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"?)
17 #include "git-compat-util.h"
18 #include "merge-ort.h"
23 #include "cache-tree.h"
25 #include "commit-reach.h"
29 #include "environment.h"
34 #include "match-trees.h"
36 #include "object-name.h"
37 #include "object-store-ll.h"
38 #include "oid-array.h"
40 #include "promisor-remote.h"
41 #include "read-cache-ll.h"
43 #include "sparse-index.h"
45 #include "submodule-config.h"
46 #include "submodule.h"
49 #include "unpack-trees.h"
50 #include "xdiff-interface.h"
53 * We have many arrays of size 3. Whenever we have such an array, the
54 * indices refer to one of the sides of the three-way merge. This is so
55 * pervasive that the constants 0, 1, and 2 are used in many places in the
56 * code (especially in arithmetic operations to find the other side's index
57 * or to compute a relevant mask), but sometimes these enum names are used
58 * to aid code clarity.
60 * See also 'filemask' and 'dirmask' in struct conflict_info; the "ith side"
61 * referred to there is one of these three sides.
69 static unsigned RESULT_INITIALIZED
= 0x1abe11ed; /* unlikely accidental value */
71 struct traversal_callback_data
{
73 unsigned long dirmask
;
74 struct name_entry names
[3];
77 struct deferred_traversal_data
{
79 * possible_trivial_merges: directories to be explored only when needed
81 * possible_trivial_merges is a map of directory names to
82 * dir_rename_mask. When we detect that a directory is unchanged on
83 * one side, we can sometimes resolve the directory without recursing
84 * into it. Renames are the only things that can prevent such an
85 * optimization. However, for rename sources:
86 * - If no parent directory needed directory rename detection, then
87 * no path under such a directory can be a relevant_source.
88 * and for rename destinations:
89 * - If no cached rename has a target path under the directory AND
90 * - If there are no unpaired relevant_sources elsewhere in the
92 * then we don't need any path under this directory for a rename
93 * destination. The only way to know the last item above is to defer
94 * handling such directories until the end of collect_merge_info(),
95 * in handle_deferred_entries().
97 * For each we store dir_rename_mask, since that's the only bit of
98 * information we need, other than the path, to resume the recursive
101 struct strintmap possible_trivial_merges
;
104 * trivial_merges_okay: if trivial directory merges are okay
106 * See possible_trivial_merges above. The "no unpaired
107 * relevant_sources elsewhere in the repository" is a single boolean
108 * per merge side, which we store here. Note that while 0 means no,
109 * 1 only means "maybe" rather than "yes"; we optimistically set it
110 * to 1 initially and only clear when we determine it is unsafe to
111 * do trivial directory merges.
113 unsigned trivial_merges_okay
;
116 * target_dirs: ancestor directories of rename targets
118 * target_dirs contains all directory names that are an ancestor of
119 * any rename destination.
121 struct strset target_dirs
;
126 * All variables that are arrays of size 3 correspond to data tracked
127 * for the sides in enum merge_side. Index 0 is almost always unused
128 * because we often only need to track information for MERGE_SIDE1 and
129 * MERGE_SIDE2 (MERGE_BASE can't have rename information since renames
130 * are determined relative to what changed since the MERGE_BASE).
134 * pairs: pairing of filenames from diffcore_rename()
136 struct diff_queue_struct pairs
[3];
139 * dirs_removed: directories removed on a given side of history.
141 * The keys of dirs_removed[side] are the directories that were removed
142 * on the given side of history. The value of the strintmap for each
143 * directory is a value from enum dir_rename_relevance.
145 struct strintmap dirs_removed
[3];
148 * dir_rename_count: tracking where parts of a directory were renamed to
150 * When files in a directory are renamed, they may not all go to the
151 * same location. Each strmap here tracks:
152 * old_dir => {new_dir => int}
153 * That is, dir_rename_count[side] is a strmap to a strintmap.
155 struct strmap dir_rename_count
[3];
158 * dir_renames: computed directory renames
160 * This is a map of old_dir => new_dir and is derived in part from
163 struct strmap dir_renames
[3];
166 * relevant_sources: deleted paths wanted in rename detection, and why
168 * relevant_sources is a set of deleted paths on each side of
169 * history for which we need rename detection. If a path is deleted
170 * on one side of history, we need to detect if it is part of a
172 * * the file is modified/deleted on the other side of history
173 * * we need to detect renames for an ancestor directory
174 * If neither of those are true, we can skip rename detection for
175 * that path. The reason is stored as a value from enum
176 * file_rename_relevance, as the reason can inform the algorithm in
177 * diffcore_rename_extended().
179 struct strintmap relevant_sources
[3];
181 struct deferred_traversal_data deferred
[3];
185 * 0: optimization removing unmodified potential rename source okay
186 * 2 or 4: optimization okay, but must check for files added to dir
187 * 7: optimization forbidden; need rename source in case of dir rename
189 unsigned dir_rename_mask
:3;
192 * callback_data_*: supporting data structures for alternate traversal
194 * We sometimes need to be able to traverse through all the files
195 * in a given tree before all immediate subdirectories within that
196 * tree. Since traverse_trees() doesn't do that naturally, we have
197 * a traverse_trees_wrapper() that stores any immediate
198 * subdirectories while traversing files, then traverses the
199 * immediate subdirectories later. These callback_data* variables
200 * store the information for the subdirectories so that we can do
201 * that traversal order.
203 struct traversal_callback_data
*callback_data
;
204 int callback_data_nr
, callback_data_alloc
;
205 char *callback_data_traverse_path
;
208 * merge_trees: trees passed to the merge algorithm for the merge
210 * merge_trees records the trees passed to the merge algorithm. But,
211 * this data also is stored in merge_result->priv. If a sequence of
212 * merges are being done (such as when cherry-picking or rebasing),
213 * the next merge can look at this and re-use information from
214 * previous merges under certain circumstances.
216 * See also all the cached_* variables.
218 struct tree
*merge_trees
[3];
221 * cached_pairs_valid_side: which side's cached info can be reused
223 * See the description for merge_trees. For repeated merges, at most
224 * only one side's cached information can be used. Valid values:
225 * MERGE_SIDE2: cached data from side2 can be reused
226 * MERGE_SIDE1: cached data from side1 can be reused
227 * 0: no cached data can be reused
228 * -1: See redo_after_renames; both sides can be reused.
230 int cached_pairs_valid_side
;
233 * cached_pairs: Caching of renames and deletions.
235 * These are mappings recording renames and deletions of individual
236 * files (not directories). They are thus a map from an old
237 * filename to either NULL (for deletions) or a new filename (for
240 struct strmap cached_pairs
[3];
243 * cached_target_names: just the destinations from cached_pairs
245 * We sometimes want a fast lookup to determine if a given filename
246 * is one of the destinations in cached_pairs. cached_target_names
247 * is thus duplicative information, but it provides a fast lookup.
249 struct strset cached_target_names
[3];
252 * cached_irrelevant: Caching of rename_sources that aren't relevant.
254 * If we try to detect a rename for a source path and succeed, it's
255 * part of a rename. If we try to detect a rename for a source path
256 * and fail, then it's a delete. If we do not try to detect a rename
257 * for a path, then we don't know if it's a rename or a delete. If
258 * merge-ort doesn't think the path is relevant, then we just won't
259 * cache anything for that path. But there's a slight problem in
260 * that merge-ort can think a path is RELEVANT_LOCATION, but due to
261 * commit 9bd342137e ("diffcore-rename: determine which
262 * relevant_sources are no longer relevant", 2021-03-13),
263 * diffcore-rename can downgrade the path to RELEVANT_NO_MORE. To
264 * avoid excessive calls to diffcore_rename_extended() we still need
265 * to cache such paths, though we cannot record them as either
266 * renames or deletes. So we cache them here as a "turned out to be
267 * irrelevant *for this commit*" as they are often also irrelevant
268 * for subsequent commits, though we will have to do some extra
269 * checking to see whether such paths become relevant for rename
270 * detection when cherry-picking/rebasing subsequent commits.
272 struct strset cached_irrelevant
[3];
275 * redo_after_renames: optimization flag for "restarting" the merge
277 * Sometimes it pays to detect renames, cache them, and then
278 * restart the merge operation from the beginning. The reason for
279 * this is that when we know where all the renames are, we know
280 * whether a certain directory has any paths under it affected --
281 * and if a directory is not affected then it permits us to do
282 * trivial tree merging in more cases. Doing trivial tree merging
283 * prevents the need to run process_entry() on every path
284 * underneath trees that can be trivially merged, and
285 * process_entry() is more expensive than collect_merge_info() --
286 * plus, the second collect_merge_info() will be much faster since
287 * it doesn't have to recurse into the relevant trees.
289 * Values for this flag:
290 * 0 = don't bother, not worth it (or conditions not yet checked)
291 * 1 = conditions for optimization met, optimization worthwhile
292 * 2 = we already did it (don't restart merge yet again)
294 unsigned redo_after_renames
;
297 * needed_limit: value needed for inexact rename detection to run
299 * If the current rename limit wasn't high enough for inexact
300 * rename detection to run, this records the limit needed. Otherwise,
301 * this value remains 0.
306 struct merge_options_internal
{
308 * paths: primary data structure in all of merge ort.
311 * * are full relative paths from the toplevel of the repository
312 * (e.g. "drivers/firmware/raspberrypi.c").
313 * * store all relevant paths in the repo, both directories and
314 * files (e.g. drivers, drivers/firmware would also be included)
315 * * these keys serve to intern all the path strings, which allows
316 * us to do pointer comparison on directory names instead of
317 * strcmp; we just have to be careful to use the interned strings.
319 * The values of paths:
320 * * either a pointer to a merged_info, or a conflict_info struct
321 * * merged_info contains all relevant information for a
322 * non-conflicted entry.
323 * * conflict_info contains a merged_info, plus any additional
324 * information about a conflict such as the higher orders stages
325 * involved and the names of the paths those came from (handy
326 * once renames get involved).
327 * * a path may start "conflicted" (i.e. point to a conflict_info)
328 * and then a later step (e.g. three-way content merge) determines
329 * it can be cleanly merged, at which point it'll be marked clean
330 * and the algorithm will ignore any data outside the contained
331 * merged_info for that entry
332 * * If an entry remains conflicted, the merged_info portion of a
333 * conflict_info will later be filled with whatever version of
334 * the file should be placed in the working directory (e.g. an
335 * as-merged-as-possible variation that contains conflict markers).
340 * conflicted: a subset of keys->values from "paths"
342 * conflicted is basically an optimization between process_entries()
343 * and record_conflicted_index_entries(); the latter could loop over
344 * ALL the entries in paths AGAIN and look for the ones that are
345 * still conflicted, but since process_entries() has to loop over
346 * all of them, it saves the ones it couldn't resolve in this strmap
347 * so that record_conflicted_index_entries() can iterate just the
350 struct strmap conflicted
;
353 * pool: memory pool for fast allocation/deallocation
355 * We allocate room for lots of filenames and auxiliary data
356 * structures in merge_options_internal, and it tends to all be
357 * freed together too. Using a memory pool for these provides a
360 struct mem_pool pool
;
363 * conflicts: logical conflicts and messages stored by _primary_ path
365 * This is a map of pathnames (a subset of the keys in "paths" above)
366 * to struct string_list, with each item's `util` containing a
367 * `struct logical_conflict_info`. Note, though, that for each path,
368 * it only stores the logical conflicts for which that path is the
369 * primary path; the path might be part of additional conflicts.
371 struct strmap conflicts
;
374 * renames: various data relating to rename detection
376 struct rename_info renames
;
379 * attr_index: hacky minimal index used for renormalization
381 * renormalization code _requires_ an index, though it only needs to
382 * find a .gitattributes file within the index. So, when
383 * renormalization is important, we create a special index with just
386 struct index_state attr_index
;
389 * current_dir_name, toplevel_dir: temporary vars
391 * These are used in collect_merge_info_callback(), and will set the
392 * various merged_info.directory_name for the various paths we get;
393 * see documentation for that variable and the requirements placed on
396 const char *current_dir_name
;
397 const char *toplevel_dir
;
399 /* call_depth: recursion level counter for merging merge bases */
402 /* field that holds submodule conflict information */
403 struct string_list conflicted_submodules
;
406 struct conflicted_submodule_item
{
411 static void conflicted_submodule_item_free(void *util
, const char *str UNUSED
)
413 struct conflicted_submodule_item
*item
= util
;
419 struct version_info
{
420 struct object_id oid
;
425 /* if is_null, ignore result. otherwise result has oid & mode */
426 struct version_info result
;
430 * clean: whether the path in question is cleanly merged.
432 * see conflict_info.merged for more details.
437 * basename_offset: offset of basename of path.
439 * perf optimization to avoid recomputing offset of final '/'
440 * character in pathname (0 if no '/' in pathname).
442 size_t basename_offset
;
445 * directory_name: containing directory name.
447 * Note that we assume directory_name is constructed such that
448 * strcmp(dir1_name, dir2_name) == 0 iff dir1_name == dir2_name,
449 * i.e. string equality is equivalent to pointer equality. For this
450 * to hold, we have to be careful setting directory_name.
452 const char *directory_name
;
455 struct conflict_info
{
457 * merged: the version of the path that will be written to working tree
459 * WARNING: It is critical to check merged.clean and ensure it is 0
460 * before reading any conflict_info fields outside of merged.
461 * Allocated merge_info structs will always have clean set to 1.
462 * Allocated conflict_info structs will have merged.clean set to 0
463 * initially. The merged.clean field is how we know if it is safe
464 * to access other parts of conflict_info besides merged; if a
465 * conflict_info's merged.clean is changed to 1, the rest of the
466 * algorithm is not allowed to look at anything outside of the
467 * merged member anymore.
469 struct merged_info merged
;
471 /* oids & modes from each of the three trees for this path */
472 struct version_info stages
[3];
474 /* pathnames for each stage; may differ due to rename detection */
475 const char *pathnames
[3];
477 /* Whether this path is/was involved in a directory/file conflict */
478 unsigned df_conflict
:1;
481 * Whether this path is/was involved in a non-content conflict other
482 * than a directory/file conflict (e.g. rename/rename, rename/delete,
483 * file location based on possible directory rename).
485 unsigned path_conflict
:1;
488 * For filemask and dirmask, the ith bit corresponds to whether the
489 * ith entry is a file (filemask) or a directory (dirmask). Thus,
490 * filemask & dirmask is always zero, and filemask | dirmask is at
491 * most 7 but can be less when a path does not appear as either a
492 * file or a directory on at least one side of history.
494 * Note that these masks are related to enum merge_side, as the ith
495 * entry corresponds to side i.
497 * These values come from a traverse_trees() call; more info may be
498 * found looking at tree-walk.h's struct traverse_info,
499 * particularly the documentation above the "fn" member (note that
500 * filemask = mask & ~dirmask from that documentation).
506 * Optimization to track which stages match, to avoid the need to
507 * recompute it in multiple steps. Either 0 or at least 2 bits are
508 * set; if at least 2 bits are set, their corresponding stages match.
510 unsigned match_mask
:3;
513 enum conflict_and_info_types
{
514 /* "Simple" conflicts and informational messages */
515 INFO_AUTO_MERGING
= 0,
516 CONFLICT_CONTENTS
, /* text file that failed to merge */
518 CONFLICT_FILE_DIRECTORY
,
519 CONFLICT_DISTINCT_MODES
,
520 CONFLICT_MODIFY_DELETE
,
523 CONFLICT_RENAME_RENAME
, /* same file renamed differently */
524 CONFLICT_RENAME_COLLIDES
, /* rename/add or two files renamed to 1 */
525 CONFLICT_RENAME_DELETE
,
527 /* Basic directory rename */
528 CONFLICT_DIR_RENAME_SUGGESTED
,
529 INFO_DIR_RENAME_APPLIED
,
531 /* Special directory rename cases */
532 INFO_DIR_RENAME_SKIPPED_DUE_TO_RERENAME
,
533 CONFLICT_DIR_RENAME_FILE_IN_WAY
,
534 CONFLICT_DIR_RENAME_COLLISION
,
535 CONFLICT_DIR_RENAME_SPLIT
,
537 /* Basic submodule */
538 INFO_SUBMODULE_FAST_FORWARDING
,
539 CONFLICT_SUBMODULE_FAILED_TO_MERGE
,
541 /* Special submodule cases broken out from FAILED_TO_MERGE */
542 CONFLICT_SUBMODULE_FAILED_TO_MERGE_BUT_POSSIBLE_RESOLUTION
,
543 CONFLICT_SUBMODULE_NOT_INITIALIZED
,
544 CONFLICT_SUBMODULE_HISTORY_NOT_AVAILABLE
,
545 CONFLICT_SUBMODULE_MAY_HAVE_REWINDS
,
546 CONFLICT_SUBMODULE_NULL_MERGE_BASE
,
548 /* Keep this entry _last_ in the list */
553 * Short description of conflict type, relied upon by external tools.
555 * We can add more entries, but DO NOT change any of these strings. Also,
556 * Order MUST match conflict_info_and_types.
558 static const char *type_short_descriptions
[] = {
559 /*** "Simple" conflicts and informational messages ***/
560 [INFO_AUTO_MERGING
] = "Auto-merging",
561 [CONFLICT_CONTENTS
] = "CONFLICT (contents)",
562 [CONFLICT_BINARY
] = "CONFLICT (binary)",
563 [CONFLICT_FILE_DIRECTORY
] = "CONFLICT (file/directory)",
564 [CONFLICT_DISTINCT_MODES
] = "CONFLICT (distinct modes)",
565 [CONFLICT_MODIFY_DELETE
] = "CONFLICT (modify/delete)",
567 /*** Regular rename ***/
568 [CONFLICT_RENAME_RENAME
] = "CONFLICT (rename/rename)",
569 [CONFLICT_RENAME_COLLIDES
] = "CONFLICT (rename involved in collision)",
570 [CONFLICT_RENAME_DELETE
] = "CONFLICT (rename/delete)",
572 /*** Basic directory rename ***/
573 [CONFLICT_DIR_RENAME_SUGGESTED
] =
574 "CONFLICT (directory rename suggested)",
575 [INFO_DIR_RENAME_APPLIED
] = "Path updated due to directory rename",
577 /*** Special directory rename cases ***/
578 [INFO_DIR_RENAME_SKIPPED_DUE_TO_RERENAME
] =
579 "Directory rename skipped since directory was renamed on both sides",
580 [CONFLICT_DIR_RENAME_FILE_IN_WAY
] =
581 "CONFLICT (file in way of directory rename)",
582 [CONFLICT_DIR_RENAME_COLLISION
] = "CONFLICT(directory rename collision)",
583 [CONFLICT_DIR_RENAME_SPLIT
] = "CONFLICT(directory rename unclear split)",
585 /*** Basic submodule ***/
586 [INFO_SUBMODULE_FAST_FORWARDING
] = "Fast forwarding submodule",
587 [CONFLICT_SUBMODULE_FAILED_TO_MERGE
] = "CONFLICT (submodule)",
589 /*** Special submodule cases broken out from FAILED_TO_MERGE ***/
590 [CONFLICT_SUBMODULE_FAILED_TO_MERGE_BUT_POSSIBLE_RESOLUTION
] =
591 "CONFLICT (submodule with possible resolution)",
592 [CONFLICT_SUBMODULE_NOT_INITIALIZED
] =
593 "CONFLICT (submodule not initialized)",
594 [CONFLICT_SUBMODULE_HISTORY_NOT_AVAILABLE
] =
595 "CONFLICT (submodule history not available)",
596 [CONFLICT_SUBMODULE_MAY_HAVE_REWINDS
] =
597 "CONFLICT (submodule may have rewinds)",
598 [CONFLICT_SUBMODULE_NULL_MERGE_BASE
] =
599 "CONFLICT (submodule lacks merge base)"
602 struct logical_conflict_info
{
603 enum conflict_and_info_types type
;
607 /*** Function Grouping: various utility functions ***/
610 * For the next three macros, see warning for conflict_info.merged.
612 * In each of the below, mi is a struct merged_info*, and ci was defined
613 * as a struct conflict_info* (but we need to verify ci isn't actually
614 * pointed at a struct merged_info*).
616 * INITIALIZE_CI: Assign ci to mi but only if it's safe; set to NULL otherwise.
617 * VERIFY_CI: Ensure that something we assigned to a conflict_info* is one.
618 * ASSIGN_AND_VERIFY_CI: Similar to VERIFY_CI but do assignment first.
620 #define INITIALIZE_CI(ci, mi) do { \
621 (ci) = (!(mi) || (mi)->clean) ? NULL : (struct conflict_info *)(mi); \
623 #define VERIFY_CI(ci) assert(ci && !ci->merged.clean);
624 #define ASSIGN_AND_VERIFY_CI(ci, mi) do { \
625 (ci) = (struct conflict_info *)(mi); \
626 assert((ci) && !(mi)->clean); \
629 static void free_strmap_strings(struct strmap
*map
)
631 struct hashmap_iter iter
;
632 struct strmap_entry
*entry
;
634 strmap_for_each_entry(map
, &iter
, entry
) {
635 free((char*)entry
->key
);
639 static void clear_or_reinit_internal_opts(struct merge_options_internal
*opti
,
642 struct rename_info
*renames
= &opti
->renames
;
644 void (*strmap_clear_func
)(struct strmap
*, int) =
645 reinitialize
? strmap_partial_clear
: strmap_clear
;
646 void (*strintmap_clear_func
)(struct strintmap
*) =
647 reinitialize
? strintmap_partial_clear
: strintmap_clear
;
648 void (*strset_clear_func
)(struct strset
*) =
649 reinitialize
? strset_partial_clear
: strset_clear
;
651 strmap_clear_func(&opti
->paths
, 0);
654 * All keys and values in opti->conflicted are a subset of those in
655 * opti->paths. We don't want to deallocate anything twice, so we
656 * don't free the keys and we pass 0 for free_values.
658 strmap_clear_func(&opti
->conflicted
, 0);
660 if (opti
->attr_index
.cache_nr
) /* true iff opt->renormalize */
661 discard_index(&opti
->attr_index
);
663 /* Free memory used by various renames maps */
664 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; ++i
) {
665 strintmap_clear_func(&renames
->dirs_removed
[i
]);
666 strmap_clear_func(&renames
->dir_renames
[i
], 0);
667 strintmap_clear_func(&renames
->relevant_sources
[i
]);
669 assert(renames
->cached_pairs_valid_side
== 0);
670 if (i
!= renames
->cached_pairs_valid_side
&&
671 -1 != renames
->cached_pairs_valid_side
) {
672 strset_clear_func(&renames
->cached_target_names
[i
]);
673 strmap_clear_func(&renames
->cached_pairs
[i
], 1);
674 strset_clear_func(&renames
->cached_irrelevant
[i
]);
675 partial_clear_dir_rename_count(&renames
->dir_rename_count
[i
]);
677 strmap_clear(&renames
->dir_rename_count
[i
], 1);
680 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; ++i
) {
681 strintmap_clear_func(&renames
->deferred
[i
].possible_trivial_merges
);
682 strset_clear_func(&renames
->deferred
[i
].target_dirs
);
683 renames
->deferred
[i
].trivial_merges_okay
= 1; /* 1 == maybe */
685 renames
->cached_pairs_valid_side
= 0;
686 renames
->dir_rename_mask
= 0;
689 struct hashmap_iter iter
;
690 struct strmap_entry
*e
;
692 /* Release and free each strbuf found in output */
693 strmap_for_each_entry(&opti
->conflicts
, &iter
, e
) {
694 struct string_list
*list
= e
->value
;
695 for (int i
= 0; i
< list
->nr
; i
++) {
696 struct logical_conflict_info
*info
=
698 strvec_clear(&info
->paths
);
701 * While strictly speaking we don't need to
702 * free(conflicts) here because we could pass
703 * free_values=1 when calling strmap_clear() on
704 * opti->conflicts, that would require strmap_clear
705 * to do another strmap_for_each_entry() loop, so we
706 * just free it while we're iterating anyway.
