2 * "Ostensibly Recursive's Twin" merge strategy, or "ort" for short. Meant
3 * as a drop-in replacement for the "recursive" merge strategy, allowing one
6 * git merge [-s recursive]
12 * Note: git's parser allows the space between '-s' and its argument to be
13 * missing. (Should I have backronymed "ham", "alsa", "kip", "nap, "alvo",
14 * "cale", "peedy", or "ins" instead of "ort"?)
18 #include "merge-ort.h"
23 #include "cache-tree.h"
25 #include "commit-reach.h"
31 #include "object-store.h"
32 #include "promisor-remote.h"
35 #include "submodule.h"
37 #include "unpack-trees.h"
38 #include "xdiff-interface.h"
40 #define USE_MEMORY_POOL 1 /* faster, but obscures memory leak hunting */
43 * We have many arrays of size 3. Whenever we have such an array, the
44 * indices refer to one of the sides of the three-way merge. This is so
45 * pervasive that the constants 0, 1, and 2 are used in many places in the
46 * code (especially in arithmetic operations to find the other side's index
47 * or to compute a relevant mask), but sometimes these enum names are used
48 * to aid code clarity.
50 * See also 'filemask' and 'dirmask' in struct conflict_info; the "ith side"
51 * referred to there is one of these three sides.
59 static unsigned RESULT_INITIALIZED
= 0x1abe11ed; /* unlikely accidental value */
61 struct traversal_callback_data
{
63 unsigned long dirmask
;
64 struct name_entry names
[3];
67 struct deferred_traversal_data
{
69 * possible_trivial_merges: directories to be explored only when needed
71 * possible_trivial_merges is a map of directory names to
72 * dir_rename_mask. When we detect that a directory is unchanged on
73 * one side, we can sometimes resolve the directory without recursing
74 * into it. Renames are the only things that can prevent such an
75 * optimization. However, for rename sources:
76 * - If no parent directory needed directory rename detection, then
77 * no path under such a directory can be a relevant_source.
78 * and for rename destinations:
79 * - If no cached rename has a target path under the directory AND
80 * - If there are no unpaired relevant_sources elsewhere in the
82 * then we don't need any path under this directory for a rename
83 * destination. The only way to know the last item above is to defer
84 * handling such directories until the end of collect_merge_info(),
85 * in handle_deferred_entries().
87 * For each we store dir_rename_mask, since that's the only bit of
88 * information we need, other than the path, to resume the recursive
91 struct strintmap possible_trivial_merges
;
94 * trivial_merges_okay: if trivial directory merges are okay
96 * See possible_trivial_merges above. The "no unpaired
97 * relevant_sources elsewhere in the repository" is a single boolean
98 * per merge side, which we store here. Note that while 0 means no,
99 * 1 only means "maybe" rather than "yes"; we optimistically set it
100 * to 1 initially and only clear when we determine it is unsafe to
101 * do trivial directory merges.
103 unsigned trivial_merges_okay
;
106 * target_dirs: ancestor directories of rename targets
108 * target_dirs contains all directory names that are an ancestor of
109 * any rename destination.
111 struct strset target_dirs
;
116 * All variables that are arrays of size 3 correspond to data tracked
117 * for the sides in enum merge_side. Index 0 is almost always unused
118 * because we often only need to track information for MERGE_SIDE1 and
119 * MERGE_SIDE2 (MERGE_BASE can't have rename information since renames
120 * are determined relative to what changed since the MERGE_BASE).
124 * pairs: pairing of filenames from diffcore_rename()
126 struct diff_queue_struct pairs
[3];
129 * dirs_removed: directories removed on a given side of history.
131 * The keys of dirs_removed[side] are the directories that were removed
132 * on the given side of history. The value of the strintmap for each
133 * directory is a value from enum dir_rename_relevance.
135 struct strintmap dirs_removed
[3];
138 * dir_rename_count: tracking where parts of a directory were renamed to
140 * When files in a directory are renamed, they may not all go to the
141 * same location. Each strmap here tracks:
142 * old_dir => {new_dir => int}
143 * That is, dir_rename_count[side] is a strmap to a strintmap.
145 struct strmap dir_rename_count
[3];
148 * dir_renames: computed directory renames
150 * This is a map of old_dir => new_dir and is derived in part from
153 struct strmap dir_renames
[3];
156 * relevant_sources: deleted paths wanted in rename detection, and why
158 * relevant_sources is a set of deleted paths on each side of
159 * history for which we need rename detection. If a path is deleted
160 * on one side of history, we need to detect if it is part of a
162 * * the file is modified/deleted on the other side of history
163 * * we need to detect renames for an ancestor directory
164 * If neither of those are true, we can skip rename detection for
165 * that path. The reason is stored as a value from enum
166 * file_rename_relevance, as the reason can inform the algorithm in
167 * diffcore_rename_extended().
169 struct strintmap relevant_sources
[3];
171 struct deferred_traversal_data deferred
[3];
175 * 0: optimization removing unmodified potential rename source okay
176 * 2 or 4: optimization okay, but must check for files added to dir
177 * 7: optimization forbidden; need rename source in case of dir rename
179 unsigned dir_rename_mask
:3;
182 * callback_data_*: supporting data structures for alternate traversal
184 * We sometimes need to be able to traverse through all the files
185 * in a given tree before all immediate subdirectories within that
186 * tree. Since traverse_trees() doesn't do that naturally, we have
187 * a traverse_trees_wrapper() that stores any immediate
188 * subdirectories while traversing files, then traverses the
189 * immediate subdirectories later. These callback_data* variables
190 * store the information for the subdirectories so that we can do
191 * that traversal order.
193 struct traversal_callback_data
*callback_data
;
194 int callback_data_nr
, callback_data_alloc
;
195 char *callback_data_traverse_path
;
198 * merge_trees: trees passed to the merge algorithm for the merge
200 * merge_trees records the trees passed to the merge algorithm. But,
201 * this data also is stored in merge_result->priv. If a sequence of
202 * merges are being done (such as when cherry-picking or rebasing),
203 * the next merge can look at this and re-use information from
204 * previous merges under certain circumstances.
206 * See also all the cached_* variables.
208 struct tree
*merge_trees
[3];
211 * cached_pairs_valid_side: which side's cached info can be reused
213 * See the description for merge_trees. For repeated merges, at most
214 * only one side's cached information can be used. Valid values:
215 * MERGE_SIDE2: cached data from side2 can be reused
216 * MERGE_SIDE1: cached data from side1 can be reused
217 * 0: no cached data can be reused
218 * -1: See redo_after_renames; both sides can be reused.
220 int cached_pairs_valid_side
;
223 * cached_pairs: Caching of renames and deletions.
225 * These are mappings recording renames and deletions of individual
226 * files (not directories). They are thus a map from an old
227 * filename to either NULL (for deletions) or a new filename (for
230 struct strmap cached_pairs
[3];
233 * cached_target_names: just the destinations from cached_pairs
235 * We sometimes want a fast lookup to determine if a given filename
236 * is one of the destinations in cached_pairs. cached_target_names
237 * is thus duplicative information, but it provides a fast lookup.
239 struct strset cached_target_names
[3];
242 * cached_irrelevant: Caching of rename_sources that aren't relevant.
244 * If we try to detect a rename for a source path and succeed, it's
245 * part of a rename. If we try to detect a rename for a source path
246 * and fail, then it's a delete. If we do not try to detect a rename
247 * for a path, then we don't know if it's a rename or a delete. If
248 * merge-ort doesn't think the path is relevant, then we just won't
249 * cache anything for that path. But there's a slight problem in
250 * that merge-ort can think a path is RELEVANT_LOCATION, but due to
251 * commit 9bd342137e ("diffcore-rename: determine which
252 * relevant_sources are no longer relevant", 2021-03-13),
253 * diffcore-rename can downgrade the path to RELEVANT_NO_MORE. To
254 * avoid excessive calls to diffcore_rename_extended() we still need
255 * to cache such paths, though we cannot record them as either
256 * renames or deletes. So we cache them here as a "turned out to be
257 * irrelevant *for this commit*" as they are often also irrelevant
258 * for subsequent commits, though we will have to do some extra
259 * checking to see whether such paths become relevant for rename
260 * detection when cherry-picking/rebasing subsequent commits.
262 struct strset cached_irrelevant
[3];
265 * redo_after_renames: optimization flag for "restarting" the merge
267 * Sometimes it pays to detect renames, cache them, and then
268 * restart the merge operation from the beginning. The reason for
269 * this is that when we know where all the renames are, we know
270 * whether a certain directory has any paths under it affected --
271 * and if a directory is not affected then it permits us to do
272 * trivial tree merging in more cases. Doing trivial tree merging
273 * prevents the need to run process_entry() on every path
274 * underneath trees that can be trivially merged, and
275 * process_entry() is more expensive than collect_merge_info() --
276 * plus, the second collect_merge_info() will be much faster since
277 * it doesn't have to recurse into the relevant trees.
279 * Values for this flag:
280 * 0 = don't bother, not worth it (or conditions not yet checked)
281 * 1 = conditions for optimization met, optimization worthwhile
282 * 2 = we already did it (don't restart merge yet again)
284 unsigned redo_after_renames
;
287 * needed_limit: value needed for inexact rename detection to run
289 * If the current rename limit wasn't high enough for inexact
290 * rename detection to run, this records the limit needed. Otherwise,
291 * this value remains 0.
296 struct merge_options_internal
{
298 * paths: primary data structure in all of merge ort.
301 * * are full relative paths from the toplevel of the repository
302 * (e.g. "drivers/firmware/raspberrypi.c").
303 * * store all relevant paths in the repo, both directories and
304 * files (e.g. drivers, drivers/firmware would also be included)
305 * * these keys serve to intern all the path strings, which allows
306 * us to do pointer comparison on directory names instead of
307 * strcmp; we just have to be careful to use the interned strings.
308 * (Technically paths_to_free may track some strings that were
309 * removed from froms paths.)
311 * The values of paths:
312 * * either a pointer to a merged_info, or a conflict_info struct
313 * * merged_info contains all relevant information for a
314 * non-conflicted entry.
315 * * conflict_info contains a merged_info, plus any additional
316 * information about a conflict such as the higher orders stages
317 * involved and the names of the paths those came from (handy
318 * once renames get involved).
319 * * a path may start "conflicted" (i.e. point to a conflict_info)
320 * and then a later step (e.g. three-way content merge) determines
321 * it can be cleanly merged, at which point it'll be marked clean
322 * and the algorithm will ignore any data outside the contained
323 * merged_info for that entry
324 * * If an entry remains conflicted, the merged_info portion of a
325 * conflict_info will later be filled with whatever version of
326 * the file should be placed in the working directory (e.g. an
327 * as-merged-as-possible variation that contains conflict markers).
332 * conflicted: a subset of keys->values from "paths"
334 * conflicted is basically an optimization between process_entries()
335 * and record_conflicted_index_entries(); the latter could loop over
336 * ALL the entries in paths AGAIN and look for the ones that are
337 * still conflicted, but since process_entries() has to loop over
338 * all of them, it saves the ones it couldn't resolve in this strmap
339 * so that record_conflicted_index_entries() can iterate just the
342 struct strmap conflicted
;
345 * pool: memory pool for fast allocation/deallocation
347 * We allocate room for lots of filenames and auxiliary data
348 * structures in merge_options_internal, and it tends to all be
349 * freed together too. Using a memory pool for these provides a
352 struct mem_pool internal_pool
;
353 struct mem_pool
*pool
; /* NULL, or pointer to internal_pool */
356 * paths_to_free: additional list of strings to free
358 * If keys are removed from "paths", they are added to paths_to_free
359 * to ensure they are later freed. We avoid free'ing immediately since
360 * other places (e.g. conflict_info.pathnames[]) may still be
361 * referencing these paths.
363 struct string_list paths_to_free
;
366 * output: special messages and conflict notices for various paths
368 * This is a map of pathnames (a subset of the keys in "paths" above)
369 * to strbufs. It gathers various warning/conflict/notice messages
370 * for later processing.
372 struct strmap output
;
375 * renames: various data relating to rename detection
377 struct rename_info renames
;
380 * attr_index: hacky minimal index used for renormalization
382 * renormalization code _requires_ an index, though it only needs to
383 * find a .gitattributes file within the index. So, when
384 * renormalization is important, we create a special index with just
387 struct index_state attr_index
;
390 * current_dir_name, toplevel_dir: temporary vars
392 * These are used in collect_merge_info_callback(), and will set the
393 * various merged_info.directory_name for the various paths we get;
394 * see documentation for that variable and the requirements placed on
397 const char *current_dir_name
;
398 const char *toplevel_dir
;
400 /* call_depth: recursion level counter for merging merge bases */
404 struct version_info
{
405 struct object_id oid
;
410 /* if is_null, ignore result. otherwise result has oid & mode */
411 struct version_info result
;
415 * clean: whether the path in question is cleanly merged.
417 * see conflict_info.merged for more details.
422 * basename_offset: offset of basename of path.
424 * perf optimization to avoid recomputing offset of final '/'
425 * character in pathname (0 if no '/' in pathname).
427 size_t basename_offset
;
430 * directory_name: containing directory name.
432 * Note that we assume directory_name is constructed such that
433 * strcmp(dir1_name, dir2_name) == 0 iff dir1_name == dir2_name,
434 * i.e. string equality is equivalent to pointer equality. For this
435 * to hold, we have to be careful setting directory_name.
437 const char *directory_name
;
440 struct conflict_info
{
442 * merged: the version of the path that will be written to working tree
444 * WARNING: It is critical to check merged.clean and ensure it is 0
445 * before reading any conflict_info fields outside of merged.
446 * Allocated merge_info structs will always have clean set to 1.
447 * Allocated conflict_info structs will have merged.clean set to 0
448 * initially. The merged.clean field is how we know if it is safe
449 * to access other parts of conflict_info besides merged; if a
450 * conflict_info's merged.clean is changed to 1, the rest of the
451 * algorithm is not allowed to look at anything outside of the
452 * merged member anymore.
454 struct merged_info merged
;
456 /* oids & modes from each of the three trees for this path */
457 struct version_info stages
[3];
459 /* pathnames for each stage; may differ due to rename detection */
460 const char *pathnames
[3];
462 /* Whether this path is/was involved in a directory/file conflict */
463 unsigned df_conflict
:1;
466 * Whether this path is/was involved in a non-content conflict other
467 * than a directory/file conflict (e.g. rename/rename, rename/delete,
468 * file location based on possible directory rename).
470 unsigned path_conflict
:1;
473 * For filemask and dirmask, the ith bit corresponds to whether the
474 * ith entry is a file (filemask) or a directory (dirmask). Thus,
475 * filemask & dirmask is always zero, and filemask | dirmask is at
476 * most 7 but can be less when a path does not appear as either a
477 * file or a directory on at least one side of history.
479 * Note that these masks are related to enum merge_side, as the ith
480 * entry corresponds to side i.
482 * These values come from a traverse_trees() call; more info may be
483 * found looking at tree-walk.h's struct traverse_info,
484 * particularly the documentation above the "fn" member (note that
485 * filemask = mask & ~dirmask from that documentation).
491 * Optimization to track which stages match, to avoid the need to
492 * recompute it in multiple steps. Either 0 or at least 2 bits are
493 * set; if at least 2 bits are set, their corresponding stages match.
495 unsigned match_mask
:3;
498 /*** Function Grouping: various utility functions ***/
501 * For the next three macros, see warning for conflict_info.merged.
503 * In each of the below, mi is a struct merged_info*, and ci was defined
504 * as a struct conflict_info* (but we need to verify ci isn't actually
505 * pointed at a struct merged_info*).
507 * INITIALIZE_CI: Assign ci to mi but only if it's safe; set to NULL otherwise.
508 * VERIFY_CI: Ensure that something we assigned to a conflict_info* is one.
509 * ASSIGN_AND_VERIFY_CI: Similar to VERIFY_CI but do assignment first.
511 #define INITIALIZE_CI(ci, mi) do { \
512 (ci) = (!(mi) || (mi)->clean) ? NULL : (struct conflict_info *)(mi); \
514 #define VERIFY_CI(ci) assert(ci && !ci->merged.clean);
515 #define ASSIGN_AND_VERIFY_CI(ci, mi) do { \
516 (ci) = (struct conflict_info *)(mi); \
517 assert((ci) && !(mi)->clean); \
520 static void free_strmap_strings(struct strmap
*map
)
522 struct hashmap_iter iter
;
523 struct strmap_entry
*entry
;
525 strmap_for_each_entry(map
, &iter
, entry
) {
526 free((char*)entry
->key
);
530 static void clear_or_reinit_internal_opts(struct merge_options_internal
*opti
,
533 struct rename_info
*renames
= &opti
->renames
;
535 void (*strmap_clear_func
)(struct strmap
*, int) =
536 reinitialize
? strmap_partial_clear
: strmap_clear
;
537 void (*strintmap_clear_func
)(struct strintmap
*) =
538 reinitialize
? strintmap_partial_clear
: strintmap_clear
;
539 void (*strset_clear_func
)(struct strset
*) =
540 reinitialize
? strset_partial_clear
: strset_clear
;
543 * We marked opti->paths with strdup_strings = 0, so that we
544 * wouldn't have to make another copy of the fullpath created by
545 * make_traverse_path from setup_path_info(). But, now that we've
546 * used it and have no other references to these strings, it is time
547 * to deallocate them.
549 free_strmap_strings(&opti
->paths
);
550 strmap_clear_func(&opti
->paths
, 1);
553 * All keys and values in opti->conflicted are a subset of those in
554 * opti->paths. We don't want to deallocate anything twice, so we
555 * don't free the keys and we pass 0 for free_values.
557 strmap_clear_func(&opti
->conflicted
, 0);
560 * opti->paths_to_free is similar to opti->paths; we created it with
561 * strdup_strings = 0 to avoid making _another_ copy of the fullpath
562 * but now that we've used it and have no other references to these
563 * strings, it is time to deallocate them. We do so by temporarily
564 * setting strdup_strings to 1.
566 opti
->paths_to_free
.strdup_strings
= 1;
567 string_list_clear(&opti
->paths_to_free
, 0);
568 opti
->paths_to_free
.strdup_strings
= 0;
570 if (opti
->attr_index
.cache_nr
) /* true iff opt->renormalize */
571 discard_index(&opti
->attr_index
);
573 /* Free memory used by various renames maps */
574 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; ++i
) {
575 strintmap_clear_func(&renames
->dirs_removed
[i
]);
576 strmap_clear_func(&renames
->dir_renames
[i
], 0);
577 strintmap_clear_func(&renames
->relevant_sources
[i
]);
579 assert(renames
->cached_pairs_valid_side
== 0);
580 if (i
!= renames
->cached_pairs_valid_side
&&
581 -1 != renames
->cached_pairs_valid_side
) {
582 strset_clear_func(&renames
->cached_target_names
[i
]);
583 strmap_clear_func(&renames
->cached_pairs
[i
], 1);
584 strset_clear_func(&renames
->cached_irrelevant
[i
]);
585 partial_clear_dir_rename_count(&renames
->dir_rename_count
[i
]);
587 strmap_clear(&renames
->dir_rename_count
[i
], 1);
590 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; ++i
) {
591 strintmap_clear_func(&renames
->deferred
[i
].possible_trivial_merges
);
592 strset_clear_func(&renames
->deferred
[i
].target_dirs
);
593 renames
->deferred
[i
].trivial_merges_okay
= 1; /* 1 == maybe */
595 renames
->cached_pairs_valid_side
= 0;
596 renames
->dir_rename_mask
= 0;
599 struct hashmap_iter iter
;
600 struct strmap_entry
*e
;
602 /* Release and free each strbuf found in output */
603 strmap_for_each_entry(&opti
->output
, &iter
, e
) {
604 struct strbuf
*sb
= e
->value
;
607 * While strictly speaking we don't need to free(sb)
608 * here because we could pass free_values=1 when
609 * calling strmap_clear() on opti->output, that would
610 * require strmap_clear to do another
611 * strmap_for_each_entry() loop, so we just free it
612 * while we're iterating anyway.
616 strmap_clear(&opti
->output
, 0);
620 mem_pool_discard(&opti
->internal_pool
, 0);
625 /* Clean out callback_data as well. */
626 FREE_AND_NULL(renames
->callback_data
);
627 renames
->callback_data_nr
= renames
->callback_data_alloc
= 0;
630 static int err(struct merge_options
*opt
, const char *err
, ...)
