Blob


1 /*
2 * Copyright (c) 2018, 2019, 2020 Stefan Sperling <stsp@openbsd.org>
3 *
4 * Permission to use, copy, modify, and distribute this software for any
5 * purpose with or without fee is hereby granted, provided that the above
6 * copyright notice and this permission notice appear in all copies.
7 *
8 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
9 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
10 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
11 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
12 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
13 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
14 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
15 */
17 #include <sys/types.h>
18 #include <sys/stat.h>
19 #include <sys/queue.h>
20 #include <sys/stdint.h>
22 #include <limits.h>
23 #include <stdio.h>
24 #include <stdlib.h>
25 #include <string.h>
26 #include <sha1.h>
27 #include <zlib.h>
28 #include <ctype.h>
30 #include "got_error.h"
31 #include "got_object.h"
32 #include "got_cancel.h"
33 #include "got_commit_graph.h"
34 #include "got_path.h"
36 #include "got_lib_delta.h"
37 #include "got_lib_inflate.h"
38 #include "got_lib_object.h"
39 #include "got_lib_object_idset.h"
41 struct got_commit_graph_node {
42 struct got_object_id id;
44 /* Used only during iteration. */
45 time_t timestamp;
46 TAILQ_ENTRY(got_commit_graph_node) entry;
47 };
49 TAILQ_HEAD(got_commit_graph_iter_list, got_commit_graph_node);
51 struct got_commit_graph_branch_tip {
52 struct got_object_id *commit_id;
53 struct got_commit_object *commit;
54 struct got_commit_graph_node *new_node;
55 };
57 struct got_commit_graph {
58 /* The set of all commits we have traversed. */
59 struct got_object_idset *node_ids;
61 int flags;
62 #define GOT_COMMIT_GRAPH_FIRST_PARENT_TRAVERSAL 0x01
64 /*
65 * A set of object IDs of known parent commits which we have not yet
66 * traversed. Each commit ID in this set represents a branch in commit
67 * history: Either the first-parent branch of the head node, or another
68 * branch corresponding to a traversed merge commit for which we have
69 * not traversed a branch point commit yet.
70 *
71 * Whenever we add a commit with a matching ID to the graph, we remove
72 * its corresponding element from this set, and add new elements for
73 * each of that commit's parent commits which were not traversed yet.
74 *
75 * When API users ask us to fetch more commits, we fetch commits from
76 * all currently open branches. This allows API users to process
77 * commits in linear order even though the history contains branches.
78 */
79 struct got_object_idset *open_branches;
81 /* Array of branch tips for fetch_commits_from_open_branches(). */
82 struct got_commit_graph_branch_tip *tips;
83 int ntips;
85 /* Path of tree entry of interest to the API user. */
86 char *path;
88 /*
89 * Nodes which will be passed to the API user next, sorted by
90 * commit timestmap.
91 */
92 struct got_commit_graph_iter_list iter_list;
93 };
95 static const struct got_error *
96 detect_changed_path(int *changed, struct got_commit_object *commit,
97 struct got_object_id *commit_id, const char *path,
98 struct got_repository *repo)
99 {
100 const struct got_error *err = NULL;
101 struct got_commit_object *pcommit = NULL;
102 struct got_tree_object *tree = NULL, *ptree = NULL;
103 struct got_object_qid *pid;
105 if (got_path_is_root_dir(path)) {
106 *changed = 1;
107 return NULL;
110 *changed = 0;
112 pid = STAILQ_FIRST(&commit->parent_ids);
113 if (pid == NULL) {
114 struct got_object_id *obj_id;
115 err = got_object_id_by_path(&obj_id, repo, commit, path);
116 if (err) {
117 if (err->code == GOT_ERR_NO_TREE_ENTRY)
118 err = NULL;
119 } else
120 *changed = 1; /* The path was created in this commit. */
121 free(obj_id);
122 return err;
125 err = got_object_open_as_tree(&tree, repo, commit->tree_id);
126 if (err)
127 return err;
129 err = got_object_open_as_commit(&pcommit, repo, &pid->id);
130 if (err)
131 goto done;
133 err = got_object_open_as_tree(&ptree, repo, pcommit->tree_id);
