Carview!

CARVIEW

MOTORHOMES

Select Language

HTTP/2 200 cache-control: max-age=43200 server: Combust/Plack (Perl) vary: Accept-Encoding content-encoding: gzip content-length: 3491 content-type: text/html; charset=utf-8 last-modified: Sat, 11 Oct 2025 21:37:41 GMT traceparent: e9053d4097f3ed6b4a18366f4ef3c58c strict-transport-security: max-age=15768000 Re: my $@ - a proposal - nntp.perl.org

Front page | perl.perl5.porters | Postings from July 2008

Re: my $@ - a proposal

Thread Previous | Thread Next

From:

David Nicol

Date:

July 1, 2008 16:29

Subject:

Re: my $@ - a proposal

Message ID:

934f64a20807011629s780e7e2ch41f7d4305fe9ff6c@mail.gmail.com

On Tue, Jul 1, 2008 at 8:48 AM, Aristotle Pagaltzis <pagaltzis@gmx.de> wrote:
>  it's not just programming language concern
> trolling, as Sean O'Rourke put it, but a problem that needs to
> be solved despite its low probability of occurence.
so if $@, $! and $? were all localized before calling DESTROY methods,
with no fancy logic, just localized, as recomended, the only pattern
that would interfere with is
L<https://en.wikipedia.org/wiki/Resource_acquisition_is_initialization
RAII>.  Existing RAII code that wants to check things set in DESTROY
could be made to use another mechanism, shadowing the globals of
concern within the DESTROY method.  The recommendation in the
documentation could get amended to state "Prior to perl 5.12 ... or
without a perl containing the 'PROTECT_PRECIOUS_FROM DESTROY' patch,
and for maximum portability, localize $@ before doing an eval within a
DESTROY method"
Destruction is not documented to occur immediately, the fact that it
does is an implementation detail, so any code that does in fact rely
on the current behavior could be considered breakable poor form
(although RAII is so cool that any kind of alternate destruction
deferment system should include some kind of attribute, perhaps
:ephemeral, to indicate that the variable in question needs to be
destroyed on scope exit instead of simply added to the recyclable pile
or whatever)
unfortuantely after merely restoring all of them, a unprotected die in
a DESTROY doesn't exit the program any more.  if it ever did.  Anyway
here's a patch against 5.8.8 to implicitly localize $!, $@ and $?
before DESTROY is called.
$ diff -u  pristine_perl-5.8.8/sv.c perl-5.8.8/sv.c
--- pristine_perl-5.8.8/sv.c    2006-01-16 06:22:21.000000000 -0600
+++ perl-5.8.8/sv.c     2008-07-01 18:23:18.777923600 -0500
@@ -5164,6 +5164,9 @@
                stash = SvSTASH(sv);
                destructor = StashHANDLER(stash,DESTROY);
                if (destructor) {
+                   SV* orig_error = newSVsv( ERRSV ); /* make local
copy of $@ */
+                   SV* orig_bang = newSVsv( get_sv("!", TRUE )); /*
make local copy of $! */
+                   SV* orig_hook = newSVsv( get_sv("?", TRUE )); /*
make local copy of $? */
                    SV* const tmpref = newRV(sv);
                    SvREADONLY_on(tmpref);   /* DESTROY() could be naughty */
                    ENTER;
@@ -5185,6 +5188,15 @@
                        SvROK_off(tmpref);
                    }
                    SvREFCNT_dec(tmpref);
+
+                    /* restore $@, $! and $? from local copies */
+                    sv_setsv(ERRSV, orig_error);
+                    sv_setsv(get_sv("!", TRUE ), orig_bang);
+                    sv_setsv(get_sv("?", TRUE ), orig_hook);
+
+                   SvREFCNT_dec(orig_error);
+                   SvREFCNT_dec(orig_bang);
+                   SvREFCNT_dec(orig_hook);  /* Harrrrr, mateys!  */
                }
            } while (SvOBJECT(sv) && SvSTASH(sv) != stash);

Thread Previous | Thread Next