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"
20 #include "cache-tree.h"
24 #include "object-store.h"
27 #include "unpack-trees.h"
28 #include "xdiff-interface.h"
31 * We have many arrays of size 3. Whenever we have such an array, the
32 * indices refer to one of the sides of the three-way merge. This is so
33 * pervasive that the constants 0, 1, and 2 are used in many places in the
34 * code (especially in arithmetic operations to find the other side's index
35 * or to compute a relevant mask), but sometimes these enum names are used
36 * to aid code clarity.
38 * See also 'filemask' and 'dirmask' in struct conflict_info; the "ith side"
39 * referred to there is one of these three sides.
47 struct merge_options_internal
{
49 * paths: primary data structure in all of merge ort.
52 * * are full relative paths from the toplevel of the repository
53 * (e.g. "drivers/firmware/raspberrypi.c").
54 * * store all relevant paths in the repo, both directories and
55 * files (e.g. drivers, drivers/firmware would also be included)
56 * * these keys serve to intern all the path strings, which allows
57 * us to do pointer comparison on directory names instead of
58 * strcmp; we just have to be careful to use the interned strings.
60 * The values of paths:
61 * * either a pointer to a merged_info, or a conflict_info struct
62 * * merged_info contains all relevant information for a
63 * non-conflicted entry.
64 * * conflict_info contains a merged_info, plus any additional
65 * information about a conflict such as the higher orders stages
66 * involved and the names of the paths those came from (handy
67 * once renames get involved).
68 * * a path may start "conflicted" (i.e. point to a conflict_info)
69 * and then a later step (e.g. three-way content merge) determines
70 * it can be cleanly merged, at which point it'll be marked clean
71 * and the algorithm will ignore any data outside the contained
72 * merged_info for that entry
73 * * If an entry remains conflicted, the merged_info portion of a
74 * conflict_info will later be filled with whatever version of
75 * the file should be placed in the working directory (e.g. an
76 * as-merged-as-possible variation that contains conflict markers).
81 * conflicted: a subset of keys->values from "paths"
83 * conflicted is basically an optimization between process_entries()
84 * and record_conflicted_index_entries(); the latter could loop over
85 * ALL the entries in paths AGAIN and look for the ones that are
86 * still conflicted, but since process_entries() has to loop over
87 * all of them, it saves the ones it couldn't resolve in this strmap
88 * so that record_conflicted_index_entries() can iterate just the
91 struct strmap conflicted
;
94 * current_dir_name: temporary var used in collect_merge_info_callback()
96 * Used to set merged_info.directory_name; see documentation for that
97 * variable and the requirements placed on that field.
99 const char *current_dir_name
;
101 /* call_depth: recursion level counter for merging merge bases */
105 struct version_info
{
106 struct object_id oid
;
111 /* if is_null, ignore result. otherwise result has oid & mode */
112 struct version_info result
;
116 * clean: whether the path in question is cleanly merged.
118 * see conflict_info.merged for more details.
123 * basename_offset: offset of basename of path.
125 * perf optimization to avoid recomputing offset of final '/'
126 * character in pathname (0 if no '/' in pathname).
128 size_t basename_offset
;
131 * directory_name: containing directory name.
133 * Note that we assume directory_name is constructed such that
134 * strcmp(dir1_name, dir2_name) == 0 iff dir1_name == dir2_name,
135 * i.e. string equality is equivalent to pointer equality. For this
136 * to hold, we have to be careful setting directory_name.
138 const char *directory_name
;
141 struct conflict_info
{
143 * merged: the version of the path that will be written to working tree
145 * WARNING: It is critical to check merged.clean and ensure it is 0
146 * before reading any conflict_info fields outside of merged.
147 * Allocated merge_info structs will always have clean set to 1.
148 * Allocated conflict_info structs will have merged.clean set to 0
149 * initially. The merged.clean field is how we know if it is safe
150 * to access other parts of conflict_info besides merged; if a
151 * conflict_info's merged.clean is changed to 1, the rest of the
152 * algorithm is not allowed to look at anything outside of the
153 * merged member anymore.
155 struct merged_info merged
;
157 /* oids & modes from each of the three trees for this path */
158 struct version_info stages
[3];
160 /* pathnames for each stage; may differ due to rename detection */
161 const char *pathnames
[3];
163 /* Whether this path is/was involved in a directory/file conflict */
164 unsigned df_conflict
:1;
167 * For filemask and dirmask, the ith bit corresponds to whether the
168 * ith entry is a file (filemask) or a directory (dirmask). Thus,
169 * filemask & dirmask is always zero, and filemask | dirmask is at
170 * most 7 but can be less when a path does not appear as either a
171 * file or a directory on at least one side of history.
173 * Note that these masks are related to enum merge_side, as the ith
174 * entry corresponds to side i.
176 * These values come from a traverse_trees() call; more info may be
177 * found looking at tree-walk.h's struct traverse_info,
178 * particularly the documentation above the "fn" member (note that
179 * filemask = mask & ~dirmask from that documentation).
185 * Optimization to track which stages match, to avoid the need to
186 * recompute it in multiple steps. Either 0 or at least 2 bits are
187 * set; if at least 2 bits are set, their corresponding stages match.
189 unsigned match_mask
:3;
193 * For the next three macros, see warning for conflict_info.merged.
195 * In each of the below, mi is a struct merged_info*, and ci was defined
196 * as a struct conflict_info* (but we need to verify ci isn't actually
197 * pointed at a struct merged_info*).
