$@ can't be relied on?

sherab has asked for the wisdom of the Perl Monks concerning the following question:

Hello esteemed monks, I was using the perl critic on the code below....

eval {
      #
      # Prepare the SQL Statment
      #
      my $sth = $dbh->prepare($sql) or die "Can't prepare SQL statemen
+t: $DBI::errstr";

      #
      # Execute the SQL Statment
      #
      my $rc = $sth->execute(@values) or die "Can't execute SQL statem
+ent: $DBI::errstr";

      #
      # Fetch the result
      #
      $data = $sth->fetchall_arrayref({});

      $sth->finish;

      #
      # check for problems which may have terminated the fetch early
      #
      die "Can't query result: $sth->errstr" if $sth->err;
   };

   if ($@) {
      print STDERR "ERROR [query_database()]: $@";
      return 0;
   }
[download]

I got the notice that "You can't depend upon the value of $@/$EVAL_ERROR to tell whether an eval failed."
Is that because this traps the last syntax error and not warning messages?. Any advice on $@ would be appreciated.

Got a LOT of great answers. Monks you are the best!

Comment on $@ can't be relied on? Download Code

Replies are listed 'Best First'.
Re: $@ can't be relied on? by jettero (Monsignor) on Nov 17, 2009 at 19:53 UTC
That's correct. There are rare cases where it doesn't work. The correct way to do this is as follows: `my $res = eval { might_explode(); 1 }; die "omgish: $@\n" unless $res; # Also, why invoke print and exit whe +n this will do?` [download] Note that `die $@ if $@` almost always works. I can't recall the example case I saw, but it's reproducible stuff. It's not like it just randomly doesn't work. Well, maybe with threads. EDIT: Oh, right. Below, JavaFan gives an example of a case where it doesn't work. Instead of `local $@; die`, you can easily see there being a function() or an $obj->method that does the equivalent of that without your realizing it. I believe that's what I actually ran into (years ago). -Paul	[reply] [d/l] [select]
Re^2: $@ can't be relied on? by Anonymous Monk on Nov 18, 2009 at 03:03 UTC
`eval { might_explode(); 1 } or die "omgish: $@\n"; __END__ omgish: Undefined subroutine &main::might_explode called at - line 2.` [download]	[reply] [d/l]
Re: $@ can't be relied on? by JavaFan (Canon) on Nov 17, 2009 at 21:38 UTC
Easiest case where it doesn't work: `eval {local $@; die}; if ($@) {say "eval failed"} else {say "eval succeeded"} __END__ eval succeeded` [download]	[reply] [d/l]
Re: $@ can't be relied on? by JadeNB (Chaplain) on Nov 17, 2009 at 20:24 UTC
jettero has already mentioned the solution; now, as to what the problem is, I found the following discussion in Devel::EvalError via a link in a post by ELISHEVA: Why $@ is unreliable.	[reply]
Re: $@ can't be relied on? by przemo (Scribe) on Nov 17, 2009 at 20:37 UTC
Take a look at a source of a neat module Try::Tiny, which has "safe eval"-ing as the only goal. Can be used as a replacement of own eval-s. Does more or less what other respondents wrote, plus localizing $@.	[reply]
Re: $@ can't be relied on? by mje (Curate) on Nov 17, 2009 at 20:31 UTC
I believe the cases where this might not work is for instance if something goes out of scope in the eval and a DESTROY is called - it could do something else that might change $@.	[reply]