[thirdparty/git.git] / match-trees.c

#include "git-compat-util.h"
#include "hex.h"
#include "match-trees.h"
#include "strbuf.h"
#include "tree.h"
#include "tree-walk.h"
#include "object-store-ll.h"

static int score_missing(unsigned mode)
{
	int score;

	if (S_ISDIR(mode))
		score = -1000;
	else if (S_ISLNK(mode))
		score = -500;
	else
		score = -50;
	return score;
}

static int score_differs(unsigned mode1, unsigned mode2)
{
	int score;

	if (S_ISDIR(mode1) != S_ISDIR(mode2))
		score = -100;
	else if (S_ISLNK(mode1) != S_ISLNK(mode2))
		score = -50;
	else
		score = -5;
	return score;
}

static int score_matches(unsigned mode1, unsigned mode2)
{
	int score;

	/* Heh, we found SHA-1 collisions between different kind of objects */
	if (S_ISDIR(mode1) != S_ISDIR(mode2))
		score = -100;
	else if (S_ISLNK(mode1) != S_ISLNK(mode2))
		score = -50;

	else if (S_ISDIR(mode1))
		score = 1000;
	else if (S_ISLNK(mode1))
		score = 500;
	else
		score = 250;
	return score;
}

static void *fill_tree_desc_strict(struct tree_desc *desc,
				   const struct object_id *hash)
{
	void *buffer;
	enum object_type type;
	unsigned long size;

	buffer = repo_read_object_file(the_repository, hash, &type, &size);
	if (!buffer)
		die("unable to read tree (%s)", oid_to_hex(hash));
	if (type != OBJ_TREE)
		die("%s is not a tree", oid_to_hex(hash));
	init_tree_desc(desc, hash, buffer, size);
	return buffer;
}

static int base_name_entries_compare(const struct name_entry *a,
				     const struct name_entry *b)
{
	return base_name_compare(a->path, tree_entry_len(a), a->mode,
				 b->path, tree_entry_len(b), b->mode);
}

/*
 * Inspect two trees, and give a score that tells how similar they are.
 */
static int score_trees(const struct object_id *hash1, const struct object_id *hash2)
{
	struct tree_desc one;
	struct tree_desc two;
	void *one_buf = fill_tree_desc_strict(&one, hash1);
	void *two_buf = fill_tree_desc_strict(&two, hash2);
	int score = 0;

	for (;;) {
		int cmp;

		if (one.size && two.size)
			cmp = base_name_entries_compare(&one.entry, &two.entry);
		else if (one.size)
			/* two lacks this entry */
			cmp = -1;
		else if (two.size)
			/* two has more entries */
			cmp = 1;
		else
			break;

		if (cmp < 0) {
			/* path1 does not appear in two */
			score += score_missing(one.entry.mode);
			update_tree_entry(&one);
		} else if (cmp > 0) {
			/* path2 does not appear in one */
			score += score_missing(two.entry.mode);
			update_tree_entry(&two);
		} else {
			/* path appears in both */
			if (!oideq(&one.entry.oid, &two.entry.oid)) {
				/* they are different */
				score += score_differs(one.entry.mode,
						       two.entry.mode);
			} else {
				/* same subtree or blob */
				score += score_matches(one.entry.mode,
						       two.entry.mode);
			}
			update_tree_entry(&one);
			update_tree_entry(&two);
		}
	}
	free(one_buf);
	free(two_buf);
	return score;
}

/*
 * Match one itself and its subtrees with two and pick the best match.
 */
static void match_trees(const struct object_id *hash1,
			const struct object_id *hash2,
			int *best_score,
			char **best_match,
			const char *base,
			int recurse_limit)
{
	struct tree_desc one;
	void *one_buf = fill_tree_desc_strict(&one, hash1);

	while (one.size) {
		const char *path;
		const struct object_id *elem;
		unsigned short mode;
		int score;

		elem = tree_entry_extract(&one, &path, &mode);
		if (!S_ISDIR(mode))
			goto next;
		score = score_trees(elem, hash2);
		if (*best_score < score) {
			free(*best_match);
			*best_match = xstrfmt("%s%s", base, path);
			*best_score = score;
		}
		if (recurse_limit) {
			char *newbase = xstrfmt("%s%s/", base, path);
			match_trees(elem, hash2, best_score, best_match,
				    newbase, recurse_limit - 1);
			free(newbase);
		}