199 * INITIALIZE_CI: Assign ci to mi but only if it's safe; set to NULL otherwise.
200 * VERIFY_CI: Ensure that something we assigned to a conflict_info* is one.
201 * ASSIGN_AND_VERIFY_CI: Similar to VERIFY_CI but do assignment first.
203 #define INITIALIZE_CI(ci, mi) do { \
204 (ci) = (!(mi) || (mi)->clean) ? NULL : (struct conflict_info *)(mi); \
206 #define VERIFY_CI(ci) assert(ci && !ci->merged.clean);
207 #define ASSIGN_AND_VERIFY_CI(ci, mi) do { \
208 (ci) = (struct conflict_info *)(mi); \
209 assert((ci) && !(mi)->clean); \
212 static void free_strmap_strings(struct strmap
*map
)
214 struct hashmap_iter iter
;
215 struct strmap_entry
*entry
;
217 strmap_for_each_entry(map
, &iter
, entry
) {
218 free((char*)entry
->key
);
222 static int err(struct merge_options
*opt
, const char *err
, ...)
225 struct strbuf sb
= STRBUF_INIT
;
227 strbuf_addstr(&sb
, "error: ");
228 va_start(params
, err
);
229 strbuf_vaddf(&sb
, err
, params
);
238 static void setup_path_info(struct merge_options
*opt
,
239 struct string_list_item
*result
,
240 const char *current_dir_name
,
241 int current_dir_name_len
,
242 char *fullpath
, /* we'll take over ownership */
243 struct name_entry
*names
,
244 struct name_entry
*merged_version
,
245 unsigned is_null
, /* boolean */
246 unsigned df_conflict
, /* boolean */
249 int resolved
/* boolean */)
251 /* result->util is void*, so mi is a convenience typed variable */
252 struct merged_info
*mi
;
254 assert(!is_null
|| resolved
);
255 assert(!df_conflict
|| !resolved
); /* df_conflict implies !resolved */
256 assert(resolved
== (merged_version
!= NULL
));
258 mi
= xcalloc(1, resolved
? sizeof(struct merged_info
) :
259 sizeof(struct conflict_info
));
260 mi
->directory_name
= current_dir_name
;
261 mi
->basename_offset
= current_dir_name_len
;
262 mi
->clean
= !!resolved
;
264 mi
->result
.mode
= merged_version
->mode
;
265 oidcpy(&mi
->result
.oid
, &merged_version
->oid
);
266 mi
->is_null
= !!is_null
;
269 struct conflict_info
*ci
;
271 ASSIGN_AND_VERIFY_CI(ci
, mi
);
272 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
273 ci
->pathnames
[i
] = fullpath
;
274 ci
->stages
[i
].mode
= names
[i
].mode
;
275 oidcpy(&ci
->stages
[i
].oid
, &names
[i
].oid
);
277 ci
->filemask
= filemask
;
278 ci
->dirmask
= dirmask
;
279 ci
->df_conflict
= !!df_conflict
;
282 * Assume is_null for now, but if we have entries
283 * under the directory then when it is complete in
284 * write_completed_directory() it'll update this.
285 * Also, for D/F conflicts, we have to handle the
286 * directory first, then clear this bit and process
287 * the file to see how it is handled -- that occurs
288 * near the top of process_entry().
292 strmap_put(&opt
->priv
->paths
, fullpath
, mi
);
293 result
->string
= fullpath
;
297 static int collect_merge_info_callback(int n
,
299 unsigned long dirmask
,
300 struct name_entry
*names
,
301 struct traverse_info
*info
)
305 * common ancestor (mbase) has mask 1, and stored in index 0 of names
306 * head of side 1 (side1) has mask 2, and stored in index 1 of names
307 * head of side 2 (side2) has mask 4, and stored in index 2 of names
309 struct merge_options
*opt
= info
->data
;
310 struct merge_options_internal
*opti
= opt
->priv
;
311 struct string_list_item pi
; /* Path Info */
312 struct conflict_info
*ci
; /* typed alias to pi.util (which is void*) */
313 struct name_entry
*p
;
316 const char *dirname
= opti
->current_dir_name
;
317 unsigned filemask
= mask
& ~dirmask
;
318 unsigned match_mask
= 0; /* will be updated below */
319 unsigned mbase_null
= !(mask
& 1);
320 unsigned side1_null
= !(mask
& 2);
321 unsigned side2_null
= !(mask
& 4);
322 unsigned side1_matches_mbase
= (!side1_null
&& !mbase_null
&&
323 names
[0].mode
== names
[1].mode
&&
324 oideq(&names
[0].oid
, &names
[1].oid
));
325 unsigned side2_matches_mbase
= (!side2_null
&& !mbase_null
&&
326 names
[0].mode
== names
[2].mode
&&
327 oideq(&names
[0].oid
, &names
[2].oid
));
328 unsigned sides_match
= (!side1_null
&& !side2_null
&&
329 names
[1].mode
== names
[2].mode
&&
330 oideq(&names
[1].oid
, &names
[2].oid
));
333 * Note: When a path is a file on one side of history and a directory
334 * in another, we have a directory/file conflict. In such cases, if
335 * the conflict doesn't resolve from renames and deletions, then we
336 * always leave directories where they are and move files out of the
337 * way. Thus, while struct conflict_info has a df_conflict field to
338 * track such conflicts, we ignore that field for any directories at
339 * a path and only pay attention to it for files at the given path.