708 string_list_clear(list
, 1);
711 strmap_clear(&opti
->conflicts
, 0);
714 mem_pool_discard(&opti
->pool
, 0);
716 string_list_clear_func(&opti
->conflicted_submodules
,
717 conflicted_submodule_item_free
);
719 /* Clean out callback_data as well. */
720 FREE_AND_NULL(renames
->callback_data
);
721 renames
->callback_data_nr
= renames
->callback_data_alloc
= 0;
724 static void format_commit(struct strbuf
*sb
,
726 struct repository
*repo
,
727 struct commit
*commit
)
729 struct merge_remote_desc
*desc
;
730 struct pretty_print_context ctx
= {0};
731 ctx
.abbrev
= DEFAULT_ABBREV
;
733 strbuf_addchars(sb
, ' ', indent
);
734 desc
= merge_remote_util(commit
);
736 strbuf_addf(sb
, "virtual %s\n", desc
->name
);
740 repo_format_commit_message(repo
, commit
, "%h %s", sb
, &ctx
);
741 strbuf_addch(sb
, '\n');
744 __attribute__((format (printf
, 8, 9)))
745 static void path_msg(struct merge_options
*opt
,
746 enum conflict_and_info_types type
,
747 int omittable_hint
, /* skippable under --remerge-diff */
748 const char *primary_path
,
749 const char *other_path_1
, /* may be NULL */
750 const char *other_path_2
, /* may be NULL */
751 struct string_list
*other_paths
, /* may be NULL */
752 const char *fmt
, ...)
755 struct string_list
*path_conflicts
;
756 struct logical_conflict_info
*info
;
757 struct strbuf buf
= STRBUF_INIT
;
759 struct strbuf tmp
= STRBUF_INIT
;
762 assert(omittable_hint
==
763 !starts_with(type_short_descriptions
[type
], "CONFLICT") ||
764 type
== CONFLICT_DIR_RENAME_SUGGESTED
);
765 if (opt
->record_conflict_msgs_as_headers
&& omittable_hint
)
766 return; /* Do not record mere hints in headers */
767 if (opt
->priv
->call_depth
&& opt
->verbosity
< 5)
768 return; /* Ignore messages from inner merges */
770 /* Ensure path_conflicts (ptr to array of logical_conflict) allocated */
771 path_conflicts
= strmap_get(&opt
->priv
->conflicts
, primary_path
);
772 if (!path_conflicts
) {
773 path_conflicts
= xmalloc(sizeof(*path_conflicts
));
774 string_list_init_dup(path_conflicts
);
775 strmap_put(&opt
->priv
->conflicts
, primary_path
, path_conflicts
);
778 /* Add a logical_conflict at the end to store info from this call */
779 info
= xcalloc(1, sizeof(*info
));
781 strvec_init(&info
->paths
);
783 /* Handle the list of paths */
784 strvec_push(&info
->paths
, primary_path
);
786 strvec_push(&info
->paths
, other_path_1
);
788 strvec_push(&info
->paths
, other_path_2
);
790 for (int i
= 0; i
< other_paths
->nr
; i
++)
791 strvec_push(&info
->paths
, other_paths
->items
[i
].string
);
793 /* Handle message and its format, in normal case */
794 dest
= (opt
->record_conflict_msgs_as_headers
? &tmp
: &buf
);
797 if (opt
->priv
->call_depth
) {
798 strbuf_addchars(dest
, ' ', 2);
799 strbuf_addstr(dest
, "From inner merge:");
800 strbuf_addchars(dest
, ' ', opt
->priv
->call_depth
* 2);
802 strbuf_vaddf(dest
, fmt
, ap
);
805 /* Handle specialized formatting of message under --remerge-diff */
806 if (opt
->record_conflict_msgs_as_headers
) {
807 int i_sb
= 0, i_tmp
= 0;
809 /* Start with the specified prefix */
810 if (opt
->msg_header_prefix
)
811 strbuf_addf(&buf
, "%s ", opt
->msg_header_prefix
);
813 /* Copy tmp to sb, adding spaces after newlines */
814 strbuf_grow(&buf
, buf
.len
+ 2*tmp
.len
); /* more than sufficient */
815 for (; i_tmp
< tmp
.len
; i_tmp
++, i_sb
++) {
816 /* Copy next character from tmp to sb */
817 buf
.buf
[buf
.len
+ i_sb
] = tmp
.buf
[i_tmp
];
819 /* If we copied a newline, add a space */
820 if (tmp
.buf
[i_tmp
] == '\n')
821 buf
.buf
[++i_sb
] = ' ';
823 /* Update length and ensure it's NUL-terminated */
825 buf
.buf
[buf
.len
] = '\0';
827 strbuf_release(&tmp
);
829 string_list_append_nodup(path_conflicts
, strbuf_detach(&buf
, NULL
))
833 static struct diff_filespec
*pool_alloc_filespec(struct mem_pool
*pool
,
836 /* Similar to alloc_filespec(), but allocate from pool and reuse path */
837 struct diff_filespec
*spec
;
839 spec
= mem_pool_calloc(pool
, 1, sizeof(*spec
));
840 spec
->path
= (char*)path
; /* spec won't modify it */
843 spec
->is_binary
= -1;
847 static struct diff_filepair
*pool_diff_queue(struct mem_pool
*pool
,
848 struct diff_queue_struct
*queue
,
849 struct diff_filespec
*one
,
850 struct diff_filespec
*two
)
852 /* Same code as diff_queue(), except allocate from pool */
853 struct diff_filepair
*dp
;
855 dp
= mem_pool_calloc(pool
, 1, sizeof(*dp
));
863 /* add a string to a strbuf, but converting "/" to "_" */
864 static void add_flattened_path(struct strbuf
*out
, const char *s
)
867 strbuf_addstr(out
, s
);
868 for (; i
< out
->len
; i
++)
869 if (out
->buf
[i
] == '/')
873 static char *unique_path(struct merge_options
*opt
,
878 struct strbuf newpath
= STRBUF_INIT
;
881 struct strmap
*existing_paths
= &opt
->priv
->paths
;
883 strbuf_addf(&newpath
, "%s~", path
);
884 add_flattened_path(&newpath
, branch
);
886 base_len
= newpath
.len
;
887 while (strmap_contains(existing_paths
, newpath
.buf
)) {
888 strbuf_setlen(&newpath
, base_len
);
889 strbuf_addf(&newpath
, "_%d", suffix
++);
892 /* Track the new path in our memory pool */
893 ret
= mem_pool_alloc(&opt
->priv
->pool
, newpath
.len
+ 1);
894 memcpy(ret
, newpath
.buf
, newpath
.len
+ 1);
895 strbuf_release(&newpath
);
899 /*** Function Grouping: functions related to collect_merge_info() ***/
901 static int traverse_trees_wrapper_callback(int n
,
903 unsigned long dirmask
,
904 struct name_entry
*names
,
905 struct traverse_info
*info
)
907 struct merge_options
*opt
= info
->data
;
908 struct rename_info
*renames
= &opt
->priv
->renames
;
909 unsigned filemask
= mask
& ~dirmask
;
913 if (!renames
->callback_data_traverse_path
)
914 renames
->callback_data_traverse_path
= xstrdup(info
->traverse_path
);
916 if (filemask
&& filemask
== renames
->dir_rename_mask
)
917 renames
->dir_rename_mask
= 0x07;
919 ALLOC_GROW(renames
->callback_data
, renames
->callback_data_nr
+ 1,
920 renames
->callback_data_alloc
);
921 renames
->callback_data
[renames
->callback_data_nr
].mask
= mask
;
922 renames
->callback_data
[renames
->callback_data_nr
].dirmask
= dirmask
;
923 COPY_ARRAY(renames
->callback_data
[renames
->callback_data_nr
].names
,
925 renames
->callback_data_nr
++;
931 * Much like traverse_trees(), BUT:
932 * - read all the tree entries FIRST, saving them
933 * - note that the above step provides an opportunity to compute necessary
934 * additional details before the "real" traversal
935 * - loop through the saved entries and call the original callback on them
937 static int traverse_trees_wrapper(struct index_state
*istate
,
940 struct traverse_info
*info
)
942 int ret
, i
, old_offset
;
943 traverse_callback_t old_fn
;
944 char *old_callback_data_traverse_path
;
945 struct merge_options
*opt
= info
->data
;
946 struct rename_info
*renames
= &opt
->priv
->renames
;
948 assert(renames
->dir_rename_mask
== 2 || renames
->dir_rename_mask
== 4);
950 old_callback_data_traverse_path
= renames
->callback_data_traverse_path
;
952 old_offset
= renames
->callback_data_nr
;
954 renames
->callback_data_traverse_path
= NULL
;
955 info
->fn
= traverse_trees_wrapper_callback
;
956 ret
= traverse_trees(istate
, n
, t
, info
);
960 info
->traverse_path
= renames
->callback_data_traverse_path
;
962 for (i
= old_offset
; i
< renames
->callback_data_nr
; ++i
) {
964 renames
->callback_data
[i
].mask
,
965 renames
->callback_data
[i
].dirmask
,
966 renames
->callback_data
[i
].names
,
970 renames
->callback_data_nr
= old_offset
;
971 free(renames
->callback_data_traverse_path
);
972 renames
->callback_data_traverse_path
= old_callback_data_traverse_path
;
973 info
->traverse_path
= NULL
;
977 static void setup_path_info(struct merge_options
*opt
,
978 struct string_list_item
*result
,
979 const char *current_dir_name
,
980 int current_dir_name_len
,
981 char *fullpath
, /* we'll take over ownership */
982 struct name_entry
*names
,
983 struct name_entry
*merged_version
,
984 unsigned is_null
, /* boolean */
985 unsigned df_conflict
, /* boolean */
988 int resolved
/* boolean */)
990 /* result->util is void*, so mi is a convenience typed variable */
991 struct merged_info
*mi
;
993 assert(!is_null
|| resolved
);
994 assert(!df_conflict
|| !resolved
); /* df_conflict implies !resolved */
995 assert(resolved
== (merged_version
!= NULL
));
997 mi
= mem_pool_calloc(&opt
->priv
->pool
, 1,
998 resolved
? sizeof(struct merged_info
) :
999 sizeof(struct conflict_info
));
1000 mi
->directory_name
= current_dir_name
;
1001 mi
->basename_offset
= current_dir_name_len
;
1002 mi
->clean
= !!resolved
;
1004 mi
->result
.mode
= merged_version
->mode
;
1005 oidcpy(&mi
->result
.oid
, &merged_version
->oid
);
1006 mi
->is_null
= !!is_null
;
1009 struct conflict_info
*ci
;
1011 ASSIGN_AND_VERIFY_CI(ci
, mi
);
1012 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
1013 ci
->pathnames
[i
] = fullpath
;
1014 ci
->stages
[i
].mode
= names
[i
].mode
;
1015 oidcpy(&ci
->stages
[i
].oid
, &names
[i
].oid
);
1017 ci
->filemask
= filemask
;
1018 ci
->dirmask
= dirmask
;
1019 ci
->df_conflict
= !!df_conflict
;
1022 * Assume is_null for now, but if we have entries
1023 * under the directory then when it is complete in
1024 * write_completed_directory() it'll update this.
1025 * Also, for D/F conflicts, we have to handle the
1026 * directory first, then clear this bit and process
1027 * the file to see how it is handled -- that occurs
1028 * near the top of process_entry().
1032 strmap_put(&opt
->priv
->paths
, fullpath
, mi
);
1033 result
->string
= fullpath
;
1037 static void add_pair(struct merge_options
*opt
,
1038 struct name_entry
*names
,
1039 const char *pathname
,
1041 unsigned is_add
/* if false, is_delete */,
1042 unsigned match_mask
,
1043 unsigned dir_rename_mask
)
1045 struct diff_filespec
*one
, *two
;
1046 struct rename_info
*renames
= &opt
->priv
->renames
;
1047 int names_idx
= is_add
? side
: 0;
1050 assert(match_mask
== 0 || match_mask
== 6);
1051 if (strset_contains(&renames
->cached_target_names
[side
],
1055 unsigned content_relevant
= (match_mask
== 0);
1056 unsigned location_relevant
= (dir_rename_mask
== 0x07);
1058 assert(match_mask
== 0 || match_mask
== 3 || match_mask
== 5);
1061 * If pathname is found in cached_irrelevant[side] due to
1062 * previous pick but for this commit content is relevant,
1063 * then we need to remove it from cached_irrelevant.
1065 if (content_relevant
)
1066 /* strset_remove is no-op if strset doesn't have key */
1067 strset_remove(&renames
->cached_irrelevant
[side
],
1071 * We do not need to re-detect renames for paths that we already
1072 * know the pairing, i.e. for cached_pairs (or
1073 * cached_irrelevant). However, handle_deferred_entries() needs
1074 * to loop over the union of keys from relevant_sources[side] and
1075 * cached_pairs[side], so for simplicity we set relevant_sources
1076 * for all the cached_pairs too and then strip them back out in
1077 * prune_cached_from_relevant() at the beginning of
1078 * detect_regular_renames().
1080 if (content_relevant
|| location_relevant
) {
1081 /* content_relevant trumps location_relevant */
1082 strintmap_set(&renames
->relevant_sources
[side
], pathname
,
1083 content_relevant
? RELEVANT_CONTENT
: RELEVANT_LOCATION
);
1087 * Avoid creating pair if we've already cached rename results.
1088 * Note that we do this after setting relevant_sources[side]
1089 * as noted in the comment above.
1091 if (strmap_contains(&renames
->cached_pairs
[side
], pathname
) ||
1092 strset_contains(&renames
->cached_irrelevant
[side
], pathname
))
1096 one
= pool_alloc_filespec(&opt
->priv
->pool
, pathname
);
1097 two
= pool_alloc_filespec(&opt
->priv
->pool
, pathname
);
1098 fill_filespec(is_add
? two
: one
,
1099 &names
[names_idx
].oid
, 1, names
[names_idx
].mode
);
1100 pool_diff_queue(&opt
->priv
->pool
, &renames
->pairs
[side
], one
, two
);
1103 static void collect_rename_info(struct merge_options
*opt
,
1104 struct name_entry
*names
,
1105 const char *dirname
,
1106 const char *fullname
,
1109 unsigned match_mask
)
1111 struct rename_info
*renames
= &opt
->priv
->renames
;
1115 * Update dir_rename_mask (determines ignore-rename-source validity)
1117 * dir_rename_mask helps us keep track of when directory rename
1118 * detection may be relevant. Basically, whenver a directory is
1119 * removed on one side of history, and a file is added to that
1120 * directory on the other side of history, directory rename
1121 * detection is relevant (meaning we have to detect renames for all
1122 * files within that directory to deduce where the directory
1123 * moved). Also, whenever a directory needs directory rename
1124 * detection, due to the "majority rules" choice for where to move
1125 * it (see t6423 testcase 1f), we also need to detect renames for
1126 * all files within subdirectories of that directory as well.
1128 * Here we haven't looked at files within the directory yet, we are
1129 * just looking at the directory itself. So, if we aren't yet in
1130 * a case where a parent directory needed directory rename detection
1131 * (i.e. dir_rename_mask != 0x07), and if the directory was removed
1132 * on one side of history, record the mask of the other side of
1133 * history in dir_rename_mask.
1135 if (renames
->dir_rename_mask
!= 0x07 &&
1136 (dirmask
== 3 || dirmask
== 5)) {
1137 /* simple sanity check */
1138 assert(renames
->dir_rename_mask
== 0 ||
1139 renames
->dir_rename_mask
== (dirmask
& ~1));
1140 /* update dir_rename_mask; have it record mask of new side */
1141 renames
->dir_rename_mask
= (dirmask
& ~1);
1144 /* Update dirs_removed, as needed */
1145 if (dirmask
== 1 || dirmask
== 3 || dirmask
== 5) {
1146 /* absent_mask = 0x07 - dirmask; sides = absent_mask/2 */
1147 unsigned sides
= (0x07 - dirmask
)/2;
1148 unsigned relevance
= (renames
->dir_rename_mask
== 0x07) ?
1149 RELEVANT_FOR_ANCESTOR
: NOT_RELEVANT
;
1151 * Record relevance of this directory. However, note that
1152 * when collect_merge_info_callback() recurses into this
1153 * directory and calls collect_rename_info() on paths
1154 * within that directory, if we find a path that was added
1155 * to this directory on the other side of history, we will
1156 * upgrade this value to RELEVANT_FOR_SELF; see below.
1159 strintmap_set(&renames
->dirs_removed
[1], fullname
,
1162 strintmap_set(&renames
->dirs_removed
[2], fullname
,
1167 * Here's the block that potentially upgrades to RELEVANT_FOR_SELF.
1168 * When we run across a file added to a directory. In such a case,
1169 * find the directory of the file and upgrade its relevance.
1171 if (renames
->dir_rename_mask
== 0x07 &&
1172 (filemask
== 2 || filemask
== 4)) {
1174 * Need directory rename for parent directory on other side
1175 * of history from added file. Thus
1176 * side = (~filemask & 0x06) >> 1
1178 * side = 3 - (filemask/2).
1180 unsigned side
= 3 - (filemask
>> 1);
1181 strintmap_set(&renames
->dirs_removed
[side
], dirname
,
1185 if (filemask
== 0 || filemask
== 7)
1188 for (side
= MERGE_SIDE1
; side
<= MERGE_SIDE2
; ++side
) {
1189 unsigned side_mask
= (1 << side
);
1191 /* Check for deletion on side */
1192 if ((filemask
& 1) && !(filemask
& side_mask
))
1193 add_pair(opt
, names
, fullname
, side
, 0 /* delete */,
1194 match_mask
& filemask
,
1195 renames
->dir_rename_mask
);
1197 /* Check for addition on side */
1198 if (!(filemask
& 1) && (filemask
& side_mask
))
1199 add_pair(opt
, names
, fullname
, side
, 1 /* add */,
1200 match_mask
& filemask
,
1201 renames
->dir_rename_mask
);
1205 static int collect_merge_info_callback(int n
,
1207 unsigned long dirmask
,
1208 struct name_entry
*names
,
1209 struct traverse_info
*info
)
1213 * common ancestor (mbase) has mask 1, and stored in index 0 of names
1214 * head of side 1 (side1) has mask 2, and stored in index 1 of names
1215 * head of side 2 (side2) has mask 4, and stored in index 2 of names
1217 struct merge_options
*opt
= info
->data
;
1218 struct merge_options_internal
*opti
= opt
->priv
;
1219 struct rename_info
*renames
= &opt
->priv
->renames
;
1220 struct string_list_item pi
; /* Path Info */
1221 struct conflict_info
*ci
; /* typed alias to pi.util (which is void*) */
1222 struct name_entry
*p
;
1225 const char *dirname
= opti
->current_dir_name
;
1226 unsigned prev_dir_rename_mask
= renames
->dir_rename_mask
;
1227 unsigned filemask
= mask
& ~dirmask
;
1228 unsigned match_mask
= 0; /* will be updated below */
1229 unsigned mbase_null
= !(mask
& 1);
1230 unsigned side1_null
= !(mask
& 2);
1231 unsigned side2_null
= !(mask
& 4);
1232 unsigned side1_matches_mbase
= (!side1_null
&& !mbase_null
&&
1233 names
[0].mode
== names
[1].mode
&&
1234 oideq(&names
[0].oid
, &names
[1].oid
));
1235 unsigned side2_matches_mbase
= (!side2_null
&& !mbase_null
&&
1236 names
[0].mode
== names
[2].mode
&&
1237 oideq(&names
[0].oid
, &names
[2].oid
));
1238 unsigned sides_match
= (!side1_null
&& !side2_null
&&
1239 names
[1].mode
== names
[2].mode
&&
1240 oideq(&names
[1].oid
, &names
[2].oid
));
1243 * Note: When a path is a file on one side of history and a directory
1244 * in another, we have a directory/file conflict. In such cases, if
1245 * the conflict doesn't resolve from renames and deletions, then we
1246 * always leave directories where they are and move files out of the
1247 * way. Thus, while struct conflict_info has a df_conflict field to
1248 * track such conflicts, we ignore that field for any directories at
1249 * a path and only pay attention to it for files at the given path.
1250 * The fact that we leave directories were they are also means that
1251 * we do not need to worry about getting additional df_conflict
1252 * information propagated from parent directories down to children
1253 * (unlike, say traverse_trees_recursive() in unpack-trees.c, which
1254 * sets a newinfo.df_conflicts field specifically to propagate it).
1256 unsigned df_conflict
= (filemask
!= 0) && (dirmask
!= 0);
1258 /* n = 3 is a fundamental assumption. */
1260 BUG("Called collect_merge_info_callback wrong");
1263 * A bunch of sanity checks verifying that traverse_trees() calls
1264 * us the way I expect. Could just remove these at some point,
1265 * though maybe they are helpful to future code readers.
1267 assert(mbase_null
== is_null_oid(&names
[0].oid
));
1268 assert(side1_null
== is_null_oid(&names
[1].oid
));
1269 assert(side2_null
== is_null_oid(&names
[2].oid
));
1270 assert(!mbase_null
|| !side1_null
|| !side2_null
);
1271 assert(mask
> 0 && mask
< 8);
1273 /* Determine match_mask */
1274 if (side1_matches_mbase
)
1275 match_mask
= (side2_matches_mbase
? 7 : 3);
1276 else if (side2_matches_mbase
)
1278 else if (sides_match
)
1282 * Get the name of the relevant filepath, which we'll pass to
1283 * setup_path_info() for tracking.
1288 len
= traverse_path_len(info
, p
->pathlen
);
1290 /* +1 in both of the following lines to include the NUL byte */
1291 fullpath
= mem_pool_alloc(&opt
->priv
->pool
, len
+ 1);
1292 make_traverse_path(fullpath
, len
+ 1, info
, p
->path
, p
->pathlen
);
1295 * If mbase, side1, and side2 all match, we can resolve early. Even
1296 * if these are trees, there will be no renames or anything
1299 if (side1_matches_mbase
&& side2_matches_mbase
) {
1300 /* mbase, side1, & side2 all match; use mbase as resolution */
1301 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1302 names
, names
+0, mbase_null
, 0 /* df_conflict */,
1303 filemask
, dirmask
, 1 /* resolved */);
1308 * If the sides match, and all three paths are present and are
1309 * files, then we can take either as the resolution. We can't do
1310 * this with trees, because there may be rename sources from the
1313 if (sides_match
&& filemask
== 0x07) {
1314 /* use side1 (== side2) version as resolution */
1315 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1316 names
, names
+1, side1_null
, 0,
1317 filemask
, dirmask
, 1);
1322 * If side1 matches mbase and all three paths are present and are
1323 * files, then we can use side2 as the resolution. We cannot
1324 * necessarily do so this for trees, because there may be rename
1325 * destinations within side2.
