November 2009 - libdispatch-dev - lists.macosforge.org

Solaris portability, POSIX memalign checking in configure.ac - PATCH [1/6]
by Joakim Johansson 06 Nov '09

06 Nov '09

Need to check whether posix_memalign exists on the target system (there is a patch for a future Solaris version to add it, so I'd rather not check for the Solaris platform, but rather for feature existence). Still a lot of work remains for a working Solaris port due to the kqueue stuff, but here are some initial patches. Joakim ----------------------------------------------- > svn diff configure.ac Index: configure.ac =================================================================== --- configure.ac (revision 140) +++ configure.ac (working copy) @@ -184,7 +198,7 @@ AC_CHECK_DECLS([FD_COPY], [], [], [[#include <sys/select.h>]]) AC_CHECK_DECLS([SIGEMT], [], [], [[#include <signal.h>]]) AC_CHECK_DECLS([VQ_UPDATE, VQ_VERYLOWDISK], [], [], [[#include <sys/ mount.h>]]) -AC_CHECK_FUNCS([pthread_key_init_np pthread_main_np mach_absolute_time malloc_create_zone]) +AC_CHECK_FUNCS([pthread_key_init_np pthread_main_np mach_absolute_time malloc_create_zone sem_init posix_memalign]) AC_CHECK_DECLS([POSIX_SPAWN_START_SUSPENDED], [have_posix_spawn_start_suspended=true],

3 2

Solaris portability, dispatch.h - PATCH [5/6]
by Joakim Johansson 06 Nov '09

06 Nov '09

Checking for sys/cdefs.h again... Joakim > svn diff private.h Index: private.h =================================================================== --- private.h (revision 140) +++ private.h (working copy) @@ -33,7 +33,9 @@ #include <mach/message.h> #endif #include <unistd.h> +#if HAVE_SYS_CDEFS_H #include <sys/cdefs.h> +#endif #include <pthread.h> #ifndef __DISPATCH_BUILDING_DISPATCH__

1 0

Solaris portability, POSIX memalign() missing - PATCH [2/6]
by Joakim Johansson 06 Nov '09

06 Nov '09

Use memalign() instead... Joakim > svn diff malloc_zone.h Index: malloc_zone.h =================================================================== --- malloc_zone.h (revision 140) +++ malloc_zone.h (working copy) @@ -80,14 +80,28 @@ return (realloc(ptr, size)); } +/* posix_memalign is not supported with the current shipping release of Solaris 10, it is + fixed in the latest available Solaris Nevada (next-gen) distribution though, see: + + http://bugs.opensolaris.org/bugdatabase/view_bug.do;jsessionid=3af20a8d00f3… + + Falling back at using memalign() here as is done by Sun in the implementation at: + http://src.opensolaris.org/source/xref/onnv/onnv-gate/usr/src/lib/libc/port… + */ + static inline void * malloc_zone_memalign(malloc_zone_t *zone, size_t alignment, size_t size) { +#ifdef HAS_POSIX_MEMALIGN void *ptr; if (posix_memalign(&ptr, alignment, size) < 0) return (NULL); + return (ptr); +#else + return memalign(alignment, size); +#endif } static inline void octo.tbricks.com:trunk/src/shims>

