2 * Copyright (c) 2018, 2019, 2020 Stefan Sperling <stsp@openbsd.org>
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.
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.
17 #include <sys/types.h>
19 #include <sys/queue.h>
28 #include "got_compat.h"
30 #include "got_error.h"
31 #include "got_object.h"
32 #include "got_cancel.h"
33 #include "got_commit_graph.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. */
46 TAILQ_ENTRY(got_commit_graph_node
) entry
;
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
;
57 struct got_commit_graph
{
58 /* The set of all commits we have traversed. */
59 struct got_object_idset
*node_ids
;
62 #define GOT_COMMIT_GRAPH_FIRST_PARENT_TRAVERSAL 0x01
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.
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.
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.
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
;
85 /* Path of tree entry of interest to the API user. */
89 * Nodes which will be passed to the API user next, sorted by
92 struct got_commit_graph_iter_list iter_list
;
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
)
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
)) {
112 pid
= STAILQ_FIRST(&commit
->parent_ids
);
114 struct got_object_id
*obj_id
;
115 err
= got_object_id_by_path(&obj_id
, repo
, commit
, path
);
117 if (err
->code
== GOT_ERR_NO_TREE_ENTRY
)
120 *changed
= 1; /* The path was created in this commit. */
125 err
= got_object_open_as_tree(&tree
, repo
, commit
->tree_id
);
129 err
= got_object_open_as_commit(&pcommit
, repo
, &pid
->id
);
133 err
= got_object_open_as_tree(&ptree
, repo
, pcommit
->tree_id
);
137 err
= got_object_tree_path_changed(changed
, tree
, ptree
, path
, repo
);
140 got_object_tree_close(tree
);
142 got_object_tree_close(ptree
);
144 got_object_commit_close(pcommit
);
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
);
158 next
= TAILQ_NEXT(n
, entry
);
159 if (next
&& node
->timestamp
>= next
->timestamp
) {
160 TAILQ_INSERT_BEFORE(next
, node
, entry
);
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
;
178 node
= calloc(1, sizeof(*node
));
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
);
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.
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
);
214 /* Add all traversed commits to the graph... */
215 STAILQ_FOREACH(qid
, &traversed_commits
, entry
) {
216 if (got_object_idset_contains(graph
->open_branches
, &qid
->id
))
218 if (got_object_idset_contains(graph
->node_ids
, &qid
->id
))
221 (*ncommits_traversed
)++;
223 /* ... except the last commit is the new branch tip. */
224 if (STAILQ_NEXT(qid
, entry
) == NULL
) {
225 err
= got_object_idset_add(graph
->open_branches
,
230 err
= got_object_idset_add(graph
->node_ids
, &qid
->id
, NULL
);
235 got_object_id_queue_free(&traversed_commits
);
239 static const struct got_error
*
240 close_branch(struct got_commit_graph
*graph
, struct got_object_id
*commit_id
)
242 const struct got_error
*err
;
244 err
= got_object_idset_remove(NULL
, graph
->open_branches
, commit_id
);
245 if (err
&& err
->code
!= GOT_ERR_NO_OBJ
)
250 static const struct got_error
*
251 advance_branch(struct got_commit_graph
*graph
, struct got_object_id
*commit_id
,
252 struct got_commit_object
*commit
, struct got_repository
*repo
)
254 const struct got_error
*err
;
255 struct got_object_qid
*qid
;
257 err
= close_branch(graph
, commit_id
);
261 if (graph
->flags
& GOT_COMMIT_GRAPH_FIRST_PARENT_TRAVERSAL
) {
262 qid
= STAILQ_FIRST(&commit
->parent_ids
);
264 got_object_idset_contains(graph
->open_branches
, &qid
->id
))
267 * The root directory always changes by definition, and when
268 * logging the root we want to traverse consecutive commits
269 * even if they point at the same tree.
270 * But if we are looking for a specific path then we can avoid
271 * fetching packed commits which did not modify the path and
272 * only fetch their IDs. This speeds up 'got blame'.
