->[0]++,"\n") : @_ } ;
@@ -72,3 +72,43 @@
print "not " unless $ok;
print "ok 15\n";
}
+
+# [perl #17650] DESTROY can unset $@
+{
+ package SomeClass;
+ sub new { return bless {}; }
+ sub DESTROY {
+ eval { 1; };
+ }
+}
+
+{
+ package SomeOtherClass;
+ sub new { return bless {}; }
+ sub DESTROY {
+ eval { die bless {}, "Moose"; }
+ }
+}
+
+{
+ local $SIG{__DIE__};
+ eval {
+ my $x = new SomeClass;
+ die;
+ };
+ print($@ ? "ok 16\n" : "not ok 16\n");
+
+ eval {
+ my $x = new SomeOtherClass;
+ die bless {}, "Thingy";
+ };
+ print($@ ? "ok 17\n" : "not ok 17\n");
+ print(ref $@ eq "Thingy" ? "ok 18\n" : "not ok 18\n");
+
+ eval {
+ my $s = new SomeOtherClass;
+ };
+ print($@ ? "not ok 19\n" : "ok 19\n");
+}
+
+
|
From @iabynOn Mon, Jul 18, 2005 at 01:14:43AM -0400, Alex Vandiver wrote:
I like the idea of localising $@ before calling DESTROY, but unfortunately lib/warnings..............................PROG: The behaviour being tested here is that a die in a DESTROY gets converted I'm not really sure offhand of the best way round this. Perhaps have an -- |
From @nwc10On Tue, Jul 19, 2005 at 01:19:19AM +0100, Dave Mitchell wrote:
Which means that it wasn't tested before being sent to the list. make test exists for a reason. Nicholas Clark |
From guest@guest.guest.xxxxxxxx
There's another problem with this work-around -- it breaks in the other tin-foil:/tmp glasser$ cat /tmp/localdie sub i_localize_dollar_at { # This could certainly be DESTROY eval { i_localize_dollar_at() }; my if ($error) { That is, localizing $@ in DESTROY solves the problem of "an eval inside I'm not sure if Alex's patch has this issue too. --dave |
From @smpeters
This bug has been resolved by documenting this behavior with a patch in |
|
@smpeters - Status changed from 'open' to 'resolved' |
Migrated from rt.perl.org#17650 (status was 'resolved')
Searchable as RT17650$
The text was updated successfully, but these errors were encountered: