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.h"
37 #include "unpack-trees.h"
38 #include "xdiff-interface.h"
40 #define USE_MEMORY_POOL 1 /* faster, but obscures memory leak hunting */
43 * We have many arrays of size 3. Whenever we have such an array, the
44 * indices refer to one of the sides of the three-way merge. This is so
45 * pervasive that the constants 0, 1, and 2 are used in many places in the
46 * code (especially in arithmetic operations to find the other side's index
47 * or to compute a relevant mask), but sometimes these enum names are used
48 * to aid code clarity.
50 * See also 'filemask' and 'dirmask' in struct conflict_info; the "ith side"
51 * referred to there is one of these three sides.
59 static unsigned RESULT_INITIALIZED
= 0x1abe11ed; /* unlikely accidental value */
61 struct traversal_callback_data
{
63 unsigned long dirmask
;
64 struct name_entry names
[3];
67 struct deferred_traversal_data
{
69 * possible_trivial_merges: directories to be explored only when needed
71 * possible_trivial_merges is a map of directory names to
72 * dir_rename_mask. When we detect that a directory is unchanged on
73 * one side, we can sometimes resolve the directory without recursing
74 * into it. Renames are the only things that can prevent such an
75 * optimization. However, for rename sources:
76 * - If no parent directory needed directory rename detection, then
77 * no path under such a directory can be a relevant_source.
78 * and for rename destinations:
79 * - If no cached rename has a target path under the directory AND
80 * - If there are no unpaired relevant_sources elsewhere in the
82 * then we don't need any path under this directory for a rename
83 * destination. The only way to know the last item above is to defer
84 * handling such directories until the end of collect_merge_info(),
85 * in handle_deferred_entries().
87 * For each we store dir_rename_mask, since that's the only bit of
88 * information we need, other than the path, to resume the recursive
91 struct strintmap possible_trivial_merges
;
94 * trivial_merges_okay: if trivial directory merges are okay
96 * See possible_trivial_merges above. The "no unpaired
97 * relevant_sources elsewhere in the repository" is a single boolean
98 * per merge side, which we store here. Note that while 0 means no,
99 * 1 only means "maybe" rather than "yes"; we optimistically set it
100 * to 1 initially and only clear when we determine it is unsafe to
101 * do trivial directory merges.
103 unsigned trivial_merges_okay
;
106 * target_dirs: ancestor directories of rename targets
108 * target_dirs contains all directory names that are an ancestor of
109 * any rename destination.
111 struct strset target_dirs
;
116 * All variables that are arrays of size 3 correspond to data tracked
117 * for the sides in enum merge_side. Index 0 is almost always unused
118 * because we often only need to track information for MERGE_SIDE1 and
119 * MERGE_SIDE2 (MERGE_BASE can't have rename information since renames
120 * are determined relative to what changed since the MERGE_BASE).
124 * pairs: pairing of filenames from diffcore_rename()
126 struct diff_queue_struct pairs
[3];
129 * dirs_removed: directories removed on a given side of history.
131 * The keys of dirs_removed[side] are the directories that were removed
132 * on the given side of history. The value of the strintmap for each
133 * directory is a value from enum dir_rename_relevance.
135 struct strintmap dirs_removed
[3];
138 * dir_rename_count: tracking where parts of a directory were renamed to
140 * When files in a directory are renamed, they may not all go to the
141 * same location. Each strmap here tracks:
142 * old_dir => {new_dir => int}
143 * That is, dir_rename_count[side] is a strmap to a strintmap.
145 struct strmap dir_rename_count
[3];
148 * dir_renames: computed directory renames
150 * This is a map of old_dir => new_dir and is derived in part from
153 struct strmap dir_renames
[3];
156 * relevant_sources: deleted paths wanted in rename detection, and why
158 * relevant_sources is a set of deleted paths on each side of
159 * history for which we need rename detection. If a path is deleted
160 * on one side of history, we need to detect if it is part of a
162 * * the file is modified/deleted on the other side of history
163 * * we need to detect renames for an ancestor directory
164 * If neither of those are true, we can skip rename detection for
165 * that path. The reason is stored as a value from enum
166 * file_rename_relevance, as the reason can inform the algorithm in
167 * diffcore_rename_extended().
169 struct strintmap relevant_sources
[3];
171 struct deferred_traversal_data deferred
[3];
175 * 0: optimization removing unmodified potential rename source okay
176 * 2 or 4: optimization okay, but must check for files added to dir
177 * 7: optimization forbidden; need rename source in case of dir rename
179 unsigned dir_rename_mask
:3;
182 * callback_data_*: supporting data structures for alternate traversal
184 * We sometimes need to be able to traverse through all the files
185 * in a given tree before all immediate subdirectories within that
186 * tree. Since traverse_trees() doesn't do that naturally, we have
187 * a traverse_trees_wrapper() that stores any immediate
188 * subdirectories while traversing files, then traverses the
189 * immediate subdirectories later. These callback_data* variables
190 * store the information for the subdirectories so that we can do
191 * that traversal order.
193 struct traversal_callback_data
*callback_data
;
194 int callback_data_nr
, callback_data_alloc
;
195 char *callback_data_traverse_path
;
198 * merge_trees: trees passed to the merge algorithm for the merge
200 * merge_trees records the trees passed to the merge algorithm. But,
201 * this data also is stored in merge_result->priv. If a sequence of
202 * merges are being done (such as when cherry-picking or rebasing),
203 * the next merge can look at this and re-use information from
204 * previous merges under certain circumstances.
206 * See also all the cached_* variables.
208 struct tree
*merge_trees
[3];
211 * cached_pairs_valid_side: which side's cached info can be reused
213 * See the description for merge_trees. For repeated merges, at most
214 * only one side's cached information can be used. Valid values:
215 * MERGE_SIDE2: cached data from side2 can be reused
216 * MERGE_SIDE1: cached data from side1 can be reused
217 * 0: no cached data can be reused
218 * -1: See redo_after_renames; both sides can be reused.
220 int cached_pairs_valid_side
;
223 * cached_pairs: Caching of renames and deletions.
225 * These are mappings recording renames and deletions of individual
226 * files (not directories). They are thus a map from an old
227 * filename to either NULL (for deletions) or a new filename (for
230 struct strmap cached_pairs
[3];
233 * cached_target_names: just the destinations from cached_pairs
235 * We sometimes want a fast lookup to determine if a given filename
236 * is one of the destinations in cached_pairs. cached_target_names
237 * is thus duplicative information, but it provides a fast lookup.
239 struct strset cached_target_names
[3];
242 * cached_irrelevant: Caching of rename_sources that aren't relevant.
244 * If we try to detect a rename for a source path and succeed, it's
245 * part of a rename. If we try to detect a rename for a source path
246 * and fail, then it's a delete. If we do not try to detect a rename
247 * for a path, then we don't know if it's a rename or a delete. If
248 * merge-ort doesn't think the path is relevant, then we just won't
249 * cache anything for that path. But there's a slight problem in
250 * that merge-ort can think a path is RELEVANT_LOCATION, but due to
251 * commit 9bd342137e ("diffcore-rename: determine which
252 * relevant_sources are no longer relevant", 2021-03-13),
253 * diffcore-rename can downgrade the path to RELEVANT_NO_MORE. To
254 * avoid excessive calls to diffcore_rename_extended() we still need
255 * to cache such paths, though we cannot record them as either
256 * renames or deletes. So we cache them here as a "turned out to be
257 * irrelevant *for this commit*" as they are often also irrelevant
258 * for subsequent commits, though we will have to do some extra
259 * checking to see whether such paths become relevant for rename
260 * detection when cherry-picking/rebasing subsequent commits.
262 struct strset cached_irrelevant
[3];
265 * redo_after_renames: optimization flag for "restarting" the merge
267 * Sometimes it pays to detect renames, cache them, and then
268 * restart the merge operation from the beginning. The reason for
269 * this is that when we know where all the renames are, we know
270 * whether a certain directory has any paths under it affected --
271 * and if a directory is not affected then it permits us to do
272 * trivial tree merging in more cases. Doing trivial tree merging
273 * prevents the need to run process_entry() on every path
274 * underneath trees that can be trivially merged, and
275 * process_entry() is more expensive than collect_merge_info() --
276 * plus, the second collect_merge_info() will be much faster since
277 * it doesn't have to recurse into the relevant trees.
279 * Values for this flag:
280 * 0 = don't bother, not worth it (or conditions not yet checked)
281 * 1 = conditions for optimization met, optimization worthwhile
282 * 2 = we already did it (don't restart merge yet again)
284 unsigned redo_after_renames
;
287 * needed_limit: value needed for inexact rename detection to run
289 * If the current rename limit wasn't high enough for inexact
290 * rename detection to run, this records the limit needed. Otherwise,
291 * this value remains 0.
296 struct merge_options_internal
{
298 * paths: primary data structure in all of merge ort.
301 * * are full relative paths from the toplevel of the repository
302 * (e.g. "drivers/firmware/raspberrypi.c").
303 * * store all relevant paths in the repo, both directories and
304 * files (e.g. drivers, drivers/firmware would also be included)
305 * * these keys serve to intern all the path strings, which allows
306 * us to do pointer comparison on directory names instead of
307 * strcmp; we just have to be careful to use the interned strings.
308 * (Technically paths_to_free may track some strings that were
309 * removed from froms paths.)
311 * The values of paths:
312 * * either a pointer to a merged_info, or a conflict_info struct
313 * * merged_info contains all relevant information for a
314 * non-conflicted entry.
315 * * conflict_info contains a merged_info, plus any additional
316 * information about a conflict such as the higher orders stages
317 * involved and the names of the paths those came from (handy
318 * once renames get involved).
319 * * a path may start "conflicted" (i.e. point to a conflict_info)
320 * and then a later step (e.g. three-way content merge) determines
321 * it can be cleanly merged, at which point it'll be marked clean
322 * and the algorithm will ignore any data outside the contained
323 * merged_info for that entry
324 * * If an entry remains conflicted, the merged_info portion of a
325 * conflict_info will later be filled with whatever version of
326 * the file should be placed in the working directory (e.g. an
327 * as-merged-as-possible variation that contains conflict markers).
332 * conflicted: a subset of keys->values from "paths"
334 * conflicted is basically an optimization between process_entries()
335 * and record_conflicted_index_entries(); the latter could loop over
336 * ALL the entries in paths AGAIN and look for the ones that are
337 * still conflicted, but since process_entries() has to loop over
338 * all of them, it saves the ones it couldn't resolve in this strmap
339 * so that record_conflicted_index_entries() can iterate just the
342 struct strmap conflicted
;
345 * pool: memory pool for fast allocation/deallocation
347 * We allocate room for lots of filenames and auxiliary data
348 * structures in merge_options_internal, and it tends to all be
349 * freed together too. Using a memory pool for these provides a
352 struct mem_pool internal_pool
;
353 struct mem_pool
*pool
; /* NULL, or pointer to internal_pool */
356 * paths_to_free: additional list of strings to free
358 * If keys are removed from "paths", they are added to paths_to_free
359 * to ensure they are later freed. We avoid free'ing immediately since
360 * other places (e.g. conflict_info.pathnames[]) may still be
361 * referencing these paths.
363 struct string_list paths_to_free
;
366 * output: special messages and conflict notices for various paths
368 * This is a map of pathnames (a subset of the keys in "paths" above)
369 * to strbufs. It gathers various warning/conflict/notice messages
370 * for later processing.
372 struct strmap output
;
375 * renames: various data relating to rename detection
377 struct rename_info renames
;
380 * attr_index: hacky minimal index used for renormalization
382 * renormalization code _requires_ an index, though it only needs to
383 * find a .gitattributes file within the index. So, when
384 * renormalization is important, we create a special index with just
387 struct index_state attr_index
;
390 * current_dir_name, toplevel_dir: temporary vars
392 * These are used in collect_merge_info_callback(), and will set the
393 * various merged_info.directory_name for the various paths we get;
394 * see documentation for that variable and the requirements placed on
397 const char *current_dir_name
;
398 const char *toplevel_dir
;
400 /* call_depth: recursion level counter for merging merge bases */
404 struct version_info
{
405 struct object_id oid
;
410 /* if is_null, ignore result. otherwise result has oid & mode */
411 struct version_info result
;
415 * clean: whether the path in question is cleanly merged.
417 * see conflict_info.merged for more details.
422 * basename_offset: offset of basename of path.
424 * perf optimization to avoid recomputing offset of final '/'
425 * character in pathname (0 if no '/' in pathname).
427 size_t basename_offset
;
430 * directory_name: containing directory name.
432 * Note that we assume directory_name is constructed such that
433 * strcmp(dir1_name, dir2_name) == 0 iff dir1_name == dir2_name,
434 * i.e. string equality is equivalent to pointer equality. For this
435 * to hold, we have to be careful setting directory_name.
437 const char *directory_name
;
440 struct conflict_info
{
442 * merged: the version of the path that will be written to working tree
444 * WARNING: It is critical to check merged.clean and ensure it is 0
445 * before reading any conflict_info fields outside of merged.
446 * Allocated merge_info structs will always have clean set to 1.
447 * Allocated conflict_info structs will have merged.clean set to 0
448 * initially. The merged.clean field is how we know if it is safe
449 * to access other parts of conflict_info besides merged; if a
450 * conflict_info's merged.clean is changed to 1, the rest of the
451 * algorithm is not allowed to look at anything outside of the
452 * merged member anymore.
454 struct merged_info merged
;
456 /* oids & modes from each of the three trees for this path */
457 struct version_info stages
[3];
459 /* pathnames for each stage; may differ due to rename detection */
460 const char *pathnames
[3];
462 /* Whether this path is/was involved in a directory/file conflict */
463 unsigned df_conflict
:1;
466 * Whether this path is/was involved in a non-content conflict other
467 * than a directory/file conflict (e.g. rename/rename, rename/delete,
468 * file location based on possible directory rename).
470 unsigned path_conflict
:1;
473 * For filemask and dirmask, the ith bit corresponds to whether the
474 * ith entry is a file (filemask) or a directory (dirmask). Thus,
475 * filemask & dirmask is always zero, and filemask | dirmask is at
476 * most 7 but can be less when a path does not appear as either a
477 * file or a directory on at least one side of history.
479 * Note that these masks are related to enum merge_side, as the ith
480 * entry corresponds to side i.
482 * These values come from a traverse_trees() call; more info may be
483 * found looking at tree-walk.h's struct traverse_info,
484 * particularly the documentation above the "fn" member (note that
485 * filemask = mask & ~dirmask from that documentation).
491 * Optimization to track which stages match, to avoid the need to
492 * recompute it in multiple steps. Either 0 or at least 2 bits are
493 * set; if at least 2 bits are set, their corresponding stages match.
495 unsigned match_mask
:3;
498 /*** Function Grouping: various utility functions ***/
501 * For the next three macros, see warning for conflict_info.merged.
503 * In each of the below, mi is a struct merged_info*, and ci was defined
504 * as a struct conflict_info* (but we need to verify ci isn't actually
505 * pointed at a struct merged_info*).
507 * INITIALIZE_CI: Assign ci to mi but only if it's safe; set to NULL otherwise.
508 * VERIFY_CI: Ensure that something we assigned to a conflict_info* is one.
509 * ASSIGN_AND_VERIFY_CI: Similar to VERIFY_CI but do assignment first.
511 #define INITIALIZE_CI(ci, mi) do { \
512 (ci) = (!(mi) || (mi)->clean) ? NULL : (struct conflict_info *)(mi); \
514 #define VERIFY_CI(ci) assert(ci && !ci->merged.clean);
515 #define ASSIGN_AND_VERIFY_CI(ci, mi) do { \
516 (ci) = (struct conflict_info *)(mi); \
517 assert((ci) && !(mi)->clean); \
520 static void free_strmap_strings(struct strmap
*map
)
522 struct hashmap_iter iter
;
523 struct strmap_entry
*entry
;
525 strmap_for_each_entry(map
, &iter
, entry
) {
526 free((char*)entry
->key
);
530 static void clear_or_reinit_internal_opts(struct merge_options_internal
*opti
,
533 struct rename_info
*renames
= &opti
->renames
;
535 void (*strmap_clear_func
)(struct strmap
*, int) =
536 reinitialize
? strmap_partial_clear
: strmap_clear
;
537 void (*strintmap_clear_func
)(struct strintmap
*) =
538 reinitialize
? strintmap_partial_clear
: strintmap_clear
;
539 void (*strset_clear_func
)(struct strset
*) =
540 reinitialize
? strset_partial_clear
: strset_clear
;
543 strmap_clear_func(&opti
->paths
, 0);
546 * We marked opti->paths with strdup_strings = 0, so that
547 * we wouldn't have to make another copy of the fullpath
548 * created by make_traverse_path from setup_path_info().
549 * But, now that we've used it and have no other references
550 * to these strings, it is time to deallocate them.
552 free_strmap_strings(&opti
->paths
);
553 strmap_clear_func(&opti
->paths
, 1);
557 * All keys and values in opti->conflicted are a subset of those in
558 * opti->paths. We don't want to deallocate anything twice, so we
559 * don't free the keys and we pass 0 for free_values.
561 strmap_clear_func(&opti
->conflicted
, 0);
565 * opti->paths_to_free is similar to opti->paths; we
566 * created it with strdup_strings = 0 to avoid making
567 * _another_ copy of the fullpath but now that we've used
568 * it and have no other references to these strings, it is
569 * time to deallocate them. We do so by temporarily
570 * setting strdup_strings to 1.
572 opti
->paths_to_free
.strdup_strings
= 1;
573 string_list_clear(&opti
->paths_to_free
, 0);
574 opti
->paths_to_free
.strdup_strings
= 0;
577 if (opti
->attr_index
.cache_nr
) /* true iff opt->renormalize */
578 discard_index(&opti
->attr_index
);
580 /* Free memory used by various renames maps */
581 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; ++i
) {
582 strintmap_clear_func(&renames
->dirs_removed
[i
]);
583 strmap_clear_func(&renames
->dir_renames
[i
], 0);
584 strintmap_clear_func(&renames
->relevant_sources
[i
]);
586 assert(renames
->cached_pairs_valid_side
== 0);
587 if (i
!= renames
->cached_pairs_valid_side
&&
588 -1 != renames
->cached_pairs_valid_side
) {
589 strset_clear_func(&renames
->cached_target_names
[i
]);
590 strmap_clear_func(&renames
->cached_pairs
[i
], 1);
591 strset_clear_func(&renames
->cached_irrelevant
[i
]);
592 partial_clear_dir_rename_count(&renames
->dir_rename_count
[i
]);
594 strmap_clear(&renames
->dir_rename_count
[i
], 1);
597 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; ++i
) {
598 strintmap_clear_func(&renames
->deferred
[i
].possible_trivial_merges
);
599 strset_clear_func(&renames
->deferred
[i
].target_dirs
);
600 renames
->deferred
[i
].trivial_merges_okay
= 1; /* 1 == maybe */
602 renames
->cached_pairs_valid_side
= 0;
603 renames
->dir_rename_mask
= 0;
606 struct hashmap_iter iter
;
607 struct strmap_entry
*e
;
609 /* Release and free each strbuf found in output */
610 strmap_for_each_entry(&opti
->output
, &iter
, e
) {
611 struct strbuf
*sb
= e
->value
;
614 * While strictly speaking we don't need to free(sb)
615 * here because we could pass free_values=1 when
616 * calling strmap_clear() on opti->output, that would
617 * require strmap_clear to do another
618 * strmap_for_each_entry() loop, so we just free it
619 * while we're iterating anyway.
623 strmap_clear(&opti
->output
, 0);
627 mem_pool_discard(&opti
->internal_pool
, 0);
632 /* Clean out callback_data as well. */
633 FREE_AND_NULL(renames
->callback_data
);
634 renames
->callback_data_nr
= renames
->callback_data_alloc
= 0;
637 static int err(struct merge_options
*opt
, const char *err
, ...)
640 struct strbuf sb
= STRBUF_INIT
;
642 strbuf_addstr(&sb
, "error: ");
643 va_start(params
, err
);
644 strbuf_vaddf(&sb
, err
, params
);
653 static void format_commit(struct strbuf
*sb
,
655 struct commit
*commit
)
657 struct merge_remote_desc
*desc
;
658 struct pretty_print_context ctx
= {0};
659 ctx
.abbrev
= DEFAULT_ABBREV
;
661 strbuf_addchars(sb
, ' ', indent
);
662 desc
= merge_remote_util(commit
);
664 strbuf_addf(sb
, "virtual %s\n", desc
->name
);
668 format_commit_message(commit
, "%h %s", sb
, &ctx
);
669 strbuf_addch(sb
, '\n');
672 __attribute__((format (printf
, 4, 5)))
673 static void path_msg(struct merge_options
*opt
,
675 int omittable_hint
, /* skippable under --remerge-diff */
676 const char *fmt
, ...)