274 if (!got_path_is_root_dir(graph
->path
) &&
275 (commit
->flags
& GOT_COMMIT_FLAG_PACKED
)) {
277 err
= packed_first_parent_traversal(&ncommits
,
278 graph
, &qid
->id
, repo
);
279 if (err
|| ncommits
> 0)
282 return got_object_idset_add(graph
->open_branches
,
287 * If we are graphing commits for a specific path, skip branches
288 * which do not contribute any content to this path.
290 if (commit
->nparents
> 1 && !got_path_is_root_dir(graph
->path
)) {
291 struct got_object_id
*merged_id
, *prev_id
= NULL
;
292 int branches_differ
= 0;
294 err
= got_object_id_by_path(&merged_id
, repo
, commit
,
299 STAILQ_FOREACH(qid
, &commit
->parent_ids
, entry
) {
300 struct got_object_id
*id
= NULL
;
301 struct got_commit_object
*pcommit
= NULL
;
303 if (got_object_idset_contains(graph
->open_branches
,
307 err
= got_object_open_as_commit(&pcommit
, repo
,
314 err
= got_object_id_by_path(&id
, repo
, pcommit
,
316 got_object_commit_close(pcommit
);
319 if (err
->code
== GOT_ERR_NO_TREE_ENTRY
) {
329 if (!branches_differ
&&
330 got_object_id_cmp(id
, prev_id
) != 0)
337 * If a branch has created the merged content we can
338 * skip any other branches.
340 if (got_object_id_cmp(merged_id
, id
) == 0) {
341 err
= got_object_idset_add(graph
->open_branches
,
355 * If the path's content is the same on all branches,
356 * follow the first parent only.
358 if (!branches_differ
) {
359 qid
= STAILQ_FIRST(&commit
->parent_ids
);
362 if (got_object_idset_contains(graph
->open_branches
,
365 if (got_object_idset_contains(graph
->node_ids
,
367 return NULL
; /* parent already traversed */
368 return got_object_idset_add(graph
->open_branches
,
373 STAILQ_FOREACH(qid
, &commit
->parent_ids
, entry
) {
374 if (got_object_idset_contains(graph
->open_branches
, &qid
->id
))
376 if (got_object_idset_contains(graph
->node_ids
, &qid
->id
))
377 continue; /* parent already traversed */
378 err
= got_object_idset_add(graph
->open_branches
, &qid
->id
,
387 const struct got_error
*
388 got_commit_graph_open(struct got_commit_graph
**graph
,
389 const char *path
, int first_parent_traversal
)
391 const struct got_error
*err
= NULL
;
393 *graph
= calloc(1, sizeof(**graph
));
395 return got_error_from_errno("calloc");
397 TAILQ_INIT(&(*graph
)->iter_list
);
399 (*graph
)->path
= strdup(path
);
400 if ((*graph
)->path
== NULL
) {
401 err
= got_error_from_errno("strdup");
405 (*graph
)->node_ids
= got_object_idset_alloc();
406 if ((*graph
)->node_ids
== NULL
) {
407 err
= got_error_from_errno("got_object_idset_alloc");
411 (*graph
)->open_branches
= got_object_idset_alloc();
412 if ((*graph
)->open_branches
== NULL
) {
413 err
= got_error_from_errno("got_object_idset_alloc");
417 if (first_parent_traversal
)
418 (*graph
)->flags
|= GOT_COMMIT_GRAPH_FIRST_PARENT_TRAVERSAL
;
421 got_commit_graph_close(*graph
);
427 struct add_branch_tip_arg
{
428 struct got_commit_graph_branch_tip
*tips
;
430 struct got_repository
*repo
;
431 struct got_commit_graph
*graph
;
434 static const struct got_error
*
435 add_branch_tip(struct got_object_id
*commit_id
, void *data
, void *arg
)
437 const struct got_error
*err
;
438 struct add_branch_tip_arg
