The sixteenth batch
[git/gitster.git] / sparse-index.c
blob5634abafaa07ed77ad24d10457b7f1f3ec6108ec
1 #define USE_THE_REPOSITORY_VARIABLE
2 #define DISABLE_SIGN_COMPARE_WARNINGS
4 #include "git-compat-util.h"
5 #include "environment.h"
6 #include "ewah/ewok.h"
7 #include "gettext.h"
8 #include "name-hash.h"
9 #include "read-cache-ll.h"
10 #include "repository.h"
11 #include "sparse-index.h"
12 #include "tree.h"
13 #include "pathspec.h"
14 #include "trace2.h"
15 #include "cache-tree.h"
16 #include "config.h"
17 #include "dir.h"
18 #include "fsmonitor-ll.h"
19 #include "advice.h"
21 /**
22 * This global is used by expand_index() to determine if we should give the
23 * advice for advice.sparseIndexExpanded when expanding a sparse index to a full
24 * one. However, this is sometimes done on purpose, such as in the sparse-checkout
25 * builtin, even when index.sparse=false. This may be disabled in
26 * convert_to_sparse() or by commands that know they will lead to a full
27 * expansion, but this message is not actionable.
29 int give_advice_on_expansion = 1;
30 #define ADVICE_MSG \
31 "The sparse index is expanding to a full index, a slow operation.\n" \
32 "Your working directory likely has contents that are outside of\n" \
33 "your sparse-checkout patterns. Use 'git sparse-checkout list' to\n" \
34 "see your sparse-checkout definition and compare it to your working\n" \
35 "directory contents. Running 'git clean' may assist in this cleanup."
37 struct modify_index_context {
38 struct index_state *write;
39 struct pattern_list *pl;
42 static struct cache_entry *construct_sparse_dir_entry(
43 struct index_state *istate,
44 const char *sparse_dir,
45 struct cache_tree *tree)
47 struct cache_entry *de;
49 de = make_cache_entry(istate, S_IFDIR, &tree->oid, sparse_dir, 0, 0);
51 de->ce_flags |= CE_SKIP_WORKTREE;
52 return de;
56 * Returns the number of entries "inserted" into the index.
58 static int convert_to_sparse_rec(struct index_state *istate,
59 int num_converted,
60 int start, int end,
61 const char *ct_path, size_t ct_pathlen,
62 struct cache_tree *ct)
64 int i, can_convert = 1;
65 int start_converted = num_converted;
66 struct strbuf child_path = STRBUF_INIT;
69 * Is the current path outside of the sparse cone?
70 * Then check if the region can be replaced by a sparse
71 * directory entry (everything is sparse and merged).
73 if (path_in_sparse_checkout(ct_path, istate))
74 can_convert = 0;
76 for (i = start; can_convert && i < end; i++) {
77 struct cache_entry *ce = istate->cache[i];
79 if (ce_stage(ce) ||
80 S_ISGITLINK(ce->ce_mode) ||
81 !(ce->ce_flags & CE_SKIP_WORKTREE))
82 can_convert = 0;
85 if (can_convert) {
86 struct cache_entry *se;
87 se = construct_sparse_dir_entry(istate, ct_path, ct);
89 istate->cache[num_converted++] = se;
90 return 1;
93 for (i = start; i < end; ) {
94 int count, span, pos = -1;
95 const char *base, *slash;
96 struct cache_entry *ce = istate->cache[i];
99 * Detect if this is a normal entry outside of any subtree
100 * entry.