679 struct strbuf
*sb
= strmap_get(&opt
->priv
->output
, path
);
681 sb
= xmalloc(sizeof(*sb
));
683 strmap_put(&opt
->priv
->output
, path
, sb
);
687 strbuf_vaddf(sb
, fmt
, ap
);
690 strbuf_addch(sb
, '\n');
693 static struct diff_filespec
*pool_alloc_filespec(struct mem_pool
*pool
,
696 struct diff_filespec
*spec
;
699 return alloc_filespec(path
);
701 /* Similar to alloc_filespec, but allocate from pool and reuse path */
702 spec
= mem_pool_calloc(pool
, 1, sizeof(*spec
));
703 spec
->path
= (char*)path
; /* spec won't modify it */
706 spec
->is_binary
= -1;
710 static struct diff_filepair
*pool_diff_queue(struct mem_pool
*pool
,
711 struct diff_queue_struct
*queue
,
712 struct diff_filespec
*one
,
713 struct diff_filespec
*two
)
715 struct diff_filepair
*dp
;
718 return diff_queue(queue
, one
, two
);
720 /* Same code as diff_queue, except allocate from pool */
721 dp
= mem_pool_calloc(pool
, 1, sizeof(*dp
));
729 static void *pool_calloc(struct mem_pool
*pool
, size_t count
, size_t size
)
732 return xcalloc(count
, size
);
733 return mem_pool_calloc(pool
, count
, size
);
736 static void *pool_alloc(struct mem_pool
*pool
, size_t size
)
739 return xmalloc(size
);
740 return mem_pool_alloc(pool
, size
);
743 static void *pool_strndup(struct mem_pool
*pool
, const char *str
, size_t len
)
746 return xstrndup(str
, len
);
747 return mem_pool_strndup(pool
, str
, len
);
750 /* add a string to a strbuf, but converting "/" to "_" */
751 static void add_flattened_path(struct strbuf
*out
, const char *s
)
754 strbuf_addstr(out
, s
);
755 for (; i
< out
->len
; i
++)
756 if (out
->buf
[i
] == '/')
760 static char *unique_path(struct strmap
*existing_paths
,
764 struct strbuf newpath
= STRBUF_INIT
;
768 strbuf_addf(&newpath
, "%s~", path
);
769 add_flattened_path(&newpath
, branch
);
771 base_len
= newpath
.len
;
772 while (strmap_contains(existing_paths
, newpath
.buf
)) {
773 strbuf_setlen(&newpath
, base_len
);
774 strbuf_addf(&newpath
, "_%d", suffix
++);
777 return strbuf_detach(&newpath
, NULL
);
780 /*** Function Grouping: functions related to collect_merge_info() ***/
782 static int traverse_trees_wrapper_callback(int n
,
784 unsigned long dirmask
,
785 struct name_entry
*names
,
786 struct traverse_info
*info
)
788 struct merge_options
*opt
= info
->data
;
789 struct rename_info
*renames
= &opt
->priv
->renames
;
790 unsigned filemask
= mask
& ~dirmask
;
794 if (!renames
->callback_data_traverse_path
)
795 renames
->callback_data_traverse_path
= xstrdup(info
->traverse_path
);
797 if (filemask
&& filemask
== renames
->dir_rename_mask
)
798 renames
->dir_rename_mask
= 0x07;
800 ALLOC_GROW(renames
->callback_data
, renames
->callback_data_nr
+ 1,
801 renames
->callback_data_alloc
);
802 renames
->callback_data
[renames
->callback_data_nr
].mask
= mask
;
803 renames
->callback_data
[renames
->callback_data_nr
].dirmask
= dirmask
;
804 COPY_ARRAY(renames
->callback_data
[renames
->callback_data_nr
].names
,
806 renames
->callback_data_nr
++;
812 * Much like traverse_trees(), BUT:
813 * - read all the tree entries FIRST, saving them
814 * - note that the above step provides an opportunity to compute necessary
815 * additional details before the "real" traversal
816 * - loop through the saved entries and call the original callback on them
818 static int traverse_trees_wrapper(struct index_state
*istate
,
821 struct traverse_info
*info
)
823 int ret
, i
, old_offset
;
824 traverse_callback_t old_fn
;
825 char *old_callback_data_traverse_path
;
826 struct merge_options
*opt
= info
->data
;
827 struct rename_info
*renames
= &opt
->priv
->renames
;
829 assert(renames
->dir_rename_mask
== 2 || renames
->dir_rename_mask
== 4);
831 old_callback_data_traverse_path
= renames
->callback_data_traverse_path
;
833 old_offset
= renames
->callback_data_nr
;
835 renames
->callback_data_traverse_path
= NULL
;
836 info
->fn
= traverse_trees_wrapper_callback
;
837 ret
= traverse_trees(istate
, n
, t
, info
);
841 info
->traverse_path
= renames
->callback_data_traverse_path
;
843 for (i
= old_offset
; i
< renames
->callback_data_nr
; ++i
) {
845 renames
->callback_data
[i
].mask
,
846 renames
->callback_data
[i
].dirmask
,
847 renames
->callback_data
[i
].names
,
851 renames
->callback_data_nr
= old_offset
;
852 free(renames
->callback_data_traverse_path
);
853 renames
->callback_data_traverse_path
= old_callback_data_traverse_path
;
854 info
->traverse_path
= NULL
;
858 static void setup_path_info(struct merge_options
*opt
,
859 struct string_list_item
*result
,
860 const char *current_dir_name
,
861 int current_dir_name_len
,
862 char *fullpath
, /* we'll take over ownership */
863 struct name_entry
*names
,
864 struct name_entry
*merged_version
,
865 unsigned is_null
, /* boolean */
866 unsigned df_conflict
, /* boolean */
869 int resolved
/* boolean */)
871 /* result->util is void*, so mi is a convenience typed variable */
872 struct merged_info
*mi
;
874 assert(!is_null
|| resolved
);
875 assert(!df_conflict
|| !resolved
); /* df_conflict implies !resolved */
876 assert(resolved
== (merged_version
!= NULL
));
878 mi
= pool_calloc(opt
->priv
->pool
, 1,
879 resolved
? sizeof(struct merged_info
) :
880 sizeof(struct conflict_info
));
881 mi
->directory_name
= current_dir_name
;
882 mi
->basename_offset
= current_dir_name_len
;
883 mi
->clean
= !!resolved
;
885 mi
->result
.mode
= merged_version
->mode
;
886 oidcpy(&mi
->result
.oid
, &merged_version
->oid
);
887 mi
->is_null
= !!is_null
;
890 struct conflict_info
*ci
;
892 ASSIGN_AND_VERIFY_CI(ci
, mi
);
893 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
894 ci
->pathnames
[i
] = fullpath
;
895 ci
->stages
[i
].mode
= names
[i
].mode
;
896 oidcpy(&ci
->stages
[i
].oid
, &names
[i
].oid
);
898 ci
->filemask
= filemask
;
899 ci
->dirmask
= dirmask
;
900 ci
->df_conflict
= !!df_conflict
;
903 * Assume is_null for now, but if we have entries
904 * under the directory then when it is complete in
905 * write_completed_directory() it'll update this.
906 * Also, for D/F conflicts, we have to handle the
907 * directory first, then clear this bit and process
908 * the file to see how it is handled -- that occurs
909 * near the top of process_entry().
913 strmap_put(&opt
->priv
->paths
, fullpath
, mi
);
914 result
->string
= fullpath
;
918 static void add_pair(struct merge_options
*opt
,
919 struct name_entry
*names
,
920 const char *pathname
,
922 unsigned is_add
/* if false, is_delete */,
924 unsigned dir_rename_mask
)
926 struct diff_filespec
*one
, *two
;
927 struct mem_pool
*pool
= opt
->priv
->pool
;
928 struct rename_info
*renames
= &opt
->priv
->renames
;
929 int names_idx
= is_add
? side
: 0;
932 assert(match_mask
== 0 || match_mask
== 6);
933 if (strset_contains(&renames
->cached_target_names
[side
],
937 unsigned content_relevant
= (match_mask
== 0);
938 unsigned location_relevant
= (dir_rename_mask
== 0x07);
940 assert(match_mask
== 0 || match_mask
== 3 || match_mask
== 5);
943 * If pathname is found in cached_irrelevant[side] due to
944 * previous pick but for this commit content is relevant,
945 * then we need to remove it from cached_irrelevant.
947 if (content_relevant
)
948 /* strset_remove is no-op if strset doesn't have key */
949 strset_remove(&renames
->cached_irrelevant
[side
],
953 * We do not need to re-detect renames for paths that we already
954 * know the pairing, i.e. for cached_pairs (or
955 * cached_irrelevant). However, handle_deferred_entries() needs
956 * to loop over the union of keys from relevant_sources[side] and
957 * cached_pairs[side], so for simplicity we set relevant_sources
958 * for all the cached_pairs too and then strip them back out in
959 * prune_cached_from_relevant() at the beginning of
960 * detect_regular_renames().
962 if (content_relevant
|| location_relevant
) {
963 /* content_relevant trumps location_relevant */
964 strintmap_set(&renames
->relevant_sources
[side
], pathname
,
965 content_relevant
? RELEVANT_CONTENT
: RELEVANT_LOCATION
);
969 * Avoid creating pair if we've already cached rename results.
970 * Note that we do this after setting relevant_sources[side]
971 * as noted in the comment above.
973 if (strmap_contains(&renames
->cached_pairs
[side
], pathname
) ||
974 strset_contains(&renames
->cached_irrelevant
[side
], pathname
))
978 one
= pool_alloc_filespec(pool
, pathname
);
979 two
= pool_alloc_filespec(pool
, pathname
);
980 fill_filespec(is_add
? two
: one
,
981 &names
[names_idx
].oid
, 1, names
[names_idx
].mode
);
982 pool_diff_queue(pool
, &renames
->pairs
[side
], one
, two
);
985 static void collect_rename_info(struct merge_options
*opt
,
986 struct name_entry
*names
,
988 const char *fullname
,
993 struct rename_info
*renames
= &opt
->priv
->renames
;
997 * Update dir_rename_mask (determines ignore-rename-source validity)
999 * dir_rename_mask helps us keep track of when directory rename
1000 * detection may be relevant. Basically, whenver a directory is
1001 * removed on one side of history, and a file is added to that
1002 * directory on the other side of history, directory rename
1003 * detection is relevant (meaning we have to detect renames for all
1004 * files within that directory to deduce where the directory
1005 * moved). Also, whenever a directory needs directory rename
1006 * detection, due to the "majority rules" choice for where to move
1007 * it (see t6423 testcase 1f), we also need to detect renames for
1008 * all files within subdirectories of that directory as well.
1010 * Here we haven't looked at files within the directory yet, we are
1011 * just looking at the directory itself. So, if we aren't yet in
1012 * a case where a parent directory needed directory rename detection
1013 * (i.e. dir_rename_mask != 0x07), and if the directory was removed
1014 * on one side of history, record the mask of the other side of
1015 * history in dir_rename_mask.
1017 if (renames
->dir_rename_mask
!= 0x07 &&
1018 (dirmask
== 3 || dirmask
== 5)) {
1019 /* simple sanity check */
1020 assert(renames
->dir_rename_mask
== 0 ||
1021 renames
->dir_rename_mask
== (dirmask
& ~1));
1022 /* update dir_rename_mask; have it record mask of new side */
1023 renames
->dir_rename_mask
= (dirmask
& ~1);
1026 /* Update dirs_removed, as needed */
1027 if (dirmask
== 1 || dirmask
== 3 || dirmask
== 5) {
1028 /* absent_mask = 0x07 - dirmask; sides = absent_mask/2 */
1029 unsigned sides
= (0x07 - dirmask
)/2;
1030 unsigned relevance
= (renames
->dir_rename_mask
== 0x07) ?
1031 RELEVANT_FOR_ANCESTOR
: NOT_RELEVANT
;
1033 * Record relevance of this directory. However, note that
1034 * when collect_merge_info_callback() recurses into this
1035 * directory and calls collect_rename_info() on paths
1036 * within that directory, if we find a path that was added
1037 * to this directory on the other side of history, we will
1038 * upgrade this value to RELEVANT_FOR_SELF; see below.
1041 strintmap_set(&renames
->dirs_removed
[1], fullname
,
1044 strintmap_set(&renames
->dirs_removed
[2], fullname
,
1049 * Here's the block that potentially upgrades to RELEVANT_FOR_SELF.
1050 * When we run across a file added to a directory. In such a case,
1051 * find the directory of the file and upgrade its relevance.
1053 if (renames
->dir_rename_mask
== 0x07 &&
1054 (filemask
== 2 || filemask
== 4)) {
1056 * Need directory rename for parent directory on other side
1057 * of history from added file. Thus
1058 * side = (~filemask & 0x06) >> 1
1060 * side = 3 - (filemask/2).
1062 unsigned side
= 3 - (filemask
>> 1);
1063 strintmap_set(&renames
->dirs_removed
[side
], dirname
,
1067 if (filemask
== 0 || filemask
== 7)
1070 for (side
= MERGE_SIDE1
; side
<= MERGE_SIDE2
; ++side
) {
1071 unsigned side_mask
= (1 << side
);
1073 /* Check for deletion on side */
1074 if ((filemask
& 1) && !(filemask
& side_mask
))
1075 add_pair(opt
, names
, fullname
, side
, 0 /* delete */,
1076 match_mask
& filemask
,
1077 renames
->dir_rename_mask
);
1079 /* Check for addition on side */
1080 if (!(filemask
& 1) && (filemask
& side_mask
))
1081 add_pair(opt
, names
, fullname
, side
, 1 /* add */,
1082 match_mask
& filemask
,
1083 renames
->dir_rename_mask
);
1087 static int collect_merge_info_callback(int n
,
1089 unsigned long dirmask
,
1090 struct name_entry
*names
,
1091 struct traverse_info
*info
)
1095 * common ancestor (mbase) has mask 1, and stored in index 0 of names
1096 * head of side 1 (side1) has mask 2, and stored in index 1 of names
1097 * head of side 2 (side2) has mask 4, and stored in index 2 of names
1099 struct merge_options
*opt
= info
->data
;
1100 struct merge_options_internal
*opti
= opt
->priv
;
1101 struct rename_info
*renames
= &opt
->priv
->renames
;
1102 struct string_list_item pi
; /* Path Info */
1103 struct conflict_info
*ci
; /* typed alias to pi.util (which is void*) */
1104 struct name_entry
*p
;
1107 const char *dirname
= opti
->current_dir_name
;
1108 unsigned prev_dir_rename_mask
= renames
->dir_rename_mask
;
1109 unsigned filemask
= mask
& ~dirmask
;
1110 unsigned match_mask
= 0; /* will be updated below */
1111 unsigned mbase_null
= !(mask
& 1);
1112 unsigned side1_null
= !(mask
& 2);
1113 unsigned side2_null
= !(mask
& 4);
1114 unsigned side1_matches_mbase
= (!side1_null
&& !mbase_null
&&
1115 names
[0].mode
== names
[1].mode
&&
1116 oideq(&names
[0].oid
, &names
[1].oid
));
1117 unsigned side2_matches_mbase
= (!side2_null
&& !mbase_null
&&
1118 names
[0].mode
== names
[2].mode
&&
1119 oideq(&names
[0].oid
, &names
[2].oid
));
1120 unsigned sides_match
= (!side1_null
&& !side2_null
&&
1121 names
[1].mode
== names
[2].mode
&&
1122 oideq(&names
[1].oid
, &names
[2].oid
));
1125 * Note: When a path is a file on one side of history and a directory
1126 * in another, we have a directory/file conflict. In such cases, if
1127 * the conflict doesn't resolve from renames and deletions, then we
1128 * always leave directories where they are and move files out of the
1129 * way. Thus, while struct conflict_info has a df_conflict field to
1130 * track such conflicts, we ignore that field for any directories at
1131 * a path and only pay attention to it for files at the given path.
1132 * The fact that we leave directories were they are also means that
1133 * we do not need to worry about getting additional df_conflict
1134 * information propagated from parent directories down to children
1135 * (unlike, say traverse_trees_recursive() in unpack-trees.c, which
1136 * sets a newinfo.df_conflicts field specifically to propagate it).
1138 unsigned df_conflict
= (filemask
!= 0) && (dirmask
!= 0);
1140 /* n = 3 is a fundamental assumption. */
1142 BUG("Called collect_merge_info_callback wrong");
1145 * A bunch of sanity checks verifying that traverse_trees() calls
1146 * us the way I expect. Could just remove these at some point,
1147 * though maybe they are helpful to future code readers.
1149 assert(mbase_null
== is_null_oid(&names
[0].oid
));
1150 assert(side1_null
== is_null_oid(&names
[1].oid
));
1151 assert(side2_null
== is_null_oid(&names
[2].oid
));
1152 assert(!mbase_null
|| !side1_null
|| !side2_null
);
1153 assert(mask
> 0 && mask
< 8);
1155 /* Determine match_mask */
1156 if (side1_matches_mbase
)
1157 match_mask
= (side2_matches_mbase
? 7 : 3);
1158 else if (side2_matches_mbase
)
1160 else if (sides_match
)
1164 * Get the name of the relevant filepath, which we'll pass to
1165 * setup_path_info() for tracking.
1170 len
= traverse_path_len(info
, p
->pathlen
);
1172 /* +1 in both of the following lines to include the NUL byte */
1173 fullpath
= pool_alloc(opt
->priv
->pool
, len
+ 1);
1174 make_traverse_path(fullpath
, len
+ 1, info
, p
->path
, p
->pathlen
);
1177 * If mbase, side1, and side2 all match, we can resolve early. Even
1178 * if these are trees, there will be no renames or anything
1181 if (side1_matches_mbase
&& side2_matches_mbase
) {
1182 /* mbase, side1, & side2 all match; use mbase as resolution */
1183 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1184 names
, names
+0, mbase_null
, 0 /* df_conflict */,
1185 filemask
, dirmask
, 1 /* resolved */);
1190 * If the sides match, and all three paths are present and are
1191 * files, then we can take either as the resolution. We can't do
1192 * this with trees, because there may be rename sources from the
1195 if (sides_match
&& filemask
== 0x07) {
1196 /* use side1 (== side2) version as resolution */
1197 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1198 names
, names
+1, side1_null
, 0,
1199 filemask
, dirmask
, 1);
1204 * If side1 matches mbase and all three paths are present and are
1205 * files, then we can use side2 as the resolution. We cannot
1206 * necessarily do so this for trees, because there may be rename
1207 * destinations within side2.
1209 if (side1_matches_mbase
&& filemask
== 0x07) {
1210 /* use side2 version as resolution */
1211 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1212 names
, names
+2, side2_null
, 0,
1213 filemask
, dirmask
, 1);
1217 /* Similar to above but swapping sides 1 and 2 */
1218 if (side2_matches_mbase
&& filemask
== 0x07) {
1219 /* use side1 version as resolution */
1220 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1221 names
, names
+1, side1_null
, 0,
1222 filemask
, dirmask
, 1);
1227 * Sometimes we can tell that a source path need not be included in
1228 * rename detection -- namely, whenever either
1229 * side1_matches_mbase && side2_null
1231 * side2_matches_mbase && side1_null
1232 * However, we call collect_rename_info() even in those cases,
1233 * because exact renames are cheap and would let us remove both a
1234 * source and destination path. We'll cull the unneeded sources
1237 collect_rename_info(opt
, names
, dirname
, fullpath
,
1238 filemask
, dirmask
, match_mask
);
1241 * None of the special cases above matched, so we have a
1242 * provisional conflict. (Rename detection might allow us to
1243 * unconflict some more cases, but that comes later so all we can
1244 * do now is record the different non-null file hashes.)
1246 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1247 names
, NULL
, 0, df_conflict
, filemask
, dirmask
, 0);
1251 ci
->match_mask
= match_mask
;
1253 /* If dirmask, recurse into subdirectories */
1255 struct traverse_info newinfo
;
1256 struct tree_desc t
[3];
1257 void *buf
[3] = {NULL
, NULL
, NULL
};
1258 const char *original_dir_name
;
1262 * Check for whether we can avoid recursing due to one side
1263 * matching the merge base. The side that does NOT match is
1264 * the one that might have a rename destination we need.
1266 assert(!side1_matches_mbase
|| !side2_matches_mbase
);
1267 side
= side1_matches_mbase
? MERGE_SIDE2
:
1268 side2_matches_mbase
? MERGE_SIDE1
: MERGE_BASE
;
1269 if (filemask
== 0 && (dirmask
== 2 || dirmask
== 4)) {
1271 * Also defer recursing into new directories; set up a
1272 * few variables to let us do so.
