Consider the following two backtraces: Backtrace 0: #0 A crash function thread0: #0 A #1 B #2 C #3 D
thread1: #0 A #1 E #2 F #3 G Backtrace 1: #0 A crash function thread0: #0 A #1 E #2 F #3 G thread1: #0 A #1 B #2 C #3 D The crash thread in Backtrace 0 is thread 0 and the crash thread in Bactrace 1 is thread 0 as well. But the crash threads are not the same. IMHO the algorithm must return same threads. What do you think about this? Regards Jakub On Friday, March 22, 2013 07:54:52 PM Martin Milata wrote: > The current algorithm cannot handle backtraces with multiple threads > containing the designated crash frame such as [1]. We just pick the > shortest one with the risk that our guess might be wrong. Fixes #3. > > [1] https://bugzilla.redhat.com/attachment.cgi?id=703269 > > Signed-off-by: Martin Milata <mmil...@redhat.com> > --- > lib/backtrace.c | 45 ++++++++++++++++++++++++++++++++++----------- > 1 file changed, 34 insertions(+), 11 deletions(-) > > diff --git a/lib/backtrace.c b/lib/backtrace.c > index a871eee..11050f8 100644 > --- a/lib/backtrace.c > +++ b/lib/backtrace.c > @@ -117,27 +117,31 @@ btp_backtrace_remove_threads_except_one(struct > btp_backtrace *backtrace, backtrace->threads = thread; > } > > +enum requirement { ONE_MATCHING, ABORT_FUNCTION, SHORTEST_MATCHING }; > /** > * Loop through all threads and if a single one contains the crash > * frame on the top, return it. Otherwise, return NULL. > * > - * If require_abort is true, it is also required that the thread > - * containing the crash frame contains some known "abort" function. In > - * this case there can be multiple threads with the crash frame on the > - * top, but only one of them might contain the abort function to > - * succeed. > + * Parameter req controlls the criteria for matching thread: > + * ONE_MATCHING - There is only one thread that contains the crash > frame. + * ABORT_FUNCTION - There may be multiple threads containing the > crash frame + * but only one contains some known > "abort" function. + * SHORTEST_MATCHING - If everything else fails, we just > return the shortest + * thread containing the crash > frame. > */ > static struct btp_thread * > btp_backtrace_find_crash_thread_from_crash_frame(struct btp_backtrace > *backtrace, - bool > require_abort) + enum > requirement req) { > if (btp_debug_parser) > - printf("%s(backtrace, %s)\n", __FUNCTION__, require_abort ? "true" > : "false"); + printf("%s(backtrace, %s)\n", __FUNCTION__, > + (req == ABORT_FUNCTION) ? "true" : "false"); > > assert(backtrace->threads); /* checked by the caller */ > if (!backtrace->crash || !backtrace->crash->function_name) > return NULL; > > + int result_frames; > struct btp_thread *result = NULL; > struct btp_thread *thread = backtrace->threads; > while (thread) > @@ -146,7 +150,7 @@ btp_backtrace_find_crash_thread_from_crash_frame(struct > btp_backtrace *backtrace bool same_name = top_frame && > top_frame->function_name && > 0 == strcmp(top_frame->function_name, > backtrace->crash->function_name); - bool abort_requirement_satisfied > = !require_abort || > + bool abort_requirement_satisfied = (req != ABORT_FUNCTION) || > btp_glibc_thread_find_exit_frame(thread); > if (btp_debug_parser) > { > @@ -158,13 +162,24 @@ > btp_backtrace_find_crash_thread_from_crash_frame(struct btp_backtrace > *backtrace > > if (same_name && abort_requirement_satisfied) > { > + int current_frames = btp_thread_get_frame_count(thread); > + > if (NULL == result) > + { > result = thread; > - else > + result_frames = current_frames; > + } > + else if (req != SHORTEST_MATCHING) > { > /* Second frame with the same function. Failure. */ > return NULL; > } > + else if (current_frames < result_frames) > + { > + /* We're just looking for the shortest frame that matches. > */ + result = thread; > + result_frames = current_frames;
