2 * "Ostensibly Recursive's Twin" merge strategy, or "ort" for short. Meant
3 * as a drop-in replacement for the "recursive" merge strategy, allowing one
6 * git merge [-s recursive]
12 * Note: git's parser allows the space between '-s' and its argument to be
13 * missing. (Should I have backronymed "ham", "alsa", "kip", "nap, "alvo",
14 * "cale", "peedy", or "ins" instead of "ort"?)
18 #include "merge-ort.h"
23 #include "cache-tree.h"
25 #include "commit-reach.h"
31 #include "object-store.h"
32 #include "promisor-remote.h"
35 #include "submodule-config.h"
36 #include "submodule.h"
38 #include "unpack-trees.h"
39 #include "xdiff-interface.h"
42 * We have many arrays of size 3. Whenever we have such an array, the
43 * indices refer to one of the sides of the three-way merge. This is so
44 * pervasive that the constants 0, 1, and 2 are used in many places in the
45 * code (especially in arithmetic operations to find the other side's index
46 * or to compute a relevant mask), but sometimes these enum names are used
47 * to aid code clarity.
49 * See also 'filemask' and 'dirmask' in struct conflict_info; the "ith side"
50 * referred to there is one of these three sides.
58 static unsigned RESULT_INITIALIZED
= 0x1abe11ed; /* unlikely accidental value */
60 struct traversal_callback_data
{
62 unsigned long dirmask
;
63 struct name_entry names
[3];
66 struct deferred_traversal_data
{
68 * possible_trivial_merges: directories to be explored only when needed
70 * possible_trivial_merges is a map of directory names to
71 * dir_rename_mask. When we detect that a directory is unchanged on
72 * one side, we can sometimes resolve the directory without recursing
73 * into it. Renames are the only things that can prevent such an
74 * optimization. However, for rename sources:
75 * - If no parent directory needed directory rename detection, then
76 * no path under such a directory can be a relevant_source.
77 * and for rename destinations:
78 * - If no cached rename has a target path under the directory AND
79 * - If there are no unpaired relevant_sources elsewhere in the
81 * then we don't need any path under this directory for a rename
82 * destination. The only way to know the last item above is to defer
83 * handling such directories until the end of collect_merge_info(),
84 * in handle_deferred_entries().
86 * For each we store dir_rename_mask, since that's the only bit of
87 * information we need, other than the path, to resume the recursive
90 struct strintmap possible_trivial_merges
;
93 * trivial_merges_okay: if trivial directory merges are okay
95 * See possible_trivial_merges above. The "no unpaired
96 * relevant_sources elsewhere in the repository" is a single boolean
97 * per merge side, which we store here. Note that while 0 means no,
98 * 1 only means "maybe" rather than "yes"; we optimistically set it
99 * to 1 initially and only clear when we determine it is unsafe to
100 * do trivial directory merges.
102 unsigned trivial_merges_okay
;
105 * target_dirs: ancestor directories of rename targets
107 * target_dirs contains all directory names that are an ancestor of
108 * any rename destination.
110 struct strset target_dirs
;
115 * All variables that are arrays of size 3 correspond to data tracked
116 * for the sides in enum merge_side. Index 0 is almost always unused
117 * because we often only need to track information for MERGE_SIDE1 and
118 * MERGE_SIDE2 (MERGE_BASE can't have rename information since renames
119 * are determined relative to what changed since the MERGE_BASE).
123 * pairs: pairing of filenames from diffcore_rename()
125 struct diff_queue_struct pairs
[3];
128 * dirs_removed: directories removed on a given side of history.
130 * The keys of dirs_removed[side] are the directories that were removed
131 * on the given side of history. The value of the strintmap for each
132 * directory is a value from enum dir_rename_relevance.
134 struct strintmap dirs_removed
[3];
137 * dir_rename_count: tracking where parts of a directory were renamed to
139 * When files in a directory are renamed, they may not all go to the
140 * same location. Each strmap here tracks:
141 * old_dir => {new_dir => int}
142 * That is, dir_rename_count[side] is a strmap to a strintmap.
144 struct strmap dir_rename_count
[3];
147 * dir_renames: computed directory renames
149 * This is a map of old_dir => new_dir and is derived in part from
152 struct strmap dir_renames
[3];
155 * relevant_sources: deleted paths wanted in rename detection, and why
157 * relevant_sources is a set of deleted paths on each side of
158 * history for which we need rename detection. If a path is deleted
159 * on one side of history, we need to detect if it is part of a
161 * * the file is modified/deleted on the other side of history
162 * * we need to detect renames for an ancestor directory
163 * If neither of those are true, we can skip rename detection for
164 * that path. The reason is stored as a value from enum
165 * file_rename_relevance, as the reason can inform the algorithm in
166 * diffcore_rename_extended().
168 struct strintmap relevant_sources
[3];
170 struct deferred_traversal_data deferred
[3];
174 * 0: optimization removing unmodified potential rename source okay
175 * 2 or 4: optimization okay, but must check for files added to dir
176 * 7: optimization forbidden; need rename source in case of dir rename
178 unsigned dir_rename_mask
:3;
181 * callback_data_*: supporting data structures for alternate traversal
183 * We sometimes need to be able to traverse through all the files
184 * in a given tree before all immediate subdirectories within that
185 * tree. Since traverse_trees() doesn't do that naturally, we have
186 * a traverse_trees_wrapper() that stores any immediate
187 * subdirectories while traversing files, then traverses the
188 * immediate subdirectories later. These callback_data* variables
189 * store the information for the subdirectories so that we can do
190 * that traversal order.
192 struct traversal_callback_data
*callback_data
;
193 int callback_data_nr
, callback_data_alloc
;
194 char *callback_data_traverse_path
;
197 * merge_trees: trees passed to the merge algorithm for the merge
199 * merge_trees records the trees passed to the merge algorithm. But,
200 * this data also is stored in merge_result->priv. If a sequence of
201 * merges are being done (such as when cherry-picking or rebasing),
202 * the next merge can look at this and re-use information from
203 * previous merges under certain circumstances.
205 * See also all the cached_* variables.
207 struct tree
*merge_trees
[3];
210 * cached_pairs_valid_side: which side's cached info can be reused
212 * See the description for merge_trees. For repeated merges, at most
213 * only one side's cached information can be used. Valid values:
214 * MERGE_SIDE2: cached data from side2 can be reused
215 * MERGE_SIDE1: cached data from side1 can be reused
216 * 0: no cached data can be reused
217 * -1: See redo_after_renames; both sides can be reused.
219 int cached_pairs_valid_side
;
222 * cached_pairs: Caching of renames and deletions.
224 * These are mappings recording renames and deletions of individual
225 * files (not directories). They are thus a map from an old
226 * filename to either NULL (for deletions) or a new filename (for
229 struct strmap cached_pairs
[3];
232 * cached_target_names: just the destinations from cached_pairs
234 * We sometimes want a fast lookup to determine if a given filename
235 * is one of the destinations in cached_pairs. cached_target_names
236 * is thus duplicative information, but it provides a fast lookup.
238 struct strset cached_target_names
[3];
241 * cached_irrelevant: Caching of rename_sources that aren't relevant.
243 * If we try to detect a rename for a source path and succeed, it's
244 * part of a rename. If we try to detect a rename for a source path
245 * and fail, then it's a delete. If we do not try to detect a rename
246 * for a path, then we don't know if it's a rename or a delete. If
247 * merge-ort doesn't think the path is relevant, then we just won't
248 * cache anything for that path. But there's a slight problem in
249 * that merge-ort can think a path is RELEVANT_LOCATION, but due to
250 * commit 9bd342137e ("diffcore-rename: determine which
251 * relevant_sources are no longer relevant", 2021-03-13),
252 * diffcore-rename can downgrade the path to RELEVANT_NO_MORE. To
253 * avoid excessive calls to diffcore_rename_extended() we still need
254 * to cache such paths, though we cannot record them as either
255 * renames or deletes. So we cache them here as a "turned out to be
256 * irrelevant *for this commit*" as they are often also irrelevant
257 * for subsequent commits, though we will have to do some extra
258 * checking to see whether such paths become relevant for rename
259 * detection when cherry-picking/rebasing subsequent commits.
261 struct strset cached_irrelevant
[3];
264 * redo_after_renames: optimization flag for "restarting" the merge
266 * Sometimes it pays to detect renames, cache them, and then
267 * restart the merge operation from the beginning. The reason for
268 * this is that when we know where all the renames are, we know
269 * whether a certain directory has any paths under it affected --
270 * and if a directory is not affected then it permits us to do
271 * trivial tree merging in more cases. Doing trivial tree merging
272 * prevents the need to run process_entry() on every path
273 * underneath trees that can be trivially merged, and
274 * process_entry() is more expensive than collect_merge_info() --
275 * plus, the second collect_merge_info() will be much faster since
276 * it doesn't have to recurse into the relevant trees.
278 * Values for this flag:
279 * 0 = don't bother, not worth it (or conditions not yet checked)
280 * 1 = conditions for optimization met, optimization worthwhile
281 * 2 = we already did it (don't restart merge yet again)
283 unsigned redo_after_renames
;
286 * needed_limit: value needed for inexact rename detection to run
288 * If the current rename limit wasn't high enough for inexact
289 * rename detection to run, this records the limit needed. Otherwise,
290 * this value remains 0.
295 struct merge_options_internal
{
297 * paths: primary data structure in all of merge ort.
300 * * are full relative paths from the toplevel of the repository
301 * (e.g. "drivers/firmware/raspberrypi.c").
302 * * store all relevant paths in the repo, both directories and
303 * files (e.g. drivers, drivers/firmware would also be included)
304 * * these keys serve to intern all the path strings, which allows
305 * us to do pointer comparison on directory names instead of
306 * strcmp; we just have to be careful to use the interned strings.
308 * The values of paths:
309 * * either a pointer to a merged_info, or a conflict_info struct
310 * * merged_info contains all relevant information for a
311 * non-conflicted entry.
312 * * conflict_info contains a merged_info, plus any additional
313 * information about a conflict such as the higher orders stages
314 * involved and the names of the paths those came from (handy
315 * once renames get involved).
316 * * a path may start "conflicted" (i.e. point to a conflict_info)
317 * and then a later step (e.g. three-way content merge) determines
318 * it can be cleanly merged, at which point it'll be marked clean
319 * and the algorithm will ignore any data outside the contained
320 * merged_info for that entry
321 * * If an entry remains conflicted, the merged_info portion of a
322 * conflict_info will later be filled with whatever version of
323 * the file should be placed in the working directory (e.g. an
324 * as-merged-as-possible variation that contains conflict markers).
329 * conflicted: a subset of keys->values from "paths"
331 * conflicted is basically an optimization between process_entries()
332 * and record_conflicted_index_entries(); the latter could loop over
333 * ALL the entries in paths AGAIN and look for the ones that are
334 * still conflicted, but since process_entries() has to loop over
335 * all of them, it saves the ones it couldn't resolve in this strmap
336 * so that record_conflicted_index_entries() can iterate just the
339 struct strmap conflicted
;
342 * pool: memory pool for fast allocation/deallocation
344 * We allocate room for lots of filenames and auxiliary data
345 * structures in merge_options_internal, and it tends to all be
346 * freed together too. Using a memory pool for these provides a
349 struct mem_pool pool
;
352 * output: special messages and conflict notices for various paths
354 * This is a map of pathnames (a subset of the keys in "paths" above)
355 * to strbufs. It gathers various warning/conflict/notice messages
356 * for later processing.
358 struct strmap output
;
361 * renames: various data relating to rename detection
363 struct rename_info renames
;
366 * attr_index: hacky minimal index used for renormalization
368 * renormalization code _requires_ an index, though it only needs to
369 * find a .gitattributes file within the index. So, when
370 * renormalization is important, we create a special index with just
373 struct index_state attr_index
;
376 * current_dir_name, toplevel_dir: temporary vars
378 * These are used in collect_merge_info_callback(), and will set the
379 * various merged_info.directory_name for the various paths we get;
380 * see documentation for that variable and the requirements placed on
383 const char *current_dir_name
;
384 const char *toplevel_dir
;
386 /* call_depth: recursion level counter for merging merge bases */
390 struct version_info
{
391 struct object_id oid
;
396 /* if is_null, ignore result. otherwise result has oid & mode */
397 struct version_info result
;
401 * clean: whether the path in question is cleanly merged.
403 * see conflict_info.merged for more details.
408 * basename_offset: offset of basename of path.
410 * perf optimization to avoid recomputing offset of final '/'
411 * character in pathname (0 if no '/' in pathname).
413 size_t basename_offset
;
416 * directory_name: containing directory name.
418 * Note that we assume directory_name is constructed such that
419 * strcmp(dir1_name, dir2_name) == 0 iff dir1_name == dir2_name,
420 * i.e. string equality is equivalent to pointer equality. For this
421 * to hold, we have to be careful setting directory_name.
423 const char *directory_name
;
426 struct conflict_info
{
428 * merged: the version of the path that will be written to working tree
430 * WARNING: It is critical to check merged.clean and ensure it is 0
431 * before reading any conflict_info fields outside of merged.
432 * Allocated merge_info structs will always have clean set to 1.
433 * Allocated conflict_info structs will have merged.clean set to 0
434 * initially. The merged.clean field is how we know if it is safe
435 * to access other parts of conflict_info besides merged; if a
436 * conflict_info's merged.clean is changed to 1, the rest of the
437 * algorithm is not allowed to look at anything outside of the
438 * merged member anymore.
440 struct merged_info merged
;
442 /* oids & modes from each of the three trees for this path */
443 struct version_info stages
[3];
445 /* pathnames for each stage; may differ due to rename detection */
446 const char *pathnames
[3];
448 /* Whether this path is/was involved in a directory/file conflict */
449 unsigned df_conflict
:1;
452 * Whether this path is/was involved in a non-content conflict other
453 * than a directory/file conflict (e.g. rename/rename, rename/delete,
454 * file location based on possible directory rename).
456 unsigned path_conflict
:1;
459 * For filemask and dirmask, the ith bit corresponds to whether the
460 * ith entry is a file (filemask) or a directory (dirmask). Thus,
461 * filemask & dirmask is always zero, and filemask | dirmask is at
462 * most 7 but can be less when a path does not appear as either a
463 * file or a directory on at least one side of history.
465 * Note that these masks are related to enum merge_side, as the ith
466 * entry corresponds to side i.
468 * These values come from a traverse_trees() call; more info may be
469 * found looking at tree-walk.h's struct traverse_info,
470 * particularly the documentation above the "fn" member (note that
471 * filemask = mask & ~dirmask from that documentation).
477 * Optimization to track which stages match, to avoid the need to
478 * recompute it in multiple steps. Either 0 or at least 2 bits are
479 * set; if at least 2 bits are set, their corresponding stages match.
481 unsigned match_mask
:3;
484 /*** Function Grouping: various utility functions ***/
487 * For the next three macros, see warning for conflict_info.merged.
489 * In each of the below, mi is a struct merged_info*, and ci was defined
490 * as a struct conflict_info* (but we need to verify ci isn't actually
491 * pointed at a struct merged_info*).
493 * INITIALIZE_CI: Assign ci to mi but only if it's safe; set to NULL otherwise.
494 * VERIFY_CI: Ensure that something we assigned to a conflict_info* is one.
495 * ASSIGN_AND_VERIFY_CI: Similar to VERIFY_CI but do assignment first.
497 #define INITIALIZE_CI(ci, mi) do { \
498 (ci) = (!(mi) || (mi)->clean) ? NULL : (struct conflict_info *)(mi); \
500 #define VERIFY_CI(ci) assert(ci && !ci->merged.clean);
501 #define ASSIGN_AND_VERIFY_CI(ci, mi) do { \
502 (ci) = (struct conflict_info *)(mi); \
503 assert((ci) && !(mi)->clean); \
506 static void free_strmap_strings(struct strmap
*map
)
508 struct hashmap_iter iter
;
509 struct strmap_entry
*entry
;
511 strmap_for_each_entry(map
, &iter
, entry
) {
512 free((char*)entry
->key
);
516 static void clear_or_reinit_internal_opts(struct merge_options_internal
*opti
,
519 struct rename_info
*renames
= &opti
->renames
;
521 void (*strmap_clear_func
)(struct strmap
*, int) =
522 reinitialize
? strmap_partial_clear
: strmap_clear
;
523 void (*strintmap_clear_func
)(struct strintmap
*) =
524 reinitialize
? strintmap_partial_clear
: strintmap_clear
;
525 void (*strset_clear_func
)(struct strset
*) =
526 reinitialize
? strset_partial_clear
: strset_clear
;
528 strmap_clear_func(&opti
->paths
, 0);
531 * All keys and values in opti->conflicted are a subset of those in
532 * opti->paths. We don't want to deallocate anything twice, so we
533 * don't free the keys and we pass 0 for free_values.
535 strmap_clear_func(&opti
->conflicted
, 0);
537 if (opti
->attr_index
.cache_nr
) /* true iff opt->renormalize */
538 discard_index(&opti
->attr_index
);
540 /* Free memory used by various renames maps */
541 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; ++i
) {
542 strintmap_clear_func(&renames
->dirs_removed
[i
]);
543 strmap_clear_func(&renames
->dir_renames
[i
], 0);
544 strintmap_clear_func(&renames
->relevant_sources
[i
]);
546 assert(renames
->cached_pairs_valid_side
== 0);
547 if (i
!= renames
->cached_pairs_valid_side
&&
548 -1 != renames
->cached_pairs_valid_side
) {
549 strset_clear_func(&renames
->cached_target_names
[i
]);
550 strmap_clear_func(&renames
->cached_pairs
[i
], 1);
551 strset_clear_func(&renames
->cached_irrelevant
[i
]);
552 partial_clear_dir_rename_count(&renames
->dir_rename_count
[i
]);
554 strmap_clear(&renames
->dir_rename_count
[i
], 1);
557 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; ++i
) {
558 strintmap_clear_func(&renames
->deferred
[i
].possible_trivial_merges
);
559 strset_clear_func(&renames
->deferred
[i
].target_dirs
);
560 renames
->deferred
[i
].trivial_merges_okay
= 1; /* 1 == maybe */
562 renames
->cached_pairs_valid_side
= 0;
563 renames
->dir_rename_mask
= 0;
566 struct hashmap_iter iter
;
567 struct strmap_entry
*e
;
569 /* Release and free each strbuf found in output */
570 strmap_for_each_entry(&opti
->output
, &iter
, e
) {
571 struct strbuf
*sb
= e
->value
;
574 * While strictly speaking we don't need to free(sb)
575 * here because we could pass free_values=1 when
576 * calling strmap_clear() on opti->output, that would
577 * require strmap_clear to do another
578 * strmap_for_each_entry() loop, so we just free it
579 * while we're iterating anyway.
583 strmap_clear(&opti
->output
, 0);
586 mem_pool_discard(&opti
->pool
, 0);
588 /* Clean out callback_data as well. */
589 FREE_AND_NULL(renames
->callback_data
);
590 renames
->callback_data_nr
= renames
->callback_data_alloc
= 0;
593 __attribute__((format (printf
, 2, 3)))
594 static int err(struct merge_options
*opt
, const char *err
, ...)
597 struct strbuf sb
= STRBUF_INIT
;
599 strbuf_addstr(&sb
, "error: ");
600 va_start(params
, err
);
601 strbuf_vaddf(&sb
, err
, params
);
610 static void format_commit(struct strbuf
*sb
,
612 struct repository
*repo
,
613 struct commit
*commit
)
615 struct merge_remote_desc
*desc
;
616 struct pretty_print_context ctx
= {0};
617 ctx
.abbrev
= DEFAULT_ABBREV
;
619 strbuf_addchars(sb
, ' ', indent
);
620 desc
= merge_remote_util(commit
);
622 strbuf_addf(sb
, "virtual %s\n", desc
->name
);
626 repo_format_commit_message(repo
, commit
, "%h %s", sb
, &ctx
);
627 strbuf_addch(sb
, '\n');
630 __attribute__((format (printf
, 4, 5)))
631 static void path_msg(struct merge_options
*opt
,
633 int omittable_hint
, /* skippable under --remerge-diff */
634 const char *fmt
, ...)
