* libguile/gc-malloc.c (scm_gc_malloc): Pass a non-zero size to
`GC_MALLOC ()' when SIZE is zero.
git-archimport-id: lcourtes@laas.fr--2006-libre/guile-core--boehm-gc--0--patch-2
* New file gc-segment-table.c: hold code for the segment table.
* Remove data that might be out of date; remove
scm_i_adjust_min_yield(). We don't store min_yields, since they
are only accurate at one point in time (when the sweep finishes).
We decide the min yield at that point from min_yield_fraction and
freelist->collected / freelist->swept
* Introduce scm_i_gc_heap_size_delta() replacing
scm_i_gc_grow_heap_p().
* Remove foo_1 fields containing penultimate results.
* After GC, count mark bit vector to discover number of live
objects. This simplifies hairy updates.
* Many formatting and layout cleanups.
* Fix in scm_i_sweep_card(): return the length of free_list returned,
rather than number of deleted objects.
* For mtrigger GCs: do not also run a full sweep after the gc() call, as
this is inconsistent with lazy sweeping.
* Remove scm_i_make_initial_segment().
* Use calloc in scm_i_make_empty_heap_segment() to save on
initialization code.
* New function scm_i_sweep_for_freelist() which sweeps, with proper
statistic variable updates.
* New segments are conceptually blocks with 100% reclaimable cells.
* Remove some useless constants/comments: SCM_HEAP_SIZE,
SCM_INIT_HEAP_SIZE, SCM_EXPHEAP, SCM_HEAP_SEG_SIZE
* Do not increment scm_cells_allocated() from the
scm_[double]cell(). This would be a race condition.
* Move some deprecation checks in separate functions to not distract
from main code flow.
(scm_block_gc, scm_gc_heap_lock): Removed. Removed all uses.
(scm_gc_running_p): Now a macro that refers to the scm_i_thread
field.
(scm_i_sweep_mutex): Now a non-recursive mutex. GC can not happen
recursively.
(scm_igc, scm_i_gc): Renamed former to latter. Changed all uses.
Do not lock scm_i_sweep_mutex, which is now non-recursive, or set
scm_gc_running_p. Do not run the scm_after_gc_c_hook.
(scm_gc): Lock scm_i_sweep_mutex, set scm_gc_running_p and run the
scm_after_gc_c_hook here.
(scm_gc_for_new_cell): Set scm_gc_running_p here and run the
scm_after_gc_c_hook when a full GC has in fact been performed.
(scm_i_expensive_validation_check): Call scm_gc, not scm_i_gc.
* gc-segment.c (scm_i_get_new_heap_segment): Do not check
scm_gc_heap_lock.
* gc-malloc.c (scm_realloc, increase_mtrigger): Set
scm_gc_running_p while the scm_i_sweep_mutex is locked.
scm_gc_mallocated, for now.
(scm_init_storage): Initialize it.
* gc-malloc.c (descrease_mtrigger, increase_mtrigger): Use it.
* gc-mark.c (scm_gc_mark_dependencies): Call scm_i_string_mark,
scm_i_stringbuf_mark and scm_i_symbol_mark, as appropriate.
* gc-card.c (scm_i_sweep_card): Call scm_i_string_free,
scm_i_stringbuf_free and scm_i_symbol_free, as appropriate.
(increase_mtrigger): new function, separate debug registering and
mtrigger administration.
(scm_gc_realloc): bugfix: do mtrigger administration before the
actual realloc, for the realloc might invalidate a GC-d segment of
memory. Thanks to Sam Hocevar for pointing this out.
(scm_gc_realloc): use scm_malloc_reregister instead of
unregistering and registering in sequence.
* gc.c (scm_gc_for_newcell), gc-malloc.c (scm_realloc,
scm_gc_register_collectable_memory): Substitute locking of
scm_i_sweep_mutex for calls to scm_i_thread_put_to_sleep.
