2000-06-30 13:02:53 +08:00
|
|
|
/*
|
SF bug #574132: Major GC related performance regression
"The regression" is actually due to that 2.2.1 had a bug that prevented
the regression (which isn't a regression at all) from showing up. "The
regression" is actually a glitch in cyclic gc that's been there forever.
As the generation being collected is analyzed, objects that can't be
collected (because, e.g., we find they're externally referenced, or
are in an unreachable cycle but have a __del__ method) are moved out
of the list of candidates. A tricksy scheme uses negative values of
gc_refs to mark such objects as being moved. However, the exact
negative value set at the start may become "more negative" over time
for objects not in the generation being collected, and the scheme was
checking for an exact match on the negative value originally assigned.
As a result, objects in generations older than the one being collected
could get scanned too, and yanked back into a younger generation. Doing
so doesn't lead to an error, but doesn't do any good, and can burn an
unbounded amount of time doing useless work.
A test case is simple (thanks to Kevin Jacobs for finding it!):
x = []
for i in xrange(200000):
x.append((1,))
Without the patch, this ends up scanning all of x on every gen0 collection,
scans all of x twice on every gen1 collection, and x gets yanked back into
gen1 on every gen0 collection. With the patch, once x gets to gen2, it's
never scanned again until another gen2 collection, and stays in gen2.
Bugfix candidate, although the code has changed enough that I think I'll
need to port it by hand. 2.2.1 also has a different bug that causes
bound method objects not to get tracked at all (so the test case doesn't
burn absurd amounts of time in 2.2.1, but *should* <wink>).
2002-07-01 01:56:40 +08:00
|
|
|
|
2000-06-30 13:02:53 +08:00
|
|
|
Reference Cycle Garbage Collection
|
|
|
|
==================================
|
|
|
|
|
2000-10-05 00:34:09 +08:00
|
|
|
Neil Schemenauer <nas@arctrix.com>
|
2000-06-30 13:02:53 +08:00
|
|
|
|
|
|
|
Based on a post on the python-dev list. Ideas from Guido van Rossum,
|
|
|
|
Eric Tiedemann, and various others.
|
|
|
|
|
2001-08-30 08:05:51 +08:00
|
|
|
http://www.arctrix.com/nas/python/gc/
|
2000-06-30 13:02:53 +08:00
|
|
|
http://www.python.org/pipermail/python-dev/2000-March/003869.html
|
|
|
|
http://www.python.org/pipermail/python-dev/2000-March/004010.html
|
|
|
|
http://www.python.org/pipermail/python-dev/2000-March/004022.html
|
|
|
|
|
|
|
|
For a highlevel view of the collection process, read the collect
|
|
|
|
function.
|
|
|
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include "Python.h"
|
|
|
|
|
|
|
|
#ifdef WITH_CYCLE_GC
|
|
|
|
|
2001-08-30 08:05:51 +08:00
|
|
|
/* Get an object's GC head */
|
|
|
|
#define AS_GC(o) ((PyGC_Head *)(o)-1)
|
|
|
|
|
|
|
|
/* Get the object given the GC head */
|
|
|
|
#define FROM_GC(g) ((PyObject *)(((PyGC_Head *)g)+1))
|
|
|
|
|
2002-05-21 23:53:24 +08:00
|
|
|
/* True if an object is tracked by the GC */
|
|
|
|
#define IS_TRACKED(o) ((AS_GC(o))->gc.gc_next != NULL)
|
2000-06-30 13:02:53 +08:00
|
|
|
|
|
|
|
/*** Global GC state ***/
|
|
|
|
|
2002-05-04 13:35:20 +08:00
|
|
|
struct gc_generation {
|
|
|
|
PyGC_Head head;
|
|
|
|
int threshold; /* collection threshold */
|
|
|
|
int count; /* count of allocations or collections of younger
|
|
|
|
generations */
|
|
|
|
};
|
|
|
|
|
|
|
|
#define NUM_GENERATIONS 3
|
|
|
|
#define GEN_HEAD(n) (&generations[n].head)
|
|
|
|
|
2000-06-30 13:02:53 +08:00
|
|
|
/* linked lists of container objects */
|
2002-05-04 13:35:20 +08:00
|
|
|
static struct gc_generation generations[NUM_GENERATIONS] = {
|
|
|
|
/* PyGC_Head, threshold, count */
|
|
|
|
{{{GEN_HEAD(0), GEN_HEAD(0), 0}}, 700, 0},
|
|
|
|
{{{GEN_HEAD(1), GEN_HEAD(1), 0}}, 10, 0},
|
|
|
|
{{{GEN_HEAD(2), GEN_HEAD(2), 0}}, 10, 0},
|
|
|
|
};
|
2000-06-30 13:02:53 +08:00
|
|
|
|
2002-05-04 13:35:20 +08:00
|
|
|
PyGC_Head *_PyGC_generation0 = GEN_HEAD(0);
|
2000-06-30 13:02:53 +08:00
|
|
|
|
2002-05-04 13:35:20 +08:00
|
|
|
static int enabled = 1; /* automatic collection enabled? */
|
2000-06-30 13:02:53 +08:00
|
|
|
|
2001-08-30 08:05:51 +08:00
|
|
|
/* true if we are currently running the collector */
|
|
|
|
static int collecting;
|
|
|
|
|
2000-06-30 13:02:53 +08:00
|
|
|
/* set for debugging information */
|
|
|
|
#define DEBUG_STATS (1<<0) /* print collection statistics */
|
|
|
|
#define DEBUG_COLLECTABLE (1<<1) /* print collectable objects */
|
|
|
|
#define DEBUG_UNCOLLECTABLE (1<<2) /* print uncollectable objects */
|
|
|
|
#define DEBUG_INSTANCES (1<<3) /* print instances */
|
|
|
|
#define DEBUG_OBJECTS (1<<4) /* print other objects */
|
2000-09-22 23:22:38 +08:00
|
|
|
#define DEBUG_SAVEALL (1<<5) /* save all garbage in gc.garbage */
|
2000-06-30 13:02:53 +08:00
|
|
|
#define DEBUG_LEAK DEBUG_COLLECTABLE | \
|
|
|
|
DEBUG_UNCOLLECTABLE | \
|
|
|
|
DEBUG_INSTANCES | \
|
2000-09-22 23:22:38 +08:00
|
|
|
DEBUG_OBJECTS | \
|
|
|
|
DEBUG_SAVEALL
|
2000-09-01 10:47:25 +08:00
|
|
|
static int debug;
|
2000-06-30 13:02:53 +08:00
|
|
|
|
SF bug #574132: Major GC related performance regression
"The regression" is actually due to that 2.2.1 had a bug that prevented
the regression (which isn't a regression at all) from showing up. "The
regression" is actually a glitch in cyclic gc that's been there forever.
As the generation being collected is analyzed, objects that can't be
collected (because, e.g., we find they're externally referenced, or
are in an unreachable cycle but have a __del__ method) are moved out
of the list of candidates. A tricksy scheme uses negative values of
gc_refs to mark such objects as being moved. However, the exact
negative value set at the start may become "more negative" over time
for objects not in the generation being collected, and the scheme was
checking for an exact match on the negative value originally assigned.
As a result, objects in generations older than the one being collected
could get scanned too, and yanked back into a younger generation. Doing
so doesn't lead to an error, but doesn't do any good, and can burn an
unbounded amount of time doing useless work.
A test case is simple (thanks to Kevin Jacobs for finding it!):
x = []
for i in xrange(200000):
x.append((1,))
Without the patch, this ends up scanning all of x on every gen0 collection,
scans all of x twice on every gen1 collection, and x gets yanked back into
gen1 on every gen0 collection. With the patch, once x gets to gen2, it's
never scanned again until another gen2 collection, and stays in gen2.
Bugfix candidate, although the code has changed enough that I think I'll
need to port it by hand. 2.2.1 also has a different bug that causes
bound method objects not to get tracked at all (so the test case doesn't
burn absurd amounts of time in 2.2.1, but *should* <wink>).
2002-07-01 01:56:40 +08:00
|
|
|
/* When a collection begins, gc_refs is set to ob_refcnt for, and only for,
|
|
|
|
* the objects in the generation being collected, called the "young"
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
* generation at that point. As collection proceeds, the gc_refs members
|
|
|
|
* of young objects are set to GC_REACHABLE when it becomes known that they're
|
|
|
|
* uncollectable, and to GC_TENTATIVELY_UNREACHABLE when the evidence
|
|
|
|
* suggests they are collectable (this can't be known for certain until all
|
|
|
|
* of the young generation is scanned).
|
SF bug #574132: Major GC related performance regression
"The regression" is actually due to that 2.2.1 had a bug that prevented
the regression (which isn't a regression at all) from showing up. "The
regression" is actually a glitch in cyclic gc that's been there forever.
As the generation being collected is analyzed, objects that can't be
collected (because, e.g., we find they're externally referenced, or
are in an unreachable cycle but have a __del__ method) are moved out
of the list of candidates. A tricksy scheme uses negative values of
gc_refs to mark such objects as being moved. However, the exact
negative value set at the start may become "more negative" over time
for objects not in the generation being collected, and the scheme was
checking for an exact match on the negative value originally assigned.
As a result, objects in generations older than the one being collected
could get scanned too, and yanked back into a younger generation. Doing
so doesn't lead to an error, but doesn't do any good, and can burn an
unbounded amount of time doing useless work.
A test case is simple (thanks to Kevin Jacobs for finding it!):
x = []
for i in xrange(200000):
x.append((1,))
Without the patch, this ends up scanning all of x on every gen0 collection,
scans all of x twice on every gen1 collection, and x gets yanked back into
gen1 on every gen0 collection. With the patch, once x gets to gen2, it's
never scanned again until another gen2 collection, and stays in gen2.
Bugfix candidate, although the code has changed enough that I think I'll
need to port it by hand. 2.2.1 also has a different bug that causes
bound method objects not to get tracked at all (so the test case doesn't
burn absurd amounts of time in 2.2.1, but *should* <wink>).