637 struct strbuf
*sb
, *dest
;
638 struct strbuf tmp
= STRBUF_INIT
;
640 if (opt
->record_conflict_msgs_as_headers
&& omittable_hint
)
641 return; /* Do not record mere hints in headers */
642 if (opt
->record_conflict_msgs_as_headers
&& opt
->priv
->call_depth
)
643 return; /* Do not record inner merge issues in headers */
644 sb
= strmap_get(&opt
->priv
->output
, path
);
646 sb
= xmalloc(sizeof(*sb
));
648 strmap_put(&opt
->priv
->output
, path
, sb
);
651 dest
= (opt
->record_conflict_msgs_as_headers
? &tmp
: sb
);
654 if (opt
->priv
->call_depth
) {
655 strbuf_addchars(dest
, ' ', 2);
656 strbuf_addstr(dest
, "From inner merge:");
657 strbuf_addchars(dest
, ' ', opt
->priv
->call_depth
* 2);
659 strbuf_vaddf(dest
, fmt
, ap
);
662 if (opt
->record_conflict_msgs_as_headers
) {
663 int i_sb
= 0, i_tmp
= 0;
665 /* Start with the specified prefix */
666 if (opt
->msg_header_prefix
)
667 strbuf_addf(sb
, "%s ", opt
->msg_header_prefix
);
669 /* Copy tmp to sb, adding spaces after newlines */
670 strbuf_grow(sb
, sb
->len
+ 2*tmp
.len
); /* more than sufficient */
671 for (; i_tmp
< tmp
.len
; i_tmp
++, i_sb
++) {
672 /* Copy next character from tmp to sb */
673 sb
->buf
[sb
->len
+ i_sb
] = tmp
.buf
[i_tmp
];
675 /* If we copied a newline, add a space */
676 if (tmp
.buf
[i_tmp
] == '\n')
677 sb
->buf
[++i_sb
] = ' ';
679 /* Update length and ensure it's NUL-terminated */
681 sb
->buf
[sb
->len
] = '\0';
683 strbuf_release(&tmp
);
686 /* Add final newline character to sb */
687 strbuf_addch(sb
, '\n');
690 static struct diff_filespec
*pool_alloc_filespec(struct mem_pool
*pool
,
693 /* Similar to alloc_filespec(), but allocate from pool and reuse path */
694 struct diff_filespec
*spec
;
696 spec
= mem_pool_calloc(pool
, 1, sizeof(*spec
));
697 spec
->path
= (char*)path
; /* spec won't modify it */
700 spec
->is_binary
= -1;
704 static struct diff_filepair
*pool_diff_queue(struct mem_pool
*pool
,
705 struct diff_queue_struct
*queue
,
706 struct diff_filespec
*one
,
707 struct diff_filespec
*two
)
709 /* Same code as diff_queue(), except allocate from pool */
710 struct diff_filepair
*dp
;
712 dp
= mem_pool_calloc(pool
, 1, sizeof(*dp
));
720 /* add a string to a strbuf, but converting "/" to "_" */
721 static void add_flattened_path(struct strbuf
*out
, const char *s
)
724 strbuf_addstr(out
, s
);
725 for (; i
< out
->len
; i
++)
726 if (out
->buf
[i
] == '/')
730 static char *unique_path(struct merge_options
*opt
,
735 struct strbuf newpath
= STRBUF_INIT
;
738 struct strmap
*existing_paths
= &opt
->priv
->paths
;
740 strbuf_addf(&newpath
, "%s~", path
);
741 add_flattened_path(&newpath
, branch
);
743 base_len
= newpath
.len
;
744 while (strmap_contains(existing_paths
, newpath
.buf
)) {
745 strbuf_setlen(&newpath
, base_len
);
746 strbuf_addf(&newpath
, "_%d", suffix
++);
749 /* Track the new path in our memory pool */
750 ret
= mem_pool_alloc(&opt
->priv
->pool
, newpath
.len
+ 1);
751 memcpy(ret
, newpath
.buf
, newpath
.len
+ 1);
752 strbuf_release(&newpath
);
756 /*** Function Grouping: functions related to collect_merge_info() ***/
758 static int traverse_trees_wrapper_callback(int n
,
760 unsigned long dirmask
,
761 struct name_entry
*names
,
762 struct traverse_info
*info
)
764 struct merge_options
*opt
= info
->data
;
765 struct rename_info
*renames
= &opt
->priv
->renames
;
766 unsigned filemask
= mask
& ~dirmask
;
770 if (!renames
->callback_data_traverse_path
)
771 renames
->callback_data_traverse_path
= xstrdup(info
->traverse_path
);
773 if (filemask
&& filemask
== renames
->dir_rename_mask
)
774 renames
->dir_rename_mask
= 0x07;
776 ALLOC_GROW(renames
->callback_data
, renames
->callback_data_nr
+ 1,
777 renames
->callback_data_alloc
);
778 renames
->callback_data
[renames
->callback_data_nr
].mask
= mask
;
779 renames
->callback_data
[renames
->callback_data_nr
].dirmask
= dirmask
;
780 COPY_ARRAY(renames
->callback_data
[renames
->callback_data_nr
].names
,
782 renames
->callback_data_nr
++;
788 * Much like traverse_trees(), BUT:
789 * - read all the tree entries FIRST, saving them
790 * - note that the above step provides an opportunity to compute necessary
791 * additional details before the "real" traversal
792 * - loop through the saved entries and call the original callback on them
794 static int traverse_trees_wrapper(struct index_state
*istate
,
797 struct traverse_info
*info
)
799 int ret
, i
, old_offset
;
800 traverse_callback_t old_fn
;
801 char *old_callback_data_traverse_path
;
802 struct merge_options
*opt
= info
->data
;
803 struct rename_info
*renames
= &opt
->priv
->renames
;
805 assert(renames
->dir_rename_mask
== 2 || renames
->dir_rename_mask
== 4);
807 old_callback_data_traverse_path
= renames
->callback_data_traverse_path
;
809 old_offset
= renames
->callback_data_nr
;
811 renames
->callback_data_traverse_path
= NULL
;
812 info
->fn
= traverse_trees_wrapper_callback
;
813 ret
= traverse_trees(istate
, n
, t
, info
);
817 info
->traverse_path
= renames
->callback_data_traverse_path
;
819 for (i
= old_offset
; i
< renames
->callback_data_nr
; ++i
) {
821 renames
->callback_data
[i
].mask
,
822 renames
->callback_data
[i
].dirmask
,
823 renames
->callback_data
[i
].names
,
827 renames
->callback_data_nr
= old_offset
;
828 free(renames
->callback_data_traverse_path
);
829 renames
->callback_data_traverse_path
= old_callback_data_traverse_path
;
830 info
->traverse_path
= NULL
;
834 static void setup_path_info(struct merge_options
*opt
,
835 struct string_list_item
*result
,
836 const char *current_dir_name
,
837 int current_dir_name_len
,
838 char *fullpath
, /* we'll take over ownership */
839 struct name_entry
*names
,
840 struct name_entry
*merged_version
,
841 unsigned is_null
, /* boolean */
842 unsigned df_conflict
, /* boolean */
845 int resolved
/* boolean */)
847 /* result->util is void*, so mi is a convenience typed variable */
848 struct merged_info
*mi
;
850 assert(!is_null
|| resolved
);
851 assert(!df_conflict
|| !resolved
); /* df_conflict implies !resolved */
852 assert(resolved
== (merged_version
!= NULL
));
854 mi
= mem_pool_calloc(&opt
->priv
->pool
, 1,
855 resolved
? sizeof(struct merged_info
) :
856 sizeof(struct conflict_info
));
857 mi
->directory_name
= current_dir_name
;
858 mi
->basename_offset
= current_dir_name_len
;
859 mi
->clean
= !!resolved
;
861 mi
->result
.mode
= merged_version
->mode
;
862 oidcpy(&mi
->result
.oid
, &merged_version
->oid
);
863 mi
->is_null
= !!is_null
;
866 struct conflict_info
*ci
;
868 ASSIGN_AND_VERIFY_CI(ci
, mi
);
869 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
870 ci
->pathnames
[i
] = fullpath
;
871 ci
->stages
[i
].mode
= names
[i
].mode
;
872 oidcpy(&ci
->stages
[i
].oid
, &names
[i
].oid
);
874 ci
->filemask
= filemask
;
875 ci
->dirmask
= dirmask
;
876 ci
->df_conflict
= !!df_conflict
;
879 * Assume is_null for now, but if we have entries
880 * under the directory then when it is complete in
881 * write_completed_directory() it'll update this.
882 * Also, for D/F conflicts, we have to handle the
883 * directory first, then clear this bit and process
884 * the file to see how it is handled -- that occurs
885 * near the top of process_entry().
889 strmap_put(&opt
->priv
->paths
, fullpath
, mi
);
890 result
->string
= fullpath
;
894 static void add_pair(struct merge_options
*opt
,
895 struct name_entry
*names
,
896 const char *pathname
,
898 unsigned is_add
/* if false, is_delete */,
900 unsigned dir_rename_mask
)
902 struct diff_filespec
*one
, *two
;
903 struct rename_info
*renames
= &opt
->priv
->renames
;
904 int names_idx
= is_add
? side
: 0;
907 assert(match_mask
== 0 || match_mask
== 6);
908 if (strset_contains(&renames
->cached_target_names
[side
],
912 unsigned content_relevant
= (match_mask
== 0);
913 unsigned location_relevant
= (dir_rename_mask
== 0x07);
915 assert(match_mask
== 0 || match_mask
== 3 || match_mask
== 5);
918 * If pathname is found in cached_irrelevant[side] due to
919 * previous pick but for this commit content is relevant,
920 * then we need to remove it from cached_irrelevant.
922 if (content_relevant
)
923 /* strset_remove is no-op if strset doesn't have key */
924 strset_remove(&renames
->cached_irrelevant
[side
],
928 * We do not need to re-detect renames for paths that we already
929 * know the pairing, i.e. for cached_pairs (or
930 * cached_irrelevant). However, handle_deferred_entries() needs
931 * to loop over the union of keys from relevant_sources[side] and
932 * cached_pairs[side], so for simplicity we set relevant_sources
933 * for all the cached_pairs too and then strip them back out in
934 * prune_cached_from_relevant() at the beginning of
935 * detect_regular_renames().
937 if (content_relevant
|| location_relevant
) {
938 /* content_relevant trumps location_relevant */
939 strintmap_set(&renames
->relevant_sources
[side
], pathname
,
940 content_relevant
? RELEVANT_CONTENT
: RELEVANT_LOCATION
);
944 * Avoid creating pair if we've already cached rename results.
945 * Note that we do this after setting relevant_sources[side]
946 * as noted in the comment above.
948 if (strmap_contains(&renames
->cached_pairs
[side
], pathname
) ||
949 strset_contains(&renames
->cached_irrelevant
[side
], pathname
))
953 one
= pool_alloc_filespec(&opt
->priv
->pool
, pathname
);
954 two
= pool_alloc_filespec(&opt
->priv
->pool
, pathname
);
955 fill_filespec(is_add
? two
: one
,
956 &names
[names_idx
].oid
, 1, names
[names_idx
].mode
);
957 pool_diff_queue(&opt
->priv
->pool
, &renames
->pairs
[side
], one
, two
);
960 static void collect_rename_info(struct merge_options
*opt
,
961 struct name_entry
*names
,
963 const char *fullname
,
968 struct rename_info
*renames
= &opt
->priv
->renames
;
972 * Update dir_rename_mask (determines ignore-rename-source validity)
974 * dir_rename_mask helps us keep track of when directory rename
975 * detection may be relevant. Basically, whenver a directory is
976 * removed on one side of history, and a file is added to that
977 * directory on the other side of history, directory rename
978 * detection is relevant (meaning we have to detect renames for all
979 * files within that directory to deduce where the directory
980 * moved). Also, whenever a directory needs directory rename
981 * detection, due to the "majority rules" choice for where to move
982 * it (see t6423 testcase 1f), we also need to detect renames for
983 * all files within subdirectories of that directory as well.
985 * Here we haven't looked at files within the directory yet, we are
986 * just looking at the directory itself. So, if we aren't yet in
987 * a case where a parent directory needed directory rename detection
988 * (i.e. dir_rename_mask != 0x07), and if the directory was removed
989 * on one side of history, record the mask of the other side of
990 * history in dir_rename_mask.
992 if (renames
->dir_rename_mask
!= 0x07 &&
993 (dirmask
== 3 || dirmask
== 5)) {
994 /* simple sanity check */
995 assert(renames
->dir_rename_mask
== 0 ||
996 renames
->dir_rename_mask
== (dirmask
& ~1));
997 /* update dir_rename_mask; have it record mask of new side */
998 renames
->dir_rename_mask
= (dirmask
& ~1);
1001 /* Update dirs_removed, as needed */
1002 if (dirmask
== 1 || dirmask
== 3 || dirmask
== 5) {
1003 /* absent_mask = 0x07 - dirmask; sides = absent_mask/2 */
1004 unsigned sides
= (0x07 - dirmask
)/2;
1005 unsigned relevance
= (renames
->dir_rename_mask
== 0x07) ?
1006 RELEVANT_FOR_ANCESTOR
: NOT_RELEVANT
;
1008 * Record relevance of this directory. However, note that
1009 * when collect_merge_info_callback() recurses into this
1010 * directory and calls collect_rename_info() on paths
1011 * within that directory, if we find a path that was added
1012 * to this directory on the other side of history, we will
1013 * upgrade this value to RELEVANT_FOR_SELF; see below.
1016 strintmap_set(&renames
->dirs_removed
[1], fullname
,
1019 strintmap_set(&renames
->dirs_removed
[2], fullname
,
1024 * Here's the block that potentially upgrades to RELEVANT_FOR_SELF.
1025 * When we run across a file added to a directory. In such a case,
1026 * find the directory of the file and upgrade its relevance.
1028 if (renames
->dir_rename_mask
== 0x07 &&
1029 (filemask
== 2 || filemask
== 4)) {
1031 * Need directory rename for parent directory on other side
1032 * of history from added file. Thus
1033 * side = (~filemask & 0x06) >> 1
1035 * side = 3 - (filemask/2).
1037 unsigned side
= 3 - (filemask
>> 1);
1038 strintmap_set(&renames
->dirs_removed
[side
], dirname
,
1042 if (filemask
== 0 || filemask
== 7)
1045 for (side
= MERGE_SIDE1
; side
<= MERGE_SIDE2
; ++side
) {
1046 unsigned side_mask
= (1 << side
);
1048 /* Check for deletion on side */
1049 if ((filemask
& 1) && !(filemask
& side_mask
))
1050 add_pair(opt
, names
, fullname
, side
, 0 /* delete */,
1051 match_mask
& filemask
,
1052 renames
->dir_rename_mask
);
1054 /* Check for addition on side */
1055 if (!(filemask
& 1) && (filemask
& side_mask
))
1056 add_pair(opt
, names
, fullname
, side
, 1 /* add */,
1057 match_mask
& filemask
,
1058 renames
->dir_rename_mask
);
1062 static int collect_merge_info_callback(int n
,
1064 unsigned long dirmask
,
1065 struct name_entry
*names
,
1066 struct traverse_info
*info
)
1070 * common ancestor (mbase) has mask 1, and stored in index 0 of names
1071 * head of side 1 (side1) has mask 2, and stored in index 1 of names
1072 * head of side 2 (side2) has mask 4, and stored in index 2 of names
1074 struct merge_options
*opt
= info
->data
;
1075 struct merge_options_internal
*opti
= opt
->priv
;
1076 struct rename_info
*renames
= &opt
->priv
->renames
;
1077 struct string_list_item pi
; /* Path Info */
1078 struct conflict_info
*ci
; /* typed alias to pi.util (which is void*) */
1079 struct name_entry
*p
;
1082 const char *dirname
= opti
->current_dir_name
;
1083 unsigned prev_dir_rename_mask
= renames
->dir_rename_mask
;
1084 unsigned filemask
= mask
& ~dirmask
;
1085 unsigned match_mask
= 0; /* will be updated below */
1086 unsigned mbase_null
= !(mask
& 1);
1087 unsigned side1_null
= !(mask
& 2);
1088 unsigned side2_null
= !(mask
& 4);
1089 unsigned side1_matches_mbase
= (!side1_null
&& !mbase_null
&&
1090 names
[0].mode
== names
[1].mode
&&
1091 oideq(&names
[0].oid
, &names
[1].oid
));
1092 unsigned side2_matches_mbase
= (!side2_null
&& !mbase_null
&&
1093 names
[0].mode
== names
[2].mode
&&
1094 oideq(&names
[0].oid
, &names
[2].oid
));
1095 unsigned sides_match
= (!side1_null
&& !side2_null
&&
1096 names
[1].mode
== names
[2].mode
&&
1097 oideq(&names
[1].oid
, &names
[2].oid
));
1100 * Note: When a path is a file on one side of history and a directory
1101 * in another, we have a directory/file conflict. In such cases, if
1102 * the conflict doesn't resolve from renames and deletions, then we
1103 * always leave directories where they are and move files out of the
1104 * way. Thus, while struct conflict_info has a df_conflict field to
1105 * track such conflicts, we ignore that field for any directories at
1106 * a path and only pay attention to it for files at the given path.
1107 * The fact that we leave directories were they are also means that
1108 * we do not need to worry about getting additional df_conflict
1109 * information propagated from parent directories down to children
1110 * (unlike, say traverse_trees_recursive() in unpack-trees.c, which
1111 * sets a newinfo.df_conflicts field specifically to propagate it).
1113 unsigned df_conflict
= (filemask
!= 0) && (dirmask
!= 0);
1115 /* n = 3 is a fundamental assumption. */
1117 BUG("Called collect_merge_info_callback wrong");
1120 * A bunch of sanity checks verifying that traverse_trees() calls
1121 * us the way I expect. Could just remove these at some point,
1122 * though maybe they are helpful to future code readers.
1124 assert(mbase_null
== is_null_oid(&names
[0].oid
));
1125 assert(side1_null
== is_null_oid(&names
[1].oid
));
1126 assert(side2_null
== is_null_oid(&names
[2].oid
));
1127 assert(!mbase_null
|| !side1_null
|| !side2_null
);
1128 assert(mask
> 0 && mask
< 8);
1130 /* Determine match_mask */
1131 if (side1_matches_mbase
)
1132 match_mask
= (side2_matches_mbase
? 7 : 3);
1133 else if (side2_matches_mbase
)
1135 else if (sides_match
)
1139 * Get the name of the relevant filepath, which we'll pass to
1140 * setup_path_info() for tracking.
1145 len
= traverse_path_len(info
, p
->pathlen
);
1147 /* +1 in both of the following lines to include the NUL byte */
1148 fullpath
= mem_pool_alloc(&opt
->priv
->pool
, len
+ 1);
1149 make_traverse_path(fullpath
, len
+ 1, info
, p
->path
, p
->pathlen
);
1152 * If mbase, side1, and side2 all match, we can resolve early. Even
1153 * if these are trees, there will be no renames or anything
1156 if (side1_matches_mbase
&& side2_matches_mbase
) {
1157 /* mbase, side1, & side2 all match; use mbase as resolution */
1158 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1159 names
, names
+0, mbase_null
, 0 /* df_conflict */,
1160 filemask
, dirmask
, 1 /* resolved */);
1165 * If the sides match, and all three paths are present and are
1166 * files, then we can take either as the resolution. We can't do
1167 * this with trees, because there may be rename sources from the
1170 if (sides_match
&& filemask
== 0x07) {
1171 /* use side1 (== side2) version as resolution */
1172 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1173 names
, names
+1, side1_null
, 0,
1174 filemask
, dirmask
, 1);
1179 * If side1 matches mbase and all three paths are present and are
1180 * files, then we can use side2 as the resolution. We cannot
1181 * necessarily do so this for trees, because there may be rename
1182 * destinations within side2.
1184 if (side1_matches_mbase
&& filemask
== 0x07) {
1185 /* use side2 version as resolution */
1186 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1187 names
, names
+2, side2_null
, 0,
1188 filemask
, dirmask
, 1);
1192 /* Similar to above but swapping sides 1 and 2 */
1193 if (side2_matches_mbase
&& filemask
== 0x07) {
1194 /* use side1 version as resolution */
1195 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1196 names
, names
+1, side1_null
, 0,
1197 filemask
, dirmask
, 1);
1202 * Sometimes we can tell that a source path need not be included in
1203 * rename detection -- namely, whenever either
1204 * side1_matches_mbase && side2_null
1206 * side2_matches_mbase && side1_null
1207 * However, we call collect_rename_info() even in those cases,
1208 * because exact renames are cheap and would let us remove both a
1209 * source and destination path. We'll cull the unneeded sources
1212 collect_rename_info(opt
, names
, dirname
, fullpath
,
1213 filemask
, dirmask
, match_mask
);
1216 * None of the special cases above matched, so we have a
1217 * provisional conflict. (Rename detection might allow us to
1218 * unconflict some more cases, but that comes later so all we can
1219 * do now is record the different non-null file hashes.)
1221 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1222 names
, NULL
, 0, df_conflict
, filemask
, dirmask
, 0);
1226 ci
->match_mask
= match_mask
;
1228 /* If dirmask, recurse into subdirectories */
1230 struct traverse_info newinfo
;
1231 struct tree_desc t
[3];
1232 void *buf
[3] = {NULL
, NULL
, NULL
};
1233 const char *original_dir_name
;
1237 * Check for whether we can avoid recursing due to one side
1238 * matching the merge base. The side that does NOT match is
1239 * the one that might have a rename destination we need.
1241 assert(!side1_matches_mbase
|| !side2_matches_mbase
);
1242 side
= side1_matches_mbase
? MERGE_SIDE2
:
1243 side2_matches_mbase
? MERGE_SIDE1
: MERGE_BASE
;
1244 if (filemask
== 0 && (dirmask
== 2 || dirmask
== 4)) {
1246 * Also defer recursing into new directories; set up a
1247 * few variables to let us do so.
