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

#include "cache.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 = read_object_file(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 = read_object_file(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;
		hashcpy(tree_oid.hash, 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, tree_type, 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(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(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(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(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(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 JH	1	#include "cache.h"
	2	#include "tree.h"
	3	#include "tree-walk.h"
cbd53a21	4	#include "object-store.h"
68faf689	5
667f71c3	6	static int score_missing(unsigned mode)
68faf689 JH	7	{
	8	int score;
	9
	10	if (S_ISDIR(mode))
	11	score = -1000;
	12	else if (S_ISLNK(mode))
	13	score = -500;
	14	else
	15	score = -50;
	16	return score;
	17	}
	18
667f71c3	19	static int score_differs(unsigned mode1, unsigned mode2)
68faf689 JH	20	{
	21	int score;
	22
	23	if (S_ISDIR(mode1) != S_ISDIR(mode2))
	24	score = -100;
	25	else if (S_ISLNK(mode1) != S_ISLNK(mode2))
	26	score = -50;
	27	else
	28	score = -5;
	29	return score;
	30	}
	31
667f71c3	32	static int score_matches(unsigned mode1, unsigned mode2)
68faf689 JH	33	{
	34	int score;
	35
	36	/* Heh, we found SHA-1 collisions between different kind of objects */
	37	if (S_ISDIR(mode1) != S_ISDIR(mode2))
	38	score = -100;
	39	else if (S_ISLNK(mode1) != S_ISLNK(mode2))
	40	score = -50;
	41
	42	else if (S_ISDIR(mode1))
	43	score = 1000;
	44	else if (S_ISLNK(mode1))
	45	score = 500;
	46	else
	47	score = 250;
	48	return score;
	49	}
	50
10a3fb00	51	static void fill_tree_desc_strict(struct tree_desc desc,
b6aec868	52	const struct object_id *hash)
10a3fb00 RS	53	{
	54	void *buffer;
	55	enum object_type type;
	56	unsigned long size;
	57
b4f5aca4	58	buffer = read_object_file(hash, &type, &size);
10a3fb00	59	if (!buffer)
b6aec868	60	die("unable to read tree (%s)", oid_to_hex(hash));
10a3fb00	61	if (type != OBJ_TREE)
b6aec868	62	die("%s is not a tree", oid_to_hex(hash));
10a3fb00 RS	63	init_tree_desc(desc, buffer, size);
	64	return buffer;
	65	}
	66
d8febde3 RS	67	static int base_name_entries_compare(const struct name_entry *a,
	68	const struct name_entry *b)
	69	{
	70	return base_name_compare(a->path, tree_entry_len(a), a->mode,
	71	b->path, tree_entry_len(b), b->mode);
	72	}
	73
68faf689 JH	74	/*
	75	* Inspect two trees, and give a score that tells how similar they are.
	76	*/
b6aec868	77	static int score_trees(const struct object_id hash1, const struct object_id hash2)
68faf689 JH	78	{
	79	struct tree_desc one;
	80	struct tree_desc two;
10a3fb00 RS	81	void *one_buf = fill_tree_desc_strict(&one, hash1);
10a3fb00 RS	82	void *two_buf = fill_tree_desc_strict(&two, hash2);
68faf689	83	int score = 0;
68faf689	84
d8febde3	85	for (;;) {
68faf689 JH	86	int cmp;
68faf689 JH	87
2ec41507 JK	88	if (one.size && two.size)
	89	cmp = base_name_entries_compare(&one.entry, &two.entry);
	90	else if (one.size)
68faf689	91	/* two lacks this entry */
d8febde3	92	cmp = -1;
2ec41507	93	else if (two.size)
d8febde3 RS	94	/* two has more entries */
	95	cmp = 1;
	96	else
	97	break;
	98
2ec41507	99	if (cmp < 0) {
68faf689	100	/* path1 does not appear in two */
667f71c3	101	score += score_missing(one.entry.mode);
2ec41507 JK	102	update_tree_entry(&one);
2ec41507 JK	103	} else if (cmp > 0) {
68faf689	104	/* path2 does not appear in one */
667f71c3	105	score += score_missing(two.entry.mode);
2ec41507 JK	106	update_tree_entry(&two);
	107	} else {
	108	/* path appears in both */
ea82b2a0	109	if (!oideq(&one.entry.oid, &two.entry.oid)) {
2ec41507 JK	110	/* they are different */
2ec41507 JK	111	score += score_differs(one.entry.mode,
667f71c3	112	two.entry.mode);
2ec41507 JK	113	} else {
	114	/* same subtree or blob */
	115	score += score_matches(one.entry.mode,
667f71c3	116	two.entry.mode);
2ec41507 JK	117	}
	118	update_tree_entry(&one);
	119	update_tree_entry(&two);
	120	}
68faf689 JH	121	}
	122	free(one_buf);
	123	free(two_buf);
	124	return score;
	125	}
	126
	127	/*
	128	* Match one itself and its subtrees with two and pick the best match.