340 * The fact that we leave directories were they are also means that
341 * we do not need to worry about getting additional df_conflict
342 * information propagated from parent directories down to children
343 * (unlike, say traverse_trees_recursive() in unpack-trees.c, which
344 * sets a newinfo.df_conflicts field specifically to propagate it).
346 unsigned df_conflict
= (filemask
!= 0) && (dirmask
!= 0);
348 /* n = 3 is a fundamental assumption. */
350 BUG("Called collect_merge_info_callback wrong");
353 * A bunch of sanity checks verifying that traverse_trees() calls
354 * us the way I expect. Could just remove these at some point,
355 * though maybe they are helpful to future code readers.
357 assert(mbase_null
== is_null_oid(&names
[0].oid
));
358 assert(side1_null
== is_null_oid(&names
[1].oid
));
359 assert(side2_null
== is_null_oid(&names
[2].oid
));
360 assert(!mbase_null
|| !side1_null
|| !side2_null
);
361 assert(mask
> 0 && mask
< 8);
363 /* Determine match_mask */
364 if (side1_matches_mbase
)
365 match_mask
= (side2_matches_mbase
? 7 : 3);
366 else if (side2_matches_mbase
)
368 else if (sides_match
)
372 * Get the name of the relevant filepath, which we'll pass to
373 * setup_path_info() for tracking.
378 len
= traverse_path_len(info
, p
->pathlen
);
380 /* +1 in both of the following lines to include the NUL byte */
381 fullpath
= xmalloc(len
+ 1);
382 make_traverse_path(fullpath
, len
+ 1, info
, p
->path
, p
->pathlen
);
385 * If mbase, side1, and side2 all match, we can resolve early. Even
386 * if these are trees, there will be no renames or anything
389 if (side1_matches_mbase
&& side2_matches_mbase
) {
390 /* mbase, side1, & side2 all match; use mbase as resolution */
391 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
392 names
, names
+0, mbase_null
, 0,
393 filemask
, dirmask
, 1);
398 * Record information about the path so we can resolve later in
401 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
402 names
, NULL
, 0, df_conflict
, filemask
, dirmask
, 0);
406 ci
->match_mask
= match_mask
;
408 /* If dirmask, recurse into subdirectories */
410 struct traverse_info newinfo
;
411 struct tree_desc t
[3];
412 void *buf
[3] = {NULL
, NULL
, NULL
};
413 const char *original_dir_name
;
416 ci
->match_mask
&= filemask
;
419 newinfo
.name
= p
->path
;
420 newinfo
.namelen
= p
->pathlen
;
421 newinfo
.pathlen
= st_add3(newinfo
.pathlen
, p
->pathlen
, 1);
423 * If this directory we are about to recurse into cared about
424 * its parent directory (the current directory) having a D/F
425 * conflict, then we'd propagate the masks in this way:
426 * newinfo.df_conflicts |= (mask & ~dirmask);
427 * But we don't worry about propagating D/F conflicts. (See
428 * comment near setting of local df_conflict variable near
429 * the beginning of this function).
432 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
433 if (i
== 1 && side1_matches_mbase
)
435 else if (i
== 2 && side2_matches_mbase
)
437 else if (i
== 2 && sides_match
)
440 const struct object_id
*oid
= NULL
;
443 buf
[i
] = fill_tree_descriptor(opt
->repo
,
449 original_dir_name
= opti
->current_dir_name
;
450 opti
->current_dir_name
= pi
.string
;
451 ret
= traverse_trees(NULL
, 3, t
, &newinfo
);
452 opti
->current_dir_name
= original_dir_name
;
454 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++)
464 static int collect_merge_info(struct merge_options
*opt
,
465 struct tree
*merge_base
,
470 struct tree_desc t
[3];
471 struct traverse_info info
;
472 const char *toplevel_dir_placeholder
= "";
474 opt
->priv
->current_dir_name
= toplevel_dir_placeholder
;
475 setup_traverse_info(&info
, toplevel_dir_placeholder
);
476 info
.fn
= collect_merge_info_callback
;
478 info
.show_all_errors
= 1;
480 parse_tree(merge_base
);
483 init_tree_desc(t
+ 0, merge_base
->buffer
, merge_base
->size
);
484 init_tree_desc(t
+ 1, side1
->buffer
, side1
->size
);
485 init_tree_desc(t
+ 2, side2
->buffer
, side2
->size
);
487 ret
= traverse_trees(NULL
, 3, t
, &info
);
492 static int detect_and_process_renames(struct merge_options
*opt
,
493 struct tree
*merge_base
,
500 * Rename detection works by detecting file similarity. Here we use
501 * a really easy-to-implement scheme: files are similar IFF they have
502 * the same filename. Therefore, by this scheme, there are no renames.
504 * TODO: Actually implement a real rename detection scheme.
509 static int string_list_df_name_compare(const char *one
, const char *two
)
511 int onelen
= strlen(one
);
512 int twolen
= strlen(two
);
514 * Here we only care that entries for D/F conflicts are
515 * adjacent, in particular with the file of the D/F conflict
516 * appearing before files below the corresponding directory.
517 * The order of the rest of the list is irrelevant for us.
519 * To achieve this, we sort with df_name_compare and provide
520 * the mode S_IFDIR so that D/F conflicts will sort correctly.
521 * We use the mode S_IFDIR for everything else for simplicity,
522 * since in other cases any changes in their order due to
523 * sorting cause no problems for us.