1249 ci
->match_mask
= (7 - dirmask
);
1252 if (renames
->dir_rename_mask
!= 0x07 &&
1253 side
!= MERGE_BASE
&&
1254 renames
->deferred
[side
].trivial_merges_okay
&&
1255 !strset_contains(&renames
->deferred
[side
].target_dirs
,
1257 strintmap_set(&renames
->deferred
[side
].possible_trivial_merges
,
1258 pi
.string
, renames
->dir_rename_mask
);
1259 renames
->dir_rename_mask
= prev_dir_rename_mask
;
1263 /* We need to recurse */
1264 ci
->match_mask
&= filemask
;
1266 newinfo
.prev
= info
;
1267 newinfo
.name
= p
->path
;
1268 newinfo
.namelen
= p
->pathlen
;
1269 newinfo
.pathlen
= st_add3(newinfo
.pathlen
, p
->pathlen
, 1);
1271 * If this directory we are about to recurse into cared about
1272 * its parent directory (the current directory) having a D/F
1273 * conflict, then we'd propagate the masks in this way:
1274 * newinfo.df_conflicts |= (mask & ~dirmask);
1275 * But we don't worry about propagating D/F conflicts. (See
1276 * comment near setting of local df_conflict variable near
1277 * the beginning of this function).
1280 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
1281 if (i
== 1 && side1_matches_mbase
)
1283 else if (i
== 2 && side2_matches_mbase
)
1285 else if (i
== 2 && sides_match
)
1288 const struct object_id
*oid
= NULL
;
1290 oid
= &names
[i
].oid
;
1291 buf
[i
] = fill_tree_descriptor(opt
->repo
,
1297 original_dir_name
= opti
->current_dir_name
;
1298 opti
->current_dir_name
= pi
.string
;
1299 if (renames
->dir_rename_mask
== 0 ||
1300 renames
->dir_rename_mask
== 0x07)
1301 ret
= traverse_trees(NULL
, 3, t
, &newinfo
);
1303 ret
= traverse_trees_wrapper(NULL
, 3, t
, &newinfo
);
1304 opti
->current_dir_name
= original_dir_name
;
1305 renames
->dir_rename_mask
= prev_dir_rename_mask
;
1307 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++)
1317 static void resolve_trivial_directory_merge(struct conflict_info
*ci
, int side
)
1320 assert((side
== 1 && ci
->match_mask
== 5) ||
1321 (side
== 2 && ci
->match_mask
== 3));
1322 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[side
].oid
);
1323 ci
->merged
.result
.mode
= ci
->stages
[side
].mode
;
1324 ci
->merged
.is_null
= is_null_oid(&ci
->stages
[side
].oid
);
1326 ci
->merged
.clean
= 1; /* (ci->filemask == 0); */
1329 static int handle_deferred_entries(struct merge_options
*opt
,
1330 struct traverse_info
*info
)
1332 struct rename_info
*renames
= &opt
->priv
->renames
;
1333 struct hashmap_iter iter
;
1334 struct strmap_entry
*entry
;
1336 int path_count_before
, path_count_after
= 0;
1338 path_count_before
= strmap_get_size(&opt
->priv
->paths
);
1339 for (side
= MERGE_SIDE1
; side
<= MERGE_SIDE2
; side
++) {
1340 unsigned optimization_okay
= 1;
1341 struct strintmap copy
;
1343 /* Loop over the set of paths we need to know rename info for */
1344 strset_for_each_entry(&renames
->relevant_sources
[side
],
1346 char *rename_target
, *dir
, *dir_marker
;
1347 struct strmap_entry
*e
;
1350 * If we don't know delete/rename info for this path,
1351 * then we need to recurse into all trees to get all
1352 * adds to make sure we have it.
1354 if (strset_contains(&renames
->cached_irrelevant
[side
],
1357 e
= strmap_get_entry(&renames
->cached_pairs
[side
],
1360 optimization_okay
= 0;
1364 /* If this is a delete, we have enough info already */
1365 rename_target
= e
->value
;
1369 /* If we already walked the rename target, we're good */
1370 if (strmap_contains(&opt
->priv
->paths
, rename_target
))
1374 * Otherwise, we need to get a list of directories that
1375 * will need to be recursed into to get this
1378 dir
= xstrdup(rename_target
);
1379 while ((dir_marker
= strrchr(dir
, '/'))) {
1381 if (strset_contains(&renames
->deferred
[side
].target_dirs
,
1384 strset_add(&renames
->deferred
[side
].target_dirs
,
1389 renames
->deferred
[side
].trivial_merges_okay
= optimization_okay
;
1391 * We need to recurse into any directories in
1392 * possible_trivial_merges[side] found in target_dirs[side].
1393 * But when we recurse, we may need to queue up some of the
1394 * subdirectories for possible_trivial_merges[side]. Since
1395 * we can't safely iterate through a hashmap while also adding
1396 * entries, move the entries into 'copy', iterate over 'copy',
1397 * and then we'll also iterate anything added into
1398 * possible_trivial_merges[side] once this loop is done.
1400 copy
= renames
->deferred
[side
].possible_trivial_merges
;
1401 strintmap_init_with_options(&renames
->deferred
[side
].possible_trivial_merges
,
1405 strintmap_for_each_entry(©
, &iter
, entry
) {
1406 const char *path
= entry
->key
;
1407 unsigned dir_rename_mask
= (intptr_t)entry
->value
;
1408 struct conflict_info
*ci
;
1410 struct tree_desc t
[3];
1411 void *buf
[3] = {NULL
,};
1414 ci
= strmap_get(&opt
->priv
->paths
, path
);
1416 dirmask
= ci
->dirmask
;
1418 if (optimization_okay
&&
1419 !strset_contains(&renames
->deferred
[side
].target_dirs
,
1421 resolve_trivial_directory_merge(ci
, side
);
1426 info
->namelen
= strlen(path
);
1427 info
->pathlen
= info
->namelen
+ 1;
1429 for (i
= 0; i
< 3; i
++, dirmask
>>= 1) {
1430 if (i
== 1 && ci
->match_mask
== 3)
1432 else if (i
== 2 && ci
->match_mask
== 5)
1434 else if (i
== 2 && ci
->match_mask
== 6)
1437 const struct object_id
*oid
= NULL
;
1439 oid
= &ci
->stages
[i
].oid
;
1440 buf
[i
] = fill_tree_descriptor(opt
->repo
,
1445 ci
->match_mask
&= ci
->filemask
;
1446 opt
->priv
->current_dir_name
= path
;
1447 renames
->dir_rename_mask
= dir_rename_mask
;
1448 if (renames
->dir_rename_mask
== 0 ||
1449 renames
->dir_rename_mask
== 0x07)
1450 ret
= traverse_trees(NULL
, 3, t
, info
);
1452 ret
= traverse_trees_wrapper(NULL
, 3, t
, info
);
1454 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++)
1460 strintmap_clear(©
);
1461 strintmap_for_each_entry(&renames
->deferred
[side
].possible_trivial_merges
,
1463 const char *path
= entry
->key
;
1464 struct conflict_info
*ci
;
1466 ci
= strmap_get(&opt
->priv
->paths
, path
);
1469 assert(renames
->deferred
[side
].trivial_merges_okay
&&
1470 !strset_contains(&renames
->deferred
[side
].target_dirs
,
1472 resolve_trivial_directory_merge(ci
, side
);
1474 if (!optimization_okay
|| path_count_after
)
1475 path_count_after
= strmap_get_size(&opt
->priv
->paths
);
1477 if (path_count_after
) {
1479 * The choice of wanted_factor here does not affect
1480 * correctness, only performance. When the
1481 * path_count_after / path_count_before
1482 * ratio is high, redoing after renames is a big
1483 * performance boost. I suspect that redoing is a wash
1484 * somewhere near a value of 2, and below that redoing will
1485 * slow things down. I applied a fudge factor and picked
1486 * 3; see the commit message when this was introduced for
1487 * back of the envelope calculations for this ratio.
1489 const int wanted_factor
= 3;
1491 /* We should only redo collect_merge_info one time */
1492 assert(renames
->redo_after_renames
== 0);
1494 if (path_count_after
/ path_count_before
>= wanted_factor
) {
1495 renames
->redo_after_renames
= 1;
1496 renames
->cached_pairs_valid_side
= -1;
1498 } else if (renames
->redo_after_renames
== 2)
1499 renames
->redo_after_renames
= 0;
1503 static int collect_merge_info(struct merge_options
*opt
,
1504 struct tree
*merge_base
,
1509 struct tree_desc t
[3];
1510 struct traverse_info info
;
1512 opt
->priv
->toplevel_dir
= "";
1513 opt
->priv
->current_dir_name
= opt
->priv
->toplevel_dir
;
1514 setup_traverse_info(&info
, opt
->priv
->toplevel_dir
);
1515 info
.fn
= collect_merge_info_callback
;
1517 info
.show_all_errors
= 1;
1519 parse_tree(merge_base
);
1522 init_tree_desc(t
+ 0, merge_base
->buffer
, merge_base
->size
);
1523 init_tree_desc(t
+ 1, side1
->buffer
, side1
->size
);
1524 init_tree_desc(t
+ 2, side2
->buffer
, side2
->size
);
1526 trace2_region_enter("merge", "traverse_trees", opt
->repo
);
1527 ret
= traverse_trees(NULL
, 3, t
, &info
);
1529 ret
= handle_deferred_entries(opt
, &info
);
1530 trace2_region_leave("merge", "traverse_trees", opt
->repo
);
1535 /*** Function Grouping: functions related to threeway content merges ***/
1537 static int find_first_merges(struct repository
*repo
,
1541 struct object_array
*result
)
1544 struct object_array merges
= OBJECT_ARRAY_INIT
;
1545 struct commit
*commit
;
1546 int contains_another
;
1548 char merged_revision
[GIT_MAX_HEXSZ
+ 2];
1549 const char *rev_args
[] = { "rev-list", "--merges", "--ancestry-path",
1550 "--all", merged_revision
, NULL
};
1551 struct rev_info revs
;
1552 struct setup_revision_opt rev_opts
;
1554 memset(result
, 0, sizeof(struct object_array
));
1555 memset(&rev_opts
, 0, sizeof(rev_opts
));
1557 /* get all revisions that merge commit a */
1558 xsnprintf(merged_revision
, sizeof(merged_revision
), "^%s",
1559 oid_to_hex(&a
->object
.oid
));
1560 repo_init_revisions(repo
, &revs
, NULL
);
1561 /* FIXME: can't handle linked worktrees in submodules yet */
1562 revs
.single_worktree
= path
!= NULL
;
1563 setup_revisions(ARRAY_SIZE(rev_args
)-1, rev_args
, &revs
, &rev_opts
);
1565 /* save all revisions from the above list that contain b */
1566 if (prepare_revision_walk(&revs
))
1567 die("revision walk setup failed");
1568 while ((commit
= get_revision(&revs
)) != NULL
) {
1569 struct object
*o
= &(commit
->object
);
1570 if (repo_in_merge_bases(repo
, b
, commit
))
1571 add_object_array(o
, NULL
, &merges
);
1573 reset_revision_walk();
1575 /* Now we've got all merges that contain a and b. Prune all
1576 * merges that contain another found merge and save them in
1579 for (i
= 0; i
< merges
.nr
; i
++) {
1580 struct commit
*m1
= (struct commit
*) merges
.objects
[i
].item
;
1582 contains_another
= 0;
1583 for (j
= 0; j
< merges
.nr
; j
++) {
1584 struct commit
*m2
= (struct commit
*) merges
.objects
[j
].item
;
1585 if (i
!= j
&& repo_in_merge_bases(repo
, m2
, m1
)) {
1586 contains_another
= 1;
1591 if (!contains_another
)
1592 add_object_array(merges
.objects
[i
].item
, NULL
, result
);
1595 object_array_clear(&merges
);
1599 static int merge_submodule(struct merge_options
*opt
,
1601 const struct object_id
*o
,
1602 const struct object_id
*a
,
1603 const struct object_id
*b
,
1604 struct object_id
*result
)
1606 struct repository subrepo
;
1607 struct strbuf sb
= STRBUF_INIT
;
1609 struct commit
*commit_o
, *commit_a
, *commit_b
;
1611 struct object_array merges
;
1614 int search
= !opt
->priv
->call_depth
;
1616 /* store fallback answer in result in case we fail */
1617 oidcpy(result
, opt
->priv
->call_depth
? o
: a
);
1619 /* we can not handle deletion conflicts */
1627 if (repo_submodule_init(&subrepo
, opt
->repo
, path
, null_oid())) {
1628 path_msg(opt
, path
, 0,
1629 _("Failed to merge submodule %s (not checked out)"),
1634 if (!(commit_o
= lookup_commit_reference(&subrepo
, o
)) ||
1635 !(commit_a
= lookup_commit_reference(&subrepo
, a
)) ||
1636 !(commit_b
= lookup_commit_reference(&subrepo
, b
))) {
1637 path_msg(opt
, path
, 0,
1638 _("Failed to merge submodule %s (commits not present)"),
1643 /* check whether both changes are forward */
1644 if (!repo_in_merge_bases(&subrepo
, commit_o
, commit_a
) ||
1645 !repo_in_merge_bases(&subrepo
, commit_o
, commit_b
)) {
1646 path_msg(opt
, path
, 0,
1647 _("Failed to merge submodule %s "
1648 "(commits don't follow merge-base)"),
1653 /* Case #1: a is contained in b or vice versa */
1654 if (repo_in_merge_bases(&subrepo
, commit_a
, commit_b
)) {
1656 path_msg(opt
, path
, 1,
1657 _("Note: Fast-forwarding submodule %s to %s"),
1658 path
, oid_to_hex(b
));
1662 if (repo_in_merge_bases(&subrepo
, commit_b
, commit_a
)) {
1664 path_msg(opt
, path
, 1,
1665 _("Note: Fast-forwarding submodule %s to %s"),
1666 path
, oid_to_hex(a
));
1672 * Case #2: There are one or more merges that contain a and b in
1673 * the submodule. If there is only one, then present it as a
1674 * suggestion to the user, but leave it marked unmerged so the
1675 * user needs to confirm the resolution.
1678 /* Skip the search if makes no sense to the calling context. */
1682 /* find commit which merges them */
1683 parent_count
= find_first_merges(&subrepo
, path
, commit_a
, commit_b
,
1685 switch (parent_count
) {
1687 path_msg(opt
, path
, 0, _("Failed to merge submodule %s"), path
);
1691 format_commit(&sb
, 4, &subrepo
,
1692 (struct commit
*)merges
.objects
[0].item
);
1693 path_msg(opt
, path
, 0,
1694 _("Failed to merge submodule %s, but a possible merge "
1695 "resolution exists:\n%s\n"),
1697 path_msg(opt
, path
, 1,
1698 _("If this is correct simply add it to the index "
1701 " git update-index --cacheinfo 160000 %s \"%s\"\n\n"
1702 "which will accept this suggestion.\n"),
1703 oid_to_hex(&merges
.objects
[0].item
->oid
), path
);
1704 strbuf_release(&sb
);
1707 for (i
= 0; i
< merges
.nr
; i
++)
1708 format_commit(&sb
, 4, &subrepo
,
1709 (struct commit
*)merges
.objects
[i
].item
);
1710 path_msg(opt
, path
, 0,
1711 _("Failed to merge submodule %s, but multiple "
1712 "possible merges exist:\n%s"), path
, sb
.buf
);
1713 strbuf_release(&sb
);
1716 object_array_clear(&merges
);
1718 repo_clear(&subrepo
);
1722 static void initialize_attr_index(struct merge_options
*opt
)
1725 * The renormalize_buffer() functions require attributes, and
1726 * annoyingly those can only be read from the working tree or from
1727 * an index_state. merge-ort doesn't have an index_state, so we
1728 * generate a fake one containing only attribute information.
1730 struct merged_info
*mi
;
1731 struct index_state
*attr_index
= &opt
->priv
->attr_index
;
1732 struct cache_entry
*ce
;
1734 attr_index
->initialized
= 1;
1736 if (!opt
->renormalize
)
1739 mi
= strmap_get(&opt
->priv
->paths
, GITATTRIBUTES_FILE
);
1744 int len
= strlen(GITATTRIBUTES_FILE
);
1745 ce
= make_empty_cache_entry(attr_index
, len
);
1746 ce
->ce_mode
= create_ce_mode(mi
->result
.mode
);
1747 ce
->ce_flags
= create_ce_flags(0);
1748 ce
->ce_namelen
= len
;
1749 oidcpy(&ce
->oid
, &mi
->result
.oid
);
1750 memcpy(ce
->name
, GITATTRIBUTES_FILE
, len
);
1751 add_index_entry(attr_index
, ce
,
1752 ADD_CACHE_OK_TO_ADD
| ADD_CACHE_OK_TO_REPLACE
);
1753 get_stream_filter(attr_index
, GITATTRIBUTES_FILE
, &ce
->oid
);
1756 struct conflict_info
*ci
;
1758 ASSIGN_AND_VERIFY_CI(ci
, mi
);
1759 for (stage
= 0; stage
< 3; stage
++) {
1760 unsigned stage_mask
= (1 << stage
);
1762 if (!(ci
->filemask
& stage_mask
))
1764 len
= strlen(GITATTRIBUTES_FILE
);
1765 ce
= make_empty_cache_entry(attr_index
, len
);
1766 ce
->ce_mode
= create_ce_mode(ci
->stages
[stage
].mode
);
1767 ce
->ce_flags
= create_ce_flags(stage
);
1768 ce
->ce_namelen
= len
;
1769 oidcpy(&ce
->oid
, &ci
->stages
[stage
].oid
);
1770 memcpy(ce
->name
, GITATTRIBUTES_FILE
, len
);
1771 add_index_entry(attr_index
, ce
,
1772 ADD_CACHE_OK_TO_ADD
| ADD_CACHE_OK_TO_REPLACE
);
1773 get_stream_filter(attr_index
, GITATTRIBUTES_FILE
,
1779 static int merge_3way(struct merge_options
*opt
,
1781 const struct object_id
*o
,
1782 const struct object_id
*a
,
1783 const struct object_id
*b
,
1784 const char *pathnames
[3],
1785 const int extra_marker_size
,
1786 mmbuffer_t
*result_buf
)
1788 mmfile_t orig
, src1
, src2
;
1789 struct ll_merge_options ll_opts
= {0};
1790 char *base
, *name1
, *name2
;
1791 enum ll_merge_result merge_status
;
1793 if (!opt
->priv
->attr_index
.initialized
)
1794 initialize_attr_index(opt
);
1796 ll_opts
.renormalize
= opt
->renormalize
;
1797 ll_opts
.extra_marker_size
= extra_marker_size
;
1798 ll_opts
.xdl_opts
= opt
->xdl_opts
;
1800 if (opt
->priv
->call_depth
) {
1801 ll_opts
.virtual_ancestor
= 1;
1802 ll_opts
.variant
= 0;
1804 switch (opt
->recursive_variant
) {
1805 case MERGE_VARIANT_OURS
:
1806 ll_opts
.variant
= XDL_MERGE_FAVOR_OURS
;
1808 case MERGE_VARIANT_THEIRS
:
1809 ll_opts
.variant
= XDL_MERGE_FAVOR_THEIRS
;
1812 ll_opts
.variant
= 0;
1817 assert(pathnames
[0] && pathnames
[1] && pathnames
[2] && opt
->ancestor
);
1818 if (pathnames
[0] == pathnames
[1] && pathnames
[1] == pathnames
[2]) {
1819 base
= mkpathdup("%s", opt
->ancestor
);
1820 name1
= mkpathdup("%s", opt
->branch1
);
1821 name2
= mkpathdup("%s", opt
->branch2
);
1823 base
= mkpathdup("%s:%s", opt
->ancestor
, pathnames
[0]);
1824 name1
= mkpathdup("%s:%s", opt
->branch1
, pathnames
[1]);
1825 name2
= mkpathdup("%s:%s", opt
->branch2
, pathnames
[2]);
1828 read_mmblob(&orig
, o
);
1829 read_mmblob(&src1
, a
);
1830 read_mmblob(&src2
, b
);
1832 merge_status
= ll_merge(result_buf
, path
, &orig
, base
,
1833 &src1
, name1
, &src2
, name2
,
1834 &opt
->priv
->attr_index
, &ll_opts
);
1835 if (merge_status
== LL_MERGE_BINARY_CONFLICT
)
1836 path_msg(opt
, path
, 0,
1837 "warning: Cannot merge binary files: %s (%s vs. %s)",
1838 path
, name1
, name2
);
1846 return merge_status
;
1849 static int handle_content_merge(struct merge_options
*opt
,
1851 const struct version_info
*o
,
1852 const struct version_info
*a
,
1853 const struct version_info
*b
,
1854 const char *pathnames
[3],
1855 const int extra_marker_size
,
1856 struct version_info
*result
)
1859 * path is the target location where we want to put the file, and
1860 * is used to determine any normalization rules in ll_merge.
1862 * The normal case is that path and all entries in pathnames are
1863 * identical, though renames can affect which path we got one of
1864 * the three blobs to merge on various sides of history.
