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"
21 #include "cache-tree.h"
22 #include "commit-reach.h"
26 #include "object-store.h"
29 #include "unpack-trees.h"
30 #include "xdiff-interface.h"
33 * We have many arrays of size 3. Whenever we have such an array, the
34 * indices refer to one of the sides of the three-way merge. This is so
35 * pervasive that the constants 0, 1, and 2 are used in many places in the
36 * code (especially in arithmetic operations to find the other side's index
37 * or to compute a relevant mask), but sometimes these enum names are used
38 * to aid code clarity.
40 * See also 'filemask' and 'dirmask' in struct conflict_info; the "ith side"
41 * referred to there is one of these three sides.
49 struct merge_options_internal
{
51 * paths: primary data structure in all of merge ort.
54 * * are full relative paths from the toplevel of the repository
55 * (e.g. "drivers/firmware/raspberrypi.c").
56 * * store all relevant paths in the repo, both directories and
57 * files (e.g. drivers, drivers/firmware would also be included)
58 * * these keys serve to intern all the path strings, which allows
59 * us to do pointer comparison on directory names instead of
60 * strcmp; we just have to be careful to use the interned strings.
61 * (Technically paths_to_free may track some strings that were
62 * removed from froms paths.)
64 * The values of paths:
65 * * either a pointer to a merged_info, or a conflict_info struct
66 * * merged_info contains all relevant information for a
67 * non-conflicted entry.
68 * * conflict_info contains a merged_info, plus any additional
69 * information about a conflict such as the higher orders stages
70 * involved and the names of the paths those came from (handy
71 * once renames get involved).
72 * * a path may start "conflicted" (i.e. point to a conflict_info)
73 * and then a later step (e.g. three-way content merge) determines
74 * it can be cleanly merged, at which point it'll be marked clean
75 * and the algorithm will ignore any data outside the contained
76 * merged_info for that entry
77 * * If an entry remains conflicted, the merged_info portion of a
78 * conflict_info will later be filled with whatever version of
79 * the file should be placed in the working directory (e.g. an
80 * as-merged-as-possible variation that contains conflict markers).
85 * conflicted: a subset of keys->values from "paths"
87 * conflicted is basically an optimization between process_entries()
88 * and record_conflicted_index_entries(); the latter could loop over
89 * ALL the entries in paths AGAIN and look for the ones that are
90 * still conflicted, but since process_entries() has to loop over
91 * all of them, it saves the ones it couldn't resolve in this strmap
92 * so that record_conflicted_index_entries() can iterate just the
95 struct strmap conflicted
;
98 * paths_to_free: additional list of strings to free
100 * If keys are removed from "paths", they are added to paths_to_free
101 * to ensure they are later freed. We avoid free'ing immediately since
102 * other places (e.g. conflict_info.pathnames[]) may still be
103 * referencing these paths.
105 struct string_list paths_to_free
;
108 * output: special messages and conflict notices for various paths
110 * This is a map of pathnames (a subset of the keys in "paths" above)
111 * to strbufs. It gathers various warning/conflict/notice messages
112 * for later processing.
114 struct strmap output
;
117 * current_dir_name: temporary var used in collect_merge_info_callback()
119 * Used to set merged_info.directory_name; see documentation for that
120 * variable and the requirements placed on that field.
122 const char *current_dir_name
;
124 /* call_depth: recursion level counter for merging merge bases */
128 struct version_info
{
129 struct object_id oid
;
134 /* if is_null, ignore result. otherwise result has oid & mode */
135 struct version_info result
;
139 * clean: whether the path in question is cleanly merged.
141 * see conflict_info.merged for more details.
146 * basename_offset: offset of basename of path.
148 * perf optimization to avoid recomputing offset of final '/'
149 * character in pathname (0 if no '/' in pathname).
151 size_t basename_offset
;
154 * directory_name: containing directory name.
156 * Note that we assume directory_name is constructed such that
157 * strcmp(dir1_name, dir2_name) == 0 iff dir1_name == dir2_name,
158 * i.e. string equality is equivalent to pointer equality. For this
159 * to hold, we have to be careful setting directory_name.
161 const char *directory_name
;
164 struct conflict_info
{
166 * merged: the version of the path that will be written to working tree
168 * WARNING: It is critical to check merged.clean and ensure it is 0
169 * before reading any conflict_info fields outside of merged.
170 * Allocated merge_info structs will always have clean set to 1.
171 * Allocated conflict_info structs will have merged.clean set to 0
172 * initially. The merged.clean field is how we know if it is safe
173 * to access other parts of conflict_info besides merged; if a
174 * conflict_info's merged.clean is changed to 1, the rest of the
175 * algorithm is not allowed to look at anything outside of the
176 * merged member anymore.
178 struct merged_info merged
;
180 /* oids & modes from each of the three trees for this path */
181 struct version_info stages
[3];
183 /* pathnames for each stage; may differ due to rename detection */
184 const char *pathnames
[3];
186 /* Whether this path is/was involved in a directory/file conflict */
187 unsigned df_conflict
:1;
190 * Whether this path is/was involved in a non-content conflict other
191 * than a directory/file conflict (e.g. rename/rename, rename/delete,
192 * file location based on possible directory rename).
194 unsigned path_conflict
:1;
197 * For filemask and dirmask, the ith bit corresponds to whether the
198 * ith entry is a file (filemask) or a directory (dirmask). Thus,
199 * filemask & dirmask is always zero, and filemask | dirmask is at
200 * most 7 but can be less when a path does not appear as either a
201 * file or a directory on at least one side of history.
203 * Note that these masks are related to enum merge_side, as the ith
204 * entry corresponds to side i.
206 * These values come from a traverse_trees() call; more info may be
207 * found looking at tree-walk.h's struct traverse_info,
208 * particularly the documentation above the "fn" member (note that
209 * filemask = mask & ~dirmask from that documentation).
215 * Optimization to track which stages match, to avoid the need to
216 * recompute it in multiple steps. Either 0 or at least 2 bits are
217 * set; if at least 2 bits are set, their corresponding stages match.
219 unsigned match_mask
:3;
222 /*** Function Grouping: various utility functions ***/
225 * For the next three macros, see warning for conflict_info.merged.
