blob: 8c524f8309d08b41c637623cf484ddd2b330804d [file] [log] [blame]
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001/*
Tim Peters88396172002-06-30 17:56:40 +00002
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00003 Reference Cycle Garbage Collection
4 ==================================
5
Neil Schemenauerb2c2c9e2000-10-04 16:34:09 +00006 Neil Schemenauer <nas@arctrix.com>
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00007
8 Based on a post on the python-dev list. Ideas from Guido van Rossum,
9 Eric Tiedemann, and various others.
10
Neil Schemenauer43411b52001-08-30 00:05:51 +000011 http://www.arctrix.com/nas/python/gc/
Neil Schemenauera7024e92008-07-15 19:24:01 +000012
13 The following mailing list threads provide a historical perspective on
14 the design of this module. Note that a fair amount of refinement has
15 occurred since those discussions.
16
17 http://mail.python.org/pipermail/python-dev/2000-March/002385.html
18 http://mail.python.org/pipermail/python-dev/2000-March/002434.html
19 http://mail.python.org/pipermail/python-dev/2000-March/002497.html
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +000020
21 For a highlevel view of the collection process, read the collect
22 function.
23
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +000024*/
25
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +000026#include "Python.h"
Antoine Pitrouf95a1b32010-05-09 15:52:27 +000027#include "frameobject.h" /* for PyFrame_ClearFreeList */
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +000028
Neil Schemenauer43411b52001-08-30 00:05:51 +000029/* Get an object's GC head */
30#define AS_GC(o) ((PyGC_Head *)(o)-1)
31
32/* Get the object given the GC head */
33#define FROM_GC(g) ((PyObject *)(((PyGC_Head *)g)+1))
34
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +000035/*** Global GC state ***/
36
Neil Schemenauer2880ae52002-05-04 05:35:20 +000037struct gc_generation {
Antoine Pitrouf95a1b32010-05-09 15:52:27 +000038 PyGC_Head head;
39 int threshold; /* collection threshold */
40 int count; /* count of allocations or collections of younger
41 generations */
Neil Schemenauer2880ae52002-05-04 05:35:20 +000042};
43
44#define NUM_GENERATIONS 3
45#define GEN_HEAD(n) (&generations[n].head)
46
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +000047/* linked lists of container objects */
Neil Schemenauer2880ae52002-05-04 05:35:20 +000048static struct gc_generation generations[NUM_GENERATIONS] = {
Antoine Pitrouf95a1b32010-05-09 15:52:27 +000049 /* PyGC_Head, threshold, count */
50 {{{GEN_HEAD(0), GEN_HEAD(0), 0}}, 700, 0},
51 {{{GEN_HEAD(1), GEN_HEAD(1), 0}}, 10, 0},
52 {{{GEN_HEAD(2), GEN_HEAD(2), 0}}, 10, 0},
Neil Schemenauer2880ae52002-05-04 05:35:20 +000053};
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +000054
Neil Schemenauer2880ae52002-05-04 05:35:20 +000055PyGC_Head *_PyGC_generation0 = GEN_HEAD(0);
56
Vladimir Marangozovf9d20c32000-08-06 22:45:31 +000057static int enabled = 1; /* automatic collection enabled? */
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +000058
Neil Schemenauer43411b52001-08-30 00:05:51 +000059/* true if we are currently running the collector */
Tim Petersbf384c22003-04-06 00:11:39 +000060static int collecting = 0;
Neil Schemenauer43411b52001-08-30 00:05:51 +000061
Tim Peters6fc13d92002-07-02 18:12:35 +000062/* list of uncollectable objects */
Tim Petersbf384c22003-04-06 00:11:39 +000063static PyObject *garbage = NULL;
Tim Peters6fc13d92002-07-02 18:12:35 +000064
65/* Python string to use if unhandled exception occurs */
Tim Petersbf384c22003-04-06 00:11:39 +000066static PyObject *gc_str = NULL;
Tim Peters6fc13d92002-07-02 18:12:35 +000067
Tim Peters93ad66d2003-04-05 17:15:44 +000068/* Python string used to look for __del__ attribute. */
69static PyObject *delstr = NULL;
Jeremy Hyltonce136e92003-04-04 19:59:06 +000070
Antoine Pitrou14b78f52009-01-09 22:27:08 +000071/* This is the number of objects who survived the last full collection. It
72 approximates the number of long lived objects tracked by the GC.
73
74 (by "full collection", we mean a collection of the oldest generation).
75*/
76static Py_ssize_t long_lived_total = 0;
77
78/* This is the number of objects who survived all "non-full" collections,
79 and are awaiting to undergo a full collection for the first time.
80
81*/
82static Py_ssize_t long_lived_pending = 0;
83
84/*
85 NOTE: about the counting of long-lived objects.
86
87 To limit the cost of garbage collection, there are two strategies;
88 - make each collection faster, e.g. by scanning fewer objects
89 - do less collections
90 This heuristic is about the latter strategy.
91
92 In addition to the various configurable thresholds, we only trigger a
93 full collection if the ratio
Antoine Pitrouf95a1b32010-05-09 15:52:27 +000094 long_lived_pending / long_lived_total
Antoine Pitrou14b78f52009-01-09 22:27:08 +000095 is above a given value (hardwired to 25%).
96
97 The reason is that, while "non-full" collections (i.e., collections of
98 the young and middle generations) will always examine roughly the same
99 number of objects -- determined by the aforementioned thresholds --,
100 the cost of a full collection is proportional to the total number of
101 long-lived objects, which is virtually unbounded.
102
103 Indeed, it has been remarked that doing a full collection every
104 <constant number> of object creations entails a dramatic performance
105 degradation in workloads which consist in creating and storing lots of
106 long-lived objects (e.g. building a large list of GC-tracked objects would
107 show quadratic performance, instead of linear as expected: see issue #4074).
108
109 Using the above ratio, instead, yields amortized linear performance in
110 the total number of objects (the effect of which can be summarized
111 thusly: "each full garbage collection is more and more costly as the
112 number of objects grows, but we do fewer and fewer of them").
113
114 This heuristic was suggested by Martin von Löwis on python-dev in
115 June 2008. His original analysis and proposal can be found at:
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000116 http://mail.python.org/pipermail/python-dev/2008-June/080579.html
Antoine Pitrou14b78f52009-01-09 22:27:08 +0000117*/
118
Antoine Pitroue1ad3da2012-05-28 22:22:34 +0200119/*
120 NOTE: about untracking of mutable objects.
121
122 Certain types of container cannot participate in a reference cycle, and
123 so do not need to be tracked by the garbage collector. Untracking these
124 objects reduces the cost of garbage collections. However, determining
125 which objects may be untracked is not free, and the costs must be
126 weighed against the benefits for garbage collection.
127
128 There are two possible strategies for when to untrack a container:
129
130 i) When the container is created.
131 ii) When the container is examined by the garbage collector.
132
133 Tuples containing only immutable objects (integers, strings etc, and
134 recursively, tuples of immutable objects) do not need to be tracked.
135 The interpreter creates a large number of tuples, many of which will
136 not survive until garbage collection. It is therefore not worthwhile
137 to untrack eligible tuples at creation time.
138
139 Instead, all tuples except the empty tuple are tracked when created.
140 During garbage collection it is determined whether any surviving tuples
141 can be untracked. A tuple can be untracked if all of its contents are
142 already not tracked. Tuples are examined for untracking in all garbage
143 collection cycles. It may take more than one cycle to untrack a tuple.
144
145 Dictionaries containing only immutable objects also do not need to be
146 tracked. Dictionaries are untracked when created. If a tracked item is
147 inserted into a dictionary (either as a key or value), the dictionary
148 becomes tracked. During a full garbage collection (all generations),
149 the collector will untrack any dictionaries whose contents are not
150 tracked.
151
152 The module provides the python function is_tracked(obj), which returns
153 the CURRENT tracking status of the object. Subsequent garbage
154 collections may change the tracking status of the object.
155
156 Untracking of certain containers was introduced in issue #4688, and
157 the algorithm was refined in response to issue #14775.
158*/
Antoine Pitrou14b78f52009-01-09 22:27:08 +0000159
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000160/* set for debugging information */
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000161#define DEBUG_STATS (1<<0) /* print collection statistics */
162#define DEBUG_COLLECTABLE (1<<1) /* print collectable objects */
163#define DEBUG_UNCOLLECTABLE (1<<2) /* print uncollectable objects */
164#define DEBUG_SAVEALL (1<<5) /* save all garbage in gc.garbage */
165#define DEBUG_LEAK DEBUG_COLLECTABLE | \
166 DEBUG_UNCOLLECTABLE | \
167 DEBUG_SAVEALL
Jeremy Hyltonb709df32000-09-01 02:47:25 +0000168static int debug;
Thomas Wouters477c8d52006-05-27 19:21:47 +0000169static PyObject *tmod = NULL;
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000170
Tim Peters6fc13d92002-07-02 18:12:35 +0000171/*--------------------------------------------------------------------------
172gc_refs values.
Neil Schemenauer43411b52001-08-30 00:05:51 +0000173
Tim Peters6fc13d92002-07-02 18:12:35 +0000174Between collections, every gc'ed object has one of two gc_refs values:
175
176GC_UNTRACKED
177 The initial state; objects returned by PyObject_GC_Malloc are in this
178 state. The object doesn't live in any generation list, and its
179 tp_traverse slot must not be called.
180
181GC_REACHABLE
182 The object lives in some generation list, and its tp_traverse is safe to
183 call. An object transitions to GC_REACHABLE when PyObject_GC_Track
184 is called.