102 base = ce->name + ct_pathlen;
103 slash = strchr(base, '/');
105 if (slash)
106 pos = cache_tree_subtree_pos(ct, base, slash - base);
108 if (pos < 0) {
109 istate->cache[num_converted++] = ce;
110 i++;
111 continue;
114 strbuf_setlen(&child_path, 0);
115 strbuf_add(&child_path, ce->name, slash - ce->name + 1);
117 span = ct->down[pos]->cache_tree->entry_count;
118 count = convert_to_sparse_rec(istate,
119 num_converted, i, i + span,
120 child_path.buf, child_path.len,
121 ct->down[pos]->cache_tree);
122 num_converted += count;
123 i += span;
126 strbuf_release(&child_path);
127 return num_converted - start_converted;
130 int set_sparse_index_config(struct repository *repo, int enable)
132 int res = repo_config_set_worktree_gently(repo,
133 "index.sparse",
134 enable ? "true" : "false");
135 prepare_repo_settings(repo);
136 repo->settings.sparse_index = enable;
137 return res;
140 static int index_has_unmerged_entries(struct index_state *istate)
142 int i;
143 for (i = 0; i < istate->cache_nr; i++) {
144 if (ce_stage(istate->cache[i]))
145 return 1;
148 return 0;
151 int is_sparse_index_allowed(struct index_state *istate, int flags)
153 if (!core_apply_sparse_checkout || !core_sparse_checkout_cone)
154 return 0;
156 if (!(flags & SPARSE_INDEX_MEMORY_ONLY)) {
157 int test_env;
160 * The sparse index is not (yet) integrated with a split index.
162 if (istate->split_index || git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
163 return 0;
165 * The GIT_TEST_SPARSE_INDEX environment variable triggers the
166 * index.sparse config variable to be on.
168 test_env = git_env_bool("GIT_TEST_SPARSE_INDEX", -1);
169 if (test_env >= 0)
170 set_sparse_index_config(istate->repo, test_env);
173 * Only convert to sparse if index.sparse is set.
175 prepare_repo_settings(istate->repo);
176 if (!istate->repo->settings.sparse_index)
177 return 0;
180 if (init_sparse_checkout_patterns(istate))
181 return 0;
184 * We need cone-mode patterns to use sparse-index. If a user edits
185 * their sparse-checkout file manually, then we can detect during
186 * parsing that they are not actually using cone-mode patterns and
187 * hence we need to abort this conversion _without error_. Warnings
188 * already exist in the pattern parsing to inform the user of their
189 * bad patterns.
191 if (!istate->sparse_checkout_patterns->use_cone_patterns)
192 return 0;
194 return 1;
197 int convert_to_sparse(struct index_state *istate, int flags)
200 * If the index is already sparse, empty, or otherwise
201 * cannot be converted to sparse, do not convert.
203 if (istate->sparse_index == INDEX_COLLAPSED || !istate->cache_nr ||
204 !is_sparse_index_allowed(istate, flags))
205 return 0;
208 * If we are purposefully collapsing a full index, then don't give
209 * advice when it is expanded later.
211 give_advice_on_expansion = 0;
214 * NEEDSWORK: If we have unmerged entries, then stay full.
215 * Unmerged entries prevent the cache-tree extension from working.
217 if (index_has_unmerged_entries(istate))
218 return 0;
220 if (!cache_tree_fully_valid(istate->cache_tree)) {
221 /* Clear and recompute the cache-tree */
222 cache_tree_free(&istate->cache_tree);
225 * Silently return if there is a problem with the cache tree update,
226 * which might just be due to a conflict state in some entry.
228 * This might create new tree objects, so be sure to use
229 * WRITE_TREE_MISSING_OK.