227 * In each of the below, mi is a struct merged_info*, and ci was defined
228 * as a struct conflict_info* (but we need to verify ci isn't actually
229 * pointed at a struct merged_info*).
231 * INITIALIZE_CI: Assign ci to mi but only if it's safe; set to NULL otherwise.
232 * VERIFY_CI: Ensure that something we assigned to a conflict_info* is one.
233 * ASSIGN_AND_VERIFY_CI: Similar to VERIFY_CI but do assignment first.
235 #define INITIALIZE_CI(ci, mi) do { \
236 (ci) = (!(mi) || (mi)->clean) ? NULL : (struct conflict_info *)(mi); \
238 #define VERIFY_CI(ci) assert(ci && !ci->merged.clean);
239 #define ASSIGN_AND_VERIFY_CI(ci, mi) do { \
240 (ci) = (struct conflict_info *)(mi); \
241 assert((ci) && !(mi)->clean); \
244 static void free_strmap_strings(struct strmap
*map
)
246 struct hashmap_iter iter
;
247 struct strmap_entry
*entry
;
249 strmap_for_each_entry(map
, &iter
, entry
) {
250 free((char*)entry
->key
);
254 static void clear_internal_opts(struct merge_options_internal
*opti
,
257 assert(!reinitialize
);
260 * We marked opti->paths with strdup_strings = 0, so that we
261 * wouldn't have to make another copy of the fullpath created by
262 * make_traverse_path from setup_path_info(). But, now that we've
263 * used it and have no other references to these strings, it is time
264 * to deallocate them.
266 free_strmap_strings(&opti
->paths
);
267 strmap_clear(&opti
->paths
, 1);
270 * All keys and values in opti->conflicted are a subset of those in
271 * opti->paths. We don't want to deallocate anything twice, so we
272 * don't free the keys and we pass 0 for free_values.
274 strmap_clear(&opti
->conflicted
, 0);
277 * opti->paths_to_free is similar to opti->paths; we created it with
278 * strdup_strings = 0 to avoid making _another_ copy of the fullpath
279 * but now that we've used it and have no other references to these
280 * strings, it is time to deallocate them. We do so by temporarily
281 * setting strdup_strings to 1.
283 opti
->paths_to_free
.strdup_strings
= 1;
284 string_list_clear(&opti
->paths_to_free
, 0);
285 opti
->paths_to_free
.strdup_strings
= 0;
288 struct hashmap_iter iter
;
289 struct strmap_entry
*e
;
291 /* Release and free each strbuf found in output */
292 strmap_for_each_entry(&opti
->output
, &iter
, e
) {
293 struct strbuf
*sb
= e
->value
;
296 * While strictly speaking we don't need to free(sb)
297 * here because we could pass free_values=1 when
298 * calling strmap_clear() on opti->output, that would
299 * require strmap_clear to do another
300 * strmap_for_each_entry() loop, so we just free it
301 * while we're iterating anyway.
305 strmap_clear(&opti
->output
, 0);
309 static int err(struct merge_options
*opt
, const char *err
, ...)
312 struct strbuf sb
= STRBUF_INIT
;
314 strbuf_addstr(&sb
, "error: ");
315 va_start(params
, err
);
316 strbuf_vaddf(&sb
, err
, params
);
325 __attribute__((format (printf
, 4, 5)))
326 static void path_msg(struct merge_options
*opt
,
328 int omittable_hint
, /* skippable under --remerge-diff */
329 const char *fmt
, ...)
332 struct strbuf
*sb
= strmap_get(&opt
->priv
->output
, path
);
334 sb
= xmalloc(sizeof(*sb
));
336 strmap_put(&opt
->priv
->output
, path
, sb
);
340 strbuf_vaddf(sb
, fmt
, ap
);
343 strbuf_addch(sb
, '\n');
346 /*** Function Grouping: functions related to collect_merge_info() ***/
348 static void setup_path_info(struct merge_options
*opt
,
349 struct string_list_item
*result
,
350 const char *current_dir_name
,
351 int current_dir_name_len
,
352 char *fullpath
, /* we'll take over ownership */
353 struct name_entry
*names
,
354 struct name_entry
*merged_version
,
355 unsigned is_null
, /* boolean */
356 unsigned df_conflict
, /* boolean */
359 int resolved
/* boolean */)
361 /* result->util is void*, so mi is a convenience typed variable */
362 struct merged_info
*mi
;
364 assert(!is_null
|| resolved
);
365 assert(!df_conflict
|| !resolved
); /* df_conflict implies !resolved */
366 assert(resolved
== (merged_version
!= NULL
));
368 mi
= xcalloc(1, resolved
? sizeof(struct merged_info
) :
369 sizeof(struct conflict_info
));
370 mi
->directory_name
= current_dir_name
;
371 mi
->basename_offset
= current_dir_name_len
;
372 mi
->clean
= !!resolved
;
374 mi
->result
.mode
= merged_version
->mode
;
375 oidcpy(&mi
->result
.oid
, &merged_version
->oid
);
376 mi
->is_null
= !!is_null
;
379 struct conflict_info
*ci
;
381 ASSIGN_AND_VERIFY_CI(ci
, mi
);
382 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
383 ci
->pathnames
[i
] = fullpath
;
384 ci
->stages
[i
].mode
= names
[i
].mode
;
385 oidcpy(&ci
->stages
[i
].oid
, &names
[i
].oid
);
387 ci
->filemask
= filemask
;
388 ci
->dirmask
= dirmask
;
389 ci
->df_conflict
= !!df_conflict
;
392 * Assume is_null for now, but if we have entries
393 * under the directory then when it is complete in
394 * write_completed_directory() it'll update this.
395 * Also, for D/F conflicts, we have to handle the
396 * directory first, then clear this bit and process
397 * the file to see how it is handled -- that occurs
398 * near the top of process_entry().
