Code

Clean up reflog unreachability pruning decision
[git.git] / builtin-reflog.c
1 #include "cache.h"
2 #include "builtin.h"
3 #include "commit.h"
4 #include "refs.h"
5 #include "dir.h"
6 #include "tree-walk.h"
7 #include "diff.h"
8 #include "revision.h"
9 #include "reachable.h"
11 /*
12  * reflog expire
13  */
15 static const char reflog_expire_usage[] =
16 "git reflog (show|expire) [--verbose] [--dry-run] [--stale-fix] [--expire=<time>] [--expire-unreachable=<time>] [--all] <refs>...";
17 static const char reflog_delete_usage[] =
18 "git reflog delete [--verbose] [--dry-run] [--rewrite] [--updateref] <refs>...";
20 static unsigned long default_reflog_expire;
21 static unsigned long default_reflog_expire_unreachable;
23 struct cmd_reflog_expire_cb {
24         struct rev_info revs;
25         int dry_run;
26         int stalefix;
27         int rewrite;
28         int updateref;
29         int verbose;
30         unsigned long expire_total;
31         unsigned long expire_unreachable;
32         int recno;
33 };
35 struct expire_reflog_cb {
36         FILE *newlog;
37         const char *ref;
38         struct commit *ref_commit;
39         struct cmd_reflog_expire_cb *cmd;
40         unsigned char last_kept_sha1[20];
41 };
43 struct collected_reflog {
44         unsigned char sha1[20];
45         char reflog[FLEX_ARRAY];
46 };
47 struct collect_reflog_cb {
48         struct collected_reflog **e;
49         int alloc;
50         int nr;
51 };
53 #define INCOMPLETE      (1u<<10)
54 #define STUDYING        (1u<<11)
56 static int tree_is_complete(const unsigned char *sha1)
57 {
58         struct tree_desc desc;
59         struct name_entry entry;
60         int complete;
61         struct tree *tree;
63         tree = lookup_tree(sha1);
64         if (!tree)
65                 return 0;
66         if (tree->object.flags & SEEN)
67                 return 1;
68         if (tree->object.flags & INCOMPLETE)
69                 return 0;
71         if (!tree->buffer) {
72                 enum object_type type;
73                 unsigned long size;
74                 void *data = read_sha1_file(sha1, &type, &size);
75                 if (!data) {
76                         tree->object.flags |= INCOMPLETE;
77                         return 0;
78                 }
79                 tree->buffer = data;
80                 tree->size = size;
81         }
82         init_tree_desc(&desc, tree->buffer, tree->size);
83         complete = 1;
84         while (tree_entry(&desc, &entry)) {
85                 if (!has_sha1_file(entry.sha1) ||
86                     (S_ISDIR(entry.mode) && !tree_is_complete(entry.sha1))) {
87                         tree->object.flags |= INCOMPLETE;
88                         complete = 0;
89                 }
90         }
91         free(tree->buffer);
92         tree->buffer = NULL;
94         if (complete)
95                 tree->object.flags |= SEEN;
96         return complete;
97 }
99 static int commit_is_complete(struct commit *commit)
101         struct object_array study;
102         struct object_array found;
103         int is_incomplete = 0;
104         int i;
106         /* early return */
107         if (commit->object.flags & SEEN)
108                 return 1;
109         if (commit->object.flags & INCOMPLETE)
110                 return 0;
111         /*
112          * Find all commits that are reachable and are not marked as
113          * SEEN.  Then make sure the trees and blobs contained are
114          * complete.  After that, mark these commits also as SEEN.
115          * If some of the objects that are needed to complete this
116          * commit are missing, mark this commit as INCOMPLETE.
117          */
118         memset(&study, 0, sizeof(study));
119         memset(&found, 0, sizeof(found));
120         add_object_array(&commit->object, NULL, &study);
121         add_object_array(&commit->object, NULL, &found);
122         commit->object.flags |= STUDYING;
123         while (study.nr) {
124                 struct commit *c;
125                 struct commit_list *parent;
127                 c = (struct commit *)study.objects[--study.nr].item;
128                 if (!c->object.parsed && !parse_object(c->object.sha1))
129                         c->object.flags |= INCOMPLETE;
131                 if (c->object.flags & INCOMPLETE) {
132                         is_incomplete = 1;
133                         break;
134                 }
135                 else if (c->object.flags & SEEN)
136                         continue;
137                 for (parent = c->parents; parent; parent = parent->next) {
138                         struct commit *p = parent->item;
139                         if (p->object.flags & STUDYING)
140                                 continue;
141                         p->object.flags |= STUDYING;
142                         add_object_array(&p->object, NULL, &study);
143                         add_object_array(&p->object, NULL, &found);
144                 }
145         }
146         if (!is_incomplete) {
147                 /*
148                  * make sure all commits in "found" array have all the
149                  * necessary objects.
150                  */
151                 for (i = 0; i < found.nr; i++) {
152                         struct commit *c =
153                                 (struct commit *)found.objects[i].item;
154                         if (!tree_is_complete(c->tree->object.sha1)) {
155                                 is_incomplete = 1;
156                                 c->object.flags |= INCOMPLETE;
157                         }
158                 }
159                 if (!is_incomplete) {
160                         /* mark all found commits as complete, iow SEEN */
161                         for (i = 0; i < found.nr; i++)
162                                 found.objects[i].item->flags |= SEEN;
163                 }
164         }
165         /* clear flags from the objects we traversed */
166         for (i = 0; i < found.nr; i++)
167                 found.objects[i].item->flags &= ~STUDYING;
168         if (is_incomplete)
169                 commit->object.flags |= INCOMPLETE;
170         else {
171                 /*
172                  * If we come here, we have (1) traversed the ancestry chain
173                  * from the "commit" until we reach SEEN commits (which are
174                  * known to be complete), and (2) made sure that the commits
175                  * encountered during the above traversal refer to trees that
176                  * are complete.  Which means that we know *all* the commits
177                  * we have seen during this process are complete.
178                  */
179                 for (i = 0; i < found.nr; i++)
180                         found.objects[i].item->flags |= SEEN;
181         }
182         /* free object arrays */
183         free(study.objects);
184         free(found.objects);
185         return !is_incomplete;
188 static int keep_entry(struct commit **it, unsigned char *sha1)
190         struct commit *commit;
192         if (is_null_sha1(sha1))
193                 return 1;
194         commit = lookup_commit_reference_gently(sha1, 1);
195         if (!commit)
196                 return 0;
198         /*
199          * Make sure everything in this commit exists.
200          *
201          * We have walked all the objects reachable from the refs
202          * and cache earlier.  The commits reachable by this commit
203          * must meet SEEN commits -- and then we should mark them as
204          * SEEN as well.
205          */
206         if (!commit_is_complete(commit))
207                 return 0;
208         *it = commit;
209         return 1;
212 static int unreachable(struct expire_reflog_cb *cb, struct commit *commit, unsigned char *sha1)
214         /*
215          * We may or may not have the commit yet - if not, look it
216          * up using the supplied sha1.