2002-07-01 01:56:40 +08:00
|
|
|
*/
|
2001-08-30 08:05:51 +08:00
|
|
|
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
/* Special gc_refs values. */
|
2002-07-02 08:52:30 +08:00
|
|
|
#define GC_UNTRACKED _PyGC_REFS_UNTRACKED
|
|
|
|
#define GC_REACHABLE _PyGC_REFS_REACHABLE
|
|
|
|
#define GC_TENTATIVELY_UNREACHABLE _PyGC_REFS_TENTATIVELY_UNREACHABLE
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
|
|
|
|
#define IS_REACHABLE(o) ((AS_GC(o))->gc.gc_refs == GC_REACHABLE)
|
|
|
|
#define IS_TENTATIVELY_UNREACHABLE(o) ( \
|
|
|
|
(AS_GC(o))->gc.gc_refs == GC_TENTATIVELY_UNREACHABLE)
|
2002-05-21 23:53:24 +08:00
|
|
|
|
2000-06-30 13:02:53 +08:00
|
|
|
/* list of uncollectable objects */
|
|
|
|
static PyObject *garbage;
|
|
|
|
|
2000-09-01 10:47:25 +08:00
|
|
|
/* Python string to use if unhandled exception occurs */
|
|
|
|
static PyObject *gc_str;
|
2000-06-30 13:02:53 +08:00
|
|
|
|
|
|
|
/*** list functions ***/
|
|
|
|
|
|
|
|
static void
|
|
|
|
gc_list_init(PyGC_Head *list)
|
|
|
|
{
|
2001-10-12 02:31:31 +08:00
|
|
|
list->gc.gc_prev = list;
|
|
|
|
list->gc.gc_next = list;
|
2000-06-30 13:02:53 +08:00
|
|
|
}
|
|
|
|
|
2002-05-04 13:35:20 +08:00
|
|
|
static int
|
|
|
|
gc_list_is_empty(PyGC_Head *list)
|
|
|
|
{
|
|
|
|
return (list->gc.gc_next == list);
|
|
|
|
}
|
|
|
|
|
2000-06-30 13:02:53 +08:00
|
|
|
static void
|
|
|
|
gc_list_append(PyGC_Head *node, PyGC_Head *list)
|
|
|
|
{
|
2001-10-12 02:31:31 +08:00
|
|
|
node->gc.gc_next = list;
|
|
|
|
node->gc.gc_prev = list->gc.gc_prev;
|
|
|
|
node->gc.gc_prev->gc.gc_next = node;
|
|
|
|
list->gc.gc_prev = node;
|
2000-06-30 13:02:53 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
gc_list_remove(PyGC_Head *node)
|
|
|
|
{
|
2001-10-12 02:31:31 +08:00
|
|
|
node->gc.gc_prev->gc.gc_next = node->gc.gc_next;
|
|
|
|
node->gc.gc_next->gc.gc_prev = node->gc.gc_prev;
|
|
|
|
node->gc.gc_next = NULL; /* object is not currently tracked */
|
2000-06-30 13:02:53 +08:00
|
|
|
}
|
|
|
|
|
SF bug #574132: Major GC related performance regression
"The regression" is actually due to that 2.2.1 had a bug that prevented
the regression (which isn't a regression at all) from showing up. "The
regression" is actually a glitch in cyclic gc that's been there forever.
As the generation being collected is analyzed, objects that can't be
collected (because, e.g., we find they're externally referenced, or
are in an unreachable cycle but have a __del__ method) are moved out
of the list of candidates. A tricksy scheme uses negative values of
gc_refs to mark such objects as being moved. However, the exact
negative value set at the start may become "more negative" over time
for objects not in the generation being collected, and the scheme was
checking for an exact match on the negative value originally assigned.
As a result, objects in generations older than the one being collected
could get scanned too, and yanked back into a younger generation. Doing
so doesn't lead to an error, but doesn't do any good, and can burn an
unbounded amount of time doing useless work.
A test case is simple (thanks to Kevin Jacobs for finding it!):
x = []
for i in xrange(200000):
x.append((1,))
Without the patch, this ends up scanning all of x on every gen0 collection,
scans all of x twice on every gen1 collection, and x gets yanked back into
gen1 on every gen0 collection. With the patch, once x gets to gen2, it's
never scanned again until another gen2 collection, and stays in gen2.
Bugfix candidate, although the code has changed enough that I think I'll
need to port it by hand. 2.2.1 also has a different bug that causes
bound method objects not to get tracked at all (so the test case doesn't
burn absurd amounts of time in 2.2.1, but *should* <wink>).
2002-07-01 01:56:40 +08:00
|
|
|
static void
|
2000-06-30 13:02:53 +08:00
|
|
|
gc_list_move(PyGC_Head *from, PyGC_Head *to)
|
|
|
|
{
|
2002-05-04 13:35:20 +08:00
|
|
|
if (gc_list_is_empty(from)) {
|
2000-06-30 13:02:53 +08:00
|
|
|
gc_list_init(to);
|
2000-09-22 23:22:38 +08:00
|
|
|
}
|
|
|
|
else {
|
2001-10-12 02:31:31 +08:00
|
|
|
to->gc.gc_next = from->gc.gc_next;
|
|
|
|
to->gc.gc_next->gc.gc_prev = to;
|
|
|
|
to->gc.gc_prev = from->gc.gc_prev;
|
|
|
|
to->gc.gc_prev->gc.gc_next = to;
|
2000-06-30 13:02:53 +08:00
|
|
|
}
|
|
|
|
gc_list_init(from);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* append a list onto another list, from becomes an empty list */
|
|
|
|
static void
|
|
|
|
gc_list_merge(PyGC_Head *from, PyGC_Head *to)
|
|
|
|
{
|
|
|
|
PyGC_Head *tail;
|
2002-05-04 13:35:20 +08:00
|
|
|
if (!gc_list_is_empty(from)) {
|
2001-10-12 02:31:31 +08:00
|
|
|
tail = to->gc.gc_prev;
|
|
|
|
tail->gc.gc_next = from->gc.gc_next;
|
|
|
|
tail->gc.gc_next->gc.gc_prev = tail;
|
|
|
|
to->gc.gc_prev = from->gc.gc_prev;
|
|
|
|
to->gc.gc_prev->gc.gc_next = to;
|
2000-06-30 13:02:53 +08:00
|
|
|
}
|
|
|
|
gc_list_init(from);
|
|
|
|
}
|
|
|
|
|
|
|
|
static long
|
|
|
|
gc_list_size(PyGC_Head *list)
|
|
|
|
{
|
|
|
|
PyGC_Head *gc;
|
|
|
|
long n = 0;
|
2001-10-12 02:31:31 +08:00
|
|
|
for (gc = list->gc.gc_next; gc != list; gc = gc->gc.gc_next) {
|
2000-06-30 13:02:53 +08:00
|
|
|
n++;
|
|
|
|
}
|
|
|
|
return n;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*** end of list stuff ***/
|
|
|
|
|
|
|
|
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
/* Set all gc_refs = ob_refcnt. After this, gc_refs is > 0 for all objects
|
|
|
|
* in containers, and is GC_REACHABLE for all tracked gc objects not in
|
|
|
|
* containers.
|
SF bug #574132: Major GC related performance regression
"The regression" is actually due to that 2.2.1 had a bug that prevented
the regression (which isn't a regression at all) from showing up. "The
regression" is actually a glitch in cyclic gc that's been there forever.
As the generation being collected is analyzed, objects that can't be
collected (because, e.g., we find they're externally referenced, or
are in an unreachable cycle but have a __del__ method) are moved out
of the list of candidates. A tricksy scheme uses negative values of
gc_refs to mark such objects as being moved. However, the exact
negative value set at the start may become "more negative" over time
for objects not in the generation being collected, and the scheme was
checking for an exact match on the negative value originally assigned.
As a result, objects in generations older than the one being collected
could get scanned too, and yanked back into a younger generation. Doing
so doesn't lead to an error, but doesn't do any good, and can burn an
unbounded amount of time doing useless work.
A test case is simple (thanks to Kevin Jacobs for finding it!):
x = []
for i in xrange(200000):
x.append((1,))
Without the patch, this ends up scanning all of x on every gen0 collection,
scans all of x twice on every gen1 collection, and x gets yanked back into
gen1 on every gen0 collection. With the patch, once x gets to gen2, it's
never scanned again until another gen2 collection, and stays in gen2.
Bugfix candidate, although the code has changed enough that I think I'll
need to port it by hand. 2.2.1 also has a different bug that causes
bound method objects not to get tracked at all (so the test case doesn't
burn absurd amounts of time in 2.2.1, but *should* <wink>).
2002-07-01 01:56:40 +08:00
|
|
|
*/
|
2000-06-30 13:02:53 +08:00
|
|
|
static void
|
|
|
|
update_refs(PyGC_Head *containers)
|
|
|
|
{
|
2001-10-12 02:31:31 +08:00
|
|
|
PyGC_Head *gc = containers->gc.gc_next;
|
2002-07-02 08:52:30 +08:00
|
|
|
for (; gc != containers; gc = gc->gc.gc_next) {
|
|
|
|
assert(gc->gc.gc_refs == GC_REACHABLE);
|
2001-10-12 02:31:31 +08:00
|
|
|
gc->gc.gc_refs = FROM_GC(gc)->ob_refcnt;
|
2002-07-02 08:52:30 +08:00
|
|
|
}
|
2000-06-30 13:02:53 +08:00
|
|
|
}
|
|
|
|
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
/* A traversal callback for subtract_refs. */
|
2000-06-30 13:02:53 +08:00
|
|
|
static int
|
|
|
|
visit_decref(PyObject *op, void *data)
|
|
|
|
{
|
2002-07-01 05:31:03 +08:00
|
|
|
assert(op != NULL);
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
if (PyObject_IS_GC(op)) {
|
|
|
|
PyGC_Head *gc = AS_GC(op);
|
|
|
|
/* We're only interested in gc_refs for objects in the
|
|
|
|
* generation being collected, which can be recognized
|
|
|
|
* because only they have positive gc_refs.
|
|
|
|
*/
|
|
|
|
if (gc->gc.gc_refs > 0)
|
|
|
|
gc->gc.gc_refs--;
|
|
|
|
}
|
2000-06-30 13:02:53 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
/* Subtract internal references from gc_refs. After this, gc_refs is >= 0
|
|
|
|
* for all objects in containers, and is GC_REACHABLE for all tracked gc
|
|
|
|
* objects not in containers. The ones with gc_refs > 0 are directly
|
|
|
|
* reachable from outside containers, and so can't be collected.