402 strmap_put(&opt
->priv
->paths
, fullpath
, mi
);
403 result
->string
= fullpath
;
407 static int collect_merge_info_callback(int n
,
409 unsigned long dirmask
,
410 struct name_entry
*names
,
411 struct traverse_info
*info
)
415 * common ancestor (mbase) has mask 1, and stored in index 0 of names
416 * head of side 1 (side1) has mask 2, and stored in index 1 of names
417 * head of side 2 (side2) has mask 4, and stored in index 2 of names
419 struct merge_options
*opt
= info
->data
;
420 struct merge_options_internal
*opti
= opt
->priv
;
421 struct string_list_item pi
; /* Path Info */
422 struct conflict_info
*ci
; /* typed alias to pi.util (which is void*) */
423 struct name_entry
*p
;
426 const char *dirname
= opti
->current_dir_name
;
427 unsigned filemask
= mask
& ~dirmask
;
428 unsigned match_mask
= 0; /* will be updated below */
429 unsigned mbase_null
= !(mask
& 1);
430 unsigned side1_null
= !(mask
& 2);
431 unsigned side2_null
= !(mask
& 4);
432 unsigned side1_matches_mbase
= (!side1_null
&& !mbase_null
&&
433 names
[0].mode
== names
[1].mode
&&
434 oideq(&names
[0].oid
, &names
[1].oid
));
435 unsigned side2_matches_mbase
= (!side2_null
&& !mbase_null
&&
436 names
[0].mode
== names
[2].mode
&&
437 oideq(&names
[0].oid
, &names
[2].oid
));
438 unsigned sides_match
= (!side1_null
&& !side2_null
&&
439 names
[1].mode
== names
[2].mode
&&
440 oideq(&names
[1].oid
, &names
[2].oid
));
443 * Note: When a path is a file on one side of history and a directory
444 * in another, we have a directory/file conflict. In such cases, if
445 * the conflict doesn't resolve from renames and deletions, then we
446 * always leave directories where they are and move files out of the
447 * way. Thus, while struct conflict_info has a df_conflict field to
448 * track such conflicts, we ignore that field for any directories at
449 * a path and only pay attention to it for files at the given path.
450 * The fact that we leave directories were they are also means that
451 * we do not need to worry about getting additional df_conflict
452 * information propagated from parent directories down to children
453 * (unlike, say traverse_trees_recursive() in unpack-trees.c, which
454 * sets a newinfo.df_conflicts field specifically to propagate it).
456 unsigned df_conflict
= (filemask
!= 0) && (dirmask
!= 0);
458 /* n = 3 is a fundamental assumption. */
460 BUG("Called collect_merge_info_callback wrong");
463 * A bunch of sanity checks verifying that traverse_trees() calls
464 * us the way I expect. Could just remove these at some point,
465 * though maybe they are helpful to future code readers.
467 assert(mbase_null
== is_null_oid(&names
[0].oid
));
468 assert(side1_null
== is_null_oid(&names
[1].oid
));
469 assert(side2_null
== is_null_oid(&names
[2].oid
));
470 assert(!mbase_null
|| !side1_null
|| !side2_null
);
471 assert(mask
> 0 && mask
< 8);
473 /* Determine match_mask */
474 if (side1_matches_mbase
)
475 match_mask
= (side2_matches_mbase
? 7 : 3);
476 else if (side2_matches_mbase
)
478 else if (sides_match
)
482 * Get the name of the relevant filepath, which we'll pass to
483 * setup_path_info() for tracking.
488 len
= traverse_path_len(info
, p
->pathlen
);
490 /* +1 in both of the following lines to include the NUL byte */
491 fullpath
= xmalloc(len
+ 1);
492 make_traverse_path(fullpath
, len
+ 1, info
, p
->path
, p
->pathlen
);
495 * If mbase, side1, and side2 all match, we can resolve early. Even
496 * if these are trees, there will be no renames or anything
499 if (side1_matches_mbase
&& side2_matches_mbase
) {
500 /* mbase, side1, & side2 all match; use mbase as resolution */
501 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
502 names
, names
+0, mbase_null
, 0,
503 filemask
, dirmask
, 1);
508 * Record information about the path so we can resolve later in
511 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
512 names
, NULL
, 0, df_conflict
, filemask
, dirmask
, 0);
516 ci
->match_mask
= match_mask
;
518 /* If dirmask, recurse into subdirectories */
520 struct traverse_info newinfo
;
521 struct tree_desc t
[3];
522 void *buf
[3] = {NULL
, NULL
, NULL
};
523 const char *original_dir_name
;
526 ci
->match_mask
&= filemask
;
529 newinfo
.name
= p
->path
;
530 newinfo
.namelen
= p
->pathlen
;
531 newinfo
.pathlen
= st_add3(newinfo
.pathlen
, p
->pathlen
, 1);
533 * If this directory we are about to recurse into cared about
534 * its parent directory (the current directory) having a D/F
535 * conflict, then we'd propagate the masks in this way:
536 * newinfo.df_conflicts |= (mask & ~dirmask);
537 * But we don't worry about propagating D/F conflicts. (See
538 * comment near setting of local df_conflict variable near
539 * the beginning of this function).
542 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
543 if (i
== 1 && side1_matches_mbase
)
545 else if (i
== 2 && side2_matches_mbase
)
547 else if (i
== 2 && sides_match
)
550 const struct object_id
*oid
= NULL
;
553 buf
[i
] = fill_tree_descriptor(opt
->repo
,
559 original_dir_name
= opti
->current_dir_name
;
560 opti
->current_dir_name
= pi
.string
;
561 ret
= traverse_trees(NULL
, 3, t
, &newinfo
);
562 opti
->current_dir_name
= original_dir_name
;
564 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++)
574 static int collect_merge_info(struct merge_options
*opt
,
575 struct tree
*merge_base
,
580 struct tree_desc t
[3];
581 struct traverse_info info
;
582 const char *toplevel_dir_placeholder
= "";
584 opt
->priv
->current_dir_name
= toplevel_dir_placeholder
;
585 setup_traverse_info(&info
, toplevel_dir_placeholder
);
586 info
.fn
= collect_merge_info_callback
;
588 info
.show_all_errors
= 1;
590 parse_tree(merge_base
);
593 init_tree_desc(t
+ 0, merge_base
->buffer
, merge_base
->size
);
594 init_tree_desc(t
+ 1, side1
->buffer
, side1
->size
);
595 init_tree_desc(t
+ 2, side2
->buffer
, side2
->size
);
597 ret
= traverse_trees(NULL
, 3, t
, &info
);
602 /*** Function Grouping: functions related to threeway content merges ***/
604 static int handle_content_merge(struct merge_options
*opt
,
606 const struct version_info
*o
,
607 const struct version_info
*a
,
608 const struct version_info
*b
,
609 const char *pathnames
[3],
610 const int extra_marker_size
,
611 struct version_info
*result
)
613 die("Not yet implemented");
616 /*** Function Grouping: functions related to detect_and_process_renames(), ***
617 *** which are split into directory and regular rename detection sections. ***/
619 /*** Function Grouping: functions related to directory rename detection ***/
621 /*** Function Grouping: functions related to regular rename detection ***/
623 static int detect_and_process_renames(struct merge_options
*opt
,
624 struct tree
*merge_base
,
631 * Rename detection works by detecting file similarity. Here we use
632 * a really easy-to-implement scheme: files are similar IFF they have
633 * the same filename. Therefore, by this scheme, there are no renames.
