Martin v. Löwis wrote: > Thomas Heller wrote: >> Why does the ia64 debian buildbot now complain about unaligned accesses, >> and how can we find out where they occur? > > I don't know why they started show up suddenly; on Debian-Itanium, it > is a configuration option (even per process, through prctl(1)) whether > they just produce a log message, or a signal, or nothing; if they > produce a log message, this also goes to the process' terminal. > > These unaligned access must have been there for some time now; > perhaps something on the machine has changed. Matthias? > > Finding where they originate from is really hard. You need to > set the process into the "signal unaligned access" mode, and > then run it under a debugger, so you know where it crashes. > This requires shell access to the machine. > > I have tried to reproduce it on my Itanium machine, and found that > that they come from the AST compiler's arena (I /knew/ it was > wrong to implement your own memory management algorithms :-): it > was returning memory blocks that were only 4-aligned, so the > the pointers in the ASDL sequences were all unaligned; hence the > errors. > > I changed that to always guarantee 8-alignment for all blocks > returned from the arena; that seems to help.
Cool - thanks. I'm anxiously waiting until the buildbot runs the ctypes-test ;-). Thomas _______________________________________________ Python-Dev mailing list Python-Dev@python.org http://mail.python.org/mailman/listinfo/python-dev Unsubscribe: http://mail.python.org/mailman/options/python-dev/archive%40mail-archive.com