185
186During a collection, gc_refs can temporarily take on other states:
187
188>= 0
189 At the start of a collection, update_refs() copies the true refcount
190 to gc_refs, for each object in the generation being collected.
191 subtract_refs() then adjusts gc_refs so that it equals the number of
192 times an object is referenced directly from outside the generation
193 being collected.
Martin v. Löwis774348c2002-11-09 19:54:06 +0000194 gc_refs remains >= 0 throughout these steps.
Tim Peters6fc13d92002-07-02 18:12:35 +0000195
196GC_TENTATIVELY_UNREACHABLE
197 move_unreachable() then moves objects not reachable (whether directly or
198 indirectly) from outside the generation into an "unreachable" set.
199 Objects that are found to be reachable have gc_refs set to GC_REACHABLE
200 again. Objects that are found to be unreachable have gc_refs set to
201 GC_TENTATIVELY_UNREACHABLE. It's "tentatively" because the pass doing
202 this can't be sure until it ends, and GC_TENTATIVELY_UNREACHABLE may
203 transition back to GC_REACHABLE.
204
205 Only objects with GC_TENTATIVELY_UNREACHABLE still set are candidates
206 for collection. If it's decided not to collect such an object (e.g.,
207 it has a __del__ method), its gc_refs is restored to GC_REACHABLE again.
208----------------------------------------------------------------------------
209*/
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000210#define GC_UNTRACKED _PyGC_REFS_UNTRACKED
211#define GC_REACHABLE _PyGC_REFS_REACHABLE
212#define GC_TENTATIVELY_UNREACHABLE _PyGC_REFS_TENTATIVELY_UNREACHABLE
Tim Peters19b74c72002-07-01 03:52:19 +0000213
Tim Peters6fc13d92002-07-02 18:12:35 +0000214#define IS_TRACKED(o) ((AS_GC(o))->gc.gc_refs != GC_UNTRACKED)
Tim Peters19b74c72002-07-01 03:52:19 +0000215#define IS_REACHABLE(o) ((AS_GC(o))->gc.gc_refs == GC_REACHABLE)
216#define IS_TENTATIVELY_UNREACHABLE(o) ( \
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000217 (AS_GC(o))->gc.gc_refs == GC_TENTATIVELY_UNREACHABLE)
Neil Schemenauera2b11ec2002-05-21 15:53:24 +0000218
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000219/*** list functions ***/
220
221static void
222gc_list_init(PyGC_Head *list)
223{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000224 list->gc.gc_prev = list;
225 list->gc.gc_next = list;
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000226}
227
Neil Schemenauer2880ae52002-05-04 05:35:20 +0000228static int
229gc_list_is_empty(PyGC_Head *list)
230{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000231 return (list->gc.gc_next == list);
Neil Schemenauer2880ae52002-05-04 05:35:20 +0000232}
233
Tim Peterse2d59182004-11-01 01:39:08 +0000234#if 0
235/* This became unused after gc_list_move() was introduced. */
236/* Append `node` to `list`. */
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000237static void
238gc_list_append(PyGC_Head *node, PyGC_Head *list)
239{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000240 node->gc.gc_next = list;
241 node->gc.gc_prev = list->gc.gc_prev;
242 node->gc.gc_prev->gc.gc_next = node;
243 list->gc.gc_prev = node;
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000244}
Tim Peterse2d59182004-11-01 01:39:08 +0000245#endif
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000246
Tim Peterse2d59182004-11-01 01:39:08 +0000247/* Remove `node` from the gc list it's currently in. */
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000248static void
249gc_list_remove(PyGC_Head *node)
250{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000251 node->gc.gc_prev->gc.gc_next = node->gc.gc_next;
252 node->gc.gc_next->gc.gc_prev = node->gc.gc_prev;
253 node->gc.gc_next = NULL; /* object is not currently tracked */
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000254}
255
Tim Peterse2d59182004-11-01 01:39:08 +0000256/* Move `node` from the gc list it's currently in (which is not explicitly
257 * named here) to the end of `list`. This is semantically the same as
258 * gc_list_remove(node) followed by gc_list_append(node, list).
259 */
260static void
261gc_list_move(PyGC_Head *node, PyGC_Head *list)
262{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000263 PyGC_Head *new_prev;
264 PyGC_Head *current_prev = node->gc.gc_prev;
265 PyGC_Head *current_next = node->gc.gc_next;
266 /* Unlink from current list. */
267 current_prev->gc.gc_next = current_next;
268 current_next->gc.gc_prev = current_prev;
269 /* Relink at end of new list. */
270 new_prev = node->gc.gc_prev = list->gc.gc_prev;
271 new_prev->gc.gc_next = list->gc.gc_prev = node;
272 node->gc.gc_next = list;
Tim Peterse2d59182004-11-01 01:39:08 +0000273}
274
275/* append list `from` onto list `to`; `from` becomes an empty list */
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000276static void
277gc_list_merge(PyGC_Head *from, PyGC_Head *to)
278{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000279 PyGC_Head *tail;
280 assert(from != to);
281 if (!gc_list_is_empty(from)) {
282 tail = to->gc.gc_prev;
283 tail->gc.gc_next = from->gc.gc_next;
284 tail->gc.gc_next->gc.gc_prev = tail;
285 to->gc.gc_prev = from->gc.gc_prev;
286 to->gc.gc_prev->gc.gc_next = to;
287 }
288 gc_list_init(from);
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000289}
290
Neal Norwitz7b216c52006-03-04 20:01:53 +0000291static Py_ssize_t
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000292gc_list_size(PyGC_Head *list)
293{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000294 PyGC_Head *gc;
295 Py_ssize_t n = 0;
296 for (gc = list->gc.gc_next; gc != list; gc = gc->gc.gc_next) {
297 n++;
298 }
299 return n;
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000300}
301
Tim Peters259272b2003-04-06 19:41:39 +0000302/* Append objects in a GC list to a Python list.
303 * Return 0 if all OK, < 0 if error (out of memory for list).
304 */
305static int
306append_objects(PyObject *py_list, PyGC_Head *gc_list)
307{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000308 PyGC_Head *gc;
309 for (gc = gc_list->gc.gc_next; gc != gc_list; gc = gc->gc.gc_next) {
310 PyObject *op = FROM_GC(gc);
311 if (op != py_list) {
312 if (PyList_Append(py_list, op)) {
313 return -1; /* exception */
314 }
315 }
316 }
317 return 0;
Tim Peters259272b2003-04-06 19:41:39 +0000318}
319
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000320/*** end of list stuff ***/
321
322
Tim Peters19b74c72002-07-01 03:52:19 +0000323/* Set all gc_refs = ob_refcnt. After this, gc_refs is > 0 for all objects
324 * in containers, and is GC_REACHABLE for all tracked gc objects not in
325 * containers.
Tim Peters88396172002-06-30 17:56:40 +0000326 */
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000327static void
328update_refs(PyGC_Head *containers)
329{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000330 PyGC_Head *gc = containers->gc.gc_next;
331 for (; gc != containers; gc = gc->gc.gc_next) {
332 assert(gc->gc.gc_refs == GC_REACHABLE);
333 gc->gc.gc_refs = Py_REFCNT(FROM_GC(gc));
334 /* Python's cyclic gc should never see an incoming refcount
335 * of 0: if something decref'ed to 0, it should have been
336 * deallocated immediately at that time.
337 * Possible cause (if the assert triggers): a tp_dealloc
338 * routine left a gc-aware object tracked during its teardown
339 * phase, and did something-- or allowed something to happen --
340 * that called back into Python. gc can trigger then, and may
341 * see the still-tracked dying object. Before this assert
342 * was added, such mistakes went on to allow gc to try to
343 * delete the object again. In a debug build, that caused
344 * a mysterious segfault, when _Py_ForgetReference tried
345 * to remove the object from the doubly-linked list of all
346 * objects a second time. In a release build, an actual
347 * double deallocation occurred, which leads to corruption
348 * of the allocator's internal bookkeeping pointers. That's
349 * so serious that maybe this should be a release-build
350 * check instead of an assert?
351 */
352 assert(gc->gc.gc_refs != 0);
353 }
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000354}
355
Tim Peters19b74c72002-07-01 03:52:19 +0000356/* A traversal callback for subtract_refs. */
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000357static int
358visit_decref(PyObject *op, void *data)
359{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000360 assert(op != NULL);
361 if (PyObject_IS_GC(op)) {
362 PyGC_Head *gc = AS_GC(op);
363 /* We're only interested in gc_refs for objects in the
364 * generation being collected, which can be recognized
365 * because only they have positive gc_refs.
366 */
367 assert(gc->gc.gc_refs != 0); /* else refcount was too small */
368 if (gc->gc.gc_refs > 0)
369 gc->gc.gc_refs--;
370 }
371 return 0;
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000372}
373
Tim Peters19b74c72002-07-01 03:52:19 +0000374/* Subtract internal references from gc_refs. After this, gc_refs is >= 0
375 * for all objects in containers, and is GC_REACHABLE for all tracked gc
376 * objects not in containers. The ones with gc_refs > 0 are directly
377 * reachable from outside containers, and so can't be collected.
378 */
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000379static void
380subtract_refs(PyGC_Head *containers)
381{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000382 traverseproc traverse;
383 PyGC_Head *gc = containers->gc.gc_next;
384 for (; gc != containers; gc=gc->gc.gc_next) {
385 traverse = Py_TYPE(FROM_GC(gc))->tp_traverse;
386 (void) traverse(FROM_GC(gc),
387 (visitproc)visit_decref,
388 NULL);
389 }
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000390}
391
Tim Peters19b74c72002-07-01 03:52:19 +0000392/* A traversal callback for move_unreachable. */
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000393static int
Tim Peters19b74c72002-07-01 03:52:19 +0000394visit_reachable(PyObject *op, PyGC_Head *reachable)
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000395{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000396 if (PyObject_IS_GC(op)) {
397 PyGC_Head *gc = AS_GC(op);
398 const Py_ssize_t gc_refs = gc->gc.gc_refs;
Tim Peters19b74c72002-07-01 03:52:19 +0000399
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000400 if (gc_refs == 0) {
401 /* This is in move_unreachable's 'young' list, but
402 * the traversal hasn't yet gotten to it. All
403 * we need to do is tell move_unreachable that it's
404 * reachable.
