On Mon, 2005-08-15 at 06:12 -0700, Dan Kegel wrote: > Peter Zijlstra <[EMAIL PROTECTED]> wrote: > > On this controversial subject, could somebody please - pretty please > > with a cherry on top - tell me what the current status is: > > - in general, > > - as implemented in the 3.4 series and > > - as implemented in the 4.0 series. > > > > At work we're using 3.4 and we have managed to shoot our foot of with > > this issue :-(, google gives a lot of hits on the issue but it is a bit > > hard to get the current impl. status for 3.4. Which in turn makes it > > hard to decide on how to bandage our foot. > > Could you provide a link to a description of the particular > problem? I looked around, and all I could find was > https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=111548 > > I suppose the controversial part is that you're using > pthread_cancel, which is somewhat frowned upon as > inherently unsafe. > - Dan >
Yes, is seems to be that problem. Discussion here: http://gcc.gnu.org/ml/gcc/2003-12/msg00743.html And this list seems dedicated to the problem: http://www.codesourcery.com/archives/c++-pthreads/maillist.html The issue seems to be that pthread_cancel is implemented using force_unwind, the same mechanism as used for exception handling. And the interaction is ill defined. The behaviour of gcc-3.4 is that the unclassified exception caused by SIGCANCEL can be caught by the catch-all clause: 'catch (...)'. And then when not rethrown causes an abort. Peter Zijlstra