1327 if (side1_matches_mbase
&& filemask
== 0x07) {
1328 /* use side2 version as resolution */
1329 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1330 names
, names
+2, side2_null
, 0,
1331 filemask
, dirmask
, 1);
1335 /* Similar to above but swapping sides 1 and 2 */
1336 if (side2_matches_mbase
&& filemask
== 0x07) {
1337 /* use side1 version as resolution */
1338 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1339 names
, names
+1, side1_null
, 0,
1340 filemask
, dirmask
, 1);
1345 * Sometimes we can tell that a source path need not be included in
1346 * rename detection -- namely, whenever either
1347 * side1_matches_mbase && side2_null
1349 * side2_matches_mbase && side1_null
1350 * However, we call collect_rename_info() even in those cases,
1351 * because exact renames are cheap and would let us remove both a
1352 * source and destination path. We'll cull the unneeded sources
1355 collect_rename_info(opt
, names
, dirname
, fullpath
,
1356 filemask
, dirmask
, match_mask
);
1359 * None of the special cases above matched, so we have a
1360 * provisional conflict. (Rename detection might allow us to
1361 * unconflict some more cases, but that comes later so all we can
1362 * do now is record the different non-null file hashes.)
1364 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1365 names
, NULL
, 0, df_conflict
, filemask
, dirmask
, 0);
1369 ci
->match_mask
= match_mask
;
1371 /* If dirmask, recurse into subdirectories */
1373 struct traverse_info newinfo
;
1374 struct tree_desc t
[3];
1375 void *buf
[3] = {NULL
, NULL
, NULL
};
1376 const char *original_dir_name
;
1380 * Check for whether we can avoid recursing due to one side
1381 * matching the merge base. The side that does NOT match is
1382 * the one that might have a rename destination we need.
1384 assert(!side1_matches_mbase
|| !side2_matches_mbase
);
1385 side
= side1_matches_mbase
? MERGE_SIDE2
:
1386 side2_matches_mbase
? MERGE_SIDE1
: MERGE_BASE
;
1387 if (filemask
== 0 && (dirmask
== 2 || dirmask
== 4)) {
1389 * Also defer recursing into new directories; set up a
1390 * few variables to let us do so.
1392 ci
->match_mask
= (7 - dirmask
);
1395 if (renames
->dir_rename_mask
!= 0x07 &&
1396 side
!= MERGE_BASE
&&
1397 renames
->deferred
[side
].trivial_merges_okay
&&
1398 !strset_contains(&renames
->deferred
[side
].target_dirs
,
1400 strintmap_set(&renames
->deferred
[side
].possible_trivial_merges
,
1401 pi
.string
, renames
->dir_rename_mask
);
1402 renames
->dir_rename_mask
= prev_dir_rename_mask
;
1406 /* We need to recurse */
1407 ci
->match_mask
&= filemask
;
1409 newinfo
.prev
= info
;
1410 newinfo
.name
= p
->path
;
1411 newinfo
.namelen
= p
->pathlen
;
1412 newinfo
.pathlen
= st_add3(newinfo
.pathlen
, p
->pathlen
, 1);
1414 * If this directory we are about to recurse into cared about
1415 * its parent directory (the current directory) having a D/F
1416 * conflict, then we'd propagate the masks in this way:
1417 * newinfo.df_conflicts |= (mask & ~dirmask);
1418 * But we don't worry about propagating D/F conflicts. (See
1419 * comment near setting of local df_conflict variable near
1420 * the beginning of this function).
1423 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
1424 if (i
== 1 && side1_matches_mbase
)
1426 else if (i
== 2 && side2_matches_mbase
)
1428 else if (i
== 2 && sides_match
)
1431 const struct object_id
*oid
= NULL
;
1433 oid
= &names
[i
].oid
;
1434 buf
[i
] = fill_tree_descriptor(opt
->repo
,
1440 original_dir_name
= opti
->current_dir_name
;
1441 opti
->current_dir_name
= pi
.string
;
1442 if (renames
->dir_rename_mask
== 0 ||
1443 renames
->dir_rename_mask
== 0x07)
1444 ret
= traverse_trees(NULL
, 3, t
, &newinfo
);
1446 ret
= traverse_trees_wrapper(NULL
, 3, t
, &newinfo
);
1447 opti
->current_dir_name
= original_dir_name
;
1448 renames
->dir_rename_mask
= prev_dir_rename_mask
;
1450 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++)
1460 static void resolve_trivial_directory_merge(struct conflict_info
*ci
, int side
)
1463 assert((side
== 1 && ci
->match_mask
== 5) ||
1464 (side
== 2 && ci
->match_mask
== 3));
1465 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[side
].oid
);
1466 ci
->merged
.result
.mode
= ci
->stages
[side
].mode
;
1467 ci
->merged
.is_null
= is_null_oid(&ci
->stages
[side
].oid
);
1469 ci
->merged
.clean
= 1; /* (ci->filemask == 0); */
1472 static int handle_deferred_entries(struct merge_options
*opt
,
1473 struct traverse_info
*info
)
1475 struct rename_info
*renames
= &opt
->priv
->renames
;
1476 struct hashmap_iter iter
;
1477 struct strmap_entry
*entry
;
1479 int path_count_before
, path_count_after
= 0;
1481 path_count_before
= strmap_get_size(&opt
->priv
->paths
);
1482 for (side
= MERGE_SIDE1
; side
<= MERGE_SIDE2
; side
++) {
1483 unsigned optimization_okay
= 1;
1484 struct strintmap copy
;
1486 /* Loop over the set of paths we need to know rename info for */
1487 strset_for_each_entry(&renames
->relevant_sources
[side
],
1489 char *rename_target
, *dir
, *dir_marker
;
1490 struct strmap_entry
*e
;
1493 * If we don't know delete/rename info for this path,
1494 * then we need to recurse into all trees to get all
1495 * adds to make sure we have it.
1497 if (strset_contains(&renames
->cached_irrelevant
[side
],
1500 e
= strmap_get_entry(&renames
->cached_pairs
[side
],
1503 optimization_okay
= 0;
1507 /* If this is a delete, we have enough info already */
1508 rename_target
= e
->value
;
1512 /* If we already walked the rename target, we're good */
1513 if (strmap_contains(&opt
->priv
->paths
, rename_target
))
1517 * Otherwise, we need to get a list of directories that
1518 * will need to be recursed into to get this
1521 dir
= xstrdup(rename_target
);
1522 while ((dir_marker
= strrchr(dir
, '/'))) {
1524 if (strset_contains(&renames
->deferred
[side
].target_dirs
,
1527 strset_add(&renames
->deferred
[side
].target_dirs
,
1532 renames
->deferred
[side
].trivial_merges_okay
= optimization_okay
;
1534 * We need to recurse into any directories in
1535 * possible_trivial_merges[side] found in target_dirs[side].
1536 * But when we recurse, we may need to queue up some of the
1537 * subdirectories for possible_trivial_merges[side]. Since
1538 * we can't safely iterate through a hashmap while also adding
1539 * entries, move the entries into 'copy', iterate over 'copy',
1540 * and then we'll also iterate anything added into
1541 * possible_trivial_merges[side] once this loop is done.
1543 copy
= renames
->deferred
[side
].possible_trivial_merges
;
1544 strintmap_init_with_options(&renames
->deferred
[side
].possible_trivial_merges
,
1548 strintmap_for_each_entry(©
, &iter
, entry
) {
1549 const char *path
= entry
->key
;
1550 unsigned dir_rename_mask
= (intptr_t)entry
->value
;
1551 struct conflict_info
*ci
;
1553 struct tree_desc t
[3];
1554 void *buf
[3] = {NULL
,};
1557 ci
= strmap_get(&opt
->priv
->paths
, path
);
1559 dirmask
= ci
->dirmask
;
1561 if (optimization_okay
&&
1562 !strset_contains(&renames
->deferred
[side
].target_dirs
,
1564 resolve_trivial_directory_merge(ci
, side
);
1569 info
->namelen
= strlen(path
);
1570 info
->pathlen
= info
->namelen
+ 1;
1572 for (i
= 0; i
< 3; i
++, dirmask
>>= 1) {
1573 if (i
== 1 && ci
->match_mask
== 3)
1575 else if (i
== 2 && ci
->match_mask
== 5)
1577 else if (i
== 2 && ci
->match_mask
== 6)
1580 const struct object_id
*oid
= NULL
;
1582 oid
= &ci
->stages
[i
].oid
;
1583 buf
[i
] = fill_tree_descriptor(opt
->repo
,
1588 ci
->match_mask
&= ci
->filemask
;
1589 opt
->priv
->current_dir_name
= path
;
1590 renames
->dir_rename_mask
= dir_rename_mask
;
1591 if (renames
->dir_rename_mask
== 0 ||
1592 renames
->dir_rename_mask
== 0x07)
1593 ret
= traverse_trees(NULL
, 3, t
, info
);
1595 ret
= traverse_trees_wrapper(NULL
, 3, t
, info
);
1597 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++)
1603 strintmap_clear(©
);
1604 strintmap_for_each_entry(&renames
->deferred
[side
].possible_trivial_merges
,
1606 const char *path
= entry
->key
;
1607 struct conflict_info
*ci
;
1609 ci
= strmap_get(&opt
->priv
->paths
, path
);
1612 assert(renames
->deferred
[side
].trivial_merges_okay
&&
1613 !strset_contains(&renames
->deferred
[side
].target_dirs
,
1615 resolve_trivial_directory_merge(ci
, side
);
1617 if (!optimization_okay
|| path_count_after
)
1618 path_count_after
= strmap_get_size(&opt
->priv
->paths
);
1620 if (path_count_after
) {
1622 * The choice of wanted_factor here does not affect
1623 * correctness, only performance. When the
1624 * path_count_after / path_count_before
1625 * ratio is high, redoing after renames is a big
1626 * performance boost. I suspect that redoing is a wash
1627 * somewhere near a value of 2, and below that redoing will
1628 * slow things down. I applied a fudge factor and picked
1629 * 3; see the commit message when this was introduced for
1630 * back of the envelope calculations for this ratio.
1632 const int wanted_factor
= 3;
1634 /* We should only redo collect_merge_info one time */
1635 assert(renames
->redo_after_renames
== 0);
1637 if (path_count_after
/ path_count_before
>= wanted_factor
) {
1638 renames
->redo_after_renames
= 1;
1639 renames
->cached_pairs_valid_side
= -1;
1641 } else if (renames
->redo_after_renames
== 2)
1642 renames
->redo_after_renames
= 0;
1646 static int collect_merge_info(struct merge_options
*opt
,
1647 struct tree
*merge_base
,
1652 struct tree_desc t
[3];
1653 struct traverse_info info
;
1655 opt
->priv
->toplevel_dir
= "";
1656 opt
->priv
->current_dir_name
= opt
->priv
->toplevel_dir
;
1657 setup_traverse_info(&info
, opt
->priv
->toplevel_dir
);
1658 info
.fn
= collect_merge_info_callback
;
1660 info
.show_all_errors
= 1;
1662 if (parse_tree(merge_base
) < 0 ||
1663 parse_tree(side1
) < 0 ||
1664 parse_tree(side2
) < 0)
1666 init_tree_desc(t
+ 0, merge_base
->buffer
, merge_base
->size
);
1667 init_tree_desc(t
+ 1, side1
->buffer
, side1
->size
);
1668 init_tree_desc(t
+ 2, side2
->buffer
, side2
->size
);
1670 trace2_region_enter("merge", "traverse_trees", opt
->repo
);
1671 ret
= traverse_trees(NULL
, 3, t
, &info
);
1673 ret
= handle_deferred_entries(opt
, &info
);
1674 trace2_region_leave("merge", "traverse_trees", opt
->repo
);
1679 /*** Function Grouping: functions related to threeway content merges ***/
1681 static int find_first_merges(struct repository
*repo
,
1685 struct object_array
*result
)
1688 struct object_array merges
= OBJECT_ARRAY_INIT
;
1689 struct commit
*commit
;
1690 int contains_another
;
1692 char merged_revision
[GIT_MAX_HEXSZ
+ 2];
1693 const char *rev_args
[] = { "rev-list", "--merges", "--ancestry-path",
1694 "--all", merged_revision
, NULL
};
1695 struct rev_info revs
;
1696 struct setup_revision_opt rev_opts
;
1698 memset(result
, 0, sizeof(struct object_array
));
1699 memset(&rev_opts
, 0, sizeof(rev_opts
));
1701 /* get all revisions that merge commit a */
1702 xsnprintf(merged_revision
, sizeof(merged_revision
), "^%s",
1703 oid_to_hex(&a
->object
.oid
));
1704 repo_init_revisions(repo
, &revs
, NULL
);
1705 /* FIXME: can't handle linked worktrees in submodules yet */
1706 revs
.single_worktree
= path
!= NULL
;
1707 setup_revisions(ARRAY_SIZE(rev_args
)-1, rev_args
, &revs
, &rev_opts
);
1709 /* save all revisions from the above list that contain b */
1710 if (prepare_revision_walk(&revs
))
1711 die("revision walk setup failed");
1712 while ((commit
= get_revision(&revs
)) != NULL
) {
1713 struct object
*o
= &(commit
->object
);
1714 if (repo_in_merge_bases(repo
, b
, commit
))
1715 add_object_array(o
, NULL
, &merges
);
1717 reset_revision_walk();
1719 /* Now we've got all merges that contain a and b. Prune all
1720 * merges that contain another found merge and save them in
1723 for (i
= 0; i
< merges
.nr
; i
++) {
1724 struct commit
*m1
= (struct commit
*) merges
.objects
[i
].item
;
1726 contains_another
= 0;
1727 for (j
= 0; j
< merges
.nr
; j
++) {
1728 struct commit
*m2
= (struct commit
*) merges
.objects
[j
].item
;
1729 if (i
!= j
&& repo_in_merge_bases(repo
, m2
, m1
)) {
1730 contains_another
= 1;
1735 if (!contains_another
)
1736 add_object_array(merges
.objects
[i
].item
, NULL
, result
);
1739 object_array_clear(&merges
);
1740 release_revisions(&revs
);
1744 static int merge_submodule(struct merge_options
*opt
,
1746 const struct object_id
*o
,
1747 const struct object_id
*a
,
1748 const struct object_id
*b
,
1749 struct object_id
*result
)
1751 struct repository subrepo
;
1752 struct strbuf sb
= STRBUF_INIT
;
1754 struct commit
*commit_o
, *commit_a
, *commit_b
;
1756 struct object_array merges
;
1759 int search
= !opt
->priv
->call_depth
;
1760 int sub_not_initialized
= 1;
1761 int sub_flag
= CONFLICT_SUBMODULE_FAILED_TO_MERGE
;
1763 /* store fallback answer in result in case we fail */
1764 oidcpy(result
, opt
->priv
->call_depth
? o
: a
);
1766 /* we can not handle deletion conflicts */
1767 if (is_null_oid(a
) || is_null_oid(b
))
1768 BUG("submodule deleted on one side; this should be handled outside of merge_submodule()");
1770 if ((sub_not_initialized
= repo_submodule_init(&subrepo
,
1771 opt
->repo
, path
, null_oid()))) {
1772 path_msg(opt
, CONFLICT_SUBMODULE_NOT_INITIALIZED
, 0,
1773 path
, NULL
, NULL
, NULL
,
1774 _("Failed to merge submodule %s (not checked out)"),
1776 sub_flag
= CONFLICT_SUBMODULE_NOT_INITIALIZED
;
1780 if (is_null_oid(o
)) {
1781 path_msg(opt
, CONFLICT_SUBMODULE_NULL_MERGE_BASE
, 0,
1782 path
, NULL
, NULL
, NULL
,
1783 _("Failed to merge submodule %s (no merge base)"),
1788 if (!(commit_o
= lookup_commit_reference(&subrepo
, o
)) ||
1789 !(commit_a
= lookup_commit_reference(&subrepo
, a
)) ||
1790 !(commit_b
= lookup_commit_reference(&subrepo
, b
))) {
1791 path_msg(opt
, CONFLICT_SUBMODULE_HISTORY_NOT_AVAILABLE
, 0,
1792 path
, NULL
, NULL
, NULL
,
1793 _("Failed to merge submodule %s (commits not present)"),
1795 sub_flag
= CONFLICT_SUBMODULE_HISTORY_NOT_AVAILABLE
;
1799 /* check whether both changes are forward */
1800 if (!repo_in_merge_bases(&subrepo
, commit_o
, commit_a
) ||
1801 !repo_in_merge_bases(&subrepo
, commit_o
, commit_b
)) {
1802 path_msg(opt
, CONFLICT_SUBMODULE_MAY_HAVE_REWINDS
, 0,
1803 path
, NULL
, NULL
, NULL
,
1804 _("Failed to merge submodule %s "
1805 "(commits don't follow merge-base)"),
1810 /* Case #1: a is contained in b or vice versa */
1811 if (repo_in_merge_bases(&subrepo
, commit_a
, commit_b
)) {
1813 path_msg(opt
, INFO_SUBMODULE_FAST_FORWARDING
, 1,
1814 path
, NULL
, NULL
, NULL
,
1815 _("Note: Fast-forwarding submodule %s to %s"),
1816 path
, oid_to_hex(b
));
1820 if (repo_in_merge_bases(&subrepo
, commit_b
, commit_a
)) {
1822 path_msg(opt
, INFO_SUBMODULE_FAST_FORWARDING
, 1,
1823 path
, NULL
, NULL
, NULL
,
1824 _("Note: Fast-forwarding submodule %s to %s"),
1825 path
, oid_to_hex(a
));
1831 * Case #2: There are one or more merges that contain a and b in
1832 * the submodule. If there is only one, then present it as a
1833 * suggestion to the user, but leave it marked unmerged so the
1834 * user needs to confirm the resolution.
1837 /* Skip the search if makes no sense to the calling context. */
1841 /* find commit which merges them */
1842 parent_count
= find_first_merges(&subrepo
, path
, commit_a
, commit_b
,
1844 switch (parent_count
) {
1846 path_msg(opt
, CONFLICT_SUBMODULE_FAILED_TO_MERGE
, 0,
1847 path
, NULL
, NULL
, NULL
,
1848 _("Failed to merge submodule %s"), path
);
1852 format_commit(&sb
, 4, &subrepo
,
1853 (struct commit
*)merges
.objects
[0].item
);
1854 path_msg(opt
, CONFLICT_SUBMODULE_FAILED_TO_MERGE_BUT_POSSIBLE_RESOLUTION
, 0,
1855 path
, NULL
, NULL
, NULL
,
1856 _("Failed to merge submodule %s, but a possible merge "
1857 "resolution exists: %s"),
1859 strbuf_release(&sb
);
1862 for (i
= 0; i
< merges
.nr
; i
++)
1863 format_commit(&sb
, 4, &subrepo
,
1864 (struct commit
*)merges
.objects
[i
].item
);
1865 path_msg(opt
, CONFLICT_SUBMODULE_FAILED_TO_MERGE_BUT_POSSIBLE_RESOLUTION
, 0,
1866 path
, NULL
, NULL
, NULL
,
1867 _("Failed to merge submodule %s, but multiple "
1868 "possible merges exist:\n%s"), path
, sb
.buf
);
1869 strbuf_release(&sb
);
1872 object_array_clear(&merges
);
1874 if (!opt
->priv
->call_depth
&& !ret
) {
1875 struct string_list
*csub
= &opt
->priv
->conflicted_submodules
;
1876 struct conflicted_submodule_item
*util
;
1879 util
= xmalloc(sizeof(*util
));
1880 util
->flag
= sub_flag
;
1881 util
->abbrev
= NULL
;
1882 if (!sub_not_initialized
) {
1883 abbrev
= repo_find_unique_abbrev(&subrepo
, b
, DEFAULT_ABBREV
);
1884 util
->abbrev
= xstrdup(abbrev
);
1886 string_list_append(csub
, path
)->util
= util
;
1889 if (!sub_not_initialized
)
1890 repo_clear(&subrepo
);
1894 static void initialize_attr_index(struct merge_options
*opt
)
1897 * The renormalize_buffer() functions require attributes, and
1898 * annoyingly those can only be read from the working tree or from
1899 * an index_state. merge-ort doesn't have an index_state, so we
1900 * generate a fake one containing only attribute information.
