Thu Dec 28 21:14:45 2006 Tim Janik <timj@imendio.com>
* glib/gslice.c: removed pthread-dependant debugging bits, the code
was already converted to GMutex. this obsoletes Tor's recent fixups.
2006-12-28 Tor Lillqvist <tml@novell.com>
* glib/gslice.c: Call InitializeCriticalSection() on the
sdt_mutex in g_slice_init_nomessage() and not in
_g_slice_thread_init_nomessage(), which is called only through
g_thread_init(). Not all programs call g_thread_init().
Thu Dec 28 12:50:31 2006 Tim Janik <timj@imendio.com>
* glib/gslice.h, glib/gslice.c: implemented static debugging
hash-tree to validate slice adresses and sizes with G_SLICE=debug-blocks.
use abort() to exit in mem_error() to allow catching of these in gdb.
abort programs with a descriptive error message if g_thread_init() is
called after GSlice was in use. previously this just silently corrupted
the magazines.
* glib/ghash.c (struct _GHashNode): reordered fields to keep 8-byte
pointer alignment on 64bit systems and request smaller slice sizes
on 32bit systems.
* tests/slice-test.c: support '~' option flag to introduce slice
allocation/release corruption with a significant probability. this
allowes testing of G_SLICE=debug-blocks.
2006-05-09 Sebastian Wilhelmi <wilhelmi@google.com>
* glib/gthreadinit.h: Renamed to glib/gthreadprivate.h and moved
system thread identifier comparision and assignment macros from
glib/gthread.c to glib/gthreadprivate.h.
* glib/Makefile.am, glib/gatomic.c, glib/gconvert.c, glib/gmain.c,
glib/gmem.c, glib/gmessages.c, glib/grand.c, glib/gslice.c,
glib/gthread.c, glib/gutils.c, gthread/gthread-impl.c: Use
glib/gthreadprivate.h instead of glib/gthreadinit.h.
* gthread/gthread-impl.c: Use GSystemThread instead of GThread for
owner determination. This fixes#311043 and is mostly modeled
after the patch from jylefort@FreeBSD.org.
Tue Feb 14 17:00:43 2006 Tim Janik <timj@imendio.com>
* glib/gslice.c: only define _XOPEN_SOURCE to 600 to get at
posix_memalign() in case we actually are going to use it, because
we detected a compliant implementation (#328997).
* configure.in (enable_included_printf): don't include malloc.h when
testing for posix_memalign() funcitonality, since this may break the
test on some systems (#328997).
Wed Jan 25 19:16:57 2006 Tim Janik <timj@imendio.com>
* fixed buglets reported by Jens Granseuer in #328254.
* configure.in: free the memory allocated in posix_memalign() tests.
* glib/gslice.c: spelling fixes.
Wed Jan 25 16:39:18 2006 Tim Janik <timj@imendio.com>
* glib/gslice.c: honour g_mem_gc_friendly settings when freeing
slices, make sure g_mem_gc_friendly is properly initialized.
* gmem.[hc]: ensure g_mem_gc_friendly is initialized from G_DEBUG upon
the first allocation. applied some branching optimizations.
* docs/macros.txt: reflected --enable-gc-friendly change and
described ENABLE_GC_FRIENDLY_DEFAULT as well as G_DEBUG=gc-friendly.
* configure.in: changed --enable-gc-friendly=yes to define
ENABLE_GC_FRIENDLY_DEFAULT.
* glib/garray.c: changed ENABLE_GC_FRIENDLY macro #ifdef-s to
if (G_UNLIKELY (g_mem_gc_friendly)).
* glib/gtree.c:
* glib/ghash.c: removed ENABLE_GC_FRIENDLY code which is now taken
care of by g_slice_free1().
* tests/slice-test.c: fixed leaks, reported by Kjartan Maraas.
Tue Jan 24 17:49:36 2006 Tim Janik <timj@imendio.com>
* glib/gslice.c: only use posix_memalign() if it's known to work,
revert to memalign() otherwise.
* configure.in: check for broken posix_memalign() implementations
to fix#328254.
