Merge branch 'ps/reftable-optimize-io'
[git/gitster.git] / builtin / fast-export.c
blobf18f0809f9c7f78874eaddfd869c5fc5d11831e0
1 /*
2 * "git fast-export" builtin command
4 * Copyright (C) 2007 Johannes E. Schindelin
5 */
6 #include "builtin.h"
7 #include "config.h"
8 #include "gettext.h"
9 #include "hex.h"
10 #include "refs.h"
11 #include "refspec.h"
12 #include "object-file.h"
13 #include "object-store-ll.h"
14 #include "commit.h"
15 #include "object.h"
16 #include "tag.h"
17 #include "diff.h"
18 #include "diffcore.h"
19 #include "log-tree.h"
20 #include "revision.h"
21 #include "decorate.h"
22 #include "string-list.h"
23 #include "utf8.h"
24 #include "parse-options.h"
25 #include "quote.h"
26 #include "remote.h"
27 #include "blob.h"
29 static const char *fast_export_usage[] = {
30 N_("git fast-export [<rev-list-opts>]"),
31 NULL
34 static int progress;
35 static enum signed_tag_mode { SIGNED_TAG_ABORT, VERBATIM, WARN, WARN_STRIP, STRIP } signed_tag_mode = SIGNED_TAG_ABORT;
36 static enum tag_of_filtered_mode { TAG_FILTERING_ABORT, DROP, REWRITE } tag_of_filtered_mode = TAG_FILTERING_ABORT;
37 static enum reencode_mode { REENCODE_ABORT, REENCODE_YES, REENCODE_NO } reencode_mode = REENCODE_ABORT;
38 static int fake_missing_tagger;
39 static int use_done_feature;
40 static int no_data;
41 static int full_tree;
42 static int reference_excluded_commits;
43 static int show_original_ids;
44 static int mark_tags;
45 static struct string_list extra_refs = STRING_LIST_INIT_NODUP;
46 static struct string_list tag_refs = STRING_LIST_INIT_NODUP;
47 static struct refspec refspecs = REFSPEC_INIT_FETCH;
48 static int anonymize;
49 static struct hashmap anonymized_seeds;
50 static struct revision_sources revision_sources;
52 static int parse_opt_signed_tag_mode(const struct option *opt,
53 const char *arg, int unset)
55 enum signed_tag_mode *val = opt->value;
57 if (unset || !strcmp(arg, "abort"))
58 *val = SIGNED_TAG_ABORT;
59 else if (!strcmp(arg, "verbatim") || !strcmp(arg, "ignore"))
60 *val = VERBATIM;
61 else if (!strcmp(arg, "warn"))
62 *val = WARN;
63 else if (!strcmp(arg, "warn-strip"))
64 *val = WARN_STRIP;
65 else if (!strcmp(arg, "strip"))
66 *val = STRIP;
67 else
68 return error("Unknown signed-tags mode: %s", arg);
69 return 0;
72 static int parse_opt_tag_of_filtered_mode(const struct option *opt,
73 const char *arg, int unset)
75 enum tag_of_filtered_mode *val = opt->value;
77 if (unset || !strcmp(arg, "abort"))
78 *val = TAG_FILTERING_ABORT;
79 else if (!strcmp(arg, "drop"))
80 *val = DROP;
81 else if (!strcmp(arg, "rewrite"))
82 *val = REWRITE;
83 else
84 return error("Unknown tag-of-filtered mode: %s", arg);
85 return 0;
88 static int parse_opt_reencode_mode(const struct option *opt,
89 const char *arg, int unset)
91 enum reencode_mode *val = opt->value;
93 if (unset) {
94 *val = REENCODE_ABORT;
95 return 0;
98 switch (git_parse_maybe_bool(arg)) {
99 case 0:
100 *val = REENCODE_NO;
101 break;
102 case 1:
103 *val = REENCODE_YES;
104 break;
105 default:
106 if (!strcasecmp(arg, "abort"))
107 *val = REENCODE_ABORT;
108 else
109 return error("Unknown reencoding mode: %s", arg);
112 return 0;
115 static struct decoration idnums;
116 static uint32_t last_idnum;
117 struct anonymized_entry {
118 struct hashmap_entry hash;
119 char *anon;
120 const char orig[FLEX_ARRAY];
123 struct anonymized_entry_key {
124 struct hashmap_entry hash;
125 const char *orig;
126 size_t orig_len;
129 static int anonymized_entry_cmp(const void *cmp_data UNUSED,
130 const struct hashmap_entry *eptr,
131 const struct hashmap_entry *entry_or_key,
132 const void *keydata)
134 const struct anonymized_entry *a, *b;
136 a = container_of(eptr, const struct anonymized_entry, hash);
137 if (keydata) {
138 const struct anonymized_entry_key *key = keydata;
139 int equal = !strncmp(a->orig, key->orig, key->orig_len) &&
140 !a->orig[key->orig_len];
141 return !equal;
144 b = container_of(entry_or_key, const struct anonymized_entry, hash);
145 return strcmp(a->orig, b->orig);
148 static struct anonymized_entry *add_anonymized_entry(struct hashmap *map,
149 unsigned hash,
150 const char *orig, size_t len,
151 char *anon)
153 struct anonymized_entry *ret, *old;
155 if (!map->cmpfn)
156 hashmap_init(map, anonymized_entry_cmp, NULL, 0);
158 FLEX_ALLOC_MEM(ret, orig, orig, len);
159 hashmap_entry_init(&ret->hash, hash);
160 ret->anon = anon;
161 old = hashmap_put_entry(map, ret, hash);
163 if (old) {
164 free(old->anon);
165 free(old);
168 return ret;
172 * Basically keep a cache of X->Y so that we can repeatedly replace
173 * the same anonymized string with another. The actual generation
174 * is farmed out to the generate function.