1866 * extra_marker_size is the amount to extend conflict markers in
1867 * ll_merge; this is neeed if we have content merges of content
1868 * merges, which happens for example with rename/rename(2to1) and
1869 * rename/add conflicts.
1874 * handle_content_merge() needs both files to be of the same type, i.e.
1875 * both files OR both submodules OR both symlinks. Conflicting types
1876 * needs to be handled elsewhere.
1878 assert((S_IFMT
& a
->mode
) == (S_IFMT
& b
->mode
));
1881 if (a
->mode
== b
->mode
|| a
->mode
== o
->mode
)
1882 result
->mode
= b
->mode
;
1884 /* must be the 100644/100755 case */
1885 assert(S_ISREG(a
->mode
));
1886 result
->mode
= a
->mode
;
1887 clean
= (b
->mode
== o
->mode
);
1889 * FIXME: If opt->priv->call_depth && !clean, then we really
1890 * should not make result->mode match either a->mode or
1891 * b->mode; that causes t6036 "check conflicting mode for
1892 * regular file" to fail. It would be best to use some other
1893 * mode, but we'll confuse all kinds of stuff if we use one
1894 * where S_ISREG(result->mode) isn't true, and if we use
1895 * something like 0100666, then tree-walk.c's calls to
1896 * canon_mode() will just normalize that to 100644 for us and
1897 * thus not solve anything.
1899 * Figure out if there's some kind of way we can work around
1905 * Trivial oid merge.
1907 * Note: While one might assume that the next four lines would
1908 * be unnecessary due to the fact that match_mask is often
1909 * setup and already handled, renames don't always take care
1912 if (oideq(&a
->oid
, &b
->oid
) || oideq(&a
->oid
, &o
->oid
))
1913 oidcpy(&result
->oid
, &b
->oid
);
1914 else if (oideq(&b
->oid
, &o
->oid
))
1915 oidcpy(&result
->oid
, &a
->oid
);
1917 /* Remaining rules depend on file vs. submodule vs. symlink. */
1918 else if (S_ISREG(a
->mode
)) {
1919 mmbuffer_t result_buf
;
1920 int ret
= 0, merge_status
;
1924 * If 'o' is different type, treat it as null so we do a
1927 two_way
= ((S_IFMT
& o
->mode
) != (S_IFMT
& a
->mode
));
1929 merge_status
= merge_3way(opt
, path
,
1930 two_way
? null_oid() : &o
->oid
,
1932 pathnames
, extra_marker_size
,
1935 if ((merge_status
< 0) || !result_buf
.ptr
)
1936 ret
= err(opt
, _("Failed to execute internal merge"));
1939 write_object_file(result_buf
.ptr
, result_buf
.size
,
1940 blob_type
, &result
->oid
))
1941 ret
= err(opt
, _("Unable to add %s to database"),
1944 free(result_buf
.ptr
);
1947 clean
&= (merge_status
== 0);
1948 path_msg(opt
, path
, 1, _("Auto-merging %s"), path
);
1949 } else if (S_ISGITLINK(a
->mode
)) {
1950 int two_way
= ((S_IFMT
& o
->mode
) != (S_IFMT
& a
->mode
));
1951 clean
= merge_submodule(opt
, pathnames
[0],
1952 two_way
? null_oid() : &o
->oid
,
1953 &a
->oid
, &b
->oid
, &result
->oid
);
1954 if (opt
->priv
->call_depth
&& two_way
&& !clean
) {
1955 result
->mode
= o
->mode
;
1956 oidcpy(&result
->oid
, &o
->oid
);
1958 } else if (S_ISLNK(a
->mode
)) {
1959 if (opt
->priv
->call_depth
) {
1961 result
->mode
= o
->mode
;
1962 oidcpy(&result
->oid
, &o
->oid
);
1964 switch (opt
->recursive_variant
) {
1965 case MERGE_VARIANT_NORMAL
:
1967 oidcpy(&result
->oid
, &a
->oid
);
1969 case MERGE_VARIANT_OURS
:
1970 oidcpy(&result
->oid
, &a
->oid
);
1972 case MERGE_VARIANT_THEIRS
:
1973 oidcpy(&result
->oid
, &b
->oid
);
1978 BUG("unsupported object type in the tree: %06o for %s",
1984 /*** Function Grouping: functions related to detect_and_process_renames(), ***
1985 *** which are split into directory and regular rename detection sections. ***/
1987 /*** Function Grouping: functions related to directory rename detection ***/
1989 struct collision_info
{
1990 struct string_list source_files
;
1991 unsigned reported_already
:1;
1995 * Return a new string that replaces the beginning portion (which matches
1996 * rename_info->key), with rename_info->util.new_dir. In perl-speak:
1997 * new_path_name = (old_path =~ s/rename_info->key/rename_info->value/);
1999 * Caller must ensure that old_path starts with rename_info->key + '/'.
2001 static char *apply_dir_rename(struct strmap_entry
*rename_info
,
2002 const char *old_path
)
2004 struct strbuf new_path
= STRBUF_INIT
;
2005 const char *old_dir
= rename_info
->key
;
2006 const char *new_dir
= rename_info
->value
;
2007 int oldlen
, newlen
, new_dir_len
;
2009 oldlen
= strlen(old_dir
);
2010 if (*new_dir
== '\0')
2012 * If someone renamed/merged a subdirectory into the root
2013 * directory (e.g. 'some/subdir' -> ''), then we want to
2016 * as the rename; we need to make old_path + oldlen advance
2017 * past the '/' character.
2020 new_dir_len
= strlen(new_dir
);
2021 newlen
= new_dir_len
+ (strlen(old_path
) - oldlen
) + 1;
2022 strbuf_grow(&new_path
, newlen
);
2023 strbuf_add(&new_path
, new_dir
, new_dir_len
);
2024 strbuf_addstr(&new_path
, &old_path
[oldlen
]);
2026 return strbuf_detach(&new_path
, NULL
);
2029 static int path_in_way(struct strmap
*paths
, const char *path
, unsigned side_mask
)
2031 struct merged_info
*mi
= strmap_get(paths
, path
);
2032 struct conflict_info
*ci
;
2035 INITIALIZE_CI(ci
, mi
);
2036 return mi
->clean
|| (side_mask
& (ci
->filemask
| ci
->dirmask
));
2040 * See if there is a directory rename for path, and if there are any file
2041 * level conflicts on the given side for the renamed location. If there is
2042 * a rename and there are no conflicts, return the new name. Otherwise,
2045 static char *handle_path_level_conflicts(struct merge_options
*opt
,
2047 unsigned side_index
,
2048 struct strmap_entry
*rename_info
,
2049 struct strmap
*collisions
)
2051 char *new_path
= NULL
;
2052 struct collision_info
*c_info
;
2054 struct strbuf collision_paths
= STRBUF_INIT
;
2057 * entry has the mapping of old directory name to new directory name
2058 * that we want to apply to path.
2060 new_path
= apply_dir_rename(rename_info
, path
);
2062 BUG("Failed to apply directory rename!");
2065 * The caller needs to have ensured that it has pre-populated
2066 * collisions with all paths that map to new_path. Do a quick check
2067 * to ensure that's the case.
2069 c_info
= strmap_get(collisions
, new_path
);
2071 BUG("c_info is NULL");
2074 * Check for one-sided add/add/.../add conflicts, i.e.
2075 * where implicit renames from the other side doing
2076 * directory rename(s) can affect this side of history
2077 * to put multiple paths into the same location. Warn
2078 * and bail on directory renames for such paths.
2080 if (c_info
->reported_already
) {
2082 } else if (path_in_way(&opt
->priv
->paths
, new_path
, 1 << side_index
)) {
2083 c_info
->reported_already
= 1;
2084 strbuf_add_separated_string_list(&collision_paths
, ", ",
2085 &c_info
->source_files
);
2086 path_msg(opt
, new_path
, 0,
2087 _("CONFLICT (implicit dir rename): Existing file/dir "
2088 "at %s in the way of implicit directory rename(s) "
2089 "putting the following path(s) there: %s."),
2090 new_path
, collision_paths
.buf
);
2092 } else if (c_info
->source_files
.nr
> 1) {
2093 c_info
->reported_already
= 1;
2094 strbuf_add_separated_string_list(&collision_paths
, ", ",
2095 &c_info
->source_files
);
2096 path_msg(opt
, new_path
, 0,
2097 _("CONFLICT (implicit dir rename): Cannot map more "
2098 "than one path to %s; implicit directory renames "
2099 "tried to put these paths there: %s"),
2100 new_path
, collision_paths
.buf
);
2104 /* Free memory we no longer need */
2105 strbuf_release(&collision_paths
);
2106 if (!clean
&& new_path
) {
2114 static void get_provisional_directory_renames(struct merge_options
*opt
,
2118 struct hashmap_iter iter
;
2119 struct strmap_entry
*entry
;
2120 struct rename_info
*renames
= &opt
->priv
->renames
;
2124 * dir_rename_count: old_directory -> {new_directory -> count}
2126 * dir_renames: old_directory -> best_new_directory
2127 * where best_new_directory is the one with the unique highest count.
2129 strmap_for_each_entry(&renames
->dir_rename_count
[side
], &iter
, entry
) {
2130 const char *source_dir
= entry
->key
;
2131 struct strintmap
*counts
= entry
->value
;
2132 struct hashmap_iter count_iter
;
2133 struct strmap_entry
*count_entry
;
2136 const char *best
= NULL
;
2138 strintmap_for_each_entry(counts
, &count_iter
, count_entry
) {
2139 const char *target_dir
= count_entry
->key
;
2140 intptr_t count
= (intptr_t)count_entry
->value
;
2144 else if (count
> max
) {
2153 if (bad_max
== max
) {
2154 path_msg(opt
, source_dir
, 0,
2155 _("CONFLICT (directory rename split): "
2156 "Unclear where to rename %s to; it was "
2157 "renamed to multiple other directories, with "
2158 "no destination getting a majority of the "
2163 strmap_put(&renames
->dir_renames
[side
],
2164 source_dir
, (void*)best
);
2169 static void handle_directory_level_conflicts(struct merge_options
*opt
)
2171 struct hashmap_iter iter
;
2172 struct strmap_entry
*entry
;
2173 struct string_list duplicated
= STRING_LIST_INIT_NODUP
;
2174 struct rename_info
*renames
= &opt
->priv
->renames
;
2175 struct strmap
*side1_dir_renames
= &renames
->dir_renames
[MERGE_SIDE1
];
2176 struct strmap
*side2_dir_renames
= &renames
->dir_renames
[MERGE_SIDE2
];
2179 strmap_for_each_entry(side1_dir_renames
, &iter
, entry
) {
2180 if (strmap_contains(side2_dir_renames
, entry
->key
))
2181 string_list_append(&duplicated
, entry
->key
);
2184 for (i
= 0; i
< duplicated
.nr
; i
++) {
2185 strmap_remove(side1_dir_renames
, duplicated
.items
[i
].string
, 0);
2186 strmap_remove(side2_dir_renames
, duplicated
.items
[i
].string
, 0);
2188 string_list_clear(&duplicated
, 0);
2191 static struct strmap_entry
*check_dir_renamed(const char *path
,
2192 struct strmap
*dir_renames
)
2194 char *temp
= xstrdup(path
);
2196 struct strmap_entry
*e
= NULL
;
2198 while ((end
= strrchr(temp
, '/'))) {
2200 e
= strmap_get_entry(dir_renames
, temp
);
2208 static void compute_collisions(struct strmap
*collisions
,
2209 struct strmap
*dir_renames
,
2210 struct diff_queue_struct
*pairs
)
2214 strmap_init_with_options(collisions
, NULL
, 0);
2215 if (strmap_empty(dir_renames
))
2219 * Multiple files can be mapped to the same path due to directory
2220 * renames done by the other side of history. Since that other
2221 * side of history could have merged multiple directories into one,
2222 * if our side of history added the same file basename to each of
2223 * those directories, then all N of them would get implicitly
2224 * renamed by the directory rename detection into the same path,
2225 * and we'd get an add/add/.../add conflict, and all those adds
2226 * from *this* side of history. This is not representable in the
2227 * index, and users aren't going to easily be able to make sense of
2228 * it. So we need to provide a good warning about what's
2229 * happening, and fall back to no-directory-rename detection
2230 * behavior for those paths.
2232 * See testcases 9e and all of section 5 from t6043 for examples.
2234 for (i
= 0; i
< pairs
->nr
; ++i
) {
2235 struct strmap_entry
*rename_info
;
2236 struct collision_info
*collision_info
;
2238 struct diff_filepair
*pair
= pairs
->queue
[i
];
2240 if (pair
->status
!= 'A' && pair
->status
!= 'R')
2242 rename_info
= check_dir_renamed(pair
->two
->path
, dir_renames
);
2246 new_path
= apply_dir_rename(rename_info
, pair
->two
->path
);
2248 collision_info
= strmap_get(collisions
, new_path
);
2249 if (collision_info
) {
2252 CALLOC_ARRAY(collision_info
, 1);
2253 string_list_init_nodup(&collision_info
->source_files
);
2254 strmap_put(collisions
, new_path
, collision_info
);
2256 string_list_insert(&collision_info
->source_files
,
2261 static char *check_for_directory_rename(struct merge_options
*opt
,
2263 unsigned side_index
,
2264 struct strmap
*dir_renames
,
2265 struct strmap
*dir_rename_exclusions
,
2266 struct strmap
*collisions
,
2269 char *new_path
= NULL
;
2270 struct strmap_entry
*rename_info
;
2271 struct strmap_entry
*otherinfo
= NULL
;
2272 const char *new_dir
;
2274 if (strmap_empty(dir_renames
))
2276 rename_info
= check_dir_renamed(path
, dir_renames
);
2279 /* old_dir = rename_info->key; */
2280 new_dir
= rename_info
->value
;
2283 * This next part is a little weird. We do not want to do an
2284 * implicit rename into a directory we renamed on our side, because
2285 * that will result in a spurious rename/rename(1to2) conflict. An
2287 * Base commit: dumbdir/afile, otherdir/bfile
2288 * Side 1: smrtdir/afile, otherdir/bfile
2289 * Side 2: dumbdir/afile, dumbdir/bfile
2290 * Here, while working on Side 1, we could notice that otherdir was
2291 * renamed/merged to dumbdir, and change the diff_filepair for
2292 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2293 * 2 will notice the rename from dumbdir to smrtdir, and do the
2294 * transitive rename to move it from dumbdir/bfile to
2295 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2296 * smrtdir, a rename/rename(1to2) conflict. We really just want
2297 * the file to end up in smrtdir. And the way to achieve that is
2298 * to not let Side1 do the rename to dumbdir, since we know that is
2299 * the source of one of our directory renames.
2301 * That's why otherinfo and dir_rename_exclusions is here.
2303 * As it turns out, this also prevents N-way transient rename
2304 * confusion; See testcases 9c and 9d of t6043.
2306 otherinfo
= strmap_get_entry(dir_rename_exclusions
, new_dir
);
2308 path_msg(opt
, rename_info
->key
, 1,
2309 _("WARNING: Avoiding applying %s -> %s rename "
2310 "to %s, because %s itself was renamed."),
2311 rename_info
->key
, new_dir
, path
, new_dir
);
2315 new_path
= handle_path_level_conflicts(opt
, path
, side_index
,
2316 rename_info
, collisions
);
2317 *clean_merge
&= (new_path
!= NULL
);
2322 static void apply_directory_rename_modifications(struct merge_options
*opt
,
2323 struct diff_filepair
*pair
,
2327 * The basic idea is to get the conflict_info from opt->priv->paths
2328 * at old path, and insert it into new_path; basically just this:
2329 * ci = strmap_get(&opt->priv->paths, old_path);
2330 * strmap_remove(&opt->priv->paths, old_path, 0);
2331 * strmap_put(&opt->priv->paths, new_path, ci);
2332 * However, there are some factors complicating this:
2333 * - opt->priv->paths may already have an entry at new_path
2334 * - Each ci tracks its containing directory, so we need to
2336 * - If another ci has the same containing directory, then
2337 * the two char*'s MUST point to the same location. See the
2338 * comment in struct merged_info. strcmp equality is not
2339 * enough; we need pointer equality.
2340 * - opt->priv->paths must hold the parent directories of any
2341 * entries that are added. So, if this directory rename
2342 * causes entirely new directories, we must recursively add
2343 * parent directories.
2344 * - For each parent directory added to opt->priv->paths, we
2345 * also need to get its parent directory stored in its
2346 * conflict_info->merged.directory_name with all the same
2347 * requirements about pointer equality.
2349 struct string_list dirs_to_insert
= STRING_LIST_INIT_NODUP
;
2350 struct conflict_info
*ci
, *new_ci
;
2351 struct strmap_entry
*entry
;
2352 const char *branch_with_new_path
, *branch_with_dir_rename
;
2353 const char *old_path
= pair
->two
->path
;
2354 const char *parent_name
;
2355 const char *cur_path
;
2358 entry
= strmap_get_entry(&opt
->priv
->paths
, old_path
);
2359 old_path
= entry
->key
;
2363 /* Find parent directories missing from opt->priv->paths */
2364 cur_path
= mem_pool_strdup(&opt
->priv
->pool
, new_path
);
2365 free((char*)new_path
);
2366 new_path
= (char *)cur_path
;
2369 /* Find the parent directory of cur_path */
2370 char *last_slash
= strrchr(cur_path
, '/');
2372 parent_name
= mem_pool_strndup(&opt
->priv
->pool
,
2374 last_slash
- cur_path
);
2376 parent_name
= opt
->priv
->toplevel_dir
;
2380 /* Look it up in opt->priv->paths */
2381 entry
= strmap_get_entry(&opt
->priv
->paths
, parent_name
);
2383 parent_name
= entry
->key
; /* reuse known pointer */
2387 /* Record this is one of the directories we need to insert */
2388 string_list_append(&dirs_to_insert
, parent_name
);
2389 cur_path
= parent_name
;
2392 /* Traverse dirs_to_insert and insert them into opt->priv->paths */
2393 for (i
= dirs_to_insert
.nr
-1; i
>= 0; --i
) {
2394 struct conflict_info
*dir_ci
;
2395 char *cur_dir
= dirs_to_insert
.items
[i
].string
;
2397 CALLOC_ARRAY(dir_ci
, 1);
2399 dir_ci
->merged
.directory_name
= parent_name
;
2400 len
= strlen(parent_name
);
2401 /* len+1 because of trailing '/' character */
2402 dir_ci
->merged
.basename_offset
= (len
> 0 ? len
+1 : len
);
2403 dir_ci
->dirmask
= ci
->filemask
;
2404 strmap_put(&opt
->priv
->paths
, cur_dir
, dir_ci
);
2406 parent_name
= cur_dir
;
2409 assert(ci
->filemask
== 2 || ci
->filemask
== 4);
2410 assert(ci
->dirmask
== 0);
2411 strmap_remove(&opt
->priv
->paths
, old_path
, 0);
2413 branch_with_new_path
= (ci
->filemask
== 2) ? opt
->branch1
: opt
->branch2
;
2414 branch_with_dir_rename
= (ci
->filemask
== 2) ? opt
->branch2
: opt
->branch1
;
2416 /* Now, finally update ci and stick it into opt->priv->paths */
2417 ci
->merged
.directory_name
= parent_name
;
2418 len
= strlen(parent_name
);
2419 ci
->merged
.basename_offset
= (len
> 0 ? len
+1 : len
);
2420 new_ci
= strmap_get(&opt
->priv
->paths
, new_path
);
2422 /* Place ci back into opt->priv->paths, but at new_path */
2423 strmap_put(&opt
->priv
->paths
, new_path
, ci
);
2427 /* A few sanity checks */
2429 assert(ci
->filemask
== 2 || ci
->filemask
== 4);
2430 assert((new_ci
->filemask
& ci
->filemask
) == 0);
2431 assert(!new_ci
->merged
.clean
);
2433 /* Copy stuff from ci into new_ci */
2434 new_ci
->filemask
|= ci
->filemask
;
2435 if (new_ci
->dirmask
)
2436 new_ci
->df_conflict
= 1;
2437 index
= (ci
->filemask
>> 1);
2438 new_ci
->pathnames
[index
] = ci
->pathnames
[index
];
2439 new_ci
->stages
[index
].mode
= ci
->stages
[index
].mode
;
2440 oidcpy(&new_ci
->stages
[index
].oid
, &ci
->stages
[index
].oid
);
2445 if (opt
->detect_directory_renames
== MERGE_DIRECTORY_RENAMES_TRUE
) {
2446 /* Notify user of updated path */
2447 if (pair
->status
== 'A')
2448 path_msg(opt
, new_path
, 1,
2449 _("Path updated: %s added in %s inside a "
2450 "directory that was renamed in %s; moving "
2452 old_path
, branch_with_new_path
,
2453 branch_with_dir_rename
, new_path
);
2455 path_msg(opt
, new_path
, 1,
2456 _("Path updated: %s renamed to %s in %s, "
2457 "inside a directory that was renamed in %s; "
2458 "moving it to %s."),
2459 pair
->one
->path
, old_path
, branch_with_new_path
,
2460 branch_with_dir_rename
, new_path
);
2463 * opt->detect_directory_renames has the value
2464 * MERGE_DIRECTORY_RENAMES_CONFLICT, so mark these as conflicts.
