Chunrong has raised HARMONY-6132 for this failure. I have reproduced the failure on my SuSE 10.2 x86_64.
I tried to simply set SIGPIPE to SIG_ING in drlvm's initialize_signals(), but the test still crashes. I'll check if this code (in vm/port/src/signals/linux) is alive, and try to investigate further. -- Ilya. 2009/5/18 Mark Hindess <mark.hind...@googlemail.com> > > In message <f9761b910903221942s4fe244f1y3297b63938ba6...@mail.gmail.com>, > chunrong lai writes: > > > > I also once investigated the issue times ago > > http://mail-archives.apache.org/mod_mbox/harmony-dev/200809.mbox/%3cf9761b910809160814s7fd8e257gd99b4a15d9700...@mail.gmail.com%3e > > and identified the guity commit which breaks SSLSocketImplTest as > > r691267. > > > > What r691267 did is "Remove the hy.no.sig option and make it the > > default." so it is basically a classlib patch. It seems that its > > modifications to vm makefiles finally make the test crashed. > > Was a JIRA raised for this? I don't see one? > > I had a look at this. The crash is caused by a SIGPIPE which isn't > being ignored. The classlib signal handler obviously configured this > signal to be ignored so that the issue could be dealt as an error from > an affected syscall. This seems reasonable behaviour to me. > > Since there was general consensus that having classlib deal with signal > handling [0] and the classlib signal handler has been removed [1], then > I think the drlvm signal handling code should be fixed to also configure > SIGPIPE to be ignored (SIG_IGN). > > Regards, > Mark. > > [0] http://markmail.org/thread/6suvaaeqal5occsl > [1] http://markmail.org/thread/n6vy37nl6andz2vx >