217          */
218         if (!commit) {
219                 if (is_null_sha1(sha1))
220                         return 0;
222                 commit = lookup_commit_reference_gently(sha1, 1);
224                 /* Not a commit -- keep it */
225                 if (!commit)
226                         return 0;
227         }
229         /* Reachable from the current ref?  Don't prune. */
230         if (in_merge_bases(commit, &cb->ref_commit, 1))
231                 return 0;
233         /* We can't reach it - prune it. */
234         return 1;
237 static int expire_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
238                 const char *email, unsigned long timestamp, int tz,
239                 const char *message, void *cb_data)
241         struct expire_reflog_cb *cb = cb_data;
242         struct commit *old, *new;
244         if (timestamp < cb->cmd->expire_total)
245                 goto prune;
247         if (cb->cmd->rewrite)
248                 osha1 = cb->last_kept_sha1;
250         old = new = NULL;
251         if (cb->cmd->stalefix &&
252             (!keep_entry(&old, osha1) || !keep_entry(&new, nsha1)))
253                 goto prune;
255         if (timestamp < cb->cmd->expire_unreachable) {
256                 if (!cb->ref_commit)
257                         goto prune;
258                 if (unreachable(cb, old, osha1) || unreachable(cb, new, nsha1))
259                         goto prune;
260         }
262         if (cb->cmd->recno && --(cb->cmd->recno) == 0)
263                 goto prune;
265         if (cb->newlog) {
266                 char sign = (tz < 0) ? '-' : '+';
267                 int zone = (tz < 0) ? (-tz) : tz;
268                 fprintf(cb->newlog, "%s %s %s %lu %c%04d\t%s",
269                         sha1_to_hex(osha1), sha1_to_hex(nsha1),
270                         email, timestamp, sign, zone,
271                         message);
272                 hashcpy(cb->last_kept_sha1, nsha1);
273         }
274         if (cb->cmd->verbose)
275                 printf("keep %s", message);
276         return 0;
277  prune:
278         if (!cb->newlog || cb->cmd->verbose)
279                 printf("%sprune %s", cb->newlog ? "" : "would ", message);
280         return 0;
283 static int expire_reflog(const char *ref, const unsigned char *sha1, int unused, void *cb_data)
285         struct cmd_reflog_expire_cb *cmd = cb_data;
286         struct expire_reflog_cb cb;
287         struct ref_lock *lock;
288         char *log_file, *newlog_path = NULL;
289         int status = 0;
291         memset(&cb, 0, sizeof(cb));
293         /*
294          * we take the lock for the ref itself to prevent it from
295          * getting updated.
296          */
297         lock = lock_any_ref_for_update(ref, sha1, 0);
298         if (!lock)
299                 return error("cannot lock ref '%s'", ref);
300         log_file = git_pathdup("logs/%s", ref);
301         if (!file_exists(log_file))
302                 goto finish;
303         if (!cmd->dry_run) {
304                 newlog_path = git_pathdup("logs/%s.lock", ref);
305                 cb.newlog = fopen(newlog_path, "w");
306         }
308         cb.ref_commit = lookup_commit_reference_gently(sha1, 1);
309         cb.ref = ref;
310         cb.cmd = cmd;
311         for_each_reflog_ent(ref, expire_reflog_ent, &cb);
312  finish:
313         if (cb.newlog) {
314                 if (fclose(cb.newlog)) {
315                         status |= error("%s: %s", strerror(errno),
316                                         newlog_path);
317                         unlink(newlog_path);
318                 } else if (cmd->updateref &&
319                         (write_in_full(lock->lock_fd,
320                                 sha1_to_hex(cb.last_kept_sha1), 40) != 40 ||
321                          write_in_full(lock->lock_fd, "\n", 1) != 1 ||
322                          close_ref(lock) < 0)) {
323                         status |= error("Couldn't write %s",
324                                 lock->lk->filename);