2466 ci
->path_conflict
= 1;
2467 if (pair
->status
== 'A')
2468 path_msg(opt
, new_path
, 1,
2469 _("CONFLICT (file location): %s added in %s "
2470 "inside a directory that was renamed in %s, "
2471 "suggesting it should perhaps be moved to "
2473 old_path
, branch_with_new_path
,
2474 branch_with_dir_rename
, new_path
);
2476 path_msg(opt
, new_path
, 1,
2477 _("CONFLICT (file location): %s renamed to %s "
2478 "in %s, inside a directory that was renamed "
2479 "in %s, suggesting it should perhaps be "
2481 pair
->one
->path
, old_path
, branch_with_new_path
,
2482 branch_with_dir_rename
, new_path
);
2486 * Finally, record the new location.
2488 pair
->two
->path
= new_path
;
2491 /*** Function Grouping: functions related to regular rename detection ***/
2493 static int process_renames(struct merge_options
*opt
,
2494 struct diff_queue_struct
*renames
)
2496 int clean_merge
= 1, i
;
2498 for (i
= 0; i
< renames
->nr
; ++i
) {
2499 const char *oldpath
= NULL
, *newpath
;
2500 struct diff_filepair
*pair
= renames
->queue
[i
];
2501 struct conflict_info
*oldinfo
= NULL
, *newinfo
= NULL
;
2502 struct strmap_entry
*old_ent
, *new_ent
;
2503 unsigned int old_sidemask
;
2504 int target_index
, other_source_index
;
2505 int source_deleted
, collision
, type_changed
;
2506 const char *rename_branch
= NULL
, *delete_branch
= NULL
;
2508 old_ent
= strmap_get_entry(&opt
->priv
->paths
, pair
->one
->path
);
2509 new_ent
= strmap_get_entry(&opt
->priv
->paths
, pair
->two
->path
);
2511 oldpath
= old_ent
->key
;
2512 oldinfo
= old_ent
->value
;
2514 newpath
= pair
->two
->path
;
2516 newpath
= new_ent
->key
;
2517 newinfo
= new_ent
->value
;
2521 * If pair->one->path isn't in opt->priv->paths, that means
2522 * that either directory rename detection removed that
2523 * path, or a parent directory of oldpath was resolved and
2524 * we don't even need the rename; in either case, we can
2525 * skip it. If oldinfo->merged.clean, then the other side
2526 * of history had no changes to oldpath and we don't need
2527 * the rename and can skip it.
2529 if (!oldinfo
|| oldinfo
->merged
.clean
)
2533 * diff_filepairs have copies of pathnames, thus we have to
2534 * use standard 'strcmp()' (negated) instead of '=='.
2536 if (i
+ 1 < renames
->nr
&&
2537 !strcmp(oldpath
, renames
->queue
[i
+1]->one
->path
)) {
2538 /* Handle rename/rename(1to2) or rename/rename(1to1) */
2539 const char *pathnames
[3];
2540 struct version_info merged
;
2541 struct conflict_info
*base
, *side1
, *side2
;
2542 unsigned was_binary_blob
= 0;
2544 pathnames
[0] = oldpath
;
2545 pathnames
[1] = newpath
;
2546 pathnames
[2] = renames
->queue
[i
+1]->two
->path
;
2548 base
= strmap_get(&opt
->priv
->paths
, pathnames
[0]);
2549 side1
= strmap_get(&opt
->priv
->paths
, pathnames
[1]);
2550 side2
= strmap_get(&opt
->priv
->paths
, pathnames
[2]);
2556 if (!strcmp(pathnames
[1], pathnames
[2])) {
2557 struct rename_info
*ri
= &opt
->priv
->renames
;
2560 /* Both sides renamed the same way */
2561 assert(side1
== side2
);
2562 memcpy(&side1
->stages
[0], &base
->stages
[0],
2564 side1
->filemask
|= (1 << MERGE_BASE
);
2565 /* Mark base as resolved by removal */
2566 base
->merged
.is_null
= 1;
2567 base
->merged
.clean
= 1;
2570 * Disable remembering renames optimization;
2571 * rename/rename(1to1) is incredibly rare, and
2572 * just disabling the optimization is easier
2573 * than purging cached_pairs,
2574 * cached_target_names, and dir_rename_counts.
2576 for (j
= 0; j
< 3; j
++)
2577 ri
->merge_trees
[j
] = NULL
;
2579 /* We handled both renames, i.e. i+1 handled */
2581 /* Move to next rename */
2585 /* This is a rename/rename(1to2) */
2586 clean_merge
= handle_content_merge(opt
,
2592 1 + 2 * opt
->priv
->call_depth
,
2595 merged
.mode
== side1
->stages
[1].mode
&&
2596 oideq(&merged
.oid
, &side1
->stages
[1].oid
))
2597 was_binary_blob
= 1;
2598 memcpy(&side1
->stages
[1], &merged
, sizeof(merged
));
2599 if (was_binary_blob
) {
2601 * Getting here means we were attempting to
2602 * merge a binary blob.
2604 * Since we can't merge binaries,
2605 * handle_content_merge() just takes one
2606 * side. But we don't want to copy the
2607 * contents of one side to both paths. We
2608 * used the contents of side1 above for
2609 * side1->stages, let's use the contents of
2610 * side2 for side2->stages below.
2612 oidcpy(&merged
.oid
, &side2
->stages
[2].oid
);
2613 merged
.mode
= side2
->stages
[2].mode
;
2615 memcpy(&side2
->stages
[2], &merged
, sizeof(merged
));
2617 side1
->path_conflict
= 1;
2618 side2
->path_conflict
= 1;
2620 * TODO: For renames we normally remove the path at the
2621 * old name. It would thus seem consistent to do the
2622 * same for rename/rename(1to2) cases, but we haven't
2623 * done so traditionally and a number of the regression
2624 * tests now encode an expectation that the file is
2625 * left there at stage 1. If we ever decide to change
2626 * this, add the following two lines here:
2627 * base->merged.is_null = 1;
2628 * base->merged.clean = 1;
2629 * and remove the setting of base->path_conflict to 1.
2631 base
->path_conflict
= 1;
2632 path_msg(opt
, oldpath
, 0,
2633 _("CONFLICT (rename/rename): %s renamed to "
2634 "%s in %s and to %s in %s."),
2636 pathnames
[1], opt
->branch1
,
2637 pathnames
[2], opt
->branch2
);
2639 i
++; /* We handled both renames, i.e. i+1 handled */
2645 target_index
= pair
->score
; /* from collect_renames() */
2646 assert(target_index
== 1 || target_index
== 2);
2647 other_source_index
= 3 - target_index
;
2648 old_sidemask
= (1 << other_source_index
); /* 2 or 4 */
2649 source_deleted
= (oldinfo
->filemask
== 1);
2650 collision
= ((newinfo
->filemask
& old_sidemask
) != 0);
2651 type_changed
= !source_deleted
&&
2652 (S_ISREG(oldinfo
->stages
[other_source_index
].mode
) !=
2653 S_ISREG(newinfo
->stages
[target_index
].mode
));
2654 if (type_changed
&& collision
) {
2656 * special handling so later blocks can handle this...
2658 * if type_changed && collision are both true, then this
2659 * was really a double rename, but one side wasn't
2660 * detected due to lack of break detection. I.e.
2662 * orig: has normal file 'foo'
2663 * side1: renames 'foo' to 'bar', adds 'foo' symlink
2664 * side2: renames 'foo' to 'bar'
2665 * In this case, the foo->bar rename on side1 won't be
2666 * detected because the new symlink named 'foo' is
2667 * there and we don't do break detection. But we detect
2668 * this here because we don't want to merge the content
2669 * of the foo symlink with the foo->bar file, so we
2670 * have some logic to handle this special case. The
2671 * easiest way to do that is make 'bar' on side1 not
2672 * be considered a colliding file but the other part
2673 * of a normal rename. If the file is very different,
2674 * well we're going to get content merge conflicts
2675 * anyway so it doesn't hurt. And if the colliding
2676 * file also has a different type, that'll be handled
2677 * by the content merge logic in process_entry() too.
2679 * See also t6430, 'rename vs. rename/symlink'
2683 if (source_deleted
) {
2684 if (target_index
== 1) {
2685 rename_branch
= opt
->branch1
;
2686 delete_branch
= opt
->branch2
;
2688 rename_branch
= opt
->branch2
;
2689 delete_branch
= opt
->branch1
;
2693 assert(source_deleted
|| oldinfo
->filemask
& old_sidemask
);
2695 /* Need to check for special types of rename conflicts... */
2696 if (collision
&& !source_deleted
) {
2697 /* collision: rename/add or rename/rename(2to1) */
2698 const char *pathnames
[3];
2699 struct version_info merged
;
2701 struct conflict_info
*base
, *side1
, *side2
;
2704 pathnames
[0] = oldpath
;
2705 pathnames
[other_source_index
] = oldpath
;
2706 pathnames
[target_index
] = newpath
;
2708 base
= strmap_get(&opt
->priv
->paths
, pathnames
[0]);
2709 side1
= strmap_get(&opt
->priv
->paths
, pathnames
[1]);
2710 side2
= strmap_get(&opt
->priv
->paths
, pathnames
[2]);
2716 clean
= handle_content_merge(opt
, pair
->one
->path
,
2721 1 + 2 * opt
->priv
->call_depth
,
2724 memcpy(&newinfo
->stages
[target_index
], &merged
,
2727 path_msg(opt
, newpath
, 0,
2728 _("CONFLICT (rename involved in "
2729 "collision): rename of %s -> %s has "
2730 "content conflicts AND collides "
2731 "with another path; this may result "
2732 "in nested conflict markers."),
2735 } else if (collision
&& source_deleted
) {
2737 * rename/add/delete or rename/rename(2to1)/delete:
2738 * since oldpath was deleted on the side that didn't
2739 * do the rename, there's not much of a content merge
2740 * we can do for the rename. oldinfo->merged.is_null
2741 * was already set, so we just leave things as-is so
2742 * they look like an add/add conflict.
2745 newinfo
->path_conflict
= 1;
2746 path_msg(opt
, newpath
, 0,
2747 _("CONFLICT (rename/delete): %s renamed "
2748 "to %s in %s, but deleted in %s."),
2749 oldpath
, newpath
, rename_branch
, delete_branch
);
2752 * a few different cases...start by copying the
2753 * existing stage(s) from oldinfo over the newinfo
2754 * and update the pathname(s).
2756 memcpy(&newinfo
->stages
[0], &oldinfo
->stages
[0],
2757 sizeof(newinfo
->stages
[0]));
2758 newinfo
->filemask
|= (1 << MERGE_BASE
);
2759 newinfo
->pathnames
[0] = oldpath
;
2761 /* rename vs. typechange */
2762 /* Mark the original as resolved by removal */
2763 memcpy(&oldinfo
->stages
[0].oid
, null_oid(),
2764 sizeof(oldinfo
->stages
[0].oid
));
2765 oldinfo
->stages
[0].mode
= 0;
2766 oldinfo
->filemask
&= 0x06;
2767 } else if (source_deleted
) {
2769 newinfo
->path_conflict
= 1;
2770 path_msg(opt
, newpath
, 0,
2771 _("CONFLICT (rename/delete): %s renamed"
2772 " to %s in %s, but deleted in %s."),
2774 rename_branch
, delete_branch
);
2777 memcpy(&newinfo
->stages
[other_source_index
],
2778 &oldinfo
->stages
[other_source_index
],
2779 sizeof(newinfo
->stages
[0]));
2780 newinfo
->filemask
|= (1 << other_source_index
);
2781 newinfo
->pathnames
[other_source_index
] = oldpath
;
2785 if (!type_changed
) {
2786 /* Mark the original as resolved by removal */
2787 oldinfo
->merged
.is_null
= 1;
2788 oldinfo
->merged
.clean
= 1;
2796 static inline int possible_side_renames(struct rename_info
*renames
,
2797 unsigned side_index
)
2799 return renames
->pairs
[side_index
].nr
> 0 &&
2800 !strintmap_empty(&renames
->relevant_sources
[side_index
]);
2803 static inline int possible_renames(struct rename_info
*renames
)
2805 return possible_side_renames(renames
, 1) ||
2806 possible_side_renames(renames
, 2) ||
2807 !strmap_empty(&renames
->cached_pairs
[1]) ||
2808 !strmap_empty(&renames
->cached_pairs
[2]);
2811 static void resolve_diffpair_statuses(struct diff_queue_struct
*q
)
2814 * A simplified version of diff_resolve_rename_copy(); would probably
2815 * just use that function but it's static...
2818 struct diff_filepair
*p
;
2820 for (i
= 0; i
< q
->nr
; ++i
) {
2822 p
->status
= 0; /* undecided */
2823 if (!DIFF_FILE_VALID(p
->one
))
2824 p
->status
= DIFF_STATUS_ADDED
;
2825 else if (!DIFF_FILE_VALID(p
->two
))
2826 p
->status
= DIFF_STATUS_DELETED
;
2827 else if (DIFF_PAIR_RENAME(p
))
2828 p
->status
= DIFF_STATUS_RENAMED
;
2832 static void prune_cached_from_relevant(struct rename_info
*renames
,
2835 /* Reason for this function described in add_pair() */
2836 struct hashmap_iter iter
;
2837 struct strmap_entry
*entry
;
2839 /* Remove from relevant_sources all entries in cached_pairs[side] */
2840 strmap_for_each_entry(&renames
->cached_pairs
[side
], &iter
, entry
) {
2841 strintmap_remove(&renames
->relevant_sources
[side
],
2844 /* Remove from relevant_sources all entries in cached_irrelevant[side] */
2845 strset_for_each_entry(&renames
->cached_irrelevant
[side
], &iter
, entry
) {
2846 strintmap_remove(&renames
->relevant_sources
[side
],
2851 static void use_cached_pairs(struct merge_options
*opt
,
2852 struct strmap
*cached_pairs
,
2853 struct diff_queue_struct
*pairs
)
2855 struct hashmap_iter iter
;
2856 struct strmap_entry
*entry
;
2859 * Add to side_pairs all entries from renames->cached_pairs[side_index].
2860 * (Info in cached_irrelevant[side_index] is not relevant here.)
2862 strmap_for_each_entry(cached_pairs
, &iter
, entry
) {
2863 struct diff_filespec
*one
, *two
;
2864 const char *old_name
= entry
->key
;
2865 const char *new_name
= entry
->value
;
2867 new_name
= old_name
;
2870 * cached_pairs has *copies* of old_name and new_name,
2871 * because it has to persist across merges. Since
2872 * pool_alloc_filespec() will just re-use the existing
2873 * filenames, which will also get re-used by
2874 * opt->priv->paths if they become renames, and then
2875 * get freed at the end of the merge, that would leave
2876 * the copy in cached_pairs dangling. Avoid this by
2877 * making a copy here.
2879 old_name
= mem_pool_strdup(&opt
->priv
->pool
, old_name
);
2880 new_name
= mem_pool_strdup(&opt
->priv
->pool
, new_name
);
2882 /* We don't care about oid/mode, only filenames and status */
2883 one
= pool_alloc_filespec(&opt
->priv
->pool
, old_name
);
2884 two
= pool_alloc_filespec(&opt
->priv
->pool
, new_name
);
2885 pool_diff_queue(&opt
->priv
->pool
, pairs
, one
, two
);
2886 pairs
->queue
[pairs
->nr
-1]->status
= entry
->value
? 'R' : 'D';
2890 static void cache_new_pair(struct rename_info
*renames
,
2897 new_path
= xstrdup(new_path
);
2898 old_value
= strmap_put(&renames
->cached_pairs
[side
],
2899 old_path
, new_path
);
2900 strset_add(&renames
->cached_target_names
[side
], new_path
);
2907 static void possibly_cache_new_pair(struct rename_info
*renames
,
2908 struct diff_filepair
*p
,
2912 int dir_renamed_side
= 0;
2916 * Directory renames happen on the other side of history from
2917 * the side that adds new files to the old directory.
2919 dir_renamed_side
= 3 - side
;
2921 int val
= strintmap_get(&renames
->relevant_sources
[side
],
2923 if (val
== RELEVANT_NO_MORE
) {
2924 assert(p
->status
== 'D');
2925 strset_add(&renames
->cached_irrelevant
[side
],
2932 if (p
->status
== 'D') {
2934 * If we already had this delete, we'll just set it's value
2935 * to NULL again, so no harm.
2937 strmap_put(&renames
->cached_pairs
[side
], p
->one
->path
, NULL
);
2938 } else if (p
->status
== 'R') {
2940 new_path
= p
->two
->path
;
2942 cache_new_pair(renames
, dir_renamed_side
,
2943 p
->two
->path
, new_path
, 0);
2944 cache_new_pair(renames
, side
, p
->one
->path
, new_path
, 1);
2945 } else if (p
->status
== 'A' && new_path
) {
2946 cache_new_pair(renames
, dir_renamed_side
,
2947 p
->two
->path
, new_path
, 0);
2951 static int compare_pairs(const void *a_
, const void *b_
)
2953 const struct diff_filepair
*a
= *((const struct diff_filepair
**)a_
);
2954 const struct diff_filepair
*b
= *((const struct diff_filepair
**)b_
);
2956 return strcmp(a
->one
->path
, b
->one
->path
);
2959 /* Call diffcore_rename() to update deleted/added pairs into rename pairs */
2960 static int detect_regular_renames(struct merge_options
*opt
,
2961 unsigned side_index
)
2963 struct diff_options diff_opts
;
2964 struct rename_info
*renames
= &opt
->priv
->renames
;
2966 prune_cached_from_relevant(renames
, side_index
);
2967 if (!possible_side_renames(renames
, side_index
)) {
2969 * No rename detection needed for this side, but we still need
2970 * to make sure 'adds' are marked correctly in case the other
2971 * side had directory renames.
2973 resolve_diffpair_statuses(&renames
->pairs
[side_index
]);
2977 partial_clear_dir_rename_count(&renames
->dir_rename_count
[side_index
]);
2978 repo_diff_setup(opt
->repo
, &diff_opts
);
2979 diff_opts
.flags
.recursive
= 1;
2980 diff_opts
.flags
.rename_empty
= 0;
2981 diff_opts
.detect_rename
= DIFF_DETECT_RENAME
;
2982 diff_opts
.rename_limit
= opt
->rename_limit
;
2983 if (opt
->rename_limit
<= 0)
2984 diff_opts
.rename_limit
= 7000;
2985 diff_opts
.rename_score
= opt
->rename_score
;
2986 diff_opts
.show_rename_progress
= opt
->show_rename_progress
;
2987 diff_opts
.output_format
= DIFF_FORMAT_NO_OUTPUT
;
2988 diff_setup_done(&diff_opts
);
2990 diff_queued_diff
= renames
->pairs
[side_index
];
2991 trace2_region_enter("diff", "diffcore_rename", opt
->repo
);
2992 diffcore_rename_extended(&diff_opts
,
2994 &renames
->relevant_sources
[side_index
],
2995 &renames
->dirs_removed
[side_index
],
2996 &renames
->dir_rename_count
[side_index
],
2997 &renames
->cached_pairs
[side_index
]);
2998 trace2_region_leave("diff", "diffcore_rename", opt
->repo
);
2999 resolve_diffpair_statuses(&diff_queued_diff
);
3001 if (diff_opts
.needed_rename_limit
> 0)
3002 renames
->redo_after_renames
= 0;
3003 if (diff_opts
.needed_rename_limit
> renames
->needed_limit
)
3004 renames
->needed_limit
= diff_opts
.needed_rename_limit
;
3006 renames
->pairs
[side_index
] = diff_queued_diff
;
3008 diff_opts
.output_format
= DIFF_FORMAT_NO_OUTPUT
;
3009 diff_queued_diff
.nr
= 0;
3010 diff_queued_diff
.queue
= NULL
;
3011 diff_flush(&diff_opts
);
3017 * Get information of all renames which occurred in 'side_pairs', making use
3018 * of any implicit directory renames in side_dir_renames (also making use of
3019 * implicit directory renames rename_exclusions as needed by
3020 * check_for_directory_rename()). Add all (updated) renames into result.
