oid.c 9.17 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
#include "common.h"
9
#include "git2/oid.h"
10
#include "repository.h"
11
#include <string.h>
12
#include <limits.h>
13

14
static char to_hex[] = "0123456789abcdef";
15

16 17 18 19 20 21
static int oid_error_invalid(const char *msg)
{
	giterr_set(GITERR_INVALID, "Unable to parse OID - %s", msg);
	return -1;
}

22
int git_oid_fromstrn(git_oid *out, const char *str, size_t length)
23
{
24
	size_t p;
25
	int v;
26 27

	if (length > GIT_OID_HEXSZ)
28
		return oid_error_invalid("too long");
29

30
	for (p = 0; p < length - 1; p += 2) {
nulltoken committed
31 32
		v = (git__fromhex(str[p + 0]) << 4)
				| git__fromhex(str[p + 1]);
33

34
		if (v < 0)
35
			return oid_error_invalid("contains invalid characters");
36 37

		out->id[p / 2] = (unsigned char)v;
38
	}
39

40
	if (length % 2) {
nulltoken committed
41
		v = (git__fromhex(str[p + 0]) << 4);
42
		if (v < 0)
43
			return oid_error_invalid("contains invalid characters");
44

45 46 47 48
		out->id[p / 2] = (unsigned char)v;
		p += 2;
	}

49
	memset(out->id + p / 2, 0, (GIT_OID_HEXSZ - p) / 2);
50

51
	return 0;
52
}
53

54 55
int git_oid_fromstrp(git_oid *out, const char *str)
{
56
	return git_oid_fromstrn(out, str, strlen(str));
57 58
}

59 60 61 62 63
int git_oid_fromstr(git_oid *out, const char *str)
{
	return git_oid_fromstrn(out, str, GIT_OID_HEXSZ);
}

64
GIT_INLINE(char) *fmt_one(char *str, unsigned int val)
65 66 67 68 69 70
{
	*str++ = to_hex[val >> 4];
	*str++ = to_hex[val & 0xf];
	return str;
}

71
void git_oid_nfmt(char *str, size_t n, const git_oid *oid)
72
{
73 74 75 76 77 78 79 80 81 82 83 84
	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;
85

86
	for (i = 0; i < max_i; i++)
87
		str = fmt_one(str, oid->id[i]);
88 89 90 91 92 93 94 95

	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);
96 97 98 99
}

void git_oid_pathfmt(char *str, const git_oid *oid)
{
100
	size_t i;
101 102 103 104 105 106 107 108 109

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

char *git_oid_allocfmt(const git_oid *oid)
{
110
	char *str = git__malloc(GIT_OID_HEXSZ + 1);
111 112
	if (!str)
		return NULL;
113
	git_oid_nfmt(str, GIT_OID_HEXSZ + 1, oid);
114 115
	return str;
}
116

117
char *git_oid_tostr(char *out, size_t n, const git_oid *oid)
118
{
119
	if (!out || n == 0)
120 121
		return "";

122 123
	if (n > GIT_OID_HEXSZ + 1)
		n = GIT_OID_HEXSZ + 1;
124

125 126
	git_oid_nfmt(out, n - 1, oid); /* allow room for terminating NUL */
	out[n - 1] = '\0';
127 128 129 130

	return out;
}

131 132 133
int git_oid__parse(
	git_oid *oid, const char **buffer_out,
	const char *buffer_end, const char *header)
134 135 136 137
{
	const size_t sha_len = GIT_OID_HEXSZ;
	const size_t header_len = strlen(header);

138
	const char *buffer = *buffer_out;
139 140

	if (buffer + (header_len + sha_len + 1) > buffer_end)
141
		return -1;
142 143

	if (memcmp(buffer, header, header_len) != 0)
144
		return -1;
145 146

	if (buffer[header_len + sha_len] != '\n')
147
		return -1;
148

149 150
	if (git_oid_fromstr(oid, buffer + header_len) < 0)
		return -1;
151 152 153

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

154
	return 0;
155 156
}

157 158 159 160 161 162 163 164 165 166
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
167
void git_oid_fromraw(git_oid *out, const unsigned char *raw)
168 169 170 171 172 173 174 175 176
{
	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));
}

177
int git_oid_cmp(const git_oid *a, const git_oid *b)
178
{
179
	return git_oid__cmp(a, b);
180 181
}