325                         unlink(newlog_path);
326                 } else if (rename(newlog_path, log_file)) {
327                         status |= error("cannot rename %s to %s",
328                                         newlog_path, log_file);
329                         unlink(newlog_path);
330                 } else if (cmd->updateref && commit_ref(lock)) {
331                         status |= error("Couldn't set %s", lock->ref_name);
332                 } else {
333                         adjust_shared_perm(log_file);
334                 }
335         }
336         free(newlog_path);
337         free(log_file);
338         unlock_ref(lock);
339         return status;
342 static int collect_reflog(const char *ref, const unsigned char *sha1, int unused, void *cb_data)
344         struct collected_reflog *e;
345         struct collect_reflog_cb *cb = cb_data;
346         size_t namelen = strlen(ref);
348         e = xmalloc(sizeof(*e) + namelen + 1);
349         hashcpy(e->sha1, sha1);
350         memcpy(e->reflog, ref, namelen + 1);
351         ALLOC_GROW(cb->e, cb->nr + 1, cb->alloc);
352         cb->e[cb->nr++] = e;
353         return 0;
356 static struct reflog_expire_cfg {
357         struct reflog_expire_cfg *next;
358         unsigned long expire_total;
359         unsigned long expire_unreachable;
360         size_t len;
361         char pattern[FLEX_ARRAY];
362 } *reflog_expire_cfg, **reflog_expire_cfg_tail;
364 static struct reflog_expire_cfg *find_cfg_ent(const char *pattern, size_t len)
366         struct reflog_expire_cfg *ent;
368         if (!reflog_expire_cfg_tail)
369                 reflog_expire_cfg_tail = &reflog_expire_cfg;
371         for (ent = reflog_expire_cfg; ent; ent = ent->next)
372                 if (ent->len == len &&
373                     !memcmp(ent->pattern, pattern, len))
374                         return ent;
376         ent = xcalloc(1, (sizeof(*ent) + len));
377         memcpy(ent->pattern, pattern, len);
378         ent->len = len;
379         *reflog_expire_cfg_tail = ent;
380         reflog_expire_cfg_tail = &(ent->next);
381         return ent;
384 static int parse_expire_cfg_value(const char *var, const char *value, unsigned long *expire)
386         if (!value)
387                 return config_error_nonbool(var);
388         if (!strcmp(value, "never") || !strcmp(value, "false")) {
389                 *expire = 0;
390                 return 0;
391         }
392         *expire = approxidate(value);
393         return 0;
396 /* expiry timer slot */
397 #define EXPIRE_TOTAL   01
398 #define EXPIRE_UNREACH 02
400 static int reflog_expire_config(const char *var, const char *value, void *cb)
402         const char *lastdot = strrchr(var, '.');
403         unsigned long expire;
404         int slot;
405         struct reflog_expire_cfg *ent;
407         if (!lastdot || prefixcmp(var, "gc."))
408                 return git_default_config(var, value, cb);
410         if (!strcmp(lastdot, ".reflogexpire")) {
411                 slot = EXPIRE_TOTAL;
412                 if (parse_expire_cfg_value(var, value, &expire))
413                         return -1;
414         } else if (!strcmp(lastdot, ".reflogexpireunreachable")) {
415                 slot = EXPIRE_UNREACH;
416                 if (parse_expire_cfg_value(var, value, &expire))
417                         return -1;
418         } else
419                 return git_default_config(var, value, cb);
421         if (lastdot == var + 2) {
422                 switch (slot) {
423                 case EXPIRE_TOTAL:
424                         default_reflog_expire = expire;
425                         break;
426                 case EXPIRE_UNREACH:
427                         default_reflog_expire_unreachable = expire;
428                         break;
429                 }
430                 return 0;
431         }
433         ent = find_cfg_ent(var + 3, lastdot - (var+3));
434         if (!ent)
435                 return -1;
436         switch (slot) {
437         case EXPIRE_TOTAL:
438                 ent->expire_total = expire;
439                 break;
440         case EXPIRE_UNREACH:
441                 ent->expire_unreachable = expire;
442                 break;
443         }
444         return 0;
447 static void set_reflog_expiry_param(struct cmd_reflog_expire_cb *cb, int slot, const char *ref)
449         struct reflog_expire_cfg *ent;
451         if (slot == (EXPIRE_TOTAL|EXPIRE_UNREACH))
452                 return; /* both given explicitly -- nothing to tweak */
454         for (ent = reflog_expire_cfg; ent; ent = ent->next) {
455                 if (!fnmatch(ent->pattern, ref, 0)) {
456                         if (!(slot & EXPIRE_TOTAL))
457                                 cb->expire_total = ent->expire_total;
458                         if (!(slot & EXPIRE_UNREACH))
459                                 cb->expire_unreachable = ent->expire_unreachable;
460                         return;
461                 }
462         }
464         /*
465          * If unconfigured, make stash never expire
466          */
467         if (!strcmp(ref, "refs/stash")) {
468                 if (!(slot & EXPIRE_TOTAL))
469                         cb->expire_total = 0;
470                 if (!(slot & EXPIRE_UNREACH))
471                         cb->expire_unreachable = 0;
472                 return;
473         }
475         /* Nothing matched -- use the default value */
476         if (!(slot & EXPIRE_TOTAL))
477                 cb->expire_total = default_reflog_expire;
478         if (!(slot & EXPIRE_UNREACH))
479                 cb->expire_unreachable = default_reflog_expire_unreachable;
482 static int cmd_reflog_expire(int argc, const char **argv, const char *prefix)
484         struct cmd_reflog_expire_cb cb;
485         unsigned long now = time(NULL);
486         int i, status, do_all;
487         int explicit_expiry = 0;
489         git_config(reflog_expire_config, NULL);
491         save_commit_buffer = 0;
492         do_all = status = 0;
493         memset(&cb, 0, sizeof(cb));
495         if (!default_reflog_expire_unreachable)
496                 default_reflog_expire_unreachable = now - 30 * 24 * 3600;
497         if (!default_reflog_expire)
498                 default_reflog_expire = now - 90 * 24 * 3600;
499         cb.expire_total = default_reflog_expire;
500         cb.expire_unreachable = default_reflog_expire_unreachable;
502         for (i = 1; i < argc; i++) {
503                 const char *arg = argv[i];
504                 if (!strcmp(arg, "--dry-run") || !strcmp(arg, "-n"))
505                         cb.dry_run = 1;
506                 else if (!prefixcmp(arg, "--expire=")) {
507                         cb.expire_total = approxidate(arg + 9);
508                         explicit_expiry |= EXPIRE_TOTAL;
509                 }
510                 else if (!prefixcmp(arg, "--expire-unreachable=")) {
511                         cb.expire_unreachable = approxidate(arg + 21);
512                         explicit_expiry |= EXPIRE_UNREACH;
513                 }
514                 else if (!strcmp(arg, "--stale-fix"))
515                         cb.stalefix = 1;
516                 else if (!strcmp(arg, "--rewrite"))
517                         cb.rewrite = 1;
518                 else if (!strcmp(arg, "--updateref"))
519                         cb.updateref = 1;
520                 else if (!strcmp(arg, "--all"))
521                         do_all = 1;
522                 else if (!strcmp(arg, "--verbose"))
523                         cb.verbose = 1;
524                 else if (!strcmp(arg, "--")) {
525                         i++;
526                         break;
527                 }
528                 else if (arg[0] == '-')
529                         usage(reflog_expire_usage);
530                 else
531                         break;
532         }
534         /*
535          * We can trust the commits and objects reachable from refs
536          * even in older repository.  We cannot trust what's reachable
537          * from reflog if the repository was pruned with older git.