231 if (cache_tree_update(istate, WRITE_TREE_MISSING_OK))
232 return 0;
235 remove_fsmonitor(istate);
237 trace2_region_enter("index", "convert_to_sparse", istate->repo);
238 istate->cache_nr = convert_to_sparse_rec(istate,
239 0, 0, istate->cache_nr,
240 "", 0, istate->cache_tree);
242 /* Clear and recompute the cache-tree */
243 cache_tree_free(&istate->cache_tree);
244 cache_tree_update(istate, 0);
246 istate->fsmonitor_has_run_once = 0;
247 ewah_free(istate->fsmonitor_dirty);
248 istate->fsmonitor_dirty = NULL;
249 FREE_AND_NULL(istate->fsmonitor_last_update);
251 istate->sparse_index = INDEX_COLLAPSED;
252 trace2_region_leave("index", "convert_to_sparse", istate->repo);
253 return 0;
256 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
258 ALLOC_GROW(istate->cache, nr + 1, istate->cache_alloc);
260 istate->cache[nr] = ce;
261 add_name_hash(istate, ce);
264 static int add_path_to_index(const struct object_id *oid,
265 struct strbuf *base, const char *path,
266 unsigned int mode, void *context)
268 struct modify_index_context *ctx = (struct modify_index_context *)context;
269 struct cache_entry *ce;
270 size_t len = base->len;
272 if (S_ISDIR(mode)) {
273 int dtype;
274 size_t baselen = base->len;
275 if (!ctx->pl)
276 return READ_TREE_RECURSIVE;
279 * Have we expanded to a point outside of the sparse-checkout?
281 * Artificially pad the path name with a slash "/" to
282 * indicate it as a directory, and add an arbitrary file
283 * name ("-") so we can consider base->buf as a file name
284 * to match against the cone-mode patterns.
286 * If we compared just "path", then we would expand more
287 * than we should. Since every file at root is always
288 * included, we would expand every directory at root at
289 * least one level deep instead of using sparse directory
290 * entries.
292 strbuf_addstr(base, path);
293 strbuf_add(base, "/-", 2);
295 if (path_matches_pattern_list(base->buf, base->len,
296 NULL, &dtype,
297 ctx->pl, ctx->write)) {
298 strbuf_setlen(base, baselen);
299 return READ_TREE_RECURSIVE;
303 * The path "{base}{path}/" is a sparse directory. Create the correct
304 * name for inserting the entry into the index.
306 strbuf_setlen(base, base->len - 1);
307 } else {
308 strbuf_addstr(base, path);
311 ce = make_cache_entry(ctx->write, mode, oid, base->buf, 0, 0);
312 ce->ce_flags |= CE_SKIP_WORKTREE | CE_EXTENDED;
313 set_index_entry(ctx->write, ctx->write->cache_nr++, ce);
315 strbuf_setlen(base, len);
316 return 0;
319 void expand_index(struct index_state *istate, struct pattern_list *pl)
321 int i;
322 struct index_state *full;
323 struct strbuf base = STRBUF_INIT;
324 const char *tr_region;
325 struct modify_index_context ctx;
328 * If the index is already full, then keep it full. We will convert
329 * it to a sparse index on write, if possible.
331 if (istate->sparse_index == INDEX_EXPANDED)
332 return;
335 * If our index is sparse, but our new pattern set does not use
336 * cone mode patterns, then we need to expand the index before we
337 * continue. A NULL pattern set indicates a full expansion to a
338 * full index.
340 if (pl && !pl->use_cone_patterns) {
341 pl = NULL;
342 } else {
344 * We might contract file entries into sparse-directory
345 * entries, and for that we will need the cache tree to
346 * be recomputed.
348 cache_tree_free(&istate->cache_tree);
351 * If there is a problem creating the cache tree, then we
352 * need to expand to a full index since we cannot satisfy
353 * the current request as a sparse index.
355 if (cache_tree_update(istate, 0))
356 pl = NULL;
359 if (!pl && give_advice_on_expansion) {
360 give_advice_on_expansion = 0;
361 advise_if_enabled(ADVICE_SPARSE_INDEX_EXPANDED,
362 _(ADVICE_MSG));
366 * A NULL pattern set indicates we are expanding a full index, so
367 * we use a special region name that indicates the full expansion.
368 * This is used by test cases, but also helps to differentiate the
369 * two cases.