	129	*/
b6aec868	130	static void match_trees(const struct object_id *hash1,
b6aec868	131	const struct object_id *hash2,
68faf689 JH	132	int *best_score,
68faf689 JH	133	char **best_match,
538dfe73	134	const char *base,
68faf689 JH	135	int recurse_limit)
	136	{
	137	struct tree_desc one;
10a3fb00	138	void *one_buf = fill_tree_desc_strict(&one, hash1);
68faf689 JH	139
	140	while (one.size) {
	141	const char *path;
ce6663a9	142	const struct object_id *elem;
5ec1e728	143	unsigned short mode;
68faf689 JH	144	int score;
	145
	146	elem = tree_entry_extract(&one, &path, &mode);
	147	if (!S_ISDIR(mode))
	148	goto next;
b6aec868	149	score = score_trees(elem, hash2);
68faf689	150	if (*best_score < score) {
68faf689	151	free(*best_match);
28310186	152	*best_match = xstrfmt("%s%s", base, path);
68faf689 JH	153	*best_score = score;
	154	}
	155	if (recurse_limit) {
28310186	156	char *newbase = xstrfmt("%s%s/", base, path);
b6aec868	157	match_trees(elem, hash2, best_score, best_match,
68faf689 JH	158	newbase, recurse_limit - 1);
	159	free(newbase);
	160	}
	161
	162	next:
	163	update_tree_entry(&one);
	164	}
	165	free(one_buf);
	166	}
	167
	168	/*
3b34934d PO	169	* A tree "oid1" has a subdirectory at "prefix". Come up with a tree object by
3b34934d PO	170	* replacing it with another tree "oid2".
68faf689	171	*/
3b34934d PO	172	static int splice_tree(const struct object_id oid1, const char prefix,
3b34934d PO	173	const struct object_id oid2, struct object_id result)
68faf689 JH	174	{
	175	char *subpath;
	176	int toplen;
	177	char *buf;
	178	unsigned long sz;
	179	struct tree_desc desc;
f55ac431	180	unsigned char *rewrite_here;
3b34934d PO	181	const struct object_id *rewrite_with;
3b34934d PO	182	struct object_id subtree;
68faf689 JH	183	enum object_type type;
	184	int status;
	185
2c5495f7 RM	186	subpath = strchrnul(prefix, '/');
	187	toplen = subpath - prefix;
	188	if (*subpath)
68faf689	189	subpath++;
68faf689	190
b4f5aca4	191	buf = read_object_file(oid1, &type, &sz);
68faf689	192	if (!buf)
3b34934d	193	die("cannot read tree %s", oid_to_hex(oid1));
68faf689 JH	194	init_tree_desc(&desc, buf, sz);
	195
	196	rewrite_here = NULL;
	197	while (desc.size) {
	198	const char *name;
5ec1e728	199	unsigned short mode;
68faf689	200
36775ab5	201	tree_entry_extract(&desc, &name, &mode);
68faf689 JH	202	if (strlen(name) == toplen &&
	203	!memcmp(name, prefix, toplen)) {
	204	if (!S_ISDIR(mode))
3b34934d PO	205	die("entry %s in tree %s is not a tree", name,
3b34934d PO	206	oid_to_hex(oid1));
f55ac431	207
	208	/*
	209	* We cast here for two reasons:
	210	*
	211	* - to flip the "char *" (for the path) to "unsigned
	212	* char *" (for the hash stored after it)
	213	*
	214	* - to discard the "const"; this is OK because we
	215	* know it points into our non-const "buf"
	216	*/
	217	rewrite_here = (unsigned char *)(desc.entry.path +
	218	strlen(desc.entry.path) +
	219	1);
68faf689 JH	220	break;
	221	}
	222	update_tree_entry(&desc);
	223	}
	224	if (!rewrite_here)
3b34934d PO	225	die("entry %.*s not found in tree %s", toplen, prefix,
3b34934d PO	226	oid_to_hex(oid1));
2c5495f7	227	if (*subpath) {
f55ac431	228	struct object_id tree_oid;
	229	hashcpy(tree_oid.hash, rewrite_here);
	230	status = splice_tree(&tree_oid, subpath, oid2, &subtree);
68faf689 JH	231	if (status)
68faf689 JH	232	return status;
3b34934d PO	233	rewrite_with = &subtree;
	234	} else {
	235	rewrite_with = oid2;
68faf689	236	}
f55ac431	237	hashcpy(rewrite_here, rewrite_with->hash);
a09c985e	238	status = write_object_file(buf, sz, tree_type, result);
68faf689 JH	239	free(buf);
	240	return status;
	241	}
	242
	243	/*
	244	* We are trying to come up with a merge between one and two that
	245	* results in a tree shape similar to one. The tree two might
	246	* correspond to a subtree of one, in which case it needs to be
	247	* shifted down by prefixing otherwise empty directories. On the
	248	* other hand, it could cover tree one and we might need to pick a
	249	* subtree of it.
