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

#include "cache.h"
#include "hex.h"
#include "tree.h"
#include "tree-walk.h"
#include "object-store.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, 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, 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
68faf689	1	#include "cache.h"
41771fa4	2	#include "hex.h"
68faf689 JH	3	#include "tree.h"
68faf689 JH	4	#include "tree-walk.h"
cbd53a21	5	#include "object-store.h"
68faf689	6
667f71c3	7	static int score_missing(unsigned mode)
68faf689 JH	8	{
	9	int score;
	10
	11	if (S_ISDIR(mode))
	12	score = -1000;
	13	else if (S_ISLNK(mode))
	14	score = -500;
	15	else
	16	score = -50;
	17	return score;
	18	}
	19
667f71c3	20	static int score_differs(unsigned mode1, unsigned mode2)
68faf689 JH	21	{
	22	int score;
	23
	24	if (S_ISDIR(mode1) != S_ISDIR(mode2))
	25	score = -100;
	26	else if (S_ISLNK(mode1) != S_ISLNK(mode2))
	27	score = -50;
	28	else
	29	score = -5;
	30	return score;
	31	}
	32
667f71c3	33	static int score_matches(unsigned mode1, unsigned mode2)
68faf689 JH	34	{
	35	int score;
	36
	37	/* Heh, we found SHA-1 collisions between different kind of objects */
	38	if (S_ISDIR(mode1) != S_ISDIR(mode2))
	39	score = -100;
	40	else if (S_ISLNK(mode1) != S_ISLNK(mode2))
	41	score = -50;
	42
	43	else if (S_ISDIR(mode1))
	44	score = 1000;
	45	else if (S_ISLNK(mode1))
	46	score = 500;
	47	else
	48	score = 250;
	49	return score;
	50	}
	51
10a3fb00	52	static void fill_tree_desc_strict(struct tree_desc desc,
b6aec868	53	const struct object_id *hash)
10a3fb00 RS	54	{
	55	void *buffer;
	56	enum object_type type;
	57	unsigned long size;
	58
bc726bd0	59	buffer = repo_read_object_file(the_repository, hash, &type, &size);
10a3fb00	60	if (!buffer)
b6aec868	61	die("unable to read tree (%s)", oid_to_hex(hash));
10a3fb00	62	if (type != OBJ_TREE)
b6aec868	63	die("%s is not a tree", oid_to_hex(hash));
10a3fb00 RS	64	init_tree_desc(desc, buffer, size);
	65	return buffer;
	66	}
	67
d8febde3 RS	68	static int base_name_entries_compare(const struct name_entry *a,
	69	const struct name_entry *b)
	70	{
	71	return base_name_compare(a->path, tree_entry_len(a), a->mode,
	72	b->path, tree_entry_len(b), b->mode);
	73	}
	74
68faf689 JH	75	/*
	76	* Inspect two trees, and give a score that tells how similar they are.
	77	*/
b6aec868	78	static int score_trees(const struct object_id hash1, const struct object_id hash2)
68faf689 JH	79	{
	80	struct tree_desc one;
	81	struct tree_desc two;
10a3fb00 RS	82	void *one_buf = fill_tree_desc_strict(&one, hash1);
10a3fb00 RS	83	void *two_buf = fill_tree_desc_strict(&two, hash2);
68faf689	84	int score = 0;
68faf689	85
d8febde3	86	for (;;) {
68faf689 JH	87	int cmp;
68faf689 JH	88
2ec41507 JK	89	if (one.size && two.size)
	90	cmp = base_name_entries_compare(&one.entry, &two.entry);
	91	else if (one.size)
68faf689	92	/* two lacks this entry */
d8febde3	93	cmp = -1;
2ec41507	94	else if (two.size)
d8febde3 RS	95	/* two has more entries */
	96	cmp = 1;
	97	else
	98	break;
	99
2ec41507	100	if (cmp < 0) {
68faf689	101	/* path1 does not appear in two */
667f71c3	102	score += score_missing(one.entry.mode);
2ec41507 JK	103	update_tree_entry(&one);
2ec41507 JK	104	} else if (cmp > 0) {
68faf689	105	/* path2 does not appear in one */
667f71c3	106	score += score_missing(two.entry.mode);
2ec41507 JK	107	update_tree_entry(&two);
	108	} else {
	109	/* path appears in both */
ea82b2a0	110	if (!oideq(&one.entry.oid, &two.entry.oid)) {
2ec41507 JK	111	/* they are different */
2ec41507 JK	112	score += score_differs(one.entry.mode,
667f71c3	113	two.entry.mode);
2ec41507 JK	114	} else {
	115	/* same subtree or blob */
	116	score += score_matches(one.entry.mode,
667f71c3	117	two.entry.mode);
2ec41507 JK	118	}
	119	update_tree_entry(&one);
	120	update_tree_entry(&two);
	121	}
68faf689 JH	122	}
	123	free(one_buf);
	124	free(two_buf);
	125	return score;
	126	}
	127
	128	/*
	129	* Match one itself and its subtrees with two and pick the best match.