635 * TODO: Actually implement a real rename detection scheme.
640 /*** Function Grouping: functions related to process_entries() ***/
642 static int string_list_df_name_compare(const char *one
, const char *two
)
644 int onelen
= strlen(one
);
645 int twolen
= strlen(two
);
647 * Here we only care that entries for D/F conflicts are
648 * adjacent, in particular with the file of the D/F conflict
649 * appearing before files below the corresponding directory.
650 * The order of the rest of the list is irrelevant for us.
652 * To achieve this, we sort with df_name_compare and provide
653 * the mode S_IFDIR so that D/F conflicts will sort correctly.
654 * We use the mode S_IFDIR for everything else for simplicity,
655 * since in other cases any changes in their order due to
656 * sorting cause no problems for us.
658 int cmp
= df_name_compare(one
, onelen
, S_IFDIR
,
659 two
, twolen
, S_IFDIR
);
661 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
662 * that 'foo' comes before 'foo/bar'.
666 return onelen
- twolen
;
669 struct directory_versions
{
671 * versions: list of (basename -> version_info)
673 * The basenames are in reverse lexicographic order of full pathnames,
674 * as processed in process_entries(). This puts all entries within
675 * a directory together, and covers the directory itself after
676 * everything within it, allowing us to write subtrees before needing
677 * to record information for the tree itself.
679 struct string_list versions
;
682 * offsets: list of (full relative path directories -> integer offsets)
684 * Since versions contains basenames from files in multiple different
685 * directories, we need to know which entries in versions correspond
686 * to which directories. Values of e.g.
690 * Would mean that entries 0-1 of versions are files in the toplevel
691 * directory, entries 2-4 are files under src/, and the remaining
692 * entries starting at index 5 are files under src/moduleA/.
694 struct string_list offsets
;
697 * last_directory: directory that previously processed file found in
699 * last_directory starts NULL, but records the directory in which the
700 * previous file was found within. As soon as
701 * directory(current_file) != last_directory
702 * then we need to start updating accounting in versions & offsets.
703 * Note that last_directory is always the last path in "offsets" (or
704 * NULL if "offsets" is empty) so this exists just for quick access.
706 const char *last_directory
;
708 /* last_directory_len: cached computation of strlen(last_directory) */
709 unsigned last_directory_len
;
712 static int tree_entry_order(const void *a_
, const void *b_
)
714 const struct string_list_item
*a
= a_
;
715 const struct string_list_item
*b
= b_
;
717 const struct merged_info
*ami
= a
->util
;
718 const struct merged_info
*bmi
= b
->util
;
719 return base_name_compare(a
->string
, strlen(a
->string
), ami
->result
.mode
,
720 b
->string
, strlen(b
->string
), bmi
->result
.mode
);
723 static void write_tree(struct object_id
*result_oid
,
724 struct string_list
*versions
,
728 size_t maxlen
= 0, extra
;
729 unsigned int nr
= versions
->nr
- offset
;
730 struct strbuf buf
= STRBUF_INIT
;
731 struct string_list relevant_entries
= STRING_LIST_INIT_NODUP
;
735 * We want to sort the last (versions->nr-offset) entries in versions.
736 * Do so by abusing the string_list API a bit: make another string_list
737 * that contains just those entries and then sort them.
739 * We won't use relevant_entries again and will let it just pop off the
740 * stack, so there won't be allocation worries or anything.
742 relevant_entries
.items
= versions
->items
+ offset
;
743 relevant_entries
.nr
= versions
->nr
- offset
;
744 QSORT(relevant_entries
.items
, relevant_entries
.nr
, tree_entry_order
);
746 /* Pre-allocate some space in buf */
747 extra
= hash_size
+ 8; /* 8: 6 for mode, 1 for space, 1 for NUL char */
748 for (i
= 0; i
< nr
; i
++) {
749 maxlen
+= strlen(versions
->items
[offset
+i
].string
) + extra
;
751 strbuf_grow(&buf
, maxlen
);
753 /* Write each entry out to buf */
754 for (i
= 0; i
< nr
; i
++) {
755 struct merged_info
*mi
= versions
->items
[offset
+i
].util
;
756 struct version_info
*ri
= &mi
->result
;
757 strbuf_addf(&buf
, "%o %s%c",
759 versions
->items
[offset
+i
].string
, '\0');
760 strbuf_add(&buf
, ri
->oid
.hash
, hash_size
);
763 /* Write this object file out, and record in result_oid */
764 write_object_file(buf
.buf
, buf
.len
, tree_type
, result_oid
);
765 strbuf_release(&buf
);
768 static void record_entry_for_tree(struct directory_versions
*dir_metadata
,
770 struct merged_info
*mi
)
772 const char *basename
;
775 /* nothing to record */
778 basename
= path
+ mi
->basename_offset
;
779 assert(strchr(basename
, '/') == NULL
);
780 string_list_append(&dir_metadata
->versions
,
781 basename
)->util
= &mi
->result
;
784 static void write_completed_directory(struct merge_options
*opt
,
785 const char *new_directory_name
,
786 struct directory_versions
*info
)
788 const char *prev_dir
;
789 struct merged_info
*dir_info
= NULL
;
793 * Some explanation of info->versions and info->offsets...