|
|
|
|
*/
|
2000-06-30 13:02:53 +08:00
|
|
|
static void
|
|
|
|
subtract_refs(PyGC_Head *containers)
|
|
|
|
{
|
|
|
|
traverseproc traverse;
|
2001-10-12 02:31:31 +08:00
|
|
|
PyGC_Head *gc = containers->gc.gc_next;
|
|
|
|
for (; gc != containers; gc=gc->gc.gc_next) {
|
2001-08-30 08:05:51 +08:00
|
|
|
traverse = FROM_GC(gc)->ob_type->tp_traverse;
|
|
|
|
(void) traverse(FROM_GC(gc),
|
2000-06-30 13:02:53 +08:00
|
|
|
(visitproc)visit_decref,
|
|
|
|
NULL);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
/* A traversal callback for move_unreachable. */
|
2000-06-30 13:02:53 +08:00
|
|
|
static int
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
visit_reachable(PyObject *op, PyGC_Head *reachable)
|
2000-06-30 13:02:53 +08:00
|
|
|
{
|
2002-07-02 08:52:30 +08:00
|
|
|
if (PyObject_IS_GC(op)) {
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
PyGC_Head *gc = AS_GC(op);
|
|
|
|
const int gc_refs = gc->gc.gc_refs;
|
|
|
|
|
|
|
|
if (gc_refs == 0) {
|
|
|
|
/* This is in move_unreachable's 'young' list, but
|
|
|
|
* the traversal hasn't yet gotten to it. All
|
|
|
|
* we need to do is tell move_unreachable that it's
|
|
|
|
* reachable.
|
|
|
|
*/
|
|
|
|
gc->gc.gc_refs = 1;
|
|
|
|
}
|
|
|
|
else if (gc_refs == GC_TENTATIVELY_UNREACHABLE) {
|
|
|
|
/* This had gc_refs = 0 when move_unreachable got
|
|
|
|
* to it, but turns out it's reachable after all.
|
|
|
|
* Move it back to move_unreachable's 'young' list,
|
|
|
|
* and move_unreachable will eventually get to it
|
|
|
|
* again.
|
|
|
|
*/
|
2000-06-30 13:02:53 +08:00
|
|
|
gc_list_remove(gc);
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
gc_list_append(gc, reachable);
|
|
|
|
gc->gc.gc_refs = 1;
|
2000-06-30 13:02:53 +08:00
|
|
|
}
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
/* Else there's nothing to do.
|
|
|
|
* If gc_refs > 0, it must be in move_unreachable's 'young'
|
|
|
|
* list, and move_unreachable will eventually get to it.
|
|
|
|
* If gc_refs == GC_REACHABLE, it's either in some other
|
|
|
|
* generation so we don't care about it, or move_unreachable
|
2002-07-02 08:52:30 +08:00
|
|
|
* already deat with it.
|
|
|
|
* If gc_refs == GC_UNTRACKED, it must be ignored.
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
*/
|
2002-07-02 08:52:30 +08:00
|
|
|
else {
|
|
|
|
assert(gc_refs > 0
|
|
|
|
|| gc_refs == GC_REACHABLE
|
|
|
|
|| gc_refs == GC_UNTRACKED);
|
|
|
|
}
|
2000-06-30 13:02:53 +08:00
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
/* Move the unreachable objects from young to unreachable. After this,
|
|
|
|
* all objects in young have gc_refs = GC_REACHABLE, and all objects in
|
|
|
|
* unreachable have gc_refs = GC_TENTATIVELY_UNREACHABLE. All tracked
|
|
|
|
* gc objects not in young or unreachable still have gc_refs = GC_REACHABLE.
|
|
|
|
* All objects in young after this are directly or indirectly reachable
|
|
|
|
* from outside the original young; and all objects in unreachable are
|
|
|
|
* not.
|
SF bug #574132: Major GC related performance regression
"The regression" is actually due to that 2.2.1 had a bug that prevented
the regression (which isn't a regression at all) from showing up. "The
regression" is actually a glitch in cyclic gc that's been there forever.
As the generation being collected is analyzed, objects that can't be
collected (because, e.g., we find they're externally referenced, or
are in an unreachable cycle but have a __del__ method) are moved out
of the list of candidates. A tricksy scheme uses negative values of
gc_refs to mark such objects as being moved. However, the exact
negative value set at the start may become "more negative" over time
for objects not in the generation being collected, and the scheme was
checking for an exact match on the negative value originally assigned.
As a result, objects in generations older than the one being collected
could get scanned too, and yanked back into a younger generation. Doing
so doesn't lead to an error, but doesn't do any good, and can burn an
unbounded amount of time doing useless work.
A test case is simple (thanks to Kevin Jacobs for finding it!):
x = []
for i in xrange(200000):
x.append((1,))
Without the patch, this ends up scanning all of x on every gen0 collection,
scans all of x twice on every gen1 collection, and x gets yanked back into
gen1 on every gen0 collection. With the patch, once x gets to gen2, it's
never scanned again until another gen2 collection, and stays in gen2.
Bugfix candidate, although the code has changed enough that I think I'll
need to port it by hand. 2.2.1 also has a different bug that causes
bound method objects not to get tracked at all (so the test case doesn't
burn absurd amounts of time in 2.2.1, but *should* <wink>).
2002-07-01 01:56:40 +08:00
|
|
|
*/
|
2000-06-30 13:02:53 +08:00
|
|
|
static void
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
move_unreachable(PyGC_Head *young, PyGC_Head *unreachable)
|
2000-06-30 13:02:53 +08:00
|
|
|
{
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
PyGC_Head *gc = young->gc.gc_next;
|
|
|
|
|
|
|
|
/* Invariants: all objects "to the left" of us in young have gc_refs
|
|
|
|
* = GC_REACHABLE, and are indeed reachable (directly or indirectly)
|
|
|
|
* from outside the young list as it was at entry. All other objects
|
|
|
|
* from the original young "to the left" of us are in unreachable now,
|
|
|
|
* and have gc_refs = GC_TENTATIVELY_UNREACHABLE. All objects to the
|
|
|
|
* left of us in 'young' now have been scanned, and no objects here
|
|
|
|
* or to the right have been scanned yet.
|
|
|
|
*/
|
|
|
|
|
|
|
|
while (gc != young) {
|
|
|
|
PyGC_Head *next;
|
|
|
|
|
|
|
|
if (gc->gc.gc_refs == 0) {
|
|
|
|
/* This *may* be unreachable. To make progress,
|
|
|
|
* assume it is. gc isn't directly reachable from
|
|
|
|
* any object we've already traversed, but may be
|
|
|
|
* reachable from an object we haven't gotten to yet.
|
|
|
|
* visit_reachable will eventually move gc back into
|
|
|
|
* young if that's so, and we'll see it again.
|
|
|
|
*/
|
|
|
|
next = gc->gc.gc_next;
|
|
|
|
gc_list_remove(gc);
|
|
|
|
gc_list_append(gc, unreachable);
|
|
|
|
gc->gc.gc_refs = GC_TENTATIVELY_UNREACHABLE;
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
/* gc is definitely reachable from outside the
|
|
|
|
* original 'young'. Mark it as such, and traverse
|
|
|
|
* its pointers to find any other objects that may
|
|
|
|
* be directly reachable from it. Note that the
|
|
|
|
* call to tp_traverse may append objects to young,
|
|
|
|
* so we have to wait until it returns to determine
|
|
|
|
* the next object to visit.
|
|
|
|
*/
|
|
|
|
PyObject *op = FROM_GC(gc);
|
|
|
|
traverseproc traverse = op->ob_type->tp_traverse;
|
|
|
|
gc->gc.gc_refs = GC_REACHABLE;
|
|
|
|
(void) traverse(op,
|
|
|
|
(visitproc)visit_reachable,
|
|
|
|
(void *)young);
|
|
|
|
next = gc->gc.gc_next;
|
|
|
|
}
|
|
|
|
gc = next;
|
2000-06-30 13:02:53 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
SF bug #574132: Major GC related performance regression
"The regression" is actually due to that 2.2.1 had a bug that prevented
the regression (which isn't a regression at all) from showing up. "The
regression" is actually a glitch in cyclic gc that's been there forever.
As the generation being collected is analyzed, objects that can't be
collected (because, e.g., we find they're externally referenced, or
are in an unreachable cycle but have a __del__ method) are moved out
of the list of candidates. A tricksy scheme uses negative values of
gc_refs to mark such objects as being moved. However, the exact
negative value set at the start may become "more negative" over time
for objects not in the generation being collected, and the scheme was
checking for an exact match on the negative value originally assigned.
As a result, objects in generations older than the one being collected
could get scanned too, and yanked back into a younger generation. Doing
so doesn't lead to an error, but doesn't do any good, and can burn an
unbounded amount of time doing useless work.
A test case is simple (thanks to Kevin Jacobs for finding it!):
x = []
for i in xrange(200000):
x.append((1,))
Without the patch, this ends up scanning all of x on every gen0 collection,
scans all of x twice on every gen1 collection, and x gets yanked back into
gen1 on every gen0 collection. With the patch, once x gets to gen2, it's
never scanned again until another gen2 collection, and stays in gen2.
Bugfix candidate, although the code has changed enough that I think I'll
need to port it by hand. 2.2.1 also has a different bug that causes
bound method objects not to get tracked at all (so the test case doesn't
burn absurd amounts of time in 2.2.1, but *should* <wink>).