405 */
406 gc->gc.gc_refs = 1;
407 }
408 else if (gc_refs == GC_TENTATIVELY_UNREACHABLE) {
409 /* This had gc_refs = 0 when move_unreachable got
410 * to it, but turns out it's reachable after all.
411 * Move it back to move_unreachable's 'young' list,
412 * and move_unreachable will eventually get to it
413 * again.
414 */
415 gc_list_move(gc, reachable);
416 gc->gc.gc_refs = 1;
417 }
418 /* Else there's nothing to do.
419 * If gc_refs > 0, it must be in move_unreachable's 'young'
420 * list, and move_unreachable will eventually get to it.
421 * If gc_refs == GC_REACHABLE, it's either in some other
422 * generation so we don't care about it, or move_unreachable
423 * already dealt with it.
424 * If gc_refs == GC_UNTRACKED, it must be ignored.
425 */
426 else {
427 assert(gc_refs > 0
428 || gc_refs == GC_REACHABLE
429 || gc_refs == GC_UNTRACKED);
430 }
431 }
432 return 0;
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000433}
434
Tim Peters19b74c72002-07-01 03:52:19 +0000435/* Move the unreachable objects from young to unreachable. After this,
436 * all objects in young have gc_refs = GC_REACHABLE, and all objects in
437 * unreachable have gc_refs = GC_TENTATIVELY_UNREACHABLE. All tracked
438 * gc objects not in young or unreachable still have gc_refs = GC_REACHABLE.
439 * All objects in young after this are directly or indirectly reachable
440 * from outside the original young; and all objects in unreachable are
441 * not.
Tim Peters88396172002-06-30 17:56:40 +0000442 */
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000443static void
Tim Peters19b74c72002-07-01 03:52:19 +0000444move_unreachable(PyGC_Head *young, PyGC_Head *unreachable)
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000445{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000446 PyGC_Head *gc = young->gc.gc_next;
Tim Peters19b74c72002-07-01 03:52:19 +0000447
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000448 /* Invariants: all objects "to the left" of us in young have gc_refs
449 * = GC_REACHABLE, and are indeed reachable (directly or indirectly)
450 * from outside the young list as it was at entry. All other objects
451 * from the original young "to the left" of us are in unreachable now,
452 * and have gc_refs = GC_TENTATIVELY_UNREACHABLE. All objects to the
453 * left of us in 'young' now have been scanned, and no objects here
454 * or to the right have been scanned yet.
455 */
Tim Peters19b74c72002-07-01 03:52:19 +0000456
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000457 while (gc != young) {
458 PyGC_Head *next;
Tim Peters19b74c72002-07-01 03:52:19 +0000459
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000460 if (gc->gc.gc_refs) {
461 /* gc is definitely reachable from outside the
462 * original 'young'. Mark it as such, and traverse
463 * its pointers to find any other objects that may
464 * be directly reachable from it. Note that the
465 * call to tp_traverse may append objects to young,
466 * so we have to wait until it returns to determine
467 * the next object to visit.
468 */
469 PyObject *op = FROM_GC(gc);
470 traverseproc traverse = Py_TYPE(op)->tp_traverse;
471 assert(gc->gc.gc_refs > 0);
472 gc->gc.gc_refs = GC_REACHABLE;
473 (void) traverse(op,
474 (visitproc)visit_reachable,
475 (void *)young);
476 next = gc->gc.gc_next;
477 if (PyTuple_CheckExact(op)) {
478 _PyTuple_MaybeUntrack(op);
479 }
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000480 }
481 else {
482 /* This *may* be unreachable. To make progress,
483 * assume it is. gc isn't directly reachable from
484 * any object we've already traversed, but may be
485 * reachable from an object we haven't gotten to yet.
486 * visit_reachable will eventually move gc back into
487 * young if that's so, and we'll see it again.
488 */
489 next = gc->gc.gc_next;
490 gc_list_move(gc, unreachable);
491 gc->gc.gc_refs = GC_TENTATIVELY_UNREACHABLE;
492 }
493 gc = next;
494 }
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000495}
496
Antoine Pitroue1ad3da2012-05-28 22:22:34 +0200497/* Try to untrack all currently tracked dictionaries */
498static void
499untrack_dicts(PyGC_Head *head)
500{
501 PyGC_Head *next, *gc = head->gc.gc_next;
502 while (gc != head) {
503 PyObject *op = FROM_GC(gc);
504 next = gc->gc.gc_next;
505 if (PyDict_CheckExact(op))
506 _PyDict_MaybeUntrack(op);
507 gc = next;
508 }
509}
510
Amaury Forgeot d'Arcad8dcd52007-12-10 23:58:35 +0000511/* Return true if object has a finalization method. */
Neil Schemenauera765c122001-11-01 17:35:23 +0000512static int
513has_finalizer(PyObject *op)
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000514{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000515 if (PyGen_CheckExact(op))
516 return PyGen_NeedsFinalizing((PyGenObject *)op);
517 else
518 return op->ob_type->tp_del != NULL;
Neil Schemenauera765c122001-11-01 17:35:23 +0000519}
520
Tim Petersead8b7a2004-10-30 23:09:22 +0000521/* Move the objects in unreachable with __del__ methods into `finalizers`.
522 * Objects moved into `finalizers` have gc_refs set to GC_REACHABLE; the
523 * objects remaining in unreachable are left at GC_TENTATIVELY_UNREACHABLE.
Jeremy Hyltonce136e92003-04-04 19:59:06 +0000524 */
Neil Schemenauera765c122001-11-01 17:35:23 +0000525static void
Tim Petersead8b7a2004-10-30 23:09:22 +0000526move_finalizers(PyGC_Head *unreachable, PyGC_Head *finalizers)
Neil Schemenauera765c122001-11-01 17:35:23 +0000527{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000528 PyGC_Head *gc;
529 PyGC_Head *next;
Tim Petersf6b80452003-04-07 19:21:15 +0000530
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000531 /* March over unreachable. Move objects with finalizers into
532 * `finalizers`.
533 */
534 for (gc = unreachable->gc.gc_next; gc != unreachable; gc = next) {
535 PyObject *op = FROM_GC(gc);
Jeremy Hyltonce136e92003-04-04 19:59:06 +0000536
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000537 assert(IS_TENTATIVELY_UNREACHABLE(op));
538 next = gc->gc.gc_next;
Tim Petersf6ae7a42003-04-05 18:40:50 +0000539
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000540 if (has_finalizer(op)) {
541 gc_list_move(gc, finalizers);
542 gc->gc.gc_refs = GC_REACHABLE;
543 }
544 }
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000545}
546
Tim Peters19b74c72002-07-01 03:52:19 +0000547/* A traversal callback for move_finalizer_reachable. */
548static int
549visit_move(PyObject *op, PyGC_Head *tolist)
550{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000551 if (PyObject_IS_GC(op)) {
552 if (IS_TENTATIVELY_UNREACHABLE(op)) {
553 PyGC_Head *gc = AS_GC(op);
554 gc_list_move(gc, tolist);
555 gc->gc.gc_refs = GC_REACHABLE;
556 }
557 }
558 return 0;
Tim Peters19b74c72002-07-01 03:52:19 +0000559}
560
561/* Move objects that are reachable from finalizers, from the unreachable set
Tim Petersf6b80452003-04-07 19:21:15 +0000562 * into finalizers set.
Tim Peters19b74c72002-07-01 03:52:19 +0000563 */
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000564static void
Tim Petersf6b80452003-04-07 19:21:15 +0000565move_finalizer_reachable(PyGC_Head *finalizers)
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000566{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000567 traverseproc traverse;
568 PyGC_Head *gc = finalizers->gc.gc_next;
569 for (; gc != finalizers; gc = gc->gc.gc_next) {
570 /* Note that the finalizers list may grow during this. */
571 traverse = Py_TYPE(FROM_GC(gc))->tp_traverse;
572 (void) traverse(FROM_GC(gc),
573 (visitproc)visit_move,
574 (void *)finalizers);
575 }
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000576}
577
Tim Petersead8b7a2004-10-30 23:09:22 +0000578/* Clear all weakrefs to unreachable objects, and if such a weakref has a
579 * callback, invoke it if necessary. Note that it's possible for such
580 * weakrefs to be outside the unreachable set -- indeed, those are precisely
581 * the weakrefs whose callbacks must be invoked. See gc_weakref.txt for
582 * overview & some details. Some weakrefs with callbacks may be reclaimed
583 * directly by this routine; the number reclaimed is the return value. Other
584 * weakrefs with callbacks may be moved into the `old` generation. Objects
585 * moved into `old` have gc_refs set to GC_REACHABLE; the objects remaining in
586 * unreachable are left at GC_TENTATIVELY_UNREACHABLE. When this returns,
587 * no object in `unreachable` is weakly referenced anymore.