633 struct strbuf sb
= STRBUF_INIT
;
635 strbuf_addstr(&sb
, "error: ");
636 va_start(params
, err
);
637 strbuf_vaddf(&sb
, err
, params
);
646 static void format_commit(struct strbuf
*sb
,
648 struct commit
*commit
)
650 struct merge_remote_desc
*desc
;
651 struct pretty_print_context ctx
= {0};
652 ctx
.abbrev
= DEFAULT_ABBREV
;
654 strbuf_addchars(sb
, ' ', indent
);
655 desc
= merge_remote_util(commit
);
657 strbuf_addf(sb
, "virtual %s\n", desc
->name
);
661 format_commit_message(commit
, "%h %s", sb
, &ctx
);
662 strbuf_addch(sb
, '\n');
665 __attribute__((format (printf
, 4, 5)))
666 static void path_msg(struct merge_options
*opt
,
668 int omittable_hint
, /* skippable under --remerge-diff */
669 const char *fmt
, ...)
672 struct strbuf
*sb
= strmap_get(&opt
->priv
->output
, path
);
674 sb
= xmalloc(sizeof(*sb
));
676 strmap_put(&opt
->priv
->output
, path
, sb
);
680 strbuf_vaddf(sb
, fmt
, ap
);
683 strbuf_addch(sb
, '\n');
687 static void *pool_calloc(struct mem_pool
*pool
, size_t count
, size_t size
)
690 return xcalloc(count
, size
);
691 return mem_pool_calloc(pool
, count
, size
);
695 static void *pool_alloc(struct mem_pool
*pool
, size_t size
)
698 return xmalloc(size
);
699 return mem_pool_alloc(pool
, size
);
703 static void *pool_strndup(struct mem_pool
*pool
, const char *str
, size_t len
)
706 return xstrndup(str
, len
);
707 return mem_pool_strndup(pool
, str
, len
);
710 /* add a string to a strbuf, but converting "/" to "_" */
711 static void add_flattened_path(struct strbuf
*out
, const char *s
)
714 strbuf_addstr(out
, s
);
715 for (; i
< out
->len
; i
++)
716 if (out
->buf
[i
] == '/')
720 static char *unique_path(struct strmap
*existing_paths
,
724 struct strbuf newpath
= STRBUF_INIT
;
728 strbuf_addf(&newpath
, "%s~", path
);
729 add_flattened_path(&newpath
, branch
);
731 base_len
= newpath
.len
;
732 while (strmap_contains(existing_paths
, newpath
.buf
)) {
733 strbuf_setlen(&newpath
, base_len
);
734 strbuf_addf(&newpath
, "_%d", suffix
++);
737 return strbuf_detach(&newpath
, NULL
);
740 /*** Function Grouping: functions related to collect_merge_info() ***/
742 static int traverse_trees_wrapper_callback(int n
,
744 unsigned long dirmask
,
745 struct name_entry
*names
,
746 struct traverse_info
*info
)
748 struct merge_options
*opt
= info
->data
;
749 struct rename_info
*renames
= &opt
->priv
->renames
;
750 unsigned filemask
= mask
& ~dirmask
;
754 if (!renames
->callback_data_traverse_path
)
755 renames
->callback_data_traverse_path
= xstrdup(info
->traverse_path
);
757 if (filemask
&& filemask
== renames
->dir_rename_mask
)
758 renames
->dir_rename_mask
= 0x07;
760 ALLOC_GROW(renames
->callback_data
, renames
->callback_data_nr
+ 1,
761 renames
->callback_data_alloc
);
762 renames
->callback_data
[renames
->callback_data_nr
].mask
= mask
;
763 renames
->callback_data
[renames
->callback_data_nr
].dirmask
= dirmask
;
764 COPY_ARRAY(renames
->callback_data
[renames
->callback_data_nr
].names
,
766 renames
->callback_data_nr
++;
772 * Much like traverse_trees(), BUT:
773 * - read all the tree entries FIRST, saving them
774 * - note that the above step provides an opportunity to compute necessary
775 * additional details before the "real" traversal
776 * - loop through the saved entries and call the original callback on them
778 static int traverse_trees_wrapper(struct index_state
*istate
,
781 struct traverse_info
*info
)
783 int ret
, i
, old_offset
;
784 traverse_callback_t old_fn
;
785 char *old_callback_data_traverse_path
;
786 struct merge_options
*opt
= info
->data
;
787 struct rename_info
*renames
= &opt
->priv
->renames
;
789 assert(renames
->dir_rename_mask
== 2 || renames
->dir_rename_mask
== 4);
791 old_callback_data_traverse_path
= renames
->callback_data_traverse_path
;
793 old_offset
= renames
->callback_data_nr
;
795 renames
->callback_data_traverse_path
= NULL
;
796 info
->fn
= traverse_trees_wrapper_callback
;
797 ret
= traverse_trees(istate
, n
, t
, info
);
801 info
->traverse_path
= renames
->callback_data_traverse_path
;
803 for (i
= old_offset
; i
< renames
->callback_data_nr
; ++i
) {
805 renames
->callback_data
[i
].mask
,
806 renames
->callback_data
[i
].dirmask
,
807 renames
->callback_data
[i
].names
,
811 renames
->callback_data_nr
= old_offset
;
812 free(renames
->callback_data_traverse_path
);
813 renames
->callback_data_traverse_path
= old_callback_data_traverse_path
;
814 info
->traverse_path
= NULL
;
818 static void setup_path_info(struct merge_options
*opt
,
819 struct string_list_item
*result
,
820 const char *current_dir_name
,
821 int current_dir_name_len
,
822 char *fullpath
, /* we'll take over ownership */
823 struct name_entry
*names
,
824 struct name_entry
*merged_version
,
825 unsigned is_null
, /* boolean */
826 unsigned df_conflict
, /* boolean */
829 int resolved
/* boolean */)
831 /* result->util is void*, so mi is a convenience typed variable */
832 struct merged_info
*mi
;
834 assert(!is_null
|| resolved
);
835 assert(!df_conflict
|| !resolved
); /* df_conflict implies !resolved */
836 assert(resolved
== (merged_version
!= NULL
));
838 mi
= xcalloc(1, resolved
? sizeof(struct merged_info
) :
839 sizeof(struct conflict_info
));
840 mi
->directory_name
= current_dir_name
;
841 mi
->basename_offset
= current_dir_name_len
;
842 mi
->clean
= !!resolved
;
844 mi
->result
.mode
= merged_version
->mode
;
845 oidcpy(&mi
->result
.oid
, &merged_version
->oid
);
846 mi
->is_null
= !!is_null
;
849 struct conflict_info
*ci
;
851 ASSIGN_AND_VERIFY_CI(ci
, mi
);
852 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
853 ci
->pathnames
[i
] = fullpath
;
854 ci
->stages
[i
].mode
= names
[i
].mode
;
855 oidcpy(&ci
->stages
[i
].oid
, &names
[i
].oid
);
857 ci
->filemask
= filemask
;
858 ci
->dirmask
= dirmask
;
859 ci
->df_conflict
= !!df_conflict
;
862 * Assume is_null for now, but if we have entries
863 * under the directory then when it is complete in
864 * write_completed_directory() it'll update this.
865 * Also, for D/F conflicts, we have to handle the
866 * directory first, then clear this bit and process
867 * the file to see how it is handled -- that occurs
868 * near the top of process_entry().
872 strmap_put(&opt
->priv
->paths
, fullpath
, mi
);
873 result
->string
= fullpath
;
877 static void add_pair(struct merge_options
*opt
,
878 struct name_entry
*names
,
879 const char *pathname
,
881 unsigned is_add
/* if false, is_delete */,
883 unsigned dir_rename_mask
)
885 struct diff_filespec
*one
, *two
;
886 struct rename_info
*renames
= &opt
->priv
->renames
;
887 int names_idx
= is_add
? side
: 0;
890 assert(match_mask
== 0 || match_mask
== 6);
891 if (strset_contains(&renames
->cached_target_names
[side
],
895 unsigned content_relevant
= (match_mask
== 0);
896 unsigned location_relevant
= (dir_rename_mask
== 0x07);
898 assert(match_mask
== 0 || match_mask
== 3 || match_mask
== 5);
901 * If pathname is found in cached_irrelevant[side] due to
902 * previous pick but for this commit content is relevant,
903 * then we need to remove it from cached_irrelevant.
905 if (content_relevant
)
906 /* strset_remove is no-op if strset doesn't have key */
907 strset_remove(&renames
->cached_irrelevant
[side
],
911 * We do not need to re-detect renames for paths that we already
912 * know the pairing, i.e. for cached_pairs (or
913 * cached_irrelevant). However, handle_deferred_entries() needs
914 * to loop over the union of keys from relevant_sources[side] and
915 * cached_pairs[side], so for simplicity we set relevant_sources
916 * for all the cached_pairs too and then strip them back out in
917 * prune_cached_from_relevant() at the beginning of
918 * detect_regular_renames().
920 if (content_relevant
|| location_relevant
) {
921 /* content_relevant trumps location_relevant */
922 strintmap_set(&renames
->relevant_sources
[side
], pathname
,
923 content_relevant
? RELEVANT_CONTENT
: RELEVANT_LOCATION
);
927 * Avoid creating pair if we've already cached rename results.
928 * Note that we do this after setting relevant_sources[side]
929 * as noted in the comment above.
931 if (strmap_contains(&renames
->cached_pairs
[side
], pathname
) ||
932 strset_contains(&renames
->cached_irrelevant
[side
], pathname
))
936 one
= alloc_filespec(pathname
);
937 two
= alloc_filespec(pathname
);
938 fill_filespec(is_add
? two
: one
,
939 &names
[names_idx
].oid
, 1, names
[names_idx
].mode
);
940 diff_queue(&renames
->pairs
[side
], one
, two
);
943 static void collect_rename_info(struct merge_options
*opt
,
944 struct name_entry
*names
,
946 const char *fullname
,
951 struct rename_info
*renames
= &opt
->priv
->renames
;
955 * Update dir_rename_mask (determines ignore-rename-source validity)
957 * dir_rename_mask helps us keep track of when directory rename
958 * detection may be relevant. Basically, whenver a directory is
959 * removed on one side of history, and a file is added to that
960 * directory on the other side of history, directory rename
961 * detection is relevant (meaning we have to detect renames for all
962 * files within that directory to deduce where the directory
963 * moved). Also, whenever a directory needs directory rename
964 * detection, due to the "majority rules" choice for where to move
965 * it (see t6423 testcase 1f), we also need to detect renames for
966 * all files within subdirectories of that directory as well.
968 * Here we haven't looked at files within the directory yet, we are
969 * just looking at the directory itself. So, if we aren't yet in
970 * a case where a parent directory needed directory rename detection
971 * (i.e. dir_rename_mask != 0x07), and if the directory was removed
972 * on one side of history, record the mask of the other side of
973 * history in dir_rename_mask.
975 if (renames
->dir_rename_mask
!= 0x07 &&
976 (dirmask
== 3 || dirmask
== 5)) {
977 /* simple sanity check */
978 assert(renames
->dir_rename_mask
== 0 ||
979 renames
->dir_rename_mask
== (dirmask
& ~1));
980 /* update dir_rename_mask; have it record mask of new side */
981 renames
->dir_rename_mask
= (dirmask
& ~1);
984 /* Update dirs_removed, as needed */
985 if (dirmask
== 1 || dirmask
== 3 || dirmask
== 5) {
986 /* absent_mask = 0x07 - dirmask; sides = absent_mask/2 */
987 unsigned sides
= (0x07 - dirmask
)/2;
988 unsigned relevance
= (renames
->dir_rename_mask
== 0x07) ?
989 RELEVANT_FOR_ANCESTOR
: NOT_RELEVANT
;
991 * Record relevance of this directory. However, note that
992 * when collect_merge_info_callback() recurses into this
993 * directory and calls collect_rename_info() on paths
994 * within that directory, if we find a path that was added
995 * to this directory on the other side of history, we will
996 * upgrade this value to RELEVANT_FOR_SELF; see below.
999 strintmap_set(&renames
->dirs_removed
[1], fullname
,
1002 strintmap_set(&renames
->dirs_removed
[2], fullname
,
1007 * Here's the block that potentially upgrades to RELEVANT_FOR_SELF.
1008 * When we run across a file added to a directory. In such a case,
1009 * find the directory of the file and upgrade its relevance.
1011 if (renames
->dir_rename_mask
== 0x07 &&
1012 (filemask
== 2 || filemask
== 4)) {
1014 * Need directory rename for parent directory on other side
1015 * of history from added file. Thus
1016 * side = (~filemask & 0x06) >> 1
1018 * side = 3 - (filemask/2).
1020 unsigned side
= 3 - (filemask
>> 1);
1021 strintmap_set(&renames
->dirs_removed
[side
], dirname
,
1025 if (filemask
== 0 || filemask
== 7)
1028 for (side
= MERGE_SIDE1
; side
<= MERGE_SIDE2
; ++side
) {
1029 unsigned side_mask
= (1 << side
);
1031 /* Check for deletion on side */
1032 if ((filemask
& 1) && !(filemask
& side_mask
))
1033 add_pair(opt
, names
, fullname
, side
, 0 /* delete */,
1034 match_mask
& filemask
,
1035 renames
->dir_rename_mask
);
1037 /* Check for addition on side */
1038 if (!(filemask
& 1) && (filemask
& side_mask
))
1039 add_pair(opt
, names
, fullname
, side
, 1 /* add */,
1040 match_mask
& filemask
,
1041 renames
->dir_rename_mask
);
1045 static int collect_merge_info_callback(int n
,
1047 unsigned long dirmask
,
1048 struct name_entry
*names
,
1049 struct traverse_info
*info
)
1053 * common ancestor (mbase) has mask 1, and stored in index 0 of names
1054 * head of side 1 (side1) has mask 2, and stored in index 1 of names
1055 * head of side 2 (side2) has mask 4, and stored in index 2 of names
1057 struct merge_options
*opt
= info
->data
;
1058 struct merge_options_internal
*opti
= opt
->priv
;
1059 struct rename_info
*renames
= &opt
->priv
->renames
;
1060 struct string_list_item pi
; /* Path Info */
1061 struct conflict_info
*ci
; /* typed alias to pi.util (which is void*) */
1062 struct name_entry
*p
;
1065 const char *dirname
= opti
->current_dir_name
;
1066 unsigned prev_dir_rename_mask
= renames
->dir_rename_mask
;
1067 unsigned filemask
= mask
& ~dirmask
;
1068 unsigned match_mask
= 0; /* will be updated below */
1069 unsigned mbase_null
= !(mask
& 1);
1070 unsigned side1_null
= !(mask
& 2);
1071 unsigned side2_null
= !(mask
& 4);
1072 unsigned side1_matches_mbase
= (!side1_null
&& !mbase_null
&&
1073 names
[0].mode
== names
[1].mode
&&
1074 oideq(&names
[0].oid
, &names
[1].oid
));
1075 unsigned side2_matches_mbase
= (!side2_null
&& !mbase_null
&&
1076 names
[0].mode
== names
[2].mode
&&
1077 oideq(&names
[0].oid
, &names
[2].oid
));
1078 unsigned sides_match
= (!side1_null
&& !side2_null
&&
1079 names
[1].mode
== names
[2].mode
&&
1080 oideq(&names
[1].oid
, &names
[2].oid
));
1083 * Note: When a path is a file on one side of history and a directory
1084 * in another, we have a directory/file conflict. In such cases, if
1085 * the conflict doesn't resolve from renames and deletions, then we
1086 * always leave directories where they are and move files out of the
1087 * way. Thus, while struct conflict_info has a df_conflict field to
1088 * track such conflicts, we ignore that field for any directories at
1089 * a path and only pay attention to it for files at the given path.
1090 * The fact that we leave directories were they are also means that
1091 * we do not need to worry about getting additional df_conflict
1092 * information propagated from parent directories down to children
1093 * (unlike, say traverse_trees_recursive() in unpack-trees.c, which
1094 * sets a newinfo.df_conflicts field specifically to propagate it).
1096 unsigned df_conflict
= (filemask
!= 0) && (dirmask
!= 0);
1098 /* n = 3 is a fundamental assumption. */
1100 BUG("Called collect_merge_info_callback wrong");
1103 * A bunch of sanity checks verifying that traverse_trees() calls
1104 * us the way I expect. Could just remove these at some point,
1105 * though maybe they are helpful to future code readers.
1107 assert(mbase_null
== is_null_oid(&names
[0].oid
));
1108 assert(side1_null
== is_null_oid(&names
[1].oid
));
1109 assert(side2_null
== is_null_oid(&names
[2].oid
));
1110 assert(!mbase_null
|| !side1_null
|| !side2_null
);
1111 assert(mask
> 0 && mask
< 8);
1113 /* Determine match_mask */
1114 if (side1_matches_mbase
)
1115 match_mask
= (side2_matches_mbase
? 7 : 3);
1116 else if (side2_matches_mbase
)
1118 else if (sides_match
)
1122 * Get the name of the relevant filepath, which we'll pass to
1123 * setup_path_info() for tracking.
1128 len
= traverse_path_len(info
, p
->pathlen
);
1130 /* +1 in both of the following lines to include the NUL byte */
1131 fullpath
= xmalloc(len
+ 1);
1132 make_traverse_path(fullpath
, len
+ 1, info
, p
->path
, p
->pathlen
);
1135 * If mbase, side1, and side2 all match, we can resolve early. Even
1136 * if these are trees, there will be no renames or anything
1139 if (side1_matches_mbase
&& side2_matches_mbase
) {
1140 /* mbase, side1, & side2 all match; use mbase as resolution */
1141 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1142 names
, names
+0, mbase_null
, 0 /* df_conflict */,
1143 filemask
, dirmask
, 1 /* resolved */);
1148 * If the sides match, and all three paths are present and are
1149 * files, then we can take either as the resolution. We can't do
1150 * this with trees, because there may be rename sources from the
1153 if (sides_match
&& filemask
== 0x07) {
1154 /* use side1 (== side2) version as resolution */
1155 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1156 names
, names
+1, side1_null
, 0,
1157 filemask
, dirmask
, 1);
1162 * If side1 matches mbase and all three paths are present and are
1163 * files, then we can use side2 as the resolution. We cannot
1164 * necessarily do so this for trees, because there may be rename
1165 * destinations within side2.
1167 if (side1_matches_mbase
&& filemask
== 0x07) {
1168 /* use side2 version as resolution */
1169 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1170 names
, names
+2, side2_null
, 0,
1171 filemask
, dirmask
, 1);
1175 /* Similar to above but swapping sides 1 and 2 */
1176 if (side2_matches_mbase
&& filemask
== 0x07) {
1177 /* use side1 version as resolution */
1178 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1179 names
, names
+1, side1_null
, 0,
1180 filemask
, dirmask
, 1);
1185 * Sometimes we can tell that a source path need not be included in
1186 * rename detection -- namely, whenever either
1187 * side1_matches_mbase && side2_null
1189 * side2_matches_mbase && side1_null
1190 * However, we call collect_rename_info() even in those cases,
1191 * because exact renames are cheap and would let us remove both a
1192 * source and destination path. We'll cull the unneeded sources
1195 collect_rename_info(opt
, names
, dirname
, fullpath
,
1196 filemask
, dirmask
, match_mask
);
1199 * None of the special cases above matched, so we have a
1200 * provisional conflict. (Rename detection might allow us to
1201 * unconflict some more cases, but that comes later so all we can
1202 * do now is record the different non-null file hashes.)
1204 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1205 names
, NULL
, 0, df_conflict
, filemask
, dirmask
, 0);
1209 ci
->match_mask
= match_mask
;
1211 /* If dirmask, recurse into subdirectories */
1213 struct traverse_info newinfo
;
1214 struct tree_desc t
[3];
1215 void *buf
[3] = {NULL
, NULL
, NULL
};
1216 const char *original_dir_name
;
1220 * Check for whether we can avoid recursing due to one side
1221 * matching the merge base. The side that does NOT match is
1222 * the one that might have a rename destination we need.
1224 assert(!side1_matches_mbase
|| !side2_matches_mbase
);
1225 side
= side1_matches_mbase
? MERGE_SIDE2
:
1226 side2_matches_mbase
? MERGE_SIDE1
: MERGE_BASE
;
1227 if (filemask
== 0 && (dirmask
== 2 || dirmask
== 4)) {
1229 * Also defer recursing into new directories; set up a
1230 * few variables to let us do so.