2002-07-01 01:56:40 +08:00
|
|
|
/* return true if object has a finalization method */
|
2001-11-02 01:35:23 +08:00
|
|
|
static int
|
|
|
|
has_finalizer(PyObject *op)
|
2000-06-30 13:02:53 +08:00
|
|
|
{
|
2000-08-31 23:10:24 +08:00
|
|
|
static PyObject *delstr = NULL;
|
2000-06-30 13:02:53 +08:00
|
|
|
if (delstr == NULL) {
|
|
|
|
delstr = PyString_InternFromString("__del__");
|
2000-08-31 23:10:24 +08:00
|
|
|
if (delstr == NULL)
|
|
|
|
Py_FatalError("PyGC: can't initialize __del__ string");
|
2000-06-30 13:02:53 +08:00
|
|
|
}
|
2001-11-02 03:35:45 +08:00
|
|
|
return (PyInstance_Check(op) ||
|
|
|
|
PyType_HasFeature(op->ob_type, Py_TPFLAGS_HEAPTYPE))
|
|
|
|
&& PyObject_HasAttr(op, delstr);
|
2001-11-02 01:35:23 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Move all objects with finalizers (instances with __del__) */
|
|
|
|
static void
|
|
|
|
move_finalizers(PyGC_Head *unreachable, PyGC_Head *finalizers)
|
|
|
|
{
|
|
|
|
PyGC_Head *next;
|
|
|
|
PyGC_Head *gc = unreachable->gc.gc_next;
|
2000-06-30 13:02:53 +08:00
|
|
|
for (; gc != unreachable; gc=next) {
|
2001-08-30 08:05:51 +08:00
|
|
|
PyObject *op = FROM_GC(gc);
|
2001-10-12 02:31:31 +08:00
|
|
|
next = gc->gc.gc_next;
|
2001-11-02 01:35:23 +08:00
|
|
|
if (has_finalizer(op)) {
|
2000-06-30 13:02:53 +08:00
|
|
|
gc_list_remove(gc);
|
|
|
|
gc_list_append(gc, finalizers);
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
gc->gc.gc_refs = GC_REACHABLE;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* A traversal callback for move_finalizer_reachable. */
|
|
|
|
static int
|
|
|
|
visit_move(PyObject *op, PyGC_Head *tolist)
|
|
|
|
{
|
|
|
|
if (PyObject_IS_GC(op)) {
|
2002-07-02 08:52:30 +08:00
|
|
|
if (IS_TENTATIVELY_UNREACHABLE(op)) {
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
PyGC_Head *gc = AS_GC(op);
|
|
|
|
gc_list_remove(gc);
|
|
|
|
gc_list_append(gc, tolist);
|
|
|
|
gc->gc.gc_refs = GC_REACHABLE;
|
2000-06-30 13:02:53 +08:00
|
|
|
}
|
|
|
|
}
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
return 0;
|
2000-06-30 13:02:53 +08:00
|
|
|
}
|
|
|
|
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
/* Move objects that are reachable from finalizers, from the unreachable set
|
|
|
|
* into the finalizers set.
|
|
|
|
*/
|
2000-06-30 13:02:53 +08:00
|
|
|
static void
|
|
|
|
move_finalizer_reachable(PyGC_Head *finalizers)
|
|
|
|
{
|
|
|
|
traverseproc traverse;
|
2001-10-12 02:31:31 +08:00
|
|
|
PyGC_Head *gc = finalizers->gc.gc_next;
|
|
|
|
for (; gc != finalizers; gc=gc->gc.gc_next) {
|
2000-06-30 13:02:53 +08:00
|
|
|
/* careful, finalizers list is growing here */
|
2001-08-30 08:05:51 +08:00
|
|
|
traverse = FROM_GC(gc)->ob_type->tp_traverse;
|
SF bug #574132: Major GC related performance regression
"The regression" is actually due to that 2.2.1 had a bug that prevented
the regression (which isn't a regression at all) from showing up. "The
regression" is actually a glitch in cyclic gc that's been there forever.
As the generation being collected is analyzed, objects that can't be
collected (because, e.g., we find they're externally referenced, or
are in an unreachable cycle but have a __del__ method) are moved out
of the list of candidates. A tricksy scheme uses negative values of
gc_refs to mark such objects as being moved. However, the exact
negative value set at the start may become "more negative" over time
for objects not in the generation being collected, and the scheme was
checking for an exact match on the negative value originally assigned.
As a result, objects in generations older than the one being collected
could get scanned too, and yanked back into a younger generation. Doing
so doesn't lead to an error, but doesn't do any good, and can burn an
unbounded amount of time doing useless work.
A test case is simple (thanks to Kevin Jacobs for finding it!):
x = []
for i in xrange(200000):
x.append((1,))
Without the patch, this ends up scanning all of x on every gen0 collection,
scans all of x twice on every gen1 collection, and x gets yanked back into
gen1 on every gen0 collection. With the patch, once x gets to gen2, it's
never scanned again until another gen2 collection, and stays in gen2.
Bugfix candidate, although the code has changed enough that I think I'll
need to port it by hand. 2.2.1 also has a different bug that causes
bound method objects not to get tracked at all (so the test case doesn't
burn absurd amounts of time in 2.2.1, but *should* <wink>).
2002-07-01 01:56:40 +08:00
|
|
|
(void) traverse(FROM_GC(gc),
|
2001-08-30 08:05:51 +08:00
|
|
|
(visitproc)visit_move,
|
2000-06-30 13:02:53 +08:00
|
|
|
(void *)finalizers);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
2000-08-31 23:10:24 +08:00
|
|
|
debug_instance(char *msg, PyInstanceObject *inst)
|
2000-06-30 13:02:53 +08:00
|
|
|
{
|
|
|
|
char *cname;
|
2001-11-02 01:35:23 +08:00
|
|
|
/* simple version of instance_repr */
|
2000-06-30 13:02:53 +08:00
|
|
|
PyObject *classname = inst->in_class->cl_name;
|
|
|
|
if (classname != NULL && PyString_Check(classname))
|
|
|
|
cname = PyString_AsString(classname);
|
|
|
|
else
|
|
|
|
cname = "?";
|
2000-08-31 23:10:24 +08:00
|
|
|
PySys_WriteStderr("gc: %.100s <%.100s instance at %p>\n",
|
|
|
|
msg, cname, inst);
|
2000-06-30 13:02:53 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
2000-08-31 23:10:24 +08:00
|
|
|
debug_cycle(char *msg, PyObject *op)
|
2000-06-30 13:02:53 +08:00
|
|
|
{
|
|
|
|
if ((debug & DEBUG_INSTANCES) && PyInstance_Check(op)) {
|
2000-08-31 23:10:24 +08:00
|
|
|
debug_instance(msg, (PyInstanceObject *)op);
|
2000-09-22 23:22:38 +08:00
|
|
|
}
|
|
|
|
else if (debug & DEBUG_OBJECTS) {
|
2000-08-31 23:10:24 +08:00
|
|
|
PySys_WriteStderr("gc: %.100s <%.100s %p>\n",
|
|
|
|
msg, op->ob_type->tp_name, op);
|
2000-06-30 13:02:53 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Handle uncollectable garbage (cycles with finalizers). */
|
|
|
|
static void
|
|
|
|
handle_finalizers(PyGC_Head *finalizers, PyGC_Head *old)
|
|
|
|
{
|
|
|
|
PyGC_Head *gc;
|
|
|
|
if (garbage == NULL) {
|
|
|
|
garbage = PyList_New(0);
|
|
|
|
}
|
2001-10-12 02:31:31 +08:00
|
|
|
for (gc = finalizers->gc.gc_next; gc != finalizers;
|
|
|
|
gc = finalizers->gc.gc_next) {
|
2001-08-30 08:05:51 +08:00
|
|
|
PyObject *op = FROM_GC(gc);
|
2001-11-02 01:35:23 +08:00
|
|
|
if ((debug & DEBUG_SAVEALL) || has_finalizer(op)) {
|
|
|
|
/* If SAVEALL is not set then just append objects with
|
|
|
|
* finalizers to the list of garbage. All objects in
|
|
|
|
* the finalizers list are reachable from those
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
* objects.
|
|
|
|
*/
|
2000-06-30 13:02:53 +08:00
|
|
|
PyList_Append(garbage, op);
|
|
|
|
}
|
SF bug #574132: Major GC related performance regression
"The regression" is actually due to that 2.2.1 had a bug that prevented
the regression (which isn't a regression at all) from showing up. "The
regression" is actually a glitch in cyclic gc that's been there forever.
As the generation being collected is analyzed, objects that can't be
collected (because, e.g., we find they're externally referenced, or
are in an unreachable cycle but have a __del__ method) are moved out
of the list of candidates. A tricksy scheme uses negative values of
gc_refs to mark such objects as being moved. However, the exact
negative value set at the start may become "more negative" over time
for objects not in the generation being collected, and the scheme was
checking for an exact match on the negative value originally assigned.
As a result, objects in generations older than the one being collected
could get scanned too, and yanked back into a younger generation. Doing
so doesn't lead to an error, but doesn't do any good, and can burn an
unbounded amount of time doing useless work.
A test case is simple (thanks to Kevin Jacobs for finding it!):
x = []
for i in xrange(200000):
x.append((1,))
Without the patch, this ends up scanning all of x on every gen0 collection,
scans all of x twice on every gen1 collection, and x gets yanked back into
gen1 on every gen0 collection. With the patch, once x gets to gen2, it's
never scanned again until another gen2 collection, and stays in gen2.
Bugfix candidate, although the code has changed enough that I think I'll
need to port it by hand. 2.2.1 also has a different bug that causes
bound method objects not to get tracked at all (so the test case doesn't
burn absurd amounts of time in 2.2.1, but *should* <wink>).
2002-07-01 01:56:40 +08:00
|
|
|
/* object is now reachable again */
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
assert(IS_REACHABLE(op));
|
2000-06-30 13:02:53 +08:00
|
|
|
gc_list_remove(gc);
|
|
|
|
gc_list_append(gc, old);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2000-09-22 23:22:38 +08:00
|
|
|
/* Break reference cycles by clearing the containers involved. This is
|
2000-06-30 13:02:53 +08:00
|
|
|
* tricky business as the lists can be changing and we don't know which
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
* objects may be freed. It is possible I screwed something up here.
|
|
|
|
*/
|
2000-06-30 13:02:53 +08:00
|
|
|
static void
|
|
|
|
delete_garbage(PyGC_Head *unreachable, PyGC_Head *old)
|
|
|
|
{
|
|
|
|
inquiry clear;
|
|
|
|
|
2002-05-04 13:35:20 +08:00
|
|
|
while (!gc_list_is_empty(unreachable)) {
|
2001-10-12 02:31:31 +08:00
|
|
|
PyGC_Head *gc = unreachable->gc.gc_next;
|
2001-08-30 08:05:51 +08:00
|
|
|
PyObject *op = FROM_GC(gc);
|
SF bug #574132: Major GC related performance regression
"The regression" is actually due to that 2.2.1 had a bug that prevented
the regression (which isn't a regression at all) from showing up. "The
regression" is actually a glitch in cyclic gc that's been there forever.