	250	*/
82db3d44	251	void shift_tree(const struct object_id *hash1,
	252	const struct object_id *hash2,
	253	struct object_id *shifted,
68faf689 JH	254	int depth_limit)
	255	{
	256	char *add_prefix;
	257	char *del_prefix;
	258	int add_score, del_score;
	259
85e51b78 JH	260	/*
	261	* NEEDSWORK: this limits the recursion depth to hardcoded
	262	* value '2' to avoid excessive overhead.
	263	*/
	264	if (!depth_limit)
	265	depth_limit = 2;
	266
b6aec868	267	add_score = del_score = score_trees(hash1, hash2);
68faf689 JH	268	add_prefix = xcalloc(1, 1);
	269	del_prefix = xcalloc(1, 1);
	270
	271	/*
	272	* See if one's subtree resembles two; if so we need to prefix
	273	* two with a few fake trees to match the prefix.
	274	*/
b6aec868	275	match_trees(hash1, hash2, &add_score, &add_prefix, "", depth_limit);
68faf689 JH	276
	277	/*
	278	* See if two's subtree resembles one; if so we need to
	279	* pick only subtree of two.
	280	*/
b6aec868	281	match_trees(hash2, hash1, &del_score, &del_prefix, "", depth_limit);
68faf689 JH	282
68faf689 JH	283	/* Assume we do not have to do any shifting */
82db3d44	284	oidcpy(shifted, hash2);
68faf689 JH	285
	286	if (add_score < del_score) {
	287	/* We need to pick a subtree of two */
5ec1e728	288	unsigned short mode;
68faf689 JH	289
	290	if (!*del_prefix)
	291	return;
	292
916bc35b	293	if (get_tree_entry(hash2, del_prefix, shifted, &mode))
68faf689	294	die("cannot find path %s in tree %s",
82db3d44	295	del_prefix, oid_to_hex(hash2));
68faf689 JH	296	return;
	297	}
	298
	299	if (!*add_prefix)
	300	return;
	301
3b34934d	302	splice_tree(hash1, add_prefix, hash2, shifted);
68faf689	303	}
85e51b78 JH	304
	305	/*
	306	* The user says the trees will be shifted by this much.
	307	* Unfortunately we cannot fundamentally tell which one to
	308	* be prefixed, as recursive merge can work in either direction.
	309	*/
82db3d44	310	void shift_tree_by(const struct object_id *hash1,
	311	const struct object_id *hash2,
	312	struct object_id *shifted,
85e51b78 JH	313	const char *shift_prefix)
85e51b78 JH	314	{
82db3d44	315	struct object_id sub1, sub2;
5ec1e728	316	unsigned short mode1, mode2;
85e51b78 JH	317	unsigned candidate = 0;
	318
	319	/* Can hash2 be a tree at shift_prefix in tree hash1? */
916bc35b	320	if (!get_tree_entry(hash1, shift_prefix, &sub1, &mode1) &&
85e51b78 JH	321	S_ISDIR(mode1))
	322	candidate \|= 1;
	323
	324	/* Can hash1 be a tree at shift_prefix in tree hash2? */
916bc35b	325	if (!get_tree_entry(hash2, shift_prefix, &sub2, &mode2) &&
85e51b78 JH	326	S_ISDIR(mode2))
	327	candidate \|= 2;
	328
	329	if (candidate == 3) {
	330	/* Both are plausible -- we need to evaluate the score */
b6aec868	331	int best_score = score_trees(hash1, hash2);
85e51b78 JH	332	int score;
	333
	334	candidate = 0;
b6aec868	335	score = score_trees(&sub1, hash2);
85e51b78 JH	336	if (score > best_score) {
	337	candidate = 1;
	338	best_score = score;
	339	}
b6aec868	340	score = score_trees(&sub2, hash1);
85e51b78 JH	341	if (score > best_score)
	342	candidate = 2;
	343	}
	344
	345	if (!candidate) {
	346	/* Neither is plausible -- do not shift */
82db3d44	347	oidcpy(shifted, hash2);
85e51b78 JH	348	return;
	349	}
	350
	351	if (candidate == 1)
	352	/*
	353	* shift tree2 down by adding shift_prefix above it
	354	* to match tree1.
	355	*/
3b34934d	356	splice_tree(hash1, shift_prefix, hash2, shifted);
85e51b78 JH	357	else
	358	/*
	359	* shift tree2 up by removing shift_prefix from it
	360	* to match tree1.
	361	*/
82db3d44	362	oidcpy(shifted, &sub2);
85e51b78	363	}