1274 ci
->match_mask
= (7 - dirmask
);
1277 if (renames
->dir_rename_mask
!= 0x07 &&
1278 side
!= MERGE_BASE
&&
1279 renames
->deferred
[side
].trivial_merges_okay
&&
1280 !strset_contains(&renames
->deferred
[side
].target_dirs
,
1282 strintmap_set(&renames
->deferred
[side
].possible_trivial_merges
,
1283 pi
.string
, renames
->dir_rename_mask
);
1284 renames
->dir_rename_mask
= prev_dir_rename_mask
;
1288 /* We need to recurse */
1289 ci
->match_mask
&= filemask
;
1291 newinfo
.prev
= info
;
1292 newinfo
.name
= p
->path
;
1293 newinfo
.namelen
= p
->pathlen
;
1294 newinfo
.pathlen
= st_add3(newinfo
.pathlen
, p
->pathlen
, 1);
1296 * If this directory we are about to recurse into cared about
1297 * its parent directory (the current directory) having a D/F
1298 * conflict, then we'd propagate the masks in this way:
1299 * newinfo.df_conflicts |= (mask & ~dirmask);
1300 * But we don't worry about propagating D/F conflicts. (See
1301 * comment near setting of local df_conflict variable near
1302 * the beginning of this function).
1305 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
1306 if (i
== 1 && side1_matches_mbase
)
1308 else if (i
== 2 && side2_matches_mbase
)
1310 else if (i
== 2 && sides_match
)
1313 const struct object_id
*oid
= NULL
;
1315 oid
= &names
[i
].oid
;
1316 buf
[i
] = fill_tree_descriptor(opt
->repo
,
1322 original_dir_name
= opti
->current_dir_name
;
1323 opti
->current_dir_name
= pi
.string
;
1324 if (renames
->dir_rename_mask
== 0 ||
1325 renames
->dir_rename_mask
== 0x07)
1326 ret
= traverse_trees(NULL
, 3, t
, &newinfo
);
1328 ret
= traverse_trees_wrapper(NULL
, 3, t
, &newinfo
);
1329 opti
->current_dir_name
= original_dir_name
;
1330 renames
->dir_rename_mask
= prev_dir_rename_mask
;
1332 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++)
1342 static void resolve_trivial_directory_merge(struct conflict_info
*ci
, int side
)
1345 assert((side
== 1 && ci
->match_mask
== 5) ||
1346 (side
== 2 && ci
->match_mask
== 3));
1347 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[side
].oid
);
1348 ci
->merged
.result
.mode
= ci
->stages
[side
].mode
;
1349 ci
->merged
.is_null
= is_null_oid(&ci
->stages
[side
].oid
);
1351 ci
->merged
.clean
= 1; /* (ci->filemask == 0); */
1354 static int handle_deferred_entries(struct merge_options
*opt
,
1355 struct traverse_info
*info
)
1357 struct rename_info
*renames
= &opt
->priv
->renames
;
1358 struct hashmap_iter iter
;
1359 struct strmap_entry
*entry
;
1361 int path_count_before
, path_count_after
= 0;
1363 path_count_before
= strmap_get_size(&opt
->priv
->paths
);
1364 for (side
= MERGE_SIDE1
; side
<= MERGE_SIDE2
; side
++) {
1365 unsigned optimization_okay
= 1;
1366 struct strintmap copy
;
1368 /* Loop over the set of paths we need to know rename info for */
1369 strset_for_each_entry(&renames
->relevant_sources
[side
],
1371 char *rename_target
, *dir
, *dir_marker
;
1372 struct strmap_entry
*e
;
1375 * If we don't know delete/rename info for this path,
1376 * then we need to recurse into all trees to get all
1377 * adds to make sure we have it.
1379 if (strset_contains(&renames
->cached_irrelevant
[side
],
1382 e
= strmap_get_entry(&renames
->cached_pairs
[side
],
1385 optimization_okay
= 0;
1389 /* If this is a delete, we have enough info already */
1390 rename_target
= e
->value
;
1394 /* If we already walked the rename target, we're good */
1395 if (strmap_contains(&opt
->priv
->paths
, rename_target
))
1399 * Otherwise, we need to get a list of directories that
1400 * will need to be recursed into to get this
1403 dir
= xstrdup(rename_target
);
1404 while ((dir_marker
= strrchr(dir
, '/'))) {
1406 if (strset_contains(&renames
->deferred
[side
].target_dirs
,
1409 strset_add(&renames
->deferred
[side
].target_dirs
,
1414 renames
->deferred
[side
].trivial_merges_okay
= optimization_okay
;
1416 * We need to recurse into any directories in
1417 * possible_trivial_merges[side] found in target_dirs[side].
1418 * But when we recurse, we may need to queue up some of the
1419 * subdirectories for possible_trivial_merges[side]. Since
1420 * we can't safely iterate through a hashmap while also adding
1421 * entries, move the entries into 'copy', iterate over 'copy',
1422 * and then we'll also iterate anything added into
1423 * possible_trivial_merges[side] once this loop is done.
1425 copy
= renames
->deferred
[side
].possible_trivial_merges
;
1426 strintmap_init_with_options(&renames
->deferred
[side
].possible_trivial_merges
,
1430 strintmap_for_each_entry(©
, &iter
, entry
) {
1431 const char *path
= entry
->key
;
1432 unsigned dir_rename_mask
= (intptr_t)entry
->value
;
1433 struct conflict_info
*ci
;
1435 struct tree_desc t
[3];
1436 void *buf
[3] = {NULL
,};
1439 ci
= strmap_get(&opt
->priv
->paths
, path
);
1441 dirmask
= ci
->dirmask
;
1443 if (optimization_okay
&&
1444 !strset_contains(&renames
->deferred
[side
].target_dirs
,
1446 resolve_trivial_directory_merge(ci
, side
);
1451 info
->namelen
= strlen(path
);
1452 info
->pathlen
= info
->namelen
+ 1;
1454 for (i
= 0; i
< 3; i
++, dirmask
>>= 1) {
1455 if (i
== 1 && ci
->match_mask
== 3)
1457 else if (i
== 2 && ci
->match_mask
== 5)
1459 else if (i
== 2 && ci
->match_mask
== 6)
1462 const struct object_id
*oid
= NULL
;
1464 oid
= &ci
->stages
[i
].oid
;
1465 buf
[i
] = fill_tree_descriptor(opt
->repo
,
1470 ci
->match_mask
&= ci
->filemask
;
1471 opt
->priv
->current_dir_name
= path
;
1472 renames
->dir_rename_mask
= dir_rename_mask
;
1473 if (renames
->dir_rename_mask
== 0 ||
1474 renames
->dir_rename_mask
== 0x07)
1475 ret
= traverse_trees(NULL
, 3, t
, info
);
1477 ret
= traverse_trees_wrapper(NULL
, 3, t
, info
);
1479 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++)
1485 strintmap_clear(©
);
1486 strintmap_for_each_entry(&renames
->deferred
[side
].possible_trivial_merges
,
1488 const char *path
= entry
->key
;
1489 struct conflict_info
*ci
;
1491 ci
= strmap_get(&opt
->priv
->paths
, path
);
1494 assert(renames
->deferred
[side
].trivial_merges_okay
&&
1495 !strset_contains(&renames
->deferred
[side
].target_dirs
,
1497 resolve_trivial_directory_merge(ci
, side
);
1499 if (!optimization_okay
|| path_count_after
)
1500 path_count_after
= strmap_get_size(&opt
->priv
->paths
);
1502 if (path_count_after
) {
1504 * The choice of wanted_factor here does not affect
1505 * correctness, only performance. When the
1506 * path_count_after / path_count_before
1507 * ratio is high, redoing after renames is a big
1508 * performance boost. I suspect that redoing is a wash
1509 * somewhere near a value of 2, and below that redoing will
1510 * slow things down. I applied a fudge factor and picked
1511 * 3; see the commit message when this was introduced for
1512 * back of the envelope calculations for this ratio.
1514 const int wanted_factor
= 3;
1516 /* We should only redo collect_merge_info one time */
1517 assert(renames
->redo_after_renames
== 0);
1519 if (path_count_after
/ path_count_before
>= wanted_factor
) {
1520 renames
->redo_after_renames
= 1;
1521 renames
->cached_pairs_valid_side
= -1;
1523 } else if (renames
->redo_after_renames
== 2)
1524 renames
->redo_after_renames
= 0;
1528 static int collect_merge_info(struct merge_options
*opt
,
1529 struct tree
*merge_base
,
1534 struct tree_desc t
[3];
1535 struct traverse_info info
;
1537 opt
->priv
->toplevel_dir
= "";
1538 opt
->priv
->current_dir_name
= opt
->priv
->toplevel_dir
;
1539 setup_traverse_info(&info
, opt
->priv
->toplevel_dir
);
1540 info
.fn
= collect_merge_info_callback
;
1542 info
.show_all_errors
= 1;
1544 parse_tree(merge_base
);
1547 init_tree_desc(t
+ 0, merge_base
->buffer
, merge_base
->size
);
1548 init_tree_desc(t
+ 1, side1
->buffer
, side1
->size
);
1549 init_tree_desc(t
+ 2, side2
->buffer
, side2
->size
);
1551 trace2_region_enter("merge", "traverse_trees", opt
->repo
);
1552 ret
= traverse_trees(NULL
, 3, t
, &info
);
1554 ret
= handle_deferred_entries(opt
, &info
);
1555 trace2_region_leave("merge", "traverse_trees", opt
->repo
);
1560 /*** Function Grouping: functions related to threeway content merges ***/
1562 static int find_first_merges(struct repository
*repo
,
1566 struct object_array
*result
)
1569 struct object_array merges
= OBJECT_ARRAY_INIT
;
1570 struct commit
*commit
;
1571 int contains_another
;
1573 char merged_revision
[GIT_MAX_HEXSZ
+ 2];
1574 const char *rev_args
[] = { "rev-list", "--merges", "--ancestry-path",
1575 "--all", merged_revision
, NULL
};
1576 struct rev_info revs
;
1577 struct setup_revision_opt rev_opts
;
1579 memset(result
, 0, sizeof(struct object_array
));
1580 memset(&rev_opts
, 0, sizeof(rev_opts
));
1582 /* get all revisions that merge commit a */
1583 xsnprintf(merged_revision
, sizeof(merged_revision
), "^%s",
1584 oid_to_hex(&a
->object
.oid
));
1585 repo_init_revisions(repo
, &revs
, NULL
);
1586 rev_opts
.submodule
= path
;
1587 /* FIXME: can't handle linked worktrees in submodules yet */
1588 revs
.single_worktree
= path
!= NULL
;
1589 setup_revisions(ARRAY_SIZE(rev_args
)-1, rev_args
, &revs
, &rev_opts
);
1591 /* save all revisions from the above list that contain b */
1592 if (prepare_revision_walk(&revs
))
1593 die("revision walk setup failed");
1594 while ((commit
= get_revision(&revs
)) != NULL
) {
1595 struct object
*o
= &(commit
->object
);
1596 if (in_merge_bases(b
, commit
))
1597 add_object_array(o
, NULL
, &merges
);
1599 reset_revision_walk();
1601 /* Now we've got all merges that contain a and b. Prune all
1602 * merges that contain another found merge and save them in
1605 for (i
= 0; i
< merges
.nr
; i
++) {
1606 struct commit
*m1
= (struct commit
*) merges
.objects
[i
].item
;
1608 contains_another
= 0;
1609 for (j
= 0; j
< merges
.nr
; j
++) {
1610 struct commit
*m2
= (struct commit
*) merges
.objects
[j
].item
;
1611 if (i
!= j
&& in_merge_bases(m2
, m1
)) {
1612 contains_another
= 1;
1617 if (!contains_another
)
1618 add_object_array(merges
.objects
[i
].item
, NULL
, result
);
1621 object_array_clear(&merges
);
1625 static int merge_submodule(struct merge_options
*opt
,
1627 const struct object_id
*o
,
1628 const struct object_id
*a
,
1629 const struct object_id
*b
,
1630 struct object_id
*result
)
1632 struct commit
*commit_o
, *commit_a
, *commit_b
;
1634 struct object_array merges
;
1635 struct strbuf sb
= STRBUF_INIT
;
1638 int search
= !opt
->priv
->call_depth
;
1640 /* store fallback answer in result in case we fail */
1641 oidcpy(result
, opt
->priv
->call_depth
? o
: a
);
1643 /* we can not handle deletion conflicts */
1651 if (add_submodule_odb(path
)) {
1652 path_msg(opt
, path
, 0,
1653 _("Failed to merge submodule %s (not checked out)"),
1658 if (!(commit_o
= lookup_commit_reference(opt
->repo
, o
)) ||
1659 !(commit_a
= lookup_commit_reference(opt
->repo
, a
)) ||
1660 !(commit_b
= lookup_commit_reference(opt
->repo
, b
))) {
1661 path_msg(opt
, path
, 0,
1662 _("Failed to merge submodule %s (commits not present)"),
1667 /* check whether both changes are forward */
1668 if (!in_merge_bases(commit_o
, commit_a
) ||
1669 !in_merge_bases(commit_o
, commit_b
)) {
1670 path_msg(opt
, path
, 0,
1671 _("Failed to merge submodule %s "
1672 "(commits don't follow merge-base)"),
1677 /* Case #1: a is contained in b or vice versa */
1678 if (in_merge_bases(commit_a
, commit_b
)) {
1680 path_msg(opt
, path
, 1,
1681 _("Note: Fast-forwarding submodule %s to %s"),
1682 path
, oid_to_hex(b
));
1685 if (in_merge_bases(commit_b
, commit_a
)) {
1687 path_msg(opt
, path
, 1,
1688 _("Note: Fast-forwarding submodule %s to %s"),
1689 path
, oid_to_hex(a
));
1694 * Case #2: There are one or more merges that contain a and b in
1695 * the submodule. If there is only one, then present it as a
1696 * suggestion to the user, but leave it marked unmerged so the
1697 * user needs to confirm the resolution.
1700 /* Skip the search if makes no sense to the calling context. */
1704 /* find commit which merges them */
1705 parent_count
= find_first_merges(opt
->repo
, path
, commit_a
, commit_b
,
1707 switch (parent_count
) {
1709 path_msg(opt
, path
, 0, _("Failed to merge submodule %s"), path
);
1713 format_commit(&sb
, 4,
1714 (struct commit
*)merges
.objects
[0].item
);
1715 path_msg(opt
, path
, 0,
1716 _("Failed to merge submodule %s, but a possible merge "
1717 "resolution exists:\n%s\n"),
1719 path_msg(opt
, path
, 1,
1720 _("If this is correct simply add it to the index "
1723 " git update-index --cacheinfo 160000 %s \"%s\"\n\n"
1724 "which will accept this suggestion.\n"),
1725 oid_to_hex(&merges
.objects
[0].item
->oid
), path
);
1726 strbuf_release(&sb
);
1729 for (i
= 0; i
< merges
.nr
; i
++)
1730 format_commit(&sb
, 4,
1731 (struct commit
*)merges
.objects
[i
].item
);
1732 path_msg(opt
, path
, 0,
1733 _("Failed to merge submodule %s, but multiple "
1734 "possible merges exist:\n%s"), path
, sb
.buf
);
1735 strbuf_release(&sb
);
1738 object_array_clear(&merges
);
1742 static void initialize_attr_index(struct merge_options
*opt
)
1745 * The renormalize_buffer() functions require attributes, and
1746 * annoyingly those can only be read from the working tree or from
1747 * an index_state. merge-ort doesn't have an index_state, so we
1748 * generate a fake one containing only attribute information.
1750 struct merged_info
*mi
;
1751 struct index_state
*attr_index
= &opt
->priv
->attr_index
;
1752 struct cache_entry
*ce
;
1754 attr_index
->initialized
= 1;
1756 if (!opt
->renormalize
)
1759 mi
= strmap_get(&opt
->priv
->paths
, GITATTRIBUTES_FILE
);
1764 int len
= strlen(GITATTRIBUTES_FILE
);
1765 ce
= make_empty_cache_entry(attr_index
, len
);
1766 ce
->ce_mode
= create_ce_mode(mi
->result
.mode
);
1767 ce
->ce_flags
= create_ce_flags(0);
1768 ce
->ce_namelen
= len
;
1769 oidcpy(&ce
->oid
, &mi
->result
.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
, &ce
->oid
);
1776 struct conflict_info
*ci
;
1778 ASSIGN_AND_VERIFY_CI(ci
, mi
);
1779 for (stage
= 0; stage
< 3; stage
++) {
1780 unsigned stage_mask
= (1 << stage
);
1782 if (!(ci
->filemask
& stage_mask
))
1784 len
= strlen(GITATTRIBUTES_FILE
);
1785 ce
= make_empty_cache_entry(attr_index
, len
);
1786 ce
->ce_mode
= create_ce_mode(ci
->stages
[stage
].mode
);
1787 ce
->ce_flags
= create_ce_flags(stage
);
1788 ce
->ce_namelen
= len
;
1789 oidcpy(&ce
->oid
, &ci
->stages
[stage
].oid
);
1790 memcpy(ce
->name
, GITATTRIBUTES_FILE
, len
);
1791 add_index_entry(attr_index
, ce
,
1792 ADD_CACHE_OK_TO_ADD
| ADD_CACHE_OK_TO_REPLACE
);
1793 get_stream_filter(attr_index
, GITATTRIBUTES_FILE
,
1799 static int merge_3way(struct merge_options
*opt
,
1801 const struct object_id
*o
,
1802 const struct object_id
*a
,
1803 const struct object_id
*b
,
1804 const char *pathnames
[3],
1805 const int extra_marker_size
,
1806 mmbuffer_t
*result_buf
)
1808 mmfile_t orig
, src1
, src2
;
1809 struct ll_merge_options ll_opts
= {0};
1810 char *base
, *name1
, *name2
;
1813 if (!opt
->priv
->attr_index
.initialized
)
1814 initialize_attr_index(opt
);
1816 ll_opts
.renormalize
= opt
->renormalize
;
1817 ll_opts
.extra_marker_size
= extra_marker_size
;
1818 ll_opts
.xdl_opts
= opt
->xdl_opts
;
1820 if (opt
->priv
->call_depth
) {
1821 ll_opts
.virtual_ancestor
= 1;
1822 ll_opts
.variant
= 0;
1824 switch (opt
->recursive_variant
) {
1825 case MERGE_VARIANT_OURS
:
1826 ll_opts
.variant
= XDL_MERGE_FAVOR_OURS
;
1828 case MERGE_VARIANT_THEIRS
:
1829 ll_opts
.variant
= XDL_MERGE_FAVOR_THEIRS
;
1832 ll_opts
.variant
= 0;
1837 assert(pathnames
[0] && pathnames
[1] && pathnames
[2] && opt
->ancestor
);
1838 if (pathnames
[0] == pathnames
[1] && pathnames
[1] == pathnames
[2]) {
1839 base
= mkpathdup("%s", opt
->ancestor
);
1840 name1
= mkpathdup("%s", opt
->branch1
);
1841 name2
= mkpathdup("%s", opt
->branch2
);
1843 base
= mkpathdup("%s:%s", opt
->ancestor
, pathnames
[0]);
1844 name1
= mkpathdup("%s:%s", opt
->branch1
, pathnames
[1]);
1845 name2
= mkpathdup("%s:%s", opt
->branch2
, pathnames
[2]);
1848 read_mmblob(&orig
, o
);
1849 read_mmblob(&src1
, a
);
1850 read_mmblob(&src2
, b
);
1852 merge_status
= ll_merge(result_buf
, path
, &orig
, base
,
1853 &src1
, name1
, &src2
, name2
,
1854 &opt
->priv
->attr_index
, &ll_opts
);
1862 return merge_status
;
1865 static int handle_content_merge(struct merge_options
*opt
,
1867 const struct version_info
*o
,
1868 const struct version_info
*a
,
1869 const struct version_info
*b
,
1870 const char *pathnames
[3],
1871 const int extra_marker_size
,
1872 struct version_info
*result
)
1875 * path is the target location where we want to put the file, and
1876 * is used to determine any normalization rules in ll_merge.
1878 * The normal case is that path and all entries in pathnames are
1879 * identical, though renames can affect which path we got one of
1880 * the three blobs to merge on various sides of history.
1882 * extra_marker_size is the amount to extend conflict markers in
1883 * ll_merge; this is neeed if we have content merges of content
1884 * merges, which happens for example with rename/rename(2to1) and
1885 * rename/add conflicts.
1890 * handle_content_merge() needs both files to be of the same type, i.e.
1891 * both files OR both submodules OR both symlinks. Conflicting types
1892 * needs to be handled elsewhere.
1894 assert((S_IFMT
& a
->mode
) == (S_IFMT
& b
->mode
));
1897 if (a
->mode
== b
->mode
|| a
->mode
== o
->mode
)
1898 result
->mode
= b
->mode
;
1900 /* must be the 100644/100755 case */
1901 assert(S_ISREG(a
->mode
));
1902 result
->mode
= a
->mode
;
1903 clean
= (b
->mode
== o
->mode
);
1905 * FIXME: If opt->priv->call_depth && !clean, then we really
1906 * should not make result->mode match either a->mode or
1907 * b->mode; that causes t6036 "check conflicting mode for
1908 * regular file" to fail. It would be best to use some other
1909 * mode, but we'll confuse all kinds of stuff if we use one
1910 * where S_ISREG(result->mode) isn't true, and if we use
1911 * something like 0100666, then tree-walk.c's calls to
1912 * canon_mode() will just normalize that to 100644 for us and
1913 * thus not solve anything.