176 static const char *anonymize_str(struct hashmap *map,
177 char *(*generate)(void),
178 const char *orig, size_t len)
180 struct anonymized_entry_key key;
181 struct anonymized_entry *ret;
183 hashmap_entry_init(&key.hash, memhash(orig, len));
184 key.orig = orig;
185 key.orig_len = len;
187 /* First check if it's a token the user configured manually... */
188 ret = hashmap_get_entry(&anonymized_seeds, &key, hash, &key);
190 /* ...otherwise check if we've already seen it in this context... */
191 if (!ret)
192 ret = hashmap_get_entry(map, &key, hash, &key);
194 /* ...and finally generate a new mapping if necessary */
195 if (!ret)
196 ret = add_anonymized_entry(map, key.hash.hash,
197 orig, len, generate());
199 return ret->anon;
203 * We anonymize each component of a path individually,
204 * so that paths a/b and a/c will share a common root.
205 * The paths are cached via anonymize_mem so that repeated
206 * lookups for "a" will yield the same value.
208 static void anonymize_path(struct strbuf *out, const char *path,
209 struct hashmap *map,
210 char *(*generate)(void))
212 while (*path) {
213 const char *end_of_component = strchrnul(path, '/');
214 size_t len = end_of_component - path;
215 const char *c = anonymize_str(map, generate, path, len);
216 strbuf_addstr(out, c);
217 path = end_of_component;
218 if (*path)
219 strbuf_addch(out, *path++);
223 static inline void *mark_to_ptr(uint32_t mark)
225 return (void *)(uintptr_t)mark;
228 static inline uint32_t ptr_to_mark(void * mark)
230 return (uint32_t)(uintptr_t)mark;
233 static inline void mark_object(struct object *object, uint32_t mark)
235 add_decoration(&idnums, object, mark_to_ptr(mark));
238 static inline void mark_next_object(struct object *object)
240 mark_object(object, ++last_idnum);
243 static int get_object_mark(struct object *object)
245 void *decoration = lookup_decoration(&idnums, object);
246 if (!decoration)
247 return 0;
248 return ptr_to_mark(decoration);
251 static struct commit *rewrite_commit(struct commit *p)
253 for (;;) {
254 if (p->parents && p->parents->next)
255 break;
256 if (p->object.flags & UNINTERESTING)
257 break;
258 if (!(p->object.flags & TREESAME))
259 break;
260 if (!p->parents)
261 return NULL;
262 p = p->parents->item;
264 return p;
267 static void show_progress(void)
269 static int counter = 0;
270 if (!progress)
271 return;
272 if ((++counter % progress) == 0)
273 printf("progress %d objects\n", counter);
277 * Ideally we would want some transformation of the blob data here
278 * that is unreversible, but would still be the same size and have
279 * the same data relationship to other blobs (so that we get the same
280 * delta and packing behavior as the original). But the first and last
281 * requirements there are probably mutually exclusive, so let's take
282 * the easy way out for now, and just generate arbitrary content.
284 * There's no need to cache this result with anonymize_mem, since
285 * we already handle blob content caching with marks.
287 static char *anonymize_blob(unsigned long *size)
289 static int counter;
290 struct strbuf out = STRBUF_INIT;
291 strbuf_addf(&out, "anonymous blob %d", counter++);
292 *size = out.len;
293 return strbuf_detach(&out, NULL);
296 static void export_blob(const struct object_id *oid)
298 unsigned long size;
299 enum object_type type;
300 char *buf;
301 struct object *object;
302 int eaten;
304 if (no_data)
305 return;
307 if (is_null_oid(oid))
308 return;
310 object = lookup_object(the_repository, oid);
311 if (object && object->flags & SHOWN)
312 return;
314 if (anonymize) {
315 buf = anonymize_blob(&size);
316 object = (struct object *)lookup_blob(the_repository, oid);
317 eaten = 0;
318 } else {
319 buf = repo_read_object_file(the_repository, oid, &type, &size);
320 if (!buf)
321 die("could not read blob %s", oid_to_hex(oid));
322 if (check_object_signature(the_repository, oid, buf, size,
323 type) < 0)
324 die("oid mismatch in blob %s", oid_to_hex(oid));
325 object = parse_object_buffer(the_repository, oid, type,
326 size, buf, &eaten);
329 if (!object)
330 die("Could not read blob %s", oid_to_hex(oid));
332 mark_next_object(object);
334 printf("blob\nmark :%"PRIu32"\n", last_idnum);
335 if (show_original_ids)
336 printf("original-oid %s\n", oid_to_hex(oid));
337 printf("data %"PRIuMAX"\n", (uintmax_t)size);
338 if (size && fwrite(buf, size, 1, stdout) != 1)
339 die_errno("could not write blob '%s'", oid_to_hex(oid));
340 printf("\n");
342 show_progress();
344 object->flags |= SHOWN;
345 if (!eaten)
346 free(buf);
349 static int depth_first(const void *a_, const void *b_)
351 const struct diff_filepair *a = *((const struct diff_filepair **)a_);
352 const struct diff_filepair *b = *((const struct diff_filepair **)b_);
353 const char *name_a, *name_b;
354 int len_a, len_b, len;
355 int cmp;
357 name_a = a->one ? a->one->path : a->two->path;
358 name_b = b->one ? b->one->path : b->two->path;
360 len_a = strlen(name_a);
361 len_b = strlen(name_b);
362 len = (len_a < len_b) ? len_a : len_b;
364 /* strcmp will sort 'd' before 'd/e', we want 'd/e' before 'd' */
365 cmp = memcmp(name_a, name_b, len);
366 if (cmp)
367 return cmp;
368 cmp = len_b - len_a;
369 if (cmp)
370 return cmp;
372 * Move 'R'ename entries last so that all references of the file
373 * appear in the output before it is renamed (e.g., when a file
374 * was copied and renamed in the same commit).
