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
;
256 struct got_object_id
*merged_id
= NULL
;
258 err
= close_branch(graph
, commit_id
);
262 if (graph
->flags
& GOT_COMMIT_GRAPH_FIRST_PARENT_TRAVERSAL
) {
263 qid
= STAILQ_FIRST(&commit
->parent_ids
);
265 got_object_idset_contains(graph
->open_branches
, &qid
->id
))
268 * The root directory always changes by definition, and when
269 * logging the root we want to traverse consecutive commits
270 * even if they point at the same tree.
271 * But if we are looking for a specific path then we can avoid
272 * fetching packed commits which did not modify the path and
273 * only fetch their IDs. This speeds up 'got blame'.
275 if (!got_path_is_root_dir(graph
->path
) &&
276 (commit
->flags
& GOT_COMMIT_FLAG_PACKED
)) {
278 err
= packed_first_parent_traversal(&ncommits
,
279 graph
, &qid
->id
, repo
);
280 if (err
|| ncommits
> 0)
283 return got_object_idset_add(graph
->open_branches
,
288 * If we are graphing commits for a specific path, skip branches
289 * which do not contribute any content to this path.
291 if (commit
->nparents
> 1 && !got_path_is_root_dir(graph
->path
)) {
292 err
= got_object_id_by_path(&merged_id
, repo
, commit
, graph
->path
);
293 if (err
&& err
->code
!= GOT_ERR_NO_TREE_ENTRY
)
295 /* The requested path does not exist in this merge commit. */
297 if (commit
->nparents
> 1 && !got_path_is_root_dir(graph
->path
) &&
299 struct got_object_id
*prev_id
= NULL
;
300 int branches_differ
= 0;
303 STAILQ_FOREACH(qid
, &commit
->parent_ids
, entry
) {
304 struct got_object_id
*id
= NULL
;
305 struct got_commit_object
*pcommit
= NULL
;
307 if (got_object_idset_contains(graph
->open_branches
,
311 err
= got_object_open_as_commit(&pcommit
, repo
,
318 err
= got_object_id_by_path(&id
, repo
, pcommit
,
320 got_object_commit_close(pcommit
);
323 if (err
->code
== GOT_ERR_NO_TREE_ENTRY
) {
333 if (!branches_differ
&&
334 got_object_id_cmp(id
, prev_id
) != 0)
341 * If a branch has created the merged content we can
342 * skip any other branches.
344 if (got_object_id_cmp(merged_id
, id
) == 0) {
345 err
= got_object_idset_add(graph
->open_branches
,
359 * If the path's content is the same on all branches,
360 * follow the first parent only.
362 if (!branches_differ
) {
363 qid
= STAILQ_FIRST(&commit
->parent_ids
);
366 if (got_object_idset_contains(graph
->open_branches
,
369 if (got_object_idset_contains(graph
->node_ids
,
371 return NULL
; /* parent already traversed */
372 return got_object_idset_add(graph
->open_branches
,
377 STAILQ_FOREACH(qid
, &commit
->parent_ids
, entry
) {
378 if (got_object_idset_contains(graph
->open_branches
, &qid
->id
))
380 if (got_object_idset_contains(graph
->node_ids
, &qid
->id
))
381 continue; /* parent already traversed */
382 err
= got_object_idset_add(graph
->open_branches
, &qid
->id
,
391 const struct got_error
*
392 got_commit_graph_open(struct got_commit_graph
**graph
,
393 const char *path
, int first_parent_traversal
)
395 const struct got_error
*err
= NULL
;
397 *graph
= calloc(1, sizeof(**graph
));
399 return got_error_from_errno("calloc");
401 TAILQ_INIT(&(*graph
)->iter_list
);
403 (*graph
)->path
= strdup(path
);
404 if ((*graph
)->path
== NULL
) {
405 err
= got_error_from_errno("strdup");
409 (*graph
)->node_ids
= got_object_idset_alloc();
410 if ((*graph
)->node_ids
== NULL
) {