795 * process_entries() iterates over all relevant files AND
796 * directories in reverse lexicographic order, and calls this
797 * function. Thus, an example of the paths that process_entries()
798 * could operate on (along with the directories for those paths
803 * src/moduleB/umm.c src/moduleB
804 * src/moduleB/stuff.h src/moduleB
805 * src/moduleB/baz.c src/moduleB
807 * src/moduleA/foo.c src/moduleA
808 * src/moduleA/bar.c src/moduleA
815 * always contains the unprocessed entries and their
816 * version_info information. For example, after the first five
817 * entries above, info->versions would be:
819 * xtract.c <xtract.c's version_info>
820 * token.txt <token.txt's version_info>
821 * umm.c <src/moduleB/umm.c's version_info>
822 * stuff.h <src/moduleB/stuff.h's version_info>
823 * baz.c <src/moduleB/baz.c's version_info>
825 * Once a subdirectory is completed we remove the entries in
826 * that subdirectory from info->versions, writing it as a tree
827 * (write_tree()). Thus, as soon as we get to src/moduleB,
828 * info->versions would be updated to
830 * xtract.c <xtract.c's version_info>
831 * token.txt <token.txt's version_info>
832 * moduleB <src/moduleB's version_info>
836 * helps us track which entries in info->versions correspond to
837 * which directories. When we are N directories deep (e.g. 4
838 * for src/modA/submod/subdir/), we have up to N+1 unprocessed
839 * directories (+1 because of toplevel dir). Corresponding to
840 * the info->versions example above, after processing five entries
841 * info->offsets will be:
846 * which is used to know that xtract.c & token.txt are from the
847 * toplevel dirctory, while umm.c & stuff.h & baz.c are from the
848 * src/moduleB directory. Again, following the example above,
849 * once we need to process src/moduleB, then info->offsets is
855 * which says that moduleB (and only moduleB so far) is in the
858 * One unique thing to note about info->offsets here is that
859 * "src" was not added to info->offsets until there was a path
860 * (a file OR directory) immediately below src/ that got
863 * Since process_entry() just appends new entries to info->versions,
864 * write_completed_directory() only needs to do work if the next path
865 * is in a directory that is different than the last directory found
870 * If we are working with the same directory as the last entry, there
871 * is no work to do. (See comments above the directory_name member of
872 * struct merged_info for why we can use pointer comparison instead of
875 if (new_directory_name
== info
->last_directory
)
879 * If we are just starting (last_directory is NULL), or last_directory
880 * is a prefix of the current directory, then we can just update
881 * info->offsets to record the offset where we started this directory
882 * and update last_directory to have quick access to it.
884 if (info
->last_directory
== NULL
||
885 !strncmp(new_directory_name
, info
->last_directory
,
886 info
->last_directory_len
)) {
887 uintptr_t offset
= info
->versions
.nr
;
889 info
->last_directory
= new_directory_name
;
890 info
->last_directory_len
= strlen(info
->last_directory
);
892 * Record the offset into info->versions where we will
893 * start recording basenames of paths found within
894 * new_directory_name.
896 string_list_append(&info
->offsets
,
897 info
->last_directory
)->util
= (void*)offset
;
902 * The next entry that will be processed will be within
903 * new_directory_name. Since at this point we know that
904 * new_directory_name is within a different directory than
905 * info->last_directory, we have all entries for info->last_directory
906 * in info->versions and we need to create a tree object for them.
908 dir_info
= strmap_get(&opt
->priv
->paths
, info
->last_directory
);
910 offset
= (uintptr_t)info
->offsets
.items
[info
->offsets
.nr
-1].util
;
911 if (offset
== info
->versions
.nr
) {
913 * Actually, we don't need to create a tree object in this
914 * case. Whenever all files within a directory disappear
915 * during the merge (e.g. unmodified on one side and
916 * deleted on the other, or files were renamed elsewhere),
917 * then we get here and the directory itself needs to be
918 * omitted from its parent tree as well.
920 dir_info
->is_null
= 1;
923 * Write out the tree to the git object directory, and also
924 * record the mode and oid in dir_info->result.
926 dir_info
->is_null
= 0;
927 dir_info
->result
.mode
= S_IFDIR
;
928 write_tree(&dir_info
->result
.oid
, &info
->versions
, offset
,
929 opt
->repo
->hash_algo
->rawsz
);
933 * We've now used several entries from info->versions and one entry
934 * from info->offsets, so we get rid of those values.
937 info
->versions
.nr
= offset
;
940 * Now we've taken care of the completed directory, but we need to
941 * prepare things since future entries will be in
942 * new_directory_name. (In particular, process_entry() will be
943 * appending new entries to info->versions.) So, we need to make
944 * sure new_directory_name is the last entry in info->offsets.
946 prev_dir
= info
->offsets
.nr
== 0 ? NULL
:
947 info
->offsets
.items
[info
->offsets
.nr
-1].string
;
948 if (new_directory_name
!= prev_dir
) {
949 uintptr_t c
= info
->versions
.nr
;
950 string_list_append(&info
->offsets
,
951 new_directory_name
)->util
= (void*)c
;
954 /* And, of course, we need to update last_directory to match. */
955 info
->last_directory
= new_directory_name
;
956 info
->last_directory_len
= strlen(info
->last_directory
);
959 /* Per entry merge function */
960 static void process_entry(struct merge_options
*opt
,
962 struct conflict_info
*ci
,
963 struct directory_versions
*dir_metadata
)
966 assert(ci
->filemask
>= 0 && ci
->filemask
<= 7);
967 /* ci->match_mask == 7 was handled in collect_merge_info_callback() */
968 assert(ci
->match_mask
== 0 || ci
->match_mask
== 3 ||
969 ci
->match_mask
== 5 || ci
->match_mask
== 6);
972 record_entry_for_tree(dir_metadata
, path
, &ci
->merged
);
973 if (ci
->filemask
== 0)
974 /* nothing else to handle */
976 assert(ci
->df_conflict
);
979 if (ci
->df_conflict
) {
980 die("Not yet implemented.");
984 * NOTE: Below there is a long switch-like if-elseif-elseif... block
985 * which the code goes through even for the df_conflict cases
986 * above. Well, it will once we don't die-not-implemented above.