1232 ci
->match_mask
= (7 - dirmask
);
1235 if (renames
->dir_rename_mask
!= 0x07 &&
1236 side
!= MERGE_BASE
&&
1237 renames
->deferred
[side
].trivial_merges_okay
&&
1238 !strset_contains(&renames
->deferred
[side
].target_dirs
,
1240 strintmap_set(&renames
->deferred
[side
].possible_trivial_merges
,
1241 pi
.string
, renames
->dir_rename_mask
);
1242 renames
->dir_rename_mask
= prev_dir_rename_mask
;
1246 /* We need to recurse */
1247 ci
->match_mask
&= filemask
;
1249 newinfo
.prev
= info
;
1250 newinfo
.name
= p
->path
;
1251 newinfo
.namelen
= p
->pathlen
;
1252 newinfo
.pathlen
= st_add3(newinfo
.pathlen
, p
->pathlen
, 1);
1254 * If this directory we are about to recurse into cared about
1255 * its parent directory (the current directory) having a D/F
1256 * conflict, then we'd propagate the masks in this way:
1257 * newinfo.df_conflicts |= (mask & ~dirmask);
1258 * But we don't worry about propagating D/F conflicts. (See
1259 * comment near setting of local df_conflict variable near
1260 * the beginning of this function).
1263 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
1264 if (i
== 1 && side1_matches_mbase
)
1266 else if (i
== 2 && side2_matches_mbase
)
1268 else if (i
== 2 && sides_match
)
1271 const struct object_id
*oid
= NULL
;
1273 oid
= &names
[i
].oid
;
1274 buf
[i
] = fill_tree_descriptor(opt
->repo
,
1280 original_dir_name
= opti
->current_dir_name
;
1281 opti
->current_dir_name
= pi
.string
;
1282 if (renames
->dir_rename_mask
== 0 ||
1283 renames
->dir_rename_mask
== 0x07)
1284 ret
= traverse_trees(NULL
, 3, t
, &newinfo
);
1286 ret
= traverse_trees_wrapper(NULL
, 3, t
, &newinfo
);
1287 opti
->current_dir_name
= original_dir_name
;
1288 renames
->dir_rename_mask
= prev_dir_rename_mask
;
1290 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++)
1300 static void resolve_trivial_directory_merge(struct conflict_info
*ci
, int side
)
1303 assert((side
== 1 && ci
->match_mask
== 5) ||
1304 (side
== 2 && ci
->match_mask
== 3));
1305 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[side
].oid
);
1306 ci
->merged
.result
.mode
= ci
->stages
[side
].mode
;
1307 ci
->merged
.is_null
= is_null_oid(&ci
->stages
[side
].oid
);
1309 ci
->merged
.clean
= 1; /* (ci->filemask == 0); */
1312 static int handle_deferred_entries(struct merge_options
*opt
,
1313 struct traverse_info
*info
)
1315 struct rename_info
*renames
= &opt
->priv
->renames
;
1316 struct hashmap_iter iter
;
1317 struct strmap_entry
*entry
;
1319 int path_count_before
, path_count_after
= 0;
1321 path_count_before
= strmap_get_size(&opt
->priv
->paths
);
1322 for (side
= MERGE_SIDE1
; side
<= MERGE_SIDE2
; side
++) {
1323 unsigned optimization_okay
= 1;
1324 struct strintmap copy
;
1326 /* Loop over the set of paths we need to know rename info for */
1327 strset_for_each_entry(&renames
->relevant_sources
[side
],
1329 char *rename_target
, *dir
, *dir_marker
;
1330 struct strmap_entry
*e
;
1333 * If we don't know delete/rename info for this path,
1334 * then we need to recurse into all trees to get all
1335 * adds to make sure we have it.
1337 if (strset_contains(&renames
->cached_irrelevant
[side
],
1340 e
= strmap_get_entry(&renames
->cached_pairs
[side
],
1343 optimization_okay
= 0;
1347 /* If this is a delete, we have enough info already */
1348 rename_target
= e
->value
;
1352 /* If we already walked the rename target, we're good */
1353 if (strmap_contains(&opt
->priv
->paths
, rename_target
))
1357 * Otherwise, we need to get a list of directories that
1358 * will need to be recursed into to get this
1361 dir
= xstrdup(rename_target
);
1362 while ((dir_marker
= strrchr(dir
, '/'))) {
1364 if (strset_contains(&renames
->deferred
[side
].target_dirs
,
1367 strset_add(&renames
->deferred
[side
].target_dirs
,
1372 renames
->deferred
[side
].trivial_merges_okay
= optimization_okay
;
1374 * We need to recurse into any directories in
1375 * possible_trivial_merges[side] found in target_dirs[side].
1376 * But when we recurse, we may need to queue up some of the
1377 * subdirectories for possible_trivial_merges[side]. Since
1378 * we can't safely iterate through a hashmap while also adding
1379 * entries, move the entries into 'copy', iterate over 'copy',
1380 * and then we'll also iterate anything added into
1381 * possible_trivial_merges[side] once this loop is done.
1383 copy
= renames
->deferred
[side
].possible_trivial_merges
;
1384 strintmap_init_with_options(&renames
->deferred
[side
].possible_trivial_merges
,
1388 strintmap_for_each_entry(©
, &iter
, entry
) {
1389 const char *path
= entry
->key
;
1390 unsigned dir_rename_mask
= (intptr_t)entry
->value
;
1391 struct conflict_info
*ci
;
1393 struct tree_desc t
[3];
1394 void *buf
[3] = {NULL
,};
1397 ci
= strmap_get(&opt
->priv
->paths
, path
);
1399 dirmask
= ci
->dirmask
;
1401 if (optimization_okay
&&
1402 !strset_contains(&renames
->deferred
[side
].target_dirs
,
1404 resolve_trivial_directory_merge(ci
, side
);
1409 info
->namelen
= strlen(path
);
1410 info
->pathlen
= info
->namelen
+ 1;
1412 for (i
= 0; i
< 3; i
++, dirmask
>>= 1) {
1413 if (i
== 1 && ci
->match_mask
== 3)
1415 else if (i
== 2 && ci
->match_mask
== 5)
1417 else if (i
== 2 && ci
->match_mask
== 6)
1420 const struct object_id
*oid
= NULL
;
1422 oid
= &ci
->stages
[i
].oid
;
1423 buf
[i
] = fill_tree_descriptor(opt
->repo
,
1428 ci
->match_mask
&= ci
->filemask
;
1429 opt
->priv
->current_dir_name
= path
;
1430 renames
->dir_rename_mask
= dir_rename_mask
;
1431 if (renames
->dir_rename_mask
== 0 ||
1432 renames
->dir_rename_mask
== 0x07)
1433 ret
= traverse_trees(NULL
, 3, t
, info
);
1435 ret
= traverse_trees_wrapper(NULL
, 3, t
, info
);
1437 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++)
1443 strintmap_clear(©
);
1444 strintmap_for_each_entry(&renames
->deferred
[side
].possible_trivial_merges
,
1446 const char *path
= entry
->key
;
1447 struct conflict_info
*ci
;
1449 ci
= strmap_get(&opt
->priv
->paths
, path
);
1452 assert(renames
->deferred
[side
].trivial_merges_okay
&&
1453 !strset_contains(&renames
->deferred
[side
].target_dirs
,
1455 resolve_trivial_directory_merge(ci
, side
);
1457 if (!optimization_okay
|| path_count_after
)
1458 path_count_after
= strmap_get_size(&opt
->priv
->paths
);
1460 if (path_count_after
) {
1462 * The choice of wanted_factor here does not affect
1463 * correctness, only performance. When the
1464 * path_count_after / path_count_before
1465 * ratio is high, redoing after renames is a big
1466 * performance boost. I suspect that redoing is a wash
1467 * somewhere near a value of 2, and below that redoing will
1468 * slow things down. I applied a fudge factor and picked
1469 * 3; see the commit message when this was introduced for
1470 * back of the envelope calculations for this ratio.
1472 const int wanted_factor
= 3;
1474 /* We should only redo collect_merge_info one time */
1475 assert(renames
->redo_after_renames
== 0);
1477 if (path_count_after
/ path_count_before
>= wanted_factor
) {
1478 renames
->redo_after_renames
= 1;
1479 renames
->cached_pairs_valid_side
= -1;
1481 } else if (renames
->redo_after_renames
== 2)
1482 renames
->redo_after_renames
= 0;
1486 static int collect_merge_info(struct merge_options
*opt
,
1487 struct tree
*merge_base
,
1492 struct tree_desc t
[3];
1493 struct traverse_info info
;
1495 opt
->priv
->toplevel_dir
= "";
1496 opt
->priv
->current_dir_name
= opt
->priv
->toplevel_dir
;
1497 setup_traverse_info(&info
, opt
->priv
->toplevel_dir
);
1498 info
.fn
= collect_merge_info_callback
;
1500 info
.show_all_errors
= 1;
1502 parse_tree(merge_base
);
1505 init_tree_desc(t
+ 0, merge_base
->buffer
, merge_base
->size
);
1506 init_tree_desc(t
+ 1, side1
->buffer
, side1
->size
);
1507 init_tree_desc(t
+ 2, side2
->buffer
, side2
->size
);
1509 trace2_region_enter("merge", "traverse_trees", opt
->repo
);
1510 ret
= traverse_trees(NULL
, 3, t
, &info
);
1512 ret
= handle_deferred_entries(opt
, &info
);
1513 trace2_region_leave("merge", "traverse_trees", opt
->repo
);
1518 /*** Function Grouping: functions related to threeway content merges ***/
1520 static int find_first_merges(struct repository
*repo
,
1524 struct object_array
*result
)
1527 struct object_array merges
= OBJECT_ARRAY_INIT
;
1528 struct commit
*commit
;
1529 int contains_another
;
1531 char merged_revision
[GIT_MAX_HEXSZ
+ 2];
1532 const char *rev_args
[] = { "rev-list", "--merges", "--ancestry-path",
1533 "--all", merged_revision
, NULL
};
1534 struct rev_info revs
;
1535 struct setup_revision_opt rev_opts
;
1537 memset(result
, 0, sizeof(struct object_array
));
1538 memset(&rev_opts
, 0, sizeof(rev_opts
));
1540 /* get all revisions that merge commit a */
1541 xsnprintf(merged_revision
, sizeof(merged_revision
), "^%s",
1542 oid_to_hex(&a
->object
.oid
));
1543 repo_init_revisions(repo
, &revs
, NULL
);
1544 rev_opts
.submodule
= path
;
1545 /* FIXME: can't handle linked worktrees in submodules yet */
1546 revs
.single_worktree
= path
!= NULL
;
1547 setup_revisions(ARRAY_SIZE(rev_args
)-1, rev_args
, &revs
, &rev_opts
);
1549 /* save all revisions from the above list that contain b */
1550 if (prepare_revision_walk(&revs
))
1551 die("revision walk setup failed");
1552 while ((commit
= get_revision(&revs
)) != NULL
) {
1553 struct object
*o
= &(commit
->object
);
1554 if (in_merge_bases(b
, commit
))
1555 add_object_array(o
, NULL
, &merges
);
1557 reset_revision_walk();
1559 /* Now we've got all merges that contain a and b. Prune all
1560 * merges that contain another found merge and save them in
1563 for (i
= 0; i
< merges
.nr
; i
++) {
1564 struct commit
*m1
= (struct commit
*) merges
.objects
[i
].item
;
1566 contains_another
= 0;
1567 for (j
= 0; j
< merges
.nr
; j
++) {
1568 struct commit
*m2
= (struct commit
*) merges
.objects
[j
].item
;
1569 if (i
!= j
&& in_merge_bases(m2
, m1
)) {
1570 contains_another
= 1;
1575 if (!contains_another
)
1576 add_object_array(merges
.objects
[i
].item
, NULL
, result
);
1579 object_array_clear(&merges
);
1583 static int merge_submodule(struct merge_options
*opt
,
1585 const struct object_id
*o
,
1586 const struct object_id
*a
,
1587 const struct object_id
*b
,
1588 struct object_id
*result
)
1590 struct commit
*commit_o
, *commit_a
, *commit_b
;
1592 struct object_array merges
;
1593 struct strbuf sb
= STRBUF_INIT
;
1596 int search
= !opt
->priv
->call_depth
;
1598 /* store fallback answer in result in case we fail */
1599 oidcpy(result
, opt
->priv
->call_depth
? o
: a
);
1601 /* we can not handle deletion conflicts */
1609 if (add_submodule_odb(path
)) {
1610 path_msg(opt
, path
, 0,
1611 _("Failed to merge submodule %s (not checked out)"),
1616 if (!(commit_o
= lookup_commit_reference(opt
->repo
, o
)) ||
1617 !(commit_a
= lookup_commit_reference(opt
->repo
, a
)) ||
1618 !(commit_b
= lookup_commit_reference(opt
->repo
, b
))) {
1619 path_msg(opt
, path
, 0,
1620 _("Failed to merge submodule %s (commits not present)"),
1625 /* check whether both changes are forward */
1626 if (!in_merge_bases(commit_o
, commit_a
) ||
1627 !in_merge_bases(commit_o
, commit_b
)) {
1628 path_msg(opt
, path
, 0,
1629 _("Failed to merge submodule %s "
1630 "(commits don't follow merge-base)"),
1635 /* Case #1: a is contained in b or vice versa */
1636 if (in_merge_bases(commit_a
, commit_b
)) {
1638 path_msg(opt
, path
, 1,
1639 _("Note: Fast-forwarding submodule %s to %s"),
1640 path
, oid_to_hex(b
));
1643 if (in_merge_bases(commit_b
, commit_a
)) {
1645 path_msg(opt
, path
, 1,
1646 _("Note: Fast-forwarding submodule %s to %s"),
1647 path
, oid_to_hex(a
));
1652 * Case #2: There are one or more merges that contain a and b in
1653 * the submodule. If there is only one, then present it as a
1654 * suggestion to the user, but leave it marked unmerged so the
1655 * user needs to confirm the resolution.
1658 /* Skip the search if makes no sense to the calling context. */
1662 /* find commit which merges them */
1663 parent_count
= find_first_merges(opt
->repo
, path
, commit_a
, commit_b
,
1665 switch (parent_count
) {
1667 path_msg(opt
, path
, 0, _("Failed to merge submodule %s"), path
);
1671 format_commit(&sb
, 4,
1672 (struct commit
*)merges
.objects
[0].item
);
1673 path_msg(opt
, path
, 0,
1674 _("Failed to merge submodule %s, but a possible merge "
1675 "resolution exists:\n%s\n"),
1677 path_msg(opt
, path
, 1,
1678 _("If this is correct simply add it to the index "
1681 " git update-index --cacheinfo 160000 %s \"%s\"\n\n"
1682 "which will accept this suggestion.\n"),
1683 oid_to_hex(&merges
.objects
[0].item
->oid
), path
);
1684 strbuf_release(&sb
);
1687 for (i
= 0; i
< merges
.nr
; i
++)
1688 format_commit(&sb
, 4,
1689 (struct commit
*)merges
.objects
[i
].item
);
1690 path_msg(opt
, path
, 0,
1691 _("Failed to merge submodule %s, but multiple "
1692 "possible merges exist:\n%s"), path
, sb
.buf
);
1693 strbuf_release(&sb
);
1696 object_array_clear(&merges
);
1700 static void initialize_attr_index(struct merge_options
*opt
)
1703 * The renormalize_buffer() functions require attributes, and
1704 * annoyingly those can only be read from the working tree or from
1705 * an index_state. merge-ort doesn't have an index_state, so we
1706 * generate a fake one containing only attribute information.
1708 struct merged_info
*mi
;
1709 struct index_state
*attr_index
= &opt
->priv
->attr_index
;
1710 struct cache_entry
*ce
;
1712 attr_index
->initialized
= 1;
1714 if (!opt
->renormalize
)
1717 mi
= strmap_get(&opt
->priv
->paths
, GITATTRIBUTES_FILE
);
1722 int len
= strlen(GITATTRIBUTES_FILE
);
1723 ce
= make_empty_cache_entry(attr_index
, len
);
1724 ce
->ce_mode
= create_ce_mode(mi
->result
.mode
);
1725 ce
->ce_flags
= create_ce_flags(0);
1726 ce
->ce_namelen
= len
;
1727 oidcpy(&ce
->oid
, &mi
->result
.oid
);
1728 memcpy(ce
->name
, GITATTRIBUTES_FILE
, len
);
1729 add_index_entry(attr_index
, ce
,
1730 ADD_CACHE_OK_TO_ADD
| ADD_CACHE_OK_TO_REPLACE
);
1731 get_stream_filter(attr_index
, GITATTRIBUTES_FILE
, &ce
->oid
);
1734 struct conflict_info
*ci
;
1736 ASSIGN_AND_VERIFY_CI(ci
, mi
);
1737 for (stage
= 0; stage
< 3; stage
++) {
1738 unsigned stage_mask
= (1 << stage
);
1740 if (!(ci
->filemask
& stage_mask
))
1742 len
= strlen(GITATTRIBUTES_FILE
);
1743 ce
= make_empty_cache_entry(attr_index
, len
);
1744 ce
->ce_mode
= create_ce_mode(ci
->stages
[stage
].mode
);
1745 ce
->ce_flags
= create_ce_flags(stage
);
1746 ce
->ce_namelen
= len
;
1747 oidcpy(&ce
->oid
, &ci
->stages
[stage
].oid
);
1748 memcpy(ce
->name
, GITATTRIBUTES_FILE
, len
);
1749 add_index_entry(attr_index
, ce
,
1750 ADD_CACHE_OK_TO_ADD
| ADD_CACHE_OK_TO_REPLACE
);
1751 get_stream_filter(attr_index
, GITATTRIBUTES_FILE
,
1757 static int merge_3way(struct merge_options
*opt
,
1759 const struct object_id
*o
,
1760 const struct object_id
*a
,
1761 const struct object_id
*b
,
1762 const char *pathnames
[3],
1763 const int extra_marker_size
,
1764 mmbuffer_t
*result_buf
)
1766 mmfile_t orig
, src1
, src2
;
1767 struct ll_merge_options ll_opts
= {0};
1768 char *base
, *name1
, *name2
;
1771 if (!opt
->priv
->attr_index
.initialized
)
1772 initialize_attr_index(opt
);
1774 ll_opts
.renormalize
= opt
->renormalize
;
1775 ll_opts
.extra_marker_size
= extra_marker_size
;
1776 ll_opts
.xdl_opts
= opt
->xdl_opts
;
1778 if (opt
->priv
->call_depth
) {
1779 ll_opts
.virtual_ancestor
= 1;
1780 ll_opts
.variant
= 0;
1782 switch (opt
->recursive_variant
) {
1783 case MERGE_VARIANT_OURS
:
1784 ll_opts
.variant
= XDL_MERGE_FAVOR_OURS
;
1786 case MERGE_VARIANT_THEIRS
:
1787 ll_opts
.variant
= XDL_MERGE_FAVOR_THEIRS
;
1790 ll_opts
.variant
= 0;
1795 assert(pathnames
[0] && pathnames
[1] && pathnames
[2] && opt
->ancestor
);
1796 if (pathnames
[0] == pathnames
[1] && pathnames
[1] == pathnames
[2]) {
1797 base
= mkpathdup("%s", opt
->ancestor
);
1798 name1
= mkpathdup("%s", opt
->branch1
);
1799 name2
= mkpathdup("%s", opt
->branch2
);
1801 base
= mkpathdup("%s:%s", opt
->ancestor
, pathnames
[0]);
1802 name1
= mkpathdup("%s:%s", opt
->branch1
, pathnames
[1]);
1803 name2
= mkpathdup("%s:%s", opt
->branch2
, pathnames
[2]);
1806 read_mmblob(&orig
, o
);
1807 read_mmblob(&src1
, a
);
1808 read_mmblob(&src2
, b
);
1810 merge_status
= ll_merge(result_buf
, path
, &orig
, base
,
1811 &src1
, name1
, &src2
, name2
,
1812 &opt
->priv
->attr_index
, &ll_opts
);
1820 return merge_status
;
1823 static int handle_content_merge(struct merge_options
*opt
,
1825 const struct version_info
*o
,
1826 const struct version_info
*a
,
1827 const struct version_info
*b
,
1828 const char *pathnames
[3],
1829 const int extra_marker_size
,
1830 struct version_info
*result
)
1833 * path is the target location where we want to put the file, and
1834 * is used to determine any normalization rules in ll_merge.
1836 * The normal case is that path and all entries in pathnames are
1837 * identical, though renames can affect which path we got one of
1838 * the three blobs to merge on various sides of history.
1840 * extra_marker_size is the amount to extend conflict markers in
1841 * ll_merge; this is neeed if we have content merges of content
1842 * merges, which happens for example with rename/rename(2to1) and
1843 * rename/add conflicts.
1848 * handle_content_merge() needs both files to be of the same type, i.e.
1849 * both files OR both submodules OR both symlinks. Conflicting types
1850 * needs to be handled elsewhere.
