oid.c 9.25 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 "global.h"
12
#include <string.h>
13
#include <limits.h>
14

15
static char to_hex[] = "0123456789abcdef";
16

17 18 19 20 21 22
static int oid_error_invalid(const char *msg)
{
	giterr_set(GITERR_INVALID, "Unable to parse OID - %s", msg);
	return -1;
}

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

28
	assert(out && str);
29

30 31
	if (!length)
		return oid_error_invalid("too short");
32

33 34
	if (length > GIT_OID_HEXSZ)
		return oid_error_invalid("too long");
35

36
	memset(out->id, 0, GIT_OID_RAWSZ);
37

38 39
	for (p = 0; p < length; p++) {
		v = git__fromhex(str[p]);
40
		if (v < 0)
41
			return oid_error_invalid("contains invalid characters");
42

43
		out->id[p / 2] |= (unsigned char)(v << (p % 2 ? 0 : 4));
44 45
	}

46
	return 0;
47
}
48

49 50
int git_oid_fromstrp(git_oid *out, const char *str)
{
51
	return git_oid_fromstrn(out, str, strlen(str));
52 53
}

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

59
GIT_INLINE(char) *fmt_one(char *str, unsigned int val)
60 61 62 63 64 65
{
	*str++ = to_hex[val >> 4];
	*str++ = to_hex[val & 0xf];
	return str;
}

66
void git_oid_nfmt(char *str, size_t n, const git_oid *oid)
67
{
68 69 70 71 72 73 74 75 76 77 78 79
	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;
80

81
	for (i = 0; i < max_i; i++)
82
		str = fmt_one(str, oid->id[i]);
83 84 85 86 87 88 89 90

	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);
91 92 93 94
}

void git_oid_pathfmt(char *str, const git_oid *oid)
{
95
	size_t i;
96 97 98 99 100 101 102

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

103 104 105 106 107 108 109
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;
}

110 111
char *git_oid_allocfmt(const git_oid *oid)
{
112
	char *str = git__malloc(GIT_OID_HEXSZ + 1);
113 114
	if (!str)
		return NULL;
115
	git_oid_nfmt(str, GIT_OID_HEXSZ + 1, oid);
116 117
	return str;
}
118

119
char *git_oid_tostr(char *out, size_t n, const git_oid *oid)
120
{
121
	if (!out || n == 0)
122 123
		return "";

124 125
	if (n > GIT_OID_HEXSZ + 1)
		n = GIT_OID_HEXSZ + 1;
126

127 128
	git_oid_nfmt(out, n - 1, oid); /* allow room for terminating NUL */
	out[n - 1] = '\0';
129 130 131 132

	return out;
}

133 134 135
int git_oid__parse(
	git_oid *oid, const char **buffer_out,
	const char *buffer_end, const char *header)
136 137 138 139
{
	const size_t sha_len = GIT_OID_HEXSZ;
	const size_t header_len = strlen(header);

140
	const char *buffer = *buffer_out;
141 142

	if (buffer + (header_len + sha_len + 1) > buffer_end)
143
		return -1;
144 145

	if (memcmp(buffer, header, header_len) != 0)
146
		return -1;
147 148

	if (buffer[header_len + sha_len] != '\n')
149
		return -1;
150

151 152
	if (git_oid_fromstr(oid, buffer + header_len) < 0)
		return -1;
153 154 155

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

156
	return 0;
157 158
}

159 160 161 162 163 164 165 166 167 168
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
169
void git_oid_fromraw(git_oid *out, const unsigned char *raw)
170 171 172 173 174 175 176 177 178
{
	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));
}

179
int git_oid_cmp(const git_oid *a, const git_oid *b)
180
{
181
	return git_oid__cmp(a, b);
182 183
}

184 185 186 187 188
int git_oid_equal(const git_oid *a, const git_oid *b)
{
	return (git_oid__cmp(a, b) == 0);
}

189
int git_oid_ncmp(const git_oid *oid_a, const git_oid *oid_b, size_t len)
190
{
191 192 193
	const unsigned char *a = oid_a->id;
	const unsigned char *b = oid_b->id;