3022 static int collect_renames(struct merge_options
*opt
,
3023 struct diff_queue_struct
*result
,
3024 unsigned side_index
,
3025 struct strmap
*dir_renames_for_side
,
3026 struct strmap
*rename_exclusions
)
3029 struct strmap collisions
;
3030 struct diff_queue_struct
*side_pairs
;
3031 struct hashmap_iter iter
;
3032 struct strmap_entry
*entry
;
3033 struct rename_info
*renames
= &opt
->priv
->renames
;
3035 side_pairs
= &renames
->pairs
[side_index
];
3036 compute_collisions(&collisions
, dir_renames_for_side
, side_pairs
);
3038 for (i
= 0; i
< side_pairs
->nr
; ++i
) {
3039 struct diff_filepair
*p
= side_pairs
->queue
[i
];
3040 char *new_path
; /* non-NULL only with directory renames */
3042 if (p
->status
!= 'A' && p
->status
!= 'R') {
3043 possibly_cache_new_pair(renames
, p
, side_index
, NULL
);
3044 pool_diff_free_filepair(&opt
->priv
->pool
, p
);
3048 new_path
= check_for_directory_rename(opt
, p
->two
->path
,
3050 dir_renames_for_side
,
3055 possibly_cache_new_pair(renames
, p
, side_index
, new_path
);
3056 if (p
->status
!= 'R' && !new_path
) {
3057 pool_diff_free_filepair(&opt
->priv
->pool
, p
);
3062 apply_directory_rename_modifications(opt
, p
, new_path
);
3065 * p->score comes back from diffcore_rename_extended() with
3066 * the similarity of the renamed file. The similarity is
3067 * was used to determine that the two files were related
3068 * and are a rename, which we have already used, but beyond
3069 * that we have no use for the similarity. So p->score is
3070 * now irrelevant. However, process_renames() will need to
3071 * know which side of the merge this rename was associated
3072 * with, so overwrite p->score with that value.
3074 p
->score
= side_index
;
3075 result
->queue
[result
->nr
++] = p
;
3078 /* Free each value in the collisions map */
3079 strmap_for_each_entry(&collisions
, &iter
, entry
) {
3080 struct collision_info
*info
= entry
->value
;
3081 string_list_clear(&info
->source_files
, 0);
3084 * In compute_collisions(), we set collisions.strdup_strings to 0
3085 * so that we wouldn't have to make another copy of the new_path
3086 * allocated by apply_dir_rename(). But now that we've used them
3087 * and have no other references to these strings, it is time to
3090 free_strmap_strings(&collisions
);
3091 strmap_clear(&collisions
, 1);
3095 static int detect_and_process_renames(struct merge_options
*opt
,
3096 struct tree
*merge_base
,
3100 struct diff_queue_struct combined
= { 0 };
3101 struct rename_info
*renames
= &opt
->priv
->renames
;
3102 int need_dir_renames
, s
, i
, clean
= 1;
3103 unsigned detection_run
= 0;
3105 if (!possible_renames(renames
))
3108 trace2_region_enter("merge", "regular renames", opt
->repo
);
3109 detection_run
|= detect_regular_renames(opt
, MERGE_SIDE1
);
3110 detection_run
|= detect_regular_renames(opt
, MERGE_SIDE2
);
3111 if (renames
->needed_limit
) {
3112 renames
->cached_pairs_valid_side
= 0;
3113 renames
->redo_after_renames
= 0;
3115 if (renames
->redo_after_renames
&& detection_run
) {
3117 struct diff_filepair
*p
;
3119 /* Cache the renames, we found */
3120 for (side
= MERGE_SIDE1
; side
<= MERGE_SIDE2
; side
++) {
3121 for (i
= 0; i
< renames
->pairs
[side
].nr
; ++i
) {
3122 p
= renames
->pairs
[side
].queue
[i
];
3123 possibly_cache_new_pair(renames
, p
, side
, NULL
);
3127 /* Restart the merge with the cached renames */
3128 renames
->redo_after_renames
= 2;
3129 trace2_region_leave("merge", "regular renames", opt
->repo
);
3132 use_cached_pairs(opt
, &renames
->cached_pairs
[1], &renames
->pairs
[1]);
3133 use_cached_pairs(opt
, &renames
->cached_pairs
[2], &renames
->pairs
[2]);
3134 trace2_region_leave("merge", "regular renames", opt
->repo
);
3136 trace2_region_enter("merge", "directory renames", opt
->repo
);
3138 !opt
->priv
->call_depth
&&
3139 (opt
->detect_directory_renames
== MERGE_DIRECTORY_RENAMES_TRUE
||
3140 opt
->detect_directory_renames
== MERGE_DIRECTORY_RENAMES_CONFLICT
);
3142 if (need_dir_renames
) {
3143 get_provisional_directory_renames(opt
, MERGE_SIDE1
, &clean
);
3144 get_provisional_directory_renames(opt
, MERGE_SIDE2
, &clean
);
3145 handle_directory_level_conflicts(opt
);
3148 ALLOC_GROW(combined
.queue
,
3149 renames
->pairs
[1].nr
+ renames
->pairs
[2].nr
,
3151 clean
&= collect_renames(opt
, &combined
, MERGE_SIDE1
,
3152 &renames
->dir_renames
[2],
3153 &renames
->dir_renames
[1]);
3154 clean
&= collect_renames(opt
, &combined
, MERGE_SIDE2
,
3155 &renames
->dir_renames
[1],
3156 &renames
->dir_renames
[2]);
3157 STABLE_QSORT(combined
.queue
, combined
.nr
, compare_pairs
);
3158 trace2_region_leave("merge", "directory renames", opt
->repo
);
3160 trace2_region_enter("merge", "process renames", opt
->repo
);
3161 clean
&= process_renames(opt
, &combined
);
3162 trace2_region_leave("merge", "process renames", opt
->repo
);
3164 goto simple_cleanup
; /* collect_renames() handles some of cleanup */
3168 * Free now unneeded filepairs, which would have been handled
3169 * in collect_renames() normally but we skipped that code.
3171 for (s
= MERGE_SIDE1
; s
<= MERGE_SIDE2
; s
++) {
3172 struct diff_queue_struct
*side_pairs
;
3175 side_pairs
= &renames
->pairs
[s
];
3176 for (i
= 0; i
< side_pairs
->nr
; ++i
) {
3177 struct diff_filepair
*p
= side_pairs
->queue
[i
];
3178 pool_diff_free_filepair(&opt
->priv
->pool
, p
);
3183 /* Free memory for renames->pairs[] and combined */
3184 for (s
= MERGE_SIDE1
; s
<= MERGE_SIDE2
; s
++) {
3185 free(renames
->pairs
[s
].queue
);
3186 DIFF_QUEUE_CLEAR(&renames
->pairs
[s
]);
3188 for (i
= 0; i
< combined
.nr
; i
++)
3189 pool_diff_free_filepair(&opt
->priv
->pool
, combined
.queue
[i
]);
3190 free(combined
.queue
);
3195 /*** Function Grouping: functions related to process_entries() ***/
3197 static int sort_dirs_next_to_their_children(const char *one
, const char *two
)
3199 unsigned char c1
, c2
;
3202 * Here we only care that entries for directories appear adjacent
3203 * to and before files underneath the directory. We can achieve
3204 * that by pretending to add a trailing slash to every file and
3205 * then sorting. In other words, we do not want the natural
3210 * Instead, we want "foo" to sort as though it were "foo/", so that
3215 * To achieve this, we basically implement our own strcmp, except that
3216 * if we get to the end of either string instead of comparing NUL to
3217 * another character, we compare '/' to it.
3219 * If this unusual "sort as though '/' were appended" perplexes
3220 * you, perhaps it will help to note that this is not the final
3221 * sort. write_tree() will sort again without the trailing slash
3222 * magic, but just on paths immediately under a given tree.
3224 * The reason to not use df_name_compare directly was that it was
3225 * just too expensive (we don't have the string lengths handy), so
3226 * it was reimplemented.
3230 * NOTE: This function will never be called with two equal strings,
3231 * because it is used to sort the keys of a strmap, and strmaps have
3232 * unique keys by construction. That simplifies our c1==c2 handling
3236 while (*one
&& (*one
== *two
)) {
3241 c1
= *one
? *one
: '/';
3242 c2
= *two
? *two
: '/';
3245 /* Getting here means one is a leading directory of the other */
3246 return (*one
) ? 1 : -1;
3251 static int read_oid_strbuf(struct merge_options
*opt
,
3252 const struct object_id
*oid
,
3256 enum object_type type
;
3258 buf
= read_object_file(oid
, &type
, &size
);
3260 return err(opt
, _("cannot read object %s"), oid_to_hex(oid
));
3261 if (type
!= OBJ_BLOB
) {
3263 return err(opt
, _("object %s is not a blob"), oid_to_hex(oid
));
3265 strbuf_attach(dst
, buf
, size
, size
+ 1);
3269 static int blob_unchanged(struct merge_options
*opt
,
3270 const struct version_info
*base
,
3271 const struct version_info
*side
,
3274 struct strbuf basebuf
= STRBUF_INIT
;
3275 struct strbuf sidebuf
= STRBUF_INIT
;
3276 int ret
= 0; /* assume changed for safety */
3277 struct index_state
*idx
= &opt
->priv
->attr_index
;
3279 if (!idx
->initialized
)
3280 initialize_attr_index(opt
);
3282 if (base
->mode
!= side
->mode
)
3284 if (oideq(&base
->oid
, &side
->oid
))
3287 if (read_oid_strbuf(opt
, &base
->oid
, &basebuf
) ||
3288 read_oid_strbuf(opt
, &side
->oid
, &sidebuf
))
3291 * Note: binary | is used so that both renormalizations are
3292 * performed. Comparison can be skipped if both files are
3293 * unchanged since their sha1s have already been compared.
3295 if (renormalize_buffer(idx
, path
, basebuf
.buf
, basebuf
.len
, &basebuf
) |
3296 renormalize_buffer(idx
, path
, sidebuf
.buf
, sidebuf
.len
, &sidebuf
))
3297 ret
= (basebuf
.len
== sidebuf
.len
&&
3298 !memcmp(basebuf
.buf
, sidebuf
.buf
, basebuf
.len
));
3301 strbuf_release(&basebuf
);
3302 strbuf_release(&sidebuf
);
3306 struct directory_versions
{
3308 * versions: list of (basename -> version_info)
3310 * The basenames are in reverse lexicographic order of full pathnames,
3311 * as processed in process_entries(). This puts all entries within
3312 * a directory together, and covers the directory itself after
3313 * everything within it, allowing us to write subtrees before needing
3314 * to record information for the tree itself.
3316 struct string_list versions
;
3319 * offsets: list of (full relative path directories -> integer offsets)
3321 * Since versions contains basenames from files in multiple different
3322 * directories, we need to know which entries in versions correspond
3323 * to which directories. Values of e.g.
3327 * Would mean that entries 0-1 of versions are files in the toplevel
3328 * directory, entries 2-4 are files under src/, and the remaining
3329 * entries starting at index 5 are files under src/moduleA/.
3331 struct string_list offsets
;
3334 * last_directory: directory that previously processed file found in
3336 * last_directory starts NULL, but records the directory in which the
3337 * previous file was found within. As soon as
3338 * directory(current_file) != last_directory
3339 * then we need to start updating accounting in versions & offsets.
3340 * Note that last_directory is always the last path in "offsets" (or
3341 * NULL if "offsets" is empty) so this exists just for quick access.
3343 const char *last_directory
;
3345 /* last_directory_len: cached computation of strlen(last_directory) */
3346 unsigned last_directory_len
;
3349 static int tree_entry_order(const void *a_
, const void *b_
)
3351 const struct string_list_item
*a
= a_
;
3352 const struct string_list_item
*b
= b_
;
3354 const struct merged_info
*ami
= a
->util
;
3355 const struct merged_info
*bmi
= b
->util
;
3356 return base_name_compare(a
->string
, strlen(a
->string
), ami
->result
.mode
,
3357 b
->string
, strlen(b
->string
), bmi
->result
.mode
);
3360 static void write_tree(struct object_id
*result_oid
,
3361 struct string_list
*versions
,
3362 unsigned int offset
,
3365 size_t maxlen
= 0, extra
;
3367 struct strbuf buf
= STRBUF_INIT
;
3370 assert(offset
<= versions
->nr
);
3371 nr
= versions
->nr
- offset
;
3373 /* No need for STABLE_QSORT -- filenames must be unique */
3374 QSORT(versions
->items
+ offset
, nr
, tree_entry_order
);
3376 /* Pre-allocate some space in buf */
3377 extra
= hash_size
+ 8; /* 8: 6 for mode, 1 for space, 1 for NUL char */
3378 for (i
= 0; i
< nr
; i
++) {
3379 maxlen
+= strlen(versions
->items
[offset
+i
].string
) + extra
;
3381 strbuf_grow(&buf
, maxlen
);
3383 /* Write each entry out to buf */
3384 for (i
= 0; i
< nr
; i
++) {
3385 struct merged_info
*mi
= versions
->items
[offset
+i
].util
;
3386 struct version_info
*ri
= &mi
->result
;
3387 strbuf_addf(&buf
, "%o %s%c",
3389 versions
->items
[offset
+i
].string
, '\0');
3390 strbuf_add(&buf
, ri
->oid
.hash
, hash_size
);
3393 /* Write this object file out, and record in result_oid */
3394 write_object_file(buf
.buf
, buf
.len
, tree_type
, result_oid
);
3395 strbuf_release(&buf
);
3398 static void record_entry_for_tree(struct directory_versions
*dir_metadata
,
3400 struct merged_info
*mi
)
3402 const char *basename
;
3405 /* nothing to record */
3408 basename
= path
+ mi
->basename_offset
;
3409 assert(strchr(basename
, '/') == NULL
);
3410 string_list_append(&dir_metadata
->versions
,
3411 basename
)->util
= &mi
->result
;
3414 static void write_completed_directory(struct merge_options
*opt
,
3415 const char *new_directory_name
,
3416 struct directory_versions
*info
)
3418 const char *prev_dir
;
3419 struct merged_info
*dir_info
= NULL
;
3420 unsigned int offset
;
3423 * Some explanation of info->versions and info->offsets...
3425 * process_entries() iterates over all relevant files AND
3426 * directories in reverse lexicographic order, and calls this
3427 * function. Thus, an example of the paths that process_entries()
3428 * could operate on (along with the directories for those paths
3433 * src/moduleB/umm.c src/moduleB
3434 * src/moduleB/stuff.h src/moduleB
3435 * src/moduleB/baz.c src/moduleB
3437 * src/moduleA/foo.c src/moduleA
3438 * src/moduleA/bar.c src/moduleA
3445 * always contains the unprocessed entries and their
3446 * version_info information. For example, after the first five
3447 * entries above, info->versions would be:
3449 * xtract.c <xtract.c's version_info>
3450 * token.txt <token.txt's version_info>
3451 * umm.c <src/moduleB/umm.c's version_info>
3452 * stuff.h <src/moduleB/stuff.h's version_info>
3453 * baz.c <src/moduleB/baz.c's version_info>
3455 * Once a subdirectory is completed we remove the entries in
3456 * that subdirectory from info->versions, writing it as a tree
3457 * (write_tree()). Thus, as soon as we get to src/moduleB,
3458 * info->versions would be updated to
3460 * xtract.c <xtract.c's version_info>
3461 * token.txt <token.txt's version_info>
3462 * moduleB <src/moduleB's version_info>
3466 * helps us track which entries in info->versions correspond to
3467 * which directories. When we are N directories deep (e.g. 4
3468 * for src/modA/submod/subdir/), we have up to N+1 unprocessed
3469 * directories (+1 because of toplevel dir). Corresponding to
3470 * the info->versions example above, after processing five entries
3471 * info->offsets will be:
3476 * which is used to know that xtract.c & token.txt are from the
3477 * toplevel dirctory, while umm.c & stuff.h & baz.c are from the
3478 * src/moduleB directory. Again, following the example above,
3479 * once we need to process src/moduleB, then info->offsets is
3485 * which says that moduleB (and only moduleB so far) is in the
3488 * One unique thing to note about info->offsets here is that
3489 * "src" was not added to info->offsets until there was a path
3490 * (a file OR directory) immediately below src/ that got
3493 * Since process_entry() just appends new entries to info->versions,
3494 * write_completed_directory() only needs to do work if the next path
3495 * is in a directory that is different than the last directory found
3500 * If we are working with the same directory as the last entry, there
3501 * is no work to do. (See comments above the directory_name member of
3502 * struct merged_info for why we can use pointer comparison instead of
3505 if (new_directory_name
== info
->last_directory
)
3509 * If we are just starting (last_directory is NULL), or last_directory
3510 * is a prefix of the current directory, then we can just update
3511 * info->offsets to record the offset where we started this directory
3512 * and update last_directory to have quick access to it.
3514 if (info
->last_directory
== NULL
||
3515 !strncmp(new_directory_name
, info
->last_directory
,
3516 info
->last_directory_len
)) {
3517 uintptr_t offset
= info
->versions
.nr
;
3519 info
->last_directory
= new_directory_name
;
3520 info
->last_directory_len
= strlen(info
->last_directory
);
3522 * Record the offset into info->versions where we will
3523 * start recording basenames of paths found within
3524 * new_directory_name.
3526 string_list_append(&info
->offsets
,
3527 info
->last_directory
)->util
= (void*)offset
;
3532 * The next entry that will be processed will be within
3533 * new_directory_name. Since at this point we know that
3534 * new_directory_name is within a different directory than
3535 * info->last_directory, we have all entries for info->last_directory
3536 * in info->versions and we need to create a tree object for them.
3538 dir_info
= strmap_get(&opt
->priv
->paths
, info
->last_directory
);
3540 offset
= (uintptr_t)info
->offsets
.items
[info
->offsets
.nr
-1].util
;
3541 if (offset
== info
->versions
.nr
) {
3543 * Actually, we don't need to create a tree object in this
3544 * case. Whenever all files within a directory disappear
3545 * during the merge (e.g. unmodified on one side and
3546 * deleted on the other, or files were renamed elsewhere),
3547 * then we get here and the directory itself needs to be
3548 * omitted from its parent tree as well.
3550 dir_info
->is_null
= 1;
3553 * Write out the tree to the git object directory, and also
3554 * record the mode and oid in dir_info->result.
3556 dir_info
->is_null
= 0;
3557 dir_info
->result
.mode
= S_IFDIR
;
3558 write_tree(&dir_info
->result
.oid
, &info
->versions
, offset
,
3559 opt
->repo
->hash_algo
->rawsz
);
3563 * We've now used several entries from info->versions and one entry
3564 * from info->offsets, so we get rid of those values.
3567 info
->versions
.nr
= offset
;
3570 * Now we've taken care of the completed directory, but we need to
3571 * prepare things since future entries will be in
3572 * new_directory_name. (In particular, process_entry() will be
3573 * appending new entries to info->versions.) So, we need to make
3574 * sure new_directory_name is the last entry in info->offsets.
3576 prev_dir
= info
->offsets
.nr
== 0 ? NULL
:
3577 info
->offsets
.items
[info
->offsets
.nr
-1].string
;
3578 if (new_directory_name
!= prev_dir
) {
3579 uintptr_t c
= info
->versions
.nr
;
3580 string_list_append(&info
->offsets
,
3581 new_directory_name
)->util
= (void*)c
;
3584 /* And, of course, we need to update last_directory to match. */
3585 info
->last_directory
= new_directory_name
;
3586 info
->last_directory_len
= strlen(info
->last_directory
);
3589 /* Per entry merge function */
3590 static void process_entry(struct merge_options
*opt
,
3592 struct conflict_info
*ci
,
3593 struct directory_versions
*dir_metadata
)
3595 int df_file_index
= 0;
3598 assert(ci
->filemask
>= 0 && ci
->filemask
<= 7);
3599 /* ci->match_mask == 7 was handled in collect_merge_info_callback() */
3600 assert(ci
->match_mask
== 0 || ci
->match_mask
== 3 ||
3601 ci
->match_mask
== 5 || ci
->match_mask
== 6);
3604 record_entry_for_tree(dir_metadata
, path
, &ci
->merged
);
3605 if (ci
->filemask
== 0)
3606 /* nothing else to handle */
3608 assert(ci
->df_conflict
);
3611 if (ci
->df_conflict
&& ci
->merged
.result
.mode
== 0) {
3615 * directory no longer in the way, but we do have a file we
3616 * need to place here so we need to clean away the "directory
3617 * merges to nothing" result.
3619 ci
->df_conflict
= 0;
3620 assert(ci
->filemask
!= 0);
3621 ci
->merged
.clean
= 0;
3622 ci
->merged
.is_null
= 0;
3623 /* and we want to zero out any directory-related entries */
3624 ci
->match_mask
= (ci
->match_mask
& ~ci
->dirmask
);
3626 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
3627 if (ci
->filemask
& (1 << i
))
3629 ci
->stages
[i
].mode
= 0;
3630 oidcpy(&ci
->stages
[i
].oid
, null_oid());
3632 } else if (ci
->df_conflict
&& ci
->merged
.result
.mode
!= 0) {
3634 * This started out as a D/F conflict, and the entries in
3635 * the competing directory were not removed by the merge as
3636 * evidenced by write_completed_directory() writing a value
3637 * to ci->merged.result.mode.
3639 struct conflict_info
*new_ci
;
3641 const char *old_path
= path
;
3644 assert(ci
->merged
.result
.mode
== S_IFDIR
);
3647 * If filemask is 1, we can just ignore the file as having
3648 * been deleted on both sides. We do not want to overwrite
3649 * ci->merged.result, since it stores the tree for all the
3652 if (ci
->filemask
== 1) {
3658 * This file still exists on at least one side, and we want
3659 * the directory to remain here, so we need to move this
3660 * path to some new location.