As the generation being collected is analyzed, objects that can't be
collected (because, e.g., we find they're externally referenced, or
are in an unreachable cycle but have a __del__ method) are moved out
of the list of candidates. A tricksy scheme uses negative values of
gc_refs to mark such objects as being moved. However, the exact
negative value set at the start may become "more negative" over time
for objects not in the generation being collected, and the scheme was
checking for an exact match on the negative value originally assigned.
As a result, objects in generations older than the one being collected
could get scanned too, and yanked back into a younger generation. Doing
so doesn't lead to an error, but doesn't do any good, and can burn an
unbounded amount of time doing useless work.
A test case is simple (thanks to Kevin Jacobs for finding it!):
x = []
for i in xrange(200000):
x.append((1,))
Without the patch, this ends up scanning all of x on every gen0 collection,
scans all of x twice on every gen1 collection, and x gets yanked back into
gen1 on every gen0 collection. With the patch, once x gets to gen2, it's
never scanned again until another gen2 collection, and stays in gen2.
Bugfix candidate, although the code has changed enough that I think I'll
need to port it by hand. 2.2.1 also has a different bug that causes
bound method objects not to get tracked at all (so the test case doesn't
burn absurd amounts of time in 2.2.1, but *should* <wink>).
2002-07-01 01:56:40 +08:00
|
|
|
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
assert(IS_TENTATIVELY_UNREACHABLE(op));
|
2000-09-22 23:22:38 +08:00
|
|
|
if (debug & DEBUG_SAVEALL) {
|
|
|
|
PyList_Append(garbage, op);
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
if ((clear = op->ob_type->tp_clear) != NULL) {
|
|
|
|
Py_INCREF(op);
|
2002-06-07 07:23:55 +08:00
|
|
|
clear(op);
|
2000-09-22 23:22:38 +08:00
|
|
|
Py_DECREF(op);
|
|
|
|
}
|
2000-06-30 13:02:53 +08:00
|
|
|
}
|
2001-10-12 02:31:31 +08:00
|
|
|
if (unreachable->gc.gc_next == gc) {
|
2000-09-22 23:22:38 +08:00
|
|
|
/* object is still alive, move it, it may die later */
|
2000-06-30 13:02:53 +08:00
|
|
|
gc_list_remove(gc);
|
|
|
|
gc_list_append(gc, old);
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
gc->gc.gc_refs = GC_REACHABLE;
|
2000-06-30 13:02:53 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* This is the main function. Read this to understand how the
|
|
|
|
* collection process works. */
|
|
|
|
static long
|
2002-05-04 13:35:20 +08:00
|
|
|
collect(int generation)
|
2000-06-30 13:02:53 +08:00
|
|
|
{
|
2002-05-04 13:35:20 +08:00
|
|
|
int i;
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
long m = 0; /* # objects collected */
|
|
|
|
long n = 0; /* # unreachable objects that couldn't be collected */
|
2002-05-04 13:35:20 +08:00
|
|
|
PyGC_Head *young; /* the generation we are examining */
|
|
|
|
PyGC_Head *old; /* next older generation */
|
2000-06-30 13:02:53 +08:00
|
|
|
PyGC_Head unreachable;
|
|
|
|
PyGC_Head finalizers;
|
|
|
|
PyGC_Head *gc;
|
|
|
|
|
|
|
|
if (debug & DEBUG_STATS) {
|
2002-05-04 13:35:20 +08:00
|
|
|
PySys_WriteStderr("gc: collecting generation %d...\n",
|
|
|
|
generation);
|
|
|
|
PySys_WriteStderr("gc: objects in each generation:");
|
|
|
|
for (i = 0; i < NUM_GENERATIONS; i++) {
|
|
|
|
PySys_WriteStderr(" %ld", gc_list_size(GEN_HEAD(i)));
|
|
|
|
}
|
|
|
|
PySys_WriteStderr("\n");
|
|
|
|
}
|
|
|
|
|
|
|
|
/* update collection and allocation counters */
|
|
|
|
if (generation+1 < NUM_GENERATIONS)
|
|
|
|
generations[generation+1].count += 1;
|
|
|
|
for (i = 0; i <= generation; i++)
|
2002-06-29 03:16:04 +08:00
|
|
|
generations[i].count = 0;
|
2002-05-04 13:35:20 +08:00
|
|
|
|
|
|
|
/* merge younger generations with one we are currently collecting */
|
|
|
|
for (i = 0; i < generation; i++) {
|
|
|
|
gc_list_merge(GEN_HEAD(i), GEN_HEAD(generation));
|
|
|
|
}
|
|
|
|
|
|
|
|
/* handy references */
|
|
|
|
young = GEN_HEAD(generation);
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
if (generation < NUM_GENERATIONS-1)
|
2002-05-04 13:35:20 +08:00
|
|
|
old = GEN_HEAD(generation+1);
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
else
|
|
|
|
old = young;
|
2000-06-30 13:02:53 +08:00
|
|
|
|
|
|
|
/* Using ob_refcnt and gc_refs, calculate which objects in the
|
|
|
|
* container set are reachable from outside the set (ie. have a
|
|
|
|
* refcount greater than 0 when all the references within the
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
* set are taken into account
|
|
|
|
*/
|
2000-06-30 13:02:53 +08:00
|
|
|
update_refs(young);
|
|
|
|
subtract_refs(young);
|
|
|
|
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
/* Leave everything reachable from outside young in young, and move
|
|
|
|
* everything else (in young) to unreachable.
|
|
|
|
* NOTE: This used to move the reachable objects into a reachable
|
|
|
|
* set instead. But most things usually turn out to be reachable,
|
|
|
|
* so it's more efficient to move the unreachable things.
|
|
|
|
*/
|
2000-06-30 13:02:53 +08:00
|
|
|
gc_list_init(&unreachable);
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
move_unreachable(young, &unreachable);
|
2000-06-30 13:02:53 +08:00
|
|
|
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
/* Move reachable objects to next generation. */
|
|
|
|
if (young != old)
|
|
|
|
gc_list_merge(young, old);
|
2000-06-30 13:02:53 +08:00
|
|
|
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
/* All objects in unreachable are trash, but objects reachable from
|
|
|
|
* finalizers can't safely be deleted. Python programmers should take
|
|
|
|
* care not to create such things. For Python, finalizers means
|
|
|
|
* instance objects with __del__ methods.
|
|
|
|
*/
|
2000-06-30 13:02:53 +08:00
|
|
|
gc_list_init(&finalizers);
|
|
|
|
move_finalizers(&unreachable, &finalizers);
|
|
|
|
move_finalizer_reachable(&finalizers);
|
|
|
|
|
|
|
|
/* Collect statistics on collectable objects found and print
|
|
|
|
* debugging information. */
|
2001-10-12 02:31:31 +08:00
|
|
|
for (gc = unreachable.gc.gc_next; gc != &unreachable;
|
|
|
|
gc = gc->gc.gc_next) {
|
2000-06-30 13:02:53 +08:00
|
|
|
m++;
|
2000-08-31 23:10:24 +08:00
|
|
|
if (debug & DEBUG_COLLECTABLE) {
|
2001-08-30 08:05:51 +08:00
|
|
|
debug_cycle("collectable", FROM_GC(gc));
|
2000-06-30 13:02:53 +08:00
|
|
|
}
|
|
|
|
}
|
OK, I couldn't stand it <0.5 wink>: removed all uncertainty about what's
in gc_refs, even at the cost of putting back a test+branch in
visit_decref.
The good news: since gc_refs became utterly tame then, it became
clear that another special value could be useful. The move_roots() and
move_root_reachable() passes have now been replaced by a single
move_unreachable() pass. Besides saving a pass over the generation, this
has a better effect: most of the time everything turns out to be
reachable, so we were breaking the generation list apart and moving it
into into the reachable list, one element at a time. Now the reachable
stuff stays in the generation list, and the unreachable stuff is moved
instead. This isn't quite as good as it sounds, since sometimes we
guess wrongly that a thing is unreachable, and have to move it back again.
Still, overall, it yields a significant (but not dramatic) boost in
collection speed.
