[adding bug-gnulib] According to Johan van Selst on 3/1/2010 2:55 PM: > Eric Blake wrote: >> According to Johan van Selst on 3/1/2010 11:02 AM: >>> Sure, it's now up on http://mud.stack.nl/~johans/193-ktrace.txt >>> (output from the test in checks/193.esyscmd) >> Actually, it looks like you posted the trace for 007.command_li > > Oops, I was rather hurried when posting this: it was 007.command_li > indeed. There is now a new http://mud.stack.nl/~johans/193-ktrace.txt > and http://mud.stack.nl/~johans/007-ktrace.txt (both with child tracing) > for reference. > > This quickly identifies a problem: > > 2143 gm4 CALL sigprocmask(SIG_BLOCK,0x545a20,0) > 2143 gm4 RET sigprocmask 0 > 2143 gm4 CALL vfork > 2144 gm4 RET fork 0 > 2144 gm4 CALL sched_setscheduler(0,<invalid=0>,0x800a0e918) > 2144 gm4 RET sched_setscheduler -1 errno 1 Operation not permitted > 2144 gm4 CALL exit(0x7f) > 2143 gm4 RET vfork 2144/0x860 > 2143 gm4 CALL sigprocmask(SIG_UNBLOCK,0x545a20,0) > 2143 gm4 RET sigprocmask 0 > > The issue seems to be with the seconds argument of sched_setscheduler() > Quoting the FreeBSD <sched.h> there is no argument defined as 0 - > Ref. http://cvsweb.freebsd.org/src/sys/sys/sched.h > > /* POSIX 1003.1b Process Scheduling */ > > /* > * POSIX scheduling policies > */ > #define SCHED_FIFO 1 > #define SCHED_OTHER 2 > #define SCHED_RR 3 > > struct sched_param { > int sched_priority; > }; > > > On FreeBSD the default seems to be SCHED_OTHER (2); whereas on Linux > this SCHED_OTHER is defined as 0. It may be that some code relies on > this value being 0 and doesn't explicitly set it as SCHED_OTHER. > But I haven't really analysed this code and should probably stop > guessing and leave debugging to those more familiar with m4.
Thanks for the additional research. Either Bruno's gnulib posix_spawnp replacement (and thus the pipe module) triggers a FreeBSD failure, or you have found a bug in FreeBSD posix_spawnp that we need to work around. Johan, can you find the portion in config.log that shows whether the native posix_spawn was deemed acceptable or buggy? The gnulib version of posix_spawnp only calls sched_setscheduler if something called posix_spawnattr_setsched*, but I don't see that happening in the m4/gnulib code base. It may be possible to work around the failure, by priming the configure cache. For example, if the bug is that your posix_spawnp is broken, but gnulib is not properly detecting that brokenness, then you could use './configure ac_cv_func_posix_spawnp=no' to force the gnulib replacement to be compiled in. -- Don't work too hard, make some time for fun as well! Eric Blake e...@byu.net
signature.asc
Description: OpenPGP digital signature
