On Mon, Oct 29, 2007 at 10:14:33PM -0400, James E Keenan wrote: > James E Keenan wrote: > >James E Keenan wrote: > >>James E Keenan wrote: > >>> > >>>t/configure/029-option_or_data....................dubious > >>> Test returned status 0 (wstat 11, 0xb) > >>> after all the subtests completed successfully > >>> > > > > > I'm still in much the same situation as I was last month. I'll run > coverage analysis on Parrot's configuration and build tools for a week > or more with no problem. Then, out of the blue, a file in t/configure/ > will generate the error message above -- notwithstanding the fact that > all its Test::More function calls have been passing for months. > > I've quieted many instances of this by undef-ing objects, making sure > that ties to STDOUT and STDERR are explicitly untied, and even inserting > one additional Test::More::pass() test before the end of the file. When > all else fails I resort to magic incantations. Sometimes the problem > clears up by itself after a lapse of time. > > Any thoughts?
I really wish I did. I've kept an eye on this thread (monologue) in the hopes of getting some clues, because I'm not at all sure what is going on. I suspect it is a problem with Devel::Cover. Over the years I have learnt to always suspect my own code, but not being able to reproduce the problem makes it reasonably tricky to narrow it down. I'm particularly annoyed that you've had to work around the problem. If I were attacking this, which I'm afraid I'm unable to do right now, I would grab the latest bleadperl, build it with debugging options, build all the required modules, and run the whole thing through valgraind. If that was clean I'd then start doing the same with 5.8.8 (or whatever you are using) and perhaps maintperl. That's quite a lot of work, so perhaps a gentler start might be to just rebuild Devel::Cover with debugging on and run it through valgrind. Or even just run what you have right now through valgrind. (Can you tell that I'm suspecting some memory corruption here?) Sorry that I have no better suggestions. I was hoping that someone more familiar with the testing infrastructure might recognise the problem, but no such luck. -- Paul Johnson - [EMAIL PROTECTED] http://www.pjcj.net