	130	*/
b6aec868	131	static void match_trees(const struct object_id *hash1,
b6aec868	132	const struct object_id *hash2,
68faf689 JH	133	int *best_score,
68faf689 JH	134	char **best_match,
538dfe73	135	const char *base,
68faf689 JH	136	int recurse_limit)
	137	{
	138	struct tree_desc one;
10a3fb00	139	void *one_buf = fill_tree_desc_strict(&one, hash1);
68faf689 JH	140
	141	while (one.size) {
	142	const char *path;
ce6663a9	143	const struct object_id *elem;
5ec1e728	144	unsigned short mode;
68faf689 JH	145	int score;
	146
	147	elem = tree_entry_extract(&one, &path, &mode);
	148	if (!S_ISDIR(mode))
	149	goto next;
b6aec868	150	score = score_trees(elem, hash2);
68faf689	151	if (*best_score < score) {
68faf689	152	free(*best_match);
28310186	153	*best_match = xstrfmt("%s%s", base, path);
68faf689 JH	154	*best_score = score;
	155	}
	156	if (recurse_limit) {
28310186	157	char *newbase = xstrfmt("%s%s/", base, path);
b6aec868	158	match_trees(elem, hash2, best_score, best_match,
68faf689 JH	159	newbase, recurse_limit - 1);
	160	free(newbase);
	161	}
	162
	163	next:
	164	update_tree_entry(&one);
	165	}
	166	free(one_buf);
	167	}
	168
	169	/*
3b34934d PO	170	* A tree "oid1" has a subdirectory at "prefix". Come up with a tree object by
3b34934d PO	171	* replacing it with another tree "oid2".
68faf689	172	*/
3b34934d PO	173	static int splice_tree(const struct object_id oid1, const char prefix,
3b34934d PO	174	const struct object_id oid2, struct object_id result)
68faf689 JH	175	{
	176	char *subpath;
	177	int toplen;
	178	char *buf;
	179	unsigned long sz;
	180	struct tree_desc desc;
f55ac431	181	unsigned char *rewrite_here;
3b34934d PO	182	const struct object_id *rewrite_with;
3b34934d PO	183	struct object_id subtree;
68faf689 JH	184	enum object_type type;
	185	int status;
	186
2c5495f7 RM	187	subpath = strchrnul(prefix, '/');
	188	toplen = subpath - prefix;
	189	if (*subpath)
68faf689	190	subpath++;
68faf689	191
bc726bd0	192	buf = repo_read_object_file(the_repository, oid1, &type, &sz);
68faf689	193	if (!buf)
3b34934d	194	die("cannot read tree %s", oid_to_hex(oid1));
68faf689 JH	195	init_tree_desc(&desc, buf, sz);
	196
	197	rewrite_here = NULL;
	198	while (desc.size) {
	199	const char *name;
5ec1e728	200	unsigned short mode;
68faf689	201
36775ab5	202	tree_entry_extract(&desc, &name, &mode);
68faf689 JH	203	if (strlen(name) == toplen &&
	204	!memcmp(name, prefix, toplen)) {
	205	if (!S_ISDIR(mode))
3b34934d PO	206	die("entry %s in tree %s is not a tree", name,
3b34934d PO	207	oid_to_hex(oid1));
f55ac431	208
	209	/*
	210	* We cast here for two reasons:
	211	*
	212	* - to flip the "char *" (for the path) to "unsigned
	213	* char *" (for the hash stored after it)
	214	*
	215	* - to discard the "const"; this is OK because we
	216	* know it points into our non-const "buf"
	217	*/
	218	rewrite_here = (unsigned char *)(desc.entry.path +
	219	strlen(desc.entry.path) +
	220	1);
68faf689 JH	221	break;
	222	}
	223	update_tree_entry(&desc);
	224	}
	225	if (!rewrite_here)
3b34934d PO	226	die("entry %.*s not found in tree %s", toplen, prefix,
3b34934d PO	227	oid_to_hex(oid1));
2c5495f7	228	if (*subpath) {
f55ac431	229	struct object_id tree_oid;
92e2cab9	230	oidread(&tree_oid, rewrite_here);
f55ac431	231	status = splice_tree(&tree_oid, subpath, oid2, &subtree);
68faf689 JH	232	if (status)
68faf689 JH	233	return status;
3b34934d PO	234	rewrite_with = &subtree;
	235	} else {
	236	rewrite_with = oid2;
68faf689	237	}
f55ac431	238	hashcpy(rewrite_here, rewrite_with->hash);
c80d226a	239	status = write_object_file(buf, sz, OBJ_TREE, result);
68faf689 JH	240	free(buf);
	241	return status;
	242	}
	243
	244	/*
	245	* We are trying to come up with a merge between one and two that
	246	* results in a tree shape similar to one. The tree two might
	247	* correspond to a subtree of one, in which case it needs to be
	248	* shifted down by prefixing otherwise empty directories. On the
	249	* other hand, it could cover tree one and we might need to pick a
	250	* subtree of it.