1915 * Figure out if there's some kind of way we can work around
1921 * Trivial oid merge.
1923 * Note: While one might assume that the next four lines would
1924 * be unnecessary due to the fact that match_mask is often
1925 * setup and already handled, renames don't always take care
1928 if (oideq(&a
->oid
, &b
->oid
) || oideq(&a
->oid
, &o
->oid
))
1929 oidcpy(&result
->oid
, &b
->oid
);
1930 else if (oideq(&b
->oid
, &o
->oid
))
1931 oidcpy(&result
->oid
, &a
->oid
);
1933 /* Remaining rules depend on file vs. submodule vs. symlink. */
1934 else if (S_ISREG(a
->mode
)) {
1935 mmbuffer_t result_buf
;
1936 int ret
= 0, merge_status
;
1940 * If 'o' is different type, treat it as null so we do a
1943 two_way
= ((S_IFMT
& o
->mode
) != (S_IFMT
& a
->mode
));
1945 merge_status
= merge_3way(opt
, path
,
1946 two_way
? null_oid() : &o
->oid
,
1948 pathnames
, extra_marker_size
,
1951 if ((merge_status
< 0) || !result_buf
.ptr
)
1952 ret
= err(opt
, _("Failed to execute internal merge"));
1955 write_object_file(result_buf
.ptr
, result_buf
.size
,
1956 blob_type
, &result
->oid
))
1957 ret
= err(opt
, _("Unable to add %s to database"),
1960 free(result_buf
.ptr
);
1963 clean
&= (merge_status
== 0);
1964 path_msg(opt
, path
, 1, _("Auto-merging %s"), path
);
1965 } else if (S_ISGITLINK(a
->mode
)) {
1966 int two_way
= ((S_IFMT
& o
->mode
) != (S_IFMT
& a
->mode
));
1967 clean
= merge_submodule(opt
, pathnames
[0],
1968 two_way
? null_oid() : &o
->oid
,
1969 &a
->oid
, &b
->oid
, &result
->oid
);
1970 if (opt
->priv
->call_depth
&& two_way
&& !clean
) {
1971 result
->mode
= o
->mode
;
1972 oidcpy(&result
->oid
, &o
->oid
);
1974 } else if (S_ISLNK(a
->mode
)) {
1975 if (opt
->priv
->call_depth
) {
1977 result
->mode
= o
->mode
;
1978 oidcpy(&result
->oid
, &o
->oid
);
1980 switch (opt
->recursive_variant
) {
1981 case MERGE_VARIANT_NORMAL
:
1983 oidcpy(&result
->oid
, &a
->oid
);
1985 case MERGE_VARIANT_OURS
:
1986 oidcpy(&result
->oid
, &a
->oid
);
1988 case MERGE_VARIANT_THEIRS
:
1989 oidcpy(&result
->oid
, &b
->oid
);
1994 BUG("unsupported object type in the tree: %06o for %s",
2000 /*** Function Grouping: functions related to detect_and_process_renames(), ***
2001 *** which are split into directory and regular rename detection sections. ***/
2003 /*** Function Grouping: functions related to directory rename detection ***/
2005 struct collision_info
{
2006 struct string_list source_files
;
2007 unsigned reported_already
:1;
2011 * Return a new string that replaces the beginning portion (which matches
2012 * rename_info->key), with rename_info->util.new_dir. In perl-speak:
2013 * new_path_name = (old_path =~ s/rename_info->key/rename_info->value/);
2015 * Caller must ensure that old_path starts with rename_info->key + '/'.
2017 static char *apply_dir_rename(struct strmap_entry
*rename_info
,
2018 const char *old_path
)
2020 struct strbuf new_path
= STRBUF_INIT
;
2021 const char *old_dir
= rename_info
->key
;
2022 const char *new_dir
= rename_info
->value
;
2023 int oldlen
, newlen
, new_dir_len
;
2025 oldlen
= strlen(old_dir
);
2026 if (*new_dir
== '\0')
2028 * If someone renamed/merged a subdirectory into the root
2029 * directory (e.g. 'some/subdir' -> ''), then we want to
2032 * as the rename; we need to make old_path + oldlen advance
2033 * past the '/' character.
2036 new_dir_len
= strlen(new_dir
);
2037 newlen
= new_dir_len
+ (strlen(old_path
) - oldlen
) + 1;
2038 strbuf_grow(&new_path
, newlen
);
2039 strbuf_add(&new_path
, new_dir
, new_dir_len
);
2040 strbuf_addstr(&new_path
, &old_path
[oldlen
]);
2042 return strbuf_detach(&new_path
, NULL
);
2045 static int path_in_way(struct strmap
*paths
, const char *path
, unsigned side_mask
)
2047 struct merged_info
*mi
= strmap_get(paths
, path
);
2048 struct conflict_info
*ci
;
2051 INITIALIZE_CI(ci
, mi
);
2052 return mi
->clean
|| (side_mask
& (ci
->filemask
| ci
->dirmask
));
2056 * See if there is a directory rename for path, and if there are any file
2057 * level conflicts on the given side for the renamed location. If there is
2058 * a rename and there are no conflicts, return the new name. Otherwise,
2061 static char *handle_path_level_conflicts(struct merge_options
*opt
,
2063 unsigned side_index
,
2064 struct strmap_entry
*rename_info
,
2065 struct strmap
*collisions
)
2067 char *new_path
= NULL
;
2068 struct collision_info
*c_info
;
2070 struct strbuf collision_paths
= STRBUF_INIT
;
2073 * entry has the mapping of old directory name to new directory name
2074 * that we want to apply to path.
2076 new_path
= apply_dir_rename(rename_info
, path
);
2078 BUG("Failed to apply directory rename!");
2081 * The caller needs to have ensured that it has pre-populated
2082 * collisions with all paths that map to new_path. Do a quick check
2083 * to ensure that's the case.
2085 c_info
= strmap_get(collisions
, new_path
);
2087 BUG("c_info is NULL");
2090 * Check for one-sided add/add/.../add conflicts, i.e.
2091 * where implicit renames from the other side doing
2092 * directory rename(s) can affect this side of history
2093 * to put multiple paths into the same location. Warn
2094 * and bail on directory renames for such paths.
2096 if (c_info
->reported_already
) {
2098 } else if (path_in_way(&opt
->priv
->paths
, new_path
, 1 << side_index
)) {
2099 c_info
->reported_already
= 1;
2100 strbuf_add_separated_string_list(&collision_paths
, ", ",
2101 &c_info
->source_files
);
2102 path_msg(opt
, new_path
, 0,
2103 _("CONFLICT (implicit dir rename): Existing file/dir "
2104 "at %s in the way of implicit directory rename(s) "
2105 "putting the following path(s) there: %s."),
2106 new_path
, collision_paths
.buf
);
2108 } else if (c_info
->source_files
.nr
> 1) {
2109 c_info
->reported_already
= 1;
2110 strbuf_add_separated_string_list(&collision_paths
, ", ",
2111 &c_info
->source_files
);
2112 path_msg(opt
, new_path
, 0,
2113 _("CONFLICT (implicit dir rename): Cannot map more "
2114 "than one path to %s; implicit directory renames "
2115 "tried to put these paths there: %s"),
2116 new_path
, collision_paths
.buf
);
2120 /* Free memory we no longer need */
2121 strbuf_release(&collision_paths
);
2122 if (!clean
&& new_path
) {
2130 static void get_provisional_directory_renames(struct merge_options
*opt
,
2134 struct hashmap_iter iter
;
2135 struct strmap_entry
*entry
;
2136 struct rename_info
*renames
= &opt
->priv
->renames
;
2140 * dir_rename_count: old_directory -> {new_directory -> count}
2142 * dir_renames: old_directory -> best_new_directory
2143 * where best_new_directory is the one with the unique highest count.
2145 strmap_for_each_entry(&renames
->dir_rename_count
[side
], &iter
, entry
) {
2146 const char *source_dir
= entry
->key
;
2147 struct strintmap
*counts
= entry
->value
;
2148 struct hashmap_iter count_iter
;
2149 struct strmap_entry
*count_entry
;
2152 const char *best
= NULL
;
2154 strintmap_for_each_entry(counts
, &count_iter
, count_entry
) {
2155 const char *target_dir
= count_entry
->key
;
2156 intptr_t count
= (intptr_t)count_entry
->value
;
2160 else if (count
> max
) {
2169 if (bad_max
== max
) {
2170 path_msg(opt
, source_dir
, 0,
2171 _("CONFLICT (directory rename split): "
2172 "Unclear where to rename %s to; it was "
2173 "renamed to multiple other directories, with "
2174 "no destination getting a majority of the "
2179 strmap_put(&renames
->dir_renames
[side
],
2180 source_dir
, (void*)best
);
2185 static void handle_directory_level_conflicts(struct merge_options
*opt
)
2187 struct hashmap_iter iter
;
2188 struct strmap_entry
*entry
;
2189 struct string_list duplicated
= STRING_LIST_INIT_NODUP
;
2190 struct rename_info
*renames
= &opt
->priv
->renames
;
2191 struct strmap
*side1_dir_renames
= &renames
->dir_renames
[MERGE_SIDE1
];
2192 struct strmap
*side2_dir_renames
= &renames
->dir_renames
[MERGE_SIDE2
];
2195 strmap_for_each_entry(side1_dir_renames
, &iter
, entry
) {
2196 if (strmap_contains(side2_dir_renames
, entry
->key
))
2197 string_list_append(&duplicated
, entry
->key
);
2200 for (i
= 0; i
< duplicated
.nr
; i
++) {
2201 strmap_remove(side1_dir_renames
, duplicated
.items
[i
].string
, 0);
2202 strmap_remove(side2_dir_renames
, duplicated
.items
[i
].string
, 0);
2204 string_list_clear(&duplicated
, 0);
2207 static struct strmap_entry
*check_dir_renamed(const char *path
,
2208 struct strmap
*dir_renames
)
2210 char *temp
= xstrdup(path
);
2212 struct strmap_entry
*e
= NULL
;
2214 while ((end
= strrchr(temp
, '/'))) {
2216 e
= strmap_get_entry(dir_renames
, temp
);
2224 static void compute_collisions(struct strmap
*collisions
,
2225 struct strmap
*dir_renames
,
2226 struct diff_queue_struct
*pairs
)
2230 strmap_init_with_options(collisions
, NULL
, 0);
2231 if (strmap_empty(dir_renames
))
2235 * Multiple files can be mapped to the same path due to directory
2236 * renames done by the other side of history. Since that other
2237 * side of history could have merged multiple directories into one,
2238 * if our side of history added the same file basename to each of
2239 * those directories, then all N of them would get implicitly
2240 * renamed by the directory rename detection into the same path,
2241 * and we'd get an add/add/.../add conflict, and all those adds
2242 * from *this* side of history. This is not representable in the
2243 * index, and users aren't going to easily be able to make sense of
2244 * it. So we need to provide a good warning about what's
2245 * happening, and fall back to no-directory-rename detection
2246 * behavior for those paths.
2248 * See testcases 9e and all of section 5 from t6043 for examples.
2250 for (i
= 0; i
< pairs
->nr
; ++i
) {
2251 struct strmap_entry
*rename_info
;
2252 struct collision_info
*collision_info
;
2254 struct diff_filepair
*pair
= pairs
->queue
[i
];
2256 if (pair
->status
!= 'A' && pair
->status
!= 'R')
2258 rename_info
= check_dir_renamed(pair
->two
->path
, dir_renames
);
2262 new_path
= apply_dir_rename(rename_info
, pair
->two
->path
);
2264 collision_info
= strmap_get(collisions
, new_path
);
2265 if (collision_info
) {
2268 CALLOC_ARRAY(collision_info
, 1);
2269 string_list_init_nodup(&collision_info
->source_files
);
2270 strmap_put(collisions
, new_path
, collision_info
);
2272 string_list_insert(&collision_info
->source_files
,
2277 static char *check_for_directory_rename(struct merge_options
*opt
,
2279 unsigned side_index
,
2280 struct strmap
*dir_renames
,
2281 struct strmap
*dir_rename_exclusions
,
2282 struct strmap
*collisions
,
2285 char *new_path
= NULL
;
2286 struct strmap_entry
*rename_info
;
2287 struct strmap_entry
*otherinfo
= NULL
;
2288 const char *new_dir
;
2290 if (strmap_empty(dir_renames
))
2292 rename_info
= check_dir_renamed(path
, dir_renames
);
2295 /* old_dir = rename_info->key; */
2296 new_dir
= rename_info
->value
;
2299 * This next part is a little weird. We do not want to do an
2300 * implicit rename into a directory we renamed on our side, because
2301 * that will result in a spurious rename/rename(1to2) conflict. An
2303 * Base commit: dumbdir/afile, otherdir/bfile
2304 * Side 1: smrtdir/afile, otherdir/bfile
2305 * Side 2: dumbdir/afile, dumbdir/bfile
2306 * Here, while working on Side 1, we could notice that otherdir was
2307 * renamed/merged to dumbdir, and change the diff_filepair for
2308 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2309 * 2 will notice the rename from dumbdir to smrtdir, and do the
2310 * transitive rename to move it from dumbdir/bfile to
2311 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2312 * smrtdir, a rename/rename(1to2) conflict. We really just want
2313 * the file to end up in smrtdir. And the way to achieve that is
2314 * to not let Side1 do the rename to dumbdir, since we know that is
2315 * the source of one of our directory renames.
2317 * That's why otherinfo and dir_rename_exclusions is here.
2319 * As it turns out, this also prevents N-way transient rename
2320 * confusion; See testcases 9c and 9d of t6043.
2322 otherinfo
= strmap_get_entry(dir_rename_exclusions
, new_dir
);
2324 path_msg(opt
, rename_info
->key
, 1,
2325 _("WARNING: Avoiding applying %s -> %s rename "
2326 "to %s, because %s itself was renamed."),
2327 rename_info
->key
, new_dir
, path
, new_dir
);
2331 new_path
= handle_path_level_conflicts(opt
, path
, side_index
,
2332 rename_info
, collisions
);
2333 *clean_merge
&= (new_path
!= NULL
);
2338 static void apply_directory_rename_modifications(struct merge_options
*opt
,
2339 struct diff_filepair
*pair
,
2343 * The basic idea is to get the conflict_info from opt->priv->paths
2344 * at old path, and insert it into new_path; basically just this:
2345 * ci = strmap_get(&opt->priv->paths, old_path);
2346 * strmap_remove(&opt->priv->paths, old_path, 0);
2347 * strmap_put(&opt->priv->paths, new_path, ci);
2348 * However, there are some factors complicating this:
2349 * - opt->priv->paths may already have an entry at new_path
2350 * - Each ci tracks its containing directory, so we need to
2352 * - If another ci has the same containing directory, then
2353 * the two char*'s MUST point to the same location. See the
2354 * comment in struct merged_info. strcmp equality is not
2355 * enough; we need pointer equality.
2356 * - opt->priv->paths must hold the parent directories of any
2357 * entries that are added. So, if this directory rename
2358 * causes entirely new directories, we must recursively add
2359 * parent directories.
2360 * - For each parent directory added to opt->priv->paths, we
2361 * also need to get its parent directory stored in its
2362 * conflict_info->merged.directory_name with all the same
2363 * requirements about pointer equality.
2365 struct string_list dirs_to_insert
= STRING_LIST_INIT_NODUP
;
2366 struct conflict_info
*ci
, *new_ci
;
2367 struct strmap_entry
*entry
;
2368 const char *branch_with_new_path
, *branch_with_dir_rename
;
2369 const char *old_path
= pair
->two
->path
;
2370 const char *parent_name
;
2371 const char *cur_path
;
2374 entry
= strmap_get_entry(&opt
->priv
->paths
, old_path
);
2375 old_path
= entry
->key
;
2379 /* Find parent directories missing from opt->priv->paths */
2380 if (opt
->priv
->pool
) {
2381 cur_path
= mem_pool_strdup(opt
->priv
->pool
, new_path
);
2382 free((char*)new_path
);
2383 new_path
= (char *)cur_path
;
2385 cur_path
= new_path
;
2389 /* Find the parent directory of cur_path */
2390 char *last_slash
= strrchr(cur_path
, '/');
2392 parent_name
= pool_strndup(opt
->priv
->pool
,
2394 last_slash
- cur_path
);
2396 parent_name
= opt
->priv
->toplevel_dir
;
2400 /* Look it up in opt->priv->paths */
2401 entry
= strmap_get_entry(&opt
->priv
->paths
, parent_name
);
2403 if (!opt
->priv
->pool
)
2404 free((char*)parent_name
);
2405 parent_name
= entry
->key
; /* reuse known pointer */
2409 /* Record this is one of the directories we need to insert */
2410 string_list_append(&dirs_to_insert
, parent_name
);
2411 cur_path
= parent_name
;
2414 /* Traverse dirs_to_insert and insert them into opt->priv->paths */
2415 for (i
= dirs_to_insert
.nr
-1; i
>= 0; --i
) {
2416 struct conflict_info
*dir_ci
;
2417 char *cur_dir
= dirs_to_insert
.items
[i
].string
;
2419 CALLOC_ARRAY(dir_ci
, 1);
2421 dir_ci
->merged
.directory_name
= parent_name
;
2422 len
= strlen(parent_name
);
2423 /* len+1 because of trailing '/' character */
2424 dir_ci
->merged
.basename_offset
= (len
> 0 ? len
+1 : len
);
2425 dir_ci
->dirmask
= ci
->filemask
;
2426 strmap_put(&opt
->priv
->paths
, cur_dir
, dir_ci
);
2428 parent_name
= cur_dir
;
2431 if (!opt
->priv
->pool
) {
2433 * We are removing old_path from opt->priv->paths.
2434 * old_path also will eventually need to be freed, but it
2435 * may still be used by e.g. ci->pathnames. So, store it
2436 * in another string-list for now.
2438 string_list_append(&opt
->priv
->paths_to_free
, old_path
);
2441 assert(ci
->filemask
== 2 || ci
->filemask
== 4);
2442 assert(ci
->dirmask
== 0);
2443 strmap_remove(&opt
->priv
->paths
, old_path
, 0);
2445 branch_with_new_path
= (ci
->filemask
== 2) ? opt
->branch1
: opt
->branch2
;
2446 branch_with_dir_rename
= (ci
->filemask
== 2) ? opt
->branch2
: opt
->branch1
;
2448 /* Now, finally update ci and stick it into opt->priv->paths */
2449 ci
->merged
.directory_name
= parent_name
;
2450 len
= strlen(parent_name
);
2451 ci
->merged
.basename_offset
= (len
> 0 ? len
+1 : len
);
2452 new_ci
= strmap_get(&opt
->priv
->paths
, new_path
);
2454 /* Place ci back into opt->priv->paths, but at new_path */
2455 strmap_put(&opt
->priv
->paths
, new_path
, ci
);
2459 /* A few sanity checks */
2461 assert(ci
->filemask
== 2 || ci
->filemask
== 4);
2462 assert((new_ci
->filemask
& ci
->filemask
) == 0);
2463 assert(!new_ci
->merged
.clean
);
2465 /* Copy stuff from ci into new_ci */
2466 new_ci
->filemask
|= ci
->filemask
;
2467 if (new_ci
->dirmask
)
2468 new_ci
->df_conflict
= 1;
2469 index
= (ci
->filemask
>> 1);
2470 new_ci
->pathnames
[index
] = ci
->pathnames
[index
];
2471 new_ci
->stages
[index
].mode
= ci
->stages
[index
].mode
;
2472 oidcpy(&new_ci
->stages
[index
].oid
, &ci
->stages
[index
].oid
);
2474 if (!opt
->priv
->pool
)
2479 if (opt
->detect_directory_renames
== MERGE_DIRECTORY_RENAMES_TRUE
) {
2480 /* Notify user of updated path */
2481 if (pair
->status
== 'A')
2482 path_msg(opt
, new_path
, 1,
2483 _("Path updated: %s added in %s inside a "
2484 "directory that was renamed in %s; moving "
2486 old_path
, branch_with_new_path
,
2487 branch_with_dir_rename
, new_path
);
2489 path_msg(opt
, new_path
, 1,
2490 _("Path updated: %s renamed to %s in %s, "
2491 "inside a directory that was renamed in %s; "
2492 "moving it to %s."),
2493 pair
->one
->path
, old_path
, branch_with_new_path
,
2494 branch_with_dir_rename
, new_path
);
2497 * opt->detect_directory_renames has the value
2498 * MERGE_DIRECTORY_RENAMES_CONFLICT, so mark these as conflicts.