371 tr_region = pl ? "expand_index" : "ensure_full_index";
372 trace2_region_enter("index", tr_region, istate->repo);
374 /* initialize basics of new index */
375 full = xcalloc(1, sizeof(struct index_state));
376 memcpy(full, istate, sizeof(struct index_state));
379 * This slightly-misnamed 'full' index might still be sparse if we
380 * are only modifying the list of sparse directories. This hinges
381 * on whether we have a non-NULL pattern list.
383 full->sparse_index = pl ? INDEX_PARTIALLY_SPARSE : INDEX_EXPANDED;
385 /* then change the necessary things */
386 full->cache_alloc = (3 * istate->cache_alloc) / 2;
387 full->cache_nr = 0;
388 ALLOC_ARRAY(full->cache, full->cache_alloc);
390 ctx.write = full;
391 ctx.pl = pl;
393 for (i = 0; i < istate->cache_nr; i++) {
394 struct cache_entry *ce = istate->cache[i];
395 struct tree *tree;
396 struct pathspec ps;
397 int dtype;
399 if (!S_ISSPARSEDIR(ce->ce_mode)) {
400 set_index_entry(full, full->cache_nr++, ce);
401 continue;
404 /* We now have a sparse directory entry. Should we expand? */
405 if (pl &&
406 path_matches_pattern_list(ce->name, ce->ce_namelen,
407 NULL, &dtype,
408 pl, istate) == NOT_MATCHED) {
409 set_index_entry(full, full->cache_nr++, ce);
410 continue;
413 if (!(ce->ce_flags & CE_SKIP_WORKTREE))
414 warning(_("index entry is a directory, but not sparse (%08x)"),
415 ce->ce_flags);
417 /* recursively walk into cd->name */
418 tree = lookup_tree(istate->repo, &ce->oid);
420 memset(&ps, 0, sizeof(ps));
421 ps.recursive = 1;
422 ps.has_wildcard = 1;
423 ps.max_depth = -1;
425 strbuf_setlen(&base, 0);
426 strbuf_add(&base, ce->name, strlen(ce->name));
428 read_tree_at(istate->repo, tree, &base, 0, &ps,
429 add_path_to_index, &ctx);
431 /* free directory entries. full entries are re-used */
432 discard_cache_entry(ce);
435 /* Copy back into original index. */
436 memcpy(&istate->name_hash, &full->name_hash, sizeof(full->name_hash));
437 memcpy(&istate->dir_hash, &full->dir_hash, sizeof(full->dir_hash));
438 istate->sparse_index = pl ? INDEX_PARTIALLY_SPARSE : INDEX_EXPANDED;
439 free(istate->cache);
440 istate->cache = full->cache;
441 istate->cache_nr = full->cache_nr;
442 istate->cache_alloc = full->cache_alloc;
443 istate->fsmonitor_has_run_once = 0;
444 ewah_free(istate->fsmonitor_dirty);
445 istate->fsmonitor_dirty = NULL;
446 FREE_AND_NULL(istate->fsmonitor_last_update);
448 strbuf_release(&base);
449 free(full);
451 /* Clear and recompute the cache-tree */
452 cache_tree_free(&istate->cache_tree);
453 cache_tree_update(istate, 0);
455 trace2_region_leave("index", tr_region, istate->repo);
458 void ensure_full_index(struct index_state *istate)
460 if (!istate)
461 BUG("ensure_full_index() must get an index!");
462 expand_index(istate, NULL);
465 void ensure_correct_sparsity(struct index_state *istate)
468 * If the index can be sparse, make it sparse. Otherwise,
469 * ensure the index is full.
471 if (is_sparse_index_allowed(istate, 0))
472 convert_to_sparse(istate, 0);
473 else
474 ensure_full_index(istate);
477 struct path_found_data {
479 * The path stored in 'dir', if non-empty, corresponds to the most-
480 * recent path that we checked where:
482 * 1. The path should be a directory, according to the index.
483 * 2. The path does not exist.
484 * 3. The parent path _does_ exist. (This may be the root of the
485 * working directory.)
