oid.c 8.4 KB
Newer Older
1
/*
Vicent Marti committed
2
 * Copyright (C) 2009-2011 the libgit2 contributors
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
int git_oid_fromstrn(git_oid *out, const char *str, size_t length)
17
{
18
	size_t p;
19
	int v;
20

21 22 23
	if (length < 4)
		return git__throw(GIT_ENOTOID, "Failed to generate sha1. Given string is too short");

24 25 26
	if (length > GIT_OID_HEXSZ)
		length = GIT_OID_HEXSZ;

27
	for (p = 0; p < length - 1; p += 2) {
nulltoken committed
28 29
		v = (git__fromhex(str[p + 0]) << 4)
				| git__fromhex(str[p + 1]);
30

31
		if (v < 0)
32
			return git__throw(GIT_ENOTOID, "Failed to generate sha1. Given string is not a valid sha1 hash");
33 34

		out->id[p / 2] = (unsigned char)v;
35
	}
36

37
	if (length % 2) {
nulltoken committed
38
		v = (git__fromhex(str[p + 0]) << 4);
39 40 41
		if (v < 0)
			return git__throw(GIT_ENOTOID, "Failed to generate sha1. Given string is not a valid sha1 hash");

42 43 44 45
		out->id[p / 2] = (unsigned char)v;
		p += 2;
	}

46
	memset(out->id + p / 2, 0, (GIT_OID_HEXSZ - p) / 2);
47

48 49
	return GIT_SUCCESS;
}
50

51 52 53 54 55
int git_oid_fromstr(git_oid *out, const char *str)
{
	return git_oid_fromstrn(out, str, GIT_OID_HEXSZ);
}

56
GIT_INLINE(char) *fmt_one(char *str, unsigned int val)
57 58 59 60 61 62 63 64
{
	*str++ = to_hex[val >> 4];
	*str++ = to_hex[val & 0xf];
	return str;
}

void git_oid_fmt(char *str, const git_oid *oid)
{
65
	size_t i;
66 67 68 69 70 71 72

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

void git_oid_pathfmt(char *str, const git_oid *oid)
{
73
	size_t i;
74 75 76 77 78 79 80 81 82

	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)
{
83
	char *str = git__malloc(GIT_OID_HEXSZ + 1);
84 85 86 87 88 89
	if (!str)
		return NULL;
	git_oid_fmt(str, oid);
	str[GIT_OID_HEXSZ] = '\0';
	return str;
}
90 91 92 93 94 95 96 97

char *git_oid_to_string(char *out, size_t n, const git_oid *oid)
{
	char str[GIT_OID_HEXSZ];

	if (!out || n == 0 || !oid)
		return "";

Vicent Marti committed
98
	n--; /* allow room for terminating NUL */
99 100 101

	if (n > 0) {
		git_oid_fmt(str, oid);
102 103 104
		if (n > GIT_OID_HEXSZ)
			n = GIT_OID_HEXSZ;
		memcpy(out, str, n);
105 106 107 108 109 110 111
	}

	out[n] = '\0';

	return out;
}

Vicent Marti committed
112
int git_oid__parse(git_oid *oid, const char **buffer_out,
113 114 115 116 117
		const char *buffer_end, const char *header)
{
	const size_t sha_len = GIT_OID_HEXSZ;
	const size_t header_len = strlen(header);

118
	const char *buffer = *buffer_out;
119 120

	if (buffer + (header_len + sha_len + 1) > buffer_end)
121
		return git__throw(GIT_EOBJCORRUPTED, "Failed to parse OID. Buffer too small");
122 123

	if (memcmp(buffer, header, header_len) != 0)
124
		return git__throw(GIT_EOBJCORRUPTED, "Failed to parse OID. Buffer and header do not match");
125 126

	if (buffer[header_len + sha_len] != '\n')
127
		return git__throw(GIT_EOBJCORRUPTED, "Failed to parse OID. Buffer not terminated correctly");
128

