Patrick SteinhardtWhen preparing the bitmap walk, we first establish the set of of have and want objects by iterating over the set of pending objects: if an object is marked as uninteresting, it's declared as an object we already have, otherwise as an object we want. These two sets are then used to compute which transitively referenced objects we need to obtain.
One special case here are tag objects: when a tag is requested, we resolve it to its first not-tag object and add both resolved objects as well as the tag itself into either the have or want set. Given that the uninteresting-property always propagates to referenced objects, it is clear that if the tag is uninteresting, so are its children and vice versa. But we fail to propagate the flag, which effectively means that referenced objects will always be interesting except for the case where they have already been marked as uninteresting explicitly.
This mislabeling does not impact correctness: we now have it in our "wants" set, and given that we later do an `AND NOT` of the bitmaps of "wants" and "haves" sets it is clear that the result must be the same. But we now start to needlessly traverse the tag's referenced objects in case it is uninteresting, even though we know that each referenced object will be uninteresting anyway. In the worst case, this can lead to a complete graph walk just to establish that we do not care for any object.
Fix the issue by propagating the `UNINTERESTING` flag to pointees of tag objects and add a benchmark with negative revisions to p5310. This shows some nice performance benefits, tested with linux.git:
Test HEAD HEAD~ -------------------------------------------------------------------------------------------------------- 5310.2: repack to disk 194.96(183.59+16.01) 195.98(184.04+16.67) +0.5% 5310.3: simulated clone 26.09(25.04+1.05) 26.11(24.95+1.16) +0.1% 5310.4: simulated fetch 2.64(5.25+0.72) 2.59(4.99+0.72) -1.9% 5310.5: pack to file (bitmap) 58.70(58.19+5.66) 58.99(58.66+5.52) +0.5% 5310.6: rev-list (commits) 1.46(1.19+0.27) 1.47(1.21+0.25) +0.7% 5310.7: rev-list (objects) 15.54(14.48+1.06) 16.17(15.06+1.11) +4.1% 5310.8: rev-list with negation (objects) 0.09(0.07+0.02) 23.13(21.53+1.60) +25600.0% 5310.9: rev-list count with blob:none 12.40(11.43+0.97) 12.38(11.33+1.04) -0.2% 5310.10: rev-list count with blob:limit=1k 18.14(16.09+2.05) 18.26(16.18+2.07) +0.7% 5310.11: rev-list count with tree:0 1.71(1.32+0.39) 1.67(1.31+0.36) -2.3% 5310.12: simulated partial clone 20.20(19.15+1.04) 20.45(19.34+1.10) +1.2% 5310.14: clone (partial bitmap) 12.68(13.89+1.03) 12.84(13.91+1.03) +1.3% 5310.15: pack to file (partial bitmap) 42.10(45.33+2.76) 42.30(45.41+2.76) +0.5% 5310.16: rev-list with tree filter (partial bitmap) 0.47(0.30+0.16) 0.45(0.30+0.14) -4.3%
While most benchmarks are probably in the range of noise, the newly added 5310.8 benchmark shows a performance improvement of 25600%.
Signed-off-by: Patrick Steinhardt <ps@pks.im>. --- pack-bitmap.c | 1 + t/perf/p5310-pack-bitmaps.sh | 4 ++++ 2 files changed, 5 insertions(+)
Show changes to 2 files +5 −0
pack-bitmap.c, t/perf/p5310-pack-bitmaps.sh
diff --git a/pack-bitmap.c b/pack-bitmap.c index 1f69b5fa85..11118a92e8 100644 --- a/pack-bitmap.c +++ b/pack-bitmap.c @@ -997,6 +997,7 @@ struct bitmap_index *prepare_bitmap_walk(struct rev_info *revs, object_list_insert(object, &wants); object = parse_object_or_die(get_tagged_oid(tag), NULL); + object->flags |= (tag->object.flags & UNINTERESTING); } if (object->flags & UNINTERESTING) diff --git a/t/perf/p5310-pack-bitmaps.sh b/t/perf/p5310-pack-bitmaps.sh index b3e725f031..2910c818ac 100755 --- a/t/perf/p5310-pack-bitmaps.sh +++ b/t/perf/p5310-pack-bitmaps.sh @@ -43,6 +43,10 @@ test_perf 'rev-list (objects)' ' git rev-list --all --use-bitmap-index --objects >/dev/null ' +test_perf 'rev-list with negation (objects)' ' + git rev-list --not --all --use-bitmap-index --objects >/dev/null +' + test_perf 'rev-list count with blob:none' ' git rev-list --use-bitmap-index --count --objects --all \ --filter=blob:none >/dev/null
-- 2.31.0