3662 new_ci
= mem_pool_calloc(&opt
->priv
->pool
, 1, sizeof(*new_ci
));
3664 /* We don't really want new_ci->merged.result copied, but it'll
3665 * be overwritten below so it doesn't matter. We also don't
3666 * want any directory mode/oid values copied, but we'll zero
3667 * those out immediately. We do want the rest of ci copied.
3669 memcpy(new_ci
, ci
, sizeof(*ci
));
3670 new_ci
->match_mask
= (new_ci
->match_mask
& ~new_ci
->dirmask
);
3671 new_ci
->dirmask
= 0;
3672 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
3673 if (new_ci
->filemask
& (1 << i
))
3675 /* zero out any entries related to directories */
3676 new_ci
->stages
[i
].mode
= 0;
3677 oidcpy(&new_ci
->stages
[i
].oid
, null_oid());
3681 * Find out which side this file came from; note that we
3682 * cannot just use ci->filemask, because renames could cause
3683 * the filemask to go back to 7. So we use dirmask, then
3684 * pick the opposite side's index.
3686 df_file_index
= (ci
->dirmask
& (1 << 1)) ? 2 : 1;
3687 branch
= (df_file_index
== 1) ? opt
->branch1
: opt
->branch2
;
3688 path
= unique_path(opt
, path
, branch
);
3689 strmap_put(&opt
->priv
->paths
, path
, new_ci
);
3691 path_msg(opt
, path
, 0,
3692 _("CONFLICT (file/directory): directory in the way "
3693 "of %s from %s; moving it to %s instead."),
3694 old_path
, branch
, path
);
3697 * Zero out the filemask for the old ci. At this point, ci
3698 * was just an entry for a directory, so we don't need to
3699 * do anything more with it.
3704 * Now note that we're working on the new entry (path was
3711 * NOTE: Below there is a long switch-like if-elseif-elseif... block
3712 * which the code goes through even for the df_conflict cases
3715 if (ci
->match_mask
) {
3716 ci
->merged
.clean
= !ci
->df_conflict
&& !ci
->path_conflict
;
3717 if (ci
->match_mask
== 6) {
3718 /* stages[1] == stages[2] */
3719 ci
->merged
.result
.mode
= ci
->stages
[1].mode
;
3720 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[1].oid
);
3722 /* determine the mask of the side that didn't match */
3723 unsigned int othermask
= 7 & ~ci
->match_mask
;
3724 int side
= (othermask
== 4) ? 2 : 1;
3726 ci
->merged
.result
.mode
= ci
->stages
[side
].mode
;
3727 ci
->merged
.is_null
= !ci
->merged
.result
.mode
;
3728 if (ci
->merged
.is_null
)
3729 ci
->merged
.clean
= 1;
3730 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[side
].oid
);
3732 assert(othermask
== 2 || othermask
== 4);
3733 assert(ci
->merged
.is_null
==
3734 (ci
->filemask
== ci
->match_mask
));
3736 } else if (ci
->filemask
>= 6 &&
3737 (S_IFMT
& ci
->stages
[1].mode
) !=
3738 (S_IFMT
& ci
->stages
[2].mode
)) {
3739 /* Two different items from (file/submodule/symlink) */
3740 if (opt
->priv
->call_depth
) {
3741 /* Just use the version from the merge base */
3742 ci
->merged
.clean
= 0;
3743 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[0].oid
);
3744 ci
->merged
.result
.mode
= ci
->stages
[0].mode
;
3745 ci
->merged
.is_null
= (ci
->merged
.result
.mode
== 0);
3747 /* Handle by renaming one or both to separate paths. */
3748 unsigned o_mode
= ci
->stages
[0].mode
;
3749 unsigned a_mode
= ci
->stages
[1].mode
;
3750 unsigned b_mode
= ci
->stages
[2].mode
;
3751 struct conflict_info
*new_ci
;
3752 const char *a_path
= NULL
, *b_path
= NULL
;
3753 int rename_a
= 0, rename_b
= 0;
3755 new_ci
= mem_pool_alloc(&opt
->priv
->pool
,
3758 if (S_ISREG(a_mode
))
3760 else if (S_ISREG(b_mode
))
3767 if (rename_a
&& rename_b
) {
3768 path_msg(opt
, path
, 0,
3769 _("CONFLICT (distinct types): %s had "
3770 "different types on each side; "
3771 "renamed both of them so each can "
3772 "be recorded somewhere."),
3775 path_msg(opt
, path
, 0,
3776 _("CONFLICT (distinct types): %s had "
3777 "different types on each side; "
3778 "renamed one of them so each can be "
3779 "recorded somewhere."),
3783 ci
->merged
.clean
= 0;
3784 memcpy(new_ci
, ci
, sizeof(*new_ci
));
3786 /* Put b into new_ci, removing a from stages */
3787 new_ci
->merged
.result
.mode
= ci
->stages
[2].mode
;
3788 oidcpy(&new_ci
->merged
.result
.oid
, &ci
->stages
[2].oid
);
3789 new_ci
->stages
[1].mode
= 0;
3790 oidcpy(&new_ci
->stages
[1].oid
, null_oid());
3791 new_ci
->filemask
= 5;
3792 if ((S_IFMT
& b_mode
) != (S_IFMT
& o_mode
)) {
3793 new_ci
->stages
[0].mode
= 0;
3794 oidcpy(&new_ci
->stages
[0].oid
, null_oid());
3795 new_ci
->filemask
= 4;
3798 /* Leave only a in ci, fixing stages. */
3799 ci
->merged
.result
.mode
= ci
->stages
[1].mode
;
3800 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[1].oid
);
3801 ci
->stages
[2].mode
= 0;
3802 oidcpy(&ci
->stages
[2].oid
, null_oid());
3804 if ((S_IFMT
& a_mode
) != (S_IFMT
& o_mode
)) {
3805 ci
->stages
[0].mode
= 0;
3806 oidcpy(&ci
->stages
[0].oid
, null_oid());
3810 /* Insert entries into opt->priv_paths */
3811 assert(rename_a
|| rename_b
);
3813 a_path
= unique_path(opt
, path
, opt
->branch1
);
3814 strmap_put(&opt
->priv
->paths
, a_path
, ci
);
3818 b_path
= unique_path(opt
, path
, opt
->branch2
);
3821 strmap_put(&opt
->priv
->paths
, b_path
, new_ci
);
3823 if (rename_a
&& rename_b
)
3824 strmap_remove(&opt
->priv
->paths
, path
, 0);
3827 * Do special handling for b_path since process_entry()
3828 * won't be called on it specially.
3830 strmap_put(&opt
->priv
->conflicted
, b_path
, new_ci
);
3831 record_entry_for_tree(dir_metadata
, b_path
,
3835 * Remaining code for processing this entry should
3836 * think in terms of processing a_path.
3841 } else if (ci
->filemask
>= 6) {
3842 /* Need a two-way or three-way content merge */
3843 struct version_info merged_file
;
3844 unsigned clean_merge
;
3845 struct version_info
*o
= &ci
->stages
[0];
3846 struct version_info
*a
= &ci
->stages
[1];
3847 struct version_info
*b
= &ci
->stages
[2];
3849 clean_merge
= handle_content_merge(opt
, path
, o
, a
, b
,
3851 opt
->priv
->call_depth
* 2,
3853 ci
->merged
.clean
= clean_merge
&&
3854 !ci
->df_conflict
&& !ci
->path_conflict
;
3855 ci
->merged
.result
.mode
= merged_file
.mode
;
3856 ci
->merged
.is_null
= (merged_file
.mode
== 0);
3857 oidcpy(&ci
->merged
.result
.oid
, &merged_file
.oid
);
3858 if (clean_merge
&& ci
->df_conflict
) {
3859 assert(df_file_index
== 1 || df_file_index
== 2);
3860 ci
->filemask
= 1 << df_file_index
;
3861 ci
->stages
[df_file_index
].mode
= merged_file
.mode
;
3862 oidcpy(&ci
->stages
[df_file_index
].oid
, &merged_file
.oid
);
3865 const char *reason
= _("content");
3866 if (ci
->filemask
== 6)
3867 reason
= _("add/add");
3868 if (S_ISGITLINK(merged_file
.mode
))
3869 reason
= _("submodule");
3870 path_msg(opt
, path
, 0,
3871 _("CONFLICT (%s): Merge conflict in %s"),
3874 } else if (ci
->filemask
== 3 || ci
->filemask
== 5) {
3876 const char *modify_branch
, *delete_branch
;
3877 int side
= (ci
->filemask
== 5) ? 2 : 1;
3878 int index
= opt
->priv
->call_depth
? 0 : side
;
3880 ci
->merged
.result
.mode
= ci
->stages
[index
].mode
;
3881 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[index
].oid
);
3882 ci
->merged
.clean
= 0;
3884 modify_branch
= (side
== 1) ? opt
->branch1
: opt
->branch2
;
3885 delete_branch
= (side
== 1) ? opt
->branch2
: opt
->branch1
;
3887 if (opt
->renormalize
&&
3888 blob_unchanged(opt
, &ci
->stages
[0], &ci
->stages
[side
],
3890 if (!ci
->path_conflict
) {
3892 * Blob unchanged after renormalization, so
3893 * there's no modify/delete conflict after all;
3894 * we can just remove the file.
3896 ci
->merged
.is_null
= 1;
3897 ci
->merged
.clean
= 1;
3899 * file goes away => even if there was a
3900 * directory/file conflict there isn't one now.
3902 ci
->df_conflict
= 0;
3904 /* rename/delete, so conflict remains */
3906 } else if (ci
->path_conflict
&&
3907 oideq(&ci
->stages
[0].oid
, &ci
->stages
[side
].oid
)) {
3909 * This came from a rename/delete; no action to take,
3910 * but avoid printing "modify/delete" conflict notice
3911 * since the contents were not modified.
3914 path_msg(opt
, path
, 0,
3915 _("CONFLICT (modify/delete): %s deleted in %s "
3916 "and modified in %s. Version %s of %s left "
3918 path
, delete_branch
, modify_branch
,
3919 modify_branch
, path
);
3921 } else if (ci
->filemask
== 2 || ci
->filemask
== 4) {
3922 /* Added on one side */
3923 int side
= (ci
->filemask
== 4) ? 2 : 1;
3924 ci
->merged
.result
.mode
= ci
->stages
[side
].mode
;
3925 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[side
].oid
);
3926 ci
->merged
.clean
= !ci
->df_conflict
&& !ci
->path_conflict
;
3927 } else if (ci
->filemask
== 1) {
3928 /* Deleted on both sides */
3929 ci
->merged
.is_null
= 1;
3930 ci
->merged
.result
.mode
= 0;
3931 oidcpy(&ci
->merged
.result
.oid
, null_oid());
3932 assert(!ci
->df_conflict
);
3933 ci
->merged
.clean
= !ci
->path_conflict
;
3937 * If still conflicted, record it separately. This allows us to later
3938 * iterate over just conflicted entries when updating the index instead
3939 * of iterating over all entries.
3941 if (!ci
->merged
.clean
)
3942 strmap_put(&opt
->priv
->conflicted
, path
, ci
);
3944 /* Record metadata for ci->merged in dir_metadata */
3945 record_entry_for_tree(dir_metadata
, path
, &ci
->merged
);
3948 static void prefetch_for_content_merges(struct merge_options
*opt
,
3949 struct string_list
*plist
)
3951 struct string_list_item
*e
;
3952 struct oid_array to_fetch
= OID_ARRAY_INIT
;
3954 if (opt
->repo
!= the_repository
|| !has_promisor_remote())
3957 for (e
= &plist
->items
[plist
->nr
-1]; e
>= plist
->items
; --e
) {
3958 /* char *path = e->string; */
3959 struct conflict_info
*ci
= e
->util
;
3962 /* Ignore clean entries */
3963 if (ci
->merged
.clean
)
3966 /* Ignore entries that don't need a content merge */
3967 if (ci
->match_mask
|| ci
->filemask
< 6 ||
3968 !S_ISREG(ci
->stages
[1].mode
) ||
3969 !S_ISREG(ci
->stages
[2].mode
) ||
3970 oideq(&ci
->stages
[1].oid
, &ci
->stages
[2].oid
))
3973 /* Also don't need content merge if base matches either side */
3974 if (ci
->filemask
== 7 &&
3975 S_ISREG(ci
->stages
[0].mode
) &&
3976 (oideq(&ci
->stages
[0].oid
, &ci
->stages
[1].oid
) ||
3977 oideq(&ci
->stages
[0].oid
, &ci
->stages
[2].oid
)))
3980 for (i
= 0; i
< 3; i
++) {
3981 unsigned side_mask
= (1 << i
);
3982 struct version_info
*vi
= &ci
->stages
[i
];
3984 if ((ci
->filemask
& side_mask
) &&
3985 S_ISREG(vi
->mode
) &&
3986 oid_object_info_extended(opt
->repo
, &vi
->oid
, NULL
,
3987 OBJECT_INFO_FOR_PREFETCH
))
3988 oid_array_append(&to_fetch
, &vi
->oid
);
3992 promisor_remote_get_direct(opt
->repo
, to_fetch
.oid
, to_fetch
.nr
);
3993 oid_array_clear(&to_fetch
);
3996 static void process_entries(struct merge_options
*opt
,
3997 struct object_id
*result_oid
)
3999 struct hashmap_iter iter
;
4000 struct strmap_entry
*e
;
4001 struct string_list plist
= STRING_LIST_INIT_NODUP
;
4002 struct string_list_item
*entry
;
4003 struct directory_versions dir_metadata
= { STRING_LIST_INIT_NODUP
,
4004 STRING_LIST_INIT_NODUP
,
4007 trace2_region_enter("merge", "process_entries setup", opt
->repo
);
4008 if (strmap_empty(&opt
->priv
->paths
)) {
4009 oidcpy(result_oid
, opt
->repo
->hash_algo
->empty_tree
);
4013 /* Hack to pre-allocate plist to the desired size */
4014 trace2_region_enter("merge", "plist grow", opt
->repo
);
4015 ALLOC_GROW(plist
.items
, strmap_get_size(&opt
->priv
->paths
), plist
.alloc
);
4016 trace2_region_leave("merge", "plist grow", opt
->repo
);
4018 /* Put every entry from paths into plist, then sort */
4019 trace2_region_enter("merge", "plist copy", opt
->repo
);
4020 strmap_for_each_entry(&opt
->priv
->paths
, &iter
, e
) {
4021 string_list_append(&plist
, e
->key
)->util
= e
->value
;
4023 trace2_region_leave("merge", "plist copy", opt
->repo
);
4025 trace2_region_enter("merge", "plist special sort", opt
->repo
);
4026 plist
.cmp
= sort_dirs_next_to_their_children
;
4027 string_list_sort(&plist
);
4028 trace2_region_leave("merge", "plist special sort", opt
->repo
);
4030 trace2_region_leave("merge", "process_entries setup", opt
->repo
);
4033 * Iterate over the items in reverse order, so we can handle paths
4034 * below a directory before needing to handle the directory itself.
4036 * This allows us to write subtrees before we need to write trees,
4037 * and it also enables sane handling of directory/file conflicts
4038 * (because it allows us to know whether the directory is still in
4039 * the way when it is time to process the file at the same path).
4041 trace2_region_enter("merge", "processing", opt
->repo
);
4042 prefetch_for_content_merges(opt
, &plist
);
4043 for (entry
= &plist
.items
[plist
.nr
-1]; entry
>= plist
.items
; --entry
) {
4044 char *path
= entry
->string
;
4046 * NOTE: mi may actually be a pointer to a conflict_info, but
4047 * we have to check mi->clean first to see if it's safe to
4048 * reassign to such a pointer type.
4050 struct merged_info
*mi
= entry
->util
;
4052 write_completed_directory(opt
, mi
->directory_name
,
4055 record_entry_for_tree(&dir_metadata
, path
, mi
);
4057 struct conflict_info
*ci
= (struct conflict_info
*)mi
;
4058 process_entry(opt
, path
, ci
, &dir_metadata
);
4061 trace2_region_leave("merge", "processing", opt
->repo
);
4063 trace2_region_enter("merge", "process_entries cleanup", opt
->repo
);
4064 if (dir_metadata
.offsets
.nr
!= 1 ||
4065 (uintptr_t)dir_metadata
.offsets
.items
[0].util
!= 0) {
4066 printf("dir_metadata.offsets.nr = %d (should be 1)\n",
4067 dir_metadata
.offsets
.nr
);
4068 printf("dir_metadata.offsets.items[0].util = %u (should be 0)\n",
4069 (unsigned)(uintptr_t)dir_metadata
.offsets
.items
[0].util
);
4071 BUG("dir_metadata accounting completely off; shouldn't happen");
4073 write_tree(result_oid
, &dir_metadata
.versions
, 0,
4074 opt
->repo
->hash_algo
->rawsz
);
4075 string_list_clear(&plist
, 0);
4076 string_list_clear(&dir_metadata
.versions
, 0);
4077 string_list_clear(&dir_metadata
.offsets
, 0);
4078 trace2_region_leave("merge", "process_entries cleanup", opt
->repo
);
4081 /*** Function Grouping: functions related to merge_switch_to_result() ***/
4083 static int checkout(struct merge_options
*opt
,
4087 /* Switch the index/working copy from old to new */
4089 struct tree_desc trees
[2];
4090 struct unpack_trees_options unpack_opts
;
4092 memset(&unpack_opts
, 0, sizeof(unpack_opts
));
4093 unpack_opts
.head_idx
= -1;
4094 unpack_opts
.src_index
= opt
->repo
->index
;
4095 unpack_opts
.dst_index
= opt
->repo
->index
;
4097 setup_unpack_trees_porcelain(&unpack_opts
, "merge");
4100 * NOTE: if this were just "git checkout" code, we would probably
4101 * read or refresh the cache and check for a conflicted index, but
4102 * builtin/merge.c or sequencer.c really needs to read the index
4103 * and check for conflicted entries before starting merging for a
4104 * good user experience (no sense waiting for merges/rebases before
4105 * erroring out), so there's no reason to duplicate that work here.
4108 /* 2-way merge to the new branch */
4109 unpack_opts
.update
= 1;
4110 unpack_opts
.merge
= 1;
4111 unpack_opts
.quiet
= 0; /* FIXME: sequencer might want quiet? */
4112 unpack_opts
.verbose_update
= (opt
->verbosity
> 2);
4113 unpack_opts
.fn
= twoway_merge
;
4114 unpack_opts
.preserve_ignored
= 0; /* FIXME: !opts->overwrite_ignore */
4116 init_tree_desc(&trees
[0], prev
->buffer
, prev
->size
);
4118 init_tree_desc(&trees
[1], next
->buffer
, next
->size
);
4120 ret
= unpack_trees(2, trees
, &unpack_opts
);
4121 clear_unpack_trees_porcelain(&unpack_opts
);
4125 static int record_conflicted_index_entries(struct merge_options
*opt
)
4127 struct hashmap_iter iter
;
4128 struct strmap_entry
*e
;
4129 struct index_state
*index
= opt
->repo
->index
;
4130 struct checkout state
= CHECKOUT_INIT
;
4132 int original_cache_nr
;
4134 if (strmap_empty(&opt
->priv
->conflicted
))
4138 * We are in a conflicted state. These conflicts might be inside
4139 * sparse-directory entries, so check if any entries are outside
4140 * of the sparse-checkout cone preemptively.
4142 * We set original_cache_nr below, but that might change if
4143 * index_name_pos() calls ask for paths within sparse directories.
4145 strmap_for_each_entry(&opt
->priv
->conflicted
, &iter
, e
) {
4146 if (!path_in_sparse_checkout(e
->key
, index
)) {
4147 ensure_full_index(index
);
4152 /* If any entries have skip_worktree set, we'll have to check 'em out */
4155 state
.refresh_cache
= 1;
4156 state
.istate
= index
;
4157 original_cache_nr
= index
->cache_nr
;
4159 /* Append every entry from conflicted into index, then sort */
4160 strmap_for_each_entry(&opt
->priv
->conflicted
, &iter
, e
) {
4161 const char *path
= e
->key
;
4162 struct conflict_info
*ci
= e
->value
;
4164 struct cache_entry
*ce
;
4170 * The index will already have a stage=0 entry for this path,
4171 * because we created an as-merged-as-possible version of the
4172 * file and checkout() moved the working copy and index over
4175 * However, previous iterations through this loop will have
4176 * added unstaged entries to the end of the cache which
4177 * ignore the standard alphabetical ordering of cache
4178 * entries and break invariants needed for index_name_pos()
4179 * to work. However, we know the entry we want is before
4180 * those appended cache entries, so do a temporary swap on
4181 * cache_nr to only look through entries of interest.