	251	*/
90d34051 NTND	252	void shift_tree(struct repository *r,
90d34051 NTND	253	const struct object_id *hash1,
82db3d44	254	const struct object_id *hash2,
82db3d44	255	struct object_id *shifted,
68faf689 JH	256	int depth_limit)
	257	{
	258	char *add_prefix;
	259	char *del_prefix;
	260	int add_score, del_score;
	261
85e51b78 JH	262	/*
	263	* NEEDSWORK: this limits the recursion depth to hardcoded
	264	* value '2' to avoid excessive overhead.
	265	*/
	266	if (!depth_limit)
	267	depth_limit = 2;
	268
b6aec868	269	add_score = del_score = score_trees(hash1, hash2);
68faf689 JH	270	add_prefix = xcalloc(1, 1);
	271	del_prefix = xcalloc(1, 1);
	272
	273	/*
	274	* See if one's subtree resembles two; if so we need to prefix
	275	* two with a few fake trees to match the prefix.
	276	*/
b6aec868	277	match_trees(hash1, hash2, &add_score, &add_prefix, "", depth_limit);
68faf689 JH	278
	279	/*
	280	* See if two's subtree resembles one; if so we need to
	281	* pick only subtree of two.
	282	*/
b6aec868	283	match_trees(hash2, hash1, &del_score, &del_prefix, "", depth_limit);
68faf689 JH	284
68faf689 JH	285	/* Assume we do not have to do any shifting */
82db3d44	286	oidcpy(shifted, hash2);
68faf689 JH	287
	288	if (add_score < del_score) {
	289	/* We need to pick a subtree of two */
5ec1e728	290	unsigned short mode;
68faf689 JH	291
	292	if (!*del_prefix)
	293	return;
	294
90d34051	295	if (get_tree_entry(r, hash2, del_prefix, shifted, &mode))
68faf689	296	die("cannot find path %s in tree %s",
82db3d44	297	del_prefix, oid_to_hex(hash2));
68faf689 JH	298	return;
	299	}
	300
	301	if (!*add_prefix)
	302	return;
	303
3b34934d	304	splice_tree(hash1, add_prefix, hash2, shifted);
68faf689	305	}
85e51b78 JH	306
	307	/*
	308	* The user says the trees will be shifted by this much.
	309	* Unfortunately we cannot fundamentally tell which one to
	310	* be prefixed, as recursive merge can work in either direction.
	311	*/
90d34051 NTND	312	void shift_tree_by(struct repository *r,
90d34051 NTND	313	const struct object_id *hash1,
82db3d44	314	const struct object_id *hash2,
82db3d44	315	struct object_id *shifted,
85e51b78 JH	316	const char *shift_prefix)
85e51b78 JH	317	{
82db3d44	318	struct object_id sub1, sub2;
5ec1e728	319	unsigned short mode1, mode2;
85e51b78 JH	320	unsigned candidate = 0;
	321
	322	/* Can hash2 be a tree at shift_prefix in tree hash1? */
90d34051	323	if (!get_tree_entry(r, hash1, shift_prefix, &sub1, &mode1) &&
85e51b78 JH	324	S_ISDIR(mode1))
	325	candidate \|= 1;
	326
	327	/* Can hash1 be a tree at shift_prefix in tree hash2? */
90d34051	328	if (!get_tree_entry(r, hash2, shift_prefix, &sub2, &mode2) &&
85e51b78 JH	329	S_ISDIR(mode2))
	330	candidate \|= 2;
	331
	332	if (candidate == 3) {
	333	/* Both are plausible -- we need to evaluate the score */
b6aec868	334	int best_score = score_trees(hash1, hash2);
85e51b78 JH	335	int score;
	336
	337	candidate = 0;
b6aec868	338	score = score_trees(&sub1, hash2);
85e51b78 JH	339	if (score > best_score) {
	340	candidate = 1;
	341	best_score = score;
	342	}
b6aec868	343	score = score_trees(&sub2, hash1);
85e51b78 JH	344	if (score > best_score)
	345	candidate = 2;
	346	}
	347
	348	if (!candidate) {
	349	/* Neither is plausible -- do not shift */
82db3d44	350	oidcpy(shifted, hash2);
85e51b78 JH	351	return;
	352	}
	353
	354	if (candidate == 1)
	355	/*
	356	* shift tree2 down by adding shift_prefix above it
	357	* to match tree1.
	358	*/
3b34934d	359	splice_tree(hash1, shift_prefix, hash2, shifted);
85e51b78 JH	360	else
	361	/*
	362	* shift tree2 up by removing shift_prefix from it
	363	* to match tree1.
	364	*/
82db3d44	365	oidcpy(shifted, &sub2);
85e51b78	366	}