194 195 196 197
	if (len > GIT_OID_HEXSZ)
		len = GIT_OID_HEXSZ;

	while (len > 1) {
198 199 200 201 202
		if (*a != *b)
			return 1;
		a++;
		b++;
		len -= 2;
203
	};
204 205 206 207 208 209

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

	return 0;
210 211
}

212
int git_oid_strcmp(const git_oid *oid_a, const char *str)
213
{
214
	const unsigned char *a;
215 216 217 218 219 220
	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
221
		strval = (unsigned char)(hexval << 4);
222 223 224 225 226 227 228 229
		if (*str) {
			if ((hexval = git__fromhex(*str++)) < 0)
				return -1;
			strval |= hexval;
		}
		if (*a != strval)
			return (*a - strval);
	}
230

231 232
	return 0;
}
233

234 235 236
int git_oid_streq(const git_oid *oid_a, const char *str)
{
	return git_oid_strcmp(oid_a, str) == 0 ? 0 : -1;
237 238
}

239 240 241 242 243 244 245 246 247 248
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;
}

249 250 251 252 253 254 255 256 257 258 259 260 261 262 263
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)
{
264
	self->nodes = git__realloc(self->nodes, new_size * sizeof(trie_node));
265
	GITERR_CHECK_ALLOC(self->nodes);
266 267 268 269 270 271

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

	self->size = new_size;
272
	return 0;
273 274 275 276 277 278 279 280
}

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) {
281
		if (resize_trie(os, os->size * 2) < 0)
282 283 284 285 286
			return NULL;
	}

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

287
	if (os->node_count == SHRT_MAX) {
288
		os->full = 1;
289 290
        return NULL;
    }
291 292 293 294 295 296 297 298 299 300 301 302 303 304

	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;

305 306
	assert((size_t)((int)min_length) == min_length);

307
	os = git__calloc(1, sizeof(git_oid_shorten));
308 309 310
	if (os == NULL)
		return NULL;

311
	if (resize_trie(os, 16) < 0) {
312
		git__free(os);
313 314 315 316
		return NULL;
	}

	os->node_count = 1;
317
	os->min_length = (int)min_length;
318 319 320 321 322 323

	return os;
}

void git_oid_shorten_free(git_oid_shorten *os)
{
324 325 326
	if (os == NULL)
		return;

327 328
	git__free(os->nodes);
	git__free(os);
329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347
}


/*
 * 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
348
 *	pointers, because in a 64-bit arch this would be sucking
349
 *	16*sizeof(void*) = 128 bytes of memory per node, which is
350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377
 *	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)
{
378 379
	int i;
	bool is_leaf;
380 381
	node_index idx;

382 383
	if (os->full) {
		giterr_set(GITERR_INVALID, "Unable to shorten OID - OID set full");
384
		return -1;
385
	}
386

387 388 389
	if (text_oid == NULL)
		return os->min_length;

390
	idx = 0;
391
	is_leaf = false;
392 393

	for (i = 0; i < GIT_OID_HEXSZ; ++i) {
nulltoken committed
394
		int c = git__fromhex(text_oid[i]);
395 396
		trie_node *node;

397 398 399 400
		if (c == -1) {
			giterr_set(GITERR_INVALID, "Unable to shorten OID - invalid hex value");
			return -1;
		}
401 402 403 404 405 406 407 408 409

		node = &os->nodes[idx];

		if (is_leaf) {
			const char *tail;

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

nulltoken committed
410
			node = push_leaf(os, idx, git__fromhex(tail[0]), &tail[1]);
411 412 413 414 415
			if (node == NULL) {
				if (os->full)
					giterr_set(GITERR_INVALID, "Unable to shorten OID - OID set full");
				return -1;
			}
416 417 418
		}

		if (node->children[c] == 0) {
419 420 421
			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");
422
				return -1;
423
			}
424 425 426 427
			break;
		}

		idx = node->children[c];
428
		is_leaf = false;
429 430 431

		if (idx < 0) {
			node->children[c] = idx = -idx;
432
			is_leaf = true;
433 434 435 436 437 438 439 440 441
		}
	}

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

	return os->min_length;
}