525 int cmp
= df_name_compare(one
, onelen
, S_IFDIR
,
526 two
, twolen
, S_IFDIR
);
528 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
529 * that 'foo' comes before 'foo/bar'.
533 return onelen
- twolen
;
536 struct directory_versions
{
538 * versions: list of (basename -> version_info)
540 * The basenames are in reverse lexicographic order of full pathnames,
541 * as processed in process_entries(). This puts all entries within
542 * a directory together, and covers the directory itself after
543 * everything within it, allowing us to write subtrees before needing
544 * to record information for the tree itself.
546 struct string_list versions
;
549 * offsets: list of (full relative path directories -> integer offsets)
551 * Since versions contains basenames from files in multiple different
552 * directories, we need to know which entries in versions correspond
553 * to which directories. Values of e.g.
557 * Would mean that entries 0-1 of versions are files in the toplevel
558 * directory, entries 2-4 are files under src/, and the remaining
559 * entries starting at index 5 are files under src/moduleA/.
561 struct string_list offsets
;
564 * last_directory: directory that previously processed file found in
566 * last_directory starts NULL, but records the directory in which the
567 * previous file was found within. As soon as
568 * directory(current_file) != last_directory
569 * then we need to start updating accounting in versions & offsets.
570 * Note that last_directory is always the last path in "offsets" (or
571 * NULL if "offsets" is empty) so this exists just for quick access.
573 const char *last_directory
;
575 /* last_directory_len: cached computation of strlen(last_directory) */
576 unsigned last_directory_len
;
579 static int tree_entry_order(const void *a_
, const void *b_
)
581 const struct string_list_item
*a
= a_
;
582 const struct string_list_item
*b
= b_
;
584 const struct merged_info
*ami
= a
->util
;
585 const struct merged_info
*bmi
= b
->util
;
586 return base_name_compare(a
->string
, strlen(a
->string
), ami
->result
.mode
,
587 b
->string
, strlen(b
->string
), bmi
->result
.mode
);
590 static void write_tree(struct object_id
*result_oid
,
591 struct string_list
*versions
,
595 size_t maxlen
= 0, extra
;
596 unsigned int nr
= versions
->nr
- offset
;
597 struct strbuf buf
= STRBUF_INIT
;
598 struct string_list relevant_entries
= STRING_LIST_INIT_NODUP
;
602 * We want to sort the last (versions->nr-offset) entries in versions.
603 * Do so by abusing the string_list API a bit: make another string_list
604 * that contains just those entries and then sort them.
606 * We won't use relevant_entries again and will let it just pop off the
607 * stack, so there won't be allocation worries or anything.
609 relevant_entries
.items
= versions
->items
+ offset
;
610 relevant_entries
.nr
= versions
->nr
- offset
;
611 QSORT(relevant_entries
.items
, relevant_entries
.nr
, tree_entry_order
);
613 /* Pre-allocate some space in buf */
614 extra
= hash_size
+ 8; /* 8: 6 for mode, 1 for space, 1 for NUL char */
615 for (i
= 0; i
< nr
; i
++) {
616 maxlen
+= strlen(versions
->items
[offset
+i
].string
) + extra
;
618 strbuf_grow(&buf
, maxlen
);
620 /* Write each entry out to buf */
621 for (i
= 0; i
< nr
; i
++) {
622 struct merged_info
*mi
= versions
->items
[offset
+i
].util
;
623 struct version_info
*ri
= &mi
->result
;
624 strbuf_addf(&buf
, "%o %s%c",
626 versions
->items
[offset
+i
].string
, '\0');
627 strbuf_add(&buf
, ri
->oid
.hash
, hash_size
);
630 /* Write this object file out, and record in result_oid */
631 write_object_file(buf
.buf
, buf
.len
, tree_type
, result_oid
);
632 strbuf_release(&buf
);
635 static void record_entry_for_tree(struct directory_versions
*dir_metadata
,
637 struct merged_info
*mi
)
639 const char *basename
;
642 /* nothing to record */
645 basename
= path
+ mi
->basename_offset
;
646 assert(strchr(basename
, '/') == NULL
);
647 string_list_append(&dir_metadata
->versions
,
648 basename
)->util
= &mi
->result
;
651 static void write_completed_directory(struct merge_options
*opt
,
652 const char *new_directory_name
,
653 struct directory_versions
*info
)
655 const char *prev_dir
;
656 struct merged_info
*dir_info
= NULL
;
660 * Some explanation of info->versions and info->offsets...
