The sixth batch
[git.git] / delta-islands.c
blob2186bd0738ed2fcbe216cf24a6c99ed4dc4ccd9b
1 #include "cache.h"
2 #include "attr.h"
3 #include "object.h"
4 #include "blob.h"
5 #include "commit.h"
6 #include "tag.h"
7 #include "tree.h"
8 #include "delta.h"
9 #include "pack.h"
10 #include "tree-walk.h"
11 #include "diff.h"
12 #include "revision.h"
13 #include "list-objects.h"
14 #include "progress.h"
15 #include "refs.h"
16 #include "khash.h"
17 #include "pack-bitmap.h"
18 #include "pack-objects.h"
19 #include "delta-islands.h"
20 #include "sha1-array.h"
21 #include "config.h"
23 KHASH_INIT(str, const char *, void *, 1, kh_str_hash_func, kh_str_hash_equal)
25 static khash_sha1 *island_marks;
26 static unsigned island_counter;
27 static unsigned island_counter_core;
29 static kh_str_t *remote_islands;
31 struct remote_island {
32 uint64_t hash;
33 struct oid_array oids;
36 struct island_bitmap {
37 uint32_t refcount;
38 uint32_t bits[FLEX_ARRAY];
41 static uint32_t island_bitmap_size;
44 * Allocate a new bitmap; if "old" is not NULL, the new bitmap will be a copy
45 * of "old". Otherwise, the new bitmap is empty.
47 static struct island_bitmap *island_bitmap_new(const struct island_bitmap *old)
49 size_t size = sizeof(struct island_bitmap) + (island_bitmap_size * 4);
50 struct island_bitmap *b = xcalloc(1, size);
52 if (old)
53 memcpy(b, old, size);
55 b->refcount = 1;
56 return b;
59 static void island_bitmap_or(struct island_bitmap *a, const struct island_bitmap *b)
61 uint32_t i;
63 for (i = 0; i < island_bitmap_size; ++i)
64 a->bits[i] |= b->bits[i];
67 static int island_bitmap_is_subset(struct island_bitmap *self,
68 struct island_bitmap *super)
70 uint32_t i;
72 if (self == super)
73 return 1;
75 for (i = 0; i < island_bitmap_size; ++i) {
76 if ((self->bits[i] & super->bits[i]) != self->bits[i])
77 return 0;
80 return 1;
83 #define ISLAND_BITMAP_BLOCK(x) (x / 32)
84 #define ISLAND_BITMAP_MASK(x) (1 << (x % 32))
86 static void island_bitmap_set(struct island_bitmap *self, uint32_t i)
88 self->bits[ISLAND_BITMAP_BLOCK(i)] |= ISLAND_BITMAP_MASK(i);
91 static int island_bitmap_get(struct island_bitmap *self, uint32_t i)
93 return (self->bits[ISLAND_BITMAP_BLOCK(i)] & ISLAND_BITMAP_MASK(i)) != 0;
96 int in_same_island(const struct object_id *trg_oid, const struct object_id *src_oid)
98 khiter_t trg_pos, src_pos;
100 /* If we aren't using islands, assume everything goes together. */
101 if (!island_marks)
102 return 1;
105 * If we don't have a bitmap for the target, we can delta it
106 * against anything -- it's not an important object
108 trg_pos = kh_get_sha1(island_marks, trg_oid->hash);
109 if (trg_pos >= kh_end(island_marks))
110 return 1;
113 * if the source (our delta base) doesn't have a bitmap,
114 * we don't want to base any deltas on it!
116 src_pos = kh_get_sha1(island_marks, src_oid->hash);
117 if (src_pos >= kh_end(island_marks))
118 return 0;
120 return island_bitmap_is_subset(kh_value(island_marks, trg_pos),
121 kh_value(island_marks, src_pos));
124 int island_delta_cmp(const struct object_id *a, const struct object_id *b)
126 khiter_t a_pos, b_pos;
127 struct island_bitmap *a_bitmap = NULL, *b_bitmap = NULL;
129 if (!island_marks)
130 return 0;
132 a_pos = kh_get_sha1(island_marks, a->hash);
133 if (a_pos < kh_end(island_marks))
134 a_bitmap = kh_value(island_marks, a_pos);
136 b_pos = kh_get_sha1(island_marks, b->hash);
137 if (b_pos < kh_end(island_marks))
138 b_bitmap = kh_value(island_marks, b_pos);
140 if (a_bitmap) {
141 if (!b_bitmap || !island_bitmap_is_subset(a_bitmap, b_bitmap))
142 return -1;
144 if (b_bitmap) {
145 if (!a_bitmap || !island_bitmap_is_subset(b_bitmap, a_bitmap))
146 return 1;
149 return 0;
152 static struct island_bitmap *create_or_get_island_marks(struct object *obj)
154 khiter_t pos;
155 int hash_ret;
157 pos = kh_put_sha1(island_marks, obj->oid.hash, &hash_ret);
158 if (hash_ret)
159 kh_value(island_marks, pos) = island_bitmap_new(NULL);
161 return kh_value(island_marks, pos);
164 static void set_island_marks(struct object *obj, struct island_bitmap *marks)
166 struct island_bitmap *b;
167 khiter_t pos;
168 int hash_ret;
170 pos = kh_put_sha1(island_marks, obj->oid.hash, &hash_ret);
171 if (hash_ret) {
173 * We don't have one yet; make a copy-on-write of the
174 * parent.