487 struct strbuf dir;
488 size_t lstat_count;
491 #define PATH_FOUND_DATA_INIT { \
492 .dir = STRBUF_INIT \
495 static void clear_path_found_data(struct path_found_data *data)
497 strbuf_release(&data->dir);
501 * Return the length of the longest common substring that ends in a
502 * slash ('/') to indicate the longest common parent directory. Returns
503 * zero if no common directory exists.
505 static size_t max_common_dir_prefix(const char *path1, const char *path2)
507 size_t common_prefix = 0;
508 for (size_t i = 0; path1[i] && path2[i]; i++) {
509 if (path1[i] != path2[i])
510 break;
513 * If they agree at a directory separator, then add one
514 * to make sure it is included in the common prefix string.
516 if (path1[i] == '/')
517 common_prefix = i + 1;
520 return common_prefix;
523 static int path_found(const char *path, struct path_found_data *data)
525 struct stat st;
526 size_t common_prefix;
529 * If data->dir is non-empty, then it contains a path that doesn't
530 * exist, including an ending slash ('/'). If it is a prefix of 'path',
531 * then we can return 0.
533 if (data->dir.len && !memcmp(path, data->dir.buf, data->dir.len))
534 return 0;
537 * Otherwise, we must check if the current path exists. If it does, then
538 * return 1. The cached directory will be skipped until we come across
539 * a missing path again.
541 data->lstat_count++;
542 if (!lstat(path, &st))
543 return 1;
546 * At this point, we know that 'path' doesn't exist, and we know that
547 * the parent directory of 'data->dir' does exist. Let's set 'data->dir'
548 * to be the top-most non-existing directory of 'path'. If the first
549 * parent of 'path' exists, then we will act as though 'path'
550 * corresponds to a directory (by adding a slash).
552 common_prefix = max_common_dir_prefix(path, data->dir.buf);
555 * At this point, 'path' and 'data->dir' have a common existing parent
556 * directory given by path[0..common_prefix] (which could have length 0).
557 * We "grow" the data->dir buffer by checking for existing directories
558 * along 'path'.
561 strbuf_setlen(&data->dir, common_prefix);
562 while (1) {
563 /* Find the next directory in 'path'. */
564 const char *rest = path + data->dir.len;
565 const char *next_slash = strchr(rest, '/');
568 * If there are no more slashes, then 'path' doesn't contain a
569 * non-existent _parent_ directory. Set 'data->dir' to be equal
570 * to 'path' plus an additional slash, so it can be used for
571 * caching in the future. The filename of 'path' is considered
572 * a non-existent directory.
574 * Note: if "{path}/" exists as a directory, then it will never
575 * appear as a prefix of other callers to this method, assuming
576 * the context from the clear_skip_worktree... methods. If this
577 * method is reused, then this must be reconsidered.
579 if (!next_slash) {
580 strbuf_addstr(&data->dir, rest);
581 strbuf_addch(&data->dir, '/');
582 break;
586 * Now that we have a slash, let's grow 'data->dir' to include
587 * this slash, then test if we should stop.
589 strbuf_add(&data->dir, rest, next_slash - rest + 1);
591 /* If the parent dir doesn't exist, then stop here. */
592 data->lstat_count++;
593 if (lstat(data->dir.buf, &st))
594 return 0;
598 * At this point, 'data->dir' is equal to 'path' plus a slash character,
599 * and the parent directory of 'path' definitely exists. Moreover, we
600 * know that 'path' doesn't exist, or we would have returned 1 earlier.
