diff --git a/pack-bitmap-write.c b/pack-bitmap-write.c index 47250398aa..6e8060f8a0 100644 --- a/pack-bitmap-write.c +++ b/pack-bitmap-write.c @@ -19,6 +19,10 @@ #include "tree-walk.h" #include "pseudo-merge.h" #include "oid-array.h" +#include "config.h" +#include "alloc.h" +#include "refs.h" +#include "strmap.h" struct bitmapped_commit { struct commit *commit; @@ -465,6 +469,7 @@ static int fill_bitmap_tree(struct bitmap_writer *writer, } static int reused_bitmaps_nr; +static int reused_pseudo_merge_bitmaps_nr; static int fill_bitmap_commit(struct bitmap_writer *writer, struct bb_commit *ent, @@ -490,7 +495,7 @@ static int fill_bitmap_commit(struct bitmap_writer *writer, struct bitmap *remapped = bitmap_new(); if (commit->object.flags & BITMAP_PSEUDO_MERGE) - old = NULL; + old = pseudo_merge_bitmap_for_commit(old_bitmap, c); else old = bitmap_for_commit(old_bitmap, c); /* @@ -501,7 +506,10 @@ static int fill_bitmap_commit(struct bitmap_writer *writer, if (old && !rebuild_bitmap(mapping, old, remapped)) { bitmap_or(ent->bitmap, remapped); bitmap_free(remapped); - reused_bitmaps_nr++; + if (commit->object.flags & BITMAP_PSEUDO_MERGE) + reused_pseudo_merge_bitmaps_nr++; + else + reused_bitmaps_nr++; continue; } bitmap_free(remapped); @@ -631,6 +639,9 @@ int bitmap_writer_build(struct bitmap_writer *writer, the_repository); trace2_data_intmax("pack-bitmap-write", the_repository, "building_bitmaps_reused", reused_bitmaps_nr); + trace2_data_intmax("pack-bitmap-write", the_repository, + "building_bitmaps_pseudo_merge_reused", + reused_pseudo_merge_bitmaps_nr); stop_progress(&writer->progress); diff --git a/pack-bitmap.c b/pack-bitmap.c index 1966b3b95f..70230e2647 100644 --- a/pack-bitmap.c +++ b/pack-bitmap.c @@ -1316,6 +1316,37 @@ static struct bitmap *find_boundary_objects(struct bitmap_index *bitmap_git, return cb.base; } +struct ewah_bitmap *pseudo_merge_bitmap_for_commit(struct bitmap_index *bitmap_git, + struct commit *commit) +{ + struct commit_list *p; + struct bitmap *parents; + struct pseudo_merge *match = NULL; + + if (!bitmap_git->pseudo_merges.nr) + return NULL; + + parents = bitmap_new(); + + for (p = commit->parents; p; p = p->next) { + int pos = bitmap_position(bitmap_git, &p->item->object.oid); + if (pos < 0 || pos >= bitmap_num_objects(bitmap_git)) + goto done; + + bitmap_set(parents, pos); + } + + match = pseudo_merge_for_parents(&bitmap_git->pseudo_merges, + parents); + +done: + bitmap_free(parents); + if (match) + return pseudo_merge_bitmap(&bitmap_git->pseudo_merges, match); + + return NULL; +} + static void unsatisfy_all_pseudo_merges(struct bitmap_index *bitmap_git) { uint32_t i; @@ -2809,6 +2840,7 @@ void free_bitmap_index(struct bitmap_index *b) */ close_midx_revindex(b->midx); } + free_pseudo_merge_map(&b->pseudo_merges); free(b); } diff --git a/pack-bitmap.h b/pack-bitmap.h index 4466b5ad0f..1171e6d989 100644 --- a/pack-bitmap.h +++ b/pack-bitmap.h @@ -142,6 +142,8 @@ int rebuild_bitmap(const uint32_t *reposition, struct bitmap *dest); struct ewah_bitmap *bitmap_for_commit(struct bitmap_index *bitmap_git, struct commit *commit); +struct ewah_bitmap *pseudo_merge_bitmap_for_commit(struct bitmap_index *bitmap_git, + struct commit *commit); void bitmap_writer_select_commits(struct bitmap_writer *writer, struct commit **indexed_commits, unsigned int indexed_commits_nr); diff --git a/pseudo-merge.c b/pseudo-merge.c index 7d13101149..a117520996 100644 --- a/pseudo-merge.c +++ b/pseudo-merge.c @@ -699,3 +699,58 @@ int cascade_pseudo_merges(const struct pseudo_merge_map *pm, return ret; } + +struct pseudo_merge *pseudo_merge_for_parents(const struct pseudo_merge_map *pm, + struct bitmap *parents) +{ + struct pseudo_merge *match = NULL; + size_t i; + + if (!pm->nr) + return NULL; + + /* + * NOTE: this loop is quadratic in the worst-case (where no + * matching pseudo-merge bitmaps are found), but in practice + * this is OK for a few reasons: + * + * - Rejecting pseudo-merge bitmaps