	next:
		update_tree_entry(&one);
	}
	free(one_buf);
}

/*
 * A tree "oid1" has a subdirectory at "prefix".  Come up with a tree object by
 * replacing it with another tree "oid2".
 */
static int splice_tree(const struct object_id *oid1, const char *prefix,
		       const struct object_id *oid2, struct object_id *result)
{
	char *subpath;
	int toplen;
	char *buf;
	unsigned long sz;
	struct tree_desc desc;
	unsigned char *rewrite_here;
	const struct object_id *rewrite_with;
	struct object_id subtree;
	enum object_type type;
	int status;

	subpath = strchrnul(prefix, '/');
	toplen = subpath - prefix;
	if (*subpath)
		subpath++;

	buf = repo_read_object_file(the_repository, oid1, &type, &sz);
	if (!buf)
		die("cannot read tree %s", oid_to_hex(oid1));
	init_tree_desc(&desc, oid1, buf, sz);

	rewrite_here = NULL;
	while (desc.size) {
		const char *name;
		unsigned short mode;

		tree_entry_extract(&desc, &name, &mode);
		if (strlen(name) == toplen &&
		    !memcmp(name, prefix, toplen)) {
			if (!S_ISDIR(mode))
				die("entry %s in tree %s is not a tree", name,
				    oid_to_hex(oid1));

			/*
			 * We cast here for two reasons:
			 *
			 *   - to flip the "char *" (for the path) to "unsigned
			 *     char *" (for the hash stored after it)
			 *
			 *   - to discard the "const"; this is OK because we
			 *     know it points into our non-const "buf"
			 */
			rewrite_here = (unsigned char *)(desc.entry.path +
							 strlen(desc.entry.path) +
							 1);
			break;
		}
		update_tree_entry(&desc);
	}
	if (!rewrite_here)
		die("entry %.*s not found in tree %s", toplen, prefix,
		    oid_to_hex(oid1));
	if (*subpath) {
		struct object_id tree_oid;
		oidread(&tree_oid, rewrite_here);
		status = splice_tree(&tree_oid, subpath, oid2, &subtree);
		if (status)
			return status;
		rewrite_with = &subtree;
	} else {
		rewrite_with = oid2;
	}
	hashcpy(rewrite_here, rewrite_with->hash);
	status = write_object_file(buf, sz, OBJ_TREE, result);
	free(buf);
	return status;
}

/*
 * We are trying to come up with a merge between one and two that
 * results in a tree shape similar to one.  The tree two might
 * correspond to a subtree of one, in which case it needs to be
 * shifted down by prefixing otherwise empty directories.  On the
 * other hand, it could cover tree one and we might need to pick a
 * subtree of it.
 */
void shift_tree(struct repository *r,
		const struct object_id *hash1,
		const struct object_id *hash2,
		struct object_id *shifted,
		int depth_limit)
{
	char *add_prefix;
	char *del_prefix;
	int add_score, del_score;

	/*
	 * NEEDSWORK: this limits the recursion depth to hardcoded
	 * value '2' to avoid excessive overhead.
	 */
	if (!depth_limit)
		depth_limit = 2;

	add_score = del_score = score_trees(hash1, hash2);
	add_prefix = xcalloc(1, 1);
	del_prefix = xcalloc(1, 1);

	/*
	 * See if one's subtree resembles two; if so we need to prefix
	 * two with a few fake trees to match the prefix.
	 */
	match_trees(hash1, hash2, &add_score, &add_prefix, "", depth_limit);

	/*
	 * See if two's subtree resembles one; if so we need to
	 * pick only subtree of two.
	 */
	match_trees(hash2, hash1, &del_score, &del_prefix, "", depth_limit);