1 0

minor libkqueue fix
by Mario Schwalbe 06 Nov '09

06 Nov '09

Hi Mark, while working with libdispatch and libkqueue on Linux, I discovered some minor issues. Could you please commit the attached patch for those who like to: (a) reconfigure libkqueue sometimes (b) do not have test.o around (c) install to a non-standard prefix thanks & ciao, Mario Schwalbe diff -urp foo/src/libkqueue/configure diplom/src/libkqueue/configure --- foo/src/libkqueue/configure 2009-11-05 13:43:37.695235346 +0100 +++ diplom/src/libkqueue/configure 2009-11-03 18:12:21.581371724 +0100 @@ -65,15 +65,15 @@ finalize mandir "${prefix}/share/man echo "Checking operating system type... $target" rm -f socket.c vnode.c signal.c timer.c user.c if [ $target = "linux" ] ; then - check_header sys/epoll.h && ln -s os/linux/socket.c \ - || ln -s os/posix/socket.c - check_header sys/inotify.h && ln -s os/linux/vnode.c \ - || ln -s os/posix/vnode.c - check_header sys/signalfd.h && ln -s os/linux/signal.c \ - || ln -s os/posix/signal.c - check_header sys/timerfd.h && ln -s os/linux/timer.c \ - || ln -s os/posix/timer.c - ln -s os/posix/user.c + check_header sys/epoll.h && ln -sf os/linux/socket.c \ + || ln -sf os/posix/socket.c + check_header sys/inotify.h && ln -sf os/linux/vnode.c \ + || ln -sf os/posix/vnode.c + check_header sys/signalfd.h && ln -sf os/linux/signal.c \ + || ln -sf os/posix/signal.c + check_header sys/timerfd.h && ln -sf os/linux/timer.c \ + || ln -sf os/posix/timer.c + ln -sf os/posix/user.c fi echo "Creating config.h" diff -urp foo/src/libkqueue/Makefile diplom/src/libkqueue/Makefile --- foo/src/libkqueue/Makefile 2009-11-05 13:43:37.705236408 +0100 +++ diplom/src/libkqueue/Makefile 2009-11-05 13:23:21.312802794 +0100 @@ -24,13 +24,14 @@ include config.mk build: gcc $(CFLAGS) -c *.c - rm test.o + -rm test.o ar rcs libkqueue.a *.o gcc -shared -Wl,-soname,libkqueue.so -o libkqueue.so *.o install: $(INSTALL) -d -m 755 $(INCLUDEDIR)/kqueue $(INSTALL) -d -m 755 $(INCLUDEDIR)/kqueue/sys + $(INSTALL) -d -m 755 $(LIBDIR)/pkgconfig $(INSTALL) -d -m 755 $(MANDIR)/man2 $(INSTALL) -m 644 sys/event.h $(INCLUDEDIR)/kqueue/sys/event.h $(INSTALL) -m 644 libkqueue.so $(LIBDIR)/libkqueue.so

2 2

libkqueue updates
by Mark Heily 06 Nov '09

06 Nov '09

A couple of quick updates about libkqueue: * Version 0.1 has been released. * Most of the problems with EVFILT_VNODE have been solved. * There is now a "stable" Subversion branch that will only contain release-quality code. * There are now two libkqueue mailing lists: one for announcements, and one for general discussion. For more details, see the libkqueue web page at: http://mark.heily.com/libkqueue/ Cheers, - Mark

1 0

testing tree now ported/automaked and committed
by Robert Watson 05 Nov '09

05 Nov '09

Dear all: Yesterday I committed an adaptation of the testing tree to the autoconf/automake build system. Most tests now run on FreeBSD, and often report success. Here's a quick summary: OK: dispatch_api dispatch_apply dispatch_c99 * But that is not the point of this test dispatch_debug dispatch_group dispatch_pingpong dispatch_read dispatch_sema dispatch_timer_bit31 Mixed success: dispatch_after * Occasionally one or two fail dispatch_drift * Jitter fails, drift passes dispatch_starfish * Latency requirements not always met Pass, but no 'leaks' testing on non-Mac OS X: dispatch_cascade dispatch_readsync dispatch_timer_bit63 Failure: dispatch_priority * Blocks complete but fails precedence test Doesn't yet run on FreeBSD: dispatch_cffd * No Core Foundation on FreeBSD dispatch_cpp * No working C++ compiler with blocks off Mac OS X dispatch_timer_set_time * Requires legacy API I've left the original Darwin Makefile as Makefile.darwin, as it contains better runtime packaging and summarization, as well as knowing about lipo on Mac OS X. Some of this packaging should make it into the portable makefiles in the future. I plan to spend time investigating dispatch_priority today. Poor latency/jitter in some tests may be a result of my using a VM for testing, which I'll also investigate, but could also be a property of timer resolution, an off-by-one error in how wakeups are calculated (perhaps by the POSIX semaphore/kqueue code on FreeBSD?) but we'll see. I also ommitted some fixes to the POSIX semaphore code in libdispatch; I had some of the timeout calculation and handling wrong. One important semantic difference between Mach and POSIX semaphores is that their timed wait primitives take different time scales: Mach accepts a relative timeout, and POSIX and absolute timeout. There are arguments for/against both approaches, but I was pondering adding a non-portable POSIX semaphore extension to FreeBSD supporting relative timeouts to avoid additional clock reads to convert between the two bases. Robert N M Watson Computer Laboratory University of Cambridge