1852 assert((S_IFMT
& a
->mode
) == (S_IFMT
& b
->mode
));
1855 if (a
->mode
== b
->mode
|| a
->mode
== o
->mode
)
1856 result
->mode
= b
->mode
;
1858 /* must be the 100644/100755 case */
1859 assert(S_ISREG(a
->mode
));
1860 result
->mode
= a
->mode
;
1861 clean
= (b
->mode
== o
->mode
);
1863 * FIXME: If opt->priv->call_depth && !clean, then we really
1864 * should not make result->mode match either a->mode or
1865 * b->mode; that causes t6036 "check conflicting mode for
1866 * regular file" to fail. It would be best to use some other
1867 * mode, but we'll confuse all kinds of stuff if we use one
1868 * where S_ISREG(result->mode) isn't true, and if we use
1869 * something like 0100666, then tree-walk.c's calls to
1870 * canon_mode() will just normalize that to 100644 for us and
1871 * thus not solve anything.
1873 * Figure out if there's some kind of way we can work around
1879 * Trivial oid merge.
1881 * Note: While one might assume that the next four lines would
1882 * be unnecessary due to the fact that match_mask is often
1883 * setup and already handled, renames don't always take care
1886 if (oideq(&a
->oid
, &b
->oid
) || oideq(&a
->oid
, &o
->oid
))
1887 oidcpy(&result
->oid
, &b
->oid
);
1888 else if (oideq(&b
->oid
, &o
->oid
))
1889 oidcpy(&result
->oid
, &a
->oid
);
1891 /* Remaining rules depend on file vs. submodule vs. symlink. */
1892 else if (S_ISREG(a
->mode
)) {
1893 mmbuffer_t result_buf
;
1894 int ret
= 0, merge_status
;
1898 * If 'o' is different type, treat it as null so we do a
1901 two_way
= ((S_IFMT
& o
->mode
) != (S_IFMT
& a
->mode
));
1903 merge_status
= merge_3way(opt
, path
,
1904 two_way
? null_oid() : &o
->oid
,
1906 pathnames
, extra_marker_size
,
1909 if ((merge_status
< 0) || !result_buf
.ptr
)
1910 ret
= err(opt
, _("Failed to execute internal merge"));
1913 write_object_file(result_buf
.ptr
, result_buf
.size
,
1914 blob_type
, &result
->oid
))
1915 ret
= err(opt
, _("Unable to add %s to database"),
1918 free(result_buf
.ptr
);
1921 clean
&= (merge_status
== 0);
1922 path_msg(opt
, path
, 1, _("Auto-merging %s"), path
);
1923 } else if (S_ISGITLINK(a
->mode
)) {
1924 int two_way
= ((S_IFMT
& o
->mode
) != (S_IFMT
& a
->mode
));
1925 clean
= merge_submodule(opt
, pathnames
[0],
1926 two_way
? null_oid() : &o
->oid
,
1927 &a
->oid
, &b
->oid
, &result
->oid
);
1928 if (opt
->priv
->call_depth
&& two_way
&& !clean
) {
1929 result
->mode
= o
->mode
;
1930 oidcpy(&result
->oid
, &o
->oid
);
1932 } else if (S_ISLNK(a
->mode
)) {
1933 if (opt
->priv
->call_depth
) {
1935 result
->mode
= o
->mode
;
1936 oidcpy(&result
->oid
, &o
->oid
);
1938 switch (opt
->recursive_variant
) {
1939 case MERGE_VARIANT_NORMAL
:
1941 oidcpy(&result
->oid
, &a
->oid
);
1943 case MERGE_VARIANT_OURS
:
1944 oidcpy(&result
->oid
, &a
->oid
);
1946 case MERGE_VARIANT_THEIRS
:
1947 oidcpy(&result
->oid
, &b
->oid
);
1952 BUG("unsupported object type in the tree: %06o for %s",
1958 /*** Function Grouping: functions related to detect_and_process_renames(), ***
1959 *** which are split into directory and regular rename detection sections. ***/
1961 /*** Function Grouping: functions related to directory rename detection ***/
1963 struct collision_info
{
1964 struct string_list source_files
;
1965 unsigned reported_already
:1;
1969 * Return a new string that replaces the beginning portion (which matches
1970 * rename_info->key), with rename_info->util.new_dir. In perl-speak:
1971 * new_path_name = (old_path =~ s/rename_info->key/rename_info->value/);
1973 * Caller must ensure that old_path starts with rename_info->key + '/'.
1975 static char *apply_dir_rename(struct strmap_entry
*rename_info
,
1976 const char *old_path
)
1978 struct strbuf new_path
= STRBUF_INIT
;
1979 const char *old_dir
= rename_info
->key
;
1980 const char *new_dir
= rename_info
->value
;
1981 int oldlen
, newlen
, new_dir_len
;
1983 oldlen
= strlen(old_dir
);
1984 if (*new_dir
== '\0')
1986 * If someone renamed/merged a subdirectory into the root
1987 * directory (e.g. 'some/subdir' -> ''), then we want to
1990 * as the rename; we need to make old_path + oldlen advance
1991 * past the '/' character.
1994 new_dir_len
= strlen(new_dir
);
1995 newlen
= new_dir_len
+ (strlen(old_path
) - oldlen
) + 1;
1996 strbuf_grow(&new_path
, newlen
);
1997 strbuf_add(&new_path
, new_dir
, new_dir_len
);
1998 strbuf_addstr(&new_path
, &old_path
[oldlen
]);
2000 return strbuf_detach(&new_path
, NULL
);
2003 static int path_in_way(struct strmap
*paths
, const char *path
, unsigned side_mask
)
2005 struct merged_info
*mi
= strmap_get(paths
, path
);
2006 struct conflict_info
*ci
;
2009 INITIALIZE_CI(ci
, mi
);
2010 return mi
->clean
|| (side_mask
& (ci
->filemask
| ci
->dirmask
));
2014 * See if there is a directory rename for path, and if there are any file
2015 * level conflicts on the given side for the renamed location. If there is
2016 * a rename and there are no conflicts, return the new name. Otherwise,
2019 static char *handle_path_level_conflicts(struct merge_options
*opt
,
2021 unsigned side_index
,
2022 struct strmap_entry
*rename_info
,
2023 struct strmap
*collisions
)
2025 char *new_path
= NULL
;
2026 struct collision_info
*c_info
;
2028 struct strbuf collision_paths
= STRBUF_INIT
;
2031 * entry has the mapping of old directory name to new directory name
2032 * that we want to apply to path.
2034 new_path
= apply_dir_rename(rename_info
, path
);
2036 BUG("Failed to apply directory rename!");
2039 * The caller needs to have ensured that it has pre-populated
2040 * collisions with all paths that map to new_path. Do a quick check
2041 * to ensure that's the case.
2043 c_info
= strmap_get(collisions
, new_path
);
2045 BUG("c_info is NULL");
2048 * Check for one-sided add/add/.../add conflicts, i.e.
2049 * where implicit renames from the other side doing
2050 * directory rename(s) can affect this side of history
2051 * to put multiple paths into the same location. Warn
2052 * and bail on directory renames for such paths.
2054 if (c_info
->reported_already
) {
2056 } else if (path_in_way(&opt
->priv
->paths
, new_path
, 1 << side_index
)) {
2057 c_info
->reported_already
= 1;
2058 strbuf_add_separated_string_list(&collision_paths
, ", ",
2059 &c_info
->source_files
);
2060 path_msg(opt
, new_path
, 0,
2061 _("CONFLICT (implicit dir rename): Existing file/dir "
2062 "at %s in the way of implicit directory rename(s) "
2063 "putting the following path(s) there: %s."),
2064 new_path
, collision_paths
.buf
);
2066 } else if (c_info
->source_files
.nr
> 1) {
2067 c_info
->reported_already
= 1;
2068 strbuf_add_separated_string_list(&collision_paths
, ", ",
2069 &c_info
->source_files
);
2070 path_msg(opt
, new_path
, 0,
2071 _("CONFLICT (implicit dir rename): Cannot map more "
2072 "than one path to %s; implicit directory renames "
2073 "tried to put these paths there: %s"),
2074 new_path
, collision_paths
.buf
);
2078 /* Free memory we no longer need */
2079 strbuf_release(&collision_paths
);
2080 if (!clean
&& new_path
) {
2088 static void get_provisional_directory_renames(struct merge_options
*opt
,
2092 struct hashmap_iter iter
;
2093 struct strmap_entry
*entry
;
2094 struct rename_info
*renames
= &opt
->priv
->renames
;
2098 * dir_rename_count: old_directory -> {new_directory -> count}
2100 * dir_renames: old_directory -> best_new_directory
2101 * where best_new_directory is the one with the unique highest count.
2103 strmap_for_each_entry(&renames
->dir_rename_count
[side
], &iter
, entry
) {
2104 const char *source_dir
= entry
->key
;
2105 struct strintmap
*counts
= entry
->value
;
2106 struct hashmap_iter count_iter
;
2107 struct strmap_entry
*count_entry
;
2110 const char *best
= NULL
;
2112 strintmap_for_each_entry(counts
, &count_iter
, count_entry
) {
2113 const char *target_dir
= count_entry
->key
;
2114 intptr_t count
= (intptr_t)count_entry
->value
;
2118 else if (count
> max
) {
2127 if (bad_max
== max
) {
2128 path_msg(opt
, source_dir
, 0,
2129 _("CONFLICT (directory rename split): "
2130 "Unclear where to rename %s to; it was "
2131 "renamed to multiple other directories, with "
2132 "no destination getting a majority of the "
2137 strmap_put(&renames
->dir_renames
[side
],
2138 source_dir
, (void*)best
);
2143 static void handle_directory_level_conflicts(struct merge_options
*opt
)
2145 struct hashmap_iter iter
;
2146 struct strmap_entry
*entry
;
2147 struct string_list duplicated
= STRING_LIST_INIT_NODUP
;
2148 struct rename_info
*renames
= &opt
->priv
->renames
;
2149 struct strmap
*side1_dir_renames
= &renames
->dir_renames
[MERGE_SIDE1
];
2150 struct strmap
*side2_dir_renames
= &renames
->dir_renames
[MERGE_SIDE2
];
2153 strmap_for_each_entry(side1_dir_renames
, &iter
, entry
) {
2154 if (strmap_contains(side2_dir_renames
, entry
->key
))
2155 string_list_append(&duplicated
, entry
->key
);
2158 for (i
= 0; i
< duplicated
.nr
; i
++) {
2159 strmap_remove(side1_dir_renames
, duplicated
.items
[i
].string
, 0);
2160 strmap_remove(side2_dir_renames
, duplicated
.items
[i
].string
, 0);
2162 string_list_clear(&duplicated
, 0);
2165 static struct strmap_entry
*check_dir_renamed(const char *path
,
2166 struct strmap
*dir_renames
)
2168 char *temp
= xstrdup(path
);
2170 struct strmap_entry
*e
= NULL
;
2172 while ((end
= strrchr(temp
, '/'))) {
2174 e
= strmap_get_entry(dir_renames
, temp
);
2182 static void compute_collisions(struct strmap
*collisions
,
2183 struct strmap
*dir_renames
,
2184 struct diff_queue_struct
*pairs
)
2188 strmap_init_with_options(collisions
, NULL
, 0);
2189 if (strmap_empty(dir_renames
))
2193 * Multiple files can be mapped to the same path due to directory
2194 * renames done by the other side of history. Since that other
2195 * side of history could have merged multiple directories into one,
2196 * if our side of history added the same file basename to each of
2197 * those directories, then all N of them would get implicitly
2198 * renamed by the directory rename detection into the same path,
2199 * and we'd get an add/add/.../add conflict, and all those adds
2200 * from *this* side of history. This is not representable in the
2201 * index, and users aren't going to easily be able to make sense of
2202 * it. So we need to provide a good warning about what's
2203 * happening, and fall back to no-directory-rename detection
2204 * behavior for those paths.
2206 * See testcases 9e and all of section 5 from t6043 for examples.
2208 for (i
= 0; i
< pairs
->nr
; ++i
) {
2209 struct strmap_entry
*rename_info
;
2210 struct collision_info
*collision_info
;
2212 struct diff_filepair
*pair
= pairs
->queue
[i
];
2214 if (pair
->status
!= 'A' && pair
->status
!= 'R')
2216 rename_info
= check_dir_renamed(pair
->two
->path
, dir_renames
);
2220 new_path
= apply_dir_rename(rename_info
, pair
->two
->path
);
2222 collision_info
= strmap_get(collisions
, new_path
);
2223 if (collision_info
) {
2226 CALLOC_ARRAY(collision_info
, 1);
2227 string_list_init_nodup(&collision_info
->source_files
);
2228 strmap_put(collisions
, new_path
, collision_info
);
2230 string_list_insert(&collision_info
->source_files
,
2235 static char *check_for_directory_rename(struct merge_options
*opt
,
2237 unsigned side_index
,
2238 struct strmap
*dir_renames
,
2239 struct strmap
*dir_rename_exclusions
,
2240 struct strmap
*collisions
,
2243 char *new_path
= NULL
;
2244 struct strmap_entry
*rename_info
;
2245 struct strmap_entry
*otherinfo
= NULL
;
2246 const char *new_dir
;
2248 if (strmap_empty(dir_renames
))
2250 rename_info
= check_dir_renamed(path
, dir_renames
);
2253 /* old_dir = rename_info->key; */
2254 new_dir
= rename_info
->value
;
2257 * This next part is a little weird. We do not want to do an
2258 * implicit rename into a directory we renamed on our side, because
2259 * that will result in a spurious rename/rename(1to2) conflict. An
2261 * Base commit: dumbdir/afile, otherdir/bfile
2262 * Side 1: smrtdir/afile, otherdir/bfile
2263 * Side 2: dumbdir/afile, dumbdir/bfile
2264 * Here, while working on Side 1, we could notice that otherdir was
2265 * renamed/merged to dumbdir, and change the diff_filepair for
2266 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2267 * 2 will notice the rename from dumbdir to smrtdir, and do the
2268 * transitive rename to move it from dumbdir/bfile to
2269 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2270 * smrtdir, a rename/rename(1to2) conflict. We really just want
2271 * the file to end up in smrtdir. And the way to achieve that is
2272 * to not let Side1 do the rename to dumbdir, since we know that is
2273 * the source of one of our directory renames.
2275 * That's why otherinfo and dir_rename_exclusions is here.
2277 * As it turns out, this also prevents N-way transient rename
2278 * confusion; See testcases 9c and 9d of t6043.
2280 otherinfo
= strmap_get_entry(dir_rename_exclusions
, new_dir
);
2282 path_msg(opt
, rename_info
->key
, 1,
2283 _("WARNING: Avoiding applying %s -> %s rename "
2284 "to %s, because %s itself was renamed."),
2285 rename_info
->key
, new_dir
, path
, new_dir
);
2289 new_path
= handle_path_level_conflicts(opt
, path
, side_index
,
2290 rename_info
, collisions
);
2291 *clean_merge
&= (new_path
!= NULL
);
2296 static void apply_directory_rename_modifications(struct merge_options
*opt
,
2297 struct diff_filepair
*pair
,
2301 * The basic idea is to get the conflict_info from opt->priv->paths
2302 * at old path, and insert it into new_path; basically just this:
2303 * ci = strmap_get(&opt->priv->paths, old_path);
2304 * strmap_remove(&opt->priv->paths, old_path, 0);
2305 * strmap_put(&opt->priv->paths, new_path, ci);
2306 * However, there are some factors complicating this:
2307 * - opt->priv->paths may already have an entry at new_path
2308 * - Each ci tracks its containing directory, so we need to
2310 * - If another ci has the same containing directory, then
2311 * the two char*'s MUST point to the same location. See the
2312 * comment in struct merged_info. strcmp equality is not
2313 * enough; we need pointer equality.
2314 * - opt->priv->paths must hold the parent directories of any
2315 * entries that are added. So, if this directory rename
2316 * causes entirely new directories, we must recursively add
2317 * parent directories.
2318 * - For each parent directory added to opt->priv->paths, we
2319 * also need to get its parent directory stored in its
2320 * conflict_info->merged.directory_name with all the same
2321 * requirements about pointer equality.
2323 struct string_list dirs_to_insert
= STRING_LIST_INIT_NODUP
;
2324 struct conflict_info
*ci
, *new_ci
;
2325 struct strmap_entry
*entry
;
2326 const char *branch_with_new_path
, *branch_with_dir_rename
;
2327 const char *old_path
= pair
->two
->path
;
2328 const char *parent_name
;
2329 const char *cur_path
;
2332 entry
= strmap_get_entry(&opt
->priv
->paths
, old_path
);
2333 old_path
= entry
->key
;
2337 /* Find parent directories missing from opt->priv->paths */
2338 cur_path
= new_path
;
2340 /* Find the parent directory of cur_path */
2341 char *last_slash
= strrchr(cur_path
, '/');
2343 parent_name
= xstrndup(cur_path
, last_slash
- cur_path
);
2345 parent_name
= opt
->priv
->toplevel_dir
;
2349 /* Look it up in opt->priv->paths */
2350 entry
= strmap_get_entry(&opt
->priv
->paths
, parent_name
);
2352 free((char*)parent_name
);
2353 parent_name
= entry
->key
; /* reuse known pointer */
2357 /* Record this is one of the directories we need to insert */
2358 string_list_append(&dirs_to_insert
, parent_name
);
2359 cur_path
= parent_name
;
2362 /* Traverse dirs_to_insert and insert them into opt->priv->paths */
2363 for (i
= dirs_to_insert
.nr
-1; i
>= 0; --i
) {
2364 struct conflict_info
*dir_ci
;
2365 char *cur_dir
= dirs_to_insert
.items
[i
].string
;
2367 CALLOC_ARRAY(dir_ci
, 1);
2369 dir_ci
->merged
.directory_name
= parent_name
;
2370 len
= strlen(parent_name
);
2371 /* len+1 because of trailing '/' character */
2372 dir_ci
->merged
.basename_offset
= (len
> 0 ? len
+1 : len
);
2373 dir_ci
->dirmask
= ci
->filemask
;
2374 strmap_put(&opt
->priv
->paths
, cur_dir
, dir_ci
);
2376 parent_name
= cur_dir
;
2380 * We are removing old_path from opt->priv->paths. old_path also will
2381 * eventually need to be freed, but it may still be used by e.g.
2382 * ci->pathnames. So, store it in another string-list for now.
2384 string_list_append(&opt
->priv
->paths_to_free
, old_path
);
2386 assert(ci
->filemask
== 2 || ci
->filemask
== 4);
2387 assert(ci
->dirmask
== 0);
2388 strmap_remove(&opt
->priv
->paths
, old_path
, 0);
2390 branch_with_new_path
= (ci
->filemask
== 2) ? opt
->branch1
: opt
->branch2
;
2391 branch_with_dir_rename
= (ci
->filemask
== 2) ? opt
->branch2
: opt
->branch1
;
2393 /* Now, finally update ci and stick it into opt->priv->paths */
2394 ci
->merged
.directory_name
= parent_name
;
2395 len
= strlen(parent_name
);
2396 ci
->merged
.basename_offset
= (len
> 0 ? len
+1 : len
);
2397 new_ci
= strmap_get(&opt
->priv
->paths
, new_path
);
2399 /* Place ci back into opt->priv->paths, but at new_path */
2400 strmap_put(&opt
->priv
->paths
, new_path
, ci
);
2404 /* A few sanity checks */
2406 assert(ci
->filemask
== 2 || ci
->filemask
== 4);
2407 assert((new_ci
->filemask
& ci
->filemask
) == 0);
2408 assert(!new_ci
->merged
.clean
);
2410 /* Copy stuff from ci into new_ci */
2411 new_ci
->filemask
|= ci
->filemask
;
2412 if (new_ci
->dirmask
)
2413 new_ci
->df_conflict
= 1;
2414 index
= (ci
->filemask
>> 1);
2415 new_ci
->pathnames
[index
] = ci
->pathnames
[index
];
2416 new_ci
->stages
[index
].mode
= ci
->stages
[index
].mode
;
2417 oidcpy(&new_ci
->stages
[index
].oid
, &ci
->stages
[index
].oid
);
2423 if (opt
->detect_directory_renames
== MERGE_DIRECTORY_RENAMES_TRUE
) {
2424 /* Notify user of updated path */
2425 if (pair
->status
== 'A')
2426 path_msg(opt
, new_path
, 1,
2427 _("Path updated: %s added in %s inside a "
2428 "directory that was renamed in %s; moving "
2430 old_path
, branch_with_new_path
,
2431 branch_with_dir_rename
, new_path
);
2433 path_msg(opt
, new_path
, 1,
2434 _("Path updated: %s renamed to %s in %s, "
2435 "inside a directory that was renamed in %s; "
2436 "moving it to %s."),
2437 pair
->one
->path
, old_path
, branch_with_new_path
,
2438 branch_with_dir_rename
, new_path
);
2441 * opt->detect_directory_renames has the value
2442 * MERGE_DIRECTORY_RENAMES_CONFLICT, so mark these as conflicts.