2002-07-01 11:52:19 +08:00
|
|
|
/* Call tp_clear on objects in the collectable set. This will cause
|
2000-06-30 13:02:53 +08:00
|
|
|
* the reference cycles to be broken. It may also cause some objects in
|
|
|
|
* finalizers to be freed */
|
|
|
|
delete_garbage(&unreachable, old);
|
|
|
|
|
|
|
|
/* Collect statistics on uncollectable objects found and print
|
|
|
|
* debugging information. */
|
2001-10-12 02:31:31 +08:00
|
|
|
for (gc = finalizers.gc.gc_next; gc != &finalizers;
|
|
|
|
gc = gc->gc.gc_next) {
|
2000-06-30 13:02:53 +08:00
|
|
|
n++;
|
2000-08-31 23:10:24 +08:00
|
|
|
if (debug & DEBUG_UNCOLLECTABLE) {
|
2001-08-30 08:05:51 +08:00
|
|
|
debug_cycle("uncollectable", FROM_GC(gc));
|
2000-06-30 13:02:53 +08:00
|
|
|
}
|
|
|
|
}
|
2000-08-31 23:10:24 +08:00
|
|
|
if (debug & DEBUG_STATS) {
|
2000-06-30 13:02:53 +08:00
|
|
|
if (m == 0 && n == 0) {
|
2000-08-31 23:10:24 +08:00
|
|
|
PySys_WriteStderr("gc: done.\n");
|
2000-09-22 23:22:38 +08:00
|
|
|
}
|
|
|
|
else {
|
2000-08-31 23:10:24 +08:00
|
|
|
PySys_WriteStderr(
|
|
|
|
"gc: done, %ld unreachable, %ld uncollectable.\n",
|
|
|
|
n+m, n);
|
2000-06-30 13:02:53 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Append instances in the uncollectable set to a Python
|
|
|
|
* reachable list of garbage. The programmer has to deal with
|
|
|
|
* this if they insist on creating this type of structure. */
|
|
|
|
handle_finalizers(&finalizers, old);
|
|
|
|
|
2000-09-01 10:47:25 +08:00
|
|
|
if (PyErr_Occurred()) {
|
2000-09-22 23:22:38 +08:00
|
|
|
if (gc_str == NULL) {
|
|
|
|
gc_str = PyString_FromString("garbage collection");
|
|
|
|
}
|
2000-09-01 10:47:25 +08:00
|
|
|
PyErr_WriteUnraisable(gc_str);
|
|
|
|
Py_FatalError("unexpected exception during garbage collection");
|
|
|
|
}
|
2000-06-30 13:02:53 +08:00
|
|
|
return n+m;
|
|
|
|
}
|
|
|
|
|
|
|
|
static long
|
|
|
|
collect_generations(void)
|
|
|
|
{
|
2002-05-04 13:35:20 +08:00
|
|
|
int i;
|
2000-07-10 13:37:39 +08:00
|
|
|
long n = 0;
|
2000-06-30 13:02:53 +08:00
|
|
|
|
2002-05-04 13:35:20 +08:00
|
|
|
/* Find the oldest generation (higest numbered) where the count
|
|
|
|
* exceeds the threshold. Objects in the that generation and
|
|
|
|
* generations younger than it will be collected. */
|
|
|
|
for (i = NUM_GENERATIONS-1; i >= 0; i--) {
|
|
|
|
if (generations[i].count > generations[i].threshold) {
|
|
|
|
n = collect(i);
|
|
|
|
break;
|
2000-06-30 13:02:53 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
return n;
|
|
|
|
}
|
|
|
|
|
2002-06-14 04:33:02 +08:00
|
|
|
PyDoc_STRVAR(gc_enable__doc__,
|
2000-08-07 06:45:31 +08:00
|
|
|
"enable() -> None\n"
|
|
|
|
"\n"
|
2002-06-14 04:33:02 +08:00
|
|
|
"Enable automatic garbage collection.\n");
|
2000-08-07 06:45:31 +08:00
|
|
|
|
|
|
|
static PyObject *
|
|
|
|
gc_enable(PyObject *self, PyObject *args)
|
|
|
|
{
|
|
|
|
|
|
|
|
if (!PyArg_ParseTuple(args, ":enable")) /* check no args */
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
enabled = 1;
|
|
|
|
|
|
|
|
Py_INCREF(Py_None);
|
|
|
|
return Py_None;
|
|
|
|
}
|
|
|
|
|
2002-06-14 04:33:02 +08:00
|
|
|
PyDoc_STRVAR(gc_disable__doc__,
|
2000-08-07 06:45:31 +08:00
|
|
|
"disable() -> None\n"
|
|
|
|
"\n"
|
2002-06-14 04:33:02 +08:00
|
|
|
"Disable automatic garbage collection.\n");
|
2000-08-07 06:45:31 +08:00
|
|
|
|
|
|
|
static PyObject *
|
|
|
|
gc_disable(PyObject *self, PyObject *args)
|
|
|
|
{
|
|
|
|
|
|
|
|
if (!PyArg_ParseTuple(args, ":disable")) /* check no args */
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
enabled = 0;
|
|
|
|
|
|
|
|
Py_INCREF(Py_None);
|
|
|
|
return Py_None;
|
|
|
|
}
|
|
|
|
|
2002-06-14 04:33:02 +08:00
|
|
|
PyDoc_STRVAR(gc_isenabled__doc__,
|
2000-08-07 06:45:31 +08:00
|
|
|
"isenabled() -> status\n"
|
|
|
|
"\n"
|
2002-06-14 04:33:02 +08:00
|
|
|
"Returns true if automatic garbage collection is enabled.\n");
|
2000-08-07 06:45:31 +08:00
|
|
|
|
|
|
|
static PyObject *
|
|
|
|
gc_isenabled(PyObject *self, PyObject *args)
|
|
|
|
{
|
|
|
|
|
|
|
|
if (!PyArg_ParseTuple(args, ":isenabled")) /* check no args */
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
return Py_BuildValue("i", enabled);
|
|
|
|
}
|
2000-06-30 13:02:53 +08:00
|
|
|
|
2002-06-14 04:33:02 +08:00
|
|
|
PyDoc_STRVAR(gc_collect__doc__,
|
2000-06-30 13:02:53 +08:00
|
|
|
"collect() -> n\n"
|
|
|
|
"\n"
|
2002-06-14 04:33:02 +08:00
|
|
|
"Run a full collection. The number of unreachable objects is returned.\n");
|
2000-06-30 13:02:53 +08:00
|
|
|
|
|
|
|
static PyObject *
|
2000-08-07 06:45:31 +08:00
|
|
|
gc_collect(PyObject *self, PyObject *args)
|
2000-06-30 13:02:53 +08:00
|
|
|
{
|
|
|
|
long n;
|
|
|
|
|
2000-07-12 12:42:23 +08:00
|
|
|
if (!PyArg_ParseTuple(args, ":collect")) /* check no args */
|
2000-06-30 13:02:53 +08:00
|
|
|
return NULL;
|
|
|
|
|
2001-11-01 07:09:35 +08:00
|
|
|
if (collecting) {
|
|
|
|
n = 0; /* already collecting, don't do anything */
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
collecting = 1;
|
2002-05-04 13:35:20 +08:00
|
|
|
n = collect(NUM_GENERATIONS - 1);
|
2001-11-01 07:09:35 +08:00
|
|
|
collecting = 0;
|
|
|
|
}
|
2000-06-30 13:02:53 +08:00
|
|
|
|
2000-09-23 06:35:36 +08:00
|
|
|
return Py_BuildValue("l", n);
|
2000-06-30 13:02:53 +08:00
|
|
|
}
|
|
|
|
|
2002-06-14 04:33:02 +08:00
|
|
|
PyDoc_STRVAR(gc_set_debug__doc__,
|
2000-06-30 13:02:53 +08:00
|
|
|
"set_debug(flags) -> None\n"
|
|
|
|
"\n"
|
|
|
|
"Set the garbage collection debugging flags. Debugging information is\n"
|
|
|
|
"written to sys.stderr.\n"
|
|
|
|
"\n"
|
|
|
|
"flags is an integer and can have the following bits turned on:\n"
|
|
|
|
"\n"
|
|
|
|
" DEBUG_STATS - Print statistics during collection.\n"
|
|
|
|
" DEBUG_COLLECTABLE - Print collectable objects found.\n"
|
|
|
|
" DEBUG_UNCOLLECTABLE - Print unreachable but uncollectable objects found.\n"
|
|
|
|
" DEBUG_INSTANCES - Print instance objects.\n"
|
|
|
|
" DEBUG_OBJECTS - Print objects other than instances.\n"
|
2000-09-22 23:22:38 +08:00
|
|
|
" DEBUG_SAVEALL - Save objects to gc.garbage rather than freeing them.\n"
|
2002-06-14 04:33:02 +08:00
|
|
|
" DEBUG_LEAK - Debug leaking programs (everything but STATS).\n");
|
2000-06-30 13:02:53 +08:00
|
|
|
|
|
|
|
static PyObject *
|
2000-08-07 06:45:31 +08:00
|
|
|
gc_set_debug(PyObject *self, PyObject *args)
|
2000-06-30 13:02:53 +08:00
|
|
|
{
|
2000-09-23 06:35:36 +08:00
|
|
|
if (!PyArg_ParseTuple(args, "i:set_debug", &debug))
|
2000-06-30 13:02:53 +08:00
|
|
|
return NULL;
|
|
|
|
|
|
|
|
Py_INCREF(Py_None);
|
|
|
|
return Py_None;
|
|
|
|
}
|
|
|
|
|
2002-06-14 04:33:02 +08:00
|
|
|
PyDoc_STRVAR(gc_get_debug__doc__,
|
2000-06-30 13:02:53 +08:00
|
|
|
"get_debug() -> flags\n"
|
|
|
|
"\n"
|
2002-06-14 04:33:02 +08:00
|
|
|
"Get the garbage collection debugging flags.\n");
|
2000-06-30 13:02:53 +08:00
|
|
|
|
|
|
|
static PyObject *
|
2000-08-07 06:45:31 +08:00
|
|
|
gc_get_debug(PyObject *self, PyObject *args)
|
2000-06-30 13:02:53 +08:00
|
|
|
{
|
2000-07-12 12:42:23 +08:00
|
|
|
if (!PyArg_ParseTuple(args, ":get_debug")) /* no args */
|
2000-06-30 13:02:53 +08:00
|
|
|
return NULL;
|
|
|
|
|
|
|
|
return Py_BuildValue("i", debug);
