Re: Test::Aggregate results at the BBC
On Jan 3, 2008 10:20 AM, Ovid [EMAIL PROTECTED] wrote: We couldn't reproduce the segfaults in the debugger or with Devel::Trace, but after a lot of work, a colleague and I found the culprit: Contextual::Return. Like Sub::Uplevel, it overrides CORE::GLOBAL::caller. Apparently the two don't play well together, but it's very difficult to reproduce a minimal test case. Merely by removing Contextual::Return from one module, the segfault went away. We don't know why. Maybe it's too late, but which versions of Sub::Uplevel and Contextual::Return were you using? I attempted to fix Sub::Uplevel as of version 0.16 to play nicely with Contextual::Return -- but as you said, finding a minimal test case is difficult. It's also a potential issue for Hook::LexWrap, so if I can find a permanent fix, that would be a good thing. David
Test::Aggregate results at the BBC
My test aggregation took a bit longer to implement than Adrian's due to several issues. Some were bugs in tests, some were in code, but one issue which took a *long* time to debug was random segfaults. We couldn't reproduce the segfaults in the debugger or with Devel::Trace, but after a lot of work, a colleague and I found the culprit: Contextual::Return. Like Sub::Uplevel, it overrides CORE::GLOBAL::caller. Apparently the two don't play well together, but it's very difficult to reproduce a minimal test case. Merely by removing Contextual::Return from one module, the segfault went away. We don't know why. The aggregated tests, when run through t/aggregate.t, take 2m2s. When run normally, they take 5m45s. Our test suite, which two months ago took 80m, is now 10m46s. Cheers, Ovid -- Buy the book - http://www.oreilly.com/catalog/perlhks/ Perl and CGI - http://users.easystreet.com/ovid/cgi_course/ Personal blog - http://publius-ovidius.livejournal.com/ Tech blog - http://use.perl.org/~Ovid/journal/
Re: Test::Aggregate results at the BBC
On Jan 3, 2008 11:09 AM, Ovid [EMAIL PROTECTED] wrote: For smaller projects, you have fewer interactions. For large projects, globally altered behavior is playing russian roulette without counting the bullets. At least Sub::Uplevel documentation makes a point of warning in several places about the dangers using it and how it overrides CORE::GLOBAL::caller. Contextual::Return is quite a bit more vague about it. The other general area of concern I would have for global vs local behavior alteration is attributes. E.g. combining custom attributes handling with anything that is based on Attribute::Handlers. David
Re: Test::Aggregate results at the BBC
On Thu, Jan 03, 2008 at 07:20:38AM -0800, Ovid wrote: We couldn't reproduce the segfaults in the debugger or with Devel::Trace, but after a lot of work, a colleague and I found the culprit: Contextual::Return. Like Sub::Uplevel, it overrides CORE::GLOBAL::caller. That probably explains the segfaults when using Hook::LexWrap in some circumstances as well. A few years ago I could reliably make perl segfault by calling a wrapped function in void context - that is, just throwing the result away. At the time, the fix was: my $rubbish = wrapped_subroutine(...); Unfortunately I forget what version of 5.8.x this was. It might be in the p5p archives. -- David Cantrell | top google result for internet beard fetish club For every vengeance, there is an equal and opposite revengeance. -- Cartoon Law X
