describe: use prio_queue_replace()

Optimize the sequence get+put to peek+replace to avoid one unnecessary heap rebalance. Do that by tracking partial get operations in a prio_queue wrapper, struct lazy_queue, and using wrapper functions that turn get into peek and put into replace as needed. This is simpler than tracking the state explicitly in the calling code. We get a nice speedup on top of the previous patch's conversion to prio_queue: Benchmark 1: ./git_2.50.1 describe $(git rev-list v2.41.0..v2.47.0) Time (mean ± σ): 1.559 s ± 0.002 s [User: 1.493 s, System: 0.051 s] Range (min … max): 1.556 s … 1.563 s 10 runs Benchmark 2: ./git_describe_pq describe $(git rev-list v2.41.0..v2.47.0) Time (mean ± σ): 1.204 s ± 0.001 s [User: 1.138 s, System: 0.051 s] Range (min … max): 1.202 s … 1.205 s 10 runs Benchmark 3: ./git describe $(git rev-list v2.41.0..v2.47.0) Time (mean ± σ): 850.9 ms ± 1.6 ms [User: 786.6 ms, System: 49.8 ms] Range (min … max): 849.1 ms … 854.1 ms 10 runs Summary ./git describe $(git rev-list v2.41.0..v2.47.0) ran 1.41 ± 0.00 times faster than ./git_describe_pq describe $(git rev-list v2.41.0..v2.47.0) 1.83 ± 0.00 times faster than ./git_2.50.1 describe $(git rev-list v2.41.0..v2.47.0) Signed-off-by: René Scharfe <l.s.r@web.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>

René Scharfe committed Aug 3, 2025 at 13:49 UTC 08bb69d70f55cb6b44cdc6aefa7bc1d9cf4eb3f3
1 file changed +52 -16
builtin/describe.c
+52 -16
@@ -250,22 +250,58 @@ static int compare_pt(const void *a_, const void *b_)
250 return 0;
251 }
252
253 -static bool all_have_flag(const struct prio_queue *queue, unsigned flag)
253 +struct lazy_queue {
254 + struct prio_queue queue;
255 + bool get_pending;
256 +};
257 +
258 +#define LAZY_QUEUE_INIT { { compare_commits_by_commit_date }, false }
259 +
260 +static void *lazy_queue_get(struct lazy_queue *queue)
261 +{
262 + if (queue->get_pending)
263 + prio_queue_get(&queue->queue);
264 + else
265 + queue->get_pending = true;
266 + return prio_queue_peek(&queue->queue);
267 +}
268 +
269 +static void lazy_queue_put(struct lazy_queue *queue, void *thing)
270 +{
271 + if (queue->get_pending)
272 + prio_queue_replace(&queue->queue, thing);
273 + else
274 + prio_queue_put(&queue->queue, thing);
275 + queue->get_pending = false;
276 +}
277 +
278 +static bool lazy_queue_empty(const struct lazy_queue *queue)
279 +{
280 + return queue->queue.nr == (queue->get_pending ? 1 : 0);
281 +}
282 +
283 +static void lazy_queue_clear(struct lazy_queue *queue)
284 +{
285 + clear_prio_queue(&queue->queue);
286 + queue->get_pending = false;
287 +}
288 +
289 +static bool all_have_flag(const struct lazy_queue *queue, unsigned flag)
290 {
255 - for (size_t i = 0; i < queue->nr; i++) {
256 - struct commit *commit = queue->array[i].data;
291 + for (size_t i = queue->get_pending ? 1 : 0; i < queue->queue.nr; i++) {
292 + struct commit *commit = queue->queue.array[i].data;
293 if (!(commit->object.flags & flag))
294 return false;
295 }
296 return true;
297 }
298
263 -static unsigned long finish_depth_computation(struct prio_queue *queue,
299 +static unsigned long finish_depth_computation(struct lazy_queue *queue,