(scm_igc): Lock sweep mutex here instead of in callers; Calls to
scm_i_thread_put_to_sleep/scm_i_thread_wake_up used to demarkate
the single-thread section (which now only contains the mark
phase).
(scm_gc): Don't lock sweeo mutex here since scm_igc locks it;
Removed SCM_DEFER/ALLOW_INTS. Simply call scm_igc directly.
* threads.c (gc_section_mutex): Removed.
* gc-malloc.c (malloc_mutex): New mutex.
(scm_gc_malloc_prehistory): Initialize it.
(scm_realloc): Serialize call to realloc
(scm_calloc): Same for calloc.
Thanks to Wolfgang Jaehrling!
(Now we have to make sure all calls to malloc/realloc are made
through scm_malloc.)
* init.c (scm_init_guile_1): Call scm_gc_malloc_prehistory.
(scm_gc_init_malloc): Initialize it.
(scm_realloc): Serialize call to realloc
(scm_calloc): Same for calloc.
Thanks to Wolfgang Jaehrling!
(Now we have to make sure all calls to malloc/realloc are made
through scm_malloc.)
(SCM_NONREC_CRITICAL_SECTION_START,
SCM_NONREC_CRITICAL_SECTION_END, SCM_REC_CRITICAL_SECTION_START,
SCM_REC_CRITICAL_SECTION_END): New macros.
(SCM_CRITICAL_SECTION_START/END): Defined here.
* eval.c: Insert SOURCE_SECTION_START / SOURCE_SECTION_END around
the three calls to scm_m_expand_body.
* gc.h: #include "libguile/pthread-threads.h";
(SCM_FREELIST_CREATE, SCM_FREELIST_LOC): New macros.
* gc.c (scm_i_freelist, scm_i_freelist2): Defined to be of type
scm_t_key;
* gc.c, gc-freelist.c, inline.h: Use SCM_FREELIST_LOC for freelist
access.
* gc-freelist.c (scm_gc_init_freelist): Create freelist keys.
* gc-freelist.c, threads.c (really_launch): Use
SCM_FREELIST_CREATE.
* gc-malloc.c (scm_realloc, scm_gc_register_collectable_memory):
* gc.c (scm_i_expensive_validation_check, scm_gc,
scm_gc_for_newcell): Put threads to sleep before doing GC-related
heap administration so that those pieces of code are executed
single-threaded. We might consider rewriting these code sections
in terms of a "call_gc_code_singly_threaded" construct instead of
calling the pair of scm_i_thread_put_to_sleep () and
scm_i_thread_wake_up (). Also, we would want to have as many of
these sections eleminated.
* init.c (scm_init_guile_1): Call scm_threads_prehistory.
* inline.h: #include "libguile/threads.h"
* pthread-threads.h: Macros now conform more closely to the
pthreads interface. Some of them now take a second argument.
* threads.c, threads.h: Many changes.
* configure.in: Temporarily replaced "copt" threads option with new
option "pthreads".
(USE_PTHREAD_THREADS): Define if pthreads configured.
for sanity.
* gc-malloc.c (scm_gc_init_malloc): check user settings for
sanity.
(scm_gc_register_collectable_memory): prevent overflow of memory
counts.
* gc-freelist.c (scm_init_freelist): check user settings for sanity.
* gc-malloc.c (scm_gc_register_collectable_memory): use floats;
these won't ever wrap around with high memory usage.
* gc-freelist.c: include <stdio.h>
* gc-malloc.c: add DEBUGINFO for mtrigger GCs.
* gc-malloc.c (scm_gc_register_collectable_memory): use floats;
these won't ever wrap around with high memory usage.
* gc-malloc.c: add DEBUGINFO for mtrigger GCs.
* gc-malloc.c (scm_calloc): try to use calloc() before calling
scm_realloc().
* gc-segment.c (scm_i_initialize_heap_segment_data): remove card
init loop; handle this from scm_init_card_freelist()
* gc-card.c (scm_init_card_freelist): init bit vector here.