1902 struct merged_info
*mi
;
1903 struct index_state
*attr_index
= &opt
->priv
->attr_index
;
1904 struct cache_entry
*ce
;
1906 attr_index
->repo
= opt
->repo
;
1907 attr_index
->initialized
= 1;
1909 if (!opt
->renormalize
)
1912 mi
= strmap_get(&opt
->priv
->paths
, GITATTRIBUTES_FILE
);
1917 int len
= strlen(GITATTRIBUTES_FILE
);
1918 ce
= make_empty_cache_entry(attr_index
, len
);
1919 ce
->ce_mode
= create_ce_mode(mi
->result
.mode
);
1920 ce
->ce_flags
= create_ce_flags(0);
1921 ce
->ce_namelen
= len
;
1922 oidcpy(&ce
->oid
, &mi
->result
.oid
);
1923 memcpy(ce
->name
, GITATTRIBUTES_FILE
, len
);
1924 add_index_entry(attr_index
, ce
,
1925 ADD_CACHE_OK_TO_ADD
| ADD_CACHE_OK_TO_REPLACE
);
1926 get_stream_filter(attr_index
, GITATTRIBUTES_FILE
, &ce
->oid
);
1929 struct conflict_info
*ci
;
1931 ASSIGN_AND_VERIFY_CI(ci
, mi
);
1932 for (stage
= 0; stage
< 3; stage
++) {
1933 unsigned stage_mask
= (1 << stage
);
1935 if (!(ci
->filemask
& stage_mask
))
1937 len
= strlen(GITATTRIBUTES_FILE
);
1938 ce
= make_empty_cache_entry(attr_index
, len
);
1939 ce
->ce_mode
= create_ce_mode(ci
->stages
[stage
].mode
);
1940 ce
->ce_flags
= create_ce_flags(stage
);
1941 ce
->ce_namelen
= len
;
1942 oidcpy(&ce
->oid
, &ci
->stages
[stage
].oid
);
1943 memcpy(ce
->name
, GITATTRIBUTES_FILE
, len
);
1944 add_index_entry(attr_index
, ce
,
1945 ADD_CACHE_OK_TO_ADD
| ADD_CACHE_OK_TO_REPLACE
);
1946 get_stream_filter(attr_index
, GITATTRIBUTES_FILE
,
1952 static int merge_3way(struct merge_options
*opt
,
1954 const struct object_id
*o
,
1955 const struct object_id
*a
,
1956 const struct object_id
*b
,
1957 const char *pathnames
[3],
1958 const int extra_marker_size
,
1959 mmbuffer_t
*result_buf
)
1961 mmfile_t orig
, src1
, src2
;
1962 struct ll_merge_options ll_opts
= {0};
1963 char *base
, *name1
, *name2
;
1964 enum ll_merge_result merge_status
;
1966 if (!opt
->priv
->attr_index
.initialized
)
1967 initialize_attr_index(opt
);
1969 ll_opts
.renormalize
= opt
->renormalize
;
1970 ll_opts
.extra_marker_size
= extra_marker_size
;
1971 ll_opts
.xdl_opts
= opt
->xdl_opts
;
1973 if (opt
->priv
->call_depth
) {
1974 ll_opts
.virtual_ancestor
= 1;
1975 ll_opts
.variant
= 0;
1977 switch (opt
->recursive_variant
) {
1978 case MERGE_VARIANT_OURS
:
1979 ll_opts
.variant
= XDL_MERGE_FAVOR_OURS
;
1981 case MERGE_VARIANT_THEIRS
:
1982 ll_opts
.variant
= XDL_MERGE_FAVOR_THEIRS
;
1985 ll_opts
.variant
= 0;
1990 assert(pathnames
[0] && pathnames
[1] && pathnames
[2] && opt
->ancestor
);
1991 if (pathnames
[0] == pathnames
[1] && pathnames
[1] == pathnames
[2]) {
1992 base
= mkpathdup("%s", opt
->ancestor
);
1993 name1
= mkpathdup("%s", opt
->branch1
);
1994 name2
= mkpathdup("%s", opt
->branch2
);
1996 base
= mkpathdup("%s:%s", opt
->ancestor
, pathnames
[0]);
1997 name1
= mkpathdup("%s:%s", opt
->branch1
, pathnames
[1]);
1998 name2
= mkpathdup("%s:%s", opt
->branch2
, pathnames
[2]);
2001 read_mmblob(&orig
, o
);
2002 read_mmblob(&src1
, a
);
2003 read_mmblob(&src2
, b
);
2005 merge_status
= ll_merge(result_buf
, path
, &orig
, base
,
2006 &src1
, name1
, &src2
, name2
,
2007 &opt
->priv
->attr_index
, &ll_opts
);
2008 if (merge_status
== LL_MERGE_BINARY_CONFLICT
)
2009 path_msg(opt
, CONFLICT_BINARY
, 0,
2010 path
, NULL
, NULL
, NULL
,
2011 "warning: Cannot merge binary files: %s (%s vs. %s)",
2012 path
, name1
, name2
);
2020 return merge_status
;
2023 static int handle_content_merge(struct merge_options
*opt
,
2025 const struct version_info
*o
,
2026 const struct version_info
*a
,
2027 const struct version_info
*b
,
2028 const char *pathnames
[3],
2029 const int extra_marker_size
,
2030 struct version_info
*result
)
2033 * path is the target location where we want to put the file, and
2034 * is used to determine any normalization rules in ll_merge.
2036 * The normal case is that path and all entries in pathnames are
2037 * identical, though renames can affect which path we got one of
2038 * the three blobs to merge on various sides of history.
2040 * extra_marker_size is the amount to extend conflict markers in
2041 * ll_merge; this is needed if we have content merges of content
2042 * merges, which happens for example with rename/rename(2to1) and
2043 * rename/add conflicts.
2048 * handle_content_merge() needs both files to be of the same type, i.e.
2049 * both files OR both submodules OR both symlinks. Conflicting types
2050 * needs to be handled elsewhere.
2052 assert((S_IFMT
& a
->mode
) == (S_IFMT
& b
->mode
));
2055 if (a
->mode
== b
->mode
|| a
->mode
== o
->mode
)
2056 result
->mode
= b
->mode
;
2058 /* must be the 100644/100755 case */
2059 assert(S_ISREG(a
->mode
));
2060 result
->mode
= a
->mode
;
2061 clean
= (b
->mode
== o
->mode
);
2063 * FIXME: If opt->priv->call_depth && !clean, then we really
2064 * should not make result->mode match either a->mode or
2065 * b->mode; that causes t6036 "check conflicting mode for
2066 * regular file" to fail. It would be best to use some other
2067 * mode, but we'll confuse all kinds of stuff if we use one
2068 * where S_ISREG(result->mode) isn't true, and if we use
2069 * something like 0100666, then tree-walk.c's calls to
2070 * canon_mode() will just normalize that to 100644 for us and
2071 * thus not solve anything.
2073 * Figure out if there's some kind of way we can work around
2079 * Trivial oid merge.
2081 * Note: While one might assume that the next four lines would
2082 * be unnecessary due to the fact that match_mask is often
2083 * setup and already handled, renames don't always take care
2086 if (oideq(&a
->oid
, &b
->oid
) || oideq(&a
->oid
, &o
->oid
))
2087 oidcpy(&result
->oid
, &b
->oid
);
2088 else if (oideq(&b
->oid
, &o
->oid
))
2089 oidcpy(&result
->oid
, &a
->oid
);
2091 /* Remaining rules depend on file vs. submodule vs. symlink. */
2092 else if (S_ISREG(a
->mode
)) {
2093 mmbuffer_t result_buf
;
2094 int ret
= 0, merge_status
;
2098 * If 'o' is different type, treat it as null so we do a
2101 two_way
= ((S_IFMT
& o
->mode
) != (S_IFMT
& a
->mode
));
2103 merge_status
= merge_3way(opt
, path
,
2104 two_way
? null_oid() : &o
->oid
,
2106 pathnames
, extra_marker_size
,
2109 if ((merge_status
< 0) || !result_buf
.ptr
)
2110 ret
= error(_("failed to execute internal merge"));
2113 write_object_file(result_buf
.ptr
, result_buf
.size
,
2114 OBJ_BLOB
, &result
->oid
))
2115 ret
= error(_("unable to add %s to database"), path
);
2117 free(result_buf
.ptr
);
2120 clean
&= (merge_status
== 0);
2121 path_msg(opt
, INFO_AUTO_MERGING
, 1, path
, NULL
, NULL
, NULL
,
2122 _("Auto-merging %s"), path
);
2123 } else if (S_ISGITLINK(a
->mode
)) {
2124 int two_way
= ((S_IFMT
& o
->mode
) != (S_IFMT
& a
->mode
));
2125 clean
= merge_submodule(opt
, pathnames
[0],
2126 two_way
? null_oid() : &o
->oid
,
2127 &a
->oid
, &b
->oid
, &result
->oid
);
2128 if (opt
->priv
->call_depth
&& two_way
&& !clean
) {
2129 result
->mode
= o
->mode
;
2130 oidcpy(&result
->oid
, &o
->oid
);
2132 } else if (S_ISLNK(a
->mode
)) {
2133 if (opt
->priv
->call_depth
) {
2135 result
->mode
= o
->mode
;
2136 oidcpy(&result
->oid
, &o
->oid
);
2138 switch (opt
->recursive_variant
) {
2139 case MERGE_VARIANT_NORMAL
:
2141 oidcpy(&result
->oid
, &a
->oid
);
2143 case MERGE_VARIANT_OURS
:
2144 oidcpy(&result
->oid
, &a
->oid
);
2146 case MERGE_VARIANT_THEIRS
:
2147 oidcpy(&result
->oid
, &b
->oid
);
2152 BUG("unsupported object type in the tree: %06o for %s",
2158 /*** Function Grouping: functions related to detect_and_process_renames(), ***
2159 *** which are split into directory and regular rename detection sections. ***/
2161 /*** Function Grouping: functions related to directory rename detection ***/
2163 struct collision_info
{
2164 struct string_list source_files
;
2165 unsigned reported_already
:1;
2169 * Return a new string that replaces the beginning portion (which matches
2170 * rename_info->key), with rename_info->util.new_dir. In perl-speak:
2171 * new_path_name = (old_path =~ s/rename_info->key/rename_info->value/);
2173 * Caller must ensure that old_path starts with rename_info->key + '/'.
2175 static char *apply_dir_rename(struct strmap_entry
*rename_info
,
2176 const char *old_path
)
2178 struct strbuf new_path
= STRBUF_INIT
;
2179 const char *old_dir
= rename_info
->key
;
2180 const char *new_dir
= rename_info
->value
;
2181 int oldlen
, newlen
, new_dir_len
;
2183 oldlen
= strlen(old_dir
);
2184 if (*new_dir
== '\0')
2186 * If someone renamed/merged a subdirectory into the root
2187 * directory (e.g. 'some/subdir' -> ''), then we want to
2190 * as the rename; we need to make old_path + oldlen advance
2191 * past the '/' character.
2194 new_dir_len
= strlen(new_dir
);
2195 newlen
= new_dir_len
+ (strlen(old_path
) - oldlen
) + 1;
2196 strbuf_grow(&new_path
, newlen
);
2197 strbuf_add(&new_path
, new_dir
, new_dir_len
);
2198 strbuf_addstr(&new_path
, &old_path
[oldlen
]);
2200 return strbuf_detach(&new_path
, NULL
);
2203 static int path_in_way(struct strmap
*paths
, const char *path
, unsigned side_mask
)
2205 struct merged_info
*mi
= strmap_get(paths
, path
);
2206 struct conflict_info
*ci
;
2209 INITIALIZE_CI(ci
, mi
);
2210 return mi
->clean
|| (side_mask
& (ci
->filemask
| ci
->dirmask
));
2214 * See if there is a directory rename for path, and if there are any file
2215 * level conflicts on the given side for the renamed location. If there is
2216 * a rename and there are no conflicts, return the new name. Otherwise,
2219 static char *handle_path_level_conflicts(struct merge_options
*opt
,
2221 unsigned side_index
,
2222 struct strmap_entry
*rename_info
,
2223 struct strmap
*collisions
)
2225 char *new_path
= NULL
;
2226 struct collision_info
*c_info
;
2228 struct strbuf collision_paths
= STRBUF_INIT
;
2231 * entry has the mapping of old directory name to new directory name
2232 * that we want to apply to path.
2234 new_path
= apply_dir_rename(rename_info
, path
);
2236 BUG("Failed to apply directory rename!");
2239 * The caller needs to have ensured that it has pre-populated
2240 * collisions with all paths that map to new_path. Do a quick check
2241 * to ensure that's the case.
2243 c_info
= strmap_get(collisions
, new_path
);
2245 BUG("c_info is NULL");
2248 * Check for one-sided add/add/.../add conflicts, i.e.
2249 * where implicit renames from the other side doing
2250 * directory rename(s) can affect this side of history
2251 * to put multiple paths into the same location. Warn
2252 * and bail on directory renames for such paths.
2254 if (c_info
->reported_already
) {
2256 } else if (path_in_way(&opt
->priv
->paths
, new_path
, 1 << side_index
)) {
2257 c_info
->reported_already
= 1;
2258 strbuf_add_separated_string_list(&collision_paths
, ", ",
2259 &c_info
->source_files
);
2260 path_msg(opt
, CONFLICT_DIR_RENAME_FILE_IN_WAY
, 0,
2261 new_path
, NULL
, NULL
, &c_info
->source_files
,
2262 _("CONFLICT (implicit dir rename): Existing "
2263 "file/dir at %s in the way of implicit "
2264 "directory rename(s) putting the following "
2265 "path(s) there: %s."),
2266 new_path
, collision_paths
.buf
);
2268 } else if (c_info
->source_files
.nr
> 1) {
2269 c_info
->reported_already
= 1;
2270 strbuf_add_separated_string_list(&collision_paths
, ", ",
2271 &c_info
->source_files
);
2272 path_msg(opt
, CONFLICT_DIR_RENAME_COLLISION
, 0,
2273 new_path
, NULL
, NULL
, &c_info
->source_files
,
2274 _("CONFLICT (implicit dir rename): Cannot map "
2275 "more than one path to %s; implicit directory "
2276 "renames tried to put these paths there: %s"),
2277 new_path
, collision_paths
.buf
);
2281 /* Free memory we no longer need */
2282 strbuf_release(&collision_paths
);
2283 if (!clean
&& new_path
) {
2291 static void get_provisional_directory_renames(struct merge_options
*opt
,
2295 struct hashmap_iter iter
;
2296 struct strmap_entry
*entry
;
2297 struct rename_info
*renames
= &opt
->priv
->renames
;
2301 * dir_rename_count: old_directory -> {new_directory -> count}
2303 * dir_renames: old_directory -> best_new_directory
2304 * where best_new_directory is the one with the unique highest count.
2306 strmap_for_each_entry(&renames
->dir_rename_count
[side
], &iter
, entry
) {
2307 const char *source_dir
= entry
->key
;
2308 struct strintmap
*counts
= entry
->value
;
2309 struct hashmap_iter count_iter
;
2310 struct strmap_entry
*count_entry
;
2313 const char *best
= NULL
;
2315 strintmap_for_each_entry(counts
, &count_iter
, count_entry
) {
2316 const char *target_dir
= count_entry
->key
;
2317 intptr_t count
= (intptr_t)count_entry
->value
;
2321 else if (count
> max
) {
2330 if (bad_max
== max
) {
2331 path_msg(opt
, CONFLICT_DIR_RENAME_SPLIT
, 0,
2332 source_dir
, NULL
, NULL
, NULL
,
2333 _("CONFLICT (directory rename split): "
2334 "Unclear where to rename %s to; it was "
2335 "renamed to multiple other directories, "
2336 "with no destination getting a majority of "
2341 strmap_put(&renames
->dir_renames
[side
],
2342 source_dir
, (void*)best
);
2347 static void handle_directory_level_conflicts(struct merge_options
*opt
)
2349 struct hashmap_iter iter
;
2350 struct strmap_entry
*entry
;
2351 struct string_list duplicated
= STRING_LIST_INIT_NODUP
;
2352 struct rename_info
*renames
= &opt
->priv
->renames
;
2353 struct strmap
*side1_dir_renames
= &renames
->dir_renames
[MERGE_SIDE1
];
2354 struct strmap
*side2_dir_renames
= &renames
->dir_renames
[MERGE_SIDE2
];
2357 strmap_for_each_entry(side1_dir_renames
, &iter
, entry
) {
2358 if (strmap_contains(side2_dir_renames
, entry
->key
))
2359 string_list_append(&duplicated
, entry
->key
);
2362 for (i
= 0; i
< duplicated
.nr
; i
++) {
2363 strmap_remove(side1_dir_renames
, duplicated
.items
[i
].string
, 0);
2364 strmap_remove(side2_dir_renames
, duplicated
.items
[i
].string
, 0);
2366 string_list_clear(&duplicated
, 0);
2369 static struct strmap_entry
*check_dir_renamed(const char *path
,
2370 struct strmap
*dir_renames
)
2372 char *temp
= xstrdup(path
);
2374 struct strmap_entry
*e
= NULL
;
2376 while ((end
= strrchr(temp
, '/'))) {
2378 e
= strmap_get_entry(dir_renames
, temp
);
2386 static void compute_collisions(struct strmap
*collisions
,
2387 struct strmap
*dir_renames
,
2388 struct diff_queue_struct
*pairs
)
2392 strmap_init_with_options(collisions
, NULL
, 0);
2393 if (strmap_empty(dir_renames
))
2397 * Multiple files can be mapped to the same path due to directory
2398 * renames done by the other side of history. Since that other
2399 * side of history could have merged multiple directories into one,
2400 * if our side of history added the same file basename to each of
2401 * those directories, then all N of them would get implicitly
2402 * renamed by the directory rename detection into the same path,
2403 * and we'd get an add/add/.../add conflict, and all those adds
2404 * from *this* side of history. This is not representable in the
2405 * index, and users aren't going to easily be able to make sense of
2406 * it. So we need to provide a good warning about what's
2407 * happening, and fall back to no-directory-rename detection
2408 * behavior for those paths.
2410 * See testcases 9e and all of section 5 from t6043 for examples.
2412 for (i
= 0; i
< pairs
->nr
; ++i
) {
2413 struct strmap_entry
*rename_info
;
2414 struct collision_info
*collision_info
;
2416 struct diff_filepair
*pair
= pairs
->queue
[i
];
2418 if (pair
->status
!= 'A' && pair
->status
!= 'R')
2420 rename_info
= check_dir_renamed(pair
->two
->path
, dir_renames
);
2424 new_path
= apply_dir_rename(rename_info
, pair
->two
->path
);
2426 collision_info
= strmap_get(collisions
, new_path
);
2427 if (collision_info
) {
2430 CALLOC_ARRAY(collision_info
, 1);
2431 string_list_init_nodup(&collision_info
->source_files
);
2432 strmap_put(collisions
, new_path
, collision_info
);
2434 string_list_insert(&collision_info
->source_files
,
2439 static void free_collisions(struct strmap
*collisions
)
2441 struct hashmap_iter iter
;
2442 struct strmap_entry
*entry
;
2444 /* Free each value in the collisions map */
2445 strmap_for_each_entry(collisions
, &iter
, entry
) {
2446 struct collision_info
*info
= entry
->value
;
2447 string_list_clear(&info
->source_files
, 0);
2450 * In compute_collisions(), we set collisions.strdup_strings to 0
2451 * so that we wouldn't have to make another copy of the new_path
2452 * allocated by apply_dir_rename(). But now that we've used them
2453 * and have no other references to these strings, it is time to
2456 free_strmap_strings(collisions
);
2457 strmap_clear(collisions
, 1);
2460 static char *check_for_directory_rename(struct merge_options
*opt
,
2462 unsigned side_index
,
2463 struct strmap
*dir_renames
,
2464 struct strmap
*dir_rename_exclusions
,
2465 struct strmap
*collisions
,
2469 struct strmap_entry
*rename_info
;
2470 struct strmap_entry
*otherinfo
;
2471 const char *new_dir
;
2472 int other_side
= 3 - side_index
;
2475 * Cases where we don't have or don't want a directory rename for
2478 if (strmap_empty(dir_renames
))
2480 if (strmap_get(&collisions
[other_side
], path
))
2482 rename_info
= check_dir_renamed(path
, dir_renames
);
2487 * This next part is a little weird. We do not want to do an
2488 * implicit rename into a directory we renamed on our side, because
2489 * that will result in a spurious rename/rename(1to2) conflict. An
2491 * Base commit: dumbdir/afile, otherdir/bfile
2492 * Side 1: smrtdir/afile, otherdir/bfile
2493 * Side 2: dumbdir/afile, dumbdir/bfile
2494 * Here, while working on Side 1, we could notice that otherdir was
2495 * renamed/merged to dumbdir, and change the diff_filepair for
2496 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2497 * 2 will notice the rename from dumbdir to smrtdir, and do the
2498 * transitive rename to move it from dumbdir/bfile to
2499 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2500 * smrtdir, a rename/rename(1to2) conflict. We really just want
2501 * the file to end up in smrtdir. And the way to achieve that is
2502 * to not let Side1 do the rename to dumbdir, since we know that is
2503 * the source of one of our directory renames.
2505 * That's why otherinfo and dir_rename_exclusions is here.
2507 * As it turns out, this also prevents N-way transient rename
2508 * confusion; See testcases 9c and 9d of t6043.
2510 new_dir
= rename_info
->value
; /* old_dir = rename_info->key; */
2511 otherinfo
= strmap_get_entry(dir_rename_exclusions
, new_dir
);
2513 path_msg(opt
, INFO_DIR_RENAME_SKIPPED_DUE_TO_RERENAME
, 1,
2514 rename_info
->key
, path
, new_dir
, NULL
,
2515 _("WARNING: Avoiding applying %s -> %s rename "
2516 "to %s, because %s itself was renamed."),
2517 rename_info
->key
, new_dir
, path
, new_dir
);
2521 new_path
= handle_path_level_conflicts(opt
, path
, side_index
,
2523 &collisions
[side_index
]);
2524 *clean_merge
&= (new_path
!= NULL
);
2529 static void apply_directory_rename_modifications(struct merge_options
*opt
,
2530 struct diff_filepair
*pair
,
2534 * The basic idea is to get the conflict_info from opt->priv->paths
2535 * at old path, and insert it into new_path; basically just this:
2536 * ci = strmap_get(&opt->priv->paths, old_path);
2537 * strmap_remove(&opt->priv->paths, old_path, 0);
2538 * strmap_put(&opt->priv->paths, new_path, ci);
2539 * However, there are some factors complicating this:
2540 * - opt->priv->paths may already have an entry at new_path
2541 * - Each ci tracks its containing directory, so we need to
2543 * - If another ci has the same containing directory, then
2544 * the two char*'s MUST point to the same location. See the
2545 * comment in struct merged_info. strcmp equality is not
2546 * enough; we need pointer equality.
2547 * - opt->priv->paths must hold the parent directories of any
2548 * entries that are added. So, if this directory rename
2549 * causes entirely new directories, we must recursively add
2550 * parent directories.
2551 * - For each parent directory added to opt->priv->paths, we
2552 * also need to get its parent directory stored in its
2553 * conflict_info->merged.directory_name with all the same
2554 * requirements about pointer equality.
2556 struct string_list dirs_to_insert
= STRING_LIST_INIT_NODUP
;
2557 struct conflict_info
*ci
, *new_ci
;
2558 struct strmap_entry
*entry
;
2559 const char *branch_with_new_path
, *branch_with_dir_rename
;
2560 const char *old_path
= pair
->two
->path
;
2561 const char *parent_name
;
2562 const char *cur_path
;
2565 entry
= strmap_get_entry(&opt
->priv
->paths
, old_path
);
2566 old_path
= entry
->key
;
2570 /* Find parent directories missing from opt->priv->paths */
2571 cur_path
= mem_pool_strdup(&opt
->priv
->pool
, new_path
);
2572 free((char*)new_path
);
2573 new_path
= (char *)cur_path
;
2576 /* Find the parent directory of cur_path */
2577 char *last_slash
= strrchr(cur_path
, '/');
2579 parent_name
= mem_pool_strndup(&opt
->priv
->pool
,
2581 last_slash
- cur_path
);
2583 parent_name
= opt
->priv
->toplevel_dir
;
2587 /* Look it up in opt->priv->paths */
2588 entry
= strmap_get_entry(&opt
->priv
->paths
, parent_name
);
2590 parent_name
= entry
->key
; /* reuse known pointer */
2594 /* Record this is one of the directories we need to insert */
2595 string_list_append(&dirs_to_insert
, parent_name
);
2596 cur_path
= parent_name
;
2599 /* Traverse dirs_to_insert and insert them into opt->priv->paths */
2600 for (i
= dirs_to_insert
.nr
-1; i
>= 0; --i
) {
2601 struct conflict_info
*dir_ci
;
2602 char *cur_dir
= dirs_to_insert
.items
[i
].string
;
2604 CALLOC_ARRAY(dir_ci
, 1);
2606 dir_ci
->merged
.directory_name
= parent_name
;
2607 len
= strlen(parent_name
);
2608 /* len+1 because of trailing '/' character */
2609 dir_ci
->merged
.basename_offset
= (len
> 0 ? len
+1 : len
);
2610 dir_ci
->dirmask
= ci
->filemask
;
2611 strmap_put(&opt
->priv
->paths
, cur_dir
, dir_ci
);
2613 parent_name
= cur_dir
;
2616 assert(ci
->filemask
== 2 || ci
->filemask
== 4);
2617 assert(ci
->dirmask
== 0 || ci
->dirmask
== 1);
2618 if (ci
->dirmask
== 0)
2619 strmap_remove(&opt
->priv
->paths
, old_path
, 0);
2622 * This file exists on one side, but we still had a directory
2623 * at the old location that we can't remove until after
2624 * processing all paths below it. So, make a copy of ci in
2625 * new_ci and only put the file information into it.