662 * process_entries() iterates over all relevant files AND
663 * directories in reverse lexicographic order, and calls this
664 * function. Thus, an example of the paths that process_entries()
665 * could operate on (along with the directories for those paths
670 * src/moduleB/umm.c src/moduleB
671 * src/moduleB/stuff.h src/moduleB
672 * src/moduleB/baz.c src/moduleB
674 * src/moduleA/foo.c src/moduleA
675 * src/moduleA/bar.c src/moduleA
682 * always contains the unprocessed entries and their
683 * version_info information. For example, after the first five
684 * entries above, info->versions would be:
686 * xtract.c <xtract.c's version_info>
687 * token.txt <token.txt's version_info>
688 * umm.c <src/moduleB/umm.c's version_info>
689 * stuff.h <src/moduleB/stuff.h's version_info>
690 * baz.c <src/moduleB/baz.c's version_info>
692 * Once a subdirectory is completed we remove the entries in
693 * that subdirectory from info->versions, writing it as a tree
694 * (write_tree()). Thus, as soon as we get to src/moduleB,
695 * info->versions would be updated to
697 * xtract.c <xtract.c's version_info>
698 * token.txt <token.txt's version_info>
699 * moduleB <src/moduleB's version_info>
703 * helps us track which entries in info->versions correspond to
704 * which directories. When we are N directories deep (e.g. 4
705 * for src/modA/submod/subdir/), we have up to N+1 unprocessed
706 * directories (+1 because of toplevel dir). Corresponding to
707 * the info->versions example above, after processing five entries
708 * info->offsets will be:
713 * which is used to know that xtract.c & token.txt are from the
714 * toplevel dirctory, while umm.c & stuff.h & baz.c are from the
715 * src/moduleB directory. Again, following the example above,
716 * once we need to process src/moduleB, then info->offsets is
722 * which says that moduleB (and only moduleB so far) is in the
725 * One unique thing to note about info->offsets here is that
726 * "src" was not added to info->offsets until there was a path
727 * (a file OR directory) immediately below src/ that got
730 * Since process_entry() just appends new entries to info->versions,
731 * write_completed_directory() only needs to do work if the next path
732 * is in a directory that is different than the last directory found
737 * If we are working with the same directory as the last entry, there
738 * is no work to do. (See comments above the directory_name member of
739 * struct merged_info for why we can use pointer comparison instead of
742 if (new_directory_name
== info
->last_directory
)
746 * If we are just starting (last_directory is NULL), or last_directory
747 * is a prefix of the current directory, then we can just update
748 * info->offsets to record the offset where we started this directory
749 * and update last_directory to have quick access to it.
751 if (info
->last_directory
== NULL
||
752 !strncmp(new_directory_name
, info
->last_directory
,
753 info
->last_directory_len
)) {
754 uintptr_t offset
= info
->versions
.nr
;
756 info
->last_directory
= new_directory_name
;
757 info
->last_directory_len
= strlen(info
->last_directory
);
759 * Record the offset into info->versions where we will
760 * start recording basenames of paths found within
761 * new_directory_name.
763 string_list_append(&info
->offsets
,
764 info
->last_directory
)->util
= (void*)offset
;
769 * The next entry that will be processed will be within
770 * new_directory_name. Since at this point we know that
771 * new_directory_name is within a different directory than
772 * info->last_directory, we have all entries for info->last_directory
773 * in info->versions and we need to create a tree object for them.
775 dir_info
= strmap_get(&opt
->priv
->paths
, info
->last_directory
);
777 offset
= (uintptr_t)info
->offsets
.items
[info
->offsets
.nr
-1].util
;
778 if (offset
== info
->versions
.nr
) {
780 * Actually, we don't need to create a tree object in this
781 * case. Whenever all files within a directory disappear
782 * during the merge (e.g. unmodified on one side and
783 * deleted on the other, or files were renamed elsewhere),
784 * then we get here and the directory itself needs to be
785 * omitted from its parent tree as well.
787 dir_info
->is_null
= 1;
790 * Write out the tree to the git object directory, and also
791 * record the mode and oid in dir_info->result.
793 dir_info
->is_null
= 0;
794 dir_info
->result
.mode
= S_IFDIR
;
795 write_tree(&dir_info
->result
.oid
, &info
->versions
, offset
,
796 opt
->repo
->hash_algo
->rawsz
);
800 * We've now used several entries from info->versions and one entry
801 * from info->offsets, so we get rid of those values.
804 info
->versions
.nr
= offset
;
807 * Now we've taken care of the completed directory, but we need to
808 * prepare things since future entries will be in
809 * new_directory_name. (In particular, process_entry() will be
810 * appending new entries to info->versions.) So, we need to make
811 * sure new_directory_name is the last entry in info->offsets.
813 prev_dir
= info
->offsets
.nr
== 0 ? NULL
:
814 info
->offsets
.items
[info
->offsets
.nr
-1].string
;
815 if (new_directory_name
!= prev_dir
) {
816 uintptr_t c
= info
->versions
.nr
;
817 string_list_append(&info
->offsets
,
818 new_directory_name
)->util
= (void*)c
;
821 /* And, of course, we need to update last_directory to match. */
822 info
->last_directory
= new_directory_name
;
823 info
->last_directory_len
= strlen(info
->last_directory
);
826 /* Per entry merge function */
827 static void process_entry(struct merge_options
*opt
,
829 struct conflict_info
*ci
,
830 struct directory_versions
*dir_metadata
)
833 assert(ci
->filemask
>= 0 && ci
->filemask
<= 7);
834 /* ci->match_mask == 7 was handled in collect_merge_info_callback() */
835 assert(ci
->match_mask
== 0 || ci
->match_mask
== 3 ||
836 ci
->match_mask
== 5 || ci
->match_mask
== 6);
839 record_entry_for_tree(dir_metadata
, path
, &ci
->merged
);
840 if (ci
->filemask
== 0)
841 /* nothing else to handle */
843 assert(ci
->df_conflict
);
846 if (ci
->df_conflict
) {
847 die("Not yet implemented.");
851 * NOTE: Below there is a long switch-like if-elseif-elseif... block
852 * which the code goes through even for the df_conflict cases
853 * above. Well, it will once we don't die-not-implemented above.