Tim Peters403a2032003-11-20 21:21:46 +0000588 */
589static int
Tim Petersead8b7a2004-10-30 23:09:22 +0000590handle_weakrefs(PyGC_Head *unreachable, PyGC_Head *old)
Tim Peters403a2032003-11-20 21:21:46 +0000591{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000592 PyGC_Head *gc;
593 PyObject *op; /* generally FROM_GC(gc) */
594 PyWeakReference *wr; /* generally a cast of op */
595 PyGC_Head wrcb_to_call; /* weakrefs with callbacks to call */
596 PyGC_Head *next;
597 int num_freed = 0;
Tim Peters403a2032003-11-20 21:21:46 +0000598
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000599 gc_list_init(&wrcb_to_call);
Tim Peters403a2032003-11-20 21:21:46 +0000600
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000601 /* Clear all weakrefs to the objects in unreachable. If such a weakref
602 * also has a callback, move it into `wrcb_to_call` if the callback
603 * needs to be invoked. Note that we cannot invoke any callbacks until
604 * all weakrefs to unreachable objects are cleared, lest the callback
605 * resurrect an unreachable object via a still-active weakref. We
606 * make another pass over wrcb_to_call, invoking callbacks, after this
607 * pass completes.
608 */
609 for (gc = unreachable->gc.gc_next; gc != unreachable; gc = next) {
610 PyWeakReference **wrlist;
Tim Petersead8b7a2004-10-30 23:09:22 +0000611
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000612 op = FROM_GC(gc);
613 assert(IS_TENTATIVELY_UNREACHABLE(op));
614 next = gc->gc.gc_next;
Tim Petersead8b7a2004-10-30 23:09:22 +0000615
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000616 if (! PyType_SUPPORTS_WEAKREFS(Py_TYPE(op)))
617 continue;
Tim Petersead8b7a2004-10-30 23:09:22 +0000618
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000619 /* It supports weakrefs. Does it have any? */
620 wrlist = (PyWeakReference **)
621 PyObject_GET_WEAKREFS_LISTPTR(op);
Tim Petersead8b7a2004-10-30 23:09:22 +0000622
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000623 /* `op` may have some weakrefs. March over the list, clear
624 * all the weakrefs, and move the weakrefs with callbacks
625 * that must be called into wrcb_to_call.
626 */
627 for (wr = *wrlist; wr != NULL; wr = *wrlist) {
628 PyGC_Head *wrasgc; /* AS_GC(wr) */
Tim Petersead8b7a2004-10-30 23:09:22 +0000629
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000630 /* _PyWeakref_ClearRef clears the weakref but leaves
631 * the callback pointer intact. Obscure: it also
632 * changes *wrlist.
633 */
634 assert(wr->wr_object == op);
635 _PyWeakref_ClearRef(wr);
636 assert(wr->wr_object == Py_None);
637 if (wr->wr_callback == NULL)
638 continue; /* no callback */
Tim Petersead8b7a2004-10-30 23:09:22 +0000639
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000640 /* Headache time. `op` is going away, and is weakly referenced by
641 * `wr`, which has a callback. Should the callback be invoked? If wr
642 * is also trash, no:
643 *
644 * 1. There's no need to call it. The object and the weakref are
645 * both going away, so it's legitimate to pretend the weakref is
646 * going away first. The user has to ensure a weakref outlives its
647 * referent if they want a guarantee that the wr callback will get
648 * invoked.
649 *
650 * 2. It may be catastrophic to call it. If the callback is also in
651 * cyclic trash (CT), then although the CT is unreachable from
652 * outside the current generation, CT may be reachable from the
653 * callback. Then the callback could resurrect insane objects.
654 *
655 * Since the callback is never needed and may be unsafe in this case,
656 * wr is simply left in the unreachable set. Note that because we
657 * already called _PyWeakref_ClearRef(wr), its callback will never
658 * trigger.
659 *
660 * OTOH, if wr isn't part of CT, we should invoke the callback: the
661 * weakref outlived the trash. Note that since wr isn't CT in this
662 * case, its callback can't be CT either -- wr acted as an external
663 * root to this generation, and therefore its callback did too. So
664 * nothing in CT is reachable from the callback either, so it's hard
665 * to imagine how calling it later could create a problem for us. wr
666 * is moved to wrcb_to_call in this case.
667 */
668 if (IS_TENTATIVELY_UNREACHABLE(wr))
669 continue;
670 assert(IS_REACHABLE(wr));
Tim Peterscc2a8662004-10-31 22:12:43 +0000671
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000672 /* Create a new reference so that wr can't go away
673 * before we can process it again.
674 */
675 Py_INCREF(wr);
Tim Petersead8b7a2004-10-30 23:09:22 +0000676
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000677 /* Move wr to wrcb_to_call, for the next pass. */
678 wrasgc = AS_GC(wr);
679 assert(wrasgc != next); /* wrasgc is reachable, but
680 next isn't, so they can't
681 be the same */
682 gc_list_move(wrasgc, &wrcb_to_call);
683 }
684 }
Tim Petersead8b7a2004-10-30 23:09:22 +0000685
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000686 /* Invoke the callbacks we decided to honor. It's safe to invoke them
687 * because they can't reference unreachable objects.
688 */
689 while (! gc_list_is_empty(&wrcb_to_call)) {
690 PyObject *temp;
691 PyObject *callback;
Tim Petersead8b7a2004-10-30 23:09:22 +0000692
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000693 gc = wrcb_to_call.gc.gc_next;
694 op = FROM_GC(gc);
695 assert(IS_REACHABLE(op));
696 assert(PyWeakref_Check(op));
697 wr = (PyWeakReference *)op;
698 callback = wr->wr_callback;
699 assert(callback != NULL);
Tim Petersead8b7a2004-10-30 23:09:22 +0000700
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000701 /* copy-paste of weakrefobject.c's handle_callback() */
702 temp = PyObject_CallFunctionObjArgs(callback, wr, NULL);
703 if (temp == NULL)
704 PyErr_WriteUnraisable(callback);
705 else
706 Py_DECREF(temp);
Tim Petersead8b7a2004-10-30 23:09:22 +0000707
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000708 /* Give up the reference we created in the first pass. When
709 * op's refcount hits 0 (which it may or may not do right now),
710 * op's tp_dealloc will decref op->wr_callback too. Note
711 * that the refcount probably will hit 0 now, and because this
712 * weakref was reachable to begin with, gc didn't already
713 * add it to its count of freed objects. Example: a reachable
714 * weak value dict maps some key to this reachable weakref.
715 * The callback removes this key->weakref mapping from the
716 * dict, leaving no other references to the weakref (excepting
717 * ours).
718 */
719 Py_DECREF(op);
720 if (wrcb_to_call.gc.gc_next == gc) {
721 /* object is still alive -- move it */
722 gc_list_move(gc, old);
723 }
724 else
725 ++num_freed;
726 }
Tim Petersead8b7a2004-10-30 23:09:22 +0000727
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000728 return num_freed;
Tim Peters403a2032003-11-20 21:21:46 +0000729}
730
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000731static void
Jeremy Hylton06257772000-08-31 15:10:24 +0000732debug_cycle(char *msg, PyObject *op)
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000733{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000734 PySys_WriteStderr("gc: %.100s <%.100s %p>\n",
735 msg, Py_TYPE(op)->tp_name, op);
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000736}
737
Tim Petersbf384c22003-04-06 00:11:39 +0000738/* Handle uncollectable garbage (cycles with finalizers, and stuff reachable
739 * only from such cycles).
Tim Petersf6b80452003-04-07 19:21:15 +0000740 * If DEBUG_SAVEALL, all objects in finalizers are appended to the module
741 * garbage list (a Python list), else only the objects in finalizers with
742 * __del__ methods are appended to garbage. All objects in finalizers are
743 * merged into the old list regardless.
Tim Peters259272b2003-04-06 19:41:39 +0000744 * Returns 0 if all OK, <0 on error (out of memory to grow the garbage list).
745 * The finalizers list is made empty on a successful return.
Tim Petersbf384c22003-04-06 00:11:39 +0000746 */
Tim Peters259272b2003-04-06 19:41:39 +0000747static int
Tim Petersf6b80452003-04-07 19:21:15 +0000748handle_finalizers(PyGC_Head *finalizers, PyGC_Head *old)
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000749{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000750 PyGC_Head *gc = finalizers->gc.gc_next;
Tim Petersf6b80452003-04-07 19:21:15 +0000751
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000752 if (garbage == NULL) {
753 garbage = PyList_New(0);
754 if (garbage == NULL)
755 Py_FatalError("gc couldn't create gc.garbage list");
756 }
757 for (; gc != finalizers; gc = gc->gc.gc_next) {
758 PyObject *op = FROM_GC(gc);
Tim Petersf6b80452003-04-07 19:21:15 +0000759
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000760 if ((debug & DEBUG_SAVEALL) || has_finalizer(op)) {
761 if (PyList_Append(garbage, op) < 0)
762 return -1;
763 }
764 }
Tim Petersf6b80452003-04-07 19:21:15 +0000765
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000766 gc_list_merge(finalizers, old);
767 return 0;
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000768}
769
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000770/* Break reference cycles by clearing the containers involved. This is
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000771 * tricky business as the lists can be changing and we don't know which
Tim Peters19b74c72002-07-01 03:52:19 +0000772 * objects may be freed. It is possible I screwed something up here.
773 */
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000774static void
Jeremy Hyltonce136e92003-04-04 19:59:06 +0000775delete_garbage(PyGC_Head *collectable, PyGC_Head *old)
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000776{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000777 inquiry clear;
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000778
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000779 while (!gc_list_is_empty(collectable)) {
780 PyGC_Head *gc = collectable->gc.gc_next;
781 PyObject *op = FROM_GC(gc);
Tim Peters88396172002-06-30 17:56:40 +0000782
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000783 assert(IS_TENTATIVELY_UNREACHABLE(op));
784 if (debug & DEBUG_SAVEALL) {
785 PyList_Append(garbage, op);
786 }
787 else {
788 if ((clear = Py_TYPE(op)->tp_clear) != NULL) {
789 Py_INCREF(op);
790 clear(op);
791 Py_DECREF(op);
792 }
793 }
794 if (collectable->gc.gc_next == gc) {
795 /* object is still alive, move it, it may die later */
796 gc_list_move(gc, old);
797 gc->gc.gc_refs = GC_REACHABLE;
798 }
799 }
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000800}
801
Christian Heimesa156e092008-02-16 07:38:31 +0000802/* Clear all free lists
803 * All free lists are cleared during the collection of the highest generation.