2500 ci
->path_conflict
= 1;
2501 if (pair
->status
== 'A')
2502 path_msg(opt
, new_path
, 0,
2503 _("CONFLICT (file location): %s added in %s "
2504 "inside a directory that was renamed in %s, "
2505 "suggesting it should perhaps be moved to "
2507 old_path
, branch_with_new_path
,
2508 branch_with_dir_rename
, new_path
);
2510 path_msg(opt
, new_path
, 0,
2511 _("CONFLICT (file location): %s renamed to %s "
2512 "in %s, inside a directory that was renamed "
2513 "in %s, suggesting it should perhaps be "
2515 pair
->one
->path
, old_path
, branch_with_new_path
,
2516 branch_with_dir_rename
, new_path
);
2520 * Finally, record the new location.
2522 pair
->two
->path
= new_path
;
2525 /*** Function Grouping: functions related to regular rename detection ***/
2527 static int process_renames(struct merge_options
*opt
,
2528 struct diff_queue_struct
*renames
)
2530 int clean_merge
= 1, i
;
2532 for (i
= 0; i
< renames
->nr
; ++i
) {
2533 const char *oldpath
= NULL
, *newpath
;
2534 struct diff_filepair
*pair
= renames
->queue
[i
];
2535 struct conflict_info
*oldinfo
= NULL
, *newinfo
= NULL
;
2536 struct strmap_entry
*old_ent
, *new_ent
;
2537 unsigned int old_sidemask
;
2538 int target_index
, other_source_index
;
2539 int source_deleted
, collision
, type_changed
;
2540 const char *rename_branch
= NULL
, *delete_branch
= NULL
;
2542 old_ent
= strmap_get_entry(&opt
->priv
->paths
, pair
->one
->path
);
2543 new_ent
= strmap_get_entry(&opt
->priv
->paths
, pair
->two
->path
);
2545 oldpath
= old_ent
->key
;
2546 oldinfo
= old_ent
->value
;
2548 newpath
= pair
->two
->path
;
2550 newpath
= new_ent
->key
;
2551 newinfo
= new_ent
->value
;
2555 * If pair->one->path isn't in opt->priv->paths, that means
2556 * that either directory rename detection removed that
2557 * path, or a parent directory of oldpath was resolved and
2558 * we don't even need the rename; in either case, we can
2559 * skip it. If oldinfo->merged.clean, then the other side
2560 * of history had no changes to oldpath and we don't need
2561 * the rename and can skip it.
2563 if (!oldinfo
|| oldinfo
->merged
.clean
)
2567 * diff_filepairs have copies of pathnames, thus we have to
2568 * use standard 'strcmp()' (negated) instead of '=='.
2570 if (i
+ 1 < renames
->nr
&&
2571 !strcmp(oldpath
, renames
->queue
[i
+1]->one
->path
)) {
2572 /* Handle rename/rename(1to2) or rename/rename(1to1) */
2573 const char *pathnames
[3];
2574 struct version_info merged
;
2575 struct conflict_info
*base
, *side1
, *side2
;
2576 unsigned was_binary_blob
= 0;
2578 pathnames
[0] = oldpath
;
2579 pathnames
[1] = newpath
;
2580 pathnames
[2] = renames
->queue
[i
+1]->two
->path
;
2582 base
= strmap_get(&opt
->priv
->paths
, pathnames
[0]);
2583 side1
= strmap_get(&opt
->priv
->paths
, pathnames
[1]);
2584 side2
= strmap_get(&opt
->priv
->paths
, pathnames
[2]);
2590 if (!strcmp(pathnames
[1], pathnames
[2])) {
2591 struct rename_info
*ri
= &opt
->priv
->renames
;
2594 /* Both sides renamed the same way */
2595 assert(side1
== side2
);
2596 memcpy(&side1
->stages
[0], &base
->stages
[0],
2598 side1
->filemask
|= (1 << MERGE_BASE
);
2599 /* Mark base as resolved by removal */
2600 base
->merged
.is_null
= 1;
2601 base
->merged
.clean
= 1;
2604 * Disable remembering renames optimization;
2605 * rename/rename(1to1) is incredibly rare, and
2606 * just disabling the optimization is easier
2607 * than purging cached_pairs,
2608 * cached_target_names, and dir_rename_counts.
2610 for (j
= 0; j
< 3; j
++)
2611 ri
->merge_trees
[j
] = NULL
;
2613 /* We handled both renames, i.e. i+1 handled */
2615 /* Move to next rename */
2619 /* This is a rename/rename(1to2) */
2620 clean_merge
= handle_content_merge(opt
,
2626 1 + 2 * opt
->priv
->call_depth
,
2629 merged
.mode
== side1
->stages
[1].mode
&&
2630 oideq(&merged
.oid
, &side1
->stages
[1].oid
))
2631 was_binary_blob
= 1;
2632 memcpy(&side1
->stages
[1], &merged
, sizeof(merged
));
2633 if (was_binary_blob
) {
2635 * Getting here means we were attempting to
2636 * merge a binary blob.
2638 * Since we can't merge binaries,
2639 * handle_content_merge() just takes one
2640 * side. But we don't want to copy the
2641 * contents of one side to both paths. We
2642 * used the contents of side1 above for
2643 * side1->stages, let's use the contents of
2644 * side2 for side2->stages below.
2646 oidcpy(&merged
.oid
, &side2
->stages
[2].oid
);
2647 merged
.mode
= side2
->stages
[2].mode
;
2649 memcpy(&side2
->stages
[2], &merged
, sizeof(merged
));
2651 side1
->path_conflict
= 1;
2652 side2
->path_conflict
= 1;
2654 * TODO: For renames we normally remove the path at the
2655 * old name. It would thus seem consistent to do the
2656 * same for rename/rename(1to2) cases, but we haven't
2657 * done so traditionally and a number of the regression
2658 * tests now encode an expectation that the file is
2659 * left there at stage 1. If we ever decide to change
2660 * this, add the following two lines here:
2661 * base->merged.is_null = 1;
2662 * base->merged.clean = 1;
2663 * and remove the setting of base->path_conflict to 1.
2665 base
->path_conflict
= 1;
2666 path_msg(opt
, oldpath
, 0,
2667 _("CONFLICT (rename/rename): %s renamed to "
2668 "%s in %s and to %s in %s."),
2670 pathnames
[1], opt
->branch1
,
2671 pathnames
[2], opt
->branch2
);
2673 i
++; /* We handled both renames, i.e. i+1 handled */
2679 target_index
= pair
->score
; /* from collect_renames() */
2680 assert(target_index
== 1 || target_index
== 2);
2681 other_source_index
= 3 - target_index
;
2682 old_sidemask
= (1 << other_source_index
); /* 2 or 4 */
2683 source_deleted
= (oldinfo
->filemask
== 1);
2684 collision
= ((newinfo
->filemask
& old_sidemask
) != 0);
2685 type_changed
= !source_deleted
&&
2686 (S_ISREG(oldinfo
->stages
[other_source_index
].mode
) !=
2687 S_ISREG(newinfo
->stages
[target_index
].mode
));
2688 if (type_changed
&& collision
) {
2690 * special handling so later blocks can handle this...
2692 * if type_changed && collision are both true, then this
2693 * was really a double rename, but one side wasn't
2694 * detected due to lack of break detection. I.e.
2696 * orig: has normal file 'foo'
2697 * side1: renames 'foo' to 'bar', adds 'foo' symlink
2698 * side2: renames 'foo' to 'bar'
2699 * In this case, the foo->bar rename on side1 won't be
2700 * detected because the new symlink named 'foo' is
2701 * there and we don't do break detection. But we detect
2702 * this here because we don't want to merge the content
2703 * of the foo symlink with the foo->bar file, so we
2704 * have some logic to handle this special case. The
2705 * easiest way to do that is make 'bar' on side1 not
2706 * be considered a colliding file but the other part
2707 * of a normal rename. If the file is very different,
2708 * well we're going to get content merge conflicts
2709 * anyway so it doesn't hurt. And if the colliding
2710 * file also has a different type, that'll be handled
2711 * by the content merge logic in process_entry() too.
2713 * See also t6430, 'rename vs. rename/symlink'
2717 if (source_deleted
) {
2718 if (target_index
== 1) {
2719 rename_branch
= opt
->branch1
;
2720 delete_branch
= opt
->branch2
;
2722 rename_branch
= opt
->branch2
;
2723 delete_branch
= opt
->branch1
;
2727 assert(source_deleted
|| oldinfo
->filemask
& old_sidemask
);
2729 /* Need to check for special types of rename conflicts... */
2730 if (collision
&& !source_deleted
) {
2731 /* collision: rename/add or rename/rename(2to1) */
2732 const char *pathnames
[3];
2733 struct version_info merged
;
2735 struct conflict_info
*base
, *side1
, *side2
;
2738 pathnames
[0] = oldpath
;
2739 pathnames
[other_source_index
] = oldpath
;
2740 pathnames
[target_index
] = newpath
;
2742 base
= strmap_get(&opt
->priv
->paths
, pathnames
[0]);
2743 side1
= strmap_get(&opt
->priv
->paths
, pathnames
[1]);
2744 side2
= strmap_get(&opt
->priv
->paths
, pathnames
[2]);
2750 clean
= handle_content_merge(opt
, pair
->one
->path
,
2755 1 + 2 * opt
->priv
->call_depth
,
2758 memcpy(&newinfo
->stages
[target_index
], &merged
,
2761 path_msg(opt
, newpath
, 0,
2762 _("CONFLICT (rename involved in "
2763 "collision): rename of %s -> %s has "
2764 "content conflicts AND collides "
2765 "with another path; this may result "
2766 "in nested conflict markers."),
2769 } else if (collision
&& source_deleted
) {
2771 * rename/add/delete or rename/rename(2to1)/delete:
2772 * since oldpath was deleted on the side that didn't
2773 * do the rename, there's not much of a content merge
2774 * we can do for the rename. oldinfo->merged.is_null
2775 * was already set, so we just leave things as-is so
2776 * they look like an add/add conflict.
2779 newinfo
->path_conflict
= 1;
2780 path_msg(opt
, newpath
, 0,
2781 _("CONFLICT (rename/delete): %s renamed "
2782 "to %s in %s, but deleted in %s."),
2783 oldpath
, newpath
, rename_branch
, delete_branch
);
2786 * a few different cases...start by copying the
2787 * existing stage(s) from oldinfo over the newinfo
2788 * and update the pathname(s).
2790 memcpy(&newinfo
->stages
[0], &oldinfo
->stages
[0],
2791 sizeof(newinfo
->stages
[0]));
2792 newinfo
->filemask
|= (1 << MERGE_BASE
);
2793 newinfo
->pathnames
[0] = oldpath
;
2795 /* rename vs. typechange */
2796 /* Mark the original as resolved by removal */
2797 memcpy(&oldinfo
->stages
[0].oid
, null_oid(),
2798 sizeof(oldinfo
->stages
[0].oid
));
2799 oldinfo
->stages
[0].mode
= 0;
2800 oldinfo
->filemask
&= 0x06;
2801 } else if (source_deleted
) {
2803 newinfo
->path_conflict
= 1;
2804 path_msg(opt
, newpath
, 0,
2805 _("CONFLICT (rename/delete): %s renamed"
2806 " to %s in %s, but deleted in %s."),
2808 rename_branch
, delete_branch
);
2811 memcpy(&newinfo
->stages
[other_source_index
],
2812 &oldinfo
->stages
[other_source_index
],
2813 sizeof(newinfo
->stages
[0]));
2814 newinfo
->filemask
|= (1 << other_source_index
);
2815 newinfo
->pathnames
[other_source_index
] = oldpath
;
2819 if (!type_changed
) {
2820 /* Mark the original as resolved by removal */
2821 oldinfo
->merged
.is_null
= 1;
2822 oldinfo
->merged
.clean
= 1;
2830 static inline int possible_side_renames(struct rename_info
*renames
,
2831 unsigned side_index
)
2833 return renames
->pairs
[side_index
].nr
> 0 &&
2834 !strintmap_empty(&renames
->relevant_sources
[side_index
]);
2837 static inline int possible_renames(struct rename_info
*renames
)
2839 return possible_side_renames(renames
, 1) ||
2840 possible_side_renames(renames
, 2) ||
2841 !strmap_empty(&renames
->cached_pairs
[1]) ||
2842 !strmap_empty(&renames
->cached_pairs
[2]);
2845 static void resolve_diffpair_statuses(struct diff_queue_struct
*q
)
2848 * A simplified version of diff_resolve_rename_copy(); would probably
2849 * just use that function but it's static...
2852 struct diff_filepair
*p
;
2854 for (i
= 0; i
< q
->nr
; ++i
) {
2856 p
->status
= 0; /* undecided */
2857 if (!DIFF_FILE_VALID(p
->one
))
2858 p
->status
= DIFF_STATUS_ADDED
;
2859 else if (!DIFF_FILE_VALID(p
->two
))
2860 p
->status
= DIFF_STATUS_DELETED
;
2861 else if (DIFF_PAIR_RENAME(p
))
2862 p
->status
= DIFF_STATUS_RENAMED
;
2866 static void prune_cached_from_relevant(struct rename_info
*renames
,
2869 /* Reason for this function described in add_pair() */
2870 struct hashmap_iter iter
;
2871 struct strmap_entry
*entry
;
2873 /* Remove from relevant_sources all entries in cached_pairs[side] */
2874 strmap_for_each_entry(&renames
->cached_pairs
[side
], &iter
, entry
) {
2875 strintmap_remove(&renames
->relevant_sources
[side
],
2878 /* Remove from relevant_sources all entries in cached_irrelevant[side] */
2879 strset_for_each_entry(&renames
->cached_irrelevant
[side
], &iter
, entry
) {
2880 strintmap_remove(&renames
->relevant_sources
[side
],
2885 static void use_cached_pairs(struct merge_options
*opt
,
2886 struct strmap
*cached_pairs
,
2887 struct diff_queue_struct
*pairs
)
2889 struct hashmap_iter iter
;
2890 struct strmap_entry
*entry
;
2891 struct mem_pool
*pool
= opt
->priv
->pool
;
2894 * Add to side_pairs all entries from renames->cached_pairs[side_index].
2895 * (Info in cached_irrelevant[side_index] is not relevant here.)
2897 strmap_for_each_entry(cached_pairs
, &iter
, entry
) {
2898 struct diff_filespec
*one
, *two
;
2899 const char *old_name
= entry
->key
;
2900 const char *new_name
= entry
->value
;
2902 new_name
= old_name
;
2905 * cached_pairs has _copies* of old_name and new_name,
2906 * because it has to persist across merges. When
2908 * pool_alloc_filespec() will just re-use the existing
2909 * filenames, which will also get re-used by
2910 * opt->priv->paths if they become renames, and then
2911 * get freed at the end of the merge, leaving the copy
2912 * in cached_pairs dangling. Avoid this by making a
2915 * When pool == NULL, pool_alloc_filespec() calls
2916 * alloc_filespec(), which makes a copy; we don't want
2919 old_name
= mem_pool_strdup(pool
, old_name
);
2920 new_name
= mem_pool_strdup(pool
, new_name
);
2923 /* We don't care about oid/mode, only filenames and status */
2924 one
= pool_alloc_filespec(pool
, old_name
);
2925 two
= pool_alloc_filespec(pool
, new_name
);
2926 pool_diff_queue(pool
, pairs
, one
, two
);
2927 pairs
->queue
[pairs
->nr
-1]->status
= entry
->value
? 'R' : 'D';
2931 static void cache_new_pair(struct rename_info
*renames
,
2938 new_path
= xstrdup(new_path
);
2939 old_value
= strmap_put(&renames
->cached_pairs
[side
],
2940 old_path
, new_path
);
2941 strset_add(&renames
->cached_target_names
[side
], new_path
);
2948 static void possibly_cache_new_pair(struct rename_info
*renames
,
2949 struct diff_filepair
*p
,
2953 int dir_renamed_side
= 0;
2957 * Directory renames happen on the other side of history from
2958 * the side that adds new files to the old directory.
2960 dir_renamed_side
= 3 - side
;
2962 int val
= strintmap_get(&renames
->relevant_sources
[side
],
2964 if (val
== RELEVANT_NO_MORE
) {
2965 assert(p
->status
== 'D');
2966 strset_add(&renames
->cached_irrelevant
[side
],
2973 if (p
->status
== 'D') {
2975 * If we already had this delete, we'll just set it's value
2976 * to NULL again, so no harm.
2978 strmap_put(&renames
->cached_pairs
[side
], p
->one
->path
, NULL
);
2979 } else if (p
->status
== 'R') {
2981 new_path
= p
->two
->path
;
2983 cache_new_pair(renames
, dir_renamed_side
,
2984 p
->two
->path
, new_path
, 0);
2985 cache_new_pair(renames
, side
, p
->one
->path
, new_path
, 1);
2986 } else if (p
->status
== 'A' && new_path
) {
2987 cache_new_pair(renames
, dir_renamed_side
,
2988 p
->two
->path
, new_path
, 0);
2992 static int compare_pairs(const void *a_
, const void *b_
)
2994 const struct diff_filepair
*a
= *((const struct diff_filepair
**)a_
);
2995 const struct diff_filepair
*b
= *((const struct diff_filepair
**)b_
);
2997 return strcmp(a
->one
->path
, b
->one
->path
);
3000 /* Call diffcore_rename() to update deleted/added pairs into rename pairs */
3001 static int detect_regular_renames(struct merge_options
*opt
,
3002 unsigned side_index
)
3004 struct diff_options diff_opts
;
3005 struct rename_info
*renames
= &opt
->priv
->renames
;
3007 prune_cached_from_relevant(renames
, side_index
);
3008 if (!possible_side_renames(renames
, side_index
)) {
3010 * No rename detection needed for this side, but we still need
3011 * to make sure 'adds' are marked correctly in case the other
3012 * side had directory renames.
3014 resolve_diffpair_statuses(&renames
->pairs
[side_index
]);
3018 partial_clear_dir_rename_count(&renames
->dir_rename_count
[side_index
]);
3019 repo_diff_setup(opt
->repo
, &diff_opts
);
3020 diff_opts
.flags
.recursive
= 1;
3021 diff_opts
.flags
.rename_empty
= 0;
3022 diff_opts
.detect_rename
= DIFF_DETECT_RENAME
;
3023 diff_opts
.rename_limit
= opt
->rename_limit
;
3024 if (opt
->rename_limit
<= 0)
3025 diff_opts
.rename_limit
= 1000;
3026 diff_opts
.rename_score
= opt
->rename_score
;
3027 diff_opts
.show_rename_progress
= opt
->show_rename_progress
;
3028 diff_opts
.output_format
= DIFF_FORMAT_NO_OUTPUT
;
3029 diff_setup_done(&diff_opts
);
3031 diff_queued_diff
= renames
->pairs
[side_index
];
3032 trace2_region_enter("diff", "diffcore_rename", opt
->repo
);
3033 diffcore_rename_extended(&diff_opts
,
3035 &renames
->relevant_sources
[side_index
],
3036 &renames
->dirs_removed
[side_index
],
3037 &renames
->dir_rename_count
[side_index
],
3038 &renames
->cached_pairs
[side_index
]);
3039 trace2_region_leave("diff", "diffcore_rename", opt
->repo
);
3040 resolve_diffpair_statuses(&diff_queued_diff
);
3042 if (diff_opts
.needed_rename_limit
> 0)
3043 renames
->redo_after_renames
= 0;
3044 if (diff_opts
.needed_rename_limit
> renames
->needed_limit
)
3045 renames
->needed_limit
= diff_opts
.needed_rename_limit
;
3047 renames
->pairs
[side_index
] = diff_queued_diff
;
3049 diff_opts
.output_format
= DIFF_FORMAT_NO_OUTPUT
;
3050 diff_queued_diff
.nr
= 0;
3051 diff_queued_diff
.queue
= NULL
;
3052 diff_flush(&diff_opts
);
3058 * Get information of all renames which occurred in 'side_pairs', making use
3059 * of any implicit directory renames in side_dir_renames (also making use of
3060 * implicit directory renames rename_exclusions as needed by
3061 * check_for_directory_rename()). Add all (updated) renames into result.