	/* Assume we do not have to do any shifting */
	oidcpy(shifted, hash2);

	if (add_score < del_score) {
		/* We need to pick a subtree of two */
		unsigned short mode;

		if (!*del_prefix)
			return;

		if (get_tree_entry(r, hash2, del_prefix, shifted, &mode))
			die("cannot find path %s in tree %s",
			    del_prefix, oid_to_hex(hash2));
		return;
	}

	if (!*add_prefix)
		return;

	splice_tree(hash1, add_prefix, hash2, shifted);
}

/*
 * The user says the trees will be shifted by this much.
 * Unfortunately we cannot fundamentally tell which one to
 * be prefixed, as recursive merge can work in either direction.
 */
void shift_tree_by(struct repository *r,
		   const struct object_id *hash1,
		   const struct object_id *hash2,
		   struct object_id *shifted,
		   const char *shift_prefix)
{
	struct object_id sub1, sub2;
	unsigned short mode1, mode2;
	unsigned candidate = 0;

	/* Can hash2 be a tree at shift_prefix in tree hash1? */
	if (!get_tree_entry(r, hash1, shift_prefix, &sub1, &mode1) &&
	    S_ISDIR(mode1))
		candidate |= 1;

	/* Can hash1 be a tree at shift_prefix in tree hash2? */
	if (!get_tree_entry(r, hash2, shift_prefix, &sub2, &mode2) &&
	    S_ISDIR(mode2))
		candidate |= 2;

	if (candidate == 3) {
		/* Both are plausible -- we need to evaluate the score */
		int best_score = score_trees(hash1, hash2);
		int score;

		candidate = 0;
		score = score_trees(&sub1, hash2);
		if (score > best_score) {
			candidate = 1;
			best_score = score;
		}
		score = score_trees(&sub2, hash1);
		if (score > best_score)
			candidate = 2;
	}

	if (!candidate) {
		/* Neither is plausible -- do not shift */
		oidcpy(shifted, hash2);
		return;
	}