855 if (ci
->match_mask
) {
856 ci
->merged
.clean
= 1;
857 if (ci
->match_mask
== 6) {
858 /* stages[1] == stages[2] */
859 ci
->merged
.result
.mode
= ci
->stages
[1].mode
;
860 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[1].oid
);
862 /* determine the mask of the side that didn't match */
863 unsigned int othermask
= 7 & ~ci
->match_mask
;
864 int side
= (othermask
== 4) ? 2 : 1;
866 ci
->merged
.result
.mode
= ci
->stages
[side
].mode
;
867 ci
->merged
.is_null
= !ci
->merged
.result
.mode
;
868 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[side
].oid
);
870 assert(othermask
== 2 || othermask
== 4);
871 assert(ci
->merged
.is_null
==
872 (ci
->filemask
== ci
->match_mask
));
874 } else if (ci
->filemask
>= 6 &&
875 (S_IFMT
& ci
->stages
[1].mode
) !=
876 (S_IFMT
& ci
->stages
[2].mode
)) {
878 * Two different items from (file/submodule/symlink)
880 die("Not yet implemented.");
881 } else if (ci
->filemask
>= 6) {
883 * TODO: Needs a two-way or three-way content merge, but we're
884 * just being lazy and copying the version from HEAD and
885 * leaving it as conflicted.
887 ci
->merged
.clean
= 0;
888 ci
->merged
.result
.mode
= ci
->stages
[1].mode
;
889 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[1].oid
);
890 } else if (ci
->filemask
== 3 || ci
->filemask
== 5) {
892 die("Not yet implemented.");
893 } else if (ci
->filemask
== 2 || ci
->filemask
== 4) {
894 /* Added on one side */
895 int side
= (ci
->filemask
== 4) ? 2 : 1;
896 ci
->merged
.result
.mode
= ci
->stages
[side
].mode
;
897 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[side
].oid
);
898 ci
->merged
.clean
= !ci
->df_conflict
;
899 } else if (ci
->filemask
== 1) {
900 /* Deleted on both sides */
901 ci
->merged
.is_null
= 1;
902 ci
->merged
.result
.mode
= 0;
903 oidcpy(&ci
->merged
.result
.oid
, &null_oid
);
904 ci
->merged
.clean
= 1;
908 * If still conflicted, record it separately. This allows us to later
909 * iterate over just conflicted entries when updating the index instead
910 * of iterating over all entries.
912 if (!ci
->merged
.clean
)
913 strmap_put(&opt
->priv
->conflicted
, path
, ci
);
914 record_entry_for_tree(dir_metadata
, path
, &ci
->merged
);
917 static void process_entries(struct merge_options
*opt
,
918 struct object_id
*result_oid
)
920 struct hashmap_iter iter
;
921 struct strmap_entry
*e
;
922 struct string_list plist
= STRING_LIST_INIT_NODUP
;
923 struct string_list_item
*entry
;
924 struct directory_versions dir_metadata
= { STRING_LIST_INIT_NODUP
,
925 STRING_LIST_INIT_NODUP
,
928 if (strmap_empty(&opt
->priv
->paths
)) {
929 oidcpy(result_oid
, opt
->repo
->hash_algo
->empty_tree
);
933 /* Hack to pre-allocate plist to the desired size */
934 ALLOC_GROW(plist
.items
, strmap_get_size(&opt
->priv
->paths
), plist
.alloc
);
936 /* Put every entry from paths into plist, then sort */
937 strmap_for_each_entry(&opt
->priv
->paths
, &iter
, e
) {
938 string_list_append(&plist
, e
->key
)->util
= e
->value
;
940 plist
.cmp
= string_list_df_name_compare
;
941 string_list_sort(&plist
);
944 * Iterate over the items in reverse order, so we can handle paths
945 * below a directory before needing to handle the directory itself.
947 * This allows us to write subtrees before we need to write trees,
948 * and it also enables sane handling of directory/file conflicts
949 * (because it allows us to know whether the directory is still in
950 * the way when it is time to process the file at the same path).
952 for (entry
= &plist
.items
[plist
.nr
-1]; entry
>= plist
.items
; --entry
) {
953 char *path
= entry
->string
;
955 * NOTE: mi may actually be a pointer to a conflict_info, but
956 * we have to check mi->clean first to see if it's safe to
957 * reassign to such a pointer type.
959 struct merged_info
*mi
= entry
->util
;
961 write_completed_directory(opt
, mi
->directory_name
,
964 record_entry_for_tree(&dir_metadata
, path
, mi
);
966 struct conflict_info
*ci
= (struct conflict_info
*)mi
;
967 process_entry(opt
, path
, ci
, &dir_metadata
);
971 if (dir_metadata
.offsets
.nr
!= 1 ||
972 (uintptr_t)dir_metadata
.offsets
.items
[0].util
!= 0) {
973 printf("dir_metadata.offsets.nr = %d (should be 1)\n",
974 dir_metadata
.offsets
.nr
);
975 printf("dir_metadata.offsets.items[0].util = %u (should be 0)\n",
976 (unsigned)(uintptr_t)dir_metadata
.offsets
.items
[0].util
);
978 BUG("dir_metadata accounting completely off; shouldn't happen");
980 write_tree(result_oid
, &dir_metadata
.versions
, 0,
981 opt
->repo
->hash_algo
->rawsz
);
982 string_list_clear(&plist
, 0);
983 string_list_clear(&dir_metadata
.versions
, 0);
984 string_list_clear(&dir_metadata
.offsets
, 0);
987 static int checkout(struct merge_options
*opt
,
991 /* Switch the index/working copy from old to new */
993 struct tree_desc trees
[2];
994 struct unpack_trees_options unpack_opts
;
996 memset(&unpack_opts
, 0, sizeof(unpack_opts
));
997 unpack_opts
.head_idx
= -1;
998 unpack_opts
.src_index
= opt
->repo
->index
;
999 unpack_opts
.dst_index
= opt
->repo
->index
;
1001 setup_unpack_trees_porcelain(&unpack_opts
, "merge");
1004 * NOTE: if this were just "git checkout" code, we would probably
1005 * read or refresh the cache and check for a conflicted index, but
1006 * builtin/merge.c or sequencer.c really needs to read the index
1007 * and check for conflicted entries before starting merging for a
1008 * good user experience (no sense waiting for merges/rebases before
1009 * erroring out), so there's no reason to duplicate that work here.