376 return (a->status == 'R') - (b->status == 'R');
379 static void print_path_1(const char *path)
381 int need_quote = quote_c_style(path, NULL, NULL, 0);
382 if (need_quote)
383 quote_c_style(path, NULL, stdout, 0);
384 else if (strchr(path, ' '))
385 printf("\"%s\"", path);
386 else
387 printf("%s", path);
390 static char *anonymize_path_component(void)
392 static int counter;
393 struct strbuf out = STRBUF_INIT;
394 strbuf_addf(&out, "path%d", counter++);
395 return strbuf_detach(&out, NULL);
398 static void print_path(const char *path)
400 if (!anonymize)
401 print_path_1(path);
402 else {
403 static struct hashmap paths;
404 static struct strbuf anon = STRBUF_INIT;
406 anonymize_path(&anon, path, &paths, anonymize_path_component);
407 print_path_1(anon.buf);
408 strbuf_reset(&anon);
412 static char *generate_fake_oid(void)
414 static uint32_t counter = 1; /* avoid null oid */
415 const unsigned hashsz = the_hash_algo->rawsz;
416 struct object_id oid;
417 char *hex = xmallocz(GIT_MAX_HEXSZ);
419 oidclr(&oid);
420 put_be32(oid.hash + hashsz - 4, counter++);
421 return oid_to_hex_r(hex, &oid);
424 static const char *anonymize_oid(const char *oid_hex)
426 static struct hashmap objs;
427 size_t len = strlen(oid_hex);
428 return anonymize_str(&objs, generate_fake_oid, oid_hex, len);
431 static void show_filemodify(struct diff_queue_struct *q,
432 struct diff_options *options UNUSED, void *data)
434 int i;
435 struct string_list *changed = data;
438 * Handle files below a directory first, in case they are all deleted
439 * and the directory changes to a file or symlink.
441 QSORT(q->queue, q->nr, depth_first);
443 for (i = 0; i < q->nr; i++) {
444 struct diff_filespec *ospec = q->queue[i]->one;
445 struct diff_filespec *spec = q->queue[i]->two;
447 switch (q->queue[i]->status) {
448 case DIFF_STATUS_DELETED:
449 printf("D ");
450 print_path(spec->path);
451 string_list_insert(changed, spec->path);
452 putchar('\n');
453 break;
455 case DIFF_STATUS_COPIED:
456 case DIFF_STATUS_RENAMED:
458 * If a change in the file corresponding to ospec->path
459 * has been observed, we cannot trust its contents
460 * because the diff is calculated based on the prior
461 * contents, not the current contents. So, declare a
462 * copy or rename only if there was no change observed.
464 if (!string_list_has_string(changed, ospec->path)) {
465 printf("%c ", q->queue[i]->status);
466 print_path(ospec->path);
467 putchar(' ');
468 print_path(spec->path);
469 string_list_insert(changed, spec->path);
470 putchar('\n');
472 if (oideq(&ospec->oid, &spec->oid) &&
473 ospec->mode == spec->mode)
474 break;
476 /* fallthrough */
478 case DIFF_STATUS_TYPE_CHANGED:
479 case DIFF_STATUS_MODIFIED:
480 case DIFF_STATUS_ADDED:
482 * Links refer to objects in another repositories;
483 * output the SHA-1 verbatim.
485 if (no_data || S_ISGITLINK(spec->mode))
486 printf("M %06o %s ", spec->mode,
487 anonymize ?
488 anonymize_oid(oid_to_hex(&spec->oid)) :
489 oid_to_hex(&spec->oid));
490 else {
491 struct object *object = lookup_object(the_repository,
492 &spec->oid);
493 printf("M %06o :%d ", spec->mode,
494 get_object_mark(object));
496 print_path(spec->path);
497 string_list_insert(changed, spec->path);
498 putchar('\n');
499 break;
501 default:
502 die("Unexpected comparison status '%c' for %s, %s",
503 q->queue[i]->status,
504 ospec->path ? ospec->path : "none",
505 spec->path ? spec->path : "none");
510 static const char *find_encoding(const char *begin, const char *end)
512 const char *needle = "\nencoding ";
513 char *bol, *eol;
515 bol = memmem(begin, end ? end - begin : strlen(begin),
516 needle, strlen(needle));
517 if (!bol)
518 return NULL;
519 bol += strlen(needle);
520 eol = strchrnul(bol, '\n');
521 *eol = '\0';
522 return bol;
525 static char *anonymize_ref_component(void)
527 static int counter;
528 struct strbuf out = STRBUF_INIT;
529 strbuf_addf(&out, "ref%d", counter++);
530 return strbuf_detach(&out, NULL);
533 static const char *anonymize_refname(const char *refname)
536 * If any of these prefixes is found, we will leave it intact
537 * so that tags remain tags and so forth.