3063 static int collect_renames(struct merge_options
*opt
,
3064 struct diff_queue_struct
*result
,
3065 unsigned side_index
,
3066 struct strmap
*dir_renames_for_side
,
3067 struct strmap
*rename_exclusions
)
3070 struct strmap collisions
;
3071 struct diff_queue_struct
*side_pairs
;
3072 struct hashmap_iter iter
;
3073 struct strmap_entry
*entry
;
3074 struct rename_info
*renames
= &opt
->priv
->renames
;
3076 side_pairs
= &renames
->pairs
[side_index
];
3077 compute_collisions(&collisions
, dir_renames_for_side
, side_pairs
);
3079 for (i
= 0; i
< side_pairs
->nr
; ++i
) {
3080 struct diff_filepair
*p
= side_pairs
->queue
[i
];
3081 char *new_path
; /* non-NULL only with directory renames */
3083 if (p
->status
!= 'A' && p
->status
!= 'R') {
3084 possibly_cache_new_pair(renames
, p
, side_index
, NULL
);
3085 pool_diff_free_filepair(opt
->priv
->pool
, p
);
3089 new_path
= check_for_directory_rename(opt
, p
->two
->path
,
3091 dir_renames_for_side
,
3096 possibly_cache_new_pair(renames
, p
, side_index
, new_path
);
3097 if (p
->status
!= 'R' && !new_path
) {
3098 pool_diff_free_filepair(opt
->priv
->pool
, p
);
3103 apply_directory_rename_modifications(opt
, p
, new_path
);
3106 * p->score comes back from diffcore_rename_extended() with
3107 * the similarity of the renamed file. The similarity is
3108 * was used to determine that the two files were related
3109 * and are a rename, which we have already used, but beyond
3110 * that we have no use for the similarity. So p->score is
3111 * now irrelevant. However, process_renames() will need to
3112 * know which side of the merge this rename was associated
3113 * with, so overwrite p->score with that value.
3115 p
->score
= side_index
;
3116 result
->queue
[result
->nr
++] = p
;
3119 /* Free each value in the collisions map */
3120 strmap_for_each_entry(&collisions
, &iter
, entry
) {
3121 struct collision_info
*info
= entry
->value
;
3122 string_list_clear(&info
->source_files
, 0);
3125 * In compute_collisions(), we set collisions.strdup_strings to 0
3126 * so that we wouldn't have to make another copy of the new_path
3127 * allocated by apply_dir_rename(). But now that we've used them
3128 * and have no other references to these strings, it is time to
3131 free_strmap_strings(&collisions
);
3132 strmap_clear(&collisions
, 1);
3136 static int detect_and_process_renames(struct merge_options
*opt
,
3137 struct tree
*merge_base
,
3141 struct diff_queue_struct combined
;
3142 struct rename_info
*renames
= &opt
->priv
->renames
;
3143 int need_dir_renames
, s
, clean
= 1;
3144 unsigned detection_run
= 0;
3146 memset(&combined
, 0, sizeof(combined
));
3147 if (!possible_renames(renames
))
3150 trace2_region_enter("merge", "regular renames", opt
->repo
);
3151 detection_run
|= detect_regular_renames(opt
, MERGE_SIDE1
);
3152 detection_run
|= detect_regular_renames(opt
, MERGE_SIDE2
);
3153 if (renames
->redo_after_renames
&& detection_run
) {
3155 struct diff_filepair
*p
;
3157 /* Cache the renames, we found */
3158 for (side
= MERGE_SIDE1
; side
<= MERGE_SIDE2
; side
++) {
3159 for (i
= 0; i
< renames
->pairs
[side
].nr
; ++i
) {
3160 p
= renames
->pairs
[side
].queue
[i
];
3161 possibly_cache_new_pair(renames
, p
, side
, NULL
);
3165 /* Restart the merge with the cached renames */
3166 renames
->redo_after_renames
= 2;
3167 trace2_region_leave("merge", "regular renames", opt
->repo
);
3170 use_cached_pairs(opt
, &renames
->cached_pairs
[1], &renames
->pairs
[1]);
3171 use_cached_pairs(opt
, &renames
->cached_pairs
[2], &renames
->pairs
[2]);
3172 trace2_region_leave("merge", "regular renames", opt
->repo
);
3174 trace2_region_enter("merge", "directory renames", opt
->repo
);
3176 !opt
->priv
->call_depth
&&
3177 (opt
->detect_directory_renames
== MERGE_DIRECTORY_RENAMES_TRUE
||
3178 opt
->detect_directory_renames
== MERGE_DIRECTORY_RENAMES_CONFLICT
);
3180 if (need_dir_renames
) {
3181 get_provisional_directory_renames(opt
, MERGE_SIDE1
, &clean
);
3182 get_provisional_directory_renames(opt
, MERGE_SIDE2
, &clean
);
3183 handle_directory_level_conflicts(opt
);
3186 ALLOC_GROW(combined
.queue
,
3187 renames
->pairs
[1].nr
+ renames
->pairs
[2].nr
,
3189 clean
&= collect_renames(opt
, &combined
, MERGE_SIDE1
,
3190 &renames
->dir_renames
[2],
3191 &renames
->dir_renames
[1]);
3192 clean
&= collect_renames(opt
, &combined
, MERGE_SIDE2
,
3193 &renames
->dir_renames
[1],
3194 &renames
->dir_renames
[2]);
3195 STABLE_QSORT(combined
.queue
, combined
.nr
, compare_pairs
);
3196 trace2_region_leave("merge", "directory renames", opt
->repo
);
3198 trace2_region_enter("merge", "process renames", opt
->repo
);
3199 clean
&= process_renames(opt
, &combined
);
3200 trace2_region_leave("merge", "process renames", opt
->repo
);
3202 goto simple_cleanup
; /* collect_renames() handles some of cleanup */
3206 * Free now unneeded filepairs, which would have been handled
3207 * in collect_renames() normally but we skipped that code.
3209 for (s
= MERGE_SIDE1
; s
<= MERGE_SIDE2
; s
++) {
3210 struct diff_queue_struct
*side_pairs
;
3213 side_pairs
= &renames
->pairs
[s
];
3214 for (i
= 0; i
< side_pairs
->nr
; ++i
) {
3215 struct diff_filepair
*p
= side_pairs
->queue
[i
];
3216 pool_diff_free_filepair(opt
->priv
->pool
, p
);
3221 /* Free memory for renames->pairs[] and combined */
3222 for (s
= MERGE_SIDE1
; s
<= MERGE_SIDE2
; s
++) {
3223 free(renames
->pairs
[s
].queue
);
3224 DIFF_QUEUE_CLEAR(&renames
->pairs
[s
]);
3228 for (i
= 0; i
< combined
.nr
; i
++)
3229 pool_diff_free_filepair(opt
->priv
->pool
,
3231 free(combined
.queue
);
3237 /*** Function Grouping: functions related to process_entries() ***/
3239 static int sort_dirs_next_to_their_children(const char *one
, const char *two
)
3241 unsigned char c1
, c2
;
3244 * Here we only care that entries for directories appear adjacent
3245 * to and before files underneath the directory. We can achieve
3246 * that by pretending to add a trailing slash to every file and
3247 * then sorting. In other words, we do not want the natural
3252 * Instead, we want "foo" to sort as though it were "foo/", so that
3257 * To achieve this, we basically implement our own strcmp, except that
3258 * if we get to the end of either string instead of comparing NUL to
3259 * another character, we compare '/' to it.
3261 * If this unusual "sort as though '/' were appended" perplexes
3262 * you, perhaps it will help to note that this is not the final
3263 * sort. write_tree() will sort again without the trailing slash
3264 * magic, but just on paths immediately under a given tree.
3266 * The reason to not use df_name_compare directly was that it was
3267 * just too expensive (we don't have the string lengths handy), so
3268 * it was reimplemented.
3272 * NOTE: This function will never be called with two equal strings,
3273 * because it is used to sort the keys of a strmap, and strmaps have
3274 * unique keys by construction. That simplifies our c1==c2 handling
3278 while (*one
&& (*one
== *two
)) {
3283 c1
= *one
? *one
: '/';
3284 c2
= *two
? *two
: '/';
3287 /* Getting here means one is a leading directory of the other */
3288 return (*one
) ? 1 : -1;
3293 static int read_oid_strbuf(struct merge_options
*opt
,
3294 const struct object_id
*oid
,
3298 enum object_type type
;
3300 buf
= read_object_file(oid
, &type
, &size
);
3302 return err(opt
, _("cannot read object %s"), oid_to_hex(oid
));
3303 if (type
!= OBJ_BLOB
) {
3305 return err(opt
, _("object %s is not a blob"), oid_to_hex(oid
));
3307 strbuf_attach(dst
, buf
, size
, size
+ 1);
3311 static int blob_unchanged(struct merge_options
*opt
,
3312 const struct version_info
*base
,
3313 const struct version_info
*side
,
3316 struct strbuf basebuf
= STRBUF_INIT
;
3317 struct strbuf sidebuf
= STRBUF_INIT
;
3318 int ret
= 0; /* assume changed for safety */
3319 struct index_state
*idx
= &opt
->priv
->attr_index
;
3321 if (!idx
->initialized
)
3322 initialize_attr_index(opt
);
3324 if (base
->mode
!= side
->mode
)
3326 if (oideq(&base
->oid
, &side
->oid
))
3329 if (read_oid_strbuf(opt
, &base
->oid
, &basebuf
) ||
3330 read_oid_strbuf(opt
, &side
->oid
, &sidebuf
))
3333 * Note: binary | is used so that both renormalizations are
3334 * performed. Comparison can be skipped if both files are
3335 * unchanged since their sha1s have already been compared.
3337 if (renormalize_buffer(idx
, path
, basebuf
.buf
, basebuf
.len
, &basebuf
) |
3338 renormalize_buffer(idx
, path
, sidebuf
.buf
, sidebuf
.len
, &sidebuf
))
3339 ret
= (basebuf
.len
== sidebuf
.len
&&
3340 !memcmp(basebuf
.buf
, sidebuf
.buf
, basebuf
.len
));
3343 strbuf_release(&basebuf
);
3344 strbuf_release(&sidebuf
);
3348 struct directory_versions
{
3350 * versions: list of (basename -> version_info)
3352 * The basenames are in reverse lexicographic order of full pathnames,
3353 * as processed in process_entries(). This puts all entries within
3354 * a directory together, and covers the directory itself after
3355 * everything within it, allowing us to write subtrees before needing
3356 * to record information for the tree itself.
3358 struct string_list versions
;
3361 * offsets: list of (full relative path directories -> integer offsets)
3363 * Since versions contains basenames from files in multiple different
3364 * directories, we need to know which entries in versions correspond
3365 * to which directories. Values of e.g.
3369 * Would mean that entries 0-1 of versions are files in the toplevel
3370 * directory, entries 2-4 are files under src/, and the remaining
3371 * entries starting at index 5 are files under src/moduleA/.
3373 struct string_list offsets
;
3376 * last_directory: directory that previously processed file found in
3378 * last_directory starts NULL, but records the directory in which the
3379 * previous file was found within. As soon as
3380 * directory(current_file) != last_directory
3381 * then we need to start updating accounting in versions & offsets.
3382 * Note that last_directory is always the last path in "offsets" (or
3383 * NULL if "offsets" is empty) so this exists just for quick access.
3385 const char *last_directory
;
3387 /* last_directory_len: cached computation of strlen(last_directory) */
3388 unsigned last_directory_len
;
3391 static int tree_entry_order(const void *a_
, const void *b_
)
3393 const struct string_list_item
*a
= a_
;
3394 const struct string_list_item
*b
= b_
;
3396 const struct merged_info
*ami
= a
->util
;
3397 const struct merged_info
*bmi
= b
->util
;
3398 return base_name_compare(a
->string
, strlen(a
->string
), ami
->result
.mode
,
3399 b
->string
, strlen(b
->string
), bmi
->result
.mode
);
3402 static void write_tree(struct object_id
*result_oid
,
3403 struct string_list
*versions
,
3404 unsigned int offset
,
3407 size_t maxlen
= 0, extra
;
3409 struct strbuf buf
= STRBUF_INIT
;
3412 assert(offset
<= versions
->nr
);
3413 nr
= versions
->nr
- offset
;
3415 /* No need for STABLE_QSORT -- filenames must be unique */
3416 QSORT(versions
->items
+ offset
, nr
, tree_entry_order
);
3418 /* Pre-allocate some space in buf */
3419 extra
= hash_size
+ 8; /* 8: 6 for mode, 1 for space, 1 for NUL char */
3420 for (i
= 0; i
< nr
; i
++) {
3421 maxlen
+= strlen(versions
->items
[offset
+i
].string
) + extra
;
3423 strbuf_grow(&buf
, maxlen
);
3425 /* Write each entry out to buf */
3426 for (i
= 0; i
< nr
; i
++) {
3427 struct merged_info
*mi
= versions
->items
[offset
+i
].util
;
3428 struct version_info
*ri
= &mi
->result
;
3429 strbuf_addf(&buf
, "%o %s%c",
3431 versions
->items
[offset
+i
].string
, '\0');
3432 strbuf_add(&buf
, ri
->oid
.hash
, hash_size
);
3435 /* Write this object file out, and record in result_oid */
3436 write_object_file(buf
.buf
, buf
.len
, tree_type
, result_oid
);
3437 strbuf_release(&buf
);
3440 static void record_entry_for_tree(struct directory_versions
*dir_metadata
,
3442 struct merged_info
*mi
)
3444 const char *basename
;
3447 /* nothing to record */
3450 basename
= path
+ mi
->basename_offset
;
3451 assert(strchr(basename
, '/') == NULL
);
3452 string_list_append(&dir_metadata
->versions
,
3453 basename
)->util
= &mi
->result
;
3456 static void write_completed_directory(struct merge_options
*opt
,
3457 const char *new_directory_name
,
3458 struct directory_versions
*info
)
3460 const char *prev_dir
;
3461 struct merged_info
*dir_info
= NULL
;
3462 unsigned int offset
;
3465 * Some explanation of info->versions and info->offsets...
3467 * process_entries() iterates over all relevant files AND
3468 * directories in reverse lexicographic order, and calls this
3469 * function. Thus, an example of the paths that process_entries()
3470 * could operate on (along with the directories for those paths
3475 * src/moduleB/umm.c src/moduleB
3476 * src/moduleB/stuff.h src/moduleB
3477 * src/moduleB/baz.c src/moduleB
3479 * src/moduleA/foo.c src/moduleA
3480 * src/moduleA/bar.c src/moduleA
3487 * always contains the unprocessed entries and their
3488 * version_info information. For example, after the first five
3489 * entries above, info->versions would be:
3491 * xtract.c <xtract.c's version_info>
3492 * token.txt <token.txt's version_info>
3493 * umm.c <src/moduleB/umm.c's version_info>
3494 * stuff.h <src/moduleB/stuff.h's version_info>
3495 * baz.c <src/moduleB/baz.c's version_info>
3497 * Once a subdirectory is completed we remove the entries in
3498 * that subdirectory from info->versions, writing it as a tree
3499 * (write_tree()). Thus, as soon as we get to src/moduleB,
3500 * info->versions would be updated to
3502 * xtract.c <xtract.c's version_info>
3503 * token.txt <token.txt's version_info>
3504 * moduleB <src/moduleB's version_info>
3508 * helps us track which entries in info->versions correspond to
3509 * which directories. When we are N directories deep (e.g. 4
3510 * for src/modA/submod/subdir/), we have up to N+1 unprocessed
3511 * directories (+1 because of toplevel dir). Corresponding to
3512 * the info->versions example above, after processing five entries
3513 * info->offsets will be:
3518 * which is used to know that xtract.c & token.txt are from the
3519 * toplevel dirctory, while umm.c & stuff.h & baz.c are from the
3520 * src/moduleB directory. Again, following the example above,
3521 * once we need to process src/moduleB, then info->offsets is
3527 * which says that moduleB (and only moduleB so far) is in the
3530 * One unique thing to note about info->offsets here is that
3531 * "src" was not added to info->offsets until there was a path
3532 * (a file OR directory) immediately below src/ that got
3535 * Since process_entry() just appends new entries to info->versions,
3536 * write_completed_directory() only needs to do work if the next path
3537 * is in a directory that is different than the last directory found
3542 * If we are working with the same directory as the last entry, there
3543 * is no work to do. (See comments above the directory_name member of
3544 * struct merged_info for why we can use pointer comparison instead of
3547 if (new_directory_name
== info
->last_directory
)
3551 * If we are just starting (last_directory is NULL), or last_directory
3552 * is a prefix of the current directory, then we can just update
3553 * info->offsets to record the offset where we started this directory
3554 * and update last_directory to have quick access to it.
3556 if (info
->last_directory
== NULL
||
3557 !strncmp(new_directory_name
, info
->last_directory
,
3558 info
->last_directory_len
)) {
3559 uintptr_t offset
= info
->versions
.nr
;
3561 info
->last_directory
= new_directory_name
;
3562 info
->last_directory_len
= strlen(info
->last_directory
);
3564 * Record the offset into info->versions where we will
3565 * start recording basenames of paths found within
3566 * new_directory_name.
3568 string_list_append(&info
->offsets
,
3569 info
->last_directory
)->util
= (void*)offset
;
3574 * The next entry that will be processed will be within
3575 * new_directory_name. Since at this point we know that
3576 * new_directory_name is within a different directory than
3577 * info->last_directory, we have all entries for info->last_directory
3578 * in info->versions and we need to create a tree object for them.
3580 dir_info
= strmap_get(&opt
->priv
->paths
, info
->last_directory
);
3582 offset
= (uintptr_t)info
->offsets
.items
[info
->offsets
.nr
-1].util
;
3583 if (offset
== info
->versions
.nr
) {
3585 * Actually, we don't need to create a tree object in this
3586 * case. Whenever all files within a directory disappear
3587 * during the merge (e.g. unmodified on one side and
3588 * deleted on the other, or files were renamed elsewhere),
3589 * then we get here and the directory itself needs to be
3590 * omitted from its parent tree as well.
3592 dir_info
->is_null
= 1;
3595 * Write out the tree to the git object directory, and also
3596 * record the mode and oid in dir_info->result.
3598 dir_info
->is_null
= 0;
3599 dir_info
->result
.mode
= S_IFDIR
;
3600 write_tree(&dir_info
->result
.oid
, &info
->versions
, offset
,
3601 opt
->repo
->hash_algo
->rawsz
);
3605 * We've now used several entries from info->versions and one entry
3606 * from info->offsets, so we get rid of those values.
3609 info
->versions
.nr
= offset
;
3612 * Now we've taken care of the completed directory, but we need to
3613 * prepare things since future entries will be in
3614 * new_directory_name. (In particular, process_entry() will be
3615 * appending new entries to info->versions.) So, we need to make
3616 * sure new_directory_name is the last entry in info->offsets.
3618 prev_dir
= info
->offsets
.nr
== 0 ? NULL
:
3619 info
->offsets
.items
[info
->offsets
.nr
-1].string
;
3620 if (new_directory_name
!= prev_dir
) {
3621 uintptr_t c
= info
->versions
.nr
;
3622 string_list_append(&info
->offsets
,
3623 new_directory_name
)->util
= (void*)c
;
3626 /* And, of course, we need to update last_directory to match. */
3627 info
->last_directory
= new_directory_name
;
3628 info
->last_directory_len
= strlen(info
->last_directory
);
3631 /* Per entry merge function */
3632 static void process_entry(struct merge_options
*opt
,
3634 struct conflict_info
*ci
,
3635 struct directory_versions
*dir_metadata
)
3637 int df_file_index
= 0;
3640 assert(ci
->filemask
>= 0 && ci
->filemask
<= 7);
3641 /* ci->match_mask == 7 was handled in collect_merge_info_callback() */
3642 assert(ci
->match_mask
== 0 || ci
->match_mask
== 3 ||
3643 ci
->match_mask
== 5 || ci
->match_mask
== 6);
3646 record_entry_for_tree(dir_metadata
, path
, &ci
->merged
);
3647 if (ci
->filemask
== 0)
3648 /* nothing else to handle */
3650 assert(ci
->df_conflict
);
3653 if (ci
->df_conflict
&& ci
->merged
.result
.mode
== 0) {
3657 * directory no longer in the way, but we do have a file we
3658 * need to place here so we need to clean away the "directory
3659 * merges to nothing" result.
3661 ci
->df_conflict
= 0;
3662 assert(ci
->filemask
!= 0);
3663 ci
->merged
.clean
= 0;
3664 ci
->merged
.is_null
= 0;
3665 /* and we want to zero out any directory-related entries */
3666 ci
->match_mask
= (ci
->match_mask
& ~ci
->dirmask
);
3668 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
3669 if (ci
->filemask
& (1 << i
))
3671 ci
->stages
[i
].mode
= 0;
3672 oidcpy(&ci
->stages
[i
].oid
, null_oid());
3674 } else if (ci
->df_conflict
&& ci
->merged
.result
.mode
!= 0) {
3676 * This started out as a D/F conflict, and the entries in
3677 * the competing directory were not removed by the merge as
3678 * evidenced by write_completed_directory() writing a value
3679 * to ci->merged.result.mode.
3681 struct conflict_info
*new_ci
;
3683 const char *old_path
= path
;
3686 assert(ci
->merged
.result
.mode
== S_IFDIR
);
3689 * If filemask is 1, we can just ignore the file as having
3690 * been deleted on both sides. We do not want to overwrite
3691 * ci->merged.result, since it stores the tree for all the
3694 if (ci
->filemask
== 1) {
3700 * This file still exists on at least one side, and we want
3701 * the directory to remain here, so we need to move this
3702 * path to some new location.