300 struct possible_tag *best)
301 {
302 unsigned long seen_commits = 0;
267 - while (queue->nr) {
268 - struct commit *c = prio_queue_get(queue);
303 + while (!lazy_queue_empty(queue)) {
304 + struct commit *c = lazy_queue_get(queue);
305 struct commit_list *parents = c->parents;
306 seen_commits++;
307 if (c->object.flags & best->flag_within) {
@@ -277,7 +313,7 @@ static unsigned long finish_depth_computation(struct prio_queue *queue,
313 struct commit *p = parents->item;
314 repo_parse_commit(the_repository, p);
315 if (!(p->object.flags & SEEN))
280 - prio_queue_put(queue, p);
316 + lazy_queue_put(queue, p);
317 p->object.flags |= c->object.flags;
318 parents = parents->next;
319 }
@@ -319,7 +355,7 @@ static void append_suffix(int depth, const struct object_id *oid, struct strbuf
355 static void describe_commit(struct object_id *oid, struct strbuf *dst)
356 {
357 struct commit *cmit, *gave_up_on = NULL;
322 - struct prio_queue queue = { compare_commits_by_commit_date };
358 + struct lazy_queue queue = LAZY_QUEUE_INIT;
359 struct commit_name *n;
360 struct possible_tag all_matches[MAX_TAGS];
361 unsigned int match_cnt = 0, annotated_cnt = 0, cur_match;
@@ -363,9 +399,9 @@ static void describe_commit(struct object_id *oid, struct strbuf *dst)
399 }
400
401 cmit->object.flags = SEEN;
366 - prio_queue_put(&queue, cmit);
367 - while (queue.nr) {
368 - struct commit *c = prio_queue_get(&queue);
402 + lazy_queue_put(&queue, cmit);
403 + while (!lazy_queue_empty(&queue)) {
404 + struct commit *c = lazy_queue_get(&queue);
405 struct commit_list *parents = c->parents;
406 struct commit_name **slot;
407
@@ -399,7 +435,7 @@ static void describe_commit(struct object_id *oid, struct strbuf *dst)
435 t->depth++;
436 }
437 /* Stop if last remaining path already covered by best candidate(s) */
402 - if (annotated_cnt && !queue.nr) {
438 + if (annotated_cnt && lazy_queue_empty(&queue)) {
439 int best_depth = INT_MAX;
440 unsigned best_within = 0;
441 for (cur_match = 0; cur_match < match_cnt; cur_match++) {
@@ -422,7 +458,7 @@ static void describe_commit(struct object_id *oid, struct strbuf *dst)
458 struct commit *p = parents->item;
459 repo_parse_commit(the_repository, p);
460 if (!(p->object.flags & SEEN))
425 - prio_queue_put(&queue, p);
461 + lazy_queue_put(&queue, p);
462 p->object.flags |= c->object.flags;
463 parents = parents->next;
464
@@ -437,7 +473,7 @@ static void describe_commit(struct object_id *oid, struct strbuf *dst)
473 strbuf_add_unique_abbrev(dst, cmit_oid, abbrev);
474 if (suffix)
475 strbuf_addstr(dst, suffix);
440 - clear_prio_queue(&queue);
476 + lazy_queue_clear(&queue);
477 return;
478 }
479 if (unannotated_cnt)
@@ -453,11 +489,11 @@ static void describe_commit(struct object_id *oid, struct strbuf *dst)
489 QSORT(all_matches, match_cnt, compare_pt);
490
491 if (gave_up_on) {
456 - prio_queue_put(&queue, gave_up_on);
492 + lazy_queue_put(&queue, gave_up_on);
493 seen_commits--;
494 }
495 seen_commits += finish_depth_computation(&queue, &all_matches[0]);
460 - clear_prio_queue(&queue);
496 + lazy_queue_clear(&queue);
497
498 if (debug) {
499 static int label_width = -1;