182
int git_oid_ncmp(const git_oid *oid_a, const git_oid *oid_b, size_t len)
183
{
184 185 186
	const unsigned char *a = oid_a->id;
	const unsigned char *b = oid_b->id;

187 188 189 190
	if (len > GIT_OID_HEXSZ)
		len = GIT_OID_HEXSZ;

	while (len > 1) {
191 192 193 194 195
		if (*a != *b)
			return 1;
		a++;
		b++;
		len -= 2;
196
	};
197 198 199 200 201 202

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

	return 0;
203 204
}

205
int git_oid_strcmp(const git_oid *oid_a, const char *str)
206
{
207 208 209 210 211 212 213
	const unsigned char *a = oid_a->id;
	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
214
		strval = (unsigned char)(hexval << 4);
215 216 217 218 219 220 221 222
		if (*str) {
			if ((hexval = git__fromhex(*str++)) < 0)
				return -1;
			strval |= hexval;
		}
		if (*a != strval)
			return (*a - strval);
	}
223

224 225
	return 0;
}
226

227 228 229
int git_oid_streq(const git_oid *oid_a, const char *str)
{
	return git_oid_strcmp(oid_a, str) == 0 ? 0 : -1;
230 231
}

232 233 234 235 236 237 238 239 240 241
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;
}

242 243 244 245 246 247 248 249 250 251 252 253 254 255 256
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)
{
257
	self->nodes = git__realloc(self->nodes, new_size * sizeof(trie_node));
258
	GITERR_CHECK_ALLOC(self->nodes);
259 260 261 262 263 264

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

	self->size = new_size;
265
	return 0;
266 267 268 269 270 271 272 273
}

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) {
274
		if (resize_trie(os, os->size * 2) < 0)
275 276 277 278 279
			return NULL;
	}

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

280
	if (os->node_count == SHRT_MAX) {
281
		os->full = 1;
282 283
        return NULL;
    }
284 285 286 287 288 289 290 291 292 293 294 295 296 297

	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;

298 299
	assert((size_t)((int)min_length) == min_length);

300
	os = git__calloc(1, sizeof(git_oid_shorten));
301 302 303
	if (os == NULL)
		return NULL;

304
	if (resize_trie(os, 16) < 0) {
305
		git__free(os);
306 307 308 309
		return NULL;
	}

	os->node_count = 1;
310
	os->min_length = (int)min_length;
311 312 313 314 315 316

	return os;
}

void git_oid_shorten_free(git_oid_shorten *os)
{
317 318 319
	if (os == NULL)
		return;

320 321
	git__free(os->nodes);
	git__free(os);
322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340
}


/*
 * 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
341
 *	pointers, because in a 64-bit arch this would be sucking
342
 *	16*sizeof(void*) = 128 bytes of memory per node, which is
343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370
 *	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)
{
371 372
	int i;
	bool is_leaf;
373 374
	node_index idx;

375 376
	if (os->full) {
		giterr_set(GITERR_INVALID, "Unable to shorten OID - OID set full");
377
		return -1;
378
	}
379

380 381 382
	if (text_oid == NULL)
		return os->min_length;

383
	idx = 0;
384
	is_leaf = false;
385 386

	for (i = 0; i < GIT_OID_HEXSZ; ++i) {
nulltoken committed
387
		int c = git__fromhex(text_oid[i]);
388 389
		trie_node *node;

390 391 392 393
		if (c == -1) {
			giterr_set(GITERR_INVALID, "Unable to shorten OID - invalid hex value");
			return -1;
		}
394 395 396 397 398 399 400 401 402

		node = &os->nodes[idx];

		if (is_leaf) {
			const char *tail;

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

nulltoken committed
403
			node = push_leaf(os, idx, git__fromhex(tail[0]), &tail[1]);
404 405 406 407 408
			if (node == NULL) {
				if (os->full)
					giterr_set(GITERR_INVALID, "Unable to shorten OID - OID set full");
				return -1;
			}
409 410 411
		}

		if (node->children[c] == 0) {
412 413 414
			if (push_leaf(os, idx, c, &text_oid[i + 1]) == NULL) {
				if (os->full)
					giterr_set(GITERR_INVALID, "Unable to shorten OID - OID set full");
415
				return -1;
416
			}
417 418 419 420
			break;
		}

		idx = node->children[c];
421
		is_leaf = false;
422 423 424

		if (idx < 0) {
			node->children[c] = idx = -idx;
425
			is_leaf = true;
426 427 428 429 430 431 432 433 434
		}
	}

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

	return os->min_length;
}