oid.c 9.26 KB
Newer Older
1
/*
Edward Thomson committed
2
 * Copyright (C) the libgit2 contributors. All rights reserved.
3
 *
Vicent Marti committed
4 5
 * This file is part of libgit2, distributed under the GNU GPL v2 with
 * a Linking Exception. For full terms see the included COPYING file.
6 7
 */

8 9
#include "oid.h"

10
#include "git2/oid.h"
11
#include "repository.h"
12
#include "global.h"
13
#include <string.h>
14
#include <limits.h>
15

16
static char to_hex[] = "0123456789abcdef";
17

18 19
static int oid_error_invalid(const char *msg)
{
20
	giterr_set(GITERR_INVALID, "unable to parse OID - %s", msg);
21 22 23
	return -1;
}

24
int git_oid_fromstrn(git_oid *out, const char *str, size_t length)
25
{
26
	size_t p;
27
	int v;
28

29
	assert(out && str);
30

31 32
	if (!length)
		return oid_error_invalid("too short");
33

34 35
	if (length > GIT_OID_HEXSZ)
		return oid_error_invalid("too long");
36

37
	memset(out->id, 0, GIT_OID_RAWSZ);
38

39 40
	for (p = 0; p < length; p++) {
		v = git__fromhex(str[p]);
41
		if (v < 0)
42
			return oid_error_invalid("contains invalid characters");
43

44
		out->id[p / 2] |= (unsigned char)(v << (p % 2 ? 0 : 4));
45 46
	}

47
	return 0;
48
}
49

50 51
int git_oid_fromstrp(git_oid *out, const char *str)
{
52
	return git_oid_fromstrn(out, str, strlen(str));
53 54
}

55 56 57 58 59
int git_oid_fromstr(git_oid *out, const char *str)
{
	return git_oid_fromstrn(out, str, GIT_OID_HEXSZ);
}

60
GIT_INLINE(char) *fmt_one(char *str, unsigned int val)
61 62 63 64 65 66
{
	*str++ = to_hex[val >> 4];
	*str++ = to_hex[val & 0xf];
	return str;
}

67
void git_oid_nfmt(char *str, size_t n, const git_oid *oid)
68
{
69 70 71 72 73 74 75 76 77 78 79 80
	size_t i, max_i;

	if (!oid) {
		memset(str, 0, n);
		return;
	}
	if (n > GIT_OID_HEXSZ) {
		memset(&str[GIT_OID_HEXSZ], 0, n - GIT_OID_HEXSZ);
		n = GIT_OID_HEXSZ;
	}

	max_i = n / 2;
81

82
	for (i = 0; i < max_i; i++)
83
		str = fmt_one(str, oid->id[i]);
84 85 86 87 88 89 90 91

	if (n & 1)
		*str++ = to_hex[oid->id[i] >> 4];
}

void git_oid_fmt(char *str, const git_oid *oid)
{
	git_oid_nfmt(str, GIT_OID_HEXSZ, oid);
92 93 94 95
}

void git_oid_pathfmt(char *str, const git_oid *oid)
{
96
	size_t i;
97 98 99 100 101 102 103

	str = fmt_one(str, oid->id[0]);
	*str++ = '/';
	for (i = 1; i < sizeof(oid->id); i++)
		str = fmt_one(str, oid->id[i]);
}

104 105 106 107 108 109 110
char *git_oid_tostr_s(const git_oid *oid)
{
	char *str = GIT_GLOBAL->oid_fmt;
	git_oid_nfmt(str, GIT_OID_HEXSZ + 1, oid);
	return str;
}

111 112
char *git_oid_allocfmt(const git_oid *oid)
{
113
	char *str = git__malloc(GIT_OID_HEXSZ + 1);
114 115
	if (!str)
		return NULL;
116
	git_oid_nfmt(str, GIT_OID_HEXSZ + 1, oid);
117 118
	return str;
}
119

120
char *git_oid_tostr(char *out, size_t n, const git_oid *oid)
121
{
122
	if (!out || n == 0)
123 124
		return "";

125 126
	if (n > GIT_OID_HEXSZ + 1)
		n = GIT_OID_HEXSZ + 1;
127

128 129
	git_oid_nfmt(out, n - 1, oid); /* allow room for terminating NUL */
	out[n - 1] = '\0';
130 131 132 133

	return out;
}

134 135 136
int git_oid__parse(
	git_oid *oid, const char **buffer_out,
	const char *buffer_end, const char *header)
137 138 139 140
{
	const size_t sha_len = GIT_OID_HEXSZ;
	const size_t header_len = strlen(header);

141
	const char *buffer = *buffer_out;
142 143

	if (buffer + (header_len + sha_len + 1) > buffer_end)
144
		return -1;
145 146

	if (memcmp(buffer, header, header_len) != 0)
147
		return -1;
148 149

	if (buffer[header_len + sha_len] != '\n')
150
		return -1;
151

152 153
	if (git_oid_fromstr(oid, buffer + header_len) < 0)
		return -1;
154 155 156

	*buffer_out = buffer + (header_len + sha_len + 1);

157
	return 0;
158 159
}

160 161 162 163 164 165 166 167 168 169
void git_oid__writebuf(git_buf *buf, const char *header, const git_oid *oid)
{
	char hex_oid[GIT_OID_HEXSZ];

	git_oid_fmt(hex_oid, oid);
	git_buf_puts(buf, header);
	git_buf_put(buf, hex_oid, GIT_OID_HEXSZ);
	git_buf_putc(buf, '\n');
}

Vicent Marti committed
170
void git_oid_fromraw(git_oid *out, const unsigned char *raw)
171 172 173 174 175 176 177 178 179
{
	memcpy(out->id, raw, sizeof(out->id));
}

void git_oid_cpy(git_oid *out, const git_oid *src)
{
	memcpy(out->id, src->id, sizeof(out->id));
}

180
int git_oid_cmp(const git_oid *a, const git_oid *b)
181
{
182
	return git_oid__cmp(a, b);
183 184
}

185 186 187 188 189
int git_oid_equal(const git_oid *a, const git_oid *b)
{
	return (git_oid__cmp(a, b) == 0);
}

190
int git_oid_ncmp(const git_oid *oid_a, const git_oid *oid_b, size_t len)
191
{
192 193 194
	const unsigned char *a = oid_a->id;
	const unsigned char *b = oid_b->id;

195 196 197 198
	if (len > GIT_OID_HEXSZ)
		len = GIT_OID_HEXSZ;

	while (len > 1) {
199 200 201 202 203
		if (*a != *b)
			return 1;
		a++;
		b++;
		len -= 2;
204
	};
205 206 207 208 209 210

	if (len)
		if ((*a ^ *b) & 0xf0)
			return 1;

	return 0;
211 212
}

213
int git_oid_strcmp(const git_oid *oid_a, const char *str)
214
{
215
	const unsigned char *a;
216 217 218 219 220 221
	unsigned char strval;
	int hexval;

	for (a = oid_a->id; *str && (a - oid_a->id) < GIT_OID_RAWSZ; ++a) {
		if ((hexval = git__fromhex(*str++)) < 0)
			return -1;
Linquize committed
222
		strval = (unsigned char)(hexval << 4);
223 224 225 226 227 228 229 230
		if (*str) {
			if ((hexval = git__fromhex(*str++)) < 0)
				return -1;
			strval |= hexval;
		}
		if (*a != strval)
			return (*a - strval);
	}
231

232 233
	return 0;
}
234

235 236 237
int git_oid_streq(const git_oid *oid_a, const char *str)
{
	return git_oid_strcmp(oid_a, str) == 0 ? 0 : -1;
238 239
}

240 241 242 243 244 245 246 247 248 249
int git_oid_iszero(const git_oid *oid_a)
{
	const unsigned char *a = oid_a->id;
	unsigned int i;
	for (i = 0; i < GIT_OID_RAWSZ; ++i, ++a)
		if (*a != 0)
			return 0;
	return 1;
}

250 251 252 253 254 255 256 257 258 259 260 261 262 263 264
typedef short node_index;

typedef union {
	const char *tail;
	node_index children[16];
} trie_node;

struct git_oid_shorten {
	trie_node *nodes;
	size_t node_count, size;
	int min_length, full;
};

static int resize_trie(git_oid_shorten *self, size_t new_size)
{
265
	self->nodes = git__reallocarray(self->nodes, new_size, sizeof(trie_node));
266
	GITERR_CHECK_ALLOC(self->nodes);
267 268 269 270 271 272

	if (new_size > self->size) {
		memset(&self->nodes[self->size], 0x0, (new_size - self->size) * sizeof(trie_node));
	}

	self->size = new_size;
273
	return 0;
274 275 276 277 278 279 280 281
}

static trie_node *push_leaf(git_oid_shorten *os, node_index idx, int push_at, const char *oid)
{
	trie_node *node, *leaf;
	node_index idx_leaf;

	if (os->node_count >= os->size) {
282
		if (resize_trie(os, os->size * 2) < 0)
283 284 285 286 287
			return NULL;
	}

	idx_leaf = (node_index)os->node_count++;