2627 new_ci
= mem_pool_calloc(&opt
->priv
->pool
, 1, sizeof(*new_ci
));
2628 memcpy(new_ci
, ci
, sizeof(*ci
));
2629 assert(!new_ci
->match_mask
);
2630 new_ci
->dirmask
= 0;
2631 new_ci
->stages
[1].mode
= 0;
2632 oidcpy(&new_ci
->stages
[1].oid
, null_oid());
2635 * Now that we have the file information in new_ci, make sure
2636 * ci only has the directory information.
2639 ci
->merged
.clean
= 1;
2640 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
2641 if (ci
->dirmask
& (1 << i
))
2643 /* zero out any entries related to files */
2644 ci
->stages
[i
].mode
= 0;
2645 oidcpy(&ci
->stages
[i
].oid
, null_oid());
2648 // Now we want to focus on new_ci, so reassign ci to it
2652 branch_with_new_path
= (ci
->filemask
== 2) ? opt
->branch1
: opt
->branch2
;
2653 branch_with_dir_rename
= (ci
->filemask
== 2) ? opt
->branch2
: opt
->branch1
;
2655 /* Now, finally update ci and stick it into opt->priv->paths */
2656 ci
->merged
.directory_name
= parent_name
;
2657 len
= strlen(parent_name
);
2658 ci
->merged
.basename_offset
= (len
> 0 ? len
+1 : len
);
2659 new_ci
= strmap_get(&opt
->priv
->paths
, new_path
);
2661 /* Place ci back into opt->priv->paths, but at new_path */
2662 strmap_put(&opt
->priv
->paths
, new_path
, ci
);
2666 /* A few sanity checks */
2668 assert(ci
->filemask
== 2 || ci
->filemask
== 4);
2669 assert((new_ci
->filemask
& ci
->filemask
) == 0);
2670 assert(!new_ci
->merged
.clean
);
2672 /* Copy stuff from ci into new_ci */
2673 new_ci
->filemask
|= ci
->filemask
;
2674 if (new_ci
->dirmask
)
2675 new_ci
->df_conflict
= 1;
2676 index
= (ci
->filemask
>> 1);
2677 new_ci
->pathnames
[index
] = ci
->pathnames
[index
];
2678 new_ci
->stages
[index
].mode
= ci
->stages
[index
].mode
;
2679 oidcpy(&new_ci
->stages
[index
].oid
, &ci
->stages
[index
].oid
);
2684 if (opt
->detect_directory_renames
== MERGE_DIRECTORY_RENAMES_TRUE
) {
2685 /* Notify user of updated path */
2686 if (pair
->status
== 'A')
2687 path_msg(opt
, INFO_DIR_RENAME_APPLIED
, 1,
2688 new_path
, old_path
, NULL
, NULL
,
2689 _("Path updated: %s added in %s inside a "
2690 "directory that was renamed in %s; moving "
2692 old_path
, branch_with_new_path
,
2693 branch_with_dir_rename
, new_path
);
2695 path_msg(opt
, INFO_DIR_RENAME_APPLIED
, 1,
2696 new_path
, old_path
, NULL
, NULL
,
2697 _("Path updated: %s renamed to %s in %s, "
2698 "inside a directory that was renamed in %s; "
2699 "moving it to %s."),
2700 pair
->one
->path
, old_path
, branch_with_new_path
,
2701 branch_with_dir_rename
, new_path
);
2704 * opt->detect_directory_renames has the value
2705 * MERGE_DIRECTORY_RENAMES_CONFLICT, so mark these as conflicts.
2707 ci
->path_conflict
= 1;
2708 if (pair
->status
== 'A')
2709 path_msg(opt
, CONFLICT_DIR_RENAME_SUGGESTED
, 1,
2710 new_path
, old_path
, NULL
, NULL
,
2711 _("CONFLICT (file location): %s added in %s "
2712 "inside a directory that was renamed in %s, "
2713 "suggesting it should perhaps be moved to "
2715 old_path
, branch_with_new_path
,
2716 branch_with_dir_rename
, new_path
);
2718 path_msg(opt
, CONFLICT_DIR_RENAME_SUGGESTED
, 1,
2719 new_path
, old_path
, NULL
, NULL
,
2720 _("CONFLICT (file location): %s renamed to %s "
2721 "in %s, inside a directory that was renamed "
2722 "in %s, suggesting it should perhaps be "
2724 pair
->one
->path
, old_path
, branch_with_new_path
,
2725 branch_with_dir_rename
, new_path
);
2729 * Finally, record the new location.
2731 pair
->two
->path
= new_path
;
2734 /*** Function Grouping: functions related to regular rename detection ***/
2736 static int process_renames(struct merge_options
*opt
,
2737 struct diff_queue_struct
*renames
)
2739 int clean_merge
= 1, i
;
2741 for (i
= 0; i
< renames
->nr
; ++i
) {
2742 const char *oldpath
= NULL
, *newpath
;
2743 struct diff_filepair
*pair
= renames
->queue
[i
];
2744 struct conflict_info
*oldinfo
= NULL
, *newinfo
= NULL
;
2745 struct strmap_entry
*old_ent
, *new_ent
;
2746 unsigned int old_sidemask
;
2747 int target_index
, other_source_index
;
2748 int source_deleted
, collision
, type_changed
;
2749 const char *rename_branch
= NULL
, *delete_branch
= NULL
;
2751 old_ent
= strmap_get_entry(&opt
->priv
->paths
, pair
->one
->path
);
2752 new_ent
= strmap_get_entry(&opt
->priv
->paths
, pair
->two
->path
);
2754 oldpath
= old_ent
->key
;
2755 oldinfo
= old_ent
->value
;
2757 newpath
= pair
->two
->path
;
2759 newpath
= new_ent
->key
;
2760 newinfo
= new_ent
->value
;
2764 * If pair->one->path isn't in opt->priv->paths, that means
2765 * that either directory rename detection removed that
2766 * path, or a parent directory of oldpath was resolved and
2767 * we don't even need the rename; in either case, we can
2768 * skip it. If oldinfo->merged.clean, then the other side
2769 * of history had no changes to oldpath and we don't need
2770 * the rename and can skip it.
2772 if (!oldinfo
|| oldinfo
->merged
.clean
)
2776 * diff_filepairs have copies of pathnames, thus we have to
2777 * use standard 'strcmp()' (negated) instead of '=='.
2779 if (i
+ 1 < renames
->nr
&&
2780 !strcmp(oldpath
, renames
->queue
[i
+1]->one
->path
)) {
2781 /* Handle rename/rename(1to2) or rename/rename(1to1) */
2782 const char *pathnames
[3];
2783 struct version_info merged
;
2784 struct conflict_info
*base
, *side1
, *side2
;
2785 unsigned was_binary_blob
= 0;
2787 pathnames
[0] = oldpath
;
2788 pathnames
[1] = newpath
;
2789 pathnames
[2] = renames
->queue
[i
+1]->two
->path
;
2791 base
= strmap_get(&opt
->priv
->paths
, pathnames
[0]);
2792 side1
= strmap_get(&opt
->priv
->paths
, pathnames
[1]);
2793 side2
= strmap_get(&opt
->priv
->paths
, pathnames
[2]);
2799 if (!strcmp(pathnames
[1], pathnames
[2])) {
2800 struct rename_info
*ri
= &opt
->priv
->renames
;
2803 /* Both sides renamed the same way */
2804 assert(side1
== side2
);
2805 memcpy(&side1
->stages
[0], &base
->stages
[0],
2807 side1
->filemask
|= (1 << MERGE_BASE
);
2808 /* Mark base as resolved by removal */
2809 base
->merged
.is_null
= 1;
2810 base
->merged
.clean
= 1;
2813 * Disable remembering renames optimization;
2814 * rename/rename(1to1) is incredibly rare, and
2815 * just disabling the optimization is easier
2816 * than purging cached_pairs,
2817 * cached_target_names, and dir_rename_counts.
2819 for (j
= 0; j
< 3; j
++)
2820 ri
->merge_trees
[j
] = NULL
;
2822 /* We handled both renames, i.e. i+1 handled */
2824 /* Move to next rename */
2828 /* This is a rename/rename(1to2) */
2829 clean_merge
= handle_content_merge(opt
,
2835 1 + 2 * opt
->priv
->call_depth
,
2837 if (clean_merge
< 0)
2840 merged
.mode
== side1
->stages
[1].mode
&&
2841 oideq(&merged
.oid
, &side1
->stages
[1].oid
))
2842 was_binary_blob
= 1;
2843 memcpy(&side1
->stages
[1], &merged
, sizeof(merged
));
2844 if (was_binary_blob
) {
2846 * Getting here means we were attempting to
2847 * merge a binary blob.
2849 * Since we can't merge binaries,
2850 * handle_content_merge() just takes one
2851 * side. But we don't want to copy the
2852 * contents of one side to both paths. We
2853 * used the contents of side1 above for
2854 * side1->stages, let's use the contents of
2855 * side2 for side2->stages below.
2857 oidcpy(&merged
.oid
, &side2
->stages
[2].oid
);
2858 merged
.mode
= side2
->stages
[2].mode
;
2860 memcpy(&side2
->stages
[2], &merged
, sizeof(merged
));
2862 side1
->path_conflict
= 1;
2863 side2
->path_conflict
= 1;
2865 * TODO: For renames we normally remove the path at the
2866 * old name. It would thus seem consistent to do the
2867 * same for rename/rename(1to2) cases, but we haven't
2868 * done so traditionally and a number of the regression
2869 * tests now encode an expectation that the file is
2870 * left there at stage 1. If we ever decide to change
2871 * this, add the following two lines here:
2872 * base->merged.is_null = 1;
2873 * base->merged.clean = 1;
2874 * and remove the setting of base->path_conflict to 1.
2876 base
->path_conflict
= 1;
2877 path_msg(opt
, CONFLICT_RENAME_RENAME
, 0,
2878 pathnames
[0], pathnames
[1], pathnames
[2], NULL
,
2879 _("CONFLICT (rename/rename): %s renamed to "
2880 "%s in %s and to %s in %s."),
2882 pathnames
[1], opt
->branch1
,
2883 pathnames
[2], opt
->branch2
);
2885 i
++; /* We handled both renames, i.e. i+1 handled */
2891 target_index
= pair
->score
; /* from collect_renames() */
2892 assert(target_index
== 1 || target_index
== 2);
2893 other_source_index
= 3 - target_index
;
2894 old_sidemask
= (1 << other_source_index
); /* 2 or 4 */
2895 source_deleted
= (oldinfo
->filemask
== 1);
2896 collision
= ((newinfo
->filemask
& old_sidemask
) != 0);
2897 type_changed
= !source_deleted
&&
2898 (S_ISREG(oldinfo
->stages
[other_source_index
].mode
) !=
2899 S_ISREG(newinfo
->stages
[target_index
].mode
));
2900 if (type_changed
&& collision
) {
2902 * special handling so later blocks can handle this...
2904 * if type_changed && collision are both true, then this
2905 * was really a double rename, but one side wasn't
2906 * detected due to lack of break detection. I.e.
2908 * orig: has normal file 'foo'
2909 * side1: renames 'foo' to 'bar', adds 'foo' symlink
2910 * side2: renames 'foo' to 'bar'
2911 * In this case, the foo->bar rename on side1 won't be
2912 * detected because the new symlink named 'foo' is
2913 * there and we don't do break detection. But we detect
2914 * this here because we don't want to merge the content
2915 * of the foo symlink with the foo->bar file, so we
2916 * have some logic to handle this special case. The
2917 * easiest way to do that is make 'bar' on side1 not
2918 * be considered a colliding file but the other part
2919 * of a normal rename. If the file is very different,
2920 * well we're going to get content merge conflicts
2921 * anyway so it doesn't hurt. And if the colliding
2922 * file also has a different type, that'll be handled
2923 * by the content merge logic in process_entry() too.
2925 * See also t6430, 'rename vs. rename/symlink'
2929 if (source_deleted
) {
2930 if (target_index
== 1) {
2931 rename_branch
= opt
->branch1
;
2932 delete_branch
= opt
->branch2
;
2934 rename_branch
= opt
->branch2
;
2935 delete_branch
= opt
->branch1
;
2939 assert(source_deleted
|| oldinfo
->filemask
& old_sidemask
);
2941 /* Need to check for special types of rename conflicts... */
2942 if (collision
&& !source_deleted
) {
2943 /* collision: rename/add or rename/rename(2to1) */
2944 const char *pathnames
[3];
2945 struct version_info merged
;
2947 struct conflict_info
*base
, *side1
, *side2
;
2950 pathnames
[0] = oldpath
;
2951 pathnames
[other_source_index
] = oldpath
;
2952 pathnames
[target_index
] = newpath
;
2954 base
= strmap_get(&opt
->priv
->paths
, pathnames
[0]);
2955 side1
= strmap_get(&opt
->priv
->paths
, pathnames
[1]);
2956 side2
= strmap_get(&opt
->priv
->paths
, pathnames
[2]);
2962 clean
= handle_content_merge(opt
, pair
->one
->path
,
2967 1 + 2 * opt
->priv
->call_depth
,
2972 memcpy(&newinfo
->stages
[target_index
], &merged
,
2975 path_msg(opt
, CONFLICT_RENAME_COLLIDES
, 0,
2976 newpath
, oldpath
, NULL
, NULL
,
2977 _("CONFLICT (rename involved in "
2978 "collision): rename of %s -> %s has "
2979 "content conflicts AND collides "
2980 "with another path; this may result "
2981 "in nested conflict markers."),
2984 } else if (collision
&& source_deleted
) {
2986 * rename/add/delete or rename/rename(2to1)/delete:
2987 * since oldpath was deleted on the side that didn't
2988 * do the rename, there's not much of a content merge
2989 * we can do for the rename. oldinfo->merged.is_null
2990 * was already set, so we just leave things as-is so
2991 * they look like an add/add conflict.
2994 newinfo
->path_conflict
= 1;
2995 path_msg(opt
, CONFLICT_RENAME_DELETE
, 0,
2996 newpath
, oldpath
, NULL
, NULL
,
2997 _("CONFLICT (rename/delete): %s renamed "
2998 "to %s in %s, but deleted in %s."),
2999 oldpath
, newpath
, rename_branch
, delete_branch
);
3002 * a few different cases...start by copying the
3003 * existing stage(s) from oldinfo over the newinfo
3004 * and update the pathname(s).
3006 memcpy(&newinfo
->stages
[0], &oldinfo
->stages
[0],
3007 sizeof(newinfo
->stages
[0]));
3008 newinfo
->filemask
|= (1 << MERGE_BASE
);
3009 newinfo
->pathnames
[0] = oldpath
;
3011 /* rename vs. typechange */
3012 /* Mark the original as resolved by removal */
3013 memcpy(&oldinfo
->stages
[0].oid
, null_oid(),
3014 sizeof(oldinfo
->stages
[0].oid
));
3015 oldinfo
->stages
[0].mode
= 0;
3016 oldinfo
->filemask
&= 0x06;
3017 } else if (source_deleted
) {
3019 newinfo
->path_conflict
= 1;
3020 path_msg(opt
, CONFLICT_RENAME_DELETE
, 0,
3021 newpath
, oldpath
, NULL
, NULL
,
3022 _("CONFLICT (rename/delete): %s renamed"
3023 " to %s in %s, but deleted in %s."),
3025 rename_branch
, delete_branch
);
3028 memcpy(&newinfo
->stages
[other_source_index
],
3029 &oldinfo
->stages
[other_source_index
],
3030 sizeof(newinfo
->stages
[0]));
3031 newinfo
->filemask
|= (1 << other_source_index
);
3032 newinfo
->pathnames
[other_source_index
] = oldpath
;
3036 if (!type_changed
) {
3037 /* Mark the original as resolved by removal */
3038 oldinfo
->merged
.is_null
= 1;
3039 oldinfo
->merged
.clean
= 1;
3047 static inline int possible_side_renames(struct rename_info
*renames
,
3048 unsigned side_index
)
3050 return renames
->pairs
[side_index
].nr
> 0 &&
3051 !strintmap_empty(&renames
->relevant_sources
[side_index
]);
3054 static inline int possible_renames(struct rename_info
*renames
)
3056 return possible_side_renames(renames
, 1) ||
3057 possible_side_renames(renames
, 2) ||
3058 !strmap_empty(&renames
->cached_pairs
[1]) ||
3059 !strmap_empty(&renames
->cached_pairs
[2]);
3062 static void resolve_diffpair_statuses(struct diff_queue_struct
*q
)
3065 * A simplified version of diff_resolve_rename_copy(); would probably
3066 * just use that function but it's static...
3069 struct diff_filepair
*p
;
3071 for (i
= 0; i
< q
->nr
; ++i
) {
3073 p
->status
= 0; /* undecided */
3074 if (!DIFF_FILE_VALID(p
->one
))
3075 p
->status
= DIFF_STATUS_ADDED
;
3076 else if (!DIFF_FILE_VALID(p
->two
))
3077 p
->status
= DIFF_STATUS_DELETED
;
3078 else if (DIFF_PAIR_RENAME(p
))
3079 p
->status
= DIFF_STATUS_RENAMED
;
3083 static void prune_cached_from_relevant(struct rename_info
*renames
,
3086 /* Reason for this function described in add_pair() */
3087 struct hashmap_iter iter
;
3088 struct strmap_entry
*entry
;
3090 /* Remove from relevant_sources all entries in cached_pairs[side] */
3091 strmap_for_each_entry(&renames
->cached_pairs
[side
], &iter
, entry
) {
3092 strintmap_remove(&renames
->relevant_sources
[side
],
3095 /* Remove from relevant_sources all entries in cached_irrelevant[side] */
3096 strset_for_each_entry(&renames
->cached_irrelevant
[side
], &iter
, entry
) {
3097 strintmap_remove(&renames
->relevant_sources
[side
],
3102 static void use_cached_pairs(struct merge_options
*opt
,
3103 struct strmap
*cached_pairs
,
3104 struct diff_queue_struct
*pairs
)
3106 struct hashmap_iter iter
;
3107 struct strmap_entry
*entry
;
3110 * Add to side_pairs all entries from renames->cached_pairs[side_index].
3111 * (Info in cached_irrelevant[side_index] is not relevant here.)
3113 strmap_for_each_entry(cached_pairs
, &iter
, entry
) {
3114 struct diff_filespec
*one
, *two
;
3115 const char *old_name
= entry
->key
;
3116 const char *new_name
= entry
->value
;
3118 new_name
= old_name
;
3121 * cached_pairs has *copies* of old_name and new_name,
3122 * because it has to persist across merges. Since
3123 * pool_alloc_filespec() will just re-use the existing
3124 * filenames, which will also get re-used by
3125 * opt->priv->paths if they become renames, and then
3126 * get freed at the end of the merge, that would leave
3127 * the copy in cached_pairs dangling. Avoid this by
3128 * making a copy here.
3130 old_name
= mem_pool_strdup(&opt
->priv
->pool
, old_name
);
3131 new_name
= mem_pool_strdup(&opt
->priv
->pool
, new_name
);
3133 /* We don't care about oid/mode, only filenames and status */
3134 one
= pool_alloc_filespec(&opt
->priv
->pool
, old_name
);
3135 two
= pool_alloc_filespec(&opt
->priv
->pool
, new_name
);
3136 pool_diff_queue(&opt
->priv
->pool
, pairs
, one
, two
);
3137 pairs
->queue
[pairs
->nr
-1]->status
= entry
->value
? 'R' : 'D';
3141 static void cache_new_pair(struct rename_info
*renames
,
3148 new_path
= xstrdup(new_path
);
3149 old_value
= strmap_put(&renames
->cached_pairs
[side
],
3150 old_path
, new_path
);
3151 strset_add(&renames
->cached_target_names
[side
], new_path
);
3158 static void possibly_cache_new_pair(struct rename_info
*renames
,
3159 struct diff_filepair
*p
,
3163 int dir_renamed_side
= 0;
3167 * Directory renames happen on the other side of history from
3168 * the side that adds new files to the old directory.
3170 dir_renamed_side
= 3 - side
;
3172 int val
= strintmap_get(&renames
->relevant_sources
[side
],
3174 if (val
== RELEVANT_NO_MORE
) {
3175 assert(p
->status
== 'D');
3176 strset_add(&renames
->cached_irrelevant
[side
],
3183 if (p
->status
== 'D') {
3185 * If we already had this delete, we'll just set it's value
3186 * to NULL again, so no harm.
3188 strmap_put(&renames
->cached_pairs
[side
], p
->one
->path
, NULL
);
3189 } else if (p
->status
== 'R') {
3191 new_path
= p
->two
->path
;
3193 cache_new_pair(renames
, dir_renamed_side
,
3194 p
->two
->path
, new_path
, 0);
3195 cache_new_pair(renames
, side
, p
->one
->path
, new_path
, 1);
3196 } else if (p
->status
== 'A' && new_path
) {
3197 cache_new_pair(renames
, dir_renamed_side
,
3198 p
->two
->path
, new_path
, 0);
3202 static int compare_pairs(const void *a_
, const void *b_
)
3204 const struct diff_filepair
*a
= *((const struct diff_filepair
**)a_
);
3205 const struct diff_filepair
*b
= *((const struct diff_filepair
**)b_
);
3207 return strcmp(a
->one
->path
, b
->one
->path
);
3210 /* Call diffcore_rename() to update deleted/added pairs into rename pairs */
3211 static int detect_regular_renames(struct merge_options
*opt
,
3212 unsigned side_index
)
3214 struct diff_options diff_opts
;
3215 struct rename_info
*renames
= &opt
->priv
->renames
;
3217 prune_cached_from_relevant(renames
, side_index
);
3218 if (!possible_side_renames(renames
, side_index
)) {
3220 * No rename detection needed for this side, but we still need
3221 * to make sure 'adds' are marked correctly in case the other
3222 * side had directory renames.