804 * Allocated items in the free list may keep a pymalloc arena occupied.
805 * Clearing the free lists may give back memory to the OS earlier.
806 */
807static void
808clear_freelists(void)
809{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000810 (void)PyMethod_ClearFreeList();
811 (void)PyFrame_ClearFreeList();
812 (void)PyCFunction_ClearFreeList();
813 (void)PyTuple_ClearFreeList();
814 (void)PyUnicode_ClearFreeList();
815 (void)PyFloat_ClearFreeList();
Christian Heimesa156e092008-02-16 07:38:31 +0000816}
817
Antoine Pitrou621601a2008-12-17 23:18:19 +0000818static double
819get_time(void)
820{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000821 double result = 0;
822 if (tmod != NULL) {
823 PyObject *f = PyObject_CallMethod(tmod, "time", NULL);
824 if (f == NULL) {
825 PyErr_Clear();
826 }
827 else {
828 if (PyFloat_Check(f))
829 result = PyFloat_AsDouble(f);
830 Py_DECREF(f);
831 }
832 }
833 return result;
Antoine Pitrou621601a2008-12-17 23:18:19 +0000834}
835
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000836/* This is the main function. Read this to understand how the
837 * collection process works. */
Neal Norwitz7b216c52006-03-04 20:01:53 +0000838static Py_ssize_t
Neil Schemenauer2880ae52002-05-04 05:35:20 +0000839collect(int generation)
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000840{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000841 int i;
842 Py_ssize_t m = 0; /* # objects collected */
843 Py_ssize_t n = 0; /* # unreachable objects that couldn't be collected */
844 PyGC_Head *young; /* the generation we are examining */
845 PyGC_Head *old; /* next older generation */
846 PyGC_Head unreachable; /* non-problematic unreachable trash */
847 PyGC_Head finalizers; /* objects with, & reachable from, __del__ */
848 PyGC_Head *gc;
849 double t1 = 0.0;
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000850
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000851 if (delstr == NULL) {
852 delstr = PyUnicode_InternFromString("__del__");
853 if (delstr == NULL)
854 Py_FatalError("gc couldn't allocate \"__del__\"");
855 }
Tim Peters93ad66d2003-04-05 17:15:44 +0000856
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000857 if (debug & DEBUG_STATS) {
858 PySys_WriteStderr("gc: collecting generation %d...\n",
859 generation);
860 PySys_WriteStderr("gc: objects in each generation:");
861 for (i = 0; i < NUM_GENERATIONS; i++)
862 PySys_WriteStderr(" %" PY_FORMAT_SIZE_T "d",
863 gc_list_size(GEN_HEAD(i)));
864 t1 = get_time();
865 PySys_WriteStderr("\n");
866 }
Neil Schemenauer2880ae52002-05-04 05:35:20 +0000867
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000868 /* update collection and allocation counters */
869 if (generation+1 < NUM_GENERATIONS)
870 generations[generation+1].count += 1;
871 for (i = 0; i <= generation; i++)
872 generations[i].count = 0;
Neil Schemenauer2880ae52002-05-04 05:35:20 +0000873
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000874 /* merge younger generations with one we are currently collecting */
875 for (i = 0; i < generation; i++) {
876 gc_list_merge(GEN_HEAD(i), GEN_HEAD(generation));
877 }
Neil Schemenauer2880ae52002-05-04 05:35:20 +0000878
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000879 /* handy references */
880 young = GEN_HEAD(generation);
881 if (generation < NUM_GENERATIONS-1)
882 old = GEN_HEAD(generation+1);
883 else
884 old = young;
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000885
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000886 /* Using ob_refcnt and gc_refs, calculate which objects in the
887 * container set are reachable from outside the set (i.e., have a
888 * refcount greater than 0 when all the references within the
889 * set are taken into account).
890 */
891 update_refs(young);
892 subtract_refs(young);
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000893
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000894 /* Leave everything reachable from outside young in young, and move
895 * everything else (in young) to unreachable.
896 * NOTE: This used to move the reachable objects into a reachable
897 * set instead. But most things usually turn out to be reachable,
898 * so it's more efficient to move the unreachable things.
899 */
900 gc_list_init(&unreachable);
901 move_unreachable(young, &unreachable);
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000902
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000903 /* Move reachable objects to next generation. */
904 if (young != old) {
905 if (generation == NUM_GENERATIONS - 2) {
906 long_lived_pending += gc_list_size(young);
907 }
908 gc_list_merge(young, old);
909 }
910 else {
Antoine Pitroue1ad3da2012-05-28 22:22:34 +0200911 /* We only untrack dicts in full collections, to avoid quadratic
912 dict build-up. See issue #14775. */
913 untrack_dicts(young);
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000914 long_lived_pending = 0;
915 long_lived_total = gc_list_size(young);
916 }
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000917
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000918 /* All objects in unreachable are trash, but objects reachable from
919 * finalizers can't safely be deleted. Python programmers should take
920 * care not to create such things. For Python, finalizers means
921 * instance objects with __del__ methods. Weakrefs with callbacks
922 * can also call arbitrary Python code but they will be dealt with by
923 * handle_weakrefs().
924 */
925 gc_list_init(&finalizers);
926 move_finalizers(&unreachable, &finalizers);
927 /* finalizers contains the unreachable objects with a finalizer;
928 * unreachable objects reachable *from* those are also uncollectable,
929 * and we move those into the finalizers list too.
930 */
931 move_finalizer_reachable(&finalizers);
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000932
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000933 /* Collect statistics on collectable objects found and print
934 * debugging information.
935 */
936 for (gc = unreachable.gc.gc_next; gc != &unreachable;
937 gc = gc->gc.gc_next) {
938 m++;
939 if (debug & DEBUG_COLLECTABLE) {
940 debug_cycle("collectable", FROM_GC(gc));
941 }
942 }
Tim Petersead8b7a2004-10-30 23:09:22 +0000943
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000944 /* Clear weakrefs and invoke callbacks as necessary. */
945 m += handle_weakrefs(&unreachable, old);
Tim Petersead8b7a2004-10-30 23:09:22 +0000946
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000947 /* Call tp_clear on objects in the unreachable set. This will cause
948 * the reference cycles to be broken. It may also cause some objects
949 * in finalizers to be freed.
950 */
951 delete_garbage(&unreachable, old);
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000952
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000953 /* Collect statistics on uncollectable objects found and print
954 * debugging information. */
955 for (gc = finalizers.gc.gc_next;
956 gc != &finalizers;
957 gc = gc->gc.gc_next) {
958 n++;
959 if (debug & DEBUG_UNCOLLECTABLE)
960 debug_cycle("uncollectable", FROM_GC(gc));
961 }
962 if (debug & DEBUG_STATS) {
963 double t2 = get_time();
964 if (m == 0 && n == 0)
965 PySys_WriteStderr("gc: done");
966 else
967 PySys_WriteStderr(
968 "gc: done, "
969 "%" PY_FORMAT_SIZE_T "d unreachable, "
970 "%" PY_FORMAT_SIZE_T "d uncollectable",
971 n+m, n);
972 if (t1 && t2) {
973 PySys_WriteStderr(", %.4fs elapsed", t2-t1);
974 }
975 PySys_WriteStderr(".\n");
976 }
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000977
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000978 /* Append instances in the uncollectable set to a Python
979 * reachable list of garbage. The programmer has to deal with
980 * this if they insist on creating this type of structure.
981 */
982 (void)handle_finalizers(&finalizers, old);
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000983
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000984 /* Clear free list only during the collection of the highest
985 * generation */
986 if (generation == NUM_GENERATIONS-1) {
987 clear_freelists();
988 }
Christian Heimesa156e092008-02-16 07:38:31 +0000989
Antoine Pitrouf95a1b32010-05-09 15:52:27 +0000990 if (PyErr_Occurred()) {
991 if (gc_str == NULL)
992 gc_str = PyUnicode_FromString("garbage collection");
993 PyErr_WriteUnraisable(gc_str);
994 Py_FatalError("unexpected exception during garbage collection");
995 }
996 return n+m;
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +0000997}
998
Neal Norwitz7b216c52006-03-04 20:01:53 +0000999static Py_ssize_t
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001000collect_generations(void)
1001{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001002 int i;
1003 Py_ssize_t n = 0;
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001004
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001005 /* Find the oldest generation (highest numbered) where the count
1006 * exceeds the threshold. Objects in the that generation and
1007 * generations younger than it will be collected. */
1008 for (i = NUM_GENERATIONS-1; i >= 0; i--) {
1009 if (generations[i].count > generations[i].threshold) {
1010 /* Avoid quadratic performance degradation in number
1011 of tracked objects. See comments at the beginning
1012 of this file, and issue #4074.