134 if (err)
135 goto done;
137 err = got_object_tree_path_changed(changed, tree, ptree, path, repo);
138 done:
139 if (tree)
140 got_object_tree_close(tree);
141 if (ptree)
142 got_object_tree_close(ptree);
143 if (pcommit)
144 got_object_commit_close(pcommit);
145 return err;
148 static void
149 add_node_to_iter_list(struct got_commit_graph *graph,
150 struct got_commit_graph_node *node, time_t committer_time)
152 struct got_commit_graph_node *n, *next;
154 node->timestamp = committer_time;
156 n = TAILQ_FIRST(&graph->iter_list);
157 while (n) {
158 next = TAILQ_NEXT(n, entry);
159 if (next && node->timestamp >= next->timestamp) {
160 TAILQ_INSERT_BEFORE(next, node, entry);
161 return;
163 n = next;
165 TAILQ_INSERT_TAIL(&graph->iter_list, node, entry);
168 static const struct got_error *
169 add_node(struct got_commit_graph_node **new_node,
170 struct got_commit_graph *graph, struct got_object_id *commit_id,
171 struct got_repository *repo)
173 const struct got_error *err = NULL;
174 struct got_commit_graph_node *node;
176 *new_node = NULL;
178 node = calloc(1, sizeof(*node));
179 if (node == NULL)
180 return got_error_from_errno("calloc");
182 memcpy(&node->id, commit_id, sizeof(node->id));
183 err = got_object_idset_add(graph->node_ids, &node->id, NULL);
184 if (err)
185 free(node);
186 else
187 *new_node = node;
188 return err;
191 /*
192 * Ask got-read-pack to traverse first-parent history until a commit is
193 * encountered which modified graph->path, or until the pack file runs
194 * out of relevant commits. This is faster than sending an individual
195 * request for each commit stored in the pack file.
196 */
197 static const struct got_error *
198 packed_first_parent_traversal(int *ncommits_traversed,
199 struct got_commit_graph *graph, struct got_object_id *commit_id,
200 struct got_repository *repo)
202 const struct got_error *err = NULL;
203 struct got_object_id_queue traversed_commits;
204 struct got_object_qid *qid;
206 STAILQ_INIT(&traversed_commits);
207 *ncommits_traversed = 0;
209 err = got_traverse_packed_commits(&traversed_commits,
210 commit_id, graph->path, repo);
211 if (err)
212 return err;
214 /* Add all traversed commits to the graph... */
215 STAILQ_FOREACH(qid, &traversed_commits, entry) {
216 struct got_commit_graph_node *node;
218 if (got_object_idset_contains(graph->open_branches, &qid->id))
219 continue;
220 if (got_object_idset_contains(graph->node_ids, &qid->id))
221 continue;
223 (*ncommits_traversed)++;
225 /* ... except the last commit is the new branch tip. */
226 if (STAILQ_NEXT(qid, entry) == NULL) {
227 err = got_object_idset_add(graph->open_branches,
228 &qid->id, NULL);
229 break;
232 err = add_node(&node, graph, &qid->id, repo);
233 if (err)
234 break;
237 got_object_id_queue_free(&traversed_commits);
238 return err;
241 static const struct got_error *
242 close_branch(struct got_commit_graph *graph, struct got_object_id *commit_id)
244 const struct got_error *err;
246 err = got_object_idset_remove(NULL, graph->open_branches, commit_id);
247 if (err && err->code != GOT_ERR_NO_OBJ)
248 return err;
249 return NULL;
252 static const struct got_error *
253 advance_branch(struct got_commit_graph *graph, struct got_object_id *commit_id,
254 struct got_commit_object *commit, struct got_repository *repo)
256 const struct got_error *err;
257 struct got_object_qid *qid;
259 err = close_branch(graph, commit_id);
260 if (err)
261 return err;
263 if (graph->flags & GOT_COMMIT_GRAPH_FIRST_PARENT_TRAVERSAL) {
264 qid = STAILQ_FIRST(&commit->parent_ids);
265 if (qid == NULL ||
266 got_object_idset_contains(graph->open_branches, &qid->id))
267 return NULL;
268 /*
269 * The root directory always changes by definition, and when
270 * logging the root we want to traverse consecutive commits
271 * even if they point at the same tree.
272 * But if we are looking for a specific path then we can avoid
273 * fetching packed commits which did not modify the path and
274 * only fetch their IDs. This speeds up 'got blame'.