3224 resolve_diffpair_statuses(&renames
->pairs
[side_index
]);
3228 partial_clear_dir_rename_count(&renames
->dir_rename_count
[side_index
]);
3229 repo_diff_setup(opt
->repo
, &diff_opts
);
3230 diff_opts
.flags
.recursive
= 1;
3231 diff_opts
.flags
.rename_empty
= 0;
3232 diff_opts
.detect_rename
= DIFF_DETECT_RENAME
;
3233 diff_opts
.rename_limit
= opt
->rename_limit
;
3234 if (opt
->rename_limit
<= 0)
3235 diff_opts
.rename_limit
= 7000;
3236 diff_opts
.rename_score
= opt
->rename_score
;
3237 diff_opts
.show_rename_progress
= opt
->show_rename_progress
;
3238 diff_opts
.output_format
= DIFF_FORMAT_NO_OUTPUT
;
3239 diff_setup_done(&diff_opts
);
3241 diff_queued_diff
= renames
->pairs
[side_index
];
3242 trace2_region_enter("diff", "diffcore_rename", opt
->repo
);
3243 diffcore_rename_extended(&diff_opts
,
3245 &renames
->relevant_sources
[side_index
],
3246 &renames
->dirs_removed
[side_index
],
3247 &renames
->dir_rename_count
[side_index
],
3248 &renames
->cached_pairs
[side_index
]);
3249 trace2_region_leave("diff", "diffcore_rename", opt
->repo
);
3250 resolve_diffpair_statuses(&diff_queued_diff
);
3252 if (diff_opts
.needed_rename_limit
> 0)
3253 renames
->redo_after_renames
= 0;
3254 if (diff_opts
.needed_rename_limit
> renames
->needed_limit
)
3255 renames
->needed_limit
= diff_opts
.needed_rename_limit
;
3257 renames
->pairs
[side_index
] = diff_queued_diff
;
3259 diff_opts
.output_format
= DIFF_FORMAT_NO_OUTPUT
;
3260 diff_queued_diff
.nr
= 0;
3261 diff_queued_diff
.queue
= NULL
;
3262 diff_flush(&diff_opts
);
3268 * Get information of all renames which occurred in 'side_pairs', making use
3269 * of any implicit directory renames in side_dir_renames (also making use of
3270 * implicit directory renames rename_exclusions as needed by
3271 * check_for_directory_rename()). Add all (updated) renames into result.
3273 static int collect_renames(struct merge_options
*opt
,
3274 struct diff_queue_struct
*result
,
3275 unsigned side_index
,
3276 struct strmap
*collisions
,
3277 struct strmap
*dir_renames_for_side
,
3278 struct strmap
*rename_exclusions
)
3281 struct diff_queue_struct
*side_pairs
;
3282 struct rename_info
*renames
= &opt
->priv
->renames
;
3284 side_pairs
= &renames
->pairs
[side_index
];
3286 for (i
= 0; i
< side_pairs
->nr
; ++i
) {
3287 struct diff_filepair
*p
= side_pairs
->queue
[i
];
3288 char *new_path
; /* non-NULL only with directory renames */
3290 if (p
->status
!= 'A' && p
->status
!= 'R') {
3291 possibly_cache_new_pair(renames
, p
, side_index
, NULL
);
3292 pool_diff_free_filepair(&opt
->priv
->pool
, p
);
3296 new_path
= check_for_directory_rename(opt
, p
->two
->path
,
3298 dir_renames_for_side
,
3303 possibly_cache_new_pair(renames
, p
, side_index
, new_path
);
3304 if (p
->status
!= 'R' && !new_path
) {
3305 pool_diff_free_filepair(&opt
->priv
->pool
, p
);
3310 apply_directory_rename_modifications(opt
, p
, new_path
);
3313 * p->score comes back from diffcore_rename_extended() with
3314 * the similarity of the renamed file. The similarity is
3315 * was used to determine that the two files were related
3316 * and are a rename, which we have already used, but beyond
3317 * that we have no use for the similarity. So p->score is
3318 * now irrelevant. However, process_renames() will need to
3319 * know which side of the merge this rename was associated
3320 * with, so overwrite p->score with that value.
3322 p
->score
= side_index
;
3323 result
->queue
[result
->nr
++] = p
;
3329 static int detect_and_process_renames(struct merge_options
*opt
)
3331 struct diff_queue_struct combined
= { 0 };
3332 struct rename_info
*renames
= &opt
->priv
->renames
;
3333 struct strmap collisions
[3];
3334 int need_dir_renames
, s
, i
, clean
= 1;
3335 unsigned detection_run
= 0;
3337 if (!possible_renames(renames
))
3340 trace2_region_enter("merge", "regular renames", opt
->repo
);
3341 detection_run
|= detect_regular_renames(opt
, MERGE_SIDE1
);
3342 detection_run
|= detect_regular_renames(opt
, MERGE_SIDE2
);
3343 if (renames
->needed_limit
) {
3344 renames
->cached_pairs_valid_side
= 0;
3345 renames
->redo_after_renames
= 0;
3347 if (renames
->redo_after_renames
&& detection_run
) {
3349 struct diff_filepair
*p
;
3351 /* Cache the renames, we found */
3352 for (side
= MERGE_SIDE1
; side
<= MERGE_SIDE2
; side
++) {
3353 for (i
= 0; i
< renames
->pairs
[side
].nr
; ++i
) {
3354 p
= renames
->pairs
[side
].queue
[i
];
3355 possibly_cache_new_pair(renames
, p
, side
, NULL
);
3359 /* Restart the merge with the cached renames */
3360 renames
->redo_after_renames
= 2;
3361 trace2_region_leave("merge", "regular renames", opt
->repo
);
3364 use_cached_pairs(opt
, &renames
->cached_pairs
[1], &renames
->pairs
[1]);
3365 use_cached_pairs(opt
, &renames
->cached_pairs
[2], &renames
->pairs
[2]);
3366 trace2_region_leave("merge", "regular renames", opt
->repo
);
3368 trace2_region_enter("merge", "directory renames", opt
->repo
);
3370 !opt
->priv
->call_depth
&&
3371 (opt
->detect_directory_renames
== MERGE_DIRECTORY_RENAMES_TRUE
||
3372 opt
->detect_directory_renames
== MERGE_DIRECTORY_RENAMES_CONFLICT
);
3374 if (need_dir_renames
) {
3375 get_provisional_directory_renames(opt
, MERGE_SIDE1
, &clean
);
3376 get_provisional_directory_renames(opt
, MERGE_SIDE2
, &clean
);
3377 handle_directory_level_conflicts(opt
);
3380 ALLOC_GROW(combined
.queue
,
3381 renames
->pairs
[1].nr
+ renames
->pairs
[2].nr
,
3383 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; i
++) {
3384 int other_side
= 3 - i
;
3385 compute_collisions(&collisions
[i
],
3386 &renames
->dir_renames
[other_side
],
3387 &renames
->pairs
[i
]);
3389 clean
&= collect_renames(opt
, &combined
, MERGE_SIDE1
,
3391 &renames
->dir_renames
[2],
3392 &renames
->dir_renames
[1]);
3393 clean
&= collect_renames(opt
, &combined
, MERGE_SIDE2
,
3395 &renames
->dir_renames
[1],
3396 &renames
->dir_renames
[2]);
3397 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; i
++)
3398 free_collisions(&collisions
[i
]);
3399 STABLE_QSORT(combined
.queue
, combined
.nr
, compare_pairs
);
3400 trace2_region_leave("merge", "directory renames", opt
->repo
);
3402 trace2_region_enter("merge", "process renames", opt
->repo
);
3403 clean
&= process_renames(opt
, &combined
);
3404 trace2_region_leave("merge", "process renames", opt
->repo
);
3406 goto simple_cleanup
; /* collect_renames() handles some of cleanup */
3410 * Free now unneeded filepairs, which would have been handled
3411 * in collect_renames() normally but we skipped that code.
3413 for (s
= MERGE_SIDE1
; s
<= MERGE_SIDE2
; s
++) {
3414 struct diff_queue_struct
*side_pairs
;
3417 side_pairs
= &renames
->pairs
[s
];
3418 for (i
= 0; i
< side_pairs
->nr
; ++i
) {
3419 struct diff_filepair
*p
= side_pairs
->queue
[i
];
3420 pool_diff_free_filepair(&opt
->priv
->pool
, p
);
3425 /* Free memory for renames->pairs[] and combined */
3426 for (s
= MERGE_SIDE1
; s
<= MERGE_SIDE2
; s
++) {
3427 free(renames
->pairs
[s
].queue
);
3428 DIFF_QUEUE_CLEAR(&renames
->pairs
[s
]);
3430 for (i
= 0; i
< combined
.nr
; i
++)
3431 pool_diff_free_filepair(&opt
->priv
->pool
, combined
.queue
[i
]);
3432 free(combined
.queue
);
3437 /*** Function Grouping: functions related to process_entries() ***/
3439 static int sort_dirs_next_to_their_children(const char *one
, const char *two
)
3441 unsigned char c1
, c2
;
3444 * Here we only care that entries for directories appear adjacent
3445 * to and before files underneath the directory. We can achieve
3446 * that by pretending to add a trailing slash to every file and
3447 * then sorting. In other words, we do not want the natural
3452 * Instead, we want "foo" to sort as though it were "foo/", so that
3457 * To achieve this, we basically implement our own strcmp, except that
3458 * if we get to the end of either string instead of comparing NUL to
3459 * another character, we compare '/' to it.
3461 * If this unusual "sort as though '/' were appended" perplexes
3462 * you, perhaps it will help to note that this is not the final
3463 * sort. write_tree() will sort again without the trailing slash
3464 * magic, but just on paths immediately under a given tree.
3466 * The reason to not use df_name_compare directly was that it was
3467 * just too expensive (we don't have the string lengths handy), so
3468 * it was reimplemented.
3472 * NOTE: This function will never be called with two equal strings,
3473 * because it is used to sort the keys of a strmap, and strmaps have
3474 * unique keys by construction. That simplifies our c1==c2 handling
3478 while (*one
&& (*one
== *two
)) {
3483 c1
= *one
? *one
: '/';
3484 c2
= *two
? *two
: '/';
3487 /* Getting here means one is a leading directory of the other */
3488 return (*one
) ? 1 : -1;
3493 static int read_oid_strbuf(const struct object_id
*oid
,
3497 enum object_type type
;
3499 buf
= repo_read_object_file(the_repository
, oid
, &type
, &size
);
3501 return error(_("cannot read object %s"), oid_to_hex(oid
));
3502 if (type
!= OBJ_BLOB
) {
3504 return error(_("object %s is not a blob"), oid_to_hex(oid
));
3506 strbuf_attach(dst
, buf
, size
, size
+ 1);
3510 static int blob_unchanged(struct merge_options
*opt
,
3511 const struct version_info
*base
,
3512 const struct version_info
*side
,
3515 struct strbuf basebuf
= STRBUF_INIT
;
3516 struct strbuf sidebuf
= STRBUF_INIT
;
3517 int ret
= 0; /* assume changed for safety */
3518 struct index_state
*idx
= &opt
->priv
->attr_index
;
3520 if (!idx
->initialized
)
3521 initialize_attr_index(opt
);
3523 if (base
->mode
!= side
->mode
)
3525 if (oideq(&base
->oid
, &side
->oid
))
3528 if (read_oid_strbuf(&base
->oid
, &basebuf
) ||
3529 read_oid_strbuf(&side
->oid
, &sidebuf
))
3532 * Note: binary | is used so that both renormalizations are
3533 * performed. Comparison can be skipped if both files are
3534 * unchanged since their sha1s have already been compared.
3536 if (renormalize_buffer(idx
, path
, basebuf
.buf
, basebuf
.len
, &basebuf
) |
3537 renormalize_buffer(idx
, path
, sidebuf
.buf
, sidebuf
.len
, &sidebuf
))
3538 ret
= (basebuf
.len
== sidebuf
.len
&&
3539 !memcmp(basebuf
.buf
, sidebuf
.buf
, basebuf
.len
));
3542 strbuf_release(&basebuf
);
3543 strbuf_release(&sidebuf
);
3547 struct directory_versions
{
3549 * versions: list of (basename -> version_info)
3551 * The basenames are in reverse lexicographic order of full pathnames,
3552 * as processed in process_entries(). This puts all entries within
3553 * a directory together, and covers the directory itself after
3554 * everything within it, allowing us to write subtrees before needing
3555 * to record information for the tree itself.
3557 struct string_list versions
;
3560 * offsets: list of (full relative path directories -> integer offsets)
3562 * Since versions contains basenames from files in multiple different
3563 * directories, we need to know which entries in versions correspond
3564 * to which directories. Values of e.g.
3568 * Would mean that entries 0-1 of versions are files in the toplevel
3569 * directory, entries 2-4 are files under src/, and the remaining
3570 * entries starting at index 5 are files under src/moduleA/.
3572 struct string_list offsets
;
3575 * last_directory: directory that previously processed file found in
3577 * last_directory starts NULL, but records the directory in which the
3578 * previous file was found within. As soon as
3579 * directory(current_file) != last_directory
3580 * then we need to start updating accounting in versions & offsets.
3581 * Note that last_directory is always the last path in "offsets" (or
3582 * NULL if "offsets" is empty) so this exists just for quick access.
3584 const char *last_directory
;
3586 /* last_directory_len: cached computation of strlen(last_directory) */
3587 unsigned last_directory_len
;
3590 static int tree_entry_order(const void *a_
, const void *b_
)
3592 const struct string_list_item
*a
= a_
;
3593 const struct string_list_item
*b
= b_
;
3595 const struct merged_info
*ami
= a
->util
;
3596 const struct merged_info
*bmi
= b
->util
;
3597 return base_name_compare(a
->string
, strlen(a
->string
), ami
->result
.mode
,
3598 b
->string
, strlen(b
->string
), bmi
->result
.mode
);
3601 static int write_tree(struct object_id
*result_oid
,
3602 struct string_list
*versions
,
3603 unsigned int offset
,
3606 size_t maxlen
= 0, extra
;
3608 struct strbuf buf
= STRBUF_INIT
;
3611 assert(offset
<= versions
->nr
);
3612 nr
= versions
->nr
- offset
;
3614 /* No need for STABLE_QSORT -- filenames must be unique */
3615 QSORT(versions
->items
+ offset
, nr
, tree_entry_order
);
3617 /* Pre-allocate some space in buf */
3618 extra
= hash_size
+ 8; /* 8: 6 for mode, 1 for space, 1 for NUL char */
3619 for (i
= 0; i
< nr
; i
++) {
3620 maxlen
+= strlen(versions
->items
[offset
+i
].string
) + extra
;
3622 strbuf_grow(&buf
, maxlen
);
3624 /* Write each entry out to buf */
3625 for (i
= 0; i
< nr
; i
++) {
3626 struct merged_info
*mi
= versions
->items
[offset
+i
].util
;
3627 struct version_info
*ri
= &mi
->result
;
3628 strbuf_addf(&buf
, "%o %s%c",
3630 versions
->items
[offset
+i
].string
, '\0');
3631 strbuf_add(&buf
, ri
->oid
.hash
, hash_size
);
3634 /* Write this object file out, and record in result_oid */
3635 if (write_object_file(buf
.buf
, buf
.len
, OBJ_TREE
, result_oid
))
3637 strbuf_release(&buf
);
3641 static void record_entry_for_tree(struct directory_versions
*dir_metadata
,
3643 struct merged_info
*mi
)
3645 const char *basename
;
3648 /* nothing to record */
3651 basename
= path
+ mi
->basename_offset
;
3652 assert(strchr(basename
, '/') == NULL
);
3653 string_list_append(&dir_metadata
->versions
,
3654 basename
)->util
= &mi
->result
;
3657 static int write_completed_directory(struct merge_options
*opt
,
3658 const char *new_directory_name
,
3659 struct directory_versions
*info
)
3661 const char *prev_dir
;
3662 struct merged_info
*dir_info
= NULL
;
3663 unsigned int offset
, ret
= 0;
3666 * Some explanation of info->versions and info->offsets...
3668 * process_entries() iterates over all relevant files AND
3669 * directories in reverse lexicographic order, and calls this
3670 * function. Thus, an example of the paths that process_entries()
3671 * could operate on (along with the directories for those paths
3676 * src/moduleB/umm.c src/moduleB
3677 * src/moduleB/stuff.h src/moduleB
3678 * src/moduleB/baz.c src/moduleB
3680 * src/moduleA/foo.c src/moduleA
3681 * src/moduleA/bar.c src/moduleA
3688 * always contains the unprocessed entries and their
3689 * version_info information. For example, after the first five
3690 * entries above, info->versions would be:
3692 * xtract.c <xtract.c's version_info>
3693 * token.txt <token.txt's version_info>
3694 * umm.c <src/moduleB/umm.c's version_info>
3695 * stuff.h <src/moduleB/stuff.h's version_info>
3696 * baz.c <src/moduleB/baz.c's version_info>
3698 * Once a subdirectory is completed we remove the entries in
3699 * that subdirectory from info->versions, writing it as a tree
3700 * (write_tree()). Thus, as soon as we get to src/moduleB,
3701 * info->versions would be updated to
3703 * xtract.c <xtract.c's version_info>
3704 * token.txt <token.txt's version_info>
3705 * moduleB <src/moduleB's version_info>
3709 * helps us track which entries in info->versions correspond to
3710 * which directories. When we are N directories deep (e.g. 4
3711 * for src/modA/submod/subdir/), we have up to N+1 unprocessed
3712 * directories (+1 because of toplevel dir). Corresponding to
3713 * the info->versions example above, after processing five entries
3714 * info->offsets will be:
3719 * which is used to know that xtract.c & token.txt are from the
3720 * toplevel dirctory, while umm.c & stuff.h & baz.c are from the
3721 * src/moduleB directory. Again, following the example above,
3722 * once we need to process src/moduleB, then info->offsets is
3728 * which says that moduleB (and only moduleB so far) is in the
3731 * One unique thing to note about info->offsets here is that
3732 * "src" was not added to info->offsets until there was a path
3733 * (a file OR directory) immediately below src/ that got
3736 * Since process_entry() just appends new entries to info->versions,
3737 * write_completed_directory() only needs to do work if the next path
3738 * is in a directory that is different than the last directory found
3743 * If we are working with the same directory as the last entry, there
3744 * is no work to do. (See comments above the directory_name member of
3745 * struct merged_info for why we can use pointer comparison instead of
3748 if (new_directory_name
== info
->last_directory
)
3752 * If we are just starting (last_directory is NULL), or last_directory
3753 * is a prefix of the current directory, then we can just update
3754 * info->offsets to record the offset where we started this directory
3755 * and update last_directory to have quick access to it.
3757 if (info
->last_directory
== NULL
||
3758 !strncmp(new_directory_name
, info
->last_directory
,
3759 info
->last_directory_len
)) {
3760 uintptr_t offset
= info
->versions
.nr
;
3762 info
->last_directory
= new_directory_name
;
3763 info
->last_directory_len
= strlen(info
->last_directory
);
3765 * Record the offset into info->versions where we will
3766 * start recording basenames of paths found within
3767 * new_directory_name.
3769 string_list_append(&info
->offsets
,
3770 info
->last_directory
)->util
= (void*)offset
;
3775 * The next entry that will be processed will be within
3776 * new_directory_name. Since at this point we know that
3777 * new_directory_name is within a different directory than
3778 * info->last_directory, we have all entries for info->last_directory
3779 * in info->versions and we need to create a tree object for them.
3781 dir_info
= strmap_get(&opt
->priv
->paths
, info
->last_directory
);
3783 offset
= (uintptr_t)info
->offsets
.items
[info
->offsets
.nr
-1].util
;
3784 if (offset
== info
->versions
.nr
) {
3786 * Actually, we don't need to create a tree object in this
3787 * case. Whenever all files within a directory disappear
3788 * during the merge (e.g. unmodified on one side and
3789 * deleted on the other, or files were renamed elsewhere),
3790 * then we get here and the directory itself needs to be
3791 * omitted from its parent tree as well.
3793 dir_info
->is_null
= 1;
3796 * Write out the tree to the git object directory, and also
3797 * record the mode and oid in dir_info->result.
3799 dir_info
->is_null
= 0;
3800 dir_info
->result
.mode
= S_IFDIR
;
3801 if (write_tree(&dir_info
->result
.oid
, &info
->versions
, offset
,
3802 opt
->repo
->hash_algo
->rawsz
) < 0)
3807 * We've now used several entries from info->versions and one entry
3808 * from info->offsets, so we get rid of those values.
3811 info
->versions
.nr
= offset
;
3814 * Now we've taken care of the completed directory, but we need to
3815 * prepare things since future entries will be in
3816 * new_directory_name. (In particular, process_entry() will be
3817 * appending new entries to info->versions.) So, we need to make
3818 * sure new_directory_name is the last entry in info->offsets.
3820 prev_dir
= info
->offsets
.nr
== 0 ? NULL
:
3821 info
->offsets
.items
[info
->offsets
.nr
-1].string
;
3822 if (new_directory_name
!= prev_dir
) {
3823 uintptr_t c
= info
->versions
.nr
;
3824 string_list_append(&info
->offsets
,
3825 new_directory_name
)->util
= (void*)c
;
3828 /* And, of course, we need to update last_directory to match. */
3829 info
->last_directory
= new_directory_name
;
3830 info
->last_directory_len
= strlen(info
->last_directory
);
3835 /* Per entry merge function */
3836 static int process_entry(struct merge_options
*opt
,
3838 struct conflict_info
*ci
,
3839 struct directory_versions
*dir_metadata
)
3841 int df_file_index
= 0;
3844 assert(ci
->filemask
>= 0 && ci
->filemask
<= 7);
3845 /* ci->match_mask == 7 was handled in collect_merge_info_callback() */
3846 assert(ci
->match_mask
== 0 || ci
->match_mask
== 3 ||
3847 ci
->match_mask
== 5 || ci
->match_mask
== 6);
3850 record_entry_for_tree(dir_metadata
, path
, &ci
->merged
);
3851 if (ci
->filemask
== 0)
3852 /* nothing else to handle */
3854 assert(ci
->df_conflict
);
3857 if (ci
->df_conflict
&& ci
->merged
.result
.mode
== 0) {
3861 * directory no longer in the way, but we do have a file we
3862 * need to place here so we need to clean away the "directory
3863 * merges to nothing" result.
3865 ci
->df_conflict
= 0;
3866 assert(ci
->filemask
!= 0);
3867 ci
->merged
.clean
= 0;
3868 ci
->merged
.is_null
= 0;
3869 /* and we want to zero out any directory-related entries */
3870 ci
->match_mask
= (ci
->match_mask
& ~ci
->dirmask
);
3872 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
3873 if (ci
->filemask
& (1 << i
))
3875 ci
->stages
[i
].mode
= 0;
3876 oidcpy(&ci
->stages
[i
].oid
, null_oid());
3878 } else if (ci
->df_conflict
&& ci
->merged
.result
.mode
!= 0) {
3880 * This started out as a D/F conflict, and the entries in
3881 * the competing directory were not removed by the merge as
3882 * evidenced by write_completed_directory() writing a value
3883 * to ci->merged.result.mode.
3885 struct conflict_info
*new_ci
;
3887 const char *old_path
= path
;
3890 assert(ci
->merged
.result
.mode
== S_IFDIR
);
3893 * If filemask is 1, we can just ignore the file as having
3894 * been deleted on both sides. We do not want to overwrite
3895 * ci->merged.result, since it stores the tree for all the
3898 if (ci
->filemask
== 1) {
3904 * This file still exists on at least one side, and we want
3905 * the directory to remain here, so we need to move this
3906 * path to some new location.
3908 new_ci
= mem_pool_calloc(&opt
->priv
->pool
, 1, sizeof(*new_ci
));
3910 /* We don't really want new_ci->merged.result copied, but it'll
3911 * be overwritten below so it doesn't matter. We also don't
3912 * want any directory mode/oid values copied, but we'll zero
3913 * those out immediately. We do want the rest of ci copied.