411 err
= got_error_from_errno("got_object_idset_alloc");
415 (*graph
)->open_branches
= got_object_idset_alloc();
416 if ((*graph
)->open_branches
== NULL
) {
417 err
= got_error_from_errno("got_object_idset_alloc");
421 if (first_parent_traversal
)
422 (*graph
)->flags
|= GOT_COMMIT_GRAPH_FIRST_PARENT_TRAVERSAL
;
425 got_commit_graph_close(*graph
);
431 struct add_branch_tip_arg
{
432 struct got_commit_graph_branch_tip
*tips
;
434 struct got_repository
*repo
;
435 struct got_commit_graph
*graph
;
438 static const struct got_error
*
439 add_branch_tip(struct got_object_id
*commit_id
, void *data
, void *arg
)
441 const struct got_error
*err
;
442 struct add_branch_tip_arg
*a
= arg
;
443 struct got_commit_graph_node
*new_node
;
444 struct got_commit_object
*commit
;
446 err
= got_object_open_as_commit(&commit
, a
->repo
, commit_id
);
450 err
= add_node(&new_node
, a
->graph
, commit_id
, a
->repo
);
452 got_object_commit_close(commit
);
456 a
->tips
[a
->ntips
].commit_id
= &new_node
->id
;
457 a
->tips
[a
->ntips
].commit
= commit
;
458 a
->tips
[a
->ntips
].new_node
= new_node
;
464 static const struct got_error
*
465 fetch_commits_from_open_branches(struct got_commit_graph
*graph
,
466 struct got_repository
*repo
, got_cancel_cb cancel_cb
, void *cancel_arg
)
468 const struct got_error
*err
;
469 struct add_branch_tip_arg arg
;
472 ntips
= got_object_idset_num_elements(graph
->open_branches
);
476 /* (Re-)allocate branch tips array if necessary. */
477 if (graph
->ntips
< ntips
) {
478 struct got_commit_graph_branch_tip
*tips
;
479 tips
= recallocarray(graph
->tips
, graph
->ntips
, ntips
,
482 return got_error_from_errno("recallocarray");
484 graph
->ntips
= ntips
;
486 arg
.tips
= graph
->tips
;
487 arg
.ntips
= 0; /* add_branch_tip() will increment */
490 err
= got_object_idset_for_each(graph
->open_branches
, add_branch_tip
,
495 for (i
= 0; i
< arg
.ntips
; i
++) {
496 struct got_object_id
*commit_id
;
497 struct got_commit_object
*commit
;
498 struct got_commit_graph_node
*new_node
;
502 err
= (*cancel_cb
)(cancel_arg
);
507 commit_id
= arg
.tips
[i
].commit_id
;
508 commit
= arg
.tips
[i
].commit
;
509 new_node
= arg
.tips
[i
].new_node
;
511 err
= detect_changed_path(&changed
, commit
, commit_id
,
514 if (err
->code
!= GOT_ERR_NO_OBJ
)
517 * History of the path stops here on the current
518 * branch. Keep going on other branches.
520 err
= close_branch(graph
, commit_id
);
526 add_node_to_iter_list(graph
, new_node
,
527 got_object_commit_get_committer_time(commit
));
528 arg
.tips
[i
].new_node
= NULL
;
530 err
= advance_branch(graph
, commit_id
, commit
, repo
);
535 for (i
= 0; i
< arg
.ntips
; i
++) {
536 got_object_commit_close(arg
.tips
[i
].commit
);
537 free(arg
.tips
[i
].new_node
);
543 got_commit_graph_close(struct got_commit_graph
*graph
)
545 struct got_commit_graph_node
*node
;
547 while ((node
= TAILQ_FIRST(&graph
->iter_list
))) {
548 TAILQ_REMOVE(&graph
->iter_list
, node
, entry
);
552 if (graph
->open_branches
)
553 got_object_idset_free(graph
->open_branches
);
555 got_object_idset_free(graph
->node_ids
);
561 const struct got_error
*
562 got_commit_graph_iter_start(struct got_commit_graph
*graph
,
563 struct got_object_id
*id
, struct got_repository
*repo
,
564 got_cancel_cb cancel_cb
, void *cancel_arg
)
566 const struct got_error
*err
= NULL
;