988 if (ci
->match_mask
) {
989 ci
->merged
.clean
= 1;
990 if (ci
->match_mask
== 6) {
991 /* stages[1] == stages[2] */
992 ci
->merged
.result
.mode
= ci
->stages
[1].mode
;
993 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[1].oid
);
995 /* determine the mask of the side that didn't match */
996 unsigned int othermask
= 7 & ~ci
->match_mask
;
997 int side
= (othermask
== 4) ? 2 : 1;
999 ci
->merged
.result
.mode
= ci
->stages
[side
].mode
;
1000 ci
->merged
.is_null
= !ci
->merged
.result
.mode
;
1001 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[side
].oid
);
1003 assert(othermask
== 2 || othermask
== 4);
1004 assert(ci
->merged
.is_null
==
1005 (ci
->filemask
== ci
->match_mask
));
1007 } else if (ci
->filemask
>= 6 &&
1008 (S_IFMT
& ci
->stages
[1].mode
) !=
1009 (S_IFMT
& ci
->stages
[2].mode
)) {
1011 * Two different items from (file/submodule/symlink)
1013 die("Not yet implemented.");
1014 } else if (ci
->filemask
>= 6) {
1016 * TODO: Needs a two-way or three-way content merge, but we're
1017 * just being lazy and copying the version from HEAD and
1018 * leaving it as conflicted.
1020 ci
->merged
.clean
= 0;
1021 ci
->merged
.result
.mode
= ci
->stages
[1].mode
;
1022 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[1].oid
);
1023 /* When we fix above, we'll call handle_content_merge() */
1024 (void)handle_content_merge
;
1025 } else if (ci
->filemask
== 3 || ci
->filemask
== 5) {
1027 const char *modify_branch
, *delete_branch
;
1028 int side
= (ci
->filemask
== 5) ? 2 : 1;
1029 int index
= opt
->priv
->call_depth
? 0 : side
;
1031 ci
->merged
.result
.mode
= ci
->stages
[index
].mode
;
1032 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[index
].oid
);
1033 ci
->merged
.clean
= 0;
1035 modify_branch
= (side
== 1) ? opt
->branch1
: opt
->branch2
;
1036 delete_branch
= (side
== 1) ? opt
->branch2
: opt
->branch1
;
1038 path_msg(opt
, path
, 0,
1039 _("CONFLICT (modify/delete): %s deleted in %s "
1040 "and modified in %s. Version %s of %s left "
1042 path
, delete_branch
, modify_branch
,
1043 modify_branch
, path
);
1044 } else if (ci
->filemask
== 2 || ci
->filemask
== 4) {
1045 /* Added on one side */
1046 int side
= (ci
->filemask
== 4) ? 2 : 1;
1047 ci
->merged
.result
.mode
= ci
->stages
[side
].mode
;
1048 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[side
].oid
);
1049 ci
->merged
.clean
= !ci
->df_conflict
;
1050 } else if (ci
->filemask
== 1) {
1051 /* Deleted on both sides */
1052 ci
->merged
.is_null
= 1;
1053 ci
->merged
.result
.mode
= 0;
1054 oidcpy(&ci
->merged
.result
.oid
, &null_oid
);
1055 ci
->merged
.clean
= 1;
1059 * If still conflicted, record it separately. This allows us to later
1060 * iterate over just conflicted entries when updating the index instead
1061 * of iterating over all entries.
1063 if (!ci
->merged
.clean
)
1064 strmap_put(&opt
->priv
->conflicted
, path
, ci
);
1065 record_entry_for_tree(dir_metadata
, path
, &ci
->merged
);
1068 static void process_entries(struct merge_options
*opt
,
1069 struct object_id
*result_oid
)
1071 struct hashmap_iter iter
;
1072 struct strmap_entry
*e
;
1073 struct string_list plist
= STRING_LIST_INIT_NODUP
;
1074 struct string_list_item
*entry
;
1075 struct directory_versions dir_metadata
= { STRING_LIST_INIT_NODUP
,
1076 STRING_LIST_INIT_NODUP
,
1079 if (strmap_empty(&opt
->priv
->paths
)) {
1080 oidcpy(result_oid
, opt
->repo
->hash_algo
->empty_tree
);
1084 /* Hack to pre-allocate plist to the desired size */
1085 ALLOC_GROW(plist
.items
, strmap_get_size(&opt
->priv
->paths
), plist
.alloc
);
1087 /* Put every entry from paths into plist, then sort */
1088 strmap_for_each_entry(&opt
->priv
->paths
, &iter
, e
) {
1089 string_list_append(&plist
, e
->key
)->util
= e
->value
;
1091 plist
.cmp
= string_list_df_name_compare
;
1092 string_list_sort(&plist
);
1095 * Iterate over the items in reverse order, so we can handle paths
1096 * below a directory before needing to handle the directory itself.
1098 * This allows us to write subtrees before we need to write trees,
1099 * and it also enables sane handling of directory/file conflicts
1100 * (because it allows us to know whether the directory is still in
1101 * the way when it is time to process the file at the same path).
1103 for (entry
= &plist
.items
[plist
.nr
-1]; entry
>= plist
.items
; --entry
) {
1104 char *path
= entry
->string
;
1106 * NOTE: mi may actually be a pointer to a conflict_info, but
1107 * we have to check mi->clean first to see if it's safe to
1108 * reassign to such a pointer type.