*a
= arg
;
439 struct got_commit_graph_node
*new_node
;
440 struct got_commit_object
*commit
;
442 err
= got_object_open_as_commit(&commit
, a
->repo
, commit_id
);
446 err
= add_node(&new_node
, a
->graph
, commit_id
, a
->repo
);
448 got_object_commit_close(commit
);
452 a
->tips
[a
->ntips
].commit_id
= &new_node
->id
;
453 a
->tips
[a
->ntips
].commit
= commit
;
454 a
->tips
[a
->ntips
].new_node
= new_node
;
460 static const struct got_error
*
461 fetch_commits_from_open_branches(struct got_commit_graph
*graph
,
462 struct got_repository
*repo
, got_cancel_cb cancel_cb
, void *cancel_arg
)
464 const struct got_error
*err
;
465 struct add_branch_tip_arg arg
;
468 ntips
= got_object_idset_num_elements(graph
->open_branches
);
472 /* (Re-)allocate branch tips array if necessary. */
473 if (graph
->ntips
< ntips
) {
474 struct got_commit_graph_branch_tip
*tips
;
475 tips
= recallocarray(graph
->tips
, graph
->ntips
, ntips
,
478 return got_error_from_errno("recallocarray");
480 graph
->ntips
= ntips
;
482 arg
.tips
= graph
->tips
;
483 arg
.ntips
= 0; /* add_branch_tip() will increment */
486 err
= got_object_idset_for_each(graph
->open_branches
, add_branch_tip
,
491 for (i
= 0; i
< arg
.ntips
; i
++) {
492 struct got_object_id
*commit_id
;
493 struct got_commit_object
*commit
;
494 struct got_commit_graph_node
*new_node
;
498 err
= (*cancel_cb
)(cancel_arg
);
503 commit_id
= arg
.tips
[i
].commit_id
;
504 commit
= arg
.tips
[i
].commit
;
505 new_node
= arg
.tips
[i
].new_node
;
507 err
= detect_changed_path(&changed
, commit
, commit_id
,
510 if (err
->code
!= GOT_ERR_NO_OBJ
)
513 * History of the path stops here on the current
514 * branch. Keep going on other branches.
516 err
= close_branch(graph
, commit_id
);
522 add_node_to_iter_list(graph
, new_node
,
523 got_object_commit_get_committer_time(commit
));
524 arg
.tips
[i
].new_node
= NULL
;
526 err
= advance_branch(graph
, commit_id
, commit
, repo
);
531 for (i
= 0; i
< arg
.ntips
; i
++) {
532 got_object_commit_close(arg
.tips
[i
].commit
);
533 free(arg
.tips
[i
].new_node
);
539 got_commit_graph_close(struct got_commit_graph
*graph
)
541 struct got_commit_graph_node
*node
;
543 while ((node
= TAILQ_FIRST(&graph
->iter_list
))) {
544 TAILQ_REMOVE(&graph
->iter_list
, node
, entry
);
548 if (graph
->open_branches
)
549 got_object_idset_free(graph
->open_branches
);
551 got_object_idset_free(graph
->node_ids
);
557 const struct got_error
*
558 got_commit_graph_iter_start(struct got_commit_graph
*graph
,
559 struct got_object_id
*id
, struct got_repository
*repo
,
560 got_cancel_cb cancel_cb
, void *cancel_arg
)
562 const struct got_error
*err
= NULL
;
564 if (!TAILQ_EMPTY(&graph
->iter_list
))
565 return got_error(GOT_ERR_ITER_BUSY
);
567 err
= got_object_idset_add(graph
->open_branches
, id
, NULL
);
571 /* Locate first commit which changed graph->path. */
572 while (TAILQ_EMPTY(&graph
->iter_list
) &&
573 got_object_idset_num_elements(graph
->open_branches
) > 0) {
574 err
= fetch_commits_from_open_branches(graph
, repo
,
575 cancel_cb
, cancel_arg
);
580 if (TAILQ_EMPTY(&graph
->iter_list
)) {
582 if (got_path_is_root_dir(graph