275 */
276 if (!got_path_is_root_dir(graph->path) &&
277 (commit->flags & GOT_COMMIT_FLAG_PACKED)) {
278 int ncommits = 0;
279 err = packed_first_parent_traversal(&ncommits,
280 graph, &qid->id, repo);
281 if (err || ncommits > 0)
282 return err;
284 return got_object_idset_add(graph->open_branches,
285 &qid->id, NULL);
288 /*
289 * If we are graphing commits for a specific path, skip branches
290 * which do not contribute any content to this path.
291 */
292 if (commit->nparents > 1 && !got_path_is_root_dir(graph->path)) {
293 struct got_object_id *merged_id, *prev_id = NULL;
294 int branches_differ = 0;
296 err = got_object_id_by_path(&merged_id, repo, commit,
297 graph->path);
298 if (err)
299 return err;
301 STAILQ_FOREACH(qid, &commit->parent_ids, entry) {
302 struct got_object_id *id = NULL;
303 struct got_commit_object *pcommit = NULL;
305 if (got_object_idset_contains(graph->open_branches,
306 &qid->id))
307 continue;
309 err = got_object_open_as_commit(&pcommit, repo,
310 &qid->id);
311 if (err) {
312 free(merged_id);
313 free(prev_id);
314 return err;
316 err = got_object_id_by_path(&id, repo, pcommit,
317 graph->path);
318 got_object_commit_close(pcommit);
319 pcommit = NULL;
320 if (err) {
321 if (err->code == GOT_ERR_NO_TREE_ENTRY) {
322 branches_differ = 1;
323 continue;
325 free(merged_id);
326 free(prev_id);
327 return err;
330 if (prev_id) {
331 if (!branches_differ &&
332 got_object_id_cmp(id, prev_id) != 0)
333 branches_differ = 1;
334 free(prev_id);
336 prev_id = id;
338 /*
339 * If a branch has created the merged content we can
340 * skip any other branches.
341 */
342 if (got_object_id_cmp(merged_id, id) == 0) {
343 err = got_object_idset_add(graph->open_branches,
344 &qid->id, NULL);
345 free(merged_id);
346 free(id);
347 return err;
351 free(prev_id);
352 prev_id = NULL;
353 free(merged_id);
354 merged_id = NULL;
356 /*
357 * If the path's content is the same on all branches,
358 * follow the first parent only.
359 */
360 if (!branches_differ) {
361 qid = STAILQ_FIRST(&commit->parent_ids);
362 if (qid == NULL)
363 return NULL;
364 if (got_object_idset_contains(graph->open_branches,
365 &qid->id))
366 return NULL;
367 if (got_object_idset_contains(graph->node_ids,
368 &qid->id))
369 return NULL; /* parent already traversed */
370 return got_object_idset_add(graph->open_branches,
371 &qid->id, NULL);
375 STAILQ_FOREACH(qid, &commit->parent_ids, entry) {
376 if (got_object_idset_contains(graph->open_branches, &qid->id))
377 continue;
378 if (got_object_idset_contains(graph->node_ids, &qid->id))
379 continue; /* parent already traversed */
380 err = got_object_idset_add(graph->open_branches, &qid->id,
381 NULL);
382 if (err)
383 return err;
386 return NULL;
389 const struct got_error *
390 got_commit_graph_open(struct got_commit_graph **graph,
391 const char *path, int first_parent_traversal)
393 const struct got_error *err = NULL;
395 *graph = calloc(1, sizeof(**graph));
396 if (*graph == NULL)
397 return got_error_from_errno("calloc");
399 TAILQ_INIT(&(*graph)->iter_list);
401 (*graph)->path = strdup(path);
402 if ((*graph)->path == NULL) {
403 err = got_error_from_errno("strdup");
404 goto done;
407 (*graph)->node_ids = got_object_idset_alloc();
408 if ((*graph)->node_ids == NULL) {
409 err = got_error_from_errno("got_object_idset_alloc");
410 goto done;
413 (*graph)->open_branches = got_object_idset_alloc();
414 if ((*graph)->open_branches == NULL) {
415 err = got_error_from_errno("got_object_idset_alloc");
416 goto done;
419 if (first_parent_traversal)
420 (*graph)->flags |= GOT_COMMIT_GRAPH_FIRST_PARENT_TRAVERSAL;
421 done:
422 if (err) {
423 got_commit_graph_close(*graph);
424 *graph = NULL;
426 return err;
429 struct add_branch_tip_arg {