3915 memcpy(new_ci
, ci
, sizeof(*ci
));
3916 new_ci
->match_mask
= (new_ci
->match_mask
& ~new_ci
->dirmask
);
3917 new_ci
->dirmask
= 0;
3918 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
3919 if (new_ci
->filemask
& (1 << i
))
3921 /* zero out any entries related to directories */
3922 new_ci
->stages
[i
].mode
= 0;
3923 oidcpy(&new_ci
->stages
[i
].oid
, null_oid());
3927 * Find out which side this file came from; note that we
3928 * cannot just use ci->filemask, because renames could cause
3929 * the filemask to go back to 7. So we use dirmask, then
3930 * pick the opposite side's index.
3932 df_file_index
= (ci
->dirmask
& (1 << 1)) ? 2 : 1;
3933 branch
= (df_file_index
== 1) ? opt
->branch1
: opt
->branch2
;
3934 path
= unique_path(opt
, path
, branch
);
3935 strmap_put(&opt
->priv
->paths
, path
, new_ci
);
3937 path_msg(opt
, CONFLICT_FILE_DIRECTORY
, 0,
3938 path
, old_path
, NULL
, NULL
,
3939 _("CONFLICT (file/directory): directory in the way "
3940 "of %s from %s; moving it to %s instead."),
3941 old_path
, branch
, path
);
3944 * Zero out the filemask for the old ci. At this point, ci
3945 * was just an entry for a directory, so we don't need to
3946 * do anything more with it.
3951 * Now note that we're working on the new entry (path was
3958 * NOTE: Below there is a long switch-like if-elseif-elseif... block
3959 * which the code goes through even for the df_conflict cases
3962 if (ci
->match_mask
) {
3963 ci
->merged
.clean
= !ci
->df_conflict
&& !ci
->path_conflict
;
3964 if (ci
->match_mask
== 6) {
3965 /* stages[1] == stages[2] */
3966 ci
->merged
.result
.mode
= ci
->stages
[1].mode
;
3967 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[1].oid
);
3969 /* determine the mask of the side that didn't match */
3970 unsigned int othermask
= 7 & ~ci
->match_mask
;
3971 int side
= (othermask
== 4) ? 2 : 1;
3973 ci
->merged
.result
.mode
= ci
->stages
[side
].mode
;
3974 ci
->merged
.is_null
= !ci
->merged
.result
.mode
;
3975 if (ci
->merged
.is_null
)
3976 ci
->merged
.clean
= 1;
3977 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[side
].oid
);
3979 assert(othermask
== 2 || othermask
== 4);
3980 assert(ci
->merged
.is_null
==
3981 (ci
->filemask
== ci
->match_mask
));
3983 } else if (ci
->filemask
>= 6 &&
3984 (S_IFMT
& ci
->stages
[1].mode
) !=
3985 (S_IFMT
& ci
->stages
[2].mode
)) {
3986 /* Two different items from (file/submodule/symlink) */
3987 if (opt
->priv
->call_depth
) {
3988 /* Just use the version from the merge base */
3989 ci
->merged
.clean
= 0;
3990 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[0].oid
);
3991 ci
->merged
.result
.mode
= ci
->stages
[0].mode
;
3992 ci
->merged
.is_null
= (ci
->merged
.result
.mode
== 0);
3994 /* Handle by renaming one or both to separate paths. */
3995 unsigned o_mode
= ci
->stages
[0].mode
;
3996 unsigned a_mode
= ci
->stages
[1].mode
;
3997 unsigned b_mode
= ci
->stages
[2].mode
;
3998 struct conflict_info
*new_ci
;
3999 const char *a_path
= NULL
, *b_path
= NULL
;
4000 int rename_a
= 0, rename_b
= 0;
4002 new_ci
= mem_pool_alloc(&opt
->priv
->pool
,
4005 if (S_ISREG(a_mode
))
4007 else if (S_ISREG(b_mode
))
4015 a_path
= unique_path(opt
, path
, opt
->branch1
);
4017 b_path
= unique_path(opt
, path
, opt
->branch2
);
4019 if (rename_a
&& rename_b
) {
4020 path_msg(opt
, CONFLICT_DISTINCT_MODES
, 0,
4021 path
, a_path
, b_path
, NULL
,
4022 _("CONFLICT (distinct types): %s had "
4023 "different types on each side; "
4024 "renamed both of them so each can "
4025 "be recorded somewhere."),
4028 path_msg(opt
, CONFLICT_DISTINCT_MODES
, 0,
4029 path
, rename_a
? a_path
: b_path
,
4031 _("CONFLICT (distinct types): %s had "
4032 "different types on each side; "
4033 "renamed one of them so each can be "
4034 "recorded somewhere."),
4038 ci
->merged
.clean
= 0;
4039 memcpy(new_ci
, ci
, sizeof(*new_ci
));
4041 /* Put b into new_ci, removing a from stages */
4042 new_ci
->merged
.result
.mode
= ci
->stages
[2].mode
;
4043 oidcpy(&new_ci
->merged
.result
.oid
, &ci
->stages
[2].oid
);
4044 new_ci
->stages
[1].mode
= 0;
4045 oidcpy(&new_ci
->stages
[1].oid
, null_oid());
4046 new_ci
->filemask
= 5;
4047 if ((S_IFMT
& b_mode
) != (S_IFMT
& o_mode
)) {
4048 new_ci
->stages
[0].mode
= 0;
4049 oidcpy(&new_ci
->stages
[0].oid
, null_oid());
4050 new_ci
->filemask
= 4;
4053 /* Leave only a in ci, fixing stages. */
4054 ci
->merged
.result
.mode
= ci
->stages
[1].mode
;
4055 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[1].oid
);
4056 ci
->stages
[2].mode
= 0;
4057 oidcpy(&ci
->stages
[2].oid
, null_oid());
4059 if ((S_IFMT
& a_mode
) != (S_IFMT
& o_mode
)) {
4060 ci
->stages
[0].mode
= 0;
4061 oidcpy(&ci
->stages
[0].oid
, null_oid());
4065 /* Insert entries into opt->priv_paths */
4066 assert(rename_a
|| rename_b
);
4068 strmap_put(&opt
->priv
->paths
, a_path
, ci
);
4072 strmap_put(&opt
->priv
->paths
, b_path
, new_ci
);
4074 if (rename_a
&& rename_b
)
4075 strmap_remove(&opt
->priv
->paths
, path
, 0);
4078 * Do special handling for b_path since process_entry()
4079 * won't be called on it specially.
4081 strmap_put(&opt
->priv
->conflicted
, b_path
, new_ci
);
4082 record_entry_for_tree(dir_metadata
, b_path
,
4086 * Remaining code for processing this entry should
4087 * think in terms of processing a_path.
4092 } else if (ci
->filemask
>= 6) {
4093 /* Need a two-way or three-way content merge */
4094 struct version_info merged_file
;
4096 struct version_info
*o
= &ci
->stages
[0];
4097 struct version_info
*a
= &ci
->stages
[1];
4098 struct version_info
*b
= &ci
->stages
[2];
4100 clean_merge
= handle_content_merge(opt
, path
, o
, a
, b
,
4102 opt
->priv
->call_depth
* 2,
4104 if (clean_merge
< 0)
4106 ci
->merged
.clean
= clean_merge
&&
4107 !ci
->df_conflict
&& !ci
->path_conflict
;
4108 ci
->merged
.result
.mode
= merged_file
.mode
;
4109 ci
->merged
.is_null
= (merged_file
.mode
== 0);
4110 oidcpy(&ci
->merged
.result
.oid
, &merged_file
.oid
);
4111 if (clean_merge
&& ci
->df_conflict
) {
4112 assert(df_file_index
== 1 || df_file_index
== 2);
4113 ci
->filemask
= 1 << df_file_index
;
4114 ci
->stages
[df_file_index
].mode
= merged_file
.mode
;
4115 oidcpy(&ci
->stages
[df_file_index
].oid
, &merged_file
.oid
);
4118 const char *reason
= _("content");
4119 if (ci
->filemask
== 6)
4120 reason
= _("add/add");
4121 if (S_ISGITLINK(merged_file
.mode
))
4122 reason
= _("submodule");
4123 path_msg(opt
, CONFLICT_CONTENTS
, 0,
4124 path
, NULL
, NULL
, NULL
,
4125 _("CONFLICT (%s): Merge conflict in %s"),
4128 } else if (ci
->filemask
== 3 || ci
->filemask
== 5) {
4130 const char *modify_branch
, *delete_branch
;
4131 int side
= (ci
->filemask
== 5) ? 2 : 1;
4132 int index
= opt
->priv
->call_depth
? 0 : side
;
4134 ci
->merged
.result
.mode
= ci
->stages
[index
].mode
;
4135 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[index
].oid
);
4136 ci
->merged
.clean
= 0;
4138 modify_branch
= (side
== 1) ? opt
->branch1
: opt
->branch2
;
4139 delete_branch
= (side
== 1) ? opt
->branch2
: opt
->branch1
;
4141 if (opt
->renormalize
&&
4142 blob_unchanged(opt
, &ci
->stages
[0], &ci
->stages
[side
],
4144 if (!ci
->path_conflict
) {
4146 * Blob unchanged after renormalization, so
4147 * there's no modify/delete conflict after all;
4148 * we can just remove the file.
4150 ci
->merged
.is_null
= 1;
4151 ci
->merged
.clean
= 1;
4153 * file goes away => even if there was a
4154 * directory/file conflict there isn't one now.
4156 ci
->df_conflict
= 0;
4158 /* rename/delete, so conflict remains */
4160 } else if (ci
->path_conflict
&&
4161 oideq(&ci
->stages
[0].oid
, &ci
->stages
[side
].oid
)) {
4163 * This came from a rename/delete; no action to take,
4164 * but avoid printing "modify/delete" conflict notice
4165 * since the contents were not modified.
4168 path_msg(opt
, CONFLICT_MODIFY_DELETE
, 0,
4169 path
, NULL
, NULL
, NULL
,
4170 _("CONFLICT (modify/delete): %s deleted in %s "
4171 "and modified in %s. Version %s of %s left "
4173 path
, delete_branch
, modify_branch
,
4174 modify_branch
, path
);
4176 } else if (ci
->filemask
== 2 || ci
->filemask
== 4) {
4177 /* Added on one side */
4178 int side
= (ci
->filemask
== 4) ? 2 : 1;
4179 ci
->merged
.result
.mode
= ci
->stages
[side
].mode
;
4180 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[side
].oid
);
4181 ci
->merged
.clean
= !ci
->df_conflict
&& !ci
->path_conflict
;
4182 } else if (ci
->filemask
== 1) {
4183 /* Deleted on both sides */
4184 ci
->merged
.is_null
= 1;
4185 ci
->merged
.result
.mode
= 0;
4186 oidcpy(&ci
->merged
.result
.oid
, null_oid());
4187 assert(!ci
->df_conflict
);
4188 ci
->merged
.clean
= !ci
->path_conflict
;
4192 * If still conflicted, record it separately. This allows us to later
4193 * iterate over just conflicted entries when updating the index instead
4194 * of iterating over all entries.
4196 if (!ci
->merged
.clean
)
4197 strmap_put(&opt
->priv
->conflicted
, path
, ci
);
4199 /* Record metadata for ci->merged in dir_metadata */
4200 record_entry_for_tree(dir_metadata
, path
, &ci
->merged
);
4204 static void prefetch_for_content_merges(struct merge_options
*opt
,
4205 struct string_list
*plist
)
4207 struct string_list_item
*e
;
4208 struct oid_array to_fetch
= OID_ARRAY_INIT
;
4210 if (opt
->repo
!= the_repository
|| !repo_has_promisor_remote(the_repository
))
4213 for (e
= &plist
->items
[plist
->nr
-1]; e
>= plist
->items
; --e
) {
4214 /* char *path = e->string; */
4215 struct conflict_info
*ci
= e
->util
;
4218 /* Ignore clean entries */
4219 if (ci
->merged
.clean
)
4222 /* Ignore entries that don't need a content merge */
4223 if (ci
->match_mask
|| ci
->filemask
< 6 ||
4224 !S_ISREG(ci
->stages
[1].mode
) ||
4225 !S_ISREG(ci
->stages
[2].mode
) ||
4226 oideq(&ci
->stages
[1].oid
, &ci
->stages
[2].oid
))
4229 /* Also don't need content merge if base matches either side */
4230 if (ci
->filemask
== 7 &&
4231 S_ISREG(ci
->stages
[0].mode
) &&
4232 (oideq(&ci
->stages
[0].oid
, &ci
->stages
[1].oid
) ||
4233 oideq(&ci
->stages
[0].oid
, &ci
->stages
[2].oid
)))
4236 for (i
= 0; i
< 3; i
++) {
4237 unsigned side_mask
= (1 << i
);
4238 struct version_info
*vi
= &ci
->stages
[i
];
4240 if ((ci
->filemask
& side_mask
) &&
4241 S_ISREG(vi
->mode
) &&
4242 oid_object_info_extended(opt
->repo
, &vi
->oid
, NULL
,
4243 OBJECT_INFO_FOR_PREFETCH
))
4244 oid_array_append(&to_fetch
, &vi
->oid
);
4248 promisor_remote_get_direct(opt
->repo
, to_fetch
.oid
, to_fetch
.nr
);
4249 oid_array_clear(&to_fetch
);
4252 static int process_entries(struct merge_options
*opt
,
4253 struct object_id
*result_oid
)
4255 struct hashmap_iter iter
;
4256 struct strmap_entry
*e
;
4257 struct string_list plist
= STRING_LIST_INIT_NODUP
;
4258 struct string_list_item
*entry
;
4259 struct directory_versions dir_metadata
= { STRING_LIST_INIT_NODUP
,
4260 STRING_LIST_INIT_NODUP
,
4264 trace2_region_enter("merge", "process_entries setup", opt
->repo
);
4265 if (strmap_empty(&opt
->priv
->paths
)) {
4266 oidcpy(result_oid
, opt
->repo
->hash_algo
->empty_tree
);
4270 /* Hack to pre-allocate plist to the desired size */
4271 trace2_region_enter("merge", "plist grow", opt
->repo
);
4272 ALLOC_GROW(plist
.items
, strmap_get_size(&opt
->priv
->paths
), plist
.alloc
);
4273 trace2_region_leave("merge", "plist grow", opt
->repo
);
4275 /* Put every entry from paths into plist, then sort */
4276 trace2_region_enter("merge", "plist copy", opt
->repo
);
4277 strmap_for_each_entry(&opt
->priv
->paths
, &iter
, e
) {
4278 string_list_append(&plist
, e
->key
)->util
= e
->value
;
4280 trace2_region_leave("merge", "plist copy", opt
->repo
);
4282 trace2_region_enter("merge", "plist special sort", opt
->repo
);
4283 plist
.cmp
= sort_dirs_next_to_their_children
;
4284 string_list_sort(&plist
);
4285 trace2_region_leave("merge", "plist special sort", opt
->repo
);
4287 trace2_region_leave("merge", "process_entries setup", opt
->repo
);
4290 * Iterate over the items in reverse order, so we can handle paths
4291 * below a directory before needing to handle the directory itself.
4293 * This allows us to write subtrees before we need to write trees,
4294 * and it also enables sane handling of directory/file conflicts
4295 * (because it allows us to know whether the directory is still in
4296 * the way when it is time to process the file at the same path).
4298 trace2_region_enter("merge", "processing", opt
->repo
);
4299 prefetch_for_content_merges(opt
, &plist
);
4300 for (entry
= &plist
.items
[plist
.nr
-1]; entry
>= plist
.items
; --entry
) {
4301 char *path
= entry
->string
;
4303 * NOTE: mi may actually be a pointer to a conflict_info, but
4304 * we have to check mi->clean first to see if it's safe to
4305 * reassign to such a pointer type.
4307 struct merged_info
*mi
= entry
->util
;
4309 if (write_completed_directory(opt
, mi
->directory_name
,
4310 &dir_metadata
) < 0) {
4315 record_entry_for_tree(&dir_metadata
, path
, mi
);
4317 struct conflict_info
*ci
= (struct conflict_info
*)mi
;
4318 if (process_entry(opt
, path
, ci
, &dir_metadata
) < 0) {
4324 trace2_region_leave("merge", "processing", opt
->repo
);
4326 trace2_region_enter("merge", "process_entries cleanup", opt
->repo
);
4327 if (dir_metadata
.offsets
.nr
!= 1 ||
4328 (uintptr_t)dir_metadata
.offsets
.items
[0].util
!= 0) {
4329 printf("dir_metadata.offsets.nr = %"PRIuMAX
" (should be 1)\n",
4330 (uintmax_t)dir_metadata
.offsets
.nr
);
4331 printf("dir_metadata.offsets.items[0].util = %u (should be 0)\n",
4332 (unsigned)(uintptr_t)dir_metadata
.offsets
.items
[0].util
);
4334 BUG("dir_metadata accounting completely off; shouldn't happen");
4336 if (write_tree(result_oid
, &dir_metadata
.versions
, 0,
4337 opt
->repo
->hash_algo
->rawsz
) < 0)
4340 string_list_clear(&plist
, 0);
4341 string_list_clear(&dir_metadata
.versions
, 0);
4342 string_list_clear(&dir_metadata
.offsets
, 0);
4343 trace2_region_leave("merge", "process_entries cleanup", opt
->repo
);
4348 /*** Function Grouping: functions related to merge_switch_to_result() ***/
4350 static int checkout(struct merge_options
*opt
,
4354 /* Switch the index/working copy from old to new */
4356 struct tree_desc trees
[2];
4357 struct unpack_trees_options unpack_opts
;
4359 memset(&unpack_opts
, 0, sizeof(unpack_opts
));
4360 unpack_opts
.head_idx
= -1;
4361 unpack_opts
.src_index
= opt
->repo
->index
;
4362 unpack_opts
.dst_index
= opt
->repo
->index
;
4364 setup_unpack_trees_porcelain(&unpack_opts
, "merge");
4367 * NOTE: if this were just "git checkout" code, we would probably
4368 * read or refresh the cache and check for a conflicted index, but
4369 * builtin/merge.c or sequencer.c really needs to read the index
4370 * and check for conflicted entries before starting merging for a
4371 * good user experience (no sense waiting for merges/rebases before
4372 * erroring out), so there's no reason to duplicate that work here.
4375 /* 2-way merge to the new branch */
4376 unpack_opts
.update
= 1;
4377 unpack_opts
.merge
= 1;
4378 unpack_opts
.quiet
= 0; /* FIXME: sequencer might want quiet? */
4379 unpack_opts
.verbose_update
= (opt
->verbosity
> 2);
4380 unpack_opts
.fn
= twoway_merge
;
4381 unpack_opts
.preserve_ignored
= 0; /* FIXME: !opts->overwrite_ignore */
4382 if (parse_tree(prev
) < 0)
4384 init_tree_desc(&trees
[0], prev
->buffer
, prev
->size
);
4385 if (parse_tree(next
) < 0)
4387 init_tree_desc(&trees
[1], next
->buffer
, next
->size
);
4389 ret
= unpack_trees(2, trees
, &unpack_opts
);
4390 clear_unpack_trees_porcelain(&unpack_opts
);
4394 static int record_conflicted_index_entries(struct merge_options
*opt
)
4396 struct hashmap_iter iter
;
4397 struct strmap_entry
*e
;
4398 struct index_state
*index
= opt
->repo
->index
;
4399 struct checkout state
= CHECKOUT_INIT
;
4401 int original_cache_nr
;
4403 if (strmap_empty(&opt
->priv
->conflicted
))
4407 * We are in a conflicted state. These conflicts might be inside
4408 * sparse-directory entries, so check if any entries are outside
4409 * of the sparse-checkout cone preemptively.
4411 * We set original_cache_nr below, but that might change if
4412 * index_name_pos() calls ask for paths within sparse directories.
4414 strmap_for_each_entry(&opt
->priv
->conflicted
, &iter
, e
) {
4415 if (!path_in_sparse_checkout(e
->key
, index
)) {
4416 ensure_full_index(index
);
4421 /* If any entries have skip_worktree set, we'll have to check 'em out */
4424 state
.refresh_cache
= 1;
4425 state
.istate
= index
;
4426 original_cache_nr
= index
->cache_nr
;
4428 /* Append every entry from conflicted into index, then sort */
4429 strmap_for_each_entry(&opt
->priv
->conflicted
, &iter
, e
) {
4430 const char *path
= e
->key
;
4431 struct conflict_info
*ci
= e
->value
;
4433 struct cache_entry
*ce
;
4439 * The index will already have a stage=0 entry for this path,
4440 * because we created an as-merged-as-possible version of the
4441 * file and checkout() moved the working copy and index over
4444 * However, previous iterations through this loop will have
4445 * added unstaged entries to the end of the cache which
4446 * ignore the standard alphabetical ordering of cache
4447 * entries and break invariants needed for index_name_pos()
4448 * to work. However, we know the entry we want is before
4449 * those appended cache entries, so do a temporary swap on
4450 * cache_nr to only look through entries of interest.
4452 SWAP(index
->cache_nr
, original_cache_nr
);
4453 pos
= index_name_pos(index
, path
, strlen(path
));
4454 SWAP(index
->cache_nr
, original_cache_nr
);
4456 if (ci
->filemask
!= 1)
4457 BUG("Conflicted %s but nothing in basic working tree or index; this shouldn't happen", path
);
4458 cache_tree_invalidate_path(index
, path
);
4460 ce
= index
->cache
[pos
];
4463 * Clean paths with CE_SKIP_WORKTREE set will not be
4464 * written to the working tree by the unpack_trees()
4465 * call in checkout(). Our conflicted entries would
4466 * have appeared clean to that code since we ignored
4467 * the higher order stages. Thus, we need override
4468 * the CE_SKIP_WORKTREE bit and manually write those
4469 * files to the working disk here.
4471 if (ce_skip_worktree(ce
))
4472 errs
|= checkout_entry(ce
, &state
, NULL
, NULL
);
4475 * Mark this cache entry for removal and instead add
4476 * new stage>0 entries corresponding to the
4477 * conflicts. If there are many conflicted entries, we
4478 * want to avoid memmove'ing O(NM) entries by
4479 * inserting the new entries one at a time. So,
4480 * instead, we just add the new cache entries to the
4481 * end (ignoring normal index requirements on sort
4482 * order) and sort the index once we're all done.
4484 ce
->ce_flags
|= CE_REMOVE
;
4487 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
4488 struct version_info
*vi
;
4489 if (!(ci
->filemask
& (1ul << i
)))
4491 vi
= &ci
->stages
[i
];
4492 ce
= make_cache_entry(index
, vi
->mode
, &vi
->oid
,
4494 add_index_entry(index
, ce
, ADD_CACHE_JUST_APPEND
);
4499 * Remove the unused cache entries (and invalidate the relevant
4500 * cache-trees), then sort the index entries to get the conflicted
4501 * entries we added to the end into their right locations.