4183 SWAP(index
->cache_nr
, original_cache_nr
);
4184 pos
= index_name_pos(index
, path
, strlen(path
));
4185 SWAP(index
->cache_nr
, original_cache_nr
);
4187 if (ci
->filemask
!= 1)
4188 BUG("Conflicted %s but nothing in basic working tree or index; this shouldn't happen", path
);
4189 cache_tree_invalidate_path(index
, path
);
4191 ce
= index
->cache
[pos
];
4194 * Clean paths with CE_SKIP_WORKTREE set will not be
4195 * written to the working tree by the unpack_trees()
4196 * call in checkout(). Our conflicted entries would
4197 * have appeared clean to that code since we ignored
4198 * the higher order stages. Thus, we need override
4199 * the CE_SKIP_WORKTREE bit and manually write those
4200 * files to the working disk here.
4202 if (ce_skip_worktree(ce
)) {
4205 if (!lstat(path
, &st
)) {
4206 char *new_name
= unique_path(opt
,
4210 path_msg(opt
, path
, 1,
4211 _("Note: %s not up to date and in way of checking out conflicted version; old copy renamed to %s"),
4213 errs
|= rename(path
, new_name
);
4215 errs
|= checkout_entry(ce
, &state
, NULL
, NULL
);
4219 * Mark this cache entry for removal and instead add
4220 * new stage>0 entries corresponding to the
4221 * conflicts. If there are many conflicted entries, we
4222 * want to avoid memmove'ing O(NM) entries by
4223 * inserting the new entries one at a time. So,
4224 * instead, we just add the new cache entries to the
4225 * end (ignoring normal index requirements on sort
4226 * order) and sort the index once we're all done.
4228 ce
->ce_flags
|= CE_REMOVE
;
4231 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
4232 struct version_info
*vi
;
4233 if (!(ci
->filemask
& (1ul << i
)))
4235 vi
= &ci
->stages
[i
];
4236 ce
= make_cache_entry(index
, vi
->mode
, &vi
->oid
,
4238 add_index_entry(index
, ce
, ADD_CACHE_JUST_APPEND
);
4243 * Remove the unused cache entries (and invalidate the relevant
4244 * cache-trees), then sort the index entries to get the conflicted
4245 * entries we added to the end into their right locations.
4247 remove_marked_cache_entries(index
, 1);
4249 * No need for STABLE_QSORT -- cmp_cache_name_compare sorts primarily
4250 * on filename and secondarily on stage, and (name, stage #) are a
4253 QSORT(index
->cache
, index
->cache_nr
, cmp_cache_name_compare
);
4258 void merge_switch_to_result(struct merge_options
*opt
,
4260 struct merge_result
*result
,
4261 int update_worktree_and_index
,
4262 int display_update_msgs
)
4264 assert(opt
->priv
== NULL
);
4265 if (result
->clean
>= 0 && update_worktree_and_index
) {
4266 const char *filename
;
4269 trace2_region_enter("merge", "checkout", opt
->repo
);
4270 if (checkout(opt
, head
, result
->tree
)) {
4271 /* failure to function */
4275 trace2_region_leave("merge", "checkout", opt
->repo
);
4277 trace2_region_enter("merge", "record_conflicted", opt
->repo
);
4278 opt
->priv
= result
->priv
;
4279 if (record_conflicted_index_entries(opt
)) {
4280 /* failure to function */
4286 trace2_region_leave("merge", "record_conflicted", opt
->repo
);
4288 trace2_region_enter("merge", "write_auto_merge", opt
->repo
);
4289 filename
= git_path_auto_merge(opt
->repo
);
4290 fp
= xfopen(filename
, "w");
4291 fprintf(fp
, "%s\n", oid_to_hex(&result
->tree
->object
.oid
));
4293 trace2_region_leave("merge", "write_auto_merge", opt
->repo
);
4296 if (display_update_msgs
) {
4297 struct merge_options_internal
*opti
= result
->priv
;
4298 struct hashmap_iter iter
;
4299 struct strmap_entry
*e
;
4300 struct string_list olist
= STRING_LIST_INIT_NODUP
;
4303 if (opt
->record_conflict_msgs_as_headers
)
4304 BUG("Either display conflict messages or record them as headers, not both");
4306 trace2_region_enter("merge", "display messages", opt
->repo
);
4308 /* Hack to pre-allocate olist to the desired size */
4309 ALLOC_GROW(olist
.items
, strmap_get_size(&opti
->output
),
4312 /* Put every entry from output into olist, then sort */
4313 strmap_for_each_entry(&opti
->output
, &iter
, e
) {
4314 string_list_append(&olist
, e
->key
)->util
= e
->value
;
4316 string_list_sort(&olist
);
4318 /* Iterate over the items, printing them */
4319 for (i
= 0; i
< olist
.nr
; ++i
) {
4320 struct strbuf
*sb
= olist
.items
[i
].util
;
4322 printf("%s", sb
->buf
);
4324 string_list_clear(&olist
, 0);
4326 /* Also include needed rename limit adjustment now */
4327 diff_warn_rename_limit("merge.renamelimit",
4328 opti
->renames
.needed_limit
, 0);
4330 trace2_region_leave("merge", "display messages", opt
->repo
);
4333 merge_finalize(opt
, result
);
4336 void merge_finalize(struct merge_options
*opt
,
4337 struct merge_result
*result
)
4339 struct merge_options_internal
*opti
= result
->priv
;
4341 if (opt
->renormalize
)
4342 git_attr_set_direction(GIT_ATTR_CHECKIN
);
4343 assert(opt
->priv
== NULL
);
4345 clear_or_reinit_internal_opts(opti
, 0);
4346 FREE_AND_NULL(opti
);
4349 /*** Function Grouping: helper functions for merge_incore_*() ***/
4351 static struct tree
*shift_tree_object(struct repository
*repo
,
4352 struct tree
*one
, struct tree
*two
,
4353 const char *subtree_shift
)
4355 struct object_id shifted
;
4357 if (!*subtree_shift
) {
4358 shift_tree(repo
, &one
->object
.oid
, &two
->object
.oid
, &shifted
, 0);
4360 shift_tree_by(repo
, &one
->object
.oid
, &two
->object
.oid
, &shifted
,
4363 if (oideq(&two
->object
.oid
, &shifted
))
4365 return lookup_tree(repo
, &shifted
);
4368 static inline void set_commit_tree(struct commit
*c
, struct tree
*t
)
4373 static struct commit
*make_virtual_commit(struct repository
*repo
,
4375 const char *comment
)
4377 struct commit
*commit
= alloc_commit_node(repo
);
4379 set_merge_remote_desc(commit
, comment
, (struct object
*)commit
);
4380 set_commit_tree(commit
, tree
);
4381 commit
->object
.parsed
= 1;
4385 static void merge_start(struct merge_options
*opt
, struct merge_result
*result
)
4387 struct rename_info
*renames
;
4389 struct mem_pool
*pool
= NULL
;
4391 /* Sanity checks on opt */
4392 trace2_region_enter("merge", "sanity checks", opt
->repo
);
4395 assert(opt
->branch1
&& opt
->branch2
);
4397 assert(opt
->detect_directory_renames
>= MERGE_DIRECTORY_RENAMES_NONE
&&
4398 opt
->detect_directory_renames
<= MERGE_DIRECTORY_RENAMES_TRUE
);
4399 assert(opt
->rename_limit
>= -1);
4400 assert(opt
->rename_score
>= 0 && opt
->rename_score
<= MAX_SCORE
);
4401 assert(opt
->show_rename_progress
>= 0 && opt
->show_rename_progress
<= 1);
4403 assert(opt
->xdl_opts
>= 0);
4404 assert(opt
->recursive_variant
>= MERGE_VARIANT_NORMAL
&&
4405 opt
->recursive_variant
<= MERGE_VARIANT_THEIRS
);
4407 if (opt
->msg_header_prefix
)
4408 assert(opt
->record_conflict_msgs_as_headers
);
4411 * detect_renames, verbosity, buffer_output, and obuf are ignored
4412 * fields that were used by "recursive" rather than "ort" -- but
4413 * sanity check them anyway.
4415 assert(opt
->detect_renames
>= -1 &&
4416 opt
->detect_renames
<= DIFF_DETECT_COPY
);
4417 assert(opt
->verbosity
>= 0 && opt
->verbosity
<= 5);
4418 assert(opt
->buffer_output
<= 2);
4419 assert(opt
->obuf
.len
== 0);
4421 assert(opt
->priv
== NULL
);
4422 if (result
->_properly_initialized
!= 0 &&
4423 result
->_properly_initialized
!= RESULT_INITIALIZED
)
4424 BUG("struct merge_result passed to merge_incore_*recursive() must be zeroed or filled with values from a previous run");
4425 assert(!!result
->priv
== !!result
->_properly_initialized
);
4427 opt
->priv
= result
->priv
;
4428 result
->priv
= NULL
;
4430 * opt->priv non-NULL means we had results from a previous
4431 * run; do a few sanity checks that user didn't mess with
4432 * it in an obvious fashion.
4434 assert(opt
->priv
->call_depth
== 0);
4435 assert(!opt
->priv
->toplevel_dir
||
4436 0 == strlen(opt
->priv
->toplevel_dir
));
4438 trace2_region_leave("merge", "sanity checks", opt
->repo
);
4440 /* Default to histogram diff. Actually, just hardcode it...for now. */
4441 opt
->xdl_opts
= DIFF_WITH_ALG(opt
, HISTOGRAM_DIFF
);
4443 /* Handle attr direction stuff for renormalization */
4444 if (opt
->renormalize
)
4445 git_attr_set_direction(GIT_ATTR_CHECKOUT
);
4447 /* Initialization of opt->priv, our internal merge data */
4448 trace2_region_enter("merge", "allocate/init", opt
->repo
);
4450 clear_or_reinit_internal_opts(opt
->priv
, 1);
4451 trace2_region_leave("merge", "allocate/init", opt
->repo
);
4454 opt
->priv
= xcalloc(1, sizeof(*opt
->priv
));
4456 /* Initialization of various renames fields */
4457 renames
= &opt
->priv
->renames
;
4458 mem_pool_init(&opt
->priv
->pool
, 0);
4459 pool
= &opt
->priv
->pool
;
4460 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; i
++) {
4461 strintmap_init_with_options(&renames
->dirs_removed
[i
],
4462 NOT_RELEVANT
, pool
, 0);
4463 strmap_init_with_options(&renames
->dir_rename_count
[i
],
4465 strmap_init_with_options(&renames
->dir_renames
[i
],
4468 * relevant_sources uses -1 for the default, because we need
4469 * to be able to distinguish not-in-strintmap from valid
4470 * relevant_source values from enum file_rename_relevance.
4471 * In particular, possibly_cache_new_pair() expects a negative
4472 * value for not-found entries.
4474 strintmap_init_with_options(&renames
->relevant_sources
[i
],
4475 -1 /* explicitly invalid */,
4477 strmap_init_with_options(&renames
->cached_pairs
[i
],
4479 strset_init_with_options(&renames
->cached_irrelevant
[i
],
4481 strset_init_with_options(&renames
->cached_target_names
[i
],
4484 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; i
++) {
4485 strintmap_init_with_options(&renames
->deferred
[i
].possible_trivial_merges
,
4487 strset_init_with_options(&renames
->deferred
[i
].target_dirs
,
4489 renames
->deferred
[i
].trivial_merges_okay
= 1; /* 1 == maybe */
4493 * Although we initialize opt->priv->paths with strdup_strings=0,
4494 * that's just to avoid making yet another copy of an allocated
4495 * string. Putting the entry into paths means we are taking
4496 * ownership, so we will later free it.
4498 * In contrast, conflicted just has a subset of keys from paths, so
4499 * we don't want to free those (it'd be a duplicate free).
4501 strmap_init_with_options(&opt
->priv
->paths
, pool
, 0);
4502 strmap_init_with_options(&opt
->priv
->conflicted
, pool
, 0);
4505 * keys & strbufs in output will sometimes need to outlive "paths",
4506 * so it will have a copy of relevant keys. It's probably a small
4507 * subset of the overall paths that have special output.
4509 strmap_init(&opt
->priv
->output
);
4511 trace2_region_leave("merge", "allocate/init", opt
->repo
);
4514 static void merge_check_renames_reusable(struct merge_options
*opt
,
4515 struct merge_result
*result
,
4516 struct tree
*merge_base
,
4520 struct rename_info
*renames
;
4521 struct tree
**merge_trees
;
4522 struct merge_options_internal
*opti
= result
->priv
;
4527 renames
= &opti
->renames
;
4528 merge_trees
= renames
->merge_trees
;
4531 * Handle case where previous merge operation did not want cache to
4532 * take effect, e.g. because rename/rename(1to1) makes it invalid.
4534 if (!merge_trees
[0]) {
4535 assert(!merge_trees
[0] && !merge_trees
[1] && !merge_trees
[2]);
4536 renames
->cached_pairs_valid_side
= 0; /* neither side valid */
4541 * Handle other cases; note that merge_trees[0..2] will only
4542 * be NULL if opti is, or if all three were manually set to
4543 * NULL by e.g. rename/rename(1to1) handling.
4545 assert(merge_trees
[0] && merge_trees
[1] && merge_trees
[2]);
4547 /* Check if we meet a condition for re-using cached_pairs */
4548 if (oideq(&merge_base
->object
.oid
, &merge_trees
[2]->object
.oid
) &&
4549 oideq(&side1
->object
.oid
, &result
->tree
->object
.oid
))
4550 renames
->cached_pairs_valid_side
= MERGE_SIDE1
;
4551 else if (oideq(&merge_base
->object
.oid
, &merge_trees
[1]->object
.oid
) &&
4552 oideq(&side2
->object
.oid
, &result
->tree
->object
.oid
))
4553 renames
->cached_pairs_valid_side
= MERGE_SIDE2
;
4555 renames
->cached_pairs_valid_side
= 0; /* neither side valid */
4558 /*** Function Grouping: merge_incore_*() and their internal variants ***/
4561 * Originally from merge_trees_internal(); heavily adapted, though.
4563 static void merge_ort_nonrecursive_internal(struct merge_options
*opt
,
4564 struct tree
*merge_base
,
4567 struct merge_result
*result
)
4569 struct object_id working_tree_oid
;
4571 if (opt
->subtree_shift
) {
4572 side2
= shift_tree_object(opt
->repo
, side1
, side2
,
4573 opt
->subtree_shift
);
4574 merge_base
= shift_tree_object(opt
->repo
, side1
, merge_base
,
4575 opt
->subtree_shift
);
4579 trace2_region_enter("merge", "collect_merge_info", opt
->repo
);
4580 if (collect_merge_info(opt
, merge_base
, side1
, side2
) != 0) {
4582 * TRANSLATORS: The %s arguments are: 1) tree hash of a merge
4583 * base, and 2-3) the trees for the two trees we're merging.
4585 err(opt
, _("collecting merge info failed for trees %s, %s, %s"),
4586 oid_to_hex(&merge_base
->object
.oid
),
4587 oid_to_hex(&side1
->object
.oid
),
4588 oid_to_hex(&side2
->object
.oid
));
4592 trace2_region_leave("merge", "collect_merge_info", opt
->repo
);
4594 trace2_region_enter("merge", "renames", opt
->repo
);
4595 result
->clean
= detect_and_process_renames(opt
, merge_base
,
4597 trace2_region_leave("merge", "renames", opt
->repo
);
4598 if (opt
->priv
->renames
.redo_after_renames
== 2) {
4599 trace2_region_enter("merge", "reset_maps", opt
->repo
);
4600 clear_or_reinit_internal_opts(opt
->priv
, 1);
4601 trace2_region_leave("merge", "reset_maps", opt
->repo
);
4605 trace2_region_enter("merge", "process_entries", opt
->repo
);
4606 process_entries(opt
, &working_tree_oid
);
4607 trace2_region_leave("merge", "process_entries", opt
->repo
);
4609 /* Set return values */
4610 result
->path_messages
= &opt
->priv
->output
;
4611 result
->tree
= parse_tree_indirect(&working_tree_oid
);
4612 /* existence of conflicted entries implies unclean */
4613 result
->clean
&= strmap_empty(&opt
->priv
->conflicted
);
4614 if (!opt
->priv
->call_depth
) {
4615 result
->priv
= opt
->priv
;
4616 result
->_properly_initialized
= RESULT_INITIALIZED
;
4622 * Originally from merge_recursive_internal(); somewhat adapted, though.
4624 static void merge_ort_internal(struct merge_options
*opt
,
4625 struct commit_list
*merge_bases
,
4628 struct merge_result
*result
)
4630 struct commit
*next
;
4631 struct commit
*merged_merge_bases
;
4632 const char *ancestor_name
;
4633 struct strbuf merge_base_abbrev
= STRBUF_INIT
;
4636 merge_bases
= get_merge_bases(h1
, h2
);
4637 /* See merge-ort.h:merge_incore_recursive() declaration NOTE */
4638 merge_bases
= reverse_commit_list(merge_bases
);
4641 merged_merge_bases
= pop_commit(&merge_bases
);
4642 if (merged_merge_bases
== NULL
) {
4643 /* if there is no common ancestor, use an empty tree */
4646 tree
= lookup_tree(opt
->repo
, opt
->repo
->hash_algo
->empty_tree
);
4647 merged_merge_bases
= make_virtual_commit(opt
->repo
, tree
,
4649 ancestor_name
= "empty tree";
4650 } else if (merge_bases
) {
4651 ancestor_name
= "merged common ancestors";
4653 strbuf_add_unique_abbrev(&merge_base_abbrev
,
4654 &merged_merge_bases
->object
.oid
,
4656 ancestor_name
= merge_base_abbrev
.buf
;
4659 for (next
= pop_commit(&merge_bases
); next
;
4660 next
= pop_commit(&merge_bases
)) {
4661 const char *saved_b1
, *saved_b2
;
4662 struct commit
*prev
= merged_merge_bases
;
4664 opt
->priv
->call_depth
++;
4666 * When the merge fails, the result contains files
4667 * with conflict markers. The cleanness flag is
4668 * ignored (unless indicating an error), it was never
4669 * actually used, as result of merge_trees has always
4670 * overwritten it: the committed "conflicts" were
4673 saved_b1
= opt
->branch1
;
4674 saved_b2
= opt
->branch2
;
4675 opt
->branch1
= "Temporary merge branch 1";
4676 opt
->branch2
= "Temporary merge branch 2";
4677 merge_ort_internal(opt
, NULL
, prev
, next
, result
);
4678 if (result
->clean
< 0)
4680 opt
->branch1
= saved_b1
;
4681 opt
->branch2
= saved_b2
;
4682 opt
->priv
->call_depth
--;
4684 merged_merge_bases
= make_virtual_commit(opt
->repo
,
4687 commit_list_insert(prev
, &merged_merge_bases
->parents
);
4688 commit_list_insert(next
, &merged_merge_bases
->parents
->next
);
4690 clear_or_reinit_internal_opts(opt
->priv
, 1);
4693 opt
->ancestor
= ancestor_name
;
4694 merge_ort_nonrecursive_internal(opt
,
4695 repo_get_commit_tree(opt
->repo
,
4696 merged_merge_bases
),
4697 repo_get_commit_tree(opt
->repo
, h1
),
4698 repo_get_commit_tree(opt
->repo
, h2
),
4700 strbuf_release(&merge_base_abbrev
);
4701 opt
->ancestor
= NULL
; /* avoid accidental re-use of opt->ancestor */
4704 void merge_incore_nonrecursive(struct merge_options
*opt
,
4705 struct tree
*merge_base
,
4708 struct merge_result
*result
)
4710 trace2_region_enter("merge", "incore_nonrecursive", opt
->repo
);
4712 trace2_region_enter("merge", "merge_start", opt
->repo
);
4713 assert(opt
->ancestor
!= NULL
);
4714 merge_check_renames_reusable(opt
, result
, merge_base
, side1
, side2
);
4715 merge_start(opt
, result
);
4717 * Record the trees used in this merge, so if there's a next merge in
4718 * a cherry-pick or rebase sequence it might be able to take advantage
4719 * of the cached_pairs in that next merge.
4721 opt
->priv
->renames
.merge_trees
[0] = merge_base
;
4722 opt
->priv
->renames
.merge_trees
[1] = side1
;
4723 opt
->priv
->renames
.merge_trees
[2] = side2
;
4724 trace2_region_leave("merge", "merge_start", opt
->repo
);
4726 merge_ort_nonrecursive_internal(opt
, merge_base
, side1
, side2
, result
);
4727 trace2_region_leave("merge", "incore_nonrecursive", opt
->repo
);
4730 void merge_incore_recursive(struct merge_options
*opt
,
4731 struct commit_list
*merge_bases
,
4732 struct commit
*side1
,
4733 struct commit
*side2
,
4734 struct merge_result
*result
)
4736 trace2_region_enter("merge", "incore_recursive", opt
->repo
);
4738 /* We set the ancestor label based on the merge_bases */
4739 assert(opt
->ancestor
== NULL
);
4741 trace2_region_enter("merge", "merge_start", opt
->repo
);
4742 merge_start(opt
, result
);
4743 trace2_region_leave("merge", "merge_start", opt
->repo
);
4745 merge_ort_internal(opt
, merge_bases
, side1
, side2
, result
);
4746 trace2_region_leave("merge", "incore_recursive", opt
->repo
);