On Sun, Oct 26, 2008 at 2:00 AM, via RT Will Coleda <[EMAIL PROTECTED]> wrote: > # New Ticket Created by Will Coleda > # Please include the string: [perl #60128] > # in the subject line of all future correspondence about this issue. > # <URL: http://rt.perl.org/rt3/Ticket/Display.html?id=60128 > > > > Trying to track down a segfault that is occurring for me in parrot > (triggered by partcl): > > After getting partcl > (http://code.google.com/p/partcl/wiki/PartclSource)[0], and then the > spectests (http://code.google.com/p/partcl/wiki/TestingPartcl), I can > trigger it with: > > ../../parrot tcl.pbc t_tcl/namespace-old.test > > I've tried stripping down the amount of tcl[1] needed to generate > that, but it's hard to keep the segfault from vanishing. > > I did find that running with -G makes it go away; > > Using the techniques here: > http://www.oreillynet.com/onlamp/blog/2007/10/how_to_debug_a_gc_problem_in_p.html, > I see the segfault here: > > Program received signal SIGSEGV, Segmentation fault. > [Switching to Thread 0xb6b428d0 (LWP 945)] > 0xb7d20ff8 in ascii_compute_hash (interp_unused=0x804f040, > source_string=0xb5489c30, seed=3793) at src/charset/ascii.c:769 > 769 hashval += *buffptr++; > > So the problem isn't with a PMC, it's with a STRING. So, instead of > setting a conditional breakpoint on pmc_new, I do so on > new_string_header; but this is glacially slow. I get no output in the > time in normally takes the whole test to segfault. > > Anyone have better ideas on how to track this segfault down? I'm going > to leave the conditional breakpoint in gdb running overnight on > feather, but don't hold out hope. > > Regards. > > [0] I'm doing this against Revision: 32150 in trunk, though. > [1] 849 lines for the test file; 3384 lines for the test harness. And > that's on top of all the PIR required to run the tcl in the first > place. > -- > Will "Coke" Coleda >
Turns out setting the conditional breakpoint never fired; left this running overnight, and it eventually came back with the segfault directly. Back to the drawing board. -- Will "Coke" Coleda