Mon Jan 23 16:46:20 2006 Tim Janik <timj@imendio.com>
* glib/gslice.c (slice_config_init): initialize GSlice config from
G_SLICE environemtn variable. we support G_SLICE=always-malloc
currently, which forces all g_slice_*() allocations to use the system
malloc instead.
* glib/gutils.c:
g_parse_debug_string(): added a note about not using g_malloc() here.
_g_getenv_nomalloc(): getenv() variant that doesn't use g_malloc or
g_slice. contains only guesswork in the WIN32 branch.
Wed Jan 4 13:33:25 2006 Tim Janik <timj@gtk.org>
* glib/gslice.c (magazine_cache_pop_magazine): don't reverse chunk
order when creating magazines, so we hand out chunks with ascending
adresses.
Tue Dec 20 18:14:14 2005 Tim Janik <timj@imendio.com>
* glib/gslice.[hc]: added mem_error() and mem_assert() to test and
handle errors without depending on gmessage.c which might not be
setup when the error occours.
removed G_SLICE_CONFIG_ALWAYS_FREE config option, fixed the code so
always freeing can be achieved by adjusting the working set time to
0 with G_SLICE_CONFIG_WORKING_SET_MSECS.
added G_SLICE_CONFIG_COLOR_INCREMENT to test different color increments
(mainly 0 and 1). reduced the minimum block size to 128 bytes, to
minimize wastage if small amounts of differently sized structrues are
allocated, this does come at a performance cost of roughly 5% though.
fixed up block alignment calculation, so it works for varying
block sizes. only use strerror() not g_strerror() since the latter
depends on working GQuark and GSlice.
mem_error(): implemented in terms of fprintf and vfprintf.
* tests/slice-color.c: new program to test cache colorization effects.
* tests/slice-test.c: trade G_SLICE_CONFIG_ALWAYS_FREE for 0 duration
G_SLICE_CONFIG_WORKING_SET_MSECS.
Mon Dec 5 15:53:20 2005 Tim Janik <timj@imendio.com>
* glib/gslice.c: implement chain walking for arbitrary ->next pointer
offsets in g_slice_free_chain_with_offset() based on a patch by behdad
in bug 323178. moved time consuming logic from g_slice_free() out of
the inner loop, so g_slice_free_chain_with_offset() provides a real
performance benefit over g_slice_free1() now.
* glib/gslice.h: renamed g_slice_free_chain() to
g_slice_free_chain_with_offset(). implemented g_slice_free_chain() as
a type-safe macro as suggested in bug 323178.
simplified the macro implementation of g_slice_free() and implemented
it in a type safe manner for all compliers as suggested by Morten
Welinder <mortenw@gnome.org>.
* glib/gmain.c:
* glib/glist.c:
* glib/gslist.c:
* glib/glib.symbols: s/g_slice_free_chain/g_slice_free_chain_with_offset/
Fri Dec 2 00:16:59 2005 Tim Janik <timj@imendio.com>
* glib/gslice.c: improved gettimeofday() timer resolution by maintaining
acceptable load distribution of the syscall.
(slab_allocator_free_chunk): reduce minimum/single-thread magazine sizes
to approximately page_size/2.5 to avoid excessive startup allocations.
Thu Dec 1 17:32:46 2005 Tim Janik <timj@imendio.com>
* glib/gslice.[hc]: new slice allocator implementation.
* tests/slice-test.c: added random slice allocation test.
* glib/gthread.[hc]: removed newly added private thread mem API.
* glib/gthreadinit.h:
* glib/gmessages.c:
* glib/gthread.c:
* glib/gmem.c: divided glib threading initialisation into three phases,
initialisation where private keys and messaging are not available (only
needed by gmem.c), initialisation without messaging but private keys
available (gslice.c, gmessage.c), and full fledged initialisers that
server the rest of glib. initialisation functions got renamed to reflect
the limitations of their corresponding phases.
* glib/gmem.c: removed memchunk code, defer allocations to
g_slice_* instead.
* glib/gmem.[hc]: removed g_slice_* skeletons.
* glib/glib.symbols: added g_slice_* symbols.
* configure.in: check for availability of posix_memalign(3), memalign(3)
and valloc(3).
* glib/Makefile.am: added gslice.[hc].