4503 remove_marked_cache_entries(index
, 1);
4505 * No need for STABLE_QSORT -- cmp_cache_name_compare sorts primarily
4506 * on filename and secondarily on stage, and (name, stage #) are a
4509 QSORT(index
->cache
, index
->cache_nr
, cmp_cache_name_compare
);
4514 static void print_submodule_conflict_suggestion(struct string_list
*csub
) {
4515 struct string_list_item
*item
;
4516 struct strbuf msg
= STRBUF_INIT
;
4517 struct strbuf tmp
= STRBUF_INIT
;
4518 struct strbuf subs
= STRBUF_INIT
;
4523 strbuf_add_separated_string_list(&subs
, " ", csub
);
4524 for_each_string_list_item(item
, csub
) {
4525 struct conflicted_submodule_item
*util
= item
->util
;
4528 * NEEDSWORK: The steps to resolve these errors deserve a more
4529 * detailed explanation than what is currently printed below.
4531 if (util
->flag
== CONFLICT_SUBMODULE_NOT_INITIALIZED
||
4532 util
->flag
== CONFLICT_SUBMODULE_HISTORY_NOT_AVAILABLE
)
4536 * TRANSLATORS: This is a line of advice to resolve a merge
4537 * conflict in a submodule. The first argument is the submodule
4538 * name, and the second argument is the abbreviated id of the
4539 * commit that needs to be merged. For example:
4540 * - go to submodule (mysubmodule), and either merge commit abc1234"
4542 strbuf_addf(&tmp
, _(" - go to submodule (%s), and either merge commit %s\n"
4543 " or update to an existing commit which has merged those changes\n"),
4544 item
->string
, util
->abbrev
);
4548 * TRANSLATORS: This is a detailed message for resolving submodule
4549 * conflicts. The first argument is string containing one step per
4550 * submodule. The second is a space-separated list of submodule names.
4553 _("Recursive merging with submodules currently only supports trivial cases.\n"
4554 "Please manually handle the merging of each conflicted submodule.\n"
4555 "This can be accomplished with the following steps:\n"
4557 " - come back to superproject and run:\n\n"
4559 " to record the above merge or update\n"
4560 " - resolve any other conflicts in the superproject\n"
4561 " - commit the resulting index in the superproject\n"),
4564 printf("%s", msg
.buf
);
4566 strbuf_release(&subs
);
4567 strbuf_release(&tmp
);
4568 strbuf_release(&msg
);
4571 void merge_display_update_messages(struct merge_options
*opt
,
4573 struct merge_result
*result
)
4575 struct merge_options_internal
*opti
= result
->priv
;
4576 struct hashmap_iter iter
;
4577 struct strmap_entry
*e
;
4578 struct string_list olist
= STRING_LIST_INIT_NODUP
;
4580 if (opt
->record_conflict_msgs_as_headers
)
4581 BUG("Either display conflict messages or record them as headers, not both");
4583 trace2_region_enter("merge", "display messages", opt
->repo
);
4585 /* Hack to pre-allocate olist to the desired size */
4586 ALLOC_GROW(olist
.items
, strmap_get_size(&opti
->conflicts
),
4589 /* Put every entry from output into olist, then sort */
4590 strmap_for_each_entry(&opti
->conflicts
, &iter
, e
) {
4591 string_list_append(&olist
, e
->key
)->util
= e
->value
;
4593 string_list_sort(&olist
);
4595 /* Iterate over the items, printing them */
4596 for (int path_nr
= 0; path_nr
< olist
.nr
; ++path_nr
) {
4597 struct string_list
*conflicts
= olist
.items
[path_nr
].util
;
4598 for (int i
= 0; i
< conflicts
->nr
; i
++) {
4599 struct logical_conflict_info
*info
=
4600 conflicts
->items
[i
].util
;
4603 printf("%lu", (unsigned long)info
->paths
.nr
);
4605 for (int n
= 0; n
< info
->paths
.nr
; n
++) {
4606 fputs(info
->paths
.v
[n
], stdout
);
4609 fputs(type_short_descriptions
[info
->type
],
4613 puts(conflicts
->items
[i
].string
);
4618 string_list_clear(&olist
, 0);
4620 print_submodule_conflict_suggestion(&opti
->conflicted_submodules
);
4622 /* Also include needed rename limit adjustment now */
4623 diff_warn_rename_limit("merge.renamelimit",
4624 opti
->renames
.needed_limit
, 0);
4626 trace2_region_leave("merge", "display messages", opt
->repo
);
4629 void merge_get_conflicted_files(struct merge_result
*result
,
4630 struct string_list
*conflicted_files
)
4632 struct hashmap_iter iter
;
4633 struct strmap_entry
*e
;
4634 struct merge_options_internal
*opti
= result
->priv
;
4636 strmap_for_each_entry(&opti
->conflicted
, &iter
, e
) {
4637 const char *path
= e
->key
;
4638 struct conflict_info
*ci
= e
->value
;
4643 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
4644 struct stage_info
*si
;
4646 if (!(ci
->filemask
& (1ul << i
)))
4649 si
= xmalloc(sizeof(*si
));
4651 si
->mode
= ci
->stages
[i
].mode
;
4652 oidcpy(&si
->oid
, &ci
->stages
[i
].oid
);
4653 string_list_append(conflicted_files
, path
)->util
= si
;
4656 /* string_list_sort() uses a stable sort, so we're good */
4657 string_list_sort(conflicted_files
);
4660 void merge_switch_to_result(struct merge_options
*opt
,
4662 struct merge_result
*result
,
4663 int update_worktree_and_index
,
4664 int display_update_msgs
)
4666 assert(opt
->priv
== NULL
);
4667 if (result
->clean
>= 0 && update_worktree_and_index
) {
4668 const char *filename
;
4671 trace2_region_enter("merge", "checkout", opt
->repo
);
4672 if (checkout(opt
, head
, result
->tree
)) {
4673 /* failure to function */
4675 merge_finalize(opt
, result
);
4676 trace2_region_leave("merge", "checkout", opt
->repo
);
4679 trace2_region_leave("merge", "checkout", opt
->repo
);
4681 trace2_region_enter("merge", "record_conflicted", opt
->repo
);
4682 opt
->priv
= result
->priv
;
4683 if (record_conflicted_index_entries(opt
)) {
4684 /* failure to function */
4687 merge_finalize(opt
, result
);
4688 trace2_region_leave("merge", "record_conflicted",
4693 trace2_region_leave("merge", "record_conflicted", opt
->repo
);
4695 trace2_region_enter("merge", "write_auto_merge", opt
->repo
);
4696 filename
= git_path_auto_merge(opt
->repo
);
4697 fp
= xfopen(filename
, "w");
4698 fprintf(fp
, "%s\n", oid_to_hex(&result
->tree
->object
.oid
));
4700 trace2_region_leave("merge", "write_auto_merge", opt
->repo
);
4702 if (display_update_msgs
)
4703 merge_display_update_messages(opt
, /* detailed */ 0, result
);
4705 merge_finalize(opt
, result
);
4708 void merge_finalize(struct merge_options
*opt
,
4709 struct merge_result
*result
)
4711 if (opt
->renormalize
)
4712 git_attr_set_direction(GIT_ATTR_CHECKIN
);
4713 assert(opt
->priv
== NULL
);
4716 clear_or_reinit_internal_opts(result
->priv
, 0);
4717 FREE_AND_NULL(result
->priv
);
4721 /*** Function Grouping: helper functions for merge_incore_*() ***/
4723 static struct tree
*shift_tree_object(struct repository
*repo
,
4724 struct tree
*one
, struct tree
*two
,
4725 const char *subtree_shift
)
4727 struct object_id shifted
;
4729 if (!*subtree_shift
) {
4730 shift_tree(repo
, &one
->object
.oid
, &two
->object
.oid
, &shifted
, 0);
4732 shift_tree_by(repo
, &one
->object
.oid
, &two
->object
.oid
, &shifted
,
4735 if (oideq(&two
->object
.oid
, &shifted
))
4737 return lookup_tree(repo
, &shifted
);
4740 static inline void set_commit_tree(struct commit
*c
, struct tree
*t
)
4745 static struct commit
*make_virtual_commit(struct repository
*repo
,
4747 const char *comment
)
4749 struct commit
*commit
= alloc_commit_node(repo
);
4751 set_merge_remote_desc(commit
, comment
, (struct object
*)commit
);
4752 set_commit_tree(commit
, tree
);
4753 commit
->object
.parsed
= 1;
4757 static void merge_start(struct merge_options
*opt
, struct merge_result
*result
)
4759 struct rename_info
*renames
;
4761 struct mem_pool
*pool
= NULL
;
4763 /* Sanity checks on opt */
4764 trace2_region_enter("merge", "sanity checks", opt
->repo
);
4767 assert(opt
->branch1
&& opt
->branch2
);
4769 assert(opt
->detect_directory_renames
>= MERGE_DIRECTORY_RENAMES_NONE
&&
4770 opt
->detect_directory_renames
<= MERGE_DIRECTORY_RENAMES_TRUE
);
4771 assert(opt
->rename_limit
>= -1);
4772 assert(opt
->rename_score
>= 0 && opt
->rename_score
<= MAX_SCORE
);
4773 assert(opt
->show_rename_progress
>= 0 && opt
->show_rename_progress
<= 1);
4775 assert(opt
->xdl_opts
>= 0);
4776 assert(opt
->recursive_variant
>= MERGE_VARIANT_NORMAL
&&
4777 opt
->recursive_variant
<= MERGE_VARIANT_THEIRS
);
4779 if (opt
->msg_header_prefix
)
4780 assert(opt
->record_conflict_msgs_as_headers
);
4783 * detect_renames, verbosity, buffer_output, and obuf are ignored
4784 * fields that were used by "recursive" rather than "ort" -- but
4785 * sanity check them anyway.
4787 assert(opt
->detect_renames
>= -1 &&
4788 opt
->detect_renames
<= DIFF_DETECT_COPY
);
4789 assert(opt
->verbosity
>= 0 && opt
->verbosity
<= 5);
4790 assert(opt
->buffer_output
<= 2);
4791 assert(opt
->obuf
.len
== 0);
4793 assert(opt
->priv
== NULL
);
4794 if (result
->_properly_initialized
!= 0 &&
4795 result
->_properly_initialized
!= RESULT_INITIALIZED
)
4796 BUG("struct merge_result passed to merge_incore_*recursive() must be zeroed or filled with values from a previous run");
4797 assert(!!result
->priv
== !!result
->_properly_initialized
);
4799 opt
->priv
= result
->priv
;
4800 result
->priv
= NULL
;
4802 * opt->priv non-NULL means we had results from a previous
4803 * run; do a few sanity checks that user didn't mess with
4804 * it in an obvious fashion.
4806 assert(opt
->priv
->call_depth
== 0);
4807 assert(!opt
->priv
->toplevel_dir
||
4808 0 == strlen(opt
->priv
->toplevel_dir
));
4810 trace2_region_leave("merge", "sanity checks", opt
->repo
);
4812 /* Default to histogram diff. Actually, just hardcode it...for now. */
4813 opt
->xdl_opts
= DIFF_WITH_ALG(opt
, HISTOGRAM_DIFF
);
4815 /* Handle attr direction stuff for renormalization */
4816 if (opt
->renormalize
)
4817 git_attr_set_direction(GIT_ATTR_CHECKOUT
);
4819 /* Initialization of opt->priv, our internal merge data */
4820 trace2_region_enter("merge", "allocate/init", opt
->repo
);
4822 clear_or_reinit_internal_opts(opt
->priv
, 1);
4823 string_list_init_nodup(&opt
->priv
->conflicted_submodules
);
4824 trace2_region_leave("merge", "allocate/init", opt
->repo
);
4827 opt
->priv
= xcalloc(1, sizeof(*opt
->priv
));
4829 /* Initialization of various renames fields */
4830 renames
= &opt
->priv
->renames
;
4831 mem_pool_init(&opt
->priv
->pool
, 0);
4832 pool
= &opt
->priv
->pool
;
4833 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; i
++) {
4834 strintmap_init_with_options(&renames
->dirs_removed
[i
],
4835 NOT_RELEVANT
, pool
, 0);
4836 strmap_init_with_options(&renames
->dir_rename_count
[i
],
4838 strmap_init_with_options(&renames
->dir_renames
[i
],
4841 * relevant_sources uses -1 for the default, because we need
4842 * to be able to distinguish not-in-strintmap from valid
4843 * relevant_source values from enum file_rename_relevance.
4844 * In particular, possibly_cache_new_pair() expects a negative
4845 * value for not-found entries.
4847 strintmap_init_with_options(&renames
->relevant_sources
[i
],
4848 -1 /* explicitly invalid */,
4850 strmap_init_with_options(&renames
->cached_pairs
[i
],
4852 strset_init_with_options(&renames
->cached_irrelevant
[i
],
4854 strset_init_with_options(&renames
->cached_target_names
[i
],
4857 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; i
++) {
4858 strintmap_init_with_options(&renames
->deferred
[i
].possible_trivial_merges
,
4860 strset_init_with_options(&renames
->deferred
[i
].target_dirs
,
4862 renames
->deferred
[i
].trivial_merges_okay
= 1; /* 1 == maybe */
4866 * Although we initialize opt->priv->paths with strdup_strings=0,
4867 * that's just to avoid making yet another copy of an allocated
4868 * string. Putting the entry into paths means we are taking
4869 * ownership, so we will later free it.
4871 * In contrast, conflicted just has a subset of keys from paths, so
4872 * we don't want to free those (it'd be a duplicate free).
4874 strmap_init_with_options(&opt
->priv
->paths
, pool
, 0);
4875 strmap_init_with_options(&opt
->priv
->conflicted
, pool
, 0);
4878 * keys & string_lists in conflicts will sometimes need to outlive
4879 * "paths", so it will have a copy of relevant keys. It's probably
4880 * a small subset of the overall paths that have special output.
4882 strmap_init(&opt
->priv
->conflicts
);
4884 trace2_region_leave("merge", "allocate/init", opt
->repo
);
4887 static void merge_check_renames_reusable(struct merge_result
*result
,
4888 struct tree
*merge_base
,
4892 struct rename_info
*renames
;
4893 struct tree
**merge_trees
;
4894 struct merge_options_internal
*opti
= result
->priv
;
4899 renames
= &opti
->renames
;
4900 merge_trees
= renames
->merge_trees
;
4903 * Handle case where previous merge operation did not want cache to
4904 * take effect, e.g. because rename/rename(1to1) makes it invalid.
4906 if (!merge_trees
[0]) {
4907 assert(!merge_trees
[0] && !merge_trees
[1] && !merge_trees
[2]);
4908 renames
->cached_pairs_valid_side
= 0; /* neither side valid */
4913 * Handle other cases; note that merge_trees[0..2] will only
4914 * be NULL if opti is, or if all three were manually set to
4915 * NULL by e.g. rename/rename(1to1) handling.
4917 assert(merge_trees
[0] && merge_trees
[1] && merge_trees
[2]);
4919 /* Check if we meet a condition for re-using cached_pairs */
4920 if (oideq(&merge_base
->object
.oid
, &merge_trees
[2]->object
.oid
) &&
4921 oideq(&side1
->object
.oid
, &result
->tree
->object
.oid
))
4922 renames
->cached_pairs_valid_side
= MERGE_SIDE1
;
4923 else if (oideq(&merge_base
->object
.oid
, &merge_trees
[1]->object
.oid
) &&
4924 oideq(&side2
->object
.oid
, &result
->tree
->object
.oid
))
4925 renames
->cached_pairs_valid_side
= MERGE_SIDE2
;
4927 renames
->cached_pairs_valid_side
= 0; /* neither side valid */
4930 /*** Function Grouping: merge_incore_*() and their internal variants ***/
4933 * Originally from merge_trees_internal(); heavily adapted, though.
4935 static void merge_ort_nonrecursive_internal(struct merge_options
*opt
,
4936 struct tree
*merge_base
,
4939 struct merge_result
*result
)
4941 struct object_id working_tree_oid
;
4943 if (opt
->subtree_shift
) {
4944 side2
= shift_tree_object(opt
->repo
, side1
, side2
,
4945 opt
->subtree_shift
);
4946 merge_base
= shift_tree_object(opt
->repo
, side1
, merge_base
,
4947 opt
->subtree_shift
);
4951 trace2_region_enter("merge", "collect_merge_info", opt
->repo
);
4952 if (collect_merge_info(opt
, merge_base
, side1
, side2
) != 0) {
4954 * TRANSLATORS: The %s arguments are: 1) tree hash of a merge
4955 * base, and 2-3) the trees for the two trees we're merging.
4957 error(_("collecting merge info failed for trees %s, %s, %s"),
4958 oid_to_hex(&merge_base
->object
.oid
),
4959 oid_to_hex(&side1
->object
.oid
),
4960 oid_to_hex(&side2
->object
.oid
));
4964 trace2_region_leave("merge", "collect_merge_info", opt
->repo
);
4966 trace2_region_enter("merge", "renames", opt
->repo
);
4967 result
->clean
= detect_and_process_renames(opt
);
4968 trace2_region_leave("merge", "renames", opt
->repo
);
4969 if (opt
->priv
->renames
.redo_after_renames
== 2) {
4970 trace2_region_enter("merge", "reset_maps", opt
->repo
);
4971 clear_or_reinit_internal_opts(opt
->priv
, 1);
4972 trace2_region_leave("merge", "reset_maps", opt
->repo
);
4976 trace2_region_enter("merge", "process_entries", opt
->repo
);
4977 if (process_entries(opt
, &working_tree_oid
) < 0)
4979 trace2_region_leave("merge", "process_entries", opt
->repo
);
4981 /* Set return values */
4982 result
->path_messages
= &opt
->priv
->conflicts
;
4984 if (result
->clean
>= 0) {
4985 result
->tree
= parse_tree_indirect(&working_tree_oid
);
4987 die(_("unable to read tree (%s)"),
4988 oid_to_hex(&working_tree_oid
));
4989 /* existence of conflicted entries implies unclean */
4990 result
->clean
&= strmap_empty(&opt
->priv
->conflicted
);
4992 if (!opt
->priv
->call_depth
) {
4993 result
->priv
= opt
->priv
;
4994 result
->_properly_initialized
= RESULT_INITIALIZED
;
5000 * Originally from merge_recursive_internal(); somewhat adapted, though.
5002 static void merge_ort_internal(struct merge_options
*opt
,
5003 struct commit_list
*merge_bases
,
5006 struct merge_result
*result
)
5008 struct commit
*next
;
5009 struct commit
*merged_merge_bases
;
5010 const char *ancestor_name
;
5011 struct strbuf merge_base_abbrev
= STRBUF_INIT
;
5014 merge_bases
= repo_get_merge_bases(the_repository
, h1
, h2
);
5015 /* See merge-ort.h:merge_incore_recursive() declaration NOTE */
5016 merge_bases
= reverse_commit_list(merge_bases
);
5019 merged_merge_bases
= pop_commit(&merge_bases
);
5020 if (!merged_merge_bases
) {
5021 /* if there is no common ancestor, use an empty tree */
5024 tree
= lookup_tree(opt
->repo
, opt
->repo
->hash_algo
->empty_tree
);
5025 merged_merge_bases
= make_virtual_commit(opt
->repo
, tree
,
5027 ancestor_name
= "empty tree";
5028 } else if (merge_bases
) {
5029 ancestor_name
= "merged common ancestors";
5031 strbuf_add_unique_abbrev(&merge_base_abbrev
,
5032 &merged_merge_bases
->object
.oid
,
5034 ancestor_name
= merge_base_abbrev
.buf
;
5037 for (next
= pop_commit(&merge_bases
); next
;
5038 next
= pop_commit(&merge_bases
)) {
5039 const char *saved_b1
, *saved_b2
;
5040 struct commit
*prev
= merged_merge_bases
;
5042 opt
->priv
->call_depth
++;
5044 * When the merge fails, the result contains files
5045 * with conflict markers. The cleanness flag is
5046 * ignored (unless indicating an error), it was never
5047 * actually used, as result of merge_trees has always
5048 * overwritten it: the committed "conflicts" were
5051 saved_b1
= opt
->branch1
;
5052 saved_b2
= opt
->branch2
;
5053 opt
->branch1
= "Temporary merge branch 1";
5054 opt
->branch2
= "Temporary merge branch 2";
5055 merge_ort_internal(opt
, NULL
, prev
, next
, result
);
5056 if (result
->clean
< 0)
5058 opt
->branch1
= saved_b1
;
5059 opt
->branch2
= saved_b2
;
5060 opt
->priv
->call_depth
--;
5062 merged_merge_bases
= make_virtual_commit(opt
->repo
,
5065 commit_list_insert(prev
, &merged_merge_bases
->parents
);
5066 commit_list_insert(next
, &merged_merge_bases
->parents
->next
);
5068 clear_or_reinit_internal_opts(opt
->priv
, 1);
5071 opt
->ancestor
= ancestor_name
;
5072 merge_ort_nonrecursive_internal(opt
,
5073 repo_get_commit_tree(opt
->repo
,
5074 merged_merge_bases
),
5075 repo_get_commit_tree(opt
->repo
, h1
),
5076 repo_get_commit_tree(opt
->repo
, h2
),
5078 strbuf_release(&merge_base_abbrev
);
5079 opt
->ancestor
= NULL
; /* avoid accidental re-use of opt->ancestor */
5082 void merge_incore_nonrecursive(struct merge_options
*opt
,
5083 struct tree
*merge_base
,
5086 struct merge_result
*result
)
5088 trace2_region_enter("merge", "incore_nonrecursive", opt
->repo
);
5090 trace2_region_enter("merge", "merge_start", opt
->repo
);
5091 assert(opt
->ancestor
!= NULL
);
5092 merge_check_renames_reusable(result
, merge_base
, side1
, side2
);
5093 merge_start(opt
, result
);
5095 * Record the trees used in this merge, so if there's a next merge in
5096 * a cherry-pick or rebase sequence it might be able to take advantage
5097 * of the cached_pairs in that next merge.
5099 opt
->priv
->renames
.merge_trees
[0] = merge_base
;
5100 opt
->priv
->renames
.merge_trees
[1] = side1
;
5101 opt
->priv
->renames
.merge_trees
[2] = side2
;
5102 trace2_region_leave("merge", "merge_start", opt
->repo
);
5104 merge_ort_nonrecursive_internal(opt
, merge_base
, side1
, side2
, result
);
5105 trace2_region_leave("merge", "incore_nonrecursive", opt
->repo
);
5108 void merge_incore_recursive(struct merge_options
*opt
,
5109 struct commit_list
*merge_bases
,
5110 struct commit
*side1
,
5111 struct commit
*side2
,
5112 struct merge_result
*result
)
5114 trace2_region_enter("merge", "incore_recursive", opt
->repo
);
5116 /* We set the ancestor label based on the merge_bases */
5117 assert(opt
->ancestor
== NULL
);
5119 trace2_region_enter("merge", "merge_start", opt
->repo
);
5120 merge_start(opt
, result
);
5121 trace2_region_leave("merge", "merge_start", opt
->repo
);
5123 merge_ort_internal(opt
, merge_bases
, side1
, side2
, result
);
5124 trace2_region_leave("merge", "incore_recursive", opt
->repo
);