Joshua Isom (via RT) wrote:
My current svn repository uses a patch that I sent to the list about a week ago, in which the pge tests would run with gc on if the file DEVELOPING existed. Since I updated to over 18323, t/compilers/pge/p5regex/p5rx.t segfaults after test 553. The current_cont in the interp's ctx is null, and results in a segfault inside of set_returns. It doesn't happen on my darwin-ppc system, so I don't know how specific it is, and if rebooting will fix it(as often happens with the gc bugs in my experience). But it is failing on my freebsd-amd64 system. Since odds are r18323 just exposed a bug instead of adding one, maybe other people will get the crash as well. You will have to remove -G in t/compilers/pge/p5regex.t/p5rx.t to get it to fail, as with -G it succeeds(and with -t too).
I can't duplicate the error on darwin-intel or ubuntu-x86. Could be specific to FreeBSD (which might help us pin it down further).
I've learned a lot lately about how to use gdb, and all I've come to learn is that if interp->ctx.state->current_cont->obj.u._ptrs._struct_val is null, parrot isn't happy, of course that's with macro expansion, and that getting the level of debugging information I'd like to be able to write little tools to process an analyze it better isn't easy.
Could you send us a stack trace? (gdb command bt) Also, if you haven't already, try a make realclean (we've made some significant fundamental changes recently).
Thanks! Allison