176 marks->refcount++;
177 kh_value(island_marks, pos) = marks;
178 return;
182 * We do have it. Make sure we split any copy-on-write before
183 * updating.
185 b = kh_value(island_marks, pos);
186 if (b->refcount > 1) {
187 b->refcount--;
188 b = kh_value(island_marks, pos) = island_bitmap_new(b);
190 island_bitmap_or(b, marks);
193 static void mark_remote_island_1(struct repository *r,
194 struct remote_island *rl,
195 int is_core_island)
197 uint32_t i;
199 for (i = 0; i < rl->oids.nr; ++i) {
200 struct island_bitmap *marks;
201 struct object *obj = parse_object(r, &rl->oids.oid[i]);
203 if (!obj)
204 continue;
206 marks = create_or_get_island_marks(obj);
207 island_bitmap_set(marks, island_counter);
209 if (is_core_island && obj->type == OBJ_COMMIT)
210 obj->flags |= NEEDS_BITMAP;
212 /* If it was a tag, also make sure we hit the underlying object. */
213 while (obj && obj->type == OBJ_TAG) {
214 obj = ((struct tag *)obj)->tagged;
215 if (obj) {
216 parse_object(r, &obj->oid);
217 marks = create_or_get_island_marks(obj);
218 island_bitmap_set(marks, island_counter);
223 if (is_core_island)
224 island_counter_core = island_counter;
226 island_counter++;
229 struct tree_islands_todo {
230 struct object_entry *entry;
231 unsigned int depth;
234 static int tree_depth_compare(const void *a, const void *b)
236 const struct tree_islands_todo *todo_a = a;
237 const struct tree_islands_todo *todo_b = b;
239 return todo_a->depth - todo_b->depth;
242 void resolve_tree_islands(struct repository *r,
243 int progress,
244 struct packing_data *to_pack)
246 struct progress *progress_state = NULL;
247 struct tree_islands_todo *todo;
248 int nr = 0;
249 int i;
251 if (!island_marks)
252 return;
255 * We process only trees, as commits and tags have already been handled
256 * (and passed their marks on to root trees, as well. We must make sure
257 * to process them in descending tree-depth order so that marks
258 * propagate down the tree properly, even if a sub-tree is found in
259 * multiple parent trees.
261 ALLOC_ARRAY(todo, to_pack->nr_objects);
262 for (i = 0; i < to_pack->nr_objects; i++) {
263 if (oe_type(&to_pack->objects[i]) == OBJ_TREE) {
264 todo[nr].entry = &to_pack->objects[i];
265 todo[nr].depth = oe_tree_depth(to_pack, &to_pack->objects[i]);
266 nr++;
269 QSORT(todo, nr, tree_depth_compare);
271 if (progress)
272 progress_state = start_progress(_("Propagating island marks"), nr);
274 for (i = 0; i < nr; i++) {
275 struct object_entry *ent = todo[i].entry;
276 struct island_bitmap *root_marks;
277 struct tree *tree;
278 struct tree_desc desc;
279 struct name_entry entry;
280 khiter_t pos;
282 pos = kh_get_sha1(island_marks, ent->idx.oid.hash);
283 if (pos >= kh_end(island_marks))
284 continue;
286 root_marks = kh_value(island_marks, pos);
288 tree = lookup_tree(r, &ent->idx.oid);
289 if (!tree || parse_tree(tree) < 0)
290 die(_("bad tree object %s"), oid_to_hex(&ent->idx.oid));
292 init_tree_desc(&desc, tree->buffer, tree->size);
293 while (tree_entry(&desc, &entry)) {
294 struct object *obj;
296 if (S_ISGITLINK(entry.mode))
297 continue;
299 obj = lookup_object(r, entry.oid.hash);
300 if (!obj)
301 continue;
303 set_island_marks(obj, root_marks);
306 free_tree_buffer(tree);
308 display_progress(progress_state, i+1);
311 stop_progress(&progress_state);
312 free(todo);
315 static regex_t *island_regexes;
316 static unsigned int island_regexes_alloc, island_regexes_nr;
317 static const char *core_island_name;
319 static int island_config_callback(const char *k, const char *v, void *cb)
321 if (!strcmp(k, "pack.island")) {
322 struct strbuf re = STRBUF_INIT;
324 if (!v)
325 return config_error_nonbool(k);
327 ALLOC_GROW(island_regexes, island_regexes_nr + 1, island_regexes_alloc);
329 if (*v != '^')
330 strbuf_addch(&re, '^');
331 strbuf_addstr(&re, v);
333 if (regcomp(&island_regexes[island_regexes_nr], re.buf, REG_EXTENDED))
334 die(_("failed to load island regex for '%s': %s"), k, re.buf);
336 strbuf_release(&re);
337 island_regexes_nr++;
338 return 0;
341 if (!strcmp(k, "pack.islandcore"))
342 return git_config_string(&core_island_name, k, v);
344 return 0;
347 static void add_ref_to_island(const char *island_name, const struct object_id *oid)
349 uint64_t sha_core;
350 struct remote_island *rl = NULL;
352 int hash_ret;
353 khiter_t pos = kh_put_str(remote_islands, island_name, &hash_ret);
355 if (hash_ret) {
356 kh_key(remote_islands, pos) = xstrdup(island_name);
357 kh_value(remote_islands, pos) = xcalloc(1, sizeof(struct remote_island));
360 rl = kh_value(remote_islands, pos);
361 oid_array_append(&rl->oids, oid);
363 memcpy(&sha_core, oid->hash, sizeof(uint64_t));
364 rl->hash += sha_core;
367 static int find_island_for_ref(const char *refname, const struct object_id *oid,
368 int flags, void *data)
371 * We should advertise 'ARRAY_SIZE(matches) - 2' as the max,
372 * so we can diagnose below a config with more capture groups
373 * than we support.