539 static const char *prefixes[] = {
540 "refs/heads/",
541 "refs/tags/",
542 "refs/remotes/",
543 "refs/"
545 static struct hashmap refs;
546 static struct strbuf anon = STRBUF_INIT;
547 int i;
549 strbuf_reset(&anon);
550 for (i = 0; i < ARRAY_SIZE(prefixes); i++) {
551 if (skip_prefix(refname, prefixes[i], &refname)) {
552 strbuf_addstr(&anon, prefixes[i]);
553 break;
557 anonymize_path(&anon, refname, &refs, anonymize_ref_component);
558 return anon.buf;
562 * We do not even bother to cache commit messages, as they are unlikely
563 * to be repeated verbatim, and it is not that interesting when they are.
565 static char *anonymize_commit_message(void)
567 static int counter;
568 return xstrfmt("subject %d\n\nbody\n", counter++);
571 static char *anonymize_ident(void)
573 static int counter;
574 struct strbuf out = STRBUF_INIT;
575 strbuf_addf(&out, "User %d <user%d@example.com>", counter, counter);
576 counter++;
577 return strbuf_detach(&out, NULL);
581 * Our strategy here is to anonymize the names and email addresses,
582 * but keep timestamps intact, as they influence things like traversal
583 * order (and by themselves should not be too revealing).
585 static void anonymize_ident_line(const char **beg, const char **end)
587 static struct hashmap idents;
588 static struct strbuf buffers[] = { STRBUF_INIT, STRBUF_INIT };
589 static unsigned which_buffer;
591 struct strbuf *out;
592 struct ident_split split;
593 const char *end_of_header;
595 out = &buffers[which_buffer++];
596 which_buffer %= ARRAY_SIZE(buffers);
597 strbuf_reset(out);
599 /* skip "committer", "author", "tagger", etc */
600 end_of_header = strchr(*beg, ' ');
601 if (!end_of_header)
602 BUG("malformed line fed to anonymize_ident_line: %.*s",
603 (int)(*end - *beg), *beg);
604 end_of_header++;
605 strbuf_add(out, *beg, end_of_header - *beg);
607 if (!split_ident_line(&split, end_of_header, *end - end_of_header) &&
608 split.date_begin) {
609 const char *ident;
610 size_t len;
612 len = split.mail_end - split.name_begin;
613 ident = anonymize_str(&idents, anonymize_ident,
614 split.name_begin, len);
615 strbuf_addstr(out, ident);
616 strbuf_addch(out, ' ');
617 strbuf_add(out, split.date_begin, split.tz_end - split.date_begin);
618 } else {
619 strbuf_addstr(out, "Malformed Ident <malformed@example.com> 0 -0000");
622 *beg = out->buf;
623 *end = out->buf + out->len;
626 static void handle_commit(struct commit *commit, struct rev_info *rev,
627 struct string_list *paths_of_changed_objects)
629 int saved_output_format = rev->diffopt.output_format;
630 const char *commit_buffer;
631 const char *author, *author_end, *committer, *committer_end;
632 const char *encoding, *message;
633 char *reencoded = NULL;
634 struct commit_list *p;
635 const char *refname;
636 int i;
638 rev->diffopt.output_format = DIFF_FORMAT_CALLBACK;
640 parse_commit_or_die(commit);
641 commit_buffer = repo_get_commit_buffer(the_repository, commit, NULL);
642 author = strstr(commit_buffer, "\nauthor ");
643 if (!author)
644 die("could not find author in commit %s",
645 oid_to_hex(&commit->object.oid));
646 author++;
647 author_end = strchrnul(author, '\n');
648 committer = strstr(author_end, "\ncommitter ");
649 if (!committer)
650 die("could not find committer in commit %s",
651 oid_to_hex(&commit->object.oid));
652 committer++;
653 committer_end = strchrnul(committer, '\n');
654 message = strstr(committer_end, "\n\n");
655 encoding = find_encoding(committer_end, message);
656 if (message)
657 message += 2;
659 if (commit->parents &&
660 (get_object_mark(&commit->parents->item->object) != 0 ||
661 reference_excluded_commits) &&
662 !full_tree) {
663 parse_commit_or_die(commit->parents->item);
664 diff_tree_oid(get_commit_tree_oid(commit->parents->item),
665 get_commit_tree_oid(commit), "", &rev->diffopt);
667 else
668 diff_root_tree_oid(get_commit_tree_oid(commit),
669 "", &rev->diffopt);
671 /* Export the referenced blobs, and remember the marks. */
672 for (i = 0; i < diff_queued_diff.nr; i++)
673 if (!S_ISGITLINK(diff_queued_diff.queue[i]->two->mode))
674 export_blob(&diff_queued_diff.queue[i]->two->oid);
676 refname = *revision_sources_at(&revision_sources, commit);
678 * FIXME: string_list_remove() below for each ref is overall
679 * O(N^2). Compared to a history walk and diffing trees, this is
680 * just lost in the noise in practice. However, theoretically a
681 * repo may have enough refs for this to become slow.