	if (candidate == 1)
		/*
		 * shift tree2 down by adding shift_prefix above it
		 * to match tree1.
		 */
		splice_tree(hash1, shift_prefix, hash2, shifted);
	else
		/*
		 * shift tree2 up by removing shift_prefix from it
		 * to match tree1.
		 */
		oidcpy(shifted, &sub2);
}
Commit	Line	Data
5e3f94df	1	#include "git-compat-util.h"
41771fa4	2	#include "hex.h"
d4ff2072	3	#include "match-trees.h"
a034e910	4	#include "strbuf.h"
68faf689 JH	5	#include "tree.h"
68faf689 JH	6	#include "tree-walk.h"
a034e910	7	#include "object-store-ll.h"
68faf689	8
667f71c3	9	static int score_missing(unsigned mode)
68faf689 JH	10	{
	11	int score;
	12
	13	if (S_ISDIR(mode))
	14	score = -1000;
	15	else if (S_ISLNK(mode))
	16	score = -500;
	17	else
	18	score = -50;
	19	return score;
	20	}
	21
667f71c3	22	static int score_differs(unsigned mode1, unsigned mode2)
68faf689 JH	23	{
	24	int score;
	25
	26	if (S_ISDIR(mode1) != S_ISDIR(mode2))
	27	score = -100;
	28	else if (S_ISLNK(mode1) != S_ISLNK(mode2))
	29	score = -50;
	30	else
	31	score = -5;
	32	return score;
	33	}
	34
667f71c3	35	static int score_matches(unsigned mode1, unsigned mode2)
68faf689 JH	36	{
	37	int score;
	38
	39	/* Heh, we found SHA-1 collisions between different kind of objects */
	40	if (S_ISDIR(mode1) != S_ISDIR(mode2))
	41	score = -100;
	42	else if (S_ISLNK(mode1) != S_ISLNK(mode2))
	43	score = -50;
	44
	45	else if (S_ISDIR(mode1))
	46	score = 1000;
	47	else if (S_ISLNK(mode1))
	48	score = 500;
	49	else
	50	score = 250;
	51	return score;
	52	}
	53
10a3fb00	54	static void fill_tree_desc_strict(struct tree_desc desc,
b6aec868	55	const struct object_id *hash)
10a3fb00 RS	56	{
	57	void *buffer;
	58	enum object_type type;
	59	unsigned long size;
	60
bc726bd0	61	buffer = repo_read_object_file(the_repository, hash, &type, &size);
10a3fb00	62	if (!buffer)
b6aec868	63	die("unable to read tree (%s)", oid_to_hex(hash));
10a3fb00	64	if (type != OBJ_TREE)
b6aec868	65	die("%s is not a tree", oid_to_hex(hash));
efed687e	66	init_tree_desc(desc, hash, buffer, size);
10a3fb00 RS	67	return buffer;
	68	}
	69
d8febde3 RS	70	static int base_name_entries_compare(const struct name_entry *a,
	71	const struct name_entry *b)
	72	{
	73	return base_name_compare(a->path, tree_entry_len(a), a->mode,
	74	b->path, tree_entry_len(b), b->mode);
	75	}
	76
68faf689 JH	77	/*
	78	* Inspect two trees, and give a score that tells how similar they are.
	79	*/
b6aec868	80	static int score_trees(const struct object_id hash1, const struct object_id hash2)
68faf689 JH	81	{
	82	struct tree_desc one;
	83	struct tree_desc two;
10a3fb00 RS	84	void *one_buf = fill_tree_desc_strict(&one, hash1);
10a3fb00 RS	85	void *two_buf = fill_tree_desc_strict(&two, hash2);
68faf689	86	int score = 0;
68faf689	87
d8febde3	88	for (;;) {
68faf689 JH	89	int cmp;
68faf689 JH	90
2ec41507 JK	91	if (one.size && two.size)
	92	cmp = base_name_entries_compare(&one.entry, &two.entry);
	93	else if (one.size)
68faf689	94	/* two lacks this entry */
d8febde3	95	cmp = -1;
2ec41507	96	else if (two.size)
d8febde3 RS	97	/* two has more entries */
	98	cmp = 1;
	99	else
	100	break;
	101
2ec41507	102	if (cmp < 0) {
68faf689	103	/* path1 does not appear in two */
667f71c3	104	score += score_missing(one.entry.mode);
2ec41507 JK	105	update_tree_entry(&one);
2ec41507 JK	106	} else if (cmp > 0) {
68faf689	107	/* path2 does not appear in one */
667f71c3	108	score += score_missing(two.entry.mode);
2ec41507 JK	109	update_tree_entry(&two);
	110	} else {
	111	/* path appears in both */
ea82b2a0	112	if (!oideq(&one.entry.oid, &two.entry.oid)) {
2ec41507 JK	113	/* they are different */
2ec41507 JK	114	score += score_differs(one.entry.mode,
667f71c3	115	two.entry.mode);
2ec41507 JK	116	} else {
	117	/* same subtree or blob */
	118	score += score_matches(one.entry.mode,
667f71c3	119	two.entry.mode);
2ec41507 JK	120	}
	121	update_tree_entry(&one);
	122	update_tree_entry(&two);
	123	}
68faf689 JH	124	}