3704 new_ci
= pool_calloc(opt
->priv
->pool
, 1, sizeof(*new_ci
));
3706 /* We don't really want new_ci->merged.result copied, but it'll
3707 * be overwritten below so it doesn't matter. We also don't
3708 * want any directory mode/oid values copied, but we'll zero
3709 * those out immediately. We do want the rest of ci copied.
3711 memcpy(new_ci
, ci
, sizeof(*ci
));
3712 new_ci
->match_mask
= (new_ci
->match_mask
& ~new_ci
->dirmask
);
3713 new_ci
->dirmask
= 0;
3714 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
3715 if (new_ci
->filemask
& (1 << i
))
3717 /* zero out any entries related to directories */
3718 new_ci
->stages
[i
].mode
= 0;
3719 oidcpy(&new_ci
->stages
[i
].oid
, null_oid());
3723 * Find out which side this file came from; note that we
3724 * cannot just use ci->filemask, because renames could cause
3725 * the filemask to go back to 7. So we use dirmask, then
3726 * pick the opposite side's index.
3728 df_file_index
= (ci
->dirmask
& (1 << 1)) ? 2 : 1;
3729 branch
= (df_file_index
== 1) ? opt
->branch1
: opt
->branch2
;
3730 path
= unique_path(&opt
->priv
->paths
, path
, branch
);
3731 strmap_put(&opt
->priv
->paths
, path
, new_ci
);
3733 path_msg(opt
, path
, 0,
3734 _("CONFLICT (file/directory): directory in the way "
3735 "of %s from %s; moving it to %s instead."),
3736 old_path
, branch
, path
);
3739 * Zero out the filemask for the old ci. At this point, ci
3740 * was just an entry for a directory, so we don't need to
3741 * do anything more with it.
3746 * Now note that we're working on the new entry (path was
3753 * NOTE: Below there is a long switch-like if-elseif-elseif... block
3754 * which the code goes through even for the df_conflict cases
3757 if (ci
->match_mask
) {
3758 ci
->merged
.clean
= !ci
->df_conflict
&& !ci
->path_conflict
;
3759 if (ci
->match_mask
== 6) {
3760 /* stages[1] == stages[2] */
3761 ci
->merged
.result
.mode
= ci
->stages
[1].mode
;
3762 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[1].oid
);
3764 /* determine the mask of the side that didn't match */
3765 unsigned int othermask
= 7 & ~ci
->match_mask
;
3766 int side
= (othermask
== 4) ? 2 : 1;
3768 ci
->merged
.result
.mode
= ci
->stages
[side
].mode
;
3769 ci
->merged
.is_null
= !ci
->merged
.result
.mode
;
3770 if (ci
->merged
.is_null
)
3771 ci
->merged
.clean
= 1;
3772 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[side
].oid
);
3774 assert(othermask
== 2 || othermask
== 4);
3775 assert(ci
->merged
.is_null
==
3776 (ci
->filemask
== ci
->match_mask
));
3778 } else if (ci
->filemask
>= 6 &&
3779 (S_IFMT
& ci
->stages
[1].mode
) !=
3780 (S_IFMT
& ci
->stages
[2].mode
)) {
3781 /* Two different items from (file/submodule/symlink) */
3782 if (opt
->priv
->call_depth
) {
3783 /* Just use the version from the merge base */
3784 ci
->merged
.clean
= 0;
3785 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[0].oid
);
3786 ci
->merged
.result
.mode
= ci
->stages
[0].mode
;
3787 ci
->merged
.is_null
= (ci
->merged
.result
.mode
== 0);
3789 /* Handle by renaming one or both to separate paths. */
3790 unsigned o_mode
= ci
->stages
[0].mode
;
3791 unsigned a_mode
= ci
->stages
[1].mode
;
3792 unsigned b_mode
= ci
->stages
[2].mode
;
3793 struct conflict_info
*new_ci
;
3794 const char *a_path
= NULL
, *b_path
= NULL
;
3795 int rename_a
= 0, rename_b
= 0;
3797 new_ci
= pool_alloc(opt
->priv
->pool
, sizeof(*new_ci
));
3799 if (S_ISREG(a_mode
))
3801 else if (S_ISREG(b_mode
))
3808 if (rename_a
&& rename_b
) {
3809 path_msg(opt
, path
, 0,
3810 _("CONFLICT (distinct types): %s had "
3811 "different types on each side; "
3812 "renamed both of them so each can "
3813 "be recorded somewhere."),
3816 path_msg(opt
, path
, 0,
3817 _("CONFLICT (distinct types): %s had "
3818 "different types on each side; "
3819 "renamed one of them so each can be "
3820 "recorded somewhere."),
3824 ci
->merged
.clean
= 0;
3825 memcpy(new_ci
, ci
, sizeof(*new_ci
));
3827 /* Put b into new_ci, removing a from stages */
3828 new_ci
->merged
.result
.mode
= ci
->stages
[2].mode
;
3829 oidcpy(&new_ci
->merged
.result
.oid
, &ci
->stages
[2].oid
);
3830 new_ci
->stages
[1].mode
= 0;
3831 oidcpy(&new_ci
->stages
[1].oid
, null_oid());
3832 new_ci
->filemask
= 5;
3833 if ((S_IFMT
& b_mode
) != (S_IFMT
& o_mode
)) {
3834 new_ci
->stages
[0].mode
= 0;
3835 oidcpy(&new_ci
->stages
[0].oid
, null_oid());
3836 new_ci
->filemask
= 4;
3839 /* Leave only a in ci, fixing stages. */
3840 ci
->merged
.result
.mode
= ci
->stages
[1].mode
;
3841 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[1].oid
);
3842 ci
->stages
[2].mode
= 0;
3843 oidcpy(&ci
->stages
[2].oid
, null_oid());
3845 if ((S_IFMT
& a_mode
) != (S_IFMT
& o_mode
)) {
3846 ci
->stages
[0].mode
= 0;
3847 oidcpy(&ci
->stages
[0].oid
, null_oid());
3851 /* Insert entries into opt->priv_paths */
3852 assert(rename_a
|| rename_b
);
3854 a_path
= unique_path(&opt
->priv
->paths
,
3855 path
, opt
->branch1
);
3856 strmap_put(&opt
->priv
->paths
, a_path
, ci
);
3860 b_path
= unique_path(&opt
->priv
->paths
,
3861 path
, opt
->branch2
);
3864 strmap_put(&opt
->priv
->paths
, b_path
, new_ci
);
3866 if (rename_a
&& rename_b
) {
3867 strmap_remove(&opt
->priv
->paths
, path
, 0);
3869 * We removed path from opt->priv->paths. path
3870 * will also eventually need to be freed if not
3871 * part of a memory pool...but it may still be
3872 * used by e.g. ci->pathnames. So, store it in
3873 * another string-list for now in that case.
3875 if (!opt
->priv
->pool
)
3876 string_list_append(&opt
->priv
->paths_to_free
,
3881 * Do special handling for b_path since process_entry()
3882 * won't be called on it specially.
3884 strmap_put(&opt
->priv
->conflicted
, b_path
, new_ci
);
3885 record_entry_for_tree(dir_metadata
, b_path
,
3889 * Remaining code for processing this entry should
3890 * think in terms of processing a_path.
3895 } else if (ci
->filemask
>= 6) {
3896 /* Need a two-way or three-way content merge */
3897 struct version_info merged_file
;
3898 unsigned clean_merge
;
3899 struct version_info
*o
= &ci
->stages
[0];
3900 struct version_info
*a
= &ci
->stages
[1];
3901 struct version_info
*b
= &ci
->stages
[2];
3903 clean_merge
= handle_content_merge(opt
, path
, o
, a
, b
,
3905 opt
->priv
->call_depth
* 2,
3907 ci
->merged
.clean
= clean_merge
&&
3908 !ci
->df_conflict
&& !ci
->path_conflict
;
3909 ci
->merged
.result
.mode
= merged_file
.mode
;
3910 ci
->merged
.is_null
= (merged_file
.mode
== 0);
3911 oidcpy(&ci
->merged
.result
.oid
, &merged_file
.oid
);
3912 if (clean_merge
&& ci
->df_conflict
) {
3913 assert(df_file_index
== 1 || df_file_index
== 2);
3914 ci
->filemask
= 1 << df_file_index
;
3915 ci
->stages
[df_file_index
].mode
= merged_file
.mode
;
3916 oidcpy(&ci
->stages
[df_file_index
].oid
, &merged_file
.oid
);
3919 const char *reason
= _("content");
3920 if (ci
->filemask
== 6)
3921 reason
= _("add/add");
3922 if (S_ISGITLINK(merged_file
.mode
))
3923 reason
= _("submodule");
3924 path_msg(opt
, path
, 0,
3925 _("CONFLICT (%s): Merge conflict in %s"),
3928 } else if (ci
->filemask
== 3 || ci
->filemask
== 5) {
3930 const char *modify_branch
, *delete_branch
;
3931 int side
= (ci
->filemask
== 5) ? 2 : 1;
3932 int index
= opt
->priv
->call_depth
? 0 : side
;
3934 ci
->merged
.result
.mode
= ci
->stages
[index
].mode
;
3935 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[index
].oid
);
3936 ci
->merged
.clean
= 0;
3938 modify_branch
= (side
== 1) ? opt
->branch1
: opt
->branch2
;
3939 delete_branch
= (side
== 1) ? opt
->branch2
: opt
->branch1
;
3941 if (opt
->renormalize
&&
3942 blob_unchanged(opt
, &ci
->stages
[0], &ci
->stages
[side
],
3944 ci
->merged
.is_null
= 1;
3945 ci
->merged
.clean
= 1;
3946 assert(!ci
->df_conflict
&& !ci
->path_conflict
);
3947 } else if (ci
->path_conflict
&&
3948 oideq(&ci
->stages
[0].oid
, &ci
->stages
[side
].oid
)) {
3950 * This came from a rename/delete; no action to take,
3951 * but avoid printing "modify/delete" conflict notice
3952 * since the contents were not modified.
3955 path_msg(opt
, path
, 0,
3956 _("CONFLICT (modify/delete): %s deleted in %s "
3957 "and modified in %s. Version %s of %s left "
3959 path
, delete_branch
, modify_branch
,
3960 modify_branch
, path
);
3962 } else if (ci
->filemask
== 2 || ci
->filemask
== 4) {
3963 /* Added on one side */
3964 int side
= (ci
->filemask
== 4) ? 2 : 1;
3965 ci
->merged
.result
.mode
= ci
->stages
[side
].mode
;
3966 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[side
].oid
);
3967 ci
->merged
.clean
= !ci
->df_conflict
&& !ci
->path_conflict
;
3968 } else if (ci
->filemask
== 1) {
3969 /* Deleted on both sides */
3970 ci
->merged
.is_null
= 1;
3971 ci
->merged
.result
.mode
= 0;
3972 oidcpy(&ci
->merged
.result
.oid
, null_oid());
3973 assert(!ci
->df_conflict
);
3974 ci
->merged
.clean
= !ci
->path_conflict
;
3978 * If still conflicted, record it separately. This allows us to later
3979 * iterate over just conflicted entries when updating the index instead
3980 * of iterating over all entries.
3982 if (!ci
->merged
.clean
)
3983 strmap_put(&opt
->priv
->conflicted
, path
, ci
);
3985 /* Record metadata for ci->merged in dir_metadata */
3986 record_entry_for_tree(dir_metadata
, path
, &ci
->merged
);
3989 static void prefetch_for_content_merges(struct merge_options
*opt
,
3990 struct string_list
*plist
)
3992 struct string_list_item
*e
;
3993 struct oid_array to_fetch
= OID_ARRAY_INIT
;
3995 if (opt
->repo
!= the_repository
|| !has_promisor_remote())
3998 for (e
= &plist
->items
[plist
->nr
-1]; e
>= plist
->items
; --e
) {
3999 /* char *path = e->string; */
4000 struct conflict_info
*ci
= e
->util
;
4003 /* Ignore clean entries */
4004 if (ci
->merged
.clean
)
4007 /* Ignore entries that don't need a content merge */
4008 if (ci
->match_mask
|| ci
->filemask
< 6 ||
4009 !S_ISREG(ci
->stages
[1].mode
) ||
4010 !S_ISREG(ci
->stages
[2].mode
) ||
4011 oideq(&ci
->stages
[1].oid
, &ci
->stages
[2].oid
))
4014 /* Also don't need content merge if base matches either side */
4015 if (ci
->filemask
== 7 &&
4016 S_ISREG(ci
->stages
[0].mode
) &&
4017 (oideq(&ci
->stages
[0].oid
, &ci
->stages
[1].oid
) ||
4018 oideq(&ci
->stages
[0].oid
, &ci
->stages
[2].oid
)))
4021 for (i
= 0; i
< 3; i
++) {
4022 unsigned side_mask
= (1 << i
);
4023 struct version_info
*vi
= &ci
->stages
[i
];
4025 if ((ci
->filemask
& side_mask
) &&
4026 S_ISREG(vi
->mode
) &&
4027 oid_object_info_extended(opt
->repo
, &vi
->oid
, NULL
,
4028 OBJECT_INFO_FOR_PREFETCH
))
4029 oid_array_append(&to_fetch
, &vi
->oid
);
4033 promisor_remote_get_direct(opt
->repo
, to_fetch
.oid
, to_fetch
.nr
);
4034 oid_array_clear(&to_fetch
);
4037 static void process_entries(struct merge_options
*opt
,
4038 struct object_id
*result_oid
)
4040 struct hashmap_iter iter
;
4041 struct strmap_entry
*e
;
4042 struct string_list plist
= STRING_LIST_INIT_NODUP
;
4043 struct string_list_item
*entry
;
4044 struct directory_versions dir_metadata
= { STRING_LIST_INIT_NODUP
,
4045 STRING_LIST_INIT_NODUP
,
4048 trace2_region_enter("merge", "process_entries setup", opt
->repo
);
4049 if (strmap_empty(&opt
->priv
->paths
)) {
4050 oidcpy(result_oid
, opt
->repo
->hash_algo
->empty_tree
);
4054 /* Hack to pre-allocate plist to the desired size */
4055 trace2_region_enter("merge", "plist grow", opt
->repo
);
4056 ALLOC_GROW(plist
.items
, strmap_get_size(&opt
->priv
->paths
), plist
.alloc
);
4057 trace2_region_leave("merge", "plist grow", opt
->repo
);
4059 /* Put every entry from paths into plist, then sort */
4060 trace2_region_enter("merge", "plist copy", opt
->repo
);
4061 strmap_for_each_entry(&opt
->priv
->paths
, &iter
, e
) {
4062 string_list_append(&plist
, e
->key
)->util
= e
->value
;
4064 trace2_region_leave("merge", "plist copy", opt
->repo
);
4066 trace2_region_enter("merge", "plist special sort", opt
->repo
);
4067 plist
.cmp
= sort_dirs_next_to_their_children
;
4068 string_list_sort(&plist
);
4069 trace2_region_leave("merge", "plist special sort", opt
->repo
);
4071 trace2_region_leave("merge", "process_entries setup", opt
->repo
);
4074 * Iterate over the items in reverse order, so we can handle paths
4075 * below a directory before needing to handle the directory itself.
4077 * This allows us to write subtrees before we need to write trees,
4078 * and it also enables sane handling of directory/file conflicts
4079 * (because it allows us to know whether the directory is still in
4080 * the way when it is time to process the file at the same path).
4082 trace2_region_enter("merge", "processing", opt
->repo
);
4083 prefetch_for_content_merges(opt
, &plist
);
4084 for (entry
= &plist
.items
[plist
.nr
-1]; entry
>= plist
.items
; --entry
) {
4085 char *path
= entry
->string
;
4087 * NOTE: mi may actually be a pointer to a conflict_info, but
4088 * we have to check mi->clean first to see if it's safe to
4089 * reassign to such a pointer type.
4091 struct merged_info
*mi
= entry
->util
;
4093 write_completed_directory(opt
, mi
->directory_name
,
4096 record_entry_for_tree(&dir_metadata
, path
, mi
);
4098 struct conflict_info
*ci
= (struct conflict_info
*)mi
;
4099 process_entry(opt
, path
, ci
, &dir_metadata
);
4102 trace2_region_leave("merge", "processing", opt
->repo
);
4104 trace2_region_enter("merge", "process_entries cleanup", opt
->repo
);
4105 if (dir_metadata
.offsets
.nr
!= 1 ||
4106 (uintptr_t)dir_metadata
.offsets
.items
[0].util
!= 0) {
4107 printf("dir_metadata.offsets.nr = %d (should be 1)\n",
4108 dir_metadata
.offsets
.nr
);
4109 printf("dir_metadata.offsets.items[0].util = %u (should be 0)\n",
4110 (unsigned)(uintptr_t)dir_metadata
.offsets
.items
[0].util
);
4112 BUG("dir_metadata accounting completely off; shouldn't happen");
4114 write_tree(result_oid
, &dir_metadata
.versions
, 0,
4115 opt
->repo
->hash_algo
->rawsz
);
4116 string_list_clear(&plist
, 0);
4117 string_list_clear(&dir_metadata
.versions
, 0);
4118 string_list_clear(&dir_metadata
.offsets
, 0);
4119 trace2_region_leave("merge", "process_entries cleanup", opt
->repo
);
4122 /*** Function Grouping: functions related to merge_switch_to_result() ***/
4124 static int checkout(struct merge_options
*opt
,
4128 /* Switch the index/working copy from old to new */
4130 struct tree_desc trees
[2];
4131 struct unpack_trees_options unpack_opts
;
4133 memset(&unpack_opts
, 0, sizeof(unpack_opts
));
4134 unpack_opts
.head_idx
= -1;
4135 unpack_opts
.src_index
= opt
->repo
->index
;
4136 unpack_opts
.dst_index
= opt
->repo
->index
;
4138 setup_unpack_trees_porcelain(&unpack_opts
, "merge");
4141 * NOTE: if this were just "git checkout" code, we would probably
4142 * read or refresh the cache and check for a conflicted index, but
4143 * builtin/merge.c or sequencer.c really needs to read the index
4144 * and check for conflicted entries before starting merging for a
4145 * good user experience (no sense waiting for merges/rebases before
4146 * erroring out), so there's no reason to duplicate that work here.
4149 /* 2-way merge to the new branch */
4150 unpack_opts
.update
= 1;
4151 unpack_opts
.merge
= 1;
4152 unpack_opts
.quiet
= 0; /* FIXME: sequencer might want quiet? */
4153 unpack_opts
.verbose_update
= (opt
->verbosity
> 2);
4154 unpack_opts
.fn
= twoway_merge
;
4155 if (1/* FIXME: opts->overwrite_ignore*/) {
4156 CALLOC_ARRAY(unpack_opts
.dir
, 1);
4157 unpack_opts
.dir
->flags
|= DIR_SHOW_IGNORED
;
4158 setup_standard_excludes(unpack_opts
.dir
);
4161 init_tree_desc(&trees
[0], prev
->buffer
, prev
->size
);
4163 init_tree_desc(&trees
[1], next
->buffer
, next
->size
);
4165 ret
= unpack_trees(2, trees
, &unpack_opts
);
4166 clear_unpack_trees_porcelain(&unpack_opts
);
4167 dir_clear(unpack_opts
.dir
);
4168 FREE_AND_NULL(unpack_opts
.dir
);
4172 static int record_conflicted_index_entries(struct merge_options
*opt
)
4174 struct hashmap_iter iter
;
4175 struct strmap_entry
*e
;
4176 struct index_state
*index
= opt
->repo
->index
;
4177 struct checkout state
= CHECKOUT_INIT
;
4179 int original_cache_nr
;
4181 if (strmap_empty(&opt
->priv
->conflicted
))
4184 /* If any entries have skip_worktree set, we'll have to check 'em out */
4187 state
.refresh_cache
= 1;
4188 state
.istate
= index
;
4189 original_cache_nr
= index
->cache_nr
;
4191 /* Put every entry from paths into plist, then sort */
4192 strmap_for_each_entry(&opt
->priv
->conflicted
, &iter
, e
) {
4193 const char *path
= e
->key
;
4194 struct conflict_info
*ci
= e
->value
;
4196 struct cache_entry
*ce
;
4202 * The index will already have a stage=0 entry for this path,
4203 * because we created an as-merged-as-possible version of the
4204 * file and checkout() moved the working copy and index over
4207 * However, previous iterations through this loop will have
4208 * added unstaged entries to the end of the cache which
4209 * ignore the standard alphabetical ordering of cache
4210 * entries and break invariants needed for index_name_pos()
4211 * to work. However, we know the entry we want is before
4212 * those appended cache entries, so do a temporary swap on
4213 * cache_nr to only look through entries of interest.
4215 SWAP(index
->cache_nr
, original_cache_nr
);
4216 pos
= index_name_pos(index
, path
, strlen(path
));
4217 SWAP(index
->cache_nr
, original_cache_nr
);
4219 if (ci
->filemask
!= 1)
4220 BUG("Conflicted %s but nothing in basic working tree or index; this shouldn't happen", path
);
4221 cache_tree_invalidate_path(index
, path
);
4223 ce
= index
->cache
[pos
];
4226 * Clean paths with CE_SKIP_WORKTREE set will not be
4227 * written to the working tree by the unpack_trees()
4228 * call in checkout(). Our conflicted entries would
4229 * have appeared clean to that code since we ignored
4230 * the higher order stages. Thus, we need override
4231 * the CE_SKIP_WORKTREE bit and manually write those
4232 * files to the working disk here.