1013 */
1014 if (i == NUM_GENERATIONS - 1
1015 && long_lived_pending < long_lived_total / 4)
1016 continue;
1017 n = collect(i);
1018 break;
1019 }
1020 }
1021 return n;
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001022}
1023
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +00001024PyDoc_STRVAR(gc_enable__doc__,
Vladimir Marangozovf9d20c32000-08-06 22:45:31 +00001025"enable() -> None\n"
1026"\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +00001027"Enable automatic garbage collection.\n");
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001028
Vladimir Marangozovf9d20c32000-08-06 22:45:31 +00001029static PyObject *
Tim Peters50c61d52003-04-06 01:50:50 +00001030gc_enable(PyObject *self, PyObject *noargs)
Vladimir Marangozovf9d20c32000-08-06 22:45:31 +00001031{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001032 enabled = 1;
1033 Py_INCREF(Py_None);
1034 return Py_None;
Vladimir Marangozovf9d20c32000-08-06 22:45:31 +00001035}
1036
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +00001037PyDoc_STRVAR(gc_disable__doc__,
Vladimir Marangozovf9d20c32000-08-06 22:45:31 +00001038"disable() -> None\n"
1039"\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +00001040"Disable automatic garbage collection.\n");
Vladimir Marangozovf9d20c32000-08-06 22:45:31 +00001041
1042static PyObject *
Tim Peters50c61d52003-04-06 01:50:50 +00001043gc_disable(PyObject *self, PyObject *noargs)
Vladimir Marangozovf9d20c32000-08-06 22:45:31 +00001044{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001045 enabled = 0;
1046 Py_INCREF(Py_None);
1047 return Py_None;
Vladimir Marangozovf9d20c32000-08-06 22:45:31 +00001048}
1049
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +00001050PyDoc_STRVAR(gc_isenabled__doc__,
Vladimir Marangozovf9d20c32000-08-06 22:45:31 +00001051"isenabled() -> status\n"
1052"\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +00001053"Returns true if automatic garbage collection is enabled.\n");
Vladimir Marangozovf9d20c32000-08-06 22:45:31 +00001054
1055static PyObject *
Tim Peters50c61d52003-04-06 01:50:50 +00001056gc_isenabled(PyObject *self, PyObject *noargs)
Vladimir Marangozovf9d20c32000-08-06 22:45:31 +00001057{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001058 return PyBool_FromLong((long)enabled);
Vladimir Marangozovf9d20c32000-08-06 22:45:31 +00001059}
1060
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +00001061PyDoc_STRVAR(gc_collect__doc__,
Barry Warsawd3c38ff2006-03-07 09:46:03 +00001062"collect([generation]) -> n\n"
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001063"\n"
Barry Warsawd3c38ff2006-03-07 09:46:03 +00001064"With no arguments, run a full collection. The optional argument\n"
1065"may be an integer specifying which generation to collect. A ValueError\n"
1066"is raised if the generation number is invalid.\n\n"
1067"The number of unreachable objects is returned.\n");
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001068
1069static PyObject *
Barry Warsawd3c38ff2006-03-07 09:46:03 +00001070gc_collect(PyObject *self, PyObject *args, PyObject *kws)
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001071{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001072 static char *keywords[] = {"generation", NULL};
1073 int genarg = NUM_GENERATIONS - 1;
1074 Py_ssize_t n;
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001075
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001076 if (!PyArg_ParseTupleAndKeywords(args, kws, "|i", keywords, &genarg))
1077 return NULL;
Barry Warsawd3c38ff2006-03-07 09:46:03 +00001078
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001079 else if (genarg < 0 || genarg >= NUM_GENERATIONS) {
1080 PyErr_SetString(PyExc_ValueError, "invalid generation");
1081 return NULL;
1082 }
Barry Warsawd3c38ff2006-03-07 09:46:03 +00001083
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001084 if (collecting)
1085 n = 0; /* already collecting, don't do anything */
1086 else {
1087 collecting = 1;
1088 n = collect(genarg);
1089 collecting = 0;
1090 }
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001091
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001092 return PyLong_FromSsize_t(n);
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001093}
1094
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +00001095PyDoc_STRVAR(gc_set_debug__doc__,
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001096"set_debug(flags) -> None\n"
1097"\n"
1098"Set the garbage collection debugging flags. Debugging information is\n"
1099"written to sys.stderr.\n"
1100"\n"
1101"flags is an integer and can have the following bits turned on:\n"
1102"\n"
1103" DEBUG_STATS - Print statistics during collection.\n"
1104" DEBUG_COLLECTABLE - Print collectable objects found.\n"
1105" DEBUG_UNCOLLECTABLE - Print unreachable but uncollectable objects found.\n"
Neil Schemenauer544de1e2000-09-22 15:22:38 +00001106" DEBUG_SAVEALL - Save objects to gc.garbage rather than freeing them.\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +00001107" DEBUG_LEAK - Debug leaking programs (everything but STATS).\n");
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001108
1109static PyObject *
Vladimir Marangozovf9d20c32000-08-06 22:45:31 +00001110gc_set_debug(PyObject *self, PyObject *args)
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001111{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001112 if (!PyArg_ParseTuple(args, "i:set_debug", &debug))
1113 return NULL;
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001114
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001115 Py_INCREF(Py_None);
1116 return Py_None;
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001117}
1118
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +00001119PyDoc_STRVAR(gc_get_debug__doc__,
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001120"get_debug() -> flags\n"
1121"\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +00001122"Get the garbage collection debugging flags.\n");
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001123
1124static PyObject *
Tim Peters50c61d52003-04-06 01:50:50 +00001125gc_get_debug(PyObject *self, PyObject *noargs)
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001126{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001127 return Py_BuildValue("i", debug);
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001128}
1129
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +00001130PyDoc_STRVAR(gc_set_thresh__doc__,
Neal Norwitz2a47c0f2002-01-29 00:53:41 +00001131"set_threshold(threshold0, [threshold1, threshold2]) -> None\n"
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001132"\n"
1133"Sets the collection thresholds. Setting threshold0 to zero disables\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +00001134"collection.\n");
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001135
1136static PyObject *
Vladimir Marangozovf9d20c32000-08-06 22:45:31 +00001137gc_set_thresh(PyObject *self, PyObject *args)
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001138{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001139 int i;
1140 if (!PyArg_ParseTuple(args, "i|ii:set_threshold",
1141 &generations[0].threshold,
1142 &generations[1].threshold,
1143 &generations[2].threshold))
1144 return NULL;
1145 for (i = 2; i < NUM_GENERATIONS; i++) {
1146 /* generations higher than 2 get the same threshold */
1147 generations[i].threshold = generations[2].threshold;
1148 }
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001149
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001150 Py_INCREF(Py_None);
1151 return Py_None;
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001152}
1153
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +00001154PyDoc_STRVAR(gc_get_thresh__doc__,
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001155"get_threshold() -> (threshold0, threshold1, threshold2)\n"
1156"\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +00001157"Return the current collection thresholds\n");
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001158
1159static PyObject *
Tim Peters50c61d52003-04-06 01:50:50 +00001160gc_get_thresh(PyObject *self, PyObject *noargs)
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001161{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001162 return Py_BuildValue("(iii)",
1163 generations[0].threshold,
1164 generations[1].threshold,
1165 generations[2].threshold);
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001166}
1167
Barry Warsawd3c38ff2006-03-07 09:46:03 +00001168PyDoc_STRVAR(gc_get_count__doc__,
1169"get_count() -> (count0, count1, count2)\n"
1170"\n"
1171"Return the current collection counts\n");
1172
1173static PyObject *
1174gc_get_count(PyObject *self, PyObject *noargs)
1175{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001176 return Py_BuildValue("(iii)",
1177 generations[0].count,
1178 generations[1].count,
1179 generations[2].count);
Barry Warsawd3c38ff2006-03-07 09:46:03 +00001180}
1181
Neil Schemenauer48c70342001-08-09 15:38:31 +00001182static int
Martin v. Löwis560da622001-11-24 09:24:51 +00001183referrersvisit(PyObject* obj, PyObject *objs)
Neil Schemenauer48c70342001-08-09 15:38:31 +00001184{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001185 Py_ssize_t i;
1186 for (i = 0; i < PyTuple_GET_SIZE(objs); i++)
1187 if (PyTuple_GET_ITEM(objs, i) == obj)
1188 return 1;
1189 return 0;
Neil Schemenauer48c70342001-08-09 15:38:31 +00001190}
1191
Neil Schemenauer17e7be62001-08-10 14:46:47 +00001192static int
Martin v. Löwis560da622001-11-24 09:24:51 +00001193gc_referrers_for(PyObject *objs, PyGC_Head *list, PyObject *resultlist)
Neil Schemenauer48c70342001-08-09 15:38:31 +00001194{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001195 PyGC_Head *gc;
1196 PyObject *obj;
1197 traverseproc traverse;
1198 for (gc = list->gc.gc_next; gc != list; gc = gc->gc.gc_next) {
1199 obj = FROM_GC(gc);
1200 traverse = Py_TYPE(obj)->tp_traverse;
1201 if (obj == objs || obj == resultlist)
1202 continue;
1203 if (traverse(obj, (visitproc)referrersvisit, objs)) {
1204 if (PyList_Append(resultlist, obj) < 0)
1205 return 0; /* error */
1206 }
1207 }
1208 return 1; /* no error */
Neil Schemenauer48c70342001-08-09 15:38:31 +00001209}
1210
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +00001211PyDoc_STRVAR(gc_get_referrers__doc__,
Martin v. Löwis560da622001-11-24 09:24:51 +00001212"get_referrers(*objs) -> list\n\
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +00001213Return the list of objects that directly refer to any of objs.");
Neil Schemenauer48c70342001-08-09 15:38:31 +00001214
Neil Schemenauer17e7be62001-08-10 14:46:47 +00001215static PyObject *
Martin v. Löwis560da622001-11-24 09:24:51 +00001216gc_get_referrers(PyObject *self, PyObject *args)
Neil Schemenauer48c70342001-08-09 15:38:31 +00001217{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001218 int i;
1219 PyObject *result = PyList_New(0);
1220 if (!result) return NULL;
Thomas Wouters49fd7fa2006-04-21 10:40:58 +00001221
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001222 for (i = 0; i < NUM_GENERATIONS; i++) {
1223 if (!(gc_referrers_for(args, GEN_HEAD(i), result))) {
1224 Py_DECREF(result);
1225 return NULL;
1226 }
1227 }
1228 return result;
Neil Schemenauer48c70342001-08-09 15:38:31 +00001229}
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001230
Tim Peters0f81ab62003-04-08 16:39:48 +00001231/* Append obj to list; return true if error (out of memory), false if OK. */
Jeremy Hylton5bd378b2003-04-03 16:28:38 +00001232static int
Tim Peters730f5532003-04-08 17:17:17 +00001233referentsvisit(PyObject *obj, PyObject *list)
Jeremy Hylton5bd378b2003-04-03 16:28:38 +00001234{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001235 return PyList_Append(list, obj) < 0;
Jeremy Hylton5bd378b2003-04-03 16:28:38 +00001236}
1237
Tim Peters730f5532003-04-08 17:17:17 +00001238PyDoc_STRVAR(gc_get_referents__doc__,
1239"get_referents(*objs) -> list\n\
Jeremy Hylton059b0942003-04-03 16:29:13 +00001240Return the list of objects that are directly referred to by objs.");
Jeremy Hylton5bd378b2003-04-03 16:28:38 +00001241
1242static PyObject *
Tim Peters730f5532003-04-08 17:17:17 +00001243gc_get_referents(PyObject *self, PyObject *args)
Jeremy Hylton5bd378b2003-04-03 16:28:38 +00001244{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001245 Py_ssize_t i;
1246 PyObject *result = PyList_New(0);
Tim Peters0f81ab62003-04-08 16:39:48 +00001247
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001248 if (result == NULL)
1249 return NULL;
Tim Peters0f81ab62003-04-08 16:39:48 +00001250
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001251 for (i = 0; i < PyTuple_GET_SIZE(args); i++) {
1252 traverseproc traverse;
1253 PyObject *obj = PyTuple_GET_ITEM(args, i);
Tim Peters0f81ab62003-04-08 16:39:48 +00001254
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001255 if (! PyObject_IS_GC(obj))
1256 continue;
1257 traverse = Py_TYPE(obj)->tp_traverse;
1258 if (! traverse)
1259 continue;
1260 if (traverse(obj, (visitproc)referentsvisit, result)) {
1261 Py_DECREF(result);
1262 return NULL;
1263 }
1264 }
1265 return result;
Jeremy Hylton5bd378b2003-04-03 16:28:38 +00001266}
1267
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +00001268PyDoc_STRVAR(gc_get_objects__doc__,
Neil Schemenauerc7c8d8e2001-08-09 15:58:59 +00001269"get_objects() -> [...]\n"
1270"\n"
1271"Return a list of objects tracked by the collector (excluding the list\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +00001272"returned).\n");
Neil Schemenauerc7c8d8e2001-08-09 15:58:59 +00001273
Neil Schemenauerc7c8d8e2001-08-09 15:58:59 +00001274static PyObject *
Tim Peters50c61d52003-04-06 01:50:50 +00001275gc_get_objects(PyObject *self, PyObject *noargs)
Neil Schemenauerc7c8d8e2001-08-09 15:58:59 +00001276{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001277 int i;
1278 PyObject* result;
Neil Schemenauerc7c8d8e2001-08-09 15:58:59 +00001279
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001280 result = PyList_New(0);
1281 if (result == NULL)
1282 return NULL;
1283 for (i = 0; i < NUM_GENERATIONS; i++) {
1284 if (append_objects(result, GEN_HEAD(i))) {
1285 Py_DECREF(result);
1286 return NULL;
1287 }
1288 }
1289 return result;
Neil Schemenauerc7c8d8e2001-08-09 15:58:59 +00001290}
1291
Antoine Pitrou3a652b12009-03-23 18:52:06 +00001292PyDoc_STRVAR(gc_is_tracked__doc__,
1293"is_tracked(obj) -> bool\n"
1294"\n"
1295"Returns true if the object is tracked by the garbage collector.\n"
1296"Simple atomic objects will return false.\n"
1297);
1298
1299static PyObject *
1300gc_is_tracked(PyObject *self, PyObject *obj)
1301{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001302 PyObject *result;
1303
1304 if (PyObject_IS_GC(obj) && IS_TRACKED(obj))
1305 result = Py_True;
1306 else
1307 result = Py_False;
1308 Py_INCREF(result);
1309 return result;
Antoine Pitrou3a652b12009-03-23 18:52:06 +00001310}
1311
Neil Schemenauerc7c8d8e2001-08-09 15:58:59 +00001312
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +00001313PyDoc_STRVAR(gc__doc__,
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001314"This module provides access to the garbage collector for reference cycles.\n"
1315"\n"
Vladimir Marangozovf9d20c32000-08-06 22:45:31 +00001316"enable() -- Enable automatic garbage collection.\n"
1317"disable() -- Disable automatic garbage collection.\n"
1318"isenabled() -- Returns true if automatic collection is enabled.\n"
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001319"collect() -- Do a full collection right now.\n"
Thomas Wouters89f507f2006-12-13 04:49:30 +00001320"get_count() -- Return the current collection counts.\n"
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001321"set_debug() -- Set debugging flags.\n"
1322"get_debug() -- Get debugging flags.\n"
1323"set_threshold() -- Set the collection thresholds.\n"
1324"get_threshold() -- Return the current the collection thresholds.\n"
Neil Schemenauerc7c8d8e2001-08-09 15:58:59 +00001325"get_objects() -- Return a list of all objects tracked by the collector.\n"
Antoine Pitrou3a652b12009-03-23 18:52:06 +00001326"is_tracked() -- Returns true if a given object is tracked.\n"
Jeremy Hylton5bd378b2003-04-03 16:28:38 +00001327"get_referrers() -- Return the list of objects that refer to an object.\n"
Tim Peters730f5532003-04-08 17:17:17 +00001328"get_referents() -- Return the list of objects that an object refers to.\n");
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001329
1330static PyMethodDef GcMethods[] = {
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001331 {"enable", gc_enable, METH_NOARGS, gc_enable__doc__},
1332 {"disable", gc_disable, METH_NOARGS, gc_disable__doc__},
1333 {"isenabled", gc_isenabled, METH_NOARGS, gc_isenabled__doc__},
1334 {"set_debug", gc_set_debug, METH_VARARGS, gc_set_debug__doc__},
1335 {"get_debug", gc_get_debug, METH_NOARGS, gc_get_debug__doc__},
1336 {"get_count", gc_get_count, METH_NOARGS, gc_get_count__doc__},
1337 {"set_threshold", gc_set_thresh, METH_VARARGS, gc_set_thresh__doc__},
1338 {"get_threshold", gc_get_thresh, METH_NOARGS, gc_get_thresh__doc__},
1339 {"collect", (PyCFunction)gc_collect,
1340 METH_VARARGS | METH_KEYWORDS, gc_collect__doc__},
1341 {"get_objects", gc_get_objects,METH_NOARGS, gc_get_objects__doc__},
1342 {"is_tracked", gc_is_tracked, METH_O, gc_is_tracked__doc__},
1343 {"get_referrers", gc_get_referrers, METH_VARARGS,
1344 gc_get_referrers__doc__},
1345 {"get_referents", gc_get_referents, METH_VARARGS,
1346 gc_get_referents__doc__},
1347 {NULL, NULL} /* Sentinel */
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001348};
1349
Martin v. Löwis1a214512008-06-11 05:26:20 +00001350static struct PyModuleDef gcmodule = {
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001351 PyModuleDef_HEAD_INIT,
Antoine Pitrou696e0352010-08-08 22:18:46 +00001352 "gc", /* m_name */
1353 gc__doc__, /* m_doc */
1354 -1, /* m_size */
1355 GcMethods, /* m_methods */
1356 NULL, /* m_reload */
1357 NULL, /* m_traverse */
1358 NULL, /* m_clear */
1359 NULL /* m_free */
Martin v. Löwis1a214512008-06-11 05:26:20 +00001360};
1361
Jason Tishler6bc06ec2003-09-04 11:59:50 +00001362PyMODINIT_FUNC
Martin v. Löwis1a214512008-06-11 05:26:20 +00001363PyInit_gc(void)
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001364{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001365 PyObject *m;
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001366
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001367 m = PyModule_Create(&gcmodule);
Martin v. Löwis1a214512008-06-11 05:26:20 +00001368
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001369 if (m == NULL)
1370 return NULL;
Tim Peters11558872003-04-06 23:30:52 +00001371
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001372 if (garbage == NULL) {
1373 garbage = PyList_New(0);
1374 if (garbage == NULL)
1375 return NULL;
1376 }
1377 Py_INCREF(garbage);
1378 if (PyModule_AddObject(m, "garbage", garbage) < 0)
1379 return NULL;
Thomas Wouters477c8d52006-05-27 19:21:47 +00001380
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001381 /* Importing can't be done in collect() because collect()
1382 * can be called via PyGC_Collect() in Py_Finalize().
1383 * This wouldn't be a problem, except that <initialized> is
1384 * reset to 0 before calling collect which trips up
1385 * the import and triggers an assertion.