->path
))
583 return got_error_no_obj(id
);
585 while (path
[0] == '/')
587 return got_error_path(path
, GOT_ERR_NO_TREE_ENTRY
);
593 const struct got_error
*
594 got_commit_graph_iter_next(struct got_object_id
*id
,
595 struct got_commit_graph
*graph
, struct got_repository
*repo
,
596 got_cancel_cb cancel_cb
, void *cancel_arg
)
598 const struct got_error
*err
= NULL
;
599 struct got_commit_graph_node
*node
;
601 node
= TAILQ_FIRST(&graph
->iter_list
);
603 /* We are done iterating, or iteration was not started. */
604 return got_error(GOT_ERR_ITER_COMPLETED
);
607 while (TAILQ_NEXT(node
, entry
) == NULL
&&
608 got_object_idset_num_elements(graph
->open_branches
) > 0) {
609 err
= fetch_commits_from_open_branches(graph
, repo
,
610 cancel_cb
, cancel_arg
);
615 memcpy(id
, &node
->id
, sizeof(*id
));
617 TAILQ_REMOVE(&graph
->iter_list
, node
, entry
);
622 static const struct got_error
*
623 find_yca_add_id(struct got_object_id
**yca_id
, struct got_commit_graph
*graph
,
624 struct got_object_idset
*commit_ids
, struct got_repository
*repo
,
625 got_cancel_cb cancel_cb
, void *cancel_arg
)
627 const struct got_error
*err
= NULL
;
628 struct got_object_id id
;
630 err
= got_commit_graph_iter_next(&id
, graph
, repo
, cancel_cb
,
635 if (got_object_idset_contains(commit_ids
, &id
)) {
636 *yca_id
= got_object_id_dup(&id
);
638 err
= got_error_from_errno("got_object_id_dup");
642 return got_object_idset_add(commit_ids
, &id
, NULL
);
646 * Sets *yca_id to the youngest common ancestor of commit_id and
647 * commit_id2. Returns got_error(GOT_ERR_ANCESTRY) if they have no
650 * If first_parent_traversal is nonzero, only linear history is considered.
652 const struct got_error
*
653 got_commit_graph_find_youngest_common_ancestor(struct got_object_id
**yca_id
,
654 struct got_object_id
*commit_id
, struct got_object_id
*commit_id2
,
655 int first_parent_traversal
,
656 struct got_repository
*repo
, got_cancel_cb cancel_cb
, void *cancel_arg
)
658 const struct got_error
*err
= NULL
;
659 struct got_commit_graph
*graph
= NULL
, *graph2
= NULL
;
660 int completed
= 0, completed2
= 0;
661 struct got_object_idset
*commit_ids
;
665 commit_ids
= got_object_idset_alloc();
666 if (commit_ids
== NULL
)
667 return got_error_from_errno("got_object_idset_alloc");
669 err
= got_commit_graph_open(&graph
, "/", first_parent_traversal
);
673 err
= got_commit_graph_open(&graph2
, "/", first_parent_traversal
);
677 err
= got_commit_graph_iter_start(graph
, commit_id
, repo
,
678 cancel_cb
, cancel_arg
);
682 err
= got_commit_graph_iter_start(graph2
, commit_id2
, repo
,
683 cancel_cb
, cancel_arg
);
689 err
= (*cancel_cb
)(cancel_arg
);
695 err
= find_yca_add_id(yca_id
, graph
, commit_ids
, repo
,
696 cancel_cb
, cancel_arg
);
698 if (err
->code
!= GOT_ERR_ITER_COMPLETED
)
708 err
= find_yca_add_id(yca_id
, graph2
, commit_ids
, repo
,
709 cancel_cb
, cancel_arg
);
711 if (err
->code
!= GOT_ERR_ITER_COMPLETED
)
720 if (completed
&& completed2
) {
721 err
= got_error(GOT_ERR_ANCESTRY
);
726 got_object_idset_free(commit_ids
);
728 got_commit_graph_close(graph
);
730 got_commit_graph_close(graph2
);