1110 struct merged_info
*mi
= entry
->util
;
1112 write_completed_directory(opt
, mi
->directory_name
,
1115 record_entry_for_tree(&dir_metadata
, path
, mi
);
1117 struct conflict_info
*ci
= (struct conflict_info
*)mi
;
1118 process_entry(opt
, path
, ci
, &dir_metadata
);
1122 if (dir_metadata
.offsets
.nr
!= 1 ||
1123 (uintptr_t)dir_metadata
.offsets
.items
[0].util
!= 0) {
1124 printf("dir_metadata.offsets.nr = %d (should be 1)\n",
1125 dir_metadata
.offsets
.nr
);
1126 printf("dir_metadata.offsets.items[0].util = %u (should be 0)\n",
1127 (unsigned)(uintptr_t)dir_metadata
.offsets
.items
[0].util
);
1129 BUG("dir_metadata accounting completely off; shouldn't happen");
1131 write_tree(result_oid
, &dir_metadata
.versions
, 0,
1132 opt
->repo
->hash_algo
->rawsz
);
1133 string_list_clear(&plist
, 0);
1134 string_list_clear(&dir_metadata
.versions
, 0);
1135 string_list_clear(&dir_metadata
.offsets
, 0);
1138 /*** Function Grouping: functions related to merge_switch_to_result() ***/
1140 static int checkout(struct merge_options
*opt
,
1144 /* Switch the index/working copy from old to new */
1146 struct tree_desc trees
[2];
1147 struct unpack_trees_options unpack_opts
;
1149 memset(&unpack_opts
, 0, sizeof(unpack_opts
));
1150 unpack_opts
.head_idx
= -1;
1151 unpack_opts
.src_index
= opt
->repo
->index
;
1152 unpack_opts
.dst_index
= opt
->repo
->index
;
1154 setup_unpack_trees_porcelain(&unpack_opts
, "merge");
1157 * NOTE: if this were just "git checkout" code, we would probably
1158 * read or refresh the cache and check for a conflicted index, but
1159 * builtin/merge.c or sequencer.c really needs to read the index
1160 * and check for conflicted entries before starting merging for a
1161 * good user experience (no sense waiting for merges/rebases before
1162 * erroring out), so there's no reason to duplicate that work here.
1165 /* 2-way merge to the new branch */
1166 unpack_opts
.update
= 1;
1167 unpack_opts
.merge
= 1;
1168 unpack_opts
.quiet
= 0; /* FIXME: sequencer might want quiet? */
1169 unpack_opts
.verbose_update
= (opt
->verbosity
> 2);
1170 unpack_opts
.fn
= twoway_merge
;
1171 if (1/* FIXME: opts->overwrite_ignore*/) {
1172 unpack_opts
.dir
= xcalloc(1, sizeof(*unpack_opts
.dir
));
1173 unpack_opts
.dir
->flags
|= DIR_SHOW_IGNORED
;
1174 setup_standard_excludes(unpack_opts
.dir
);
1177 init_tree_desc(&trees
[0], prev
->buffer
, prev
->size
);
1179 init_tree_desc(&trees
[1], next
->buffer
, next
->size
);
1181 ret
= unpack_trees(2, trees
, &unpack_opts
);
1182 clear_unpack_trees_porcelain(&unpack_opts
);
1183 dir_clear(unpack_opts
.dir
);
1184 FREE_AND_NULL(unpack_opts
.dir
);
1188 static int record_conflicted_index_entries(struct merge_options
*opt
,
1189 struct index_state
*index
,
1190 struct strmap
*paths
,
1191 struct strmap
*conflicted
)
1193 struct hashmap_iter iter
;
1194 struct strmap_entry
*e
;
1196 int original_cache_nr
;
1198 if (strmap_empty(conflicted
))
1201 original_cache_nr
= index
->cache_nr
;
1203 /* Put every entry from paths into plist, then sort */
1204 strmap_for_each_entry(conflicted
, &iter
, e
) {
1205 const char *path
= e
->key
;
1206 struct conflict_info
*ci
= e
->value
;
1208 struct cache_entry
*ce
;
1214 * The index will already have a stage=0 entry for this path,
1215 * because we created an as-merged-as-possible version of the
1216 * file and checkout() moved the working copy and index over
1219 * However, previous iterations through this loop will have
1220 * added unstaged entries to the end of the cache which
1221 * ignore the standard alphabetical ordering of cache
1222 * entries and break invariants needed for index_name_pos()
1223 * to work. However, we know the entry we want is before
1224 * those appended cache entries, so do a temporary swap on
1225 * cache_nr to only look through entries of interest.
1227 SWAP(index
->cache_nr
, original_cache_nr
);
1228 pos
= index_name_pos(index
, path
, strlen(path
));
1229 SWAP(index
->cache_nr
, original_cache_nr
);
1231 if (ci
->filemask
!= 1)
1232 BUG("Conflicted %s but nothing in basic working tree or index; this shouldn't happen", path
);
1233 cache_tree_invalidate_path(index
, path
);
1235 ce
= index
->cache
[pos
];
1238 * Clean paths with CE_SKIP_WORKTREE set will not be
1239 * written to the working tree by the unpack_trees()
1240 * call in checkout(). Our conflicted entries would
1241 * have appeared clean to that code since we ignored
1242 * the higher order stages. Thus, we need override
1243 * the CE_SKIP_WORKTREE bit and manually write those
1244 * files to the working disk here.
1246 * TODO: Implement this CE_SKIP_WORKTREE fixup.
1250 * Mark this cache entry for removal and instead add
1251 * new stage>0 entries corresponding to the
1252 * conflicts. If there are many conflicted entries, we
1253 * want to avoid memmove'ing O(NM) entries by
1254 * inserting the new entries one at a time. So,
1255 * instead, we just add the new cache entries to the
1256 * end (ignoring normal index requirements on sort
1257 * order) and sort the index once we're all done.
1259 ce
->ce_flags
|= CE_REMOVE
;
1262 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
1263 struct version_info
*vi
;
1264 if (!(ci
->filemask
& (1ul << i
)))
1266 vi
= &ci
->stages
[i
];
1267 ce
= make_cache_entry(index
, vi
->mode
, &vi
->oid
,
1269 add_index_entry(index
, ce
, ADD_CACHE_JUST_APPEND
);
1274 * Remove the unused cache entries (and invalidate the relevant
1275 * cache-trees), then sort the index entries to get the conflicted
1276 * entries we added to the end into their right locations.