1012 /* 2-way merge to the new branch */
1013 unpack_opts
.update
= 1;
1014 unpack_opts
.merge
= 1;
1015 unpack_opts
.quiet
= 0; /* FIXME: sequencer might want quiet? */
1016 unpack_opts
.verbose_update
= (opt
->verbosity
> 2);
1017 unpack_opts
.fn
= twoway_merge
;
1018 if (1/* FIXME: opts->overwrite_ignore*/) {
1019 unpack_opts
.dir
= xcalloc(1, sizeof(*unpack_opts
.dir
));
1020 unpack_opts
.dir
->flags
|= DIR_SHOW_IGNORED
;
1021 setup_standard_excludes(unpack_opts
.dir
);
1024 init_tree_desc(&trees
[0], prev
->buffer
, prev
->size
);
1026 init_tree_desc(&trees
[1], next
->buffer
, next
->size
);
1028 ret
= unpack_trees(2, trees
, &unpack_opts
);
1029 clear_unpack_trees_porcelain(&unpack_opts
);
1030 dir_clear(unpack_opts
.dir
);
1031 FREE_AND_NULL(unpack_opts
.dir
);
1035 static int record_conflicted_index_entries(struct merge_options
*opt
,
1036 struct index_state
*index
,
1037 struct strmap
*paths
,
1038 struct strmap
*conflicted
)
1040 struct hashmap_iter iter
;
1041 struct strmap_entry
*e
;
1043 int original_cache_nr
;
1045 if (strmap_empty(conflicted
))
1048 original_cache_nr
= index
->cache_nr
;
1050 /* Put every entry from paths into plist, then sort */
1051 strmap_for_each_entry(conflicted
, &iter
, e
) {
1052 const char *path
= e
->key
;
1053 struct conflict_info
*ci
= e
->value
;
1055 struct cache_entry
*ce
;
1061 * The index will already have a stage=0 entry for this path,
1062 * because we created an as-merged-as-possible version of the
1063 * file and checkout() moved the working copy and index over
1066 * However, previous iterations through this loop will have
1067 * added unstaged entries to the end of the cache which
1068 * ignore the standard alphabetical ordering of cache
1069 * entries and break invariants needed for index_name_pos()
1070 * to work. However, we know the entry we want is before
1071 * those appended cache entries, so do a temporary swap on
1072 * cache_nr to only look through entries of interest.
1074 SWAP(index
->cache_nr
, original_cache_nr
);
1075 pos
= index_name_pos(index
, path
, strlen(path
));
1076 SWAP(index
->cache_nr
, original_cache_nr
);
1078 if (ci
->filemask
!= 1)
1079 BUG("Conflicted %s but nothing in basic working tree or index; this shouldn't happen", path
);
1080 cache_tree_invalidate_path(index
, path
);
1082 ce
= index
->cache
[pos
];
1085 * Clean paths with CE_SKIP_WORKTREE set will not be
1086 * written to the working tree by the unpack_trees()
1087 * call in checkout(). Our conflicted entries would
1088 * have appeared clean to that code since we ignored
1089 * the higher order stages. Thus, we need override
1090 * the CE_SKIP_WORKTREE bit and manually write those
1091 * files to the working disk here.
1093 * TODO: Implement this CE_SKIP_WORKTREE fixup.
1097 * Mark this cache entry for removal and instead add
1098 * new stage>0 entries corresponding to the
1099 * conflicts. If there are many conflicted entries, we
1100 * want to avoid memmove'ing O(NM) entries by
1101 * inserting the new entries one at a time. So,
1102 * instead, we just add the new cache entries to the
1103 * end (ignoring normal index requirements on sort
1104 * order) and sort the index once we're all done.
1106 ce
->ce_flags
|= CE_REMOVE
;
1109 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
1110 struct version_info
*vi
;
1111 if (!(ci
->filemask
& (1ul << i
)))
1113 vi
= &ci
->stages
[i
];
1114 ce
= make_cache_entry(index
, vi
->mode
, &vi
->oid
,
1116 add_index_entry(index
, ce
, ADD_CACHE_JUST_APPEND
);
1121 * Remove the unused cache entries (and invalidate the relevant
1122 * cache-trees), then sort the index entries to get the conflicted
1123 * entries we added to the end into their right locations.