683 string_list_remove(&extra_refs, refname, 0);
684 if (anonymize) {
685 refname = anonymize_refname(refname);
686 anonymize_ident_line(&committer, &committer_end);
687 anonymize_ident_line(&author, &author_end);
690 mark_next_object(&commit->object);
691 if (anonymize) {
692 reencoded = anonymize_commit_message();
693 } else if (encoding) {
694 switch(reencode_mode) {
695 case REENCODE_YES:
696 reencoded = reencode_string(message, "UTF-8", encoding);
697 break;
698 case REENCODE_NO:
699 break;
700 case REENCODE_ABORT:
701 die("Encountered commit-specific encoding %s in commit "
702 "%s; use --reencode=[yes|no] to handle it",
703 encoding, oid_to_hex(&commit->object.oid));
706 if (!commit->parents)
707 printf("reset %s\n", refname);
708 printf("commit %s\nmark :%"PRIu32"\n", refname, last_idnum);
709 if (show_original_ids)
710 printf("original-oid %s\n", oid_to_hex(&commit->object.oid));
711 printf("%.*s\n%.*s\n",
712 (int)(author_end - author), author,
713 (int)(committer_end - committer), committer);
714 if (!reencoded && encoding)
715 printf("encoding %s\n", encoding);
716 printf("data %u\n%s",
717 (unsigned)(reencoded
718 ? strlen(reencoded) : message
719 ? strlen(message) : 0),
720 reencoded ? reencoded : message ? message : "");
721 free(reencoded);
722 repo_unuse_commit_buffer(the_repository, commit, commit_buffer);
724 for (i = 0, p = commit->parents; p; p = p->next) {
725 struct object *obj = &p->item->object;
726 int mark = get_object_mark(obj);
728 if (!mark && !reference_excluded_commits)
729 continue;
730 if (i == 0)
731 printf("from ");
732 else
733 printf("merge ");
734 if (mark)
735 printf(":%d\n", mark);
736 else
737 printf("%s\n",
738 anonymize ?
739 anonymize_oid(oid_to_hex(&obj->oid)) :
740 oid_to_hex(&obj->oid));
741 i++;
744 if (full_tree)
745 printf("deleteall\n");
746 log_tree_diff_flush(rev);
747 string_list_clear(paths_of_changed_objects, 0);
748 rev->diffopt.output_format = saved_output_format;
750 printf("\n");
752 show_progress();
755 static char *anonymize_tag(void)
757 static int counter;
758 struct strbuf out = STRBUF_INIT;
759 strbuf_addf(&out, "tag message %d", counter++);
760 return strbuf_detach(&out, NULL);
764 static void handle_tag(const char *name, struct tag *tag)
766 unsigned long size;
767 enum object_type type;
768 char *buf;
769 const char *tagger, *tagger_end, *message;
770 size_t message_size = 0;
771 struct object *tagged;
772 int tagged_mark;
773 struct commit *p;
775 /* Trees have no identifier in fast-export output, thus we have no way
776 * to output tags of trees, tags of tags of trees, etc. Simply omit
777 * such tags.
779 tagged = tag->tagged;
780 while (tagged->type == OBJ_TAG) {
781 tagged = ((struct tag *)tagged)->tagged;
783 if (tagged->type == OBJ_TREE) {
784 warning("Omitting tag %s,\nsince tags of trees (or tags of tags of trees, etc.) are not supported.",
785 oid_to_hex(&tag->object.oid));
786 return;
789 buf = repo_read_object_file(the_repository, &tag->object.oid, &type,
790 &size);
791 if (!buf)
792 die("could not read tag %s", oid_to_hex(&tag->object.oid));
793 message = memmem(buf, size, "\n\n", 2);
794 if (message) {
795 message += 2;
796 message_size = strlen(message);
798 tagger = memmem(buf, message ? message - buf : size, "\ntagger ", 8);
799 if (!tagger) {
800 if (fake_missing_tagger)
801 tagger = "tagger Unspecified Tagger "
802 "<unspecified-tagger> 0 +0000";
803 else
804 tagger = "";
805 tagger_end = tagger + strlen(tagger);
806 } else {
807 tagger++;
808 tagger_end = strchrnul(tagger, '\n');
809 if (anonymize)
810 anonymize_ident_line(&tagger, &tagger_end);
813 if (anonymize) {
814 name = anonymize_refname(name);
815 if (message) {
816 static struct hashmap tags;
817 message = anonymize_str(&tags, anonymize_tag,
818 message, message_size);
819 message_size = strlen(message);
823 /* handle signed tags */
824 if (message) {
825 const char *signature = strstr(message,
826 "\n-----BEGIN PGP SIGNATURE-----\n");
827 if (signature)
828 switch(signed_tag_mode) {
829 case SIGNED_TAG_ABORT:
830 die("encountered signed tag %s; use "
831 "--signed-tags=<mode> to handle it",
832 oid_to_hex(&tag->object.oid));
833 case WARN:
834 warning("exporting signed tag %s",
835 oid_to_hex(&tag->object.oid));
836 /* fallthru */
837 case VERBATIM:
838 break;
839 case WARN_STRIP:
840 warning("stripping signature from tag %s",
841 oid_to_hex(&tag->object.oid));
842 /* fallthru */
843 case STRIP:
844 message_size = signature + 1 - message;
