On 5/23/2012 10:18 AM, Corinna Vinschen wrote: > On May 23 18:06, Corinna Vinschen wrote: >> On May 23 10:52, Corinna Vinschen wrote: >>> On May 22 15:25, Otto Meta wrote: >>>>>> Testcase cancel deferred: >>>>>> Works with 1.7.9 and 20120517 snapshot, fails (hangs) with 1.7.12-1 >>>>>> and 1.7.15-1. >>>>> If that works in the snapshot anyway, I'm not going to look into that >>>>> one. >>>> >>>> It worked in the reduced testcase with sem_wait(). With read() it’s >>>> still half-broken. See below. >>>> >>>>>> Testcase cancel asynchronous: >>>>>> Async cancel seems to have no effect with any tested version. >>>>> I think I found a solution for this problem. See the comment in the >>>>> patch at >>>>> http://sourceware.org/cgi-bin/cvsweb.cgi/src/winsup/cygwin/thread.cc.diff?cvsroot=src&r1=1.258&r2=1.259 >>>>> Please test the today's developer snapshot. >>>> >>>> Asynchronous cancel seems to work as well as deferred cancel now. Thanks. >>>> >>>> Both cancel types work with sem_wait() and pause() now, but for threads >>>> blocked in read() they’re still unreliable. Only one of three blocked >>>> threads is killed in the attached updated testcases. >>> >>> Just to let you know I'm working on it. This will take some time. >>> During debugging I stumbled over a heisenbug. As soon as I remove the >>> "Thread %i woke up just fine" printf from your test application, the >>> read function reurns with a "Bad address" error. So far I have onlya >>> vague hunch what the cause may be. Stay tuned. >> >> Half of the solution is now checked in to CVS. The other half is a bad >> hack to workaround a shortcoming in newlib's stdio functions. I'd rather >> have a solution in newlib. see my request on the newlib ML: >> http://cygwin.com/ml/newlib/2012/msg00188.html > > Ok, for the time being I checked in my workaround. Please test the > today's developer snapshot.
I tried installing this snapshot and found most things hung. Specifically, I ran ash in a Windows cmd window, then tried /bin/echo foo I tried mintty too but bash hangs before I get a prompt. "echo foo" worked fine, though. -- David Rothenberger ---- daver...@acm.org Glib's Fourth Law of Unreliability: Investment in reliability will increase until it exceeds the probable cost of errors, or until someone insists on getting some useful work done. -- Problem reports: http://cygwin.com/problems.html FAQ: http://cygwin.com/faq/ Documentation: http://cygwin.com/docs.html Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple