A bit more information: as far as we know there are problems only on Linux: the logs badlog, badlog1, badlog2 and badlog3 are made by one machine (a Xeon box running Ubuntu 18.04) and badlog-match is another machne (an i7 also running Ubuntu 18.04). In all the logs except badlog, there is a segmentation fault. In badlog3, gdb attaches the running process and produces a backtrace. We are currently not seeing crashes on MacOS.
Daniel Bump On Saturday, May 14, 2022 at 8:06:18 PM UTC-7 Travis Scrimshaw wrote: > Hi everyone, > On ticket #30423 <https://trac.sagemath.org/ticket/30423>, Dan, > Willie, and I have been working on a parallel-computation based > implementation for computing F-matrices that are used in math physics. > However, we have been seeing some doctest failures sporadically that > involve segfaults and linked-list corruption from (cy)PARI. Here are the > logs from testing with the first and the last having full tracebacks. > > http://sporadic.stanford.edu/badlog-match > http://sporadic.stanford.edu/badlog > http://sporadic.stanford.edu/badlog1 > http://sporadic.stanford.edu/badlog2 > http://sporadic.stanford.edu/badlog3 > > The first question would be if anyone has an idea about what is causing > this. I have this impression that PARI is thread-safe, but I am wondering > if cypari is also thread/parallel-safe or if there are any specific things > that we should be careful about. (We’ve already had to work around a > pickling issue with polynomials IIRC.) > > Second question is that because this is a Heisenbug and I suspect it is > something upstream (and so far, nobody has been able to reproducing it > during an interactive version of Sage), I was wondering what the policy > would be for merging the ticket. I recall in the past that we have merged > tickets with Heisenbugs with followup tickets noting the behavior, but I am > not 100% sure about that (and I wouldn’t necessarily know how to find any > explicit examples). I was wondering if we could merge the ticket in an > early beta version so that many people/systems can test it to see if it > becomes more reproducible; of course this is assuming that the build bots > are not consistent in reproducing this. Should we just mark any offending > test(s) as “# known bug” and is there some general policy about this? > > Thanks, > Travis > > > -- You received this message because you are subscribed to the Google Groups "sage-devel" group. To unsubscribe from this group and stop receiving emails from it, send an email to sage-devel+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/sage-devel/ed1a0c3a-ed0e-41c3-825a-153b738a7afcn%40googlegroups.com.