602 return 0;
605 static int clear_skip_worktree_from_present_files_sparse(struct index_state *istate)
607 struct path_found_data data = PATH_FOUND_DATA_INIT;
609 int path_count = 0;
610 int to_restart = 0;
612 trace2_region_enter("index", "clear_skip_worktree_from_present_files_sparse",
613 istate->repo);
614 for (int i = 0; i < istate->cache_nr; i++) {
615 struct cache_entry *ce = istate->cache[i];
617 if (ce_skip_worktree(ce)) {
618 path_count++;
619 if (path_found(ce->name, &data)) {
620 if (S_ISSPARSEDIR(ce->ce_mode)) {
621 to_restart = 1;
622 break;
624 ce->ce_flags &= ~CE_SKIP_WORKTREE;
629 trace2_data_intmax("index", istate->repo,
630 "sparse_path_count", path_count);
631 trace2_data_intmax("index", istate->repo,
632 "sparse_lstat_count", data.lstat_count);
633 trace2_region_leave("index", "clear_skip_worktree_from_present_files_sparse",
634 istate->repo);
635 clear_path_found_data(&data);
636 return to_restart;
639 static void clear_skip_worktree_from_present_files_full(struct index_state *istate)
641 struct path_found_data data = PATH_FOUND_DATA_INIT;
643 int path_count = 0;
645 trace2_region_enter("index", "clear_skip_worktree_from_present_files_full",
646 istate->repo);
647 for (int i = 0; i < istate->cache_nr; i++) {
648 struct cache_entry *ce = istate->cache[i];
650 if (S_ISSPARSEDIR(ce->ce_mode))
651 BUG("ensure-full-index did not fully flatten?");
653 if (ce_skip_worktree(ce)) {
654 path_count++;
655 if (path_found(ce->name, &data))
656 ce->ce_flags &= ~CE_SKIP_WORKTREE;
660 trace2_data_intmax("index", istate->repo,
661 "full_path_count", path_count);
662 trace2_data_intmax("index", istate->repo,
663 "full_lstat_count", data.lstat_count);
664 trace2_region_leave("index", "clear_skip_worktree_from_present_files_full",
665 istate->repo);
666 clear_path_found_data(&data);
669 void clear_skip_worktree_from_present_files(struct index_state *istate)
671 if (!core_apply_sparse_checkout ||
672 sparse_expect_files_outside_of_patterns)
673 return;
675 if (clear_skip_worktree_from_present_files_sparse(istate)) {
676 ensure_full_index(istate);
677 clear_skip_worktree_from_present_files_full(istate);
682 * This static global helps avoid infinite recursion between
683 * expand_to_path() and index_file_exists().
685 static int in_expand_to_path = 0;
687 void expand_to_path(struct index_state *istate,
688 const char *path, size_t pathlen, int icase)
690 struct strbuf path_mutable = STRBUF_INIT;
691 size_t substr_len;
693 /* prevent extra recursion */
694 if (in_expand_to_path)
695 return;
697 if (!istate->sparse_index)
698 return;
700 in_expand_to_path = 1;
703 * We only need to actually expand a region if the
704 * following are both true:
706 * 1. 'path' is not already in the index.
707 * 2. Some parent directory of 'path' is a sparse directory.
710 if (index_file_exists(istate, path, pathlen, icase))
711 goto cleanup;
713 strbuf_add(&path_mutable, path, pathlen);
714 strbuf_addch(&path_mutable, '/');
716 /* Check the name hash for all parent directories */
717 substr_len = 0;
718 while (substr_len < pathlen) {
719 char temp;
720 char *replace = strchr(path_mutable.buf + substr_len, '/');
722 if (!replace)
723 break;
725 /* replace the character _after_ the slash */
726 replace++;
727 temp = *replace;
728 *replace = '\0';
729 substr_len = replace - path_mutable.buf;
730 if (index_file_exists(istate, path_mutable.buf,
731 substr_len, icase)) {
733 * We found a parent directory in the name-hash
734 * hashtable, because only sparse directory entries
735 * have a trailing '/' character. Since "path" wasn't
736 * in the index, perhaps it exists within this
737 * sparse-directory. Expand accordingly.
739 ensure_full_index(istate);
740 break;
743 *replace = temp;
746 cleanup:
747 strbuf_release(&path_mutable);
748 in_expand_to_path = 0;