538          */
539         if (cb.stalefix) {
540                 init_revisions(&cb.revs, prefix);
541                 if (cb.verbose)
542                         printf("Marking reachable objects...");
543                 mark_reachable_objects(&cb.revs, 0);
544                 if (cb.verbose)
545                         putchar('\n');
546         }
548         if (do_all) {
549                 struct collect_reflog_cb collected;
550                 int i;
552                 memset(&collected, 0, sizeof(collected));
553                 for_each_reflog(collect_reflog, &collected);
554                 for (i = 0; i < collected.nr; i++) {
555                         struct collected_reflog *e = collected.e[i];
556                         set_reflog_expiry_param(&cb, explicit_expiry, e->reflog);
557                         status |= expire_reflog(e->reflog, e->sha1, 0, &cb);
558                         free(e);
559                 }
560                 free(collected.e);
561         }
563         for (; i < argc; i++) {
564                 char *ref;
565                 unsigned char sha1[20];
566                 if (!dwim_log(argv[i], strlen(argv[i]), sha1, &ref)) {
567                         status |= error("%s points nowhere!", argv[i]);
568                         continue;
569                 }
570                 set_reflog_expiry_param(&cb, explicit_expiry, ref);
571                 status |= expire_reflog(ref, sha1, 0, &cb);
572         }
573         return status;
576 static int count_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
577                 const char *email, unsigned long timestamp, int tz,
578                 const char *message, void *cb_data)
580         struct cmd_reflog_expire_cb *cb = cb_data;
581         if (!cb->expire_total || timestamp < cb->expire_total)
582                 cb->recno++;
583         return 0;
586 static int cmd_reflog_delete(int argc, const char **argv, const char *prefix)
588         struct cmd_reflog_expire_cb cb;
589         int i, status = 0;
591         memset(&cb, 0, sizeof(cb));
593         for (i = 1; i < argc; i++) {
594                 const char *arg = argv[i];
595                 if (!strcmp(arg, "--dry-run") || !strcmp(arg, "-n"))
596                         cb.dry_run = 1;
597                 else if (!strcmp(arg, "--rewrite"))
598                         cb.rewrite = 1;
599                 else if (!strcmp(arg, "--updateref"))
600                         cb.updateref = 1;
601                 else if (!strcmp(arg, "--verbose"))
602                         cb.verbose = 1;
603                 else if (!strcmp(arg, "--")) {
604                         i++;
605                         break;
606                 }
607                 else if (arg[0] == '-')
608                         usage(reflog_delete_usage);
609                 else
610                         break;
611         }
613         if (argc - i < 1)
614                 return error("Nothing to delete?");
616         for ( ; i < argc; i++) {
617                 const char *spec = strstr(argv[i], "@{");
618                 unsigned char sha1[20];
619                 char *ep, *ref;
620                 int recno;
622                 if (!spec) {
623                         status |= error("Not a reflog: %s", argv[i]);
624                         continue;
625                 }
627                 if (!dwim_log(argv[i], spec - argv[i], sha1, &ref)) {
628                         status |= error("no reflog for '%s'", argv[i]);
629                         continue;
630                 }
632                 recno = strtoul(spec + 2, &ep, 10);
633                 if (*ep == '}') {
634                         cb.recno = -recno;
635                         for_each_reflog_ent(ref, count_reflog_ent, &cb);
636                 } else {
637                         cb.expire_total = approxidate(spec + 2);
638                         for_each_reflog_ent(ref, count_reflog_ent, &cb);
639                         cb.expire_total = 0;
640                 }
642                 status |= expire_reflog(ref, sha1, 0, &cb);
643                 free(ref);
644         }
645         return status;
648 /*
649  * main "reflog"
650  */
652 static const char reflog_usage[] =
653 "git reflog (expire | ...)";
655 int cmd_reflog(int argc, const char **argv, const char *prefix)
657         /* With no command, we default to showing it. */
658         if (argc < 2 || *argv[1] == '-')
659                 return cmd_log_reflog(argc, argv, prefix);
661         if (!strcmp(argv[1], "show"))
662                 return cmd_log_reflog(argc - 1, argv + 1, prefix);
664         if (!strcmp(argv[1], "expire"))
665                 return cmd_reflog_expire(argc - 1, argv + 1, prefix);
667         if (!strcmp(argv[1], "delete"))
668                 return cmd_reflog_delete(argc - 1, argv + 1, prefix);
670         /* Not a recognized reflog command..*/
671         usage(reflog_usage);