perl-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Philippe M. Chiasson" <go...@ectoplasm.org>
Subject Re: [PATCH] prototypes/constant redefinition warnings from ModPerl::Util
Date Thu, 08 Jan 2009 05:43:29 GMT
On 6/11/08 15:39, Casey West wrote:
> Hi,
>
> Apache2::Reload relies on ModPerl::Util::unload_package_pp to
> disassemble a stash entry before importing the code again. I like how
> it does this with two exceptions.

Thanks.

> First, prototypes. unload_package_pp() takes care to retain the
> prototype when redefining a code stash entry. That's a good thing
> because it (typically) avoids a warning. Then it removes the
> subroutine from the stash using undef(). That's not so hot because
> undef takes the liberty to explicitly undefine the prototype of that
> subroutine. In the case of Apache2::Reload, what happens next is we
> remake a prototyped subroutine - lets use try() from Error.pm for
> illustration - and get warnings about mismatched prototypes. perl
> believes the prototype for this subroutine is not defined, Error tries
> to define it as (&;$), and perl warns about "none vs. (&;$)" prototype
> mismatch.
>
> This can be avoided by explicitly deleting the CODE slot in the stash
> entry, like this: "delete ${$fullname}{CODE};" This will remove the
> body of the coderef but leave the prototype in tact.

Works great for me, but I'd want to make sure this works/is safe back
to the oldest perl we support. In the worse case, that logic could be
made conditionnal on Perl version.

Seems this also fixes David's problem
http://marc.info/?l=apache-modperl&m=123034201530079

> Second, constant redefinitions. In this case the assignment of an
> empty subroutine (with proper prototype) throws a warning, only in the
> case of constant subroutines. Take something created by "use constant"
> as an example, and combine that with Apache2::Reload.
> unload_package_pp() makes every attempt to turn off warnings but
> can't, for whatever reason, make it work for this one. The reason, I
> found through copious trial and error and a pairing session with Adam
> Foxson, is to turn off the warning directly in the eval, like this:
>
>    *{$fullname} = eval "no warmings 'redefine'; sub ($p) {}";

Cool, that was one of the warnings I couldn't figure out how to silence
in pure-perl.

> I am not sure why the redefinition warning is thrown from this lexical
> scope. Nevertheless, a solution is found.
>
> Attached is a patch which accounts for these two problems. Please consider it.

Again, if this can be verified/adjusted for older Perls, +1

For bonus points, include a test case!

-- 
Philippe M. Chiasson     GPG: F9BFE0C2480E7680 1AE53631CB32A107 88C3A5A5
http://gozer.ectoplasm.org/       m/gozer\@(apache|cpan|ectoplasm)\.org/

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@perl.apache.org
For additional commands, e-mail: dev-help@perl.apache.org


Mime
View raw message