that do not match the + * given commit is done quickly (i.e. `bitmap_equals_ewah()` + * returns early when we know the two bitmaps aren't equal. + * + * - Already matched pseudo-merge bitmaps (which we track with + * the `->satisfied` bit here) are skipped as potential + * candidates. + * + * - The number of pseudo-merges should be small (in the + * hundreds for most repositories). + * + * If in the future this semi-quadratic behavior does become a + * problem, another approach would be to keep track of which + * pseudo-merges are still "viable" after enumerating the + * pseudo-merge commit's parents: + * + * - A pseudo-merge bitmap becomes non-viable when the bit(s) + * corresponding to one or more parent(s) of the given + * commit are not set in a candidate pseudo-merge's commits + * bitmap. + * + * - After processing all bits, enumerate the remaining set of + * viable pseudo-merge bitmaps, and check that their + * popcount() matches the number of parents in the given + * commit. + */ + for (i = 0; i < pm->nr; i++) { + struct pseudo_merge *candidate = use_pseudo_merge(pm, &pm->v[i]); + if (!candidate || candidate->satisfied) + continue; + if (!bitmap_equals_ewah(parents, candidate->commits)) + continue; + + match = candidate; + match->satisfied = 1; + break; + } + + return match; +} diff --git a/pseudo-merge.h b/pseudo-merge.h index 755edc054a..2aca01d056 100644 --- a/pseudo-merge.h +++ b/pseudo-merge.h @@ -206,4 +206,11 @@ int cascade_pseudo_merges(const struct pseudo_merge_map *pm, struct bitmap *result, struct bitmap *roots); +/* + * Returns a pseudo-merge which contains the exact set of commits + * listed in the "parents" bitamp, or NULL if none could be found. + */ +struct pseudo_merge *pseudo_merge_for_parents(const struct pseudo_merge_map *pm, + struct bitmap *parents); + #endif diff --git a/t/t5333-pseudo-merge-bitmaps.sh b/t/t5333-pseudo-merge-bitmaps.sh index 4c9aebcffd..f052f395a7 100755 --- a/t/t5333-pseudo-merge-bitmaps.sh +++ b/t/t5333-pseudo-merge-bitmaps.sh @@ -22,6 +22,10 @@ test_pseudo_merges_cascades () { test_trace2_data bitmap pseudo_merges_cascades "$1" } +test_pseudo_merges_reused () { + test_trace2_data pack-bitmap-write building_bitmaps_pseudo_merge_reused "$1" +} + tag_everything () { git rev-list --all --no-object-names >in && perl -lne ' @@ -325,4 +329,65 @@ test_expect_success 'pseudo-merge overlap stale traversal' ' ) ' +test_expect_success 'pseudo-merge reuse' ' + git init pseudo-merge-reuse && + ( + cd pseudo-merge-reuse && + + stable="1641013200" && # 2022-01-01 + unstable="1672549200" && # 2023-01-01 + + GIT_COMMITTER_DATE="$stable +0000" && + export GIT_COMMITTER_DATE && + test_commit_bulk --notick 128 && + GIT_COMMITTER_DATE="$unstable +0000" && + export GIT_COMMITTER_DATE && + test_commit_bulk --notick 128 && + + tag_everything && + + git \ + -c bitmapPseudoMerge.test.pattern="refs/tags/" \ + -c bitmapPseudoMerge.test.maxMerges=1 \ + -c bitmapPseudoMerge.test.threshold=now \ + -c bitmapPseudoMerge.test.stableThreshold=$(($unstable - 1)) \ + -c bitmapPseudoMerge.test.stableSize=512 \ + repack -adb && + + test_pseudo_merges >merges && + test_line_count = 2 merges && + + test_pseudo_merge_commits 0 >stable-oids.before && + test_pseudo_merge_commits 1 >unstable-oids.before && + + : >trace2.txt && + GIT_TRACE2_EVENT=$PWD/trace2.txt git \ + -c bitmapPseudoMerge.test.pattern="refs/tags/" \ + -c bitmapPseudoMerge.test.maxMerges=2 \ + -c bitmapPseudoMerge.test.threshold=now \ + -c bitmapPseudoMerge.test.stableThreshold=$(($unstable - 1)) \ + -c bitmapPseudoMerge.test.stableSize=512 \ + repack -adb && + + test_pseudo_merges_reused 1 merges && + test_line_count = 3 merges && + + test_pseudo_merge_commits 0 >stable-oids.after && + for i in 1 2 + do + test_pseudo_merge_commits $i || return 1 + done >unstable-oids.after && + + sort -u expect && + sort -u actual && + test_cmp expect actual && + + sort -u expect && + sort -u actual && + test_cmp expect actual + ) +' + test_done