845 break;
849 /* handle tag->tagged having been filtered out due to paths specified */
850 tagged = tag->tagged;
851 tagged_mark = get_object_mark(tagged);
852 if (!tagged_mark) {
853 switch(tag_of_filtered_mode) {
854 case TAG_FILTERING_ABORT:
855 die("tag %s tags unexported object; use "
856 "--tag-of-filtered-object=<mode> to handle it",
857 oid_to_hex(&tag->object.oid));
858 case DROP:
859 /* Ignore this tag altogether */
860 free(buf);
861 return;
862 case REWRITE:
863 if (tagged->type == OBJ_TAG && !mark_tags) {
864 die(_("Error: Cannot export nested tags unless --mark-tags is specified."));
865 } else if (tagged->type == OBJ_COMMIT) {
866 p = rewrite_commit((struct commit *)tagged);
867 if (!p) {
868 printf("reset %s\nfrom %s\n\n",
869 name, oid_to_hex(null_oid()));
870 free(buf);
871 return;
873 tagged_mark = get_object_mark(&p->object);
874 } else {
875 /* tagged->type is either OBJ_BLOB or OBJ_TAG */
876 tagged_mark = get_object_mark(tagged);
881 if (tagged->type == OBJ_TAG) {
882 printf("reset %s\nfrom %s\n\n",
883 name, oid_to_hex(null_oid()));
885 skip_prefix(name, "refs/tags/", &name);
886 printf("tag %s\n", name);
887 if (mark_tags) {
888 mark_next_object(&tag->object);
889 printf("mark :%"PRIu32"\n", last_idnum);
891 if (tagged_mark)
892 printf("from :%d\n", tagged_mark);
893 else
894 printf("from %s\n", oid_to_hex(&tagged->oid));
896 if (show_original_ids)
897 printf("original-oid %s\n", oid_to_hex(&tag->object.oid));
898 printf("%.*s%sdata %d\n%.*s\n",
899 (int)(tagger_end - tagger), tagger,
900 tagger == tagger_end ? "" : "\n",
901 (int)message_size, (int)message_size, message ? message : "");
902 free(buf);
905 static struct commit *get_commit(struct rev_cmdline_entry *e, char *full_name)
907 switch (e->item->type) {
908 case OBJ_COMMIT:
909 return (struct commit *)e->item;
910 case OBJ_TAG: {
911 struct tag *tag = (struct tag *)e->item;
913 /* handle nested tags */
914 while (tag && tag->object.type == OBJ_TAG) {
915 parse_object(the_repository, &tag->object.oid);
916 string_list_append(&tag_refs, full_name)->util = tag;
917 tag = (struct tag *)tag->tagged;
919 if (!tag)
920 die("Tag %s points nowhere?", e->name);
921 return (struct commit *)tag;
923 default:
924 return NULL;
928 static void get_tags_and_duplicates(struct rev_cmdline_info *info)
930 int i;
932 for (i = 0; i < info->nr; i++) {
933 struct rev_cmdline_entry *e = info->rev + i;
934 struct object_id oid;
935 struct commit *commit;
936 char *full_name;
938 if (e->flags & UNINTERESTING)
939 continue;
941 if (repo_dwim_ref(the_repository, e->name, strlen(e->name),
942 &oid, &full_name, 0) != 1)
943 continue;
945 if (refspecs.nr) {
946 char *private;
947 private = apply_refspecs(&refspecs, full_name);
948 if (private) {
949 free(full_name);
950 full_name = private;
954 commit = get_commit(e, full_name);
955 if (!commit) {
956 warning("%s: Unexpected object of type %s, skipping.",
957 e->name,
958 type_name(e->item->type));
959 continue;
962 switch(commit->object.type) {
963 case OBJ_COMMIT:
964 break;
965 case OBJ_BLOB:
966 export_blob(&commit->object.oid);
967 continue;
968 default: /* OBJ_TAG (nested tags) is already handled */
969 warning("Tag points to object of unexpected type %s, skipping.",
970 type_name(commit->object.type));
971 continue;
975 * Make sure this ref gets properly updated eventually, whether
976 * through a commit or manually at the end.
978 if (e->item->type != OBJ_TAG)
979 string_list_append(&extra_refs, full_name)->util = commit;
981 if (!*revision_sources_at(&revision_sources, commit))
982 *revision_sources_at(&revision_sources, commit) = full_name;
985 string_list_sort(&extra_refs);
986 string_list_remove_duplicates(&extra_refs, 0);
989 static void handle_tags_and_duplicates(struct string_list *extras)
991 struct commit *commit;
992 int i;
994 for (i = extras->nr - 1; i >= 0; i--) {
995 const char *name = extras->items[i].string;
996 struct object *object = extras->items[i].util;
997 int mark;
999 switch (object->type) {
1000 case OBJ_TAG:
1001 handle_tag(name, (struct tag *)object);
1002 break;
1003 case OBJ_COMMIT:
1004 if (anonymize)
1005 name = anonymize_refname(name);
1006 /* create refs pointing to already seen commits */
1007 commit = rewrite_commit((struct commit *)object);
1008 if (!commit) {
1010 * Neither this object nor any of its
1011 * ancestors touch any relevant paths, so
1012 * it has been filtered to nothing. Delete
1013 * it.