375 regmatch_t matches[16];
376 int i, m;
377 struct strbuf island_name = STRBUF_INIT;
379 /* walk backwards to get last-one-wins ordering */
380 for (i = island_regexes_nr - 1; i >= 0; i--) {
381 if (!regexec(&island_regexes[i], refname,
382 ARRAY_SIZE(matches), matches, 0))
383 break;
386 if (i < 0)
387 return 0;
389 if (matches[ARRAY_SIZE(matches) - 1].rm_so != -1)
390 warning(_("island regex from config has "
391 "too many capture groups (max=%d)"),
392 (int)ARRAY_SIZE(matches) - 2);
394 for (m = 1; m < ARRAY_SIZE(matches); m++) {
395 regmatch_t *match = &matches[m];
397 if (match->rm_so == -1)
398 continue;
400 if (island_name.len)
401 strbuf_addch(&island_name, '-');
403 strbuf_add(&island_name, refname + match->rm_so, match->rm_eo - match->rm_so);
406 add_ref_to_island(island_name.buf, oid);
407 strbuf_release(&island_name);
408 return 0;
411 static struct remote_island *get_core_island(void)
413 if (core_island_name) {
414 khiter_t pos = kh_get_str(remote_islands, core_island_name);
415 if (pos < kh_end(remote_islands))
416 return kh_value(remote_islands, pos);
419 return NULL;
422 static void deduplicate_islands(struct repository *r)
424 struct remote_island *island, *core = NULL, **list;
425 unsigned int island_count, dst, src, ref, i = 0;
427 island_count = kh_size(remote_islands);
428 ALLOC_ARRAY(list, island_count);
430 kh_foreach_value(remote_islands, island, {
431 list[i++] = island;
434 for (ref = 0; ref + 1 < island_count; ref++) {
435 for (src = ref + 1, dst = src; src < island_count; src++) {
436 if (list[ref]->hash == list[src]->hash)
437 continue;
439 if (src != dst)
440 list[dst] = list[src];
442 dst++;
444 island_count = dst;
447 island_bitmap_size = (island_count / 32) + 1;
448 core = get_core_island();
450 for (i = 0; i < island_count; ++i) {
451 mark_remote_island_1(r, list[i], core && list[i]->hash == core->hash);
454 free(list);
457 void load_delta_islands(struct repository *r)
459 island_marks = kh_init_sha1();
460 remote_islands = kh_init_str();
462 git_config(island_config_callback, NULL);
463 for_each_ref(find_island_for_ref, NULL);
464 deduplicate_islands(r);
466 fprintf(stderr, _("Marked %d islands, done.\n"), island_counter);
469 void propagate_island_marks(struct commit *commit)
471 khiter_t pos = kh_get_sha1(island_marks, commit->object.oid.hash);
473 if (pos < kh_end(island_marks)) {
474 struct commit_list *p;
475 struct island_bitmap *root_marks = kh_value(island_marks, pos);
477 parse_commit(commit);
478 set_island_marks(&get_commit_tree(commit)->object, root_marks);
479 for (p = commit->parents; p; p = p->next)
480 set_island_marks(&p->item->object, root_marks);
484 int compute_pack_layers(struct packing_data *to_pack)
486 uint32_t i;
488 if (!core_island_name || !island_marks)
489 return 1;
491 for (i = 0; i < to_pack->nr_objects; ++i) {
492 struct object_entry *entry = &to_pack->objects[i];
493 khiter_t pos = kh_get_sha1(island_marks, entry->idx.oid.hash);
495 oe_set_layer(to_pack, entry, 1);
497 if (pos < kh_end(island_marks)) {
498 struct island_bitmap *bitmap = kh_value(island_marks, pos);
500 if (island_bitmap_get(bitmap, island_counter_core))
501 oe_set_layer(to_pack, entry, 0);
505 return 2;