1278 remove_marked_cache_entries(index
, 1);
1279 QSORT(index
->cache
, index
->cache_nr
, cmp_cache_name_compare
);
1284 void merge_switch_to_result(struct merge_options
*opt
,
1286 struct merge_result
*result
,
1287 int update_worktree_and_index
,
1288 int display_update_msgs
)
1290 assert(opt
->priv
== NULL
);
1291 if (result
->clean
>= 0 && update_worktree_and_index
) {
1292 struct merge_options_internal
*opti
= result
->priv
;
1294 if (checkout(opt
, head
, result
->tree
)) {
1295 /* failure to function */
1300 if (record_conflicted_index_entries(opt
, opt
->repo
->index
,
1302 &opti
->conflicted
)) {
1303 /* failure to function */
1309 if (display_update_msgs
) {
1310 struct merge_options_internal
*opti
= result
->priv
;
1311 struct hashmap_iter iter
;
1312 struct strmap_entry
*e
;
1313 struct string_list olist
= STRING_LIST_INIT_NODUP
;
1316 /* Hack to pre-allocate olist to the desired size */
1317 ALLOC_GROW(olist
.items
, strmap_get_size(&opti
->output
),
1320 /* Put every entry from output into olist, then sort */
1321 strmap_for_each_entry(&opti
->output
, &iter
, e
) {
1322 string_list_append(&olist
, e
->key
)->util
= e
->value
;
1324 string_list_sort(&olist
);
1326 /* Iterate over the items, printing them */
1327 for (i
= 0; i
< olist
.nr
; ++i
) {
1328 struct strbuf
*sb
= olist
.items
[i
].util
;
1330 printf("%s", sb
->buf
);
1332 string_list_clear(&olist
, 0);
1335 merge_finalize(opt
, result
);
1338 void merge_finalize(struct merge_options
*opt
,
1339 struct merge_result
*result
)
1341 struct merge_options_internal
*opti
= result
->priv
;
1343 assert(opt
->priv
== NULL
);
1345 clear_internal_opts(opti
, 0);
1346 FREE_AND_NULL(opti
);
1349 /*** Function Grouping: helper functions for merge_incore_*() ***/
1351 static void merge_start(struct merge_options
*opt
, struct merge_result
*result
)
1353 /* Sanity checks on opt */
1356 assert(opt
->branch1
&& opt
->branch2
);
1358 assert(opt
->detect_directory_renames
>= MERGE_DIRECTORY_RENAMES_NONE
&&
1359 opt
->detect_directory_renames
<= MERGE_DIRECTORY_RENAMES_TRUE
);
1360 assert(opt
->rename_limit
>= -1);
1361 assert(opt
->rename_score
>= 0 && opt
->rename_score
<= MAX_SCORE
);
1362 assert(opt
->show_rename_progress
>= 0 && opt
->show_rename_progress
<= 1);
1364 assert(opt
->xdl_opts
>= 0);
1365 assert(opt
->recursive_variant
>= MERGE_VARIANT_NORMAL
&&
1366 opt
->recursive_variant
<= MERGE_VARIANT_THEIRS
);
1369 * detect_renames, verbosity, buffer_output, and obuf are ignored
1370 * fields that were used by "recursive" rather than "ort" -- but
1371 * sanity check them anyway.
1373 assert(opt
->detect_renames
>= -1 &&
1374 opt
->detect_renames
<= DIFF_DETECT_COPY
);
1375 assert(opt
->verbosity
>= 0 && opt
->verbosity
<= 5);
1376 assert(opt
->buffer_output
<= 2);
1377 assert(opt
->obuf
.len
== 0);
1379 assert(opt
->priv
== NULL
);
1381 /* Default to histogram diff. Actually, just hardcode it...for now. */
1382 opt
->xdl_opts
= DIFF_WITH_ALG(opt
, HISTOGRAM_DIFF
);
1384 /* Initialization of opt->priv, our internal merge data */
1385 opt
->priv
= xcalloc(1, sizeof(*opt
->priv
));
1388 * Although we initialize opt->priv->paths with strdup_strings=0,
1389 * that's just to avoid making yet another copy of an allocated
1390 * string. Putting the entry into paths means we are taking
1391 * ownership, so we will later free it. paths_to_free is similar.
1393 * In contrast, conflicted just has a subset of keys from paths, so
1394 * we don't want to free those (it'd be a duplicate free).
1396 strmap_init_with_options(&opt
->priv
->paths
, NULL
, 0);
1397 strmap_init_with_options(&opt
->priv
->conflicted
, NULL
, 0);
1398 string_list_init(&opt
->priv
->paths_to_free
, 0);
1401 * keys & strbufs in output will sometimes need to outlive "paths",
1402 * so it will have a copy of relevant keys. It's probably a small
1403 * subset of the overall paths that have special output.
1405 strmap_init(&opt
->priv
->output
);
1408 /*** Function Grouping: merge_incore_*() and their internal variants ***/
1411 * Originally from merge_trees_internal(); heavily adapted, though.
1413 static void merge_ort_nonrecursive_internal(struct merge_options
*opt
,
1414 struct tree
*merge_base
,
1417 struct merge_result
*result
)
1419 struct object_id working_tree_oid
;
1421 if (collect_merge_info(opt
, merge_base
, side1
, side2
) != 0) {
1423 * TRANSLATORS: The %s arguments are: 1) tree hash of a merge
1424 * base, and 2-3) the trees for the two trees we're merging.
1426 err(opt
, _("collecting merge info failed for trees %s, %s, %s"),
1427 oid_to_hex(&merge_base
->object
.oid
),
1428 oid_to_hex(&side1
->object
.oid
),
1429 oid_to_hex(&side2
->object
.oid
));
1434 result
->clean
= detect_and_process_renames(opt
, merge_base
,
1436 process_entries(opt
, &working_tree_oid
);
1438 /* Set return values */
1439 result
->tree
= parse_tree_indirect(&working_tree_oid
);
1440 /* existence of conflicted entries implies unclean */
1441 result
->clean
&= strmap_empty(&opt
->priv
->conflicted
);
1442 if (!opt
->priv
->call_depth
) {
1443 result
->priv
= opt
->priv
;
1448 void merge_incore_nonrecursive(struct merge_options
*opt
,
1449 struct tree
*merge_base
,
1452 struct merge_result
*result
)
1454 assert(opt
->ancestor
!= NULL
);
1455 merge_start(opt
, result
);
1456 merge_ort_nonrecursive_internal(opt
, merge_base
, side1
, side2
, result
);
1459 void merge_incore_recursive(struct merge_options
*opt
,
1460 struct commit_list
*merge_bases
,
1461 struct commit
*side1
,
1462 struct commit
*side2
,
1463 struct merge_result
*result
)
1465 die("Not yet implemented");