]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Copyright (C) 2009-2011 the libgit2 contributors | |
3 | * | |
4 | * This file is part of libgit2, distributed under the GNU GPL v2 with | |
5 | * 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 signed char from_hex[] = { | |
15 | -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 00 */ | |
16 | -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10 */ | |
17 | -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20 */ | |
18 | 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, -1, -1, -1, -1, -1, -1, /* 30 */ | |
19 | -1, 10, 11, 12, 13, 14, 15, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 40 */ | |
20 | -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 50 */ | |
21 | -1, 10, 11, 12, 13, 14, 15, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 60 */ | |
22 | -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 70 */ | |
23 | -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 80 */ | |
24 | -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 90 */ | |
25 | -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* a0 */ | |
26 | -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* b0 */ | |
27 | -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* c0 */ | |
28 | -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* d0 */ | |
29 | -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* e0 */ | |
30 | -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* f0 */ | |
31 | }; | |
32 | static char to_hex[] = "0123456789abcdef"; | |
33 | ||
34 | int git_oid_fromstrn(git_oid *out, const char *str, size_t length) | |
35 | { | |
36 | size_t p; | |
37 | int v; | |
38 | ||
39 | if (length < 4) | |
40 | return git__throw(GIT_ENOTOID, "Failed to generate sha1. Given string is too short"); | |
41 | ||
42 | if (length > GIT_OID_HEXSZ) | |
43 | length = GIT_OID_HEXSZ; | |
44 | ||
45 | for (p = 0; p < length - 1; p += 2) { | |
46 | v = (from_hex[(unsigned char)str[p + 0]] << 4) | |
47 | | from_hex[(unsigned char)str[p + 1]]; | |
48 | ||
49 | if (v < 0) | |
50 | return git__throw(GIT_ENOTOID, "Failed to generate sha1. Given string is not a valid sha1 hash"); | |
51 | ||
52 | out->id[p / 2] = (unsigned char)v; | |
53 | } | |
54 | ||
55 | if (length % 2) { | |
56 | v = (from_hex[(unsigned char)str[p + 0]] << 4); | |
57 | if (v < 0) | |
58 | return git__throw(GIT_ENOTOID, "Failed to generate sha1. Given string is not a valid sha1 hash"); | |
59 | ||
60 | out->id[p / 2] = (unsigned char)v; | |
61 | p += 2; | |
62 | } | |
63 | ||
64 | memset(out->id + p / 2, 0, (GIT_OID_HEXSZ - p) / 2); | |
65 | ||
66 | return GIT_SUCCESS; | |
67 | } | |
68 | ||
69 | int git_oid_fromstr(git_oid *out, const char *str) | |
70 | { | |
71 | return git_oid_fromstrn(out, str, GIT_OID_HEXSZ); | |
72 | } | |
73 | ||
74 | GIT_INLINE(char) *fmt_one(char *str, unsigned int val) | |
75 | { | |
76 | *str++ = to_hex[val >> 4]; | |
77 | *str++ = to_hex[val & 0xf]; | |
78 | return str; | |
79 | } | |
80 | ||
81 | void git_oid_fmt(char *str, const git_oid *oid) | |
82 | { | |
83 | size_t i; | |
84 | ||
85 | for (i = 0; i < sizeof(oid->id); i++) | |
86 | str = fmt_one(str, oid->id[i]); | |
87 | } | |
88 | ||
89 | void git_oid_pathfmt(char *str, const git_oid *oid) | |
90 | { | |
91 | size_t i; | |
92 | ||
93 | str = fmt_one(str, oid->id[0]); | |
94 | *str++ = '/'; | |
95 | for (i = 1; i < sizeof(oid->id); i++) | |
96 | str = fmt_one(str, oid->id[i]); | |
97 | } | |
98 | ||
99 | char *git_oid_allocfmt(const git_oid *oid) | |
100 | { | |