430 struct got_commit_graph_branch_tip *tips;
431 int ntips;
432 struct got_repository *repo;
433 struct got_commit_graph *graph;
434 };
436 static const struct got_error *
437 add_branch_tip(struct got_object_id *commit_id, void *data, void *arg)
439 const struct got_error *err;
440 struct add_branch_tip_arg *a = arg;
441 struct got_commit_graph_node *new_node;
442 struct got_commit_object *commit;
444 err = got_object_open_as_commit(&commit, a->repo, commit_id);
445 if (err)
446 return err;
448 err = add_node(&new_node, a->graph, commit_id, a->repo);
449 if (err) {
450 got_object_commit_close(commit);
451 return err;
454 a->tips[a->ntips].commit_id = &new_node->id;
455 a->tips[a->ntips].commit = commit;
456 a->tips[a->ntips].new_node = new_node;
457 a->ntips++;
459 return NULL;
462 static const struct got_error *
463 fetch_commits_from_open_branches(struct got_commit_graph *graph,
464 struct got_repository *repo, got_cancel_cb cancel_cb, void *cancel_arg)
466 const struct got_error *err;
467 struct add_branch_tip_arg arg;
468 int i, ntips;
470 ntips = got_object_idset_num_elements(graph->open_branches);
471 if (ntips == 0)
472 return NULL;
474 /* (Re-)allocate branch tips array if necessary. */
475 if (graph->ntips < ntips) {
476 struct got_commit_graph_branch_tip *tips;
477 tips = recallocarray(graph->tips, graph->ntips, ntips,
478 sizeof(*tips));
479 if (tips == NULL)
480 return got_error_from_errno("recallocarray");
481 graph->tips = tips;
482 graph->ntips = ntips;
484 arg.tips = graph->tips;
485 arg.ntips = 0; /* add_branch_tip() will increment */
486 arg.repo = repo;
487 arg.graph = graph;
488 err = got_object_idset_for_each(graph->open_branches, add_branch_tip,
489 &arg);
490 if (err)
491 goto done;
493 for (i = 0; i < arg.ntips; i++) {
494 struct got_object_id *commit_id;
495 struct got_commit_object *commit;
496 struct got_commit_graph_node *new_node;
497 int changed;
499 if (cancel_cb) {
500 err = (*cancel_cb)(cancel_arg);
501 if (err)
502 break;
505 commit_id = arg.tips[i].commit_id;
506 commit = arg.tips[i].commit;
507 new_node = arg.tips[i].new_node;
509 err = detect_changed_path(&changed, commit, commit_id,
510 graph->path, repo);
511 if (err) {
512 if (err->code != GOT_ERR_NO_OBJ)
513 break;
514 /*
515 * History of the path stops here on the current
516 * branch. Keep going on other branches.
517 */
518 err = close_branch(graph, commit_id);
519 if (err)
520 break;
521 continue;
523 if (changed) {
524 add_node_to_iter_list(graph, new_node,
525 got_object_commit_get_committer_time(commit));
526 arg.tips[i].new_node = NULL;
528 err = advance_branch(graph, commit_id, commit, repo);
529 if (err)
530 break;
532 done:
533 for (i = 0; i < arg.ntips; i++) {
534 got_object_commit_close(arg.tips[i].commit);
535 free(arg.tips[i].new_node);
537 return err;
540 void
541 got_commit_graph_close(struct got_commit_graph *graph)
543 struct got_commit_graph_node *node;
545 while ((node = TAILQ_FIRST(&graph->iter_list))) {
546 TAILQ_REMOVE(&graph->iter_list, node, entry);
547 free(node);
550 if (graph->open_branches)
551 got_object_idset_free(graph->open_branches);
552 if (graph->node_ids)
553 got_object_idset_free(graph->node_ids);
554 free(graph->tips);
555 free(graph->path);
556 free(graph);
559 const struct got_error *
560 got_commit_graph_iter_start(struct got_commit_graph *graph,
561 struct got_object_id *id, struct got_repository *repo,
562 got_cancel_cb cancel_cb, void *cancel_arg)
564 const struct got_error *err = NULL;
566 if (!TAILQ_EMPTY(&graph->iter_list))
567 return got_error(GOT_ERR_ITER_BUSY);
569 err = got_object_idset_add(graph->open_branches, id, NULL);
570 if (err)
571 return err;
573 /* Locate first commit which changed graph->path. */