	125	free(one_buf);
	126	free(two_buf);
	127	return score;
	128	}
	129
	130	/*
	131	* Match one itself and its subtrees with two and pick the best match.
	132	*/
b6aec868	133	static void match_trees(const struct object_id *hash1,
b6aec868	134	const struct object_id *hash2,
68faf689 JH	135	int *best_score,
68faf689 JH	136	char **best_match,
538dfe73	137	const char *base,
68faf689 JH	138	int recurse_limit)
	139	{
	140	struct tree_desc one;
10a3fb00	141	void *one_buf = fill_tree_desc_strict(&one, hash1);
68faf689 JH	142
	143	while (one.size) {
	144	const char *path;
ce6663a9	145	const struct object_id *elem;
5ec1e728	146	unsigned short mode;
68faf689 JH	147	int score;
	148
	149	elem = tree_entry_extract(&one, &path, &mode);
	150	if (!S_ISDIR(mode))
	151	goto next;
b6aec868	152	score = score_trees(elem, hash2);
68faf689	153	if (*best_score < score) {
68faf689	154	free(*best_match);
28310186	155	*best_match = xstrfmt("%s%s", base, path);
68faf689 JH	156	*best_score = score;
	157	}
	158	if (recurse_limit) {
28310186	159	char *newbase = xstrfmt("%s%s/", base, path);
b6aec868	160	match_trees(elem, hash2, best_score, best_match,
68faf689 JH	161	newbase, recurse_limit - 1);
	162	free(newbase);
	163	}
	164
	165	next:
	166	update_tree_entry(&one);
	167	}
	168	free(one_buf);
	169	}
	170
	171	/*
3b34934d PO	172	* A tree "oid1" has a subdirectory at "prefix". Come up with a tree object by
3b34934d PO	173	* replacing it with another tree "oid2".
68faf689	174	*/
3b34934d PO	175	static int splice_tree(const struct object_id oid1, const char prefix,
3b34934d PO	176	const struct object_id oid2, struct object_id result)
68faf689 JH	177	{
	178	char *subpath;
	179	int toplen;
	180	char *buf;
	181	unsigned long sz;
	182	struct tree_desc desc;
f55ac431	183	unsigned char *rewrite_here;
3b34934d PO	184	const struct object_id *rewrite_with;
3b34934d PO	185	struct object_id subtree;
68faf689 JH	186	enum object_type type;
	187	int status;
	188
2c5495f7 RM	189	subpath = strchrnul(prefix, '/');
	190	toplen = subpath - prefix;
	191	if (*subpath)
68faf689	192	subpath++;
68faf689	193
bc726bd0	194	buf = repo_read_object_file(the_repository, oid1, &type, &sz);
68faf689	195	if (!buf)
3b34934d	196	die("cannot read tree %s", oid_to_hex(oid1));
efed687e	197	init_tree_desc(&desc, oid1, buf, sz);
68faf689 JH	198
	199	rewrite_here = NULL;
	200	while (desc.size) {
	201	const char *name;
5ec1e728	202	unsigned short mode;
68faf689	203
36775ab5	204	tree_entry_extract(&desc, &name, &mode);
68faf689 JH	205	if (strlen(name) == toplen &&
	206	!memcmp(name, prefix, toplen)) {
	207	if (!S_ISDIR(mode))
3b34934d PO	208	die("entry %s in tree %s is not a tree", name,
3b34934d PO	209	oid_to_hex(oid1));
f55ac431	210
	211	/*
	212	* We cast here for two reasons:
	213	*
	214	* - to flip the "char *" (for the path) to "unsigned
	215	* char *" (for the hash stored after it)
	216	*
	217	* - to discard the "const"; this is OK because we
	218	* know it points into our non-const "buf"
	219	*/
	220	rewrite_here = (unsigned char *)(desc.entry.path +
	221	strlen(desc.entry.path) +
	222	1);
68faf689 JH	223	break;
	224	}
	225	update_tree_entry(&desc);
	226	}
	227	if (!rewrite_here)
3b34934d PO	228	die("entry %.*s not found in tree %s", toplen, prefix,
3b34934d PO	229	oid_to_hex(oid1));
2c5495f7	230	if (*subpath) {
f55ac431	231	struct object_id tree_oid;
92e2cab9	232	oidread(&tree_oid, rewrite_here);
f55ac431	233	status = splice_tree(&tree_oid, subpath, oid2, &subtree);
68faf689 JH	234	if (status)
68faf689 JH	235	return status;
3b34934d PO	236	rewrite_with = &subtree;
	237	} else {
	238	rewrite_with = oid2;
68faf689	239	}
f55ac431	240	hashcpy(rewrite_here, rewrite_with->hash);
c80d226a	241	status = write_object_file(buf, sz, OBJ_TREE, result);
68faf689 JH	242	free(buf);
	243	return status;
	244	}
	245
	246	/*
	247	* We are trying to come up with a merge between one and two that
	248	* results in a tree shape similar to one. The tree two might