2444 ci
->path_conflict
= 1;
2445 if (pair
->status
== 'A')
2446 path_msg(opt
, new_path
, 0,
2447 _("CONFLICT (file location): %s added in %s "
2448 "inside a directory that was renamed in %s, "
2449 "suggesting it should perhaps be moved to "
2451 old_path
, branch_with_new_path
,
2452 branch_with_dir_rename
, new_path
);
2454 path_msg(opt
, new_path
, 0,
2455 _("CONFLICT (file location): %s renamed to %s "
2456 "in %s, inside a directory that was renamed "
2457 "in %s, suggesting it should perhaps be "
2459 pair
->one
->path
, old_path
, branch_with_new_path
,
2460 branch_with_dir_rename
, new_path
);
2464 * Finally, record the new location.
2466 pair
->two
->path
= new_path
;
2469 /*** Function Grouping: functions related to regular rename detection ***/
2471 static int process_renames(struct merge_options
*opt
,
2472 struct diff_queue_struct
*renames
)
2474 int clean_merge
= 1, i
;
2476 for (i
= 0; i
< renames
->nr
; ++i
) {
2477 const char *oldpath
= NULL
, *newpath
;
2478 struct diff_filepair
*pair
= renames
->queue
[i
];
2479 struct conflict_info
*oldinfo
= NULL
, *newinfo
= NULL
;
2480 struct strmap_entry
*old_ent
, *new_ent
;
2481 unsigned int old_sidemask
;
2482 int target_index
, other_source_index
;
2483 int source_deleted
, collision
, type_changed
;
2484 const char *rename_branch
= NULL
, *delete_branch
= NULL
;
2486 old_ent
= strmap_get_entry(&opt
->priv
->paths
, pair
->one
->path
);
2487 new_ent
= strmap_get_entry(&opt
->priv
->paths
, pair
->two
->path
);
2489 oldpath
= old_ent
->key
;
2490 oldinfo
= old_ent
->value
;
2492 newpath
= pair
->two
->path
;
2494 newpath
= new_ent
->key
;
2495 newinfo
= new_ent
->value
;
2499 * If pair->one->path isn't in opt->priv->paths, that means
2500 * that either directory rename detection removed that
2501 * path, or a parent directory of oldpath was resolved and
2502 * we don't even need the rename; in either case, we can
2503 * skip it. If oldinfo->merged.clean, then the other side
2504 * of history had no changes to oldpath and we don't need
2505 * the rename and can skip it.
2507 if (!oldinfo
|| oldinfo
->merged
.clean
)
2511 * diff_filepairs have copies of pathnames, thus we have to
2512 * use standard 'strcmp()' (negated) instead of '=='.
2514 if (i
+ 1 < renames
->nr
&&
2515 !strcmp(oldpath
, renames
->queue
[i
+1]->one
->path
)) {
2516 /* Handle rename/rename(1to2) or rename/rename(1to1) */
2517 const char *pathnames
[3];
2518 struct version_info merged
;
2519 struct conflict_info
*base
, *side1
, *side2
;
2520 unsigned was_binary_blob
= 0;
2522 pathnames
[0] = oldpath
;
2523 pathnames
[1] = newpath
;
2524 pathnames
[2] = renames
->queue
[i
+1]->two
->path
;
2526 base
= strmap_get(&opt
->priv
->paths
, pathnames
[0]);
2527 side1
= strmap_get(&opt
->priv
->paths
, pathnames
[1]);
2528 side2
= strmap_get(&opt
->priv
->paths
, pathnames
[2]);
2534 if (!strcmp(pathnames
[1], pathnames
[2])) {
2535 struct rename_info
*ri
= &opt
->priv
->renames
;
2538 /* Both sides renamed the same way */
2539 assert(side1
== side2
);
2540 memcpy(&side1
->stages
[0], &base
->stages
[0],
2542 side1
->filemask
|= (1 << MERGE_BASE
);
2543 /* Mark base as resolved by removal */
2544 base
->merged
.is_null
= 1;
2545 base
->merged
.clean
= 1;
2548 * Disable remembering renames optimization;
2549 * rename/rename(1to1) is incredibly rare, and
2550 * just disabling the optimization is easier
2551 * than purging cached_pairs,
2552 * cached_target_names, and dir_rename_counts.
2554 for (j
= 0; j
< 3; j
++)
2555 ri
->merge_trees
[j
] = NULL
;
2557 /* We handled both renames, i.e. i+1 handled */
2559 /* Move to next rename */
2563 /* This is a rename/rename(1to2) */
2564 clean_merge
= handle_content_merge(opt
,
2570 1 + 2 * opt
->priv
->call_depth
,
2573 merged
.mode
== side1
->stages
[1].mode
&&
2574 oideq(&merged
.oid
, &side1
->stages
[1].oid
))
2575 was_binary_blob
= 1;
2576 memcpy(&side1
->stages
[1], &merged
, sizeof(merged
));
2577 if (was_binary_blob
) {
2579 * Getting here means we were attempting to
2580 * merge a binary blob.
2582 * Since we can't merge binaries,
2583 * handle_content_merge() just takes one
2584 * side. But we don't want to copy the
2585 * contents of one side to both paths. We
2586 * used the contents of side1 above for
2587 * side1->stages, let's use the contents of
2588 * side2 for side2->stages below.
2590 oidcpy(&merged
.oid
, &side2
->stages
[2].oid
);
2591 merged
.mode
= side2
->stages
[2].mode
;
2593 memcpy(&side2
->stages
[2], &merged
, sizeof(merged
));
2595 side1
->path_conflict
= 1;
2596 side2
->path_conflict
= 1;
2598 * TODO: For renames we normally remove the path at the
2599 * old name. It would thus seem consistent to do the
2600 * same for rename/rename(1to2) cases, but we haven't
2601 * done so traditionally and a number of the regression
2602 * tests now encode an expectation that the file is
2603 * left there at stage 1. If we ever decide to change
2604 * this, add the following two lines here:
2605 * base->merged.is_null = 1;
2606 * base->merged.clean = 1;
2607 * and remove the setting of base->path_conflict to 1.
2609 base
->path_conflict
= 1;
2610 path_msg(opt
, oldpath
, 0,
2611 _("CONFLICT (rename/rename): %s renamed to "
2612 "%s in %s and to %s in %s."),
2614 pathnames
[1], opt
->branch1
,
2615 pathnames
[2], opt
->branch2
);
2617 i
++; /* We handled both renames, i.e. i+1 handled */
2623 target_index
= pair
->score
; /* from collect_renames() */
2624 assert(target_index
== 1 || target_index
== 2);
2625 other_source_index
= 3 - target_index
;
2626 old_sidemask
= (1 << other_source_index
); /* 2 or 4 */
2627 source_deleted
= (oldinfo
->filemask
== 1);
2628 collision
= ((newinfo
->filemask
& old_sidemask
) != 0);
2629 type_changed
= !source_deleted
&&
2630 (S_ISREG(oldinfo
->stages
[other_source_index
].mode
) !=
2631 S_ISREG(newinfo
->stages
[target_index
].mode
));
2632 if (type_changed
&& collision
) {
2634 * special handling so later blocks can handle this...
2636 * if type_changed && collision are both true, then this
2637 * was really a double rename, but one side wasn't
2638 * detected due to lack of break detection. I.e.
2640 * orig: has normal file 'foo'
2641 * side1: renames 'foo' to 'bar', adds 'foo' symlink
2642 * side2: renames 'foo' to 'bar'
2643 * In this case, the foo->bar rename on side1 won't be
2644 * detected because the new symlink named 'foo' is
2645 * there and we don't do break detection. But we detect
2646 * this here because we don't want to merge the content
2647 * of the foo symlink with the foo->bar file, so we
2648 * have some logic to handle this special case. The
2649 * easiest way to do that is make 'bar' on side1 not
2650 * be considered a colliding file but the other part
2651 * of a normal rename. If the file is very different,
2652 * well we're going to get content merge conflicts
2653 * anyway so it doesn't hurt. And if the colliding
2654 * file also has a different type, that'll be handled
2655 * by the content merge logic in process_entry() too.
2657 * See also t6430, 'rename vs. rename/symlink'
2661 if (source_deleted
) {
2662 if (target_index
== 1) {
2663 rename_branch
= opt
->branch1
;
2664 delete_branch
= opt
->branch2
;
2666 rename_branch
= opt
->branch2
;
2667 delete_branch
= opt
->branch1
;
2671 assert(source_deleted
|| oldinfo
->filemask
& old_sidemask
);
2673 /* Need to check for special types of rename conflicts... */
2674 if (collision
&& !source_deleted
) {
2675 /* collision: rename/add or rename/rename(2to1) */
2676 const char *pathnames
[3];
2677 struct version_info merged
;
2679 struct conflict_info
*base
, *side1
, *side2
;
2682 pathnames
[0] = oldpath
;
2683 pathnames
[other_source_index
] = oldpath
;
2684 pathnames
[target_index
] = newpath
;
2686 base
= strmap_get(&opt
->priv
->paths
, pathnames
[0]);
2687 side1
= strmap_get(&opt
->priv
->paths
, pathnames
[1]);
2688 side2
= strmap_get(&opt
->priv
->paths
, pathnames
[2]);
2694 clean
= handle_content_merge(opt
, pair
->one
->path
,
2699 1 + 2 * opt
->priv
->call_depth
,
2702 memcpy(&newinfo
->stages
[target_index
], &merged
,
2705 path_msg(opt
, newpath
, 0,
2706 _("CONFLICT (rename involved in "
2707 "collision): rename of %s -> %s has "
2708 "content conflicts AND collides "
2709 "with another path; this may result "
2710 "in nested conflict markers."),
2713 } else if (collision
&& source_deleted
) {
2715 * rename/add/delete or rename/rename(2to1)/delete:
2716 * since oldpath was deleted on the side that didn't
2717 * do the rename, there's not much of a content merge
2718 * we can do for the rename. oldinfo->merged.is_null
2719 * was already set, so we just leave things as-is so
2720 * they look like an add/add conflict.
2723 newinfo
->path_conflict
= 1;
2724 path_msg(opt
, newpath
, 0,
2725 _("CONFLICT (rename/delete): %s renamed "
2726 "to %s in %s, but deleted in %s."),
2727 oldpath
, newpath
, rename_branch
, delete_branch
);
2730 * a few different cases...start by copying the
2731 * existing stage(s) from oldinfo over the newinfo
2732 * and update the pathname(s).
2734 memcpy(&newinfo
->stages
[0], &oldinfo
->stages
[0],
2735 sizeof(newinfo
->stages
[0]));
2736 newinfo
->filemask
|= (1 << MERGE_BASE
);
2737 newinfo
->pathnames
[0] = oldpath
;
2739 /* rename vs. typechange */
2740 /* Mark the original as resolved by removal */
2741 memcpy(&oldinfo
->stages
[0].oid
, null_oid(),
2742 sizeof(oldinfo
->stages
[0].oid
));
2743 oldinfo
->stages
[0].mode
= 0;
2744 oldinfo
->filemask
&= 0x06;
2745 } else if (source_deleted
) {
2747 newinfo
->path_conflict
= 1;
2748 path_msg(opt
, newpath
, 0,
2749 _("CONFLICT (rename/delete): %s renamed"
2750 " to %s in %s, but deleted in %s."),
2752 rename_branch
, delete_branch
);
2755 memcpy(&newinfo
->stages
[other_source_index
],
2756 &oldinfo
->stages
[other_source_index
],
2757 sizeof(newinfo
->stages
[0]));
2758 newinfo
->filemask
|= (1 << other_source_index
);
2759 newinfo
->pathnames
[other_source_index
] = oldpath
;
2763 if (!type_changed
) {
2764 /* Mark the original as resolved by removal */
2765 oldinfo
->merged
.is_null
= 1;
2766 oldinfo
->merged
.clean
= 1;
2774 static inline int possible_side_renames(struct rename_info
*renames
,
2775 unsigned side_index
)
2777 return renames
->pairs
[side_index
].nr
> 0 &&
2778 !strintmap_empty(&renames
->relevant_sources
[side_index
]);
2781 static inline int possible_renames(struct rename_info
*renames
)
2783 return possible_side_renames(renames
, 1) ||
2784 possible_side_renames(renames
, 2) ||
2785 !strmap_empty(&renames
->cached_pairs
[1]) ||
2786 !strmap_empty(&renames
->cached_pairs
[2]);
2789 static void resolve_diffpair_statuses(struct diff_queue_struct
*q
)
2792 * A simplified version of diff_resolve_rename_copy(); would probably
2793 * just use that function but it's static...
2796 struct diff_filepair
*p
;
2798 for (i
= 0; i
< q
->nr
; ++i
) {
2800 p
->status
= 0; /* undecided */
2801 if (!DIFF_FILE_VALID(p
->one
))
2802 p
->status
= DIFF_STATUS_ADDED
;
2803 else if (!DIFF_FILE_VALID(p
->two
))
2804 p
->status
= DIFF_STATUS_DELETED
;
2805 else if (DIFF_PAIR_RENAME(p
))
2806 p
->status
= DIFF_STATUS_RENAMED
;
2810 static void prune_cached_from_relevant(struct rename_info
*renames
,
2813 /* Reason for this function described in add_pair() */
2814 struct hashmap_iter iter
;
2815 struct strmap_entry
*entry
;
2817 /* Remove from relevant_sources all entries in cached_pairs[side] */
2818 strmap_for_each_entry(&renames
->cached_pairs
[side
], &iter
, entry
) {
2819 strintmap_remove(&renames
->relevant_sources
[side
],
2822 /* Remove from relevant_sources all entries in cached_irrelevant[side] */
2823 strset_for_each_entry(&renames
->cached_irrelevant
[side
], &iter
, entry
) {
2824 strintmap_remove(&renames
->relevant_sources
[side
],
2829 static void use_cached_pairs(struct merge_options
*opt
,
2830 struct strmap
*cached_pairs
,
2831 struct diff_queue_struct
*pairs
)
2833 struct hashmap_iter iter
;
2834 struct strmap_entry
*entry
;
2837 * Add to side_pairs all entries from renames->cached_pairs[side_index].
2838 * (Info in cached_irrelevant[side_index] is not relevant here.)
2840 strmap_for_each_entry(cached_pairs
, &iter
, entry
) {
2841 struct diff_filespec
*one
, *two
;
2842 const char *old_name
= entry
->key
;
2843 const char *new_name
= entry
->value
;
2845 new_name
= old_name
;
2847 /* We don't care about oid/mode, only filenames and status */
2848 one
= alloc_filespec(old_name
);
2849 two
= alloc_filespec(new_name
);
2850 diff_queue(pairs
, one
, two
);
2851 pairs
->queue
[pairs
->nr
-1]->status
= entry
->value
? 'R' : 'D';
2855 static void cache_new_pair(struct rename_info
*renames
,
2862 new_path
= xstrdup(new_path
);
2863 old_value
= strmap_put(&renames
->cached_pairs
[side
],
2864 old_path
, new_path
);
2865 strset_add(&renames
->cached_target_names
[side
], new_path
);
2872 static void possibly_cache_new_pair(struct rename_info
*renames
,
2873 struct diff_filepair
*p
,
2877 int dir_renamed_side
= 0;
2881 * Directory renames happen on the other side of history from
2882 * the side that adds new files to the old directory.
2884 dir_renamed_side
= 3 - side
;
2886 int val
= strintmap_get(&renames
->relevant_sources
[side
],
2888 if (val
== RELEVANT_NO_MORE
) {
2889 assert(p
->status
== 'D');
2890 strset_add(&renames
->cached_irrelevant
[side
],
2897 if (p
->status
== 'D') {
2899 * If we already had this delete, we'll just set it's value
2900 * to NULL again, so no harm.
2902 strmap_put(&renames
->cached_pairs
[side
], p
->one
->path
, NULL
);
2903 } else if (p
->status
== 'R') {
2905 new_path
= p
->two
->path
;
2907 cache_new_pair(renames
, dir_renamed_side
,
2908 p
->two
->path
, new_path
, 0);
2909 cache_new_pair(renames
, side
, p
->one
->path
, new_path
, 1);
2910 } else if (p
->status
== 'A' && new_path
) {
2911 cache_new_pair(renames
, dir_renamed_side
,
2912 p
->two
->path
, new_path
, 0);
2916 static int compare_pairs(const void *a_
, const void *b_
)
2918 const struct diff_filepair
*a
= *((const struct diff_filepair
**)a_
);
2919 const struct diff_filepair
*b
= *((const struct diff_filepair
**)b_
);
2921 return strcmp(a
->one
->path
, b
->one
->path
);
2924 /* Call diffcore_rename() to update deleted/added pairs into rename pairs */
2925 static int detect_regular_renames(struct merge_options
*opt
,
2926 unsigned side_index
)
2928 struct diff_options diff_opts
;
2929 struct rename_info
*renames
= &opt
->priv
->renames
;
2931 prune_cached_from_relevant(renames
, side_index
);
2932 if (!possible_side_renames(renames
, side_index
)) {
2934 * No rename detection needed for this side, but we still need
2935 * to make sure 'adds' are marked correctly in case the other
2936 * side had directory renames.
2938 resolve_diffpair_statuses(&renames
->pairs
[side_index
]);
2942 partial_clear_dir_rename_count(&renames
->dir_rename_count
[side_index
]);
2943 repo_diff_setup(opt
->repo
, &diff_opts
);
2944 diff_opts
.flags
.recursive
= 1;
2945 diff_opts
.flags
.rename_empty
= 0;
2946 diff_opts
.detect_rename
= DIFF_DETECT_RENAME
;
2947 diff_opts
.rename_limit
= opt
->rename_limit
;
2948 if (opt
->rename_limit
<= 0)
2949 diff_opts
.rename_limit
= 1000;
2950 diff_opts
.rename_score
= opt
->rename_score
;
2951 diff_opts
.show_rename_progress
= opt
->show_rename_progress
;
2952 diff_opts
.output_format
= DIFF_FORMAT_NO_OUTPUT
;
2953 diff_setup_done(&diff_opts
);
2955 diff_queued_diff
= renames
->pairs
[side_index
];
2956 trace2_region_enter("diff", "diffcore_rename", opt
->repo
);
2957 diffcore_rename_extended(&diff_opts
,
2958 &renames
->relevant_sources
[side_index
],
2959 &renames
->dirs_removed
[side_index
],
2960 &renames
->dir_rename_count
[side_index
],
2961 &renames
->cached_pairs
[side_index
]);
2962 trace2_region_leave("diff", "diffcore_rename", opt
->repo
);
2963 resolve_diffpair_statuses(&diff_queued_diff
);
2965 if (diff_opts
.needed_rename_limit
> 0)
2966 renames
->redo_after_renames
= 0;
2967 if (diff_opts
.needed_rename_limit
> renames
->needed_limit
)
2968 renames
->needed_limit
= diff_opts
.needed_rename_limit
;
2970 renames
->pairs
[side_index
] = diff_queued_diff
;
2972 diff_opts
.output_format
= DIFF_FORMAT_NO_OUTPUT
;
2973 diff_queued_diff
.nr
= 0;
2974 diff_queued_diff
.queue
= NULL
;
2975 diff_flush(&diff_opts
);
2981 * Get information of all renames which occurred in 'side_pairs', making use
2982 * of any implicit directory renames in side_dir_renames (also making use of
2983 * implicit directory renames rename_exclusions as needed by
2984 * check_for_directory_rename()). Add all (updated) renames into result.
2986 static int collect_renames(struct merge_options
*opt
,
2987 struct diff_queue_struct
*result
,
2988 unsigned side_index
,
2989 struct strmap
*dir_renames_for_side
,
2990 struct strmap
*rename_exclusions
)
2993 struct strmap collisions
;
2994 struct diff_queue_struct
*side_pairs
;
2995 struct hashmap_iter iter
;
2996 struct strmap_entry
*entry
;
2997 struct rename_info
*renames
= &opt
->priv
->renames
;
2999 side_pairs
= &renames
->pairs
[side_index
];
3000 compute_collisions(&collisions
, dir_renames_for_side
, side_pairs
);
3002 for (i
= 0; i
< side_pairs
->nr
; ++i
) {
3003 struct diff_filepair
*p
= side_pairs
->queue
[i
];
3004 char *new_path
; /* non-NULL only with directory renames */
3006 if (p
->status
!= 'A' && p
->status
!= 'R') {
3007 possibly_cache_new_pair(renames
, p
, side_index
, NULL
);
3008 diff_free_filepair(p
);
3012 new_path
= check_for_directory_rename(opt
, p
->two
->path
,
3014 dir_renames_for_side
,
3019 possibly_cache_new_pair(renames
, p
, side_index
, new_path
);
3020 if (p
->status
!= 'R' && !new_path
) {
3021 diff_free_filepair(p
);
3026 apply_directory_rename_modifications(opt
, p
, new_path
);
3029 * p->score comes back from diffcore_rename_extended() with
3030 * the similarity of the renamed file. The similarity is
3031 * was used to determine that the two files were related
3032 * and are a rename, which we have already used, but beyond
3033 * that we have no use for the similarity. So p->score is
3034 * now irrelevant. However, process_renames() will need to
3035 * know which side of the merge this rename was associated
3036 * with, so overwrite p->score with that value.
