Bug #63013 for Class-Load: False positive for is_class_loaded despite module death

On Mon Nov 15 19:52:32 2010, KENTNL wrote: Show quoted text
I just looked into this, and I'm not sure it's entirely fixable. In the case where module causes a syntax error, everything works as expected. I assume this happens because the syntax error happens at compile time, before @ISA is set at run time. If @ISA is set before the module dies (either in a BEGIN block or because of a runtime die) there's really no way to detect that. The only way I can think of to fix this is to maintain a black list of failed modules, but that makes everything more complex, because then we'd have to account for the module being changed on disk. Ugh.

On Mon Sep 12 21:59:47 2011, DROLSKY wrote: Show quoted text
What about doing a Class::Unload pass if require dies? I can hardly imagine someone *relying* on being abl to find the parts of the symtable populated before require had a chance to die.

So any word on the Class::Unload idea?

On Tue, Jul 09, 2013 at 11:15:59AM -0400, Dave Rolsky via RT wrote: Show quoted text
Perhaps this is enough reason to give it a try? We could do a devrel and kindly ask Andreas to run a cpan smoke with it (it takes about a week given his scheduling). Additionally this is low-level-enough dependency, and the problem scope is rather limited - thus I feel if there *are* issues with this they will pop out rather quickly. As usual I have no problem delivering an actual patch, but that's a minor detail relative to the design decision. Thoughts?

On 2011-09-12 18:59:47, DROLSKY wrote: Show quoted text
23:21 < ether> I just got bitten by Class::Load::load_class 23:22 < ether> tried to load Foo inside an eval; Foo died. tried to load Foo again later, also in an eval, but load_class('Foo') returned true this time, so I went on to do something with the class and exploded 23:22 < ether> $INC{'Foo.pm'} existed, but was undef 23:23 < ether> it's not htat helpful if load_class('Foo') only dies the first time but pretends to work thereafter 23:28 < ether> this looks to be because Foo managed to define some symbols first before it croaked.. and _is_class_loaded takes a non-empty stash to mean that it's loaded 23:31 < ether> ...and then, later on when I try "do stuff" with the class, Module::Runtime::use_module is called on the class - and it correctly sees that the module isn't loaded, tries to load it, and so require() blows up with "Attempt to reload %s aborted." 23:32 < ether> and the perldiag explanation exactly sums up what happened - $INC{'Foo.pm'} existed but was empty, signalling a failed load attempt 23:32 < ether> so clearly _is_class_loaded ought to be doing the same 23:33 * ether sees kentnl already filed this - #63013 To summarize: this is the same problem as kentnl, but it is indeed fixable, because the %INC entry exists *but is not defined* (at least it is in my case). This is something we can actually check for.