in struct _machdep_state from libpthread.
The offset is hardcoded in boehm-gc because it's a private struct so we
cannot use offsetof(). This will probably not change soon again but I am
writing this as a note for people who will run into the same problem in
the future.
All the regression tests are passing now and mono builds fine too.
Thanks to kurt@ for pointers.
The latter contains libatomic_ops-1.2, which is developed inside boehm-gc,
as stated on libatomic_ops homepage. No change on -main subpackage.
libatomic_ops may be used by a future eventual pulseaudio port, which is
being worked on.
Feedback, tweaks and ok kurt@
and adjust boehm-gc to deal with any number of holes that may appear
between __data_start and _end. (requires -current binutils)
- work-around libpthread bugs that prevent sigsuspend() from working
in a signal handler by using pthread_suspend_np() + an arch/libpthread
specific offset to obtain the suspended thread's stack pointer.
thanks to sthen@ and jasper@ for access to other arches. discussed
with todd@
Most work done by kurt@, who got bleeding eyes and has been trapped
into pthread issues because some of us (me and Stuart Henderson)
were slopyy and applied some patches only partial. A big sorry to
kurt@ for this waste of time!
Tested by many. Intermediate ok's (IIRC) from kurt@, naddy@ and todd@.
This should have been done a while ago.
Fixing this is beyond me, and until this works, w3m and gcj are pointless as
well (opencm will be free of boehm-gc soon enough).
ELF-based archs as having an old boehm-gc can cause programs to stop working
after the WorX addition. Noticed after trying to build w3m and it failed
miserably.