568 if (!TAILQ_EMPTY(&graph
->iter_list
))
569 return got_error(GOT_ERR_ITER_BUSY
);
571 err
= got_object_idset_add(graph
->open_branches
, id
, NULL
);
575 /* Locate first commit which changed graph->path. */
576 while (TAILQ_EMPTY(&graph
->iter_list
) &&
577 got_object_idset_num_elements(graph
->open_branches
) > 0) {
578 err
= fetch_commits_from_open_branches(graph
, repo
,
579 cancel_cb
, cancel_arg
);
584 if (TAILQ_EMPTY(&graph
->iter_list
)) {
586 if (got_path_is_root_dir(graph
->path
))
587 return got_error_no_obj(id
);
589 while (path
[0] == '/')
591 return got_error_path(path
, GOT_ERR_NO_TREE_ENTRY
);
597 const struct got_error
*
598 got_commit_graph_iter_next(struct got_object_id
*id
,
599 struct got_commit_graph
*graph
, struct got_repository
*repo
,
600 got_cancel_cb cancel_cb
, void *cancel_arg
)
602 const struct got_error
*err
= NULL
;
603 struct got_commit_graph_node
*node
;
605 node
= TAILQ_FIRST(&graph
->iter_list
);
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
);
619 memcpy(id
, &node
->id
, sizeof(*id
));
621 TAILQ_REMOVE(&graph
->iter_list
, node
, entry
);
626 static const struct got_error
*
627 find_yca_add_id(struct got_object_id
**yca_id
, struct got_commit_graph
*graph
,
628 struct got_object_idset
*commit_ids
, struct got_repository
*repo
,
629 got_cancel_cb cancel_cb
, void *cancel_arg
)
631 const struct got_error
*err
= NULL
;
632 struct got_object_id id
;
634 err
= got_commit_graph_iter_next(&id
, graph
, repo
, cancel_cb
,
639 if (got_object_idset_contains(commit_ids
, &id
)) {
640 *yca_id
= got_object_id_dup(&id
);
642 err
= got_error_from_errno("got_object_id_dup");
646 return got_object_idset_add(commit_ids
, &id
, NULL
);
650 * Sets *yca_id to the youngest common ancestor of commit_id and
651 * commit_id2. Returns got_error(GOT_ERR_ANCESTRY) if they have no
654 * If first_parent_traversal is nonzero, only linear history is considered.
656 const struct got_error
*
657 got_commit_graph_find_youngest_common_ancestor(struct got_object_id
**yca_id
,
658 struct got_object_id
*commit_id
, struct got_object_id
*commit_id2
,
659 int first_parent_traversal
,
660 struct got_repository
*repo
, got_cancel_cb cancel_cb
, void *cancel_arg
)
662 const struct got_error
*err
= NULL
;
663 struct got_commit_graph
*graph
= NULL
, *graph2
= NULL
;
664 int completed
= 0, completed2
= 0;
665 struct got_object_idset
*commit_ids
;
669 commit_ids
= got_object_idset_alloc();
670 if (commit_ids
== NULL
)
671 return got_error_from_errno("got_object_idset_alloc");
673 err
= got_commit_graph_open(&graph
, "/", first_parent_traversal
);
677 err
= got_commit_graph_open(&graph2
, "/", first_parent_traversal
);
681 err
= got_commit_graph_iter_start(graph
, commit_id
, repo
,
682 cancel_cb
, cancel_arg
);
686 err
= got_commit_graph_iter_start(graph2
, commit_id2
, repo
,
687 cancel_cb
, cancel_arg
);
693 err
= (*cancel_cb
)(cancel_arg
);
699 err
= find_yca_add_id(yca_id
, graph
, commit_ids
, repo
,
700 cancel_cb
, cancel_arg
);
702 if (err
->code
!= GOT_ERR_ITER_COMPLETED
)
712 err
= find_yca_add_id(yca_id
, graph2
, commit_ids
, repo
,
713 cancel_cb
, cancel_arg
);
715 if (err
->code
!= GOT_ERR_ITER_COMPLETED
)
724 if (completed
&& completed2
) {
725 err
= got_error(GOT_ERR_ANCESTRY
);
730 got_object_idset_free(commit_ids
);
732 got_commit_graph_close(graph
);
734 got_commit_graph_close(graph2
);