|
|
|
|
}
|
|
|
|
|
2002-06-14 04:33:02 +08:00
|
|
|
PyDoc_STRVAR(gc_set_thresh__doc__,
|
2002-01-29 08:53:41 +08:00
|
|
|
"set_threshold(threshold0, [threshold1, threshold2]) -> None\n"
|
2000-06-30 13:02:53 +08:00
|
|
|
"\n"
|
|
|
|
"Sets the collection thresholds. Setting threshold0 to zero disables\n"
|
2002-06-14 04:33:02 +08:00
|
|
|
"collection.\n");
|
2000-06-30 13:02:53 +08:00
|
|
|
|
|
|
|
static PyObject *
|
2000-08-07 06:45:31 +08:00
|
|
|
gc_set_thresh(PyObject *self, PyObject *args)
|
2000-06-30 13:02:53 +08:00
|
|
|
{
|
2002-05-04 13:35:20 +08:00
|
|
|
int i;
|
|
|
|
if (!PyArg_ParseTuple(args, "i|ii:set_threshold",
|
|
|
|
&generations[0].threshold,
|
|
|
|
&generations[1].threshold,
|
|
|
|
&generations[2].threshold))
|
2000-06-30 13:02:53 +08:00
|
|
|
return NULL;
|
2002-05-04 13:35:20 +08:00
|
|
|
for (i = 2; i < NUM_GENERATIONS; i++) {
|
|
|
|
/* generations higher than 2 get the same threshold */
|
|
|
|
generations[i].threshold = generations[2].threshold;
|
|
|
|
}
|
2000-06-30 13:02:53 +08:00
|
|
|
|
|
|
|
Py_INCREF(Py_None);
|
|
|
|
return Py_None;
|
|
|
|
}
|
|
|
|
|
2002-06-14 04:33:02 +08:00
|
|
|
PyDoc_STRVAR(gc_get_thresh__doc__,
|
2000-06-30 13:02:53 +08:00
|
|
|
"get_threshold() -> (threshold0, threshold1, threshold2)\n"
|
|
|
|
"\n"
|
2002-06-14 04:33:02 +08:00
|
|
|
"Return the current collection thresholds\n");
|
2000-06-30 13:02:53 +08:00
|
|
|
|
|
|
|
static PyObject *
|
2000-08-07 06:45:31 +08:00
|
|
|
gc_get_thresh(PyObject *self, PyObject *args)
|
2000-06-30 13:02:53 +08:00
|
|
|
{
|
2000-07-12 12:42:23 +08:00
|
|
|
if (!PyArg_ParseTuple(args, ":get_threshold")) /* no args */
|
2000-06-30 13:02:53 +08:00
|
|
|
return NULL;
|
|
|
|
|
2002-05-04 13:35:20 +08:00
|
|
|
return Py_BuildValue("(iii)",
|
|
|
|
generations[0].threshold,
|
|
|
|
generations[1].threshold,
|
|
|
|
generations[2].threshold);
|
2000-06-30 13:02:53 +08:00
|
|
|
}
|
|
|
|
|
2001-08-09 23:38:31 +08:00
|
|
|
static int
|
2001-11-24 17:24:51 +08:00
|
|
|
referrersvisit(PyObject* obj, PyObject *objs)
|
2001-08-09 23:38:31 +08:00
|
|
|
{
|
2001-11-30 02:08:31 +08:00
|
|
|
int i;
|
|
|
|
for (i = 0; i < PyTuple_GET_SIZE(objs); i++)
|
|
|
|
if (PyTuple_GET_ITEM(objs, i) == obj)
|
|
|
|
return 1;
|
2001-08-09 23:38:31 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2001-08-10 22:46:47 +08:00
|
|
|
static int
|
2001-11-24 17:24:51 +08:00
|
|
|
gc_referrers_for(PyObject *objs, PyGC_Head *list, PyObject *resultlist)
|
2001-08-09 23:38:31 +08:00
|
|
|
{
|
|
|
|
PyGC_Head *gc;
|
|
|
|
PyObject *obj;
|
|
|
|
traverseproc traverse;
|
2001-10-12 02:31:31 +08:00
|
|
|
for (gc = list->gc.gc_next; gc != list; gc = gc->gc.gc_next) {
|
2001-08-30 08:05:51 +08:00
|
|
|
obj = FROM_GC(gc);
|
2001-08-09 23:38:31 +08:00
|
|
|
traverse = obj->ob_type->tp_traverse;
|
|
|
|
if (obj == objs || obj == resultlist)
|
|
|
|
continue;
|
2001-11-24 17:24:51 +08:00
|
|
|
if (traverse(obj, (visitproc)referrersvisit, objs)) {
|
2001-08-10 22:46:47 +08:00
|
|
|
if (PyList_Append(resultlist, obj) < 0)
|
|
|
|
return 0; /* error */
|
2001-08-09 23:38:31 +08:00
|
|
|
}
|
|
|
|
}
|
2001-08-10 22:46:47 +08:00
|
|
|
return 1; /* no error */
|
2001-08-09 23:38:31 +08:00
|
|
|
}
|
|
|
|
|
2002-06-14 04:33:02 +08:00
|
|
|
PyDoc_STRVAR(gc_get_referrers__doc__,
|
2001-11-24 17:24:51 +08:00
|
|
|
"get_referrers(*objs) -> list\n\
|
2002-06-14 04:33:02 +08:00
|
|
|
Return the list of objects that directly refer to any of objs.");
|
2001-08-09 23:38:31 +08:00
|
|
|
|
2001-08-10 22:46:47 +08:00
|
|
|
static PyObject *
|
2001-11-24 17:24:51 +08:00
|
|
|
gc_get_referrers(PyObject *self, PyObject *args)
|
2001-08-09 23:38:31 +08:00
|
|
|
{
|
2002-05-04 13:35:20 +08:00
|
|
|
int i;
|
2001-08-09 23:38:31 +08:00
|
|
|
PyObject *result = PyList_New(0);
|
2002-05-04 13:35:20 +08:00
|
|
|
for (i = 0; i < NUM_GENERATIONS; i++) {
|
|
|
|
if (!(gc_referrers_for(args, GEN_HEAD(i), result))) {
|
|
|
|
Py_DECREF(result);
|
|
|
|
return NULL;
|
|
|
|
}
|
2001-08-10 22:46:47 +08:00
|
|
|
}
|
2001-08-09 23:38:31 +08:00
|
|
|
return result;
|
|
|
|
}
|
2000-06-30 13:02:53 +08:00
|
|
|
|
2002-06-14 04:33:02 +08:00
|
|
|
PyDoc_STRVAR(gc_get_objects__doc__,
|
2001-08-09 23:58:59 +08:00
|
|
|
"get_objects() -> [...]\n"
|
|
|
|
"\n"
|
|
|
|
"Return a list of objects tracked by the collector (excluding the list\n"
|
2002-06-14 04:33:02 +08:00
|
|
|
"returned).\n");
|
2001-08-09 23:58:59 +08:00
|
|
|
|
|
|
|
/* appending objects in a GC list to a Python list */
|
2001-12-02 20:21:34 +08:00
|
|
|
static int
|
2001-08-09 23:58:59 +08:00
|
|
|
append_objects(PyObject *py_list, PyGC_Head *gc_list)
|
|
|
|
{
|
|
|
|
PyGC_Head *gc;
|
2001-10-12 02:31:31 +08:00
|
|
|
for (gc = gc_list->gc.gc_next; gc != gc_list; gc = gc->gc.gc_next) {
|
2001-08-30 08:05:51 +08:00
|
|
|
PyObject *op = FROM_GC(gc);
|
2001-08-09 23:58:59 +08:00
|
|
|
if (op != py_list) {
|
2001-12-02 20:21:34 +08:00
|
|
|
if (PyList_Append(py_list, op)) {
|
|
|
|
return -1; /* exception */
|
|
|
|
}
|
2001-08-09 23:58:59 +08:00
|
|
|
}
|
|
|
|
}
|
2001-12-02 20:21:34 +08:00
|
|
|
return 0;
|
2001-08-09 23:58:59 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static PyObject *
|
|
|
|
gc_get_objects(PyObject *self, PyObject *args)
|
|
|
|
{
|
2002-05-04 13:35:20 +08:00
|
|
|
int i;
|
2001-08-09 23:58:59 +08:00
|
|
|
PyObject* result;
|
|
|
|
|
|
|
|
if (!PyArg_ParseTuple(args, ":get_objects")) /* check no args */
|
|
|
|
return NULL;
|
|
|
|
result = PyList_New(0);
|
2001-12-03 02:31:02 +08:00
|
|
|
if (result == NULL) {
|
|
|
|
return NULL;
|
|
|
|
}
|
2002-05-04 13:35:20 +08:00
|
|
|
for (i = 0; i < NUM_GENERATIONS; i++) {
|
|
|
|
if (append_objects(result, GEN_HEAD(i))) {
|
|
|
|
Py_DECREF(result);
|
|
|
|
return NULL;
|
|
|
|
}
|
2001-12-02 20:21:34 +08:00
|
|
|
}
|
2001-08-09 23:58:59 +08:00
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2002-06-14 04:33:02 +08:00
|
|
|
PyDoc_STRVAR(gc__doc__,
|
2000-06-30 13:02:53 +08:00
|
|
|
"This module provides access to the garbage collector for reference cycles.\n"
|
|
|
|
"\n"
|
2000-08-07 06:45:31 +08:00
|
|
|
"enable() -- Enable automatic garbage collection.\n"
|
|
|
|
"disable() -- Disable automatic garbage collection.\n"
|
|
|
|
"isenabled() -- Returns true if automatic collection is enabled.\n"
|
2000-06-30 13:02:53 +08:00
|
|
|
"collect() -- Do a full collection right now.\n"
|
|
|
|
"set_debug() -- Set debugging flags.\n"
|
|
|
|
"get_debug() -- Get debugging flags.\n"
|
|
|
|
"set_threshold() -- Set the collection thresholds.\n"
|
|
|
|
"get_threshold() -- Return the current the collection thresholds.\n"
|
2001-08-09 23:58:59 +08:00
|
|
|
"get_objects() -- Return a list of all objects tracked by the collector.\n"
|
2002-06-14 04:33:02 +08:00
|
|
|
"get_referrers() -- Return the list of objects that refer to an object.\n");
|
2000-06-30 13:02:53 +08:00
|
|
|
|
|
|
|
static PyMethodDef GcMethods[] = {
|
2000-09-22 23:22:38 +08:00
|
|
|
{"enable", gc_enable, METH_VARARGS, gc_enable__doc__},
|
|
|
|
{"disable", gc_disable, METH_VARARGS, gc_disable__doc__},
|
2000-08-07 06:45:31 +08:00
|
|
|
{"isenabled", gc_isenabled, METH_VARARGS, gc_isenabled__doc__},
|
|
|
|