288
	if (os->node_count == SHRT_MAX) {
289
		os->full = 1;
290 291
        return NULL;
    }
292 293 294 295 296 297 298 299 300 301 302 303 304 305

	node = &os->nodes[idx];
	node->children[push_at] = -idx_leaf;

	leaf = &os->nodes[idx_leaf];
	leaf->tail = oid;

	return node;
}

git_oid_shorten *git_oid_shorten_new(size_t min_length)
{
	git_oid_shorten *os;

306 307
	assert((size_t)((int)min_length) == min_length);

308
	os = git__calloc(1, sizeof(git_oid_shorten));
309 310 311
	if (os == NULL)
		return NULL;

312
	if (resize_trie(os, 16) < 0) {
313
		git__free(os);
314 315 316 317
		return NULL;
	}

	os->node_count = 1;
318
	os->min_length = (int)min_length;
319 320 321 322 323 324

	return os;
}

void git_oid_shorten_free(git_oid_shorten *os)
{
325 326 327
	if (os == NULL)
		return;

328 329
	git__free(os->nodes);
	git__free(os);
330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348
}


/*
 * What wizardry is this?
 *
 * This is just a memory-optimized trie: basically a very fancy
 * 16-ary tree, which is used to store the prefixes of the OID
 * strings.
 *
 * Read more: http://en.wikipedia.org/wiki/Trie
 *
 * Magic that happens in this method:
 *
 *	- Each node in the trie is an union, so it can work both as
 *	a normal node, or as a leaf.
 *
 *	- Each normal node points to 16 children (one for each possible
 *	character in the oid). This is *not* stored in an array of
349
 *	pointers, because in a 64-bit arch this would be sucking
350
 *	16*sizeof(void*) = 128 bytes of memory per node, which is
351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378
 *	insane. What we do is store Node Indexes, and use these indexes
 *	to look up each node in the om->index array. These indexes are
 *	signed shorts, so this limits the amount of unique OIDs that
 *	fit in the structure to about 20000 (assuming a more or less uniform
 *	distribution).
 *
 *	- All the nodes in om->index array are stored contiguously in
 *	memory, and each of them is 32 bytes, so we fit 2x nodes per
 *	cache line. Convenient for speed.
 *
 *	- To differentiate the leafs from the normal nodes, we store all
 *	the indexes towards a leaf as a negative index (indexes to normal
 *	nodes are positives). When we find that one of the children for
 *	a node has a negative value, that means it's going to be a leaf.
 *	This reduces the amount of indexes we have by two, but also reduces
 *	the size of each node by 1-4 bytes (the amount we would need to
 *	add a `is_leaf` field): this is good because it allows the nodes
 *	to fit cleanly in cache lines.
 *
 *	- Once we reach an empty children, instead of continuing to insert
 *	new nodes for each remaining character of the OID, we store a pointer
 *	to the tail in the leaf; if the leaf is reached again, we turn it
 *	into a normal node and use the tail to create a new leaf.
 *
 *	This is a pretty good balance between performance and memory usage.
 */
int git_oid_shorten_add(git_oid_shorten *os, const char *text_oid)
{
379 380
	int i;
	bool is_leaf;
381 382
	node_index idx;

383
	if (os->full) {
384
		giterr_set(GITERR_INVALID, "unable to shorten OID - OID set full");
385
		return -1;
386
	}
387

388 389 390
	if (text_oid == NULL)
		return os->min_length;

391
	idx = 0;
392
	is_leaf = false;
393 394

	for (i = 0; i < GIT_OID_HEXSZ; ++i) {
nulltoken committed
395
		int c = git__fromhex(text_oid[i]);
396 397
		trie_node *node;

398
		if (c == -1) {
399
			giterr_set(GITERR_INVALID, "unable to shorten OID - invalid hex value");
400 401
			return -1;
		}
402 403 404 405 406 407 408 409 410

		node = &os->nodes[idx];

		if (is_leaf) {
			const char *tail;

			tail = node->tail;
			node->tail = NULL;

nulltoken committed
411
			node = push_leaf(os, idx, git__fromhex(tail[0]), &tail[1]);
412 413
			if (node == NULL) {
				if (os->full)
414
					giterr_set(GITERR_INVALID, "unable to shorten OID - OID set full");
415 416
				return -1;
			}
417 418 419
		}

		if (node->children[c] == 0) {
420 421
			if (push_leaf(os, idx, c, &text_oid[i + 1]) == NULL) {
				if (os->full)
422
					giterr_set(GITERR_INVALID, "unable to shorten OID - OID set full");
423
				return -1;
424
			}
425 426 427 428
			break;
		}

		idx = node->children[c];
429
		is_leaf = false;
430 431 432

		if (idx < 0) {
			node->children[c] = idx = -idx;
433
			is_leaf = true;
434 435 436 437 438 439 440 441 442
		}
	}

	if (++i > os->min_length)
		os->min_length = i;

	return os->min_length;
}