1015 printf("reset %s\nfrom %s\n\n",
1016 name, oid_to_hex(null_oid()));
1017 continue;
1020 mark = get_object_mark(&commit->object);
1021 if (!mark) {
1023 * Getting here means we have a commit which
1024 * was excluded by a negative refspec (e.g.
1025 * fast-export ^HEAD HEAD). If we are
1026 * referencing excluded commits, set the ref
1027 * to the exact commit. Otherwise, the user
1028 * wants the branch exported but every commit
1029 * in its history to be deleted, which basically
1030 * just means deletion of the ref.
1032 if (!reference_excluded_commits) {
1033 /* delete the ref */
1034 printf("reset %s\nfrom %s\n\n",
1035 name, oid_to_hex(null_oid()));
1036 continue;
1038 /* set ref to commit using oid, not mark */
1039 printf("reset %s\nfrom %s\n\n", name,
1040 oid_to_hex(&commit->object.oid));
1041 continue;
1044 printf("reset %s\nfrom :%d\n\n", name, mark
1046 show_progress();
1047 break;
1052 static void export_marks(char *file)
1054 unsigned int i;
1055 uint32_t mark;
1056 struct decoration_entry *deco = idnums.entries;
1057 FILE *f;
1058 int e = 0;
1060 f = fopen_for_writing(file);
1061 if (!f)
1062 die_errno("Unable to open marks file %s for writing.", file);
1064 for (i = 0; i < idnums.size; i++) {
1065 if (deco->base && deco->base->type == 1) {
1066 mark = ptr_to_mark(deco->decoration);
1067 if (fprintf(f, ":%"PRIu32" %s\n", mark,
1068 oid_to_hex(&deco->base->oid)) < 0) {
1069 e = 1;
1070 break;
1073 deco++;
1076 e |= ferror(f);
1077 e |= fclose(f);
1078 if (e)
1079 error("Unable to write marks file %s.", file);
1082 static void import_marks(char *input_file, int check_exists)
1084 char line[512];
1085 FILE *f;
1086 struct stat sb;
1088 if (check_exists && stat(input_file, &sb))
1089 return;
1091 f = xfopen(input_file, "r");
1092 while (fgets(line, sizeof(line), f)) {
1093 uint32_t mark;
1094 char *line_end, *mark_end;
1095 struct object_id oid;
1096 struct object *object;
1097 struct commit *commit;
1098 enum object_type type;
1100 line_end = strchr(line, '\n');
1101 if (line[0] != ':' || !line_end)
1102 die("corrupt mark line: %s", line);
1103 *line_end = '\0';
1105 mark = strtoumax(line + 1, &mark_end, 10);
1106 if (!mark || mark_end == line + 1
1107 || *mark_end != ' ' || get_oid_hex(mark_end + 1, &oid))
1108 die("corrupt mark line: %s", line);
1110 if (last_idnum < mark)
1111 last_idnum = mark;
1113 type = oid_object_info(the_repository, &oid, NULL);
1114 if (type < 0)
1115 die("object not found: %s", oid_to_hex(&oid));
1117 if (type != OBJ_COMMIT)
1118 /* only commits */
1119 continue;
1121 commit = lookup_commit(the_repository, &oid);
1122 if (!commit)
1123 die("not a commit? can't happen: %s", oid_to_hex(&oid));
1125 object = &commit->object;
1127 if (object->flags & SHOWN)
1128 error("Object %s already has a mark", oid_to_hex(&oid));
1130 mark_object(object, mark);
1132 object->flags |= SHOWN;
1134 fclose(f);
1137 static void handle_deletes(void)
1139 int i;
1140 for (i = 0; i < refspecs.nr; i++) {
1141 struct refspec_item *refspec = &refspecs.items[i];
1142 if (*refspec->src)
1143 continue;
1145 printf("reset %s\nfrom %s\n\n",
1146 refspec->dst, oid_to_hex(null_oid()));
1150 static int parse_opt_anonymize_map(const struct option *opt,
1151 const char *arg, int unset)
1153 struct hashmap *map = opt->value;
1154 const char *delim, *value;
1155 size_t keylen;
1157 BUG_ON_OPT_NEG(unset);
1159 delim = strchr(arg, ':');
1160 if (delim) {
1161 keylen = delim - arg;
1162 value = delim + 1;
1163 } else {
1164 keylen = strlen(arg);
1165 value = arg;
1168 if (!keylen || !*value)
1169 return error(_("--anonymize-map token cannot be empty"));
1171 add_anonymized_entry(map, memhash(arg, keylen), arg, keylen,
1172 xstrdup(value));
1174 return 0;
1177 int cmd_fast_export(int argc, const char **argv, const char *prefix)
1179 struct rev_info revs;
1180 struct commit *commit;
1181 char *export_filename = NULL,
1182 *import_filename = NULL,
1183 *import_filename_if_exists = NULL;
1184 uint32_t lastimportid;
1185 struct string_list refspecs_list = STRING_LIST_INIT_NODUP;
1186 struct string_list paths_of_changed_objects = STRING_LIST_INIT_DUP;
1187 struct option options[] = {
1188 OPT_INTEGER(0, "progress", &progress,