Vicent Marti committed
129
	if (git_oid_fromstr(oid, buffer + header_len) < GIT_SUCCESS)
130
		return git__throw(GIT_EOBJCORRUPTED, "Failed to parse OID. Failed to generate sha1");
131 132 133

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

134
	return GIT_SUCCESS;
135 136
}

137 138 139 140 141 142 143 144 145 146
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
147
void git_oid_fromraw(git_oid *out, const unsigned char *raw)
148 149 150 151 152 153 154 155 156 157 158 159 160
{
	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));
}

int git_oid_cmp(const git_oid *a, const git_oid *b)
{
	return memcmp(a->id, b->id, sizeof(a->id));
}
161

162
int git_oid_ncmp(const git_oid *oid_a, const git_oid *oid_b, unsigned int len)
163
{
164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179
	const unsigned char *a = oid_a->id;
	const unsigned char *b = oid_b->id;

	do {
		if (*a != *b)
			return 1;
		a++;
		b++;
		len -= 2;
	} while (len > 1);

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

	return 0;
180 181
}

182 183 184 185 186 187 188 189 190 191 192
int git_oid_streq(const git_oid *a, const char *str)
{
	git_oid id;
	int error;

	if ((error = git_oid_fromstr(&id, str)) < GIT_SUCCESS)
		return git__rethrow(error, "Failed to convert '%s' to oid.", str);

	return git_oid_cmp(a, &id) == 0 ? GIT_SUCCESS : GIT_ERROR;
}

193 194 195 196 197 198 199 200 201 202 203 204 205 206 207
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)
{
208
	self->nodes = git__realloc(self->nodes, new_size * sizeof(trie_node));
209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254
	if (self->nodes == NULL)
		return GIT_ENOMEM;

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

	self->size = new_size;
	return GIT_SUCCESS;
}

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) {
		if (resize_trie(os, os->size * 2) < GIT_SUCCESS)
			return NULL;
	}

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

	if (os->node_count == SHRT_MAX)
		os->full = 1;

	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;

	os = git__malloc(sizeof(git_oid_shorten));
	if (os == NULL)
		return NULL;

	memset(os, 0x0, sizeof(git_oid_shorten));

	if (resize_trie(os, 16) < GIT_SUCCESS) {
255
		git__free(os);
256 257 258 259 260 261 262 263 264 265 266
		return NULL;
	}

	os->node_count = 1;
	os->min_length = min_length;

	return os;
}

void git_oid_shorten_free(git_oid_shorten *os)
{
267 268
	git__free(os->nodes);
	git__free(os);
269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287
}


/*
 * 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
288
 *	pointers, because in a 64-bit arch this would be sucking
289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323
 *	16*sizeof(void*) = 128 bytes of memory per node, which is fucking
 *	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)
{
	int i, is_leaf;
	node_index idx;

	if (os->full)
		return GIT_ENOMEM;

324 325 326
	if (text_oid == NULL)
		return os->min_length;

327 328 329 330
	idx = 0;
	is_leaf = 0;

	for (i = 0; i < GIT_OID_HEXSZ; ++i) {
nulltoken committed
331
		int c = git__fromhex(text_oid[i]);
332 333 334
		trie_node *node;

		if (c == -1)
335
			return git__throw(GIT_ENOTOID, "Failed to shorten OID. Invalid hex value");
336 337 338 339 340 341 342 343 344

		node = &os->nodes[idx];

		if (is_leaf) {
			const char *tail;

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

nulltoken committed
345
			node = push_leaf(os, idx, git__fromhex(tail[0]), &tail[1]);
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
			if (node == NULL)
				return GIT_ENOMEM;
		}

		if (node->children[c] == 0) {
			if (push_leaf(os, idx, c, &text_oid[i + 1]) == NULL)
				return GIT_ENOMEM;
			break;
		}

		idx = node->children[c];
		is_leaf = 0;

		if (idx < 0) {
			node->children[c] = idx = -idx;
			is_leaf = 1;
		}
	}

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

	return os->min_length;
}