3038 p
->score
= side_index
;
3039 result
->queue
[result
->nr
++] = p
;
3042 /* Free each value in the collisions map */
3043 strmap_for_each_entry(&collisions
, &iter
, entry
) {
3044 struct collision_info
*info
= entry
->value
;
3045 string_list_clear(&info
->source_files
, 0);
3048 * In compute_collisions(), we set collisions.strdup_strings to 0
3049 * so that we wouldn't have to make another copy of the new_path
3050 * allocated by apply_dir_rename(). But now that we've used them
3051 * and have no other references to these strings, it is time to
3054 free_strmap_strings(&collisions
);
3055 strmap_clear(&collisions
, 1);
3059 static int detect_and_process_renames(struct merge_options
*opt
,
3060 struct tree
*merge_base
,
3064 struct diff_queue_struct combined
;
3065 struct rename_info
*renames
= &opt
->priv
->renames
;
3066 int need_dir_renames
, s
, clean
= 1;
3067 unsigned detection_run
= 0;
3069 memset(&combined
, 0, sizeof(combined
));
3070 if (!possible_renames(renames
))
3073 trace2_region_enter("merge", "regular renames", opt
->repo
);
3074 detection_run
|= detect_regular_renames(opt
, MERGE_SIDE1
);
3075 detection_run
|= detect_regular_renames(opt
, MERGE_SIDE2
);
3076 if (renames
->redo_after_renames
&& detection_run
) {
3078 struct diff_filepair
*p
;
3080 /* Cache the renames, we found */
3081 for (side
= MERGE_SIDE1
; side
<= MERGE_SIDE2
; side
++) {
3082 for (i
= 0; i
< renames
->pairs
[side
].nr
; ++i
) {
3083 p
= renames
->pairs
[side
].queue
[i
];
3084 possibly_cache_new_pair(renames
, p
, side
, NULL
);
3088 /* Restart the merge with the cached renames */
3089 renames
->redo_after_renames
= 2;
3090 trace2_region_leave("merge", "regular renames", opt
->repo
);
3093 use_cached_pairs(opt
, &renames
->cached_pairs
[1], &renames
->pairs
[1]);
3094 use_cached_pairs(opt
, &renames
->cached_pairs
[2], &renames
->pairs
[2]);
3095 trace2_region_leave("merge", "regular renames", opt
->repo
);
3097 trace2_region_enter("merge", "directory renames", opt
->repo
);
3099 !opt
->priv
->call_depth
&&
3100 (opt
->detect_directory_renames
== MERGE_DIRECTORY_RENAMES_TRUE
||
3101 opt
->detect_directory_renames
== MERGE_DIRECTORY_RENAMES_CONFLICT
);
3103 if (need_dir_renames
) {
3104 get_provisional_directory_renames(opt
, MERGE_SIDE1
, &clean
);
3105 get_provisional_directory_renames(opt
, MERGE_SIDE2
, &clean
);
3106 handle_directory_level_conflicts(opt
);
3109 ALLOC_GROW(combined
.queue
,
3110 renames
->pairs
[1].nr
+ renames
->pairs
[2].nr
,
3112 clean
&= collect_renames(opt
, &combined
, MERGE_SIDE1
,
3113 &renames
->dir_renames
[2],
3114 &renames
->dir_renames
[1]);
3115 clean
&= collect_renames(opt
, &combined
, MERGE_SIDE2
,
3116 &renames
->dir_renames
[1],
3117 &renames
->dir_renames
[2]);
3118 STABLE_QSORT(combined
.queue
, combined
.nr
, compare_pairs
);
3119 trace2_region_leave("merge", "directory renames", opt
->repo
);
3121 trace2_region_enter("merge", "process renames", opt
->repo
);
3122 clean
&= process_renames(opt
, &combined
);
3123 trace2_region_leave("merge", "process renames", opt
->repo
);
3125 goto simple_cleanup
; /* collect_renames() handles some of cleanup */
3129 * Free now unneeded filepairs, which would have been handled
3130 * in collect_renames() normally but we skipped that code.
3132 for (s
= MERGE_SIDE1
; s
<= MERGE_SIDE2
; s
++) {
3133 struct diff_queue_struct
*side_pairs
;
3136 side_pairs
= &renames
->pairs
[s
];
3137 for (i
= 0; i
< side_pairs
->nr
; ++i
) {
3138 struct diff_filepair
*p
= side_pairs
->queue
[i
];
3139 diff_free_filepair(p
);
3144 /* Free memory for renames->pairs[] and combined */
3145 for (s
= MERGE_SIDE1
; s
<= MERGE_SIDE2
; s
++) {
3146 free(renames
->pairs
[s
].queue
);
3147 DIFF_QUEUE_CLEAR(&renames
->pairs
[s
]);
3151 for (i
= 0; i
< combined
.nr
; i
++)
3152 diff_free_filepair(combined
.queue
[i
]);
3153 free(combined
.queue
);
3159 /*** Function Grouping: functions related to process_entries() ***/
3161 static int sort_dirs_next_to_their_children(const char *one
, const char *two
)
3163 unsigned char c1
, c2
;
3166 * Here we only care that entries for directories appear adjacent
3167 * to and before files underneath the directory. We can achieve
3168 * that by pretending to add a trailing slash to every file and
3169 * then sorting. In other words, we do not want the natural
3174 * Instead, we want "foo" to sort as though it were "foo/", so that
3179 * To achieve this, we basically implement our own strcmp, except that
3180 * if we get to the end of either string instead of comparing NUL to
3181 * another character, we compare '/' to it.
3183 * If this unusual "sort as though '/' were appended" perplexes
3184 * you, perhaps it will help to note that this is not the final
3185 * sort. write_tree() will sort again without the trailing slash
3186 * magic, but just on paths immediately under a given tree.
3188 * The reason to not use df_name_compare directly was that it was
3189 * just too expensive (we don't have the string lengths handy), so
3190 * it was reimplemented.
3194 * NOTE: This function will never be called with two equal strings,
3195 * because it is used to sort the keys of a strmap, and strmaps have
3196 * unique keys by construction. That simplifies our c1==c2 handling
3200 while (*one
&& (*one
== *two
)) {
3205 c1
= *one
? *one
: '/';
3206 c2
= *two
? *two
: '/';
3209 /* Getting here means one is a leading directory of the other */
3210 return (*one
) ? 1 : -1;
3215 static int read_oid_strbuf(struct merge_options
*opt
,
3216 const struct object_id
*oid
,
3220 enum object_type type
;
3222 buf
= read_object_file(oid
, &type
, &size
);
3224 return err(opt
, _("cannot read object %s"), oid_to_hex(oid
));
3225 if (type
!= OBJ_BLOB
) {
3227 return err(opt
, _("object %s is not a blob"), oid_to_hex(oid
));
3229 strbuf_attach(dst
, buf
, size
, size
+ 1);
3233 static int blob_unchanged(struct merge_options
*opt
,
3234 const struct version_info
*base
,
3235 const struct version_info
*side
,
3238 struct strbuf basebuf
= STRBUF_INIT
;
3239 struct strbuf sidebuf
= STRBUF_INIT
;
3240 int ret
= 0; /* assume changed for safety */
3241 struct index_state
*idx
= &opt
->priv
->attr_index
;
3243 if (!idx
->initialized
)
3244 initialize_attr_index(opt
);
3246 if (base
->mode
!= side
->mode
)
3248 if (oideq(&base
->oid
, &side
->oid
))
3251 if (read_oid_strbuf(opt
, &base
->oid
, &basebuf
) ||
3252 read_oid_strbuf(opt
, &side
->oid
, &sidebuf
))
3255 * Note: binary | is used so that both renormalizations are
3256 * performed. Comparison can be skipped if both files are
3257 * unchanged since their sha1s have already been compared.
3259 if (renormalize_buffer(idx
, path
, basebuf
.buf
, basebuf
.len
, &basebuf
) |
3260 renormalize_buffer(idx
, path
, sidebuf
.buf
, sidebuf
.len
, &sidebuf
))
3261 ret
= (basebuf
.len
== sidebuf
.len
&&
3262 !memcmp(basebuf
.buf
, sidebuf
.buf
, basebuf
.len
));
3265 strbuf_release(&basebuf
);
3266 strbuf_release(&sidebuf
);
3270 struct directory_versions
{
3272 * versions: list of (basename -> version_info)
3274 * The basenames are in reverse lexicographic order of full pathnames,
3275 * as processed in process_entries(). This puts all entries within
3276 * a directory together, and covers the directory itself after
3277 * everything within it, allowing us to write subtrees before needing
3278 * to record information for the tree itself.
3280 struct string_list versions
;
3283 * offsets: list of (full relative path directories -> integer offsets)
3285 * Since versions contains basenames from files in multiple different
3286 * directories, we need to know which entries in versions correspond
3287 * to which directories. Values of e.g.
3291 * Would mean that entries 0-1 of versions are files in the toplevel
3292 * directory, entries 2-4 are files under src/, and the remaining
3293 * entries starting at index 5 are files under src/moduleA/.
3295 struct string_list offsets
;
3298 * last_directory: directory that previously processed file found in
3300 * last_directory starts NULL, but records the directory in which the
3301 * previous file was found within. As soon as
3302 * directory(current_file) != last_directory
3303 * then we need to start updating accounting in versions & offsets.
3304 * Note that last_directory is always the last path in "offsets" (or
3305 * NULL if "offsets" is empty) so this exists just for quick access.
3307 const char *last_directory
;
3309 /* last_directory_len: cached computation of strlen(last_directory) */
3310 unsigned last_directory_len
;
3313 static int tree_entry_order(const void *a_
, const void *b_
)
3315 const struct string_list_item
*a
= a_
;
3316 const struct string_list_item
*b
= b_
;
3318 const struct merged_info
*ami
= a
->util
;
3319 const struct merged_info
*bmi
= b
->util
;
3320 return base_name_compare(a
->string
, strlen(a
->string
), ami
->result
.mode
,
3321 b
->string
, strlen(b
->string
), bmi
->result
.mode
);
3324 static void write_tree(struct object_id
*result_oid
,
3325 struct string_list
*versions
,
3326 unsigned int offset
,
3329 size_t maxlen
= 0, extra
;
3331 struct strbuf buf
= STRBUF_INIT
;
3334 assert(offset
<= versions
->nr
);
3335 nr
= versions
->nr
- offset
;
3337 /* No need for STABLE_QSORT -- filenames must be unique */
3338 QSORT(versions
->items
+ offset
, nr
, tree_entry_order
);
3340 /* Pre-allocate some space in buf */
3341 extra
= hash_size
+ 8; /* 8: 6 for mode, 1 for space, 1 for NUL char */
3342 for (i
= 0; i
< nr
; i
++) {
3343 maxlen
+= strlen(versions
->items
[offset
+i
].string
) + extra
;
3345 strbuf_grow(&buf
, maxlen
);
3347 /* Write each entry out to buf */
3348 for (i
= 0; i
< nr
; i
++) {
3349 struct merged_info
*mi
= versions
->items
[offset
+i
].util
;
3350 struct version_info
*ri
= &mi
->result
;
3351 strbuf_addf(&buf
, "%o %s%c",
3353 versions
->items
[offset
+i
].string
, '\0');
3354 strbuf_add(&buf
, ri
->oid
.hash
, hash_size
);
3357 /* Write this object file out, and record in result_oid */
3358 write_object_file(buf
.buf
, buf
.len
, tree_type
, result_oid
);
3359 strbuf_release(&buf
);
3362 static void record_entry_for_tree(struct directory_versions
*dir_metadata
,
3364 struct merged_info
*mi
)
3366 const char *basename
;
3369 /* nothing to record */
3372 basename
= path
+ mi
->basename_offset
;
3373 assert(strchr(basename
, '/') == NULL
);
3374 string_list_append(&dir_metadata
->versions
,
3375 basename
)->util
= &mi
->result
;
3378 static void write_completed_directory(struct merge_options
*opt
,
3379 const char *new_directory_name
,
3380 struct directory_versions
*info
)
3382 const char *prev_dir
;
3383 struct merged_info
*dir_info
= NULL
;
3384 unsigned int offset
;
3387 * Some explanation of info->versions and info->offsets...
3389 * process_entries() iterates over all relevant files AND
3390 * directories in reverse lexicographic order, and calls this
3391 * function. Thus, an example of the paths that process_entries()
3392 * could operate on (along with the directories for those paths
3397 * src/moduleB/umm.c src/moduleB
3398 * src/moduleB/stuff.h src/moduleB
3399 * src/moduleB/baz.c src/moduleB
3401 * src/moduleA/foo.c src/moduleA
3402 * src/moduleA/bar.c src/moduleA
3409 * always contains the unprocessed entries and their
3410 * version_info information. For example, after the first five
3411 * entries above, info->versions would be:
3413 * xtract.c <xtract.c's version_info>
3414 * token.txt <token.txt's version_info>
3415 * umm.c <src/moduleB/umm.c's version_info>
3416 * stuff.h <src/moduleB/stuff.h's version_info>
3417 * baz.c <src/moduleB/baz.c's version_info>
3419 * Once a subdirectory is completed we remove the entries in
3420 * that subdirectory from info->versions, writing it as a tree
3421 * (write_tree()). Thus, as soon as we get to src/moduleB,
3422 * info->versions would be updated to
3424 * xtract.c <xtract.c's version_info>
3425 * token.txt <token.txt's version_info>
3426 * moduleB <src/moduleB's version_info>
3430 * helps us track which entries in info->versions correspond to
3431 * which directories. When we are N directories deep (e.g. 4
3432 * for src/modA/submod/subdir/), we have up to N+1 unprocessed
3433 * directories (+1 because of toplevel dir). Corresponding to
3434 * the info->versions example above, after processing five entries
3435 * info->offsets will be:
3440 * which is used to know that xtract.c & token.txt are from the
3441 * toplevel dirctory, while umm.c & stuff.h & baz.c are from the
3442 * src/moduleB directory. Again, following the example above,
3443 * once we need to process src/moduleB, then info->offsets is
3449 * which says that moduleB (and only moduleB so far) is in the
3452 * One unique thing to note about info->offsets here is that
3453 * "src" was not added to info->offsets until there was a path
3454 * (a file OR directory) immediately below src/ that got
3457 * Since process_entry() just appends new entries to info->versions,
3458 * write_completed_directory() only needs to do work if the next path
3459 * is in a directory that is different than the last directory found
3464 * If we are working with the same directory as the last entry, there
3465 * is no work to do. (See comments above the directory_name member of
3466 * struct merged_info for why we can use pointer comparison instead of
3469 if (new_directory_name
== info
->last_directory
)
3473 * If we are just starting (last_directory is NULL), or last_directory
3474 * is a prefix of the current directory, then we can just update
3475 * info->offsets to record the offset where we started this directory
3476 * and update last_directory to have quick access to it.
3478 if (info
->last_directory
== NULL
||
3479 !strncmp(new_directory_name
, info
->last_directory
,
3480 info
->last_directory_len
)) {
3481 uintptr_t offset
= info
->versions
.nr
;
3483 info
->last_directory
= new_directory_name
;
3484 info
->last_directory_len
= strlen(info
->last_directory
);
3486 * Record the offset into info->versions where we will
3487 * start recording basenames of paths found within
3488 * new_directory_name.
3490 string_list_append(&info
->offsets
,
3491 info
->last_directory
)->util
= (void*)offset
;
3496 * The next entry that will be processed will be within
3497 * new_directory_name. Since at this point we know that
3498 * new_directory_name is within a different directory than
3499 * info->last_directory, we have all entries for info->last_directory
3500 * in info->versions and we need to create a tree object for them.
3502 dir_info
= strmap_get(&opt
->priv
->paths
, info
->last_directory
);
3504 offset
= (uintptr_t)info
->offsets
.items
[info
->offsets
.nr
-1].util
;
3505 if (offset
== info
->versions
.nr
) {
3507 * Actually, we don't need to create a tree object in this
3508 * case. Whenever all files within a directory disappear
3509 * during the merge (e.g. unmodified on one side and
3510 * deleted on the other, or files were renamed elsewhere),
3511 * then we get here and the directory itself needs to be
3512 * omitted from its parent tree as well.
3514 dir_info
->is_null
= 1;
3517 * Write out the tree to the git object directory, and also
3518 * record the mode and oid in dir_info->result.
3520 dir_info
->is_null
= 0;
3521 dir_info
->result
.mode
= S_IFDIR
;
3522 write_tree(&dir_info
->result
.oid
, &info
->versions
, offset
,
3523 opt
->repo
->hash_algo
->rawsz
);
3527 * We've now used several entries from info->versions and one entry
3528 * from info->offsets, so we get rid of those values.
3531 info
->versions
.nr
= offset
;
3534 * Now we've taken care of the completed directory, but we need to
3535 * prepare things since future entries will be in
3536 * new_directory_name. (In particular, process_entry() will be
3537 * appending new entries to info->versions.) So, we need to make
3538 * sure new_directory_name is the last entry in info->offsets.
3540 prev_dir
= info
->offsets
.nr
== 0 ? NULL
:
3541 info
->offsets
.items
[info
->offsets
.nr
-1].string
;
3542 if (new_directory_name
!= prev_dir
) {
3543 uintptr_t c
= info
->versions
.nr
;
3544 string_list_append(&info
->offsets
,
3545 new_directory_name
)->util
= (void*)c
;
3548 /* And, of course, we need to update last_directory to match. */
3549 info
->last_directory
= new_directory_name
;
3550 info
->last_directory_len
= strlen(info
->last_directory
);
3553 /* Per entry merge function */
3554 static void process_entry(struct merge_options
*opt
,
3556 struct conflict_info
*ci
,
3557 struct directory_versions
*dir_metadata
)
3559 int df_file_index
= 0;
3562 assert(ci
->filemask
>= 0 && ci
->filemask
<= 7);
3563 /* ci->match_mask == 7 was handled in collect_merge_info_callback() */
3564 assert(ci
->match_mask
== 0 || ci
->match_mask
== 3 ||
3565 ci
->match_mask
== 5 || ci
->match_mask
== 6);
3568 record_entry_for_tree(dir_metadata
, path
, &ci
->merged
);
3569 if (ci
->filemask
== 0)
3570 /* nothing else to handle */
3572 assert(ci
->df_conflict
);
3575 if (ci
->df_conflict
&& ci
->merged
.result
.mode
== 0) {
3579 * directory no longer in the way, but we do have a file we
3580 * need to place here so we need to clean away the "directory
3581 * merges to nothing" result.
3583 ci
->df_conflict
= 0;
3584 assert(ci
->filemask
!= 0);
3585 ci
->merged
.clean
= 0;
3586 ci
->merged
.is_null
= 0;
3587 /* and we want to zero out any directory-related entries */
3588 ci
->match_mask
= (ci
->match_mask
& ~ci
->dirmask
);
3590 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
3591 if (ci
->filemask
& (1 << i
))
3593 ci
->stages
[i
].mode
= 0;
3594 oidcpy(&ci
->stages
[i
].oid
, null_oid());
3596 } else if (ci
->df_conflict
&& ci
->merged
.result
.mode
!= 0) {
3598 * This started out as a D/F conflict, and the entries in
3599 * the competing directory were not removed by the merge as
3600 * evidenced by write_completed_directory() writing a value
3601 * to ci->merged.result.mode.
3603 struct conflict_info
*new_ci
;
3605 const char *old_path
= path
;
3608 assert(ci
->merged
.result
.mode
== S_IFDIR
);
3611 * If filemask is 1, we can just ignore the file as having
3612 * been deleted on both sides. We do not want to overwrite
3613 * ci->merged.result, since it stores the tree for all the
3616 if (ci
->filemask
== 1) {
3622 * This file still exists on at least one side, and we want
3623 * the directory to remain here, so we need to move this
3624 * path to some new location.
3626 CALLOC_ARRAY(new_ci
, 1);
3627 /* We don't really want new_ci->merged.result copied, but it'll
3628 * be overwritten below so it doesn't matter. We also don't
3629 * want any directory mode/oid values copied, but we'll zero
3630 * those out immediately. We do want the rest of ci copied.
3632 memcpy(new_ci
, ci
, sizeof(*ci
));
3633 new_ci
->match_mask
= (new_ci
->match_mask
& ~new_ci
->dirmask
);
3634 new_ci
->dirmask
= 0;
3635 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
3636 if (new_ci
->filemask
& (1 << i
))
3638 /* zero out any entries related to directories */
3639 new_ci
->stages
[i
].mode
= 0;
3640 oidcpy(&new_ci
->stages
[i
].oid
, null_oid());
3644 * Find out which side this file came from; note that we
3645 * cannot just use ci->filemask, because renames could cause
3646 * the filemask to go back to 7. So we use dirmask, then
3647 * pick the opposite side's index.
3649 df_file_index
= (ci
->dirmask
& (1 << 1)) ? 2 : 1;
3650 branch
= (df_file_index
== 1) ? opt
->branch1
: opt
->branch2
;
3651 path
= unique_path(&opt
->priv
->paths
, path
, branch
);
3652 strmap_put(&opt
->priv
->paths
, path
, new_ci
);
3654 path_msg(opt
, path
, 0,
3655 _("CONFLICT (file/directory): directory in the way "
3656 "of %s from %s; moving it to %s instead."),
3657 old_path
, branch
, path
);
3660 * Zero out the filemask for the old ci. At this point, ci
3661 * was just an entry for a directory, so we don't need to
3662 * do anything more with it.
3667 * Now note that we're working on the new entry (path was
3674 * NOTE: Below there is a long switch-like if-elseif-elseif... block
3675 * which the code goes through even for the df_conflict cases
3678 if (ci
->match_mask
) {
3679 ci
->merged
.clean
= !ci
->df_conflict
&& !ci
->path_conflict
;
3680 if (ci
->match_mask
== 6) {
3681 /* stages[1] == stages[2] */
3682 ci
->merged
.result
.mode
= ci
->stages
[1].mode
;
3683 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[1].oid
);
3685 /* determine the mask of the side that didn't match */
3686 unsigned int othermask
= 7 & ~ci
->match_mask
;
3687 int side
= (othermask
== 4) ? 2 : 1;
3689 ci
->merged
.result
.mode
= ci
->stages
[side
].mode
;
3690 ci
->merged
.is_null
= !ci
->merged
.result
.mode
;
3691 if (ci
->merged
.is_null
)
3692 ci
->merged
.clean
= 1;
3693 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[side
].oid
);
3695 assert(othermask
== 2 || othermask
== 4);
3696 assert(ci
->merged
.is_null
==
3697 (ci
->filemask
== ci
->match_mask
));
3699 } else if (ci
->filemask
>= 6 &&
3700 (S_IFMT
& ci
->stages
[1].mode
) !=
3701 (S_IFMT
& ci
->stages
[2].mode
)) {
3702 /* Two different items from (file/submodule/symlink) */
3703 if (opt
->priv
->call_depth
) {
3704 /* Just use the version from the merge base */
3705 ci
->merged
.clean
= 0;
3706 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[0].oid
);
3707 ci
->merged
.result
.mode
= ci
->stages
[0].mode
;
3708 ci
->merged
.is_null
= (ci
->merged
.result
.mode
== 0);
3710 /* Handle by renaming one or both to separate paths. */
3711 unsigned o_mode
= ci
->stages
[0].mode
;
3712 unsigned a_mode
= ci
->stages
[1].mode
;
3713 unsigned b_mode
= ci
->stages
[2].mode
;
3714 struct conflict_info
*new_ci
;
3715 const char *a_path
= NULL
, *b_path
= NULL
;
3716 int rename_a
= 0, rename_b
= 0;
3718 new_ci
= xmalloc(sizeof(*new_ci
));
3720 if (S_ISREG(a_mode
))
3722 else if (S_ISREG(b_mode
))
3729 if (rename_a
&& rename_b
) {
3730 path_msg(opt
, path
, 0,
3731 _("CONFLICT (distinct types): %s had "
3732 "different types on each side; "
3733 "renamed both of them so each can "
3734 "be recorded somewhere."),
3737 path_msg(opt
, path
, 0,
3738 _("CONFLICT (distinct types): %s had "
3739 "different types on each side; "
3740 "renamed one of them so each can be "
3741 "recorded somewhere."),
3745 ci
->merged
.clean
= 0;
3746 memcpy(new_ci
, ci
, sizeof(*new_ci
));
3748 /* Put b into new_ci, removing a from stages */
3749 new_ci
->merged
.result
.mode
= ci
->stages
[2].mode
;
3750 oidcpy(&new_ci
->merged
.result
.oid
, &ci
->stages
[2].oid
);
3751 new_ci
->stages
[1].mode
= 0;
3752 oidcpy(&new_ci
->stages
[1].oid
, null_oid());
3753 new_ci
->filemask
= 5;
3754 if ((S_IFMT
& b_mode
) != (S_IFMT
& o_mode
)) {
3755 new_ci
->stages
[0].mode
= 0;
3756 oidcpy(&new_ci
->stages
[0].oid
, null_oid());
3757 new_ci
->filemask
= 4;
3760 /* Leave only a in ci, fixing stages. */
3761 ci
->merged
.result
.mode
= ci
->stages
[1].mode
;
3762 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[1].oid
);
3763 ci
->stages
[2].mode
= 0;
3764 oidcpy(&ci
->stages
[2].oid
, null_oid());
3766 if ((S_IFMT
& a_mode
) != (S_IFMT
& o_mode
)) {
3767 ci
->stages
[0].mode
= 0;
3768 oidcpy(&ci
->stages
[0].oid
, null_oid());
3772 /* Insert entries into opt->priv_paths */
3773 assert(rename_a
|| rename_b
);
3775 a_path
= unique_path(&opt
->priv
->paths
,
3776 path
, opt
->branch1
);
3777 strmap_put(&opt
->priv
->paths
, a_path
, ci
);
3781 b_path
= unique_path(&opt
->priv
->paths
,
3782 path
, opt
->branch2
);
3785 strmap_put(&opt
->priv
->paths
, b_path
, new_ci
);
3787 if (rename_a
&& rename_b
) {
3788 strmap_remove(&opt
->priv
->paths
, path
, 0);
3790 * We removed path from opt->priv->paths. path
3791 * will also eventually need to be freed, but
3792 * it may still be used by e.g. ci->pathnames.
3793 * So, store it in another string-list for now.
3795 string_list_append(&opt
->priv
->paths_to_free
,
3800 * Do special handling for b_path since process_entry()
3801 * won't be called on it specially.
3803 strmap_put(&opt
->priv
->conflicted
, b_path
, new_ci
);
3804 record_entry_for_tree(dir_metadata
, b_path
,
3808 * Remaining code for processing this entry should
3809 * think in terms of processing a_path.
3814 } else if (ci
->filemask
>= 6) {
3815 /* Need a two-way or three-way content merge */
3816 struct version_info merged_file
;
3817 unsigned clean_merge
;
3818 struct version_info
*o
= &ci
->stages
[0];
3819 struct version_info
*a
= &ci
->stages
[1];
3820 struct version_info
*b
= &ci
->stages
[2];
3822 clean_merge
= handle_content_merge(opt
, path
, o
, a
, b
,
3824 opt
->priv
->call_depth
* 2,
3826 ci
->merged
.clean
= clean_merge
&&
3827 !ci
->df_conflict
&& !ci
->path_conflict
;
3828 ci
->merged
.result
.mode
= merged_file
.mode
;
3829 ci
->merged
.is_null
= (merged_file
.mode
== 0);
3830 oidcpy(&ci
->merged
.result
.oid
, &merged_file
.oid
);
3831 if (clean_merge
&& ci
->df_conflict
) {
3832 assert(df_file_index
== 1 || df_file_index
== 2);
3833 ci
->filemask
= 1 << df_file_index
;
3834 ci
->stages
[df_file_index
].mode
= merged_file
.mode
;
3835 oidcpy(&ci
->stages
[df_file_index
].oid
, &merged_file
.oid
);
3838 const char *reason
= _("content");
3839 if (ci
->filemask
== 6)
3840 reason
= _("add/add");
3841 if (S_ISGITLINK(merged_file
.mode
))
3842 reason
= _("submodule");
3843 path_msg(opt
, path
, 0,
3844 _("CONFLICT (%s): Merge conflict in %s"),
3847 } else if (ci
->filemask
== 3 || ci
->filemask
== 5) {
3849 const char *modify_branch
, *delete_branch
;
3850 int side
= (ci
->filemask
== 5) ? 2 : 1;
3851 int index
= opt
->priv
->call_depth
? 0 : side
;
3853 ci
->merged
.result
.mode
= ci
->stages
[index
].mode
;
3854 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[index
].oid
);
3855 ci
->merged
.clean
= 0;
3857 modify_branch
= (side
== 1) ? opt
->branch1
: opt
->branch2
;
3858 delete_branch
= (side
== 1) ? opt
->branch2
: opt
->branch1
;
3860 if (opt
->renormalize
&&
3861 blob_unchanged(opt
, &ci
->stages
[0], &ci
->stages
[side
],
3863 ci
->merged
.is_null
= 1;
3864 ci
->merged
.clean
= 1;
3865 assert(!ci
->df_conflict
&& !ci
->path_conflict
);
3866 } else if (ci
->path_conflict
&&
3867 oideq(&ci
->stages
[0].oid
, &ci
->stages
[side
].oid
)) {
3869 * This came from a rename/delete; no action to take,
3870 * but avoid printing "modify/delete" conflict notice
3871 * since the contents were not modified.
3874 path_msg(opt
, path
, 0,
3875 _("CONFLICT (modify/delete): %s deleted in %s "
3876 "and modified in %s. Version %s of %s left "
3878 path
, delete_branch
, modify_branch
,
3879 modify_branch
, path
);
3881 } else if (ci
->filemask
== 2 || ci
->filemask
== 4) {
3882 /* Added on one side */
3883 int side
= (ci
->filemask
== 4) ? 2 : 1;
3884 ci
->merged
.result
.mode
= ci
->stages
[side
].mode
;
3885 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[side
].oid
);
3886 ci
->merged
.clean
= !ci
->df_conflict
&& !ci
->path_conflict
;
3887 } else if (ci
->filemask
== 1) {
3888 /* Deleted on both sides */
3889 ci
->merged
.is_null
= 1;
3890 ci
->merged
.result
.mode
= 0;
3891 oidcpy(&ci
->merged
.result
.oid
, null_oid());
3892 assert(!ci
->df_conflict
);
3893 ci
->merged
.clean
= !ci
->path_conflict
;
3897 * If still conflicted, record it separately. This allows us to later
3898 * iterate over just conflicted entries when updating the index instead
3899 * of iterating over all entries.
3901 if (!ci
->merged
.clean
)
3902 strmap_put(&opt
->priv
->conflicted
, path
, ci
);
3904 /* Record metadata for ci->merged in dir_metadata */
3905 record_entry_for_tree(dir_metadata
, path
, &ci
->merged
);
3908 static void prefetch_for_content_merges(struct merge_options
*opt
,
3909 struct string_list
*plist
)
3911 struct string_list_item
*e
;
3912 struct oid_array to_fetch
= OID_ARRAY_INIT
;
3914 if (opt
->repo
!= the_repository
|| !has_promisor_remote())
3917 for (e
= &plist
->items
[plist
->nr
-1]; e
>= plist
->items
; --e
) {
3918 /* char *path = e->string; */
3919 struct conflict_info
*ci
= e
->util
;
3922 /* Ignore clean entries */
3923 if (ci
->merged
.clean
)
3926 /* Ignore entries that don't need a content merge */
3927 if (ci
->match_mask
|| ci
->filemask
< 6 ||
3928 !S_ISREG(ci
->stages
[1].mode
) ||
3929 !S_ISREG(ci
->stages
[2].mode
) ||
3930 oideq(&ci
->stages
[1].oid
, &ci
->stages
[2].oid
))
3933 /* Also don't need content merge if base matches either side */
3934 if (ci
->filemask
== 7 &&
3935 S_ISREG(ci
->stages
[0].mode
) &&
3936 (oideq(&ci
->stages
[0].oid
, &ci
->stages
[1].oid
) ||
3937 oideq(&ci
->stages
[0].oid
, &ci
->stages
[2].oid
)))
3940 for (i
= 0; i
< 3; i
++) {
3941 unsigned side_mask
= (1 << i
);
3942 struct version_info
*vi
= &ci
->stages
[i
];
3944 if ((ci
->filemask
& side_mask
) &&
3945 S_ISREG(vi
->mode
) &&
3946 oid_object_info_extended(opt
->repo
, &vi
->oid
, NULL
,
3947 OBJECT_INFO_FOR_PREFETCH
))
3948 oid_array_append(&to_fetch
, &vi
->oid
);
3952 promisor_remote_get_direct(opt
->repo
, to_fetch
.oid
, to_fetch
.nr
);
3953 oid_array_clear(&to_fetch
);
3956 static void process_entries(struct merge_options
*opt
,
3957 struct object_id
*result_oid
)
3959 struct hashmap_iter iter
;
3960 struct strmap_entry
*e
;
3961 struct string_list plist
= STRING_LIST_INIT_NODUP
;
3962 struct string_list_item
*entry
;
3963 struct directory_versions dir_metadata
= { STRING_LIST_INIT_NODUP
,
3964 STRING_LIST_INIT_NODUP
,
3967 trace2_region_enter("merge", "process_entries setup", opt
->repo
);
3968 if (strmap_empty(&opt
->priv
->paths
)) {
3969 oidcpy(result_oid
, opt
->repo
->hash_algo
->empty_tree
);
3973 /* Hack to pre-allocate plist to the desired size */
3974 trace2_region_enter("merge", "plist grow", opt
->repo
);
3975 ALLOC_GROW(plist
.items
, strmap_get_size(&opt
->priv
->paths
), plist
.alloc
);
3976 trace2_region_leave("merge", "plist grow", opt
->repo
);
3978 /* Put every entry from paths into plist, then sort */
3979 trace2_region_enter("merge", "plist copy", opt
->repo
);
3980 strmap_for_each_entry(&opt
->priv
->paths
, &iter
, e
) {
3981 string_list_append(&plist
, e
->key
)->util
= e
->value
;
3983 trace2_region_leave("merge", "plist copy", opt
->repo
);
3985 trace2_region_enter("merge", "plist special sort", opt
->repo
);
3986 plist
.cmp
= sort_dirs_next_to_their_children
;
3987 string_list_sort(&plist
);
3988 trace2_region_leave("merge", "plist special sort", opt
->repo
);
3990 trace2_region_leave("merge", "process_entries setup", opt
->repo
);
3993 * Iterate over the items in reverse order, so we can handle paths
3994 * below a directory before needing to handle the directory itself.
3996 * This allows us to write subtrees before we need to write trees,
3997 * and it also enables sane handling of directory/file conflicts
3998 * (because it allows us to know whether the directory is still in
3999 * the way when it is time to process the file at the same path).
4001 trace2_region_enter("merge", "processing", opt
->repo
);
4002 prefetch_for_content_merges(opt
, &plist
);
4003 for (entry
= &plist
.items
[plist
.nr
-1]; entry
>= plist
.items
; --entry
) {
4004 char *path
= entry
->string
;
4006 * NOTE: mi may actually be a pointer to a conflict_info, but
4007 * we have to check mi->clean first to see if it's safe to
4008 * reassign to such a pointer type.
4010 struct merged_info
*mi
= entry
->util
;
4012 write_completed_directory(opt
, mi
->directory_name
,
4015 record_entry_for_tree(&dir_metadata
, path
, mi
);
4017 struct conflict_info
*ci
= (struct conflict_info
*)mi
;
4018 process_entry(opt
, path
, ci
, &dir_metadata
);
4021 trace2_region_leave("merge", "processing", opt
->repo
);
4023 trace2_region_enter("merge", "process_entries cleanup", opt
->repo
);
4024 if (dir_metadata
.offsets
.nr
!= 1 ||
4025 (uintptr_t)dir_metadata
.offsets
.items
[0].util
!= 0) {
4026 printf("dir_metadata.offsets.nr = %d (should be 1)\n",
4027 dir_metadata
.offsets
.nr
);
4028 printf("dir_metadata.offsets.items[0].util = %u (should be 0)\n",
4029 (unsigned)(uintptr_t)dir_metadata
.offsets
.items
[0].util
);
4031 BUG("dir_metadata accounting completely off; shouldn't happen");
4033 write_tree(result_oid
, &dir_metadata
.versions
, 0,
4034 opt
->repo
->hash_algo
->rawsz
);
4035 string_list_clear(&plist
, 0);
4036 string_list_clear(&dir_metadata
.versions
, 0);
4037 string_list_clear(&dir_metadata
.offsets
, 0);
4038 trace2_region_leave("merge", "process_entries cleanup", opt
->repo
);
4041 /*** Function Grouping: functions related to merge_switch_to_result() ***/
4043 static int checkout(struct merge_options
*opt
,
4047 /* Switch the index/working copy from old to new */
4049 struct tree_desc trees
[2];
4050 struct unpack_trees_options unpack_opts
;
4052 memset(&unpack_opts
, 0, sizeof(unpack_opts
));
4053 unpack_opts
.head_idx
= -1;
4054 unpack_opts
.src_index
= opt
->repo
->index
;
4055 unpack_opts
.dst_index
= opt
->repo
->index
;
4057 setup_unpack_trees_porcelain(&unpack_opts
, "merge");
4060 * NOTE: if this were just "git checkout" code, we would probably
4061 * read or refresh the cache and check for a conflicted index, but
4062 * builtin/merge.c or sequencer.c really needs to read the index
4063 * and check for conflicted entries before starting merging for a
4064 * good user experience (no sense waiting for merges/rebases before
4065 * erroring out), so there's no reason to duplicate that work here.
4068 /* 2-way merge to the new branch */
4069 unpack_opts
.update
= 1;
4070 unpack_opts
.merge
= 1;
4071 unpack_opts
.quiet
= 0; /* FIXME: sequencer might want quiet? */
4072 unpack_opts
.verbose_update
= (opt
->verbosity
> 2);
4073 unpack_opts
.fn
= twoway_merge
;
4074 if (1/* FIXME: opts->overwrite_ignore*/) {
4075 CALLOC_ARRAY(unpack_opts
.dir
, 1);
4076 unpack_opts
.dir
->flags
|= DIR_SHOW_IGNORED
;
4077 setup_standard_excludes(unpack_opts
.dir
);
4080 init_tree_desc(&trees
[0], prev
->buffer
, prev
->size
);
4082 init_tree_desc(&trees
[1], next
->buffer
, next
->size
);
4084 ret
= unpack_trees(2, trees
, &unpack_opts
);
4085 clear_unpack_trees_porcelain(&unpack_opts
);
4086 dir_clear(unpack_opts
.dir
);
4087 FREE_AND_NULL(unpack_opts
.dir
);
4091 static int record_conflicted_index_entries(struct merge_options
*opt
)
4093 struct hashmap_iter iter
;
4094 struct strmap_entry
*e
;
4095 struct index_state
*index
= opt
->repo
->index
;
4096 struct checkout state
= CHECKOUT_INIT
;
4098 int original_cache_nr
;
4100 if (strmap_empty(&opt
->priv
->conflicted
))
4103 /* If any entries have skip_worktree set, we'll have to check 'em out */
4106 state
.refresh_cache
= 1;
4107 state
.istate
= index
;
4108 original_cache_nr
= index
->cache_nr
;
4110 /* Put every entry from paths into plist, then sort */
4111 strmap_for_each_entry(&opt
->priv
->conflicted
, &iter
, e
) {
4112 const char *path
= e
->key
;
4113 struct conflict_info
*ci
= e
->value
;
4115 struct cache_entry
*ce
;
4121 * The index will already have a stage=0 entry for this path,
4122 * because we created an as-merged-as-possible version of the
4123 * file and checkout() moved the working copy and index over
4126 * However, previous iterations through this loop will have
4127 * added unstaged entries to the end of the cache which
4128 * ignore the standard alphabetical ordering of cache
4129 * entries and break invariants needed for index_name_pos()
4130 * to work. However, we know the entry we want is before
4131 * those appended cache entries, so do a temporary swap on
4132 * cache_nr to only look through entries of interest.
4134 SWAP(index
->cache_nr
, original_cache_nr
);
4135 pos
= index_name_pos(index
, path
, strlen(path
));
4136 SWAP(index
->cache_nr
, original_cache_nr
);
4138 if (ci
->filemask
!= 1)
4139 BUG("Conflicted %s but nothing in basic working tree or index; this shouldn't happen", path
);
4140 cache_tree_invalidate_path(index
, path
);
4142 ce
= index
->cache
[pos
];
4145 * Clean paths with CE_SKIP_WORKTREE set will not be
4146 * written to the working tree by the unpack_trees()
4147 * call in checkout(). Our conflicted entries would
4148 * have appeared clean to that code since we ignored
4149 * the higher order stages. Thus, we need override
4150 * the CE_SKIP_WORKTREE bit and manually write those
4151 * files to the working disk here.
4153 if (ce_skip_worktree(ce
)) {
4156 if (!lstat(path
, &st
)) {
4157 char *new_name
= unique_path(&opt
->priv
->paths
,
4161 path_msg(opt
, path
, 1,
4162 _("Note: %s not up to date and in way of checking out conflicted version; old copy renamed to %s"),
4164 errs
|= rename(path
, new_name
);
4167 errs
|= checkout_entry(ce
, &state
, NULL
, NULL
);
4171 * Mark this cache entry for removal and instead add
4172 * new stage>0 entries corresponding to the
4173 * conflicts. If there are many conflicted entries, we
4174 * want to avoid memmove'ing O(NM) entries by
4175 * inserting the new entries one at a time. So,
4176 * instead, we just add the new cache entries to the
4177 * end (ignoring normal index requirements on sort
4178 * order) and sort the index once we're all done.
4180 ce
->ce_flags
|= CE_REMOVE
;
4183 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
4184 struct version_info
*vi
;
4185 if (!(ci
->filemask
& (1ul << i
)))
4187 vi
= &ci
->stages
[i
];
4188 ce
= make_cache_entry(index
, vi
->mode
, &vi
->oid
,
4190 add_index_entry(index
, ce
, ADD_CACHE_JUST_APPEND
);
4195 * Remove the unused cache entries (and invalidate the relevant
4196 * cache-trees), then sort the index entries to get the conflicted
4197 * entries we added to the end into their right locations.
4199 remove_marked_cache_entries(index
, 1);
4201 * No need for STABLE_QSORT -- cmp_cache_name_compare sorts primarily
4202 * on filename and secondarily on stage, and (name, stage #) are a
4205 QSORT(index
->cache
, index
->cache_nr
, cmp_cache_name_compare
);
4210 void merge_switch_to_result(struct merge_options
*opt
,
4212 struct merge_result
*result
,
4213 int update_worktree_and_index
,
4214 int display_update_msgs
)
4216 assert(opt
->priv
== NULL
);
4217 if (result
->clean
>= 0 && update_worktree_and_index
) {
4218 const char *filename
;
4221 trace2_region_enter("merge", "checkout", opt
->repo
);
4222 if (checkout(opt
, head
, result
->tree
)) {
4223 /* failure to function */
4227 trace2_region_leave("merge", "checkout", opt
->repo
);
4229 trace2_region_enter("merge", "record_conflicted", opt
->repo
);
4230 opt
->priv
= result
->priv
;
4231 if (record_conflicted_index_entries(opt
)) {
4232 /* failure to function */
4238 trace2_region_leave("merge", "record_conflicted", opt
->repo
);
4240 trace2_region_enter("merge", "write_auto_merge", opt
->repo
);
4241 filename
= git_path_auto_merge(opt
->repo
);
4242 fp
= xfopen(filename
, "w");
4243 fprintf(fp
, "%s\n", oid_to_hex(&result
->tree
->object
.oid
));
4245 trace2_region_leave("merge", "write_auto_merge", opt
->repo
);
4248 if (display_update_msgs
) {
4249 struct merge_options_internal
*opti
= result
->priv
;
4250 struct hashmap_iter iter
;
4251 struct strmap_entry
*e
;
4252 struct string_list olist
= STRING_LIST_INIT_NODUP
;
4255 trace2_region_enter("merge", "display messages", opt
->repo
);
4257 /* Hack to pre-allocate olist to the desired size */
4258 ALLOC_GROW(olist
.items
, strmap_get_size(&opti
->output
),
4261 /* Put every entry from output into olist, then sort */
4262 strmap_for_each_entry(&opti
->output
, &iter
, e
) {
4263 string_list_append(&olist
, e
->key
)->util
= e
->value
;
4265 string_list_sort(&olist
);
4267 /* Iterate over the items, printing them */
4268 for (i
= 0; i
< olist
.nr
; ++i
) {
4269 struct strbuf
*sb
= olist
.items
[i
].util
;
4271 printf("%s", sb
->buf
);
4273 string_list_clear(&olist
, 0);
4275 /* Also include needed rename limit adjustment now */
4276 diff_warn_rename_limit("merge.renamelimit",
4277 opti
->renames
.needed_limit
, 0);
4279 trace2_region_leave("merge", "display messages", opt
->repo
);
4282 merge_finalize(opt
, result
);
4285 void merge_finalize(struct merge_options
*opt
,
4286 struct merge_result
*result
)
4288 struct merge_options_internal
*opti
= result
->priv
;
4290 if (opt
->renormalize
)
4291 git_attr_set_direction(GIT_ATTR_CHECKIN
);
4292 assert(opt
->priv
== NULL
);
4294 clear_or_reinit_internal_opts(opti
, 0);
4295 FREE_AND_NULL(opti
);
4298 /*** Function Grouping: helper functions for merge_incore_*() ***/
4300 static struct tree
*shift_tree_object(struct repository
*repo
,
4301 struct tree
*one
, struct tree
*two
,
4302 const char *subtree_shift
)
4304 struct object_id shifted
;
4306 if (!*subtree_shift
) {
4307 shift_tree(repo
, &one
->object
.oid
, &two
->object
.oid
, &shifted
, 0);
4309 shift_tree_by(repo
, &one
->object
.oid
, &two
->object
.oid
, &shifted
,
4312 if (oideq(&two
->object
.oid
, &shifted
))
4314 return lookup_tree(repo
, &shifted
);
4317 static inline void set_commit_tree(struct commit
*c
, struct tree
*t
)
4322 static struct commit
*make_virtual_commit(struct repository
*repo
,
4324 const char *comment
)
4326 struct commit
*commit
= alloc_commit_node(repo
);
4328 set_merge_remote_desc(commit
, comment
, (struct object
*)commit
);
4329 set_commit_tree(commit
, tree
);
4330 commit
->object
.parsed
= 1;
4334 static void merge_start(struct merge_options
*opt
, struct merge_result
*result
)
4336 struct rename_info
*renames
;
4339 /* Sanity checks on opt */
4340 trace2_region_enter("merge", "sanity checks", opt
->repo
);
4343 assert(opt
->branch1
&& opt
->branch2
);
4345 assert(opt
->detect_directory_renames
>= MERGE_DIRECTORY_RENAMES_NONE
&&
4346 opt
->detect_directory_renames
<= MERGE_DIRECTORY_RENAMES_TRUE
);
4347 assert(opt
->rename_limit
>= -1);
4348 assert(opt
->rename_score
>= 0 && opt
->rename_score
<= MAX_SCORE
);
4349 assert(opt
->show_rename_progress
>= 0 && opt
->show_rename_progress
<= 1);
4351 assert(opt
->xdl_opts
>= 0);
4352 assert(opt
->recursive_variant
>= MERGE_VARIANT_NORMAL
&&
4353 opt
->recursive_variant
<= MERGE_VARIANT_THEIRS
);
4356 * detect_renames, verbosity, buffer_output, and obuf are ignored
4357 * fields that were used by "recursive" rather than "ort" -- but
4358 * sanity check them anyway.
4360 assert(opt
->detect_renames
>= -1 &&
4361 opt
->detect_renames
<= DIFF_DETECT_COPY
);
4362 assert(opt
->verbosity
>= 0 && opt
->verbosity
<= 5);
4363 assert(opt
->buffer_output
<= 2);
4364 assert(opt
->obuf
.len
== 0);
4366 assert(opt
->priv
== NULL
);
4367 if (result
->_properly_initialized
!= 0 &&
4368 result
->_properly_initialized
!= RESULT_INITIALIZED
)
4369 BUG("struct merge_result passed to merge_incore_*recursive() must be zeroed or filled with values from a previous run");
4370 assert(!!result
->priv
== !!result
->_properly_initialized
);
4372 opt
->priv
= result
->priv
;
4373 result
->priv
= NULL
;
4375 * opt->priv non-NULL means we had results from a previous
4376 * run; do a few sanity checks that user didn't mess with
4377 * it in an obvious fashion.
4379 assert(opt
->priv
->call_depth
== 0);
4380 assert(!opt
->priv
->toplevel_dir
||
4381 0 == strlen(opt
->priv
->toplevel_dir
));
4383 trace2_region_leave("merge", "sanity checks", opt
->repo
);
4385 /* Default to histogram diff. Actually, just hardcode it...for now. */
4386 opt
->xdl_opts
= DIFF_WITH_ALG(opt
, HISTOGRAM_DIFF
);
4388 /* Handle attr direction stuff for renormalization */
4389 if (opt
->renormalize
)
4390 git_attr_set_direction(GIT_ATTR_CHECKOUT
);
4392 /* Initialization of opt->priv, our internal merge data */
4393 trace2_region_enter("merge", "allocate/init", opt
->repo
);
4395 clear_or_reinit_internal_opts(opt
->priv
, 1);
4396 trace2_region_leave("merge", "allocate/init", opt
->repo
);
4399 opt
->priv
= xcalloc(1, sizeof(*opt
->priv
));
4401 /* Initialization of various renames fields */
4402 renames
= &opt
->priv
->renames
;
4404 mem_pool_init(&opt
->priv
->internal_pool
, 0);
4405 opt
->priv
->pool
= &opt
->priv
->internal_pool
;
4407 opt
->priv
->pool
= NULL
;
4409 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; i
++) {
4410 strintmap_init_with_options(&renames
->dirs_removed
[i
],
4411 NOT_RELEVANT
, NULL
, 0);
4412 strmap_init_with_options(&renames
->dir_rename_count
[i
],
4414 strmap_init_with_options(&renames
->dir_renames
[i
],
4417 * relevant_sources uses -1 for the default, because we need
4418 * to be able to distinguish not-in-strintmap from valid
4419 * relevant_source values from enum file_rename_relevance.
4420 * In particular, possibly_cache_new_pair() expects a negative
4421 * value for not-found entries.
4423 strintmap_init_with_options(&renames
->relevant_sources
[i
],
4424 -1 /* explicitly invalid */,
4426 strmap_init_with_options(&renames
->cached_pairs
[i
],
4428 strset_init_with_options(&renames
->cached_irrelevant
[i
],
4430 strset_init_with_options(&renames
->cached_target_names
[i
],
4433 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; i
++) {
4434 strintmap_init_with_options(&renames
->deferred
[i
].possible_trivial_merges
,
4436 strset_init_with_options(&renames
->deferred
[i
].target_dirs
,
4438 renames
->deferred
[i
].trivial_merges_okay
= 1; /* 1 == maybe */
4442 * Although we initialize opt->priv->paths with strdup_strings=0,
4443 * that's just to avoid making yet another copy of an allocated
4444 * string. Putting the entry into paths means we are taking
4445 * ownership, so we will later free it. paths_to_free is similar.
4447 * In contrast, conflicted just has a subset of keys from paths, so
4448 * we don't want to free those (it'd be a duplicate free).
4450 strmap_init_with_options(&opt
->priv
->paths
, NULL
, 0);
4451 strmap_init_with_options(&opt
->priv
->conflicted
, NULL
, 0);
4452 string_list_init_nodup(&opt
->priv
->paths_to_free
);
4455 * keys & strbufs in output will sometimes need to outlive "paths",
4456 * so it will have a copy of relevant keys. It's probably a small
4457 * subset of the overall paths that have special output.
4459 strmap_init(&opt
->priv
->output
);
4461 trace2_region_leave("merge", "allocate/init", opt
->repo
);
4464 static void merge_check_renames_reusable(struct merge_options
*opt
,
4465 struct merge_result
*result
,
4466 struct tree
*merge_base
,
4470 struct rename_info
*renames
;
4471 struct tree
**merge_trees
;
4472 struct merge_options_internal
*opti
= result
->priv
;
4477 renames
= &opti
->renames
;
4478 merge_trees
= renames
->merge_trees
;
4481 * Handle case where previous merge operation did not want cache to
4482 * take effect, e.g. because rename/rename(1to1) makes it invalid.
4484 if (!merge_trees
[0]) {
4485 assert(!merge_trees
[0] && !merge_trees
[1] && !merge_trees
[2]);
4486 renames
->cached_pairs_valid_side
= 0; /* neither side valid */
4491 * Handle other cases; note that merge_trees[0..2] will only
4492 * be NULL if opti is, or if all three were manually set to
4493 * NULL by e.g. rename/rename(1to1) handling.
4495 assert(merge_trees
[0] && merge_trees
[1] && merge_trees
[2]);
4497 /* Check if we meet a condition for re-using cached_pairs */
4498 if (oideq(&merge_base
->object
.oid
, &merge_trees
[2]->object
.oid
) &&
4499 oideq(&side1
->object
.oid
, &result
->tree
->object
.oid
))
4500 renames
->cached_pairs_valid_side
= MERGE_SIDE1
;
4501 else if (oideq(&merge_base
->object
.oid
, &merge_trees
[1]->object
.oid
) &&
4502 oideq(&side2
->object
.oid
, &result
->tree
->object
.oid
))
4503 renames
->cached_pairs_valid_side
= MERGE_SIDE2
;
4505 renames
->cached_pairs_valid_side
= 0; /* neither side valid */
4508 /*** Function Grouping: merge_incore_*() and their internal variants ***/
4511 * Originally from merge_trees_internal(); heavily adapted, though.
4513 static void merge_ort_nonrecursive_internal(struct merge_options
*opt
,
4514 struct tree
*merge_base
,
4517 struct merge_result
*result
)
4519 struct object_id working_tree_oid
;
4521 if (opt
->subtree_shift
) {
4522 side2
= shift_tree_object(opt
->repo
, side1
, side2
,
4523 opt
->subtree_shift
);
4524 merge_base
= shift_tree_object(opt
->repo
, side1
, merge_base
,
4525 opt
->subtree_shift
);
4529 trace2_region_enter("merge", "collect_merge_info", opt
->repo
);
4530 if (collect_merge_info(opt
, merge_base
, side1
, side2
) != 0) {
4532 * TRANSLATORS: The %s arguments are: 1) tree hash of a merge
4533 * base, and 2-3) the trees for the two trees we're merging.
4535 err(opt
, _("collecting merge info failed for trees %s, %s, %s"),
4536 oid_to_hex(&merge_base
->object
.oid
),
4537 oid_to_hex(&side1
->object
.oid
),
4538 oid_to_hex(&side2
->object
.oid
));
4542 trace2_region_leave("merge", "collect_merge_info", opt
->repo
);
4544 trace2_region_enter("merge", "renames", opt
->repo
);
4545 result
->clean
= detect_and_process_renames(opt
, merge_base
,
4547 trace2_region_leave("merge", "renames", opt
->repo
);
4548 if (opt
->priv
->renames
.redo_after_renames
== 2) {
4549 trace2_region_enter("merge", "reset_maps", opt
->repo
);
4550 clear_or_reinit_internal_opts(opt
->priv
, 1);
4551 trace2_region_leave("merge", "reset_maps", opt
->repo
);
4555 trace2_region_enter("merge", "process_entries", opt
->repo
);
4556 process_entries(opt
, &working_tree_oid
);
4557 trace2_region_leave("merge", "process_entries", opt
->repo
);
4559 /* Set return values */
4560 result
->tree
= parse_tree_indirect(&working_tree_oid
);
4561 /* existence of conflicted entries implies unclean */
4562 result
->clean
&= strmap_empty(&opt
->priv
->conflicted
);
4563 if (!opt
->priv
->call_depth
) {
4564 result
->priv
= opt
->priv
;
4565 result
->_properly_initialized
= RESULT_INITIALIZED
;
4571 * Originally from merge_recursive_internal(); somewhat adapted, though.
4573 static void merge_ort_internal(struct merge_options
*opt
,
4574 struct commit_list
*merge_bases
,
4577 struct merge_result
*result
)
4579 struct commit_list
*iter
;
4580 struct commit
*merged_merge_bases
;
4581 const char *ancestor_name
;
4582 struct strbuf merge_base_abbrev
= STRBUF_INIT
;
4585 merge_bases
= get_merge_bases(h1
, h2
);
4586 /* See merge-ort.h:merge_incore_recursive() declaration NOTE */
4587 merge_bases
= reverse_commit_list(merge_bases
);
4590 merged_merge_bases
= pop_commit(&merge_bases
);
4591 if (merged_merge_bases
== NULL
) {
4592 /* if there is no common ancestor, use an empty tree */
4595 tree
= lookup_tree(opt
->repo
, opt
->repo
->hash_algo
->empty_tree
);
4596 merged_merge_bases
= make_virtual_commit(opt
->repo
, tree
,
4598 ancestor_name
= "empty tree";
4599 } else if (merge_bases
) {
4600 ancestor_name
= "merged common ancestors";
4602 strbuf_add_unique_abbrev(&merge_base_abbrev
,
4603 &merged_merge_bases
->object
.oid
,
4605 ancestor_name
= merge_base_abbrev
.buf
;
4608 for (iter
= merge_bases
; iter
; iter
= iter
->next
) {
4609 const char *saved_b1
, *saved_b2
;
4610 struct commit
*prev
= merged_merge_bases
;
4612 opt
->priv
->call_depth
++;
4614 * When the merge fails, the result contains files
4615 * with conflict markers. The cleanness flag is
4616 * ignored (unless indicating an error), it was never
4617 * actually used, as result of merge_trees has always
4618 * overwritten it: the committed "conflicts" were
4621 saved_b1
= opt
->branch1
;
4622 saved_b2
= opt
->branch2
;
4623 opt
->branch1
= "Temporary merge branch 1";
4624 opt
->branch2
= "Temporary merge branch 2";
4625 merge_ort_internal(opt
, NULL
, prev
, iter
->item
, result
);
4626 if (result
->clean
< 0)
4628 opt
->branch1
= saved_b1
;
4629 opt
->branch2
= saved_b2
;
4630 opt
->priv
->call_depth
--;
4632 merged_merge_bases
= make_virtual_commit(opt
->repo
,
4635 commit_list_insert(prev
, &merged_merge_bases
->parents
);
4636 commit_list_insert(iter
->item
,
4637 &merged_merge_bases
->parents
->next
);
4639 clear_or_reinit_internal_opts(opt
->priv
, 1);
4642 opt
->ancestor
= ancestor_name
;
4643 merge_ort_nonrecursive_internal(opt
,
4644 repo_get_commit_tree(opt
->repo
,
4645 merged_merge_bases
),
4646 repo_get_commit_tree(opt
->repo
, h1
),
4647 repo_get_commit_tree(opt
->repo
, h2
),
4649 strbuf_release(&merge_base_abbrev
);
4650 opt
->ancestor
= NULL
; /* avoid accidental re-use of opt->ancestor */
4653 void merge_incore_nonrecursive(struct merge_options
*opt
,
4654 struct tree
*merge_base
,
4657 struct merge_result
*result
)
4659 trace2_region_enter("merge", "incore_nonrecursive", opt
->repo
);
4661 trace2_region_enter("merge", "merge_start", opt
->repo
);
4662 assert(opt
->ancestor
!= NULL
);
4663 merge_check_renames_reusable(opt
, result
, merge_base
, side1
, side2
);
4664 merge_start(opt
, result
);
4666 * Record the trees used in this merge, so if there's a next merge in
4667 * a cherry-pick or rebase sequence it might be able to take advantage
4668 * of the cached_pairs in that next merge.
4670 opt
->priv
->renames
.merge_trees
[0] = merge_base
;
4671 opt
->priv
->renames
.merge_trees
[1] = side1
;
4672 opt
->priv
->renames
.merge_trees
[2] = side2
;
4673 trace2_region_leave("merge", "merge_start", opt
->repo
);
4675 merge_ort_nonrecursive_internal(opt
, merge_base
, side1
, side2
, result
);
4676 trace2_region_leave("merge", "incore_nonrecursive", opt
->repo
);
4679 void merge_incore_recursive(struct merge_options
*opt
,
4680 struct commit_list
*merge_bases
,
4681 struct commit
*side1
,
4682 struct commit
*side2
,
4683 struct merge_result
*result
)
4685 trace2_region_enter("merge", "incore_recursive", opt
->repo
);
4687 /* We set the ancestor label based on the merge_bases */
4688 assert(opt
->ancestor
== NULL
);
4690 trace2_region_enter("merge", "merge_start", opt
->repo
);
4691 merge_start(opt
, result
);
4692 trace2_region_leave("merge", "merge_start", opt
->repo
);
4694 merge_ort_internal(opt
, merge_bases
, side1
, side2
, result
);
4695 trace2_region_leave("merge", "incore_recursive", opt
->repo
);