3 4

autoconf HAVE macro conventions
by Kevin Van Vechten 02 Nov '09

02 Nov '09

I'm not very familiar with autoconf conventions and was wondering why some config.h values are merely defined and some are explicitly defined to 1. Is there a strong convention to do #ifdef HAVE_FOO instead of #if HAVE_FOO when testing for configure features? For various reasons, the ability to define a macro to 0 or 1 would be useful for the Mac OS X version of config.h — it would let us define certain features in terms of other features without as much conditional logic. Kevin

4 11

[PATCH 00/17] portability + chainsaw
by Paolo Bonzini 02 Nov '09

02 Nov '09

Here is my attempt at improving libdispatch's portability. Of course it's still a kevent-only show, but it should be easier now to implement an epoll-based (or even SIGIO+SIGALRM+poll-based) replacement. At least it makes more clear where the boundaries are between the kevent code and the rest. Because of the very precise requirements of libdispatch on kevent's semantics, it may be easier to go this way rather than implement kevent as a wrapper. I have no answer, and libnqkqueue is very interesting, but anyway these patches are quite safe. Here are them: - add .gitignore files - move __private_extern__ detection to an autoconf macro - move atomic builtin detections to separate autoconf macro - move blocks detection to separate autoconf macro - more precise detection of blocks Just configury. - split os_shims.h into separate files - remove compat/fd_copy.h - finish moving stuff into src/shims Unifying a bit the terminology, see individual comments. - portability After which the non-kqueue parts compile on Linux. - consolidate hashing of dispatch_kevent_t - move mach2nano/nano2mach from source.c to shims/ Two small cleanups. - extract source-type-specific initialization to per-type functions More completely developed later in the series. - abstract some mentions of ds->ds_dkev Also preparing for later. - extract kevent-specific parts of queue.c - almost completely separate kevent/source.c Moving code to new files without any change. You can check this with a command like diff -b <(git diff HEAD^^^! src/source.c | sed -n s/^-//p) <(git diff HEAD^^^! src/source_kevent.c | sed -n s/^+//p) - complete separation - remove last mention of struct kevent from "generic" code These finish the cleanup. Note: this was tested on FreeBSD only. HTH, Paolo

9 64

clang bugfix corrects libdispatch crashes on FreeBSD/amd64
by Robert Watson 01 Nov '09

01 Nov '09

Just an FYI for those who have been trying to use libdispatch on FreeBSD/amd64 with blocks: A bug in clang's __builtin_expect() was corrected in clang r85666, allowing it to be used with values/pointers > 32-bit. The bug didn't (visibly) affect Mac OS X as its malloc(3) does not (quickly) return pointers >32-bit, whereas the FreeBSD/amd64 malloc(3) uses them immediately on program start. If you've been seeing this segfault, try updating clang and see if that helps. Thanks to Ed Schouten to responding so rapidly to my bug report :-). If using the FreeBSD clang port: it hasn't yet been updated past that point (as of this morning), but you can force it to use the latest subversion revision by doing 'make BOOTSTRAP=yes makesum' before doing make/make install. Robert N M Watson Computer Laboratory University of Cambridge

1 0