4234 if (ce_skip_worktree(ce
)) {
4237 if (!lstat(path
, &st
)) {
4238 char *new_name
= unique_path(&opt
->priv
->paths
,
4242 path_msg(opt
, path
, 1,
4243 _("Note: %s not up to date and in way of checking out conflicted version; old copy renamed to %s"),
4245 errs
|= rename(path
, new_name
);
4248 errs
|= checkout_entry(ce
, &state
, NULL
, NULL
);
4252 * Mark this cache entry for removal and instead add
4253 * new stage>0 entries corresponding to the
4254 * conflicts. If there are many conflicted entries, we
4255 * want to avoid memmove'ing O(NM) entries by
4256 * inserting the new entries one at a time. So,
4257 * instead, we just add the new cache entries to the
4258 * end (ignoring normal index requirements on sort
4259 * order) and sort the index once we're all done.
4261 ce
->ce_flags
|= CE_REMOVE
;
4264 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
4265 struct version_info
*vi
;
4266 if (!(ci
->filemask
& (1ul << i
)))
4268 vi
= &ci
->stages
[i
];
4269 ce
= make_cache_entry(index
, vi
->mode
, &vi
->oid
,
4271 add_index_entry(index
, ce
, ADD_CACHE_JUST_APPEND
);
4276 * Remove the unused cache entries (and invalidate the relevant
4277 * cache-trees), then sort the index entries to get the conflicted
4278 * entries we added to the end into their right locations.
4280 remove_marked_cache_entries(index
, 1);
4282 * No need for STABLE_QSORT -- cmp_cache_name_compare sorts primarily
4283 * on filename and secondarily on stage, and (name, stage #) are a
4286 QSORT(index
->cache
, index
->cache_nr
, cmp_cache_name_compare
);
4291 void merge_switch_to_result(struct merge_options
*opt
,
4293 struct merge_result
*result
,
4294 int update_worktree_and_index
,
4295 int display_update_msgs
)
4297 assert(opt
->priv
== NULL
);
4298 if (result
->clean
>= 0 && update_worktree_and_index
) {
4299 const char *filename
;
4302 trace2_region_enter("merge", "checkout", opt
->repo
);
4303 if (checkout(opt
, head
, result
->tree
)) {
4304 /* failure to function */
4308 trace2_region_leave("merge", "checkout", opt
->repo
);
4310 trace2_region_enter("merge", "record_conflicted", opt
->repo
);
4311 opt
->priv
= result
->priv
;
4312 if (record_conflicted_index_entries(opt
)) {
4313 /* failure to function */
4319 trace2_region_leave("merge", "record_conflicted", opt
->repo
);
4321 trace2_region_enter("merge", "write_auto_merge", opt
->repo
);
4322 filename
= git_path_auto_merge(opt
->repo
);
4323 fp
= xfopen(filename
, "w");
4324 fprintf(fp
, "%s\n", oid_to_hex(&result
->tree
->object
.oid
));
4326 trace2_region_leave("merge", "write_auto_merge", opt
->repo
);
4329 if (display_update_msgs
) {
4330 struct merge_options_internal
*opti
= result
->priv
;
4331 struct hashmap_iter iter
;
4332 struct strmap_entry
*e
;
4333 struct string_list olist
= STRING_LIST_INIT_NODUP
;
4336 trace2_region_enter("merge", "display messages", opt
->repo
);
4338 /* Hack to pre-allocate olist to the desired size */
4339 ALLOC_GROW(olist
.items
, strmap_get_size(&opti
->output
),
4342 /* Put every entry from output into olist, then sort */
4343 strmap_for_each_entry(&opti
->output
, &iter
, e
) {
4344 string_list_append(&olist
, e
->key
)->util
= e
->value
;
4346 string_list_sort(&olist
);
4348 /* Iterate over the items, printing them */
4349 for (i
= 0; i
< olist
.nr
; ++i
) {
4350 struct strbuf
*sb
= olist
.items
[i
].util
;
4352 printf("%s", sb
->buf
);
4354 string_list_clear(&olist
, 0);
4356 /* Also include needed rename limit adjustment now */
4357 diff_warn_rename_limit("merge.renamelimit",
4358 opti
->renames
.needed_limit
, 0);
4360 trace2_region_leave("merge", "display messages", opt
->repo
);
4363 merge_finalize(opt
, result
);
4366 void merge_finalize(struct merge_options
*opt
,
4367 struct merge_result
*result
)
4369 struct merge_options_internal
*opti
= result
->priv
;
4371 if (opt
->renormalize
)
4372 git_attr_set_direction(GIT_ATTR_CHECKIN
);
4373 assert(opt
->priv
== NULL
);
4375 clear_or_reinit_internal_opts(opti
, 0);
4376 FREE_AND_NULL(opti
);
4379 /*** Function Grouping: helper functions for merge_incore_*() ***/
4381 static struct tree
*shift_tree_object(struct repository
*repo
,
4382 struct tree
*one
, struct tree
*two
,
4383 const char *subtree_shift
)
4385 struct object_id shifted
;
4387 if (!*subtree_shift
) {
4388 shift_tree(repo
, &one
->object
.oid
, &two
->object
.oid
, &shifted
, 0);
4390 shift_tree_by(repo
, &one
->object
.oid
, &two
->object
.oid
, &shifted
,
4393 if (oideq(&two
->object
.oid
, &shifted
))
4395 return lookup_tree(repo
, &shifted
);
4398 static inline void set_commit_tree(struct commit
*c
, struct tree
*t
)
4403 static struct commit
*make_virtual_commit(struct repository
*repo
,
4405 const char *comment
)
4407 struct commit
*commit
= alloc_commit_node(repo
);
4409 set_merge_remote_desc(commit
, comment
, (struct object
*)commit
);
4410 set_commit_tree(commit
, tree
);
4411 commit
->object
.parsed
= 1;
4415 static void merge_start(struct merge_options
*opt
, struct merge_result
*result
)
4417 struct rename_info
*renames
;
4419 struct mem_pool
*pool
= NULL
;
4421 /* Sanity checks on opt */
4422 trace2_region_enter("merge", "sanity checks", opt
->repo
);
4425 assert(opt
->branch1
&& opt
->branch2
);
4427 assert(opt
->detect_directory_renames
>= MERGE_DIRECTORY_RENAMES_NONE
&&
4428 opt
->detect_directory_renames
<= MERGE_DIRECTORY_RENAMES_TRUE
);
4429 assert(opt
->rename_limit
>= -1);
4430 assert(opt
->rename_score
>= 0 && opt
->rename_score
<= MAX_SCORE
);
4431 assert(opt
->show_rename_progress
>= 0 && opt
->show_rename_progress
<= 1);
4433 assert(opt
->xdl_opts
>= 0);
4434 assert(opt
->recursive_variant
>= MERGE_VARIANT_NORMAL
&&
4435 opt
->recursive_variant
<= MERGE_VARIANT_THEIRS
);
4438 * detect_renames, verbosity, buffer_output, and obuf are ignored
4439 * fields that were used by "recursive" rather than "ort" -- but
4440 * sanity check them anyway.
4442 assert(opt
->detect_renames
>= -1 &&
4443 opt
->detect_renames
<= DIFF_DETECT_COPY
);
4444 assert(opt
->verbosity
>= 0 && opt
->verbosity
<= 5);
4445 assert(opt
->buffer_output
<= 2);
4446 assert(opt
->obuf
.len
== 0);
4448 assert(opt
->priv
== NULL
);
4449 if (result
->_properly_initialized
!= 0 &&
4450 result
->_properly_initialized
!= RESULT_INITIALIZED
)
4451 BUG("struct merge_result passed to merge_incore_*recursive() must be zeroed or filled with values from a previous run");
4452 assert(!!result
->priv
== !!result
->_properly_initialized
);
4454 opt
->priv
= result
->priv
;
4455 result
->priv
= NULL
;
4457 * opt->priv non-NULL means we had results from a previous
4458 * run; do a few sanity checks that user didn't mess with
4459 * it in an obvious fashion.
4461 assert(opt
->priv
->call_depth
== 0);
4462 assert(!opt
->priv
->toplevel_dir
||
4463 0 == strlen(opt
->priv
->toplevel_dir
));
4465 trace2_region_leave("merge", "sanity checks", opt
->repo
);
4467 /* Default to histogram diff. Actually, just hardcode it...for now. */
4468 opt
->xdl_opts
= DIFF_WITH_ALG(opt
, HISTOGRAM_DIFF
);
4470 /* Handle attr direction stuff for renormalization */
4471 if (opt
->renormalize
)
4472 git_attr_set_direction(GIT_ATTR_CHECKOUT
);
4474 /* Initialization of opt->priv, our internal merge data */
4475 trace2_region_enter("merge", "allocate/init", opt
->repo
);
4477 clear_or_reinit_internal_opts(opt
->priv
, 1);
4478 trace2_region_leave("merge", "allocate/init", opt
->repo
);
4481 opt
->priv
= xcalloc(1, sizeof(*opt
->priv
));
4483 /* Initialization of various renames fields */
4484 renames
= &opt
->priv
->renames
;
4486 mem_pool_init(&opt
->priv
->internal_pool
, 0);
4487 opt
->priv
->pool
= &opt
->priv
->internal_pool
;
4489 opt
->priv
->pool
= NULL
;
4491 pool
= opt
->priv
->pool
;
4492 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; i
++) {
4493 strintmap_init_with_options(&renames
->dirs_removed
[i
],
4494 NOT_RELEVANT
, pool
, 0);
4495 strmap_init_with_options(&renames
->dir_rename_count
[i
],
4497 strmap_init_with_options(&renames
->dir_renames
[i
],
4500 * relevant_sources uses -1 for the default, because we need
4501 * to be able to distinguish not-in-strintmap from valid
4502 * relevant_source values from enum file_rename_relevance.
4503 * In particular, possibly_cache_new_pair() expects a negative
4504 * value for not-found entries.
4506 strintmap_init_with_options(&renames
->relevant_sources
[i
],
4507 -1 /* explicitly invalid */,
4509 strmap_init_with_options(&renames
->cached_pairs
[i
],
4511 strset_init_with_options(&renames
->cached_irrelevant
[i
],
4513 strset_init_with_options(&renames
->cached_target_names
[i
],
4516 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; i
++) {
4517 strintmap_init_with_options(&renames
->deferred
[i
].possible_trivial_merges
,
4519 strset_init_with_options(&renames
->deferred
[i
].target_dirs
,
4521 renames
->deferred
[i
].trivial_merges_okay
= 1; /* 1 == maybe */
4525 * Although we initialize opt->priv->paths with strdup_strings=0,
4526 * that's just to avoid making yet another copy of an allocated
4527 * string. Putting the entry into paths means we are taking
4528 * ownership, so we will later free it. paths_to_free is similar.
4530 * In contrast, conflicted just has a subset of keys from paths, so
4531 * we don't want to free those (it'd be a duplicate free).
4533 strmap_init_with_options(&opt
->priv
->paths
, pool
, 0);
4534 strmap_init_with_options(&opt
->priv
->conflicted
, pool
, 0);
4535 if (!opt
->priv
->pool
)
4536 string_list_init_nodup(&opt
->priv
->paths_to_free
);
4539 * keys & strbufs in output will sometimes need to outlive "paths",
4540 * so it will have a copy of relevant keys. It's probably a small
4541 * subset of the overall paths that have special output.
4543 strmap_init(&opt
->priv
->output
);
4545 trace2_region_leave("merge", "allocate/init", opt
->repo
);
4548 static void merge_check_renames_reusable(struct merge_options
*opt
,
4549 struct merge_result
*result
,
4550 struct tree
*merge_base
,
4554 struct rename_info
*renames
;
4555 struct tree
**merge_trees
;
4556 struct merge_options_internal
*opti
= result
->priv
;
4561 renames
= &opti
->renames
;
4562 merge_trees
= renames
->merge_trees
;
4565 * Handle case where previous merge operation did not want cache to
4566 * take effect, e.g. because rename/rename(1to1) makes it invalid.
4568 if (!merge_trees
[0]) {
4569 assert(!merge_trees
[0] && !merge_trees
[1] && !merge_trees
[2]);
4570 renames
->cached_pairs_valid_side
= 0; /* neither side valid */
4575 * Handle other cases; note that merge_trees[0..2] will only
4576 * be NULL if opti is, or if all three were manually set to
4577 * NULL by e.g. rename/rename(1to1) handling.
4579 assert(merge_trees
[0] && merge_trees
[1] && merge_trees
[2]);
4581 /* Check if we meet a condition for re-using cached_pairs */
4582 if (oideq(&merge_base
->object
.oid
, &merge_trees
[2]->object
.oid
) &&
4583 oideq(&side1
->object
.oid
, &result
->tree
->object
.oid
))
4584 renames
->cached_pairs_valid_side
= MERGE_SIDE1
;
4585 else if (oideq(&merge_base
->object
.oid
, &merge_trees
[1]->object
.oid
) &&
4586 oideq(&side2
->object
.oid
, &result
->tree
->object
.oid
))
4587 renames
->cached_pairs_valid_side
= MERGE_SIDE2
;
4589 renames
->cached_pairs_valid_side
= 0; /* neither side valid */
4592 /*** Function Grouping: merge_incore_*() and their internal variants ***/
4595 * Originally from merge_trees_internal(); heavily adapted, though.
4597 static void merge_ort_nonrecursive_internal(struct merge_options
*opt
,
4598 struct tree
*merge_base
,
4601 struct merge_result
*result
)
4603 struct object_id working_tree_oid
;
4605 if (opt
->subtree_shift
) {
4606 side2
= shift_tree_object(opt
->repo
, side1
, side2
,
4607 opt
->subtree_shift
);
4608 merge_base
= shift_tree_object(opt
->repo
, side1
, merge_base
,
4609 opt
->subtree_shift
);
4613 trace2_region_enter("merge", "collect_merge_info", opt
->repo
);
4614 if (collect_merge_info(opt
, merge_base
, side1
, side2
) != 0) {
4616 * TRANSLATORS: The %s arguments are: 1) tree hash of a merge
4617 * base, and 2-3) the trees for the two trees we're merging.
4619 err(opt
, _("collecting merge info failed for trees %s, %s, %s"),
4620 oid_to_hex(&merge_base
->object
.oid
),
4621 oid_to_hex(&side1
->object
.oid
),
4622 oid_to_hex(&side2
->object
.oid
));
4626 trace2_region_leave("merge", "collect_merge_info", opt
->repo
);
4628 trace2_region_enter("merge", "renames", opt
->repo
);
4629 result
->clean
= detect_and_process_renames(opt
, merge_base
,
4631 trace2_region_leave("merge", "renames", opt
->repo
);
4632 if (opt
->priv
->renames
.redo_after_renames
== 2) {
4633 trace2_region_enter("merge", "reset_maps", opt
->repo
);
4634 clear_or_reinit_internal_opts(opt
->priv
, 1);
4635 trace2_region_leave("merge", "reset_maps", opt
->repo
);
4639 trace2_region_enter("merge", "process_entries", opt
->repo
);
4640 process_entries(opt
, &working_tree_oid
);
4641 trace2_region_leave("merge", "process_entries", opt
->repo
);
4643 /* Set return values */
4644 result
->tree
= parse_tree_indirect(&working_tree_oid
);
4645 /* existence of conflicted entries implies unclean */
4646 result
->clean
&= strmap_empty(&opt
->priv
->conflicted
);
4647 if (!opt
->priv
->call_depth
) {
4648 result
->priv
= opt
->priv
;
4649 result
->_properly_initialized
= RESULT_INITIALIZED
;
4655 * Originally from merge_recursive_internal(); somewhat adapted, though.
4657 static void merge_ort_internal(struct merge_options
*opt
,
4658 struct commit_list
*merge_bases
,
4661 struct merge_result
*result
)
4663 struct commit_list
*iter
;
4664 struct commit
*merged_merge_bases
;
4665 const char *ancestor_name
;
4666 struct strbuf merge_base_abbrev
= STRBUF_INIT
;
4669 merge_bases
= get_merge_bases(h1
, h2
);
4670 /* See merge-ort.h:merge_incore_recursive() declaration NOTE */
4671 merge_bases
= reverse_commit_list(merge_bases
);
4674 merged_merge_bases
= pop_commit(&merge_bases
);
4675 if (merged_merge_bases
== NULL
) {
4676 /* if there is no common ancestor, use an empty tree */
4679 tree
= lookup_tree(opt
->repo
, opt
->repo
->hash_algo
->empty_tree
);
4680 merged_merge_bases
= make_virtual_commit(opt
->repo
, tree
,
4682 ancestor_name
= "empty tree";
4683 } else if (merge_bases
) {
4684 ancestor_name
= "merged common ancestors";
4686 strbuf_add_unique_abbrev(&merge_base_abbrev
,
4687 &merged_merge_bases
->object
.oid
,
4689 ancestor_name
= merge_base_abbrev
.buf
;
4692 for (iter
= merge_bases
; iter
; iter
= iter
->next
) {
4693 const char *saved_b1
, *saved_b2
;
4694 struct commit
*prev
= merged_merge_bases
;
4696 opt
->priv
->call_depth
++;
4698 * When the merge fails, the result contains files
4699 * with conflict markers. The cleanness flag is
4700 * ignored (unless indicating an error), it was never
4701 * actually used, as result of merge_trees has always
4702 * overwritten it: the committed "conflicts" were
4705 saved_b1
= opt
->branch1
;
4706 saved_b2
= opt
->branch2
;
4707 opt
->branch1
= "Temporary merge branch 1";
4708 opt
->branch2
= "Temporary merge branch 2";
4709 merge_ort_internal(opt
, NULL
, prev
, iter
->item
, result
);
4710 if (result
->clean
< 0)
4712 opt
->branch1
= saved_b1
;
4713 opt
->branch2
= saved_b2
;
4714 opt
->priv
->call_depth
--;
4716 merged_merge_bases
= make_virtual_commit(opt
->repo
,
4719 commit_list_insert(prev
, &merged_merge_bases
->parents
);
4720 commit_list_insert(iter
->item
,
4721 &merged_merge_bases
->parents
->next
);
4723 clear_or_reinit_internal_opts(opt
->priv
, 1);
4726 opt
->ancestor
= ancestor_name
;
4727 merge_ort_nonrecursive_internal(opt
,
4728 repo_get_commit_tree(opt
->repo
,
4729 merged_merge_bases
),
4730 repo_get_commit_tree(opt
->repo
, h1
),
4731 repo_get_commit_tree(opt
->repo
, h2
),
4733 strbuf_release(&merge_base_abbrev
);
4734 opt
->ancestor
= NULL
; /* avoid accidental re-use of opt->ancestor */
4737 void merge_incore_nonrecursive(struct merge_options
*opt
,
4738 struct tree
*merge_base
,
4741 struct merge_result
*result
)
4743 trace2_region_enter("merge", "incore_nonrecursive", opt
->repo
);
4745 trace2_region_enter("merge", "merge_start", opt
->repo
);
4746 assert(opt
->ancestor
!= NULL
);
4747 merge_check_renames_reusable(opt
, result
, merge_base
, side1
, side2
);
4748 merge_start(opt
, result
);
4750 * Record the trees used in this merge, so if there's a next merge in
4751 * a cherry-pick or rebase sequence it might be able to take advantage
4752 * of the cached_pairs in that next merge.
4754 opt
->priv
->renames
.merge_trees
[0] = merge_base
;
4755 opt
->priv
->renames
.merge_trees
[1] = side1
;
4756 opt
->priv
->renames
.merge_trees
[2] = side2
;
4757 trace2_region_leave("merge", "merge_start", opt
->repo
);
4759 merge_ort_nonrecursive_internal(opt
, merge_base
, side1
, side2
, result
);
4760 trace2_region_leave("merge", "incore_nonrecursive", opt
->repo
);
4763 void merge_incore_recursive(struct merge_options
*opt
,
4764 struct commit_list
*merge_bases
,
4765 struct commit
*side1
,
4766 struct commit
*side2
,
4767 struct merge_result
*result
)
4769 trace2_region_enter("merge", "incore_recursive", opt
->repo
);
4771 /* We set the ancestor label based on the merge_bases */
4772 assert(opt
->ancestor
== NULL
);
4774 trace2_region_enter("merge", "merge_start", opt
->repo
);
4775 merge_start(opt
, result
);
4776 trace2_region_leave("merge", "merge_start", opt
->repo
);
4778 merge_ort_internal(opt
, merge_bases
, side1
, side2
, result
);
4779 trace2_region_leave("merge", "incore_recursive", opt
->repo
);