{"set_debug", gc_set_debug, METH_VARARGS, gc_set_debug__doc__},
|
|
|
|
{"get_debug", gc_get_debug, METH_VARARGS, gc_get_debug__doc__},
|
|
|
|
{"set_threshold", gc_set_thresh, METH_VARARGS, gc_set_thresh__doc__},
|
|
|
|
{"get_threshold", gc_get_thresh, METH_VARARGS, gc_get_thresh__doc__},
|
2000-09-22 23:22:38 +08:00
|
|
|
{"collect", gc_collect, METH_VARARGS, gc_collect__doc__},
|
2001-08-09 23:58:59 +08:00
|
|
|
{"get_objects", gc_get_objects,METH_VARARGS, gc_get_objects__doc__},
|
2001-11-24 17:24:51 +08:00
|
|
|
{"get_referrers", gc_get_referrers, METH_VARARGS,
|
|
|
|
gc_get_referrers__doc__},
|
2000-06-30 13:02:53 +08:00
|
|
|
{NULL, NULL} /* Sentinel */
|
|
|
|
};
|
|
|
|
|
|
|
|
void
|
|
|
|
initgc(void)
|
|
|
|
{
|
|
|
|
PyObject *m;
|
|
|
|
PyObject *d;
|
|
|
|
|
|
|
|
m = Py_InitModule4("gc",
|
|
|
|
GcMethods,
|
|
|
|
gc__doc__,
|
|
|
|
NULL,
|
|
|
|
PYTHON_API_VERSION);
|
|
|
|
d = PyModule_GetDict(m);
|
|
|
|
if (garbage == NULL) {
|
|
|
|
garbage = PyList_New(0);
|
|
|
|
}
|
|
|
|
PyDict_SetItemString(d, "garbage", garbage);
|
|
|
|
PyDict_SetItemString(d, "DEBUG_STATS",
|
|
|
|
PyInt_FromLong(DEBUG_STATS));
|
|
|
|
PyDict_SetItemString(d, "DEBUG_COLLECTABLE",
|
|
|
|
PyInt_FromLong(DEBUG_COLLECTABLE));
|
|
|
|
PyDict_SetItemString(d, "DEBUG_UNCOLLECTABLE",
|
|
|
|
PyInt_FromLong(DEBUG_UNCOLLECTABLE));
|
|
|
|
PyDict_SetItemString(d, "DEBUG_INSTANCES",
|
|
|
|
PyInt_FromLong(DEBUG_INSTANCES));
|
|
|
|
PyDict_SetItemString(d, "DEBUG_OBJECTS",
|
|
|
|
PyInt_FromLong(DEBUG_OBJECTS));
|
2000-09-22 23:22:38 +08:00
|
|
|
PyDict_SetItemString(d, "DEBUG_SAVEALL",
|
|
|
|
PyInt_FromLong(DEBUG_SAVEALL));
|
2000-06-30 13:02:53 +08:00
|
|
|
PyDict_SetItemString(d, "DEBUG_LEAK",
|
|
|
|
PyInt_FromLong(DEBUG_LEAK));
|
|
|
|
}
|
|
|
|
|
2001-08-30 08:05:51 +08:00
|
|
|
/* for debugging */
|
|
|
|
void _PyGC_Dump(PyGC_Head *g)
|
|
|
|
{
|
|
|
|
_PyObject_Dump(FROM_GC(g));
|
|
|
|
}
|
|
|
|
|
2000-06-30 13:02:53 +08:00
|
|
|
#endif /* WITH_CYCLE_GC */
|
2001-08-30 08:05:51 +08:00
|
|
|
|
|
|
|
/* extension modules might be compiled with GC support so these
|
|
|
|
functions must always be available */
|
|
|
|
|
2002-04-12 10:41:03 +08:00
|
|
|
#undef PyObject_GC_Track
|
|
|
|
#undef PyObject_GC_UnTrack
|
|
|
|
#undef PyObject_GC_Del
|
|
|
|
#undef _PyObject_GC_Malloc
|
|
|
|
|
2001-08-30 08:05:51 +08:00
|
|
|
void
|
2002-04-12 10:41:03 +08:00
|
|
|
PyObject_GC_Track(void *op)
|
2001-08-30 08:05:51 +08:00
|
|
|
{
|
|
|
|
_PyObject_GC_TRACK(op);
|
|
|
|
}
|
|
|
|
|
2002-04-12 10:41:03 +08:00
|
|
|
/* for binary compatibility with 2.2 */
|
2001-08-30 08:05:51 +08:00
|
|
|
void
|
2002-04-12 10:41:03 +08:00
|
|
|
_PyObject_GC_Track(PyObject *op)
|
|
|
|
{
|
|
|
|
PyObject_GC_Track(op);
|
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
PyObject_GC_UnTrack(void *op)
|
2001-08-30 08:05:51 +08:00
|
|
|
{
|
2002-03-29 11:04:25 +08:00
|
|
|
#ifdef WITH_CYCLE_GC
|
2002-05-21 23:53:24 +08:00
|
|
|
if (IS_TRACKED(op))
|
2002-03-29 04:34:59 +08:00
|
|
|
_PyObject_GC_UNTRACK(op);
|
2002-03-29 11:04:25 +08:00
|
|
|
#endif
|
2001-08-30 08:05:51 +08:00
|
|
|
}
|
|
|
|
|
2002-04-12 10:41:03 +08:00
|
|
|
/* for binary compatibility with 2.2 */
|
|
|
|
void
|
|
|
|
_PyObject_GC_UnTrack(PyObject *op)
|
|
|
|
{
|
|
|
|
PyObject_GC_UnTrack(op);
|
|
|
|
}
|
|
|
|
|
2001-08-30 08:05:51 +08:00
|
|
|
PyObject *
|
2002-04-12 10:41:03 +08:00
|
|
|
_PyObject_GC_Malloc(size_t basicsize)
|
2001-08-30 08:05:51 +08:00
|
|
|
{
|
|
|
|
PyObject *op;
|
|
|
|
#ifdef WITH_CYCLE_GC
|
2002-04-12 10:41:03 +08:00
|
|
|
PyGC_Head *g = PyObject_MALLOC(sizeof(PyGC_Head) + basicsize);
|
2001-08-30 08:05:51 +08:00
|
|
|
if (g == NULL)
|
2002-06-07 07:23:55 +08:00
|
|
|
return PyErr_NoMemory();
|
2001-10-12 02:31:31 +08:00
|
|
|
g->gc.gc_next = NULL;
|
2002-07-02 08:52:30 +08:00
|
|
|
g->gc.gc_refs = GC_UNTRACKED;
|
2002-05-04 13:35:20 +08:00
|
|
|
generations[0].count++; /* number of allocated GC objects */
|
|
|
|
if (generations[0].count > generations[0].threshold &&
|
2001-08-30 08:05:51 +08:00
|
|
|
enabled &&
|
2002-05-04 13:35:20 +08:00
|
|
|
generations[0].threshold &&
|
2001-08-30 08:05:51 +08:00
|
|
|
!collecting &&
|
|
|
|
!PyErr_Occurred()) {
|
|
|
|
collecting = 1;
|
2002-05-04 13:35:20 +08:00
|
|
|
collect_generations();
|
2001-08-30 08:05:51 +08:00
|
|
|
collecting = 0;
|
|
|
|
}
|
|
|
|
op = FROM_GC(g);
|
|
|
|
#else
|
2002-04-12 10:41:03 +08:00
|
|
|
op = PyObject_MALLOC(basicsize);
|
2001-08-30 08:05:51 +08:00
|
|
|
if (op == NULL)
|
2002-06-07 07:23:55 +08:00
|
|
|
return PyErr_NoMemory();
|
2001-08-30 08:05:51 +08:00
|
|
|
|
|
|
|
#endif
|
|
|
|
return op;
|
|
|
|
}
|
|
|
|
|
|
|
|
PyObject *
|
|
|
|
_PyObject_GC_New(PyTypeObject *tp)
|
|
|
|
{
|
2002-04-12 10:41:03 +08:00
|
|
|
PyObject *op = _PyObject_GC_Malloc(_PyObject_SIZE(tp));
|
2002-04-28 09:57:25 +08:00
|
|
|
if (op != NULL)
|
|
|
|
op = PyObject_INIT(op, tp);
|
|
|
|
return op;
|
2001-08-30 08:05:51 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
PyVarObject *
|
2001-10-07 05:27:34 +08:00
|
|
|
_PyObject_GC_NewVar(PyTypeObject *tp, int nitems)
|
2001-08-30 08:05:51 +08:00
|
|
|
{
|
2002-04-12 10:41:03 +08:00
|
|
|
const size_t size = _PyObject_VAR_SIZE(tp, nitems);
|
|
|
|
PyVarObject *op = (PyVarObject *) _PyObject_GC_Malloc(size);
|
2002-04-28 09:57:25 +08:00
|
|
|
if (op != NULL)
|
|
|
|
op = PyObject_INIT_VAR(op, tp, nitems);
|
|
|
|
return op;
|
2001-08-30 08:05:51 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
PyVarObject *
|
2001-10-07 05:27:34 +08:00
|
|
|
_PyObject_GC_Resize(PyVarObject *op, int nitems)
|
2001-08-30 08:05:51 +08:00
|
|
|
{
|
2001-10-07 11:54:51 +08:00
|
|
|
const size_t basicsize = _PyObject_VAR_SIZE(op->ob_type, nitems);
|
2001-08-30 08:05:51 +08:00
|
|
|
#ifdef WITH_CYCLE_GC
|
|
|
|
PyGC_Head *g = AS_GC(op);
|
2002-04-12 10:41:03 +08:00
|
|
|
g = PyObject_REALLOC(g, sizeof(PyGC_Head) + basicsize);
|
2001-08-30 08:05:51 +08:00
|
|
|
if (g == NULL)
|
|
|
|
return (PyVarObject *)PyErr_NoMemory();
|
|
|
|
op = (PyVarObject *) FROM_GC(g);
|
|
|
|
#else
|
2002-04-12 10:41:03 +08:00
|
|
|
op = PyObject_REALLOC(op, basicsize);
|
2001-08-30 08:05:51 +08:00
|
|
|
if (op == NULL)
|
|
|
|
return (PyVarObject *)PyErr_NoMemory();
|
|
|
|
#endif
|
2001-10-07 05:27:34 +08:00
|
|
|
op->ob_size = nitems;
|
2001-08-30 08:05:51 +08:00
|
|
|
return op;
|
|
|
|
}
|
|
|
|
|
|
|
|
void
|
2002-04-12 10:41:03 +08:00
|
|
|
PyObject_GC_Del(void *op)
|
2001-08-30 08:05:51 +08:00
|
|
|
{
|
|
|
|
#ifdef WITH_CYCLE_GC
|
|
|
|
PyGC_Head *g = AS_GC(op);
|
2002-05-21 23:53:24 +08:00
|
|
|
if (IS_TRACKED(op))
|
2001-08-30 08:05:51 +08:00
|
|
|
gc_list_remove(g);
|
2002-05-04 13:35:20 +08:00
|
|
|
if (generations[0].count > 0) {
|
|
|
|
generations[0].count--;
|
2001-08-30 08:05:51 +08:00
|
|
|
}
|
2002-04-12 10:41:03 +08:00
|
|
|
PyObject_FREE(g);
|
2001-08-30 08:05:51 +08:00
|
|
|
#else
|
2002-04-12 10:41:03 +08:00
|
|
|
PyObject_FREE(op);
|
2001-08-30 08:05:51 +08:00
|
|
|
#endif
|
|
|
|
}
|
|
|
|
|
2002-04-12 10:41:03 +08:00
|
|
|
/* for binary compatibility with 2.2 */
|
|
|
|
#undef _PyObject_GC_Del
|
|
|
|
void
|
|
|
|
_PyObject_GC_Del(PyObject *op)
|
|
|
|
{
|
|
|
|
PyObject_GC_Del(op);
|
|
|
|
}
|