	249	* correspond to a subtree of one, in which case it needs to be
	250	* shifted down by prefixing otherwise empty directories. On the
	251	* other hand, it could cover tree one and we might need to pick a
	252	* subtree of it.
	253	*/
90d34051 NTND	254	void shift_tree(struct repository *r,
90d34051 NTND	255	const struct object_id *hash1,
82db3d44	256	const struct object_id *hash2,
82db3d44	257	struct object_id *shifted,
68faf689 JH	258	int depth_limit)
	259	{
	260	char *add_prefix;
	261	char *del_prefix;
	262	int add_score, del_score;
	263
85e51b78 JH	264	/*
	265	* NEEDSWORK: this limits the recursion depth to hardcoded
	266	* value '2' to avoid excessive overhead.
	267	*/
	268	if (!depth_limit)
	269	depth_limit = 2;
	270
b6aec868	271	add_score = del_score = score_trees(hash1, hash2);
68faf689 JH	272	add_prefix = xcalloc(1, 1);
	273	del_prefix = xcalloc(1, 1);
	274
	275	/*
	276	* See if one's subtree resembles two; if so we need to prefix
	277	* two with a few fake trees to match the prefix.
	278	*/
b6aec868	279	match_trees(hash1, hash2, &add_score, &add_prefix, "", depth_limit);
68faf689 JH	280
	281	/*
	282	* See if two's subtree resembles one; if so we need to
	283	* pick only subtree of two.
	284	*/
b6aec868	285	match_trees(hash2, hash1, &del_score, &del_prefix, "", depth_limit);
68faf689 JH	286
68faf689 JH	287	/* Assume we do not have to do any shifting */
82db3d44	288	oidcpy(shifted, hash2);
68faf689 JH	289
	290	if (add_score < del_score) {
	291	/* We need to pick a subtree of two */
5ec1e728	292	unsigned short mode;
68faf689 JH	293
	294	if (!*del_prefix)
	295	return;
	296
90d34051	297	if (get_tree_entry(r, hash2, del_prefix, shifted, &mode))
68faf689	298	die("cannot find path %s in tree %s",
82db3d44	299	del_prefix, oid_to_hex(hash2));
68faf689 JH	300	return;
	301	}
	302
	303	if (!*add_prefix)
	304	return;
	305
3b34934d	306	splice_tree(hash1, add_prefix, hash2, shifted);
68faf689	307	}
85e51b78 JH	308
	309	/*
	310	* The user says the trees will be shifted by this much.
	311	* Unfortunately we cannot fundamentally tell which one to
	312	* be prefixed, as recursive merge can work in either direction.
	313	*/
90d34051 NTND	314	void shift_tree_by(struct repository *r,
90d34051 NTND	315	const struct object_id *hash1,
82db3d44	316	const struct object_id *hash2,
82db3d44	317	struct object_id *shifted,
85e51b78 JH	318	const char *shift_prefix)
85e51b78 JH	319	{
82db3d44	320	struct object_id sub1, sub2;
5ec1e728	321	unsigned short mode1, mode2;
85e51b78 JH	322	unsigned candidate = 0;
	323
	324	/* Can hash2 be a tree at shift_prefix in tree hash1? */
90d34051	325	if (!get_tree_entry(r, hash1, shift_prefix, &sub1, &mode1) &&
85e51b78 JH	326	S_ISDIR(mode1))
	327	candidate \|= 1;
	328
	329	/* Can hash1 be a tree at shift_prefix in tree hash2? */
90d34051	330	if (!get_tree_entry(r, hash2, shift_prefix, &sub2, &mode2) &&
85e51b78 JH	331	S_ISDIR(mode2))
	332	candidate \|= 2;
	333
	334	if (candidate == 3) {
	335	/* Both are plausible -- we need to evaluate the score */
b6aec868	336	int best_score = score_trees(hash1, hash2);
85e51b78 JH	337	int score;
	338
	339	candidate = 0;
b6aec868	340	score = score_trees(&sub1, hash2);
85e51b78 JH	341	if (score > best_score) {
	342	candidate = 1;
	343	best_score = score;
	344	}
b6aec868	345	score = score_trees(&sub2, hash1);
85e51b78 JH	346	if (score > best_score)
	347	candidate = 2;
	348	}
	349
	350	if (!candidate) {
	351	/* Neither is plausible -- do not shift */
82db3d44	352	oidcpy(shifted, hash2);
85e51b78 JH	353	return;
	354	}
	355
	356	if (candidate == 1)
	357	/*
	358	* shift tree2 down by adding shift_prefix above it
	359	* to match tree1.
	360	*/
3b34934d	361	splice_tree(hash1, shift_prefix, hash2, shifted);
85e51b78 JH	362	else
	363	/*
	364	* shift tree2 up by removing shift_prefix from it
	365	* to match tree1.
	366	*/
82db3d44	367	oidcpy(shifted, &sub2);
85e51b78	368	}