574 while (TAILQ_EMPTY(&graph->iter_list) &&
575 got_object_idset_num_elements(graph->open_branches) > 0) {
576 err = fetch_commits_from_open_branches(graph, repo,
577 cancel_cb, cancel_arg);
578 if (err)
579 return err;
582 if (TAILQ_EMPTY(&graph->iter_list)) {
583 const char *path;
584 if (got_path_is_root_dir(graph->path))
585 return got_error_no_obj(id);
586 path = graph->path;
587 while (path[0] == '/')
588 path++;
589 return got_error_path(path, GOT_ERR_NO_TREE_ENTRY);
592 return NULL;
595 const struct got_error *
596 got_commit_graph_iter_next(struct got_object_id **id,
597 struct got_commit_graph *graph, struct got_repository *repo,
598 got_cancel_cb cancel_cb, void *cancel_arg)
600 const struct got_error *err = NULL;
601 struct got_commit_graph_node *node;
603 *id = NULL;
605 node = TAILQ_FIRST(&graph->iter_list);
606 if (node == NULL) {
607 /* We are done iterating, or iteration was not started. */
608 return got_error(GOT_ERR_ITER_COMPLETED);
611 while (TAILQ_NEXT(node, entry) == NULL &&
612 got_object_idset_num_elements(graph->open_branches) > 0) {
613 err = fetch_commits_from_open_branches(graph, repo,
614 cancel_cb, cancel_arg);
615 if (err)
616 return err;
619 *id = &node->id;
620 TAILQ_REMOVE(&graph->iter_list, node, entry);
621 return NULL;
624 const struct got_error *
625 got_commit_graph_find_youngest_common_ancestor(struct got_object_id **yca_id,
626 struct got_object_id *commit_id, struct got_object_id *commit_id2,
627 int first_parent_traversal,
628 struct got_repository *repo, got_cancel_cb cancel_cb, void *cancel_arg)
630 const struct got_error *err = NULL;
631 struct got_commit_graph *graph = NULL, *graph2 = NULL;
632 int completed = 0, completed2 = 0;
633 struct got_object_idset *commit_ids;
635 *yca_id = NULL;
637 commit_ids = got_object_idset_alloc();
638 if (commit_ids == NULL)
639 return got_error_from_errno("got_object_idset_alloc");
641 err = got_commit_graph_open(&graph, "/", first_parent_traversal);
642 if (err)
643 goto done;
645 err = got_commit_graph_open(&graph2, "/", first_parent_traversal);
646 if (err)
647 goto done;
649 err = got_commit_graph_iter_start(graph, commit_id, repo,
650 cancel_cb, cancel_arg);
651 if (err)
652 goto done;
654 err = got_commit_graph_iter_start(graph2, commit_id2, repo,
655 cancel_cb, cancel_arg);
656 if (err)
657 goto done;
659 for (;;) {
660 struct got_object_id *id = NULL, *id2 = NULL;
662 if (cancel_cb) {
663 err = (*cancel_cb)(cancel_arg);
664 if (err)
665 break;
668 if (!completed) {
669 err = got_commit_graph_iter_next(&id, graph, repo,
670 cancel_cb, cancel_arg);
671 if (err) {
672 if (err->code != GOT_ERR_ITER_COMPLETED)
673 break;
674 err = NULL;
675 completed = 1;
679 if (!completed2) {
680 err = got_commit_graph_iter_next(&id2, graph2, repo,
681 cancel_cb, cancel_arg);
682 if (err) {
683 if (err->code != GOT_ERR_ITER_COMPLETED)
684 break;
685 err = NULL;
686 completed2 = 1;
690 if (id) {
691 if (got_object_idset_contains(commit_ids, id)) {
692 *yca_id = got_object_id_dup(id);
693 if (*yca_id)
694 break;
695 err = got_error_from_errno("got_object_id_dup");
696 break;
699 err = got_object_idset_add(commit_ids, id, NULL);
700 if (err)
701 break;
703 if (id2) {
704 if (got_object_idset_contains(commit_ids, id2)) {
705 *yca_id = got_object_id_dup(id2);
706 if (*yca_id)
707 break;
708 err = got_error_from_errno("got_object_id_dup");
709 break;
712 err = got_object_idset_add(commit_ids, id2, NULL);
713 if (err)
714 break;
717 if (completed && completed2) {
718 err = got_error(GOT_ERR_ANCESTRY);
719 break;
723 done:
724 got_object_idset_free(commit_ids);
725 if (graph)
726 got_commit_graph_close(graph);
727 if (graph2)
728 got_commit_graph_close(graph2);
729 return err;