Hilary Holz wrote: > Hi Geoff, > > I found the problem with Apache::TestMB that was causing the unstable > behavior I was seeing with D::C results (actually, I found several related > problems...)
cool. > > I didn't know if this should go to you or to David Wheeler - I thought I'd > run it by you as I haven't done this sort of thing before and you and I had > been talking. I'm CC'ing here the appropriate places. if you could please resend the patch to the list that would be great. we greatly prefer unified diffs directly against svn $ svn diff but if that doesn't work for you $ diff -u ... is better than just plain diff, especially a diff that doesn't list the relevant files :) > > Anyway, there are three distinct changes (annotated as patch 1, patch 2 and > patch 3 in my comments in the patch, appended to this email.) > > - Patch 1 was the easiest to diagnose - without it the cover_db is not > deleted when running ./Build testcover repeatedly (you can just check file > mod times to see.) ok. we do in MakeMaker land, so that's probably an oversight in M::B land. > > - I've attached a minimal test case that shows the effect of patch 3, where > the minimal test case is an extension of the one you have, translated into > Module::Build and expanded a bit to show the problem. Just build and run > ./Build test, and you'll see it doesn't descend, but with the patch it does. if you say so :) I'll assume this is MB::Specific. > > - I haven't been able to construct a minimal test case for patch 2 - the > effects show in the distribution we're developing in the lab, in fact, this > patch is what actually fixes the unstable D::C results. I'm hoping that > isn't a problem, as the patch brings A::TestMB into line with the way A::T > works under MakeMaker, so you might have a minimal test case? The patch > invokes test with APACHE_TEST_EXTRA_ARGS set to -one-process (without the > patch the variable is set but not used.) ok. we added APACHE_TEST_EXTRA_ARGS as a way to add signals specifically through make, so I doubt anyone used or tested it under M::B. separarely, as I mentioned before D::C + mod_perl segfaults for me maybe 50% of the time, but I always write that off to the unusual stuff I'm trying to cover and just how twisted both tools happen to be. > > Of course, none of this fixes the problem with the select/ignores not being > processed :-/ as you suspected. Any thoughts on that one? I'll keep poking > away at it in the background, but I haven't had any luck so far. well, I'm not really sure about them being ignored. but as I said before, I'd like to see if using MakeMaker makes a difference to you, since MM behaves just fine for me. if there is a difference between MM and M::B for you then at least we know to look at gaps, whereas if MM misbehaves that's a different issue. --Geoff