1386 */
1387 if (tmod == NULL) {
1388 tmod = PyImport_ImportModuleNoBlock("time");
1389 if (tmod == NULL)
1390 PyErr_Clear();
1391 }
Thomas Wouters477c8d52006-05-27 19:21:47 +00001392
Martin v. Löwis1a214512008-06-11 05:26:20 +00001393#define ADD_INT(NAME) if (PyModule_AddIntConstant(m, #NAME, NAME) < 0) return NULL
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001394 ADD_INT(DEBUG_STATS);
1395 ADD_INT(DEBUG_COLLECTABLE);
1396 ADD_INT(DEBUG_UNCOLLECTABLE);
1397 ADD_INT(DEBUG_SAVEALL);
1398 ADD_INT(DEBUG_LEAK);
Tim Peters11558872003-04-06 23:30:52 +00001399#undef ADD_INT
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001400 return m;
Jeremy Hyltonc5007aa2000-06-30 05:02:53 +00001401}
1402
Guido van Rossume13ddc92003-04-17 17:29:22 +00001403/* API to invoke gc.collect() from C */
Neal Norwitz7b216c52006-03-04 20:01:53 +00001404Py_ssize_t
Guido van Rossume13ddc92003-04-17 17:29:22 +00001405PyGC_Collect(void)
1406{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001407 Py_ssize_t n;
Guido van Rossume13ddc92003-04-17 17:29:22 +00001408
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001409 if (collecting)
1410 n = 0; /* already collecting, don't do anything */
1411 else {
1412 collecting = 1;
1413 n = collect(NUM_GENERATIONS - 1);
1414 collecting = 0;
1415 }
Guido van Rossume13ddc92003-04-17 17:29:22 +00001416
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001417 return n;
Guido van Rossume13ddc92003-04-17 17:29:22 +00001418}
1419
Antoine Pitrou696e0352010-08-08 22:18:46 +00001420void
1421_PyGC_Fini(void)
1422{
Antoine Pitrou2ed94eb2010-09-14 09:48:39 +00001423 if (!(debug & DEBUG_SAVEALL)
1424 && garbage != NULL && PyList_GET_SIZE(garbage) > 0) {
Georg Brandl08be72d2010-10-24 15:11:22 +00001425 char *message;
1426 if (debug & DEBUG_UNCOLLECTABLE)
Antoine Pitroub5d82042010-11-05 00:05:25 +00001427 message = "gc: %zd uncollectable objects at " \
Georg Brandl08be72d2010-10-24 15:11:22 +00001428 "shutdown";
1429 else
Antoine Pitroub5d82042010-11-05 00:05:25 +00001430 message = "gc: %zd uncollectable objects at " \
Georg Brandl08be72d2010-10-24 15:11:22 +00001431 "shutdown; use gc.set_debug(gc.DEBUG_UNCOLLECTABLE) to list them";
1432 if (PyErr_WarnFormat(PyExc_ResourceWarning, 0, message,
1433 PyList_GET_SIZE(garbage)) < 0)
1434 PyErr_WriteUnraisable(NULL);
Antoine Pitrou696e0352010-08-08 22:18:46 +00001435 if (debug & DEBUG_UNCOLLECTABLE) {
1436 PyObject *repr = NULL, *bytes = NULL;
1437 repr = PyObject_Repr(garbage);
1438 if (!repr || !(bytes = PyUnicode_EncodeFSDefault(repr)))
1439 PyErr_WriteUnraisable(garbage);
1440 else {
1441 PySys_WriteStderr(
1442 " %s\n",
1443 PyBytes_AS_STRING(bytes)
1444 );
1445 }
1446 Py_XDECREF(repr);
1447 Py_XDECREF(bytes);
1448 }
Antoine Pitrou696e0352010-08-08 22:18:46 +00001449 }
1450}
1451
Neil Schemenauer43411b52001-08-30 00:05:51 +00001452/* for debugging */
Guido van Rossume13ddc92003-04-17 17:29:22 +00001453void
1454_PyGC_Dump(PyGC_Head *g)
Neil Schemenauer43411b52001-08-30 00:05:51 +00001455{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001456 _PyObject_Dump(FROM_GC(g));
Neil Schemenauer43411b52001-08-30 00:05:51 +00001457}
1458
Neil Schemenauer43411b52001-08-30 00:05:51 +00001459/* extension modules might be compiled with GC support so these
1460 functions must always be available */
1461
Neil Schemenauerfec4eb12002-04-12 02:41:03 +00001462#undef PyObject_GC_Track
1463#undef PyObject_GC_UnTrack
1464#undef PyObject_GC_Del
1465#undef _PyObject_GC_Malloc
1466
Neil Schemenauer43411b52001-08-30 00:05:51 +00001467void
Neil Schemenauerfec4eb12002-04-12 02:41:03 +00001468PyObject_GC_Track(void *op)
Neil Schemenauer43411b52001-08-30 00:05:51 +00001469{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001470 _PyObject_GC_TRACK(op);
Neil Schemenauer43411b52001-08-30 00:05:51 +00001471}
1472
Neil Schemenauerfec4eb12002-04-12 02:41:03 +00001473/* for binary compatibility with 2.2 */
Neil Schemenauer43411b52001-08-30 00:05:51 +00001474void
Neil Schemenauerfec4eb12002-04-12 02:41:03 +00001475_PyObject_GC_Track(PyObject *op)
1476{
1477 PyObject_GC_Track(op);
1478}
1479
1480void
1481PyObject_GC_UnTrack(void *op)
Neil Schemenauer43411b52001-08-30 00:05:51 +00001482{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001483 /* Obscure: the Py_TRASHCAN mechanism requires that we be able to
1484 * call PyObject_GC_UnTrack twice on an object.
1485 */
1486 if (IS_TRACKED(op))
1487 _PyObject_GC_UNTRACK(op);
Neil Schemenauer43411b52001-08-30 00:05:51 +00001488}
1489
Neil Schemenauerfec4eb12002-04-12 02:41:03 +00001490/* for binary compatibility with 2.2 */
1491void
1492_PyObject_GC_UnTrack(PyObject *op)
1493{
1494 PyObject_GC_UnTrack(op);
1495}
1496
Neil Schemenauer43411b52001-08-30 00:05:51 +00001497PyObject *
Neil Schemenauerfec4eb12002-04-12 02:41:03 +00001498_PyObject_GC_Malloc(size_t basicsize)
Neil Schemenauer43411b52001-08-30 00:05:51 +00001499{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001500 PyObject *op;
1501 PyGC_Head *g;
1502 if (basicsize > PY_SSIZE_T_MAX - sizeof(PyGC_Head))
1503 return PyErr_NoMemory();
1504 g = (PyGC_Head *)PyObject_MALLOC(
1505 sizeof(PyGC_Head) + basicsize);
1506 if (g == NULL)
1507 return PyErr_NoMemory();
1508 g->gc.gc_refs = GC_UNTRACKED;
1509 generations[0].count++; /* number of allocated GC objects */
1510 if (generations[0].count > generations[0].threshold &&
1511 enabled &&
1512 generations[0].threshold &&
1513 !collecting &&
1514 !PyErr_Occurred()) {
1515 collecting = 1;
1516 collect_generations();
1517 collecting = 0;
1518 }
1519 op = FROM_GC(g);
1520 return op;
Neil Schemenauer43411b52001-08-30 00:05:51 +00001521}
1522
1523PyObject *
1524_PyObject_GC_New(PyTypeObject *tp)
1525{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001526 PyObject *op = _PyObject_GC_Malloc(_PyObject_SIZE(tp));
1527 if (op != NULL)
1528 op = PyObject_INIT(op, tp);
1529 return op;
Neil Schemenauer43411b52001-08-30 00:05:51 +00001530}
1531
1532PyVarObject *
Martin v. Löwis18e16552006-02-15 17:27:45 +00001533_PyObject_GC_NewVar(PyTypeObject *tp, Py_ssize_t nitems)
Neil Schemenauer43411b52001-08-30 00:05:51 +00001534{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001535 const size_t size = _PyObject_VAR_SIZE(tp, nitems);
1536 PyVarObject *op = (PyVarObject *) _PyObject_GC_Malloc(size);
1537 if (op != NULL)
1538 op = PyObject_INIT_VAR(op, tp, nitems);
1539 return op;
Neil Schemenauer43411b52001-08-30 00:05:51 +00001540}
1541
1542PyVarObject *
Martin v. Löwis41290682006-02-16 14:56:14 +00001543_PyObject_GC_Resize(PyVarObject *op, Py_ssize_t nitems)
Neil Schemenauer43411b52001-08-30 00:05:51 +00001544{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001545 const size_t basicsize = _PyObject_VAR_SIZE(Py_TYPE(op), nitems);
1546 PyGC_Head *g = AS_GC(op);
1547 if (basicsize > PY_SSIZE_T_MAX - sizeof(PyGC_Head))
1548 return (PyVarObject *)PyErr_NoMemory();
1549 g = (PyGC_Head *)PyObject_REALLOC(g, sizeof(PyGC_Head) + basicsize);
1550 if (g == NULL)
1551 return (PyVarObject *)PyErr_NoMemory();
1552 op = (PyVarObject *) FROM_GC(g);
1553 Py_SIZE(op) = nitems;
1554 return op;
Neil Schemenauer43411b52001-08-30 00:05:51 +00001555}
1556
1557void
Neil Schemenauerfec4eb12002-04-12 02:41:03 +00001558PyObject_GC_Del(void *op)
Neil Schemenauer43411b52001-08-30 00:05:51 +00001559{
Antoine Pitrouf95a1b32010-05-09 15:52:27 +00001560 PyGC_Head *g = AS_GC(op);
1561 if (IS_TRACKED(op))
1562 gc_list_remove(g);
1563 if (generations[0].count > 0) {
1564 generations[0].count--;
1565 }
1566 PyObject_FREE(g);
Neil Schemenauer43411b52001-08-30 00:05:51 +00001567}