1125 remove_marked_cache_entries(index
, 1);
1126 QSORT(index
->cache
, index
->cache_nr
, cmp_cache_name_compare
);
1131 void merge_switch_to_result(struct merge_options
*opt
,
1133 struct merge_result
*result
,
1134 int update_worktree_and_index
,
1135 int display_update_msgs
)
1137 assert(opt
->priv
== NULL
);
1138 if (result
->clean
>= 0 && update_worktree_and_index
) {
1139 struct merge_options_internal
*opti
= result
->priv
;
1141 if (checkout(opt
, head
, result
->tree
)) {
1142 /* failure to function */
1147 if (record_conflicted_index_entries(opt
, opt
->repo
->index
,
1149 &opti
->conflicted
)) {
1150 /* failure to function */
1156 if (display_update_msgs
) {
1157 /* TODO: print out CONFLICT and other informational messages. */
1160 merge_finalize(opt
, result
);
1163 void merge_finalize(struct merge_options
*opt
,
1164 struct merge_result
*result
)
1166 struct merge_options_internal
*opti
= result
->priv
;
1168 assert(opt
->priv
== NULL
);
1171 * We marked opti->paths with strdup_strings = 0, so that we
1172 * wouldn't have to make another copy of the fullpath created by
1173 * make_traverse_path from setup_path_info(). But, now that we've
1174 * used it and have no other references to these strings, it is time
1175 * to deallocate them.
1177 free_strmap_strings(&opti
->paths
);
1178 strmap_clear(&opti
->paths
, 1);
1181 * All keys and values in opti->conflicted are a subset of those in
1182 * opti->paths. We don't want to deallocate anything twice, so we
1183 * don't free the keys and we pass 0 for free_values.
1185 strmap_clear(&opti
->conflicted
, 0);
1186 FREE_AND_NULL(opti
);
1189 static void merge_start(struct merge_options
*opt
, struct merge_result
*result
)
1191 /* Sanity checks on opt */
1194 assert(opt
->branch1
&& opt
->branch2
);
1196 assert(opt
->detect_directory_renames
>= MERGE_DIRECTORY_RENAMES_NONE
&&
1197 opt
->detect_directory_renames
<= MERGE_DIRECTORY_RENAMES_TRUE
);
1198 assert(opt
->rename_limit
>= -1);
1199 assert(opt
->rename_score
>= 0 && opt
->rename_score
<= MAX_SCORE
);
1200 assert(opt
->show_rename_progress
>= 0 && opt
->show_rename_progress
<= 1);
1202 assert(opt
->xdl_opts
>= 0);
1203 assert(opt
->recursive_variant
>= MERGE_VARIANT_NORMAL
&&
1204 opt
->recursive_variant
<= MERGE_VARIANT_THEIRS
);
1207 * detect_renames, verbosity, buffer_output, and obuf are ignored
1208 * fields that were used by "recursive" rather than "ort" -- but
1209 * sanity check them anyway.
1211 assert(opt
->detect_renames
>= -1 &&
1212 opt
->detect_renames
<= DIFF_DETECT_COPY
);
1213 assert(opt
->verbosity
>= 0 && opt
->verbosity
<= 5);
1214 assert(opt
->buffer_output
<= 2);
1215 assert(opt
->obuf
.len
== 0);
1217 assert(opt
->priv
== NULL
);
1219 /* Default to histogram diff. Actually, just hardcode it...for now. */
1220 opt
->xdl_opts
= DIFF_WITH_ALG(opt
, HISTOGRAM_DIFF
);
1222 /* Initialization of opt->priv, our internal merge data */
1223 opt
->priv
= xcalloc(1, sizeof(*opt
->priv
));
1226 * Although we initialize opt->priv->paths with strdup_strings=0,
1227 * that's just to avoid making yet another copy of an allocated
1228 * string. Putting the entry into paths means we are taking
1229 * ownership, so we will later free it.
1231 * In contrast, conflicted just has a subset of keys from paths, so
1232 * we don't want to free those (it'd be a duplicate free).
1234 strmap_init_with_options(&opt
->priv
->paths
, NULL
, 0);
1235 strmap_init_with_options(&opt
->priv
->conflicted
, NULL
, 0);
1239 * Originally from merge_trees_internal(); heavily adapted, though.
1241 static void merge_ort_nonrecursive_internal(struct merge_options
*opt
,
1242 struct tree
*merge_base
,
1245 struct merge_result
*result
)
1247 struct object_id working_tree_oid
;
1249 if (collect_merge_info(opt
, merge_base
, side1
, side2
) != 0) {
1251 * TRANSLATORS: The %s arguments are: 1) tree hash of a merge
1252 * base, and 2-3) the trees for the two trees we're merging.
1254 err(opt
, _("collecting merge info failed for trees %s, %s, %s"),
1255 oid_to_hex(&merge_base
->object
.oid
),
1256 oid_to_hex(&side1
->object
.oid
),
1257 oid_to_hex(&side2
->object
.oid
));
1262 result
->clean
= detect_and_process_renames(opt
, merge_base
,
1264 process_entries(opt
, &working_tree_oid
);
1266 /* Set return values */
1267 result
->tree
= parse_tree_indirect(&working_tree_oid
);
1268 /* existence of conflicted entries implies unclean */
1269 result
->clean
&= strmap_empty(&opt
->priv
->conflicted
);
1270 if (!opt
->priv
->call_depth
) {
1271 result
->priv
= opt
->priv
;
1276 void merge_incore_nonrecursive(struct merge_options
*opt
,
1277 struct tree
*merge_base
,
1280 struct merge_result
*result
)
1282 assert(opt
->ancestor
!= NULL
);
1283 merge_start(opt
, result
);
1284 merge_ort_nonrecursive_internal(opt
, merge_base
, side1
, side2
, result
);
1287 void merge_incore_recursive(struct merge_options
*opt
,
1288 struct commit_list
*merge_bases
,
1289 struct commit
*side1
,
1290 struct commit
*side2
,
1291 struct merge_result
*result
)
1293 die("Not yet implemented");