1189 N_("show progress after <n> objects")),
1190 OPT_CALLBACK(0, "signed-tags", &signed_tag_mode, N_("mode"),
1191 N_("select handling of signed tags"),
1192 parse_opt_signed_tag_mode),
1193 OPT_CALLBACK(0, "tag-of-filtered-object", &tag_of_filtered_mode, N_("mode"),
1194 N_("select handling of tags that tag filtered objects"),
1195 parse_opt_tag_of_filtered_mode),
1196 OPT_CALLBACK(0, "reencode", &reencode_mode, N_("mode"),
1197 N_("select handling of commit messages in an alternate encoding"),
1198 parse_opt_reencode_mode),
1199 OPT_STRING(0, "export-marks", &export_filename, N_("file"),
1200 N_("dump marks to this file")),
1201 OPT_STRING(0, "import-marks", &import_filename, N_("file"),
1202 N_("import marks from this file")),
1203 OPT_STRING(0, "import-marks-if-exists",
1204 &import_filename_if_exists,
1205 N_("file"),
1206 N_("import marks from this file if it exists")),
1207 OPT_BOOL(0, "fake-missing-tagger", &fake_missing_tagger,
1208 N_("fake a tagger when tags lack one")),
1209 OPT_BOOL(0, "full-tree", &full_tree,
1210 N_("output full tree for each commit")),
1211 OPT_BOOL(0, "use-done-feature", &use_done_feature,
1212 N_("use the done feature to terminate the stream")),
1213 OPT_BOOL(0, "no-data", &no_data, N_("skip output of blob data")),
1214 OPT_STRING_LIST(0, "refspec", &refspecs_list, N_("refspec"),
1215 N_("apply refspec to exported refs")),
1216 OPT_BOOL(0, "anonymize", &anonymize, N_("anonymize output")),
1217 OPT_CALLBACK_F(0, "anonymize-map", &anonymized_seeds, N_("from:to"),
1218 N_("convert <from> to <to> in anonymized output"),
1219 PARSE_OPT_NONEG, parse_opt_anonymize_map),
1220 OPT_BOOL(0, "reference-excluded-parents",
1221 &reference_excluded_commits, N_("reference parents which are not in fast-export stream by object id")),
1222 OPT_BOOL(0, "show-original-ids", &show_original_ids,
1223 N_("show original object ids of blobs/commits")),
1224 OPT_BOOL(0, "mark-tags", &mark_tags,
1225 N_("label tags with mark ids")),
1227 OPT_END()
1230 if (argc == 1)
1231 usage_with_options (fast_export_usage, options);
1233 /* we handle encodings */
1234 git_config(git_default_config, NULL);
1236 repo_init_revisions(the_repository, &revs, prefix);
1237 init_revision_sources(&revision_sources);
1238 revs.topo_order = 1;
1239 revs.sources = &revision_sources;
1240 revs.rewrite_parents = 1;
1241 argc = parse_options(argc, argv, prefix, options, fast_export_usage,
1242 PARSE_OPT_KEEP_ARGV0 | PARSE_OPT_KEEP_UNKNOWN_OPT);
1243 argc = setup_revisions(argc, argv, &revs, NULL);
1244 if (argc > 1)
1245 usage_with_options (fast_export_usage, options);
1247 if (anonymized_seeds.cmpfn && !anonymize)
1248 die(_("the option '%s' requires '%s'"), "--anonymize-map", "--anonymize");
1250 if (refspecs_list.nr) {
1251 int i;
1253 for (i = 0; i < refspecs_list.nr; i++)
1254 refspec_append(&refspecs, refspecs_list.items[i].string);
1256 string_list_clear(&refspecs_list, 1);
1259 if (use_done_feature)
1260 printf("feature done\n");
1262 if (import_filename && import_filename_if_exists)
1263 die(_("options '%s' and '%s' cannot be used together"), "--import-marks", "--import-marks-if-exists");
1264 if (import_filename)
1265 import_marks(import_filename, 0);
1266 else if (import_filename_if_exists)
1267 import_marks(import_filename_if_exists, 1);
1268 lastimportid = last_idnum;
1270 if (import_filename && revs.prune_data.nr)
1271 full_tree = 1;
1273 get_tags_and_duplicates(&revs.cmdline);
1275 if (prepare_revision_walk(&revs))
1276 die("revision walk setup failed");
1278 revs.reverse = 1;
1279 revs.diffopt.format_callback = show_filemodify;
1280 revs.diffopt.format_callback_data = &paths_of_changed_objects;
1281 revs.diffopt.flags.recursive = 1;
1282 revs.diffopt.no_free = 1;
1283 while ((commit = get_revision(&revs)))
1284 handle_commit(commit, &revs, &paths_of_changed_objects);
1286 handle_tags_and_duplicates(&extra_refs);
1287 handle_tags_and_duplicates(&tag_refs);
1288 handle_deletes();
1290 if (export_filename && lastimportid != last_idnum)
1291 export_marks(export_filename);
1293 if (use_done_feature)
1294 printf("done\n");
1296 refspec_clear(&refspecs);
1297 release_revisions(&revs);
1299 return 0;