|
[text/plain 2.1k]
This patch was manually applied in the latest version (0.17006) after
a less optimal fix was applied in 0.17005. Thanks for the fix, the
testcase, and the report!
Regards,
Shlomi Fish
On Fri Sep 22 07:58:57 2006, waba[...]waba.be wrote:
> Error version: Error-0.17004
> Perl version: v5.8.6
> OS version: OpenBSD 3.9 / i386
> (but confirmed on other versions/OSes, such as 5.8.8/Linux/amd64).
>
> Most OO languages allow for "exception translation" - that is,
catching
> an exception, encapsulating the information in an other class and
> re-throwing it. Error.pm supports that... but not always!
>
> Generally speaking, using "plain" calls to die is equivalent to
throwing
> an Error::Simple. Yet, when doing this:
>
> try { die "oops" } otherwise { die "oops2" };
>
> we get:
>
> at /usr/local/libdata/perl5/site_perl/Error.pm line 38.
>
> instead of the expected:
>
> oops2 at -e line 1.
>
> that we get when doing
>
> try { die "oops" } otherwise { throw Error::Simple "oops2" };
>
> Oops. Exception translation works, but only when rethrowing objects.
> That's inconsistent...
>
>
> Some research shows that if the error is correctly detected by
Error.pm,
> it's mysteriously lost when it comes to rethrowing it. After some
more
> work, it appears that the following snippet does not do what one
might
> think:
>
> eval { local $@; die "oops" }; print "no error\n" unless $@;
>
> This was probably unforeseen at the time of writing of version
0.17003,
> which introduced this bug with the intend of making the exception
> available in $@ in catch/otherwise clauses. Of course the alteration
of
> this variable was to be as local as possible and resulted in a local
> declaration in the eval block.
>
> But since actually we still want to have the modified version of $@
> past the eval, the fix is simply to move the local declaration out
of
> this scope. It will not affect the program logic, as the success of
the
> eval is determined using its return value ($ok), not $@.
>
>
> Attached is a patch for the catch and otherwise code, with a
testcase
> that illustrates the problem.
>
>
> Best regards,
> -Thomas Equeter.
|