101 | char *str = git__malloc(GIT_OID_HEXSZ + 1); | |
102 | if (!str) | |
103 | return NULL; | |
104 | git_oid_fmt(str, oid); | |
105 | str[GIT_OID_HEXSZ] = '\0'; | |
106 | return str; | |
107 | } | |
108 | ||
109 | char *git_oid_to_string(char *out, size_t n, const git_oid *oid) | |
110 | { | |
111 | char str[GIT_OID_HEXSZ]; | |
112 | ||
113 | if (!out || n == 0 || !oid) | |
114 | return ""; | |
115 | ||
116 | n--; /* allow room for terminating NUL */ | |
117 | ||
118 | if (n > 0) { | |
119 | git_oid_fmt(str, oid); | |
120 | if (n > GIT_OID_HEXSZ) | |
121 | n = GIT_OID_HEXSZ; | |
122 | memcpy(out, str, n); | |
123 | } | |
124 | ||
125 | out[n] = '\0'; | |
126 | ||
127 | return out; | |
128 | } | |
129 | ||
130 | int git_oid__parse(git_oid *oid, const char **buffer_out, | |
131 | const char *buffer_end, const char *header) | |
132 | { | |
133 | const size_t sha_len = GIT_OID_HEXSZ; | |
134 | const size_t header_len = strlen(header); | |
135 | ||
136 | const char *buffer = *buffer_out; | |
137 | ||
138 | if (buffer + (header_len + sha_len + 1) > buffer_end) | |
139 | return git__throw(GIT_EOBJCORRUPTED, "Failed to parse OID. Buffer too small"); | |
140 | ||
141 | if (memcmp(buffer, header, header_len) != 0) | |
142 | return git__throw(GIT_EOBJCORRUPTED, "Failed to parse OID. Buffer and header do not match"); | |
143 | ||
144 | if (buffer[header_len + sha_len] != '\n') | |
145 | return git__throw(GIT_EOBJCORRUPTED, "Failed to parse OID. Buffer not terminated correctly"); | |
146 | ||
147 | if (git_oid_fromstr(oid, buffer + header_len) < GIT_SUCCESS) | |
148 | return git__throw(GIT_EOBJCORRUPTED, "Failed to parse OID. Failed to generate sha1"); | |
149 | ||
150 | *buffer_out = buffer + (header_len + sha_len + 1); | |
151 | ||
152 | return GIT_SUCCESS; | |
153 | } | |
154 | ||
155 | void git_oid__writebuf(git_buf *buf, const char *header, const git_oid *oid) | |
156 | { | |
157 | char hex_oid[GIT_OID_HEXSZ]; | |
158 | ||
159 | git_oid_fmt(hex_oid, oid); | |
160 | git_buf_puts(buf, header); | |
161 | git_buf_put(buf, hex_oid, GIT_OID_HEXSZ); | |
162 | git_buf_putc(buf, '\n'); | |
163 | } | |
164 | ||
165 | void git_oid_fromraw(git_oid *out, const unsigned char *raw) | |
166 | { | |
167 | memcpy(out->id, raw, sizeof(out->id)); | |
168 | } | |
169 | ||
170 | void git_oid_cpy(git_oid *out, const git_oid *src) | |
171 | { | |
172 | memcpy(out->id, src->id, sizeof(out->id)); | |
173 | } | |
174 | ||
175 | int git_oid_cmp(const git_oid *a, const git_oid *b) | |
176 | { | |
177 | return memcmp(a->id, b->id, sizeof(a->id)); | |
178 | } | |
179 | ||
180 | int git_oid_ncmp(const git_oid *oid_a, const git_oid *oid_b, unsigned int len) | |
181 | { | |
182 | const unsigned char *a = oid_a->id; | |
183 | const unsigned char *b = oid_b->id; | |
184 | ||
185 | do { | |
186 | if (*a != *b) | |
187 | return 1; | |
188 | a++; | |
189 | b++; | |
190 | len -= 2; | |
191 | } while (len > 1); | |
192 | ||
193 | if (len) | |
194 | if ((*a ^ *b) & 0xf0) | |
195 | return 1; | |
196 | ||
197 | return 0; | |
198 | } | |
199 | ||
200 | typedef short node_index; | |
201 | ||
202 | typedef union { | |
203 | const char *tail; | |
204 | node_index children[16]; | |
205 | } trie_node; | |
206 | ||
207 | struct git_oid_shorten { | |
208 | trie_node *nodes; | |
209 | size_t node_count, size; | |
210 | int min_length, full; | |
211 | }; | |
212 | ||
213 | static int resize_trie(git_oid_shorten *self, size_t new_size) | |
214 | { | |
215 | self->nodes = realloc(self->nodes, new_size * sizeof(trie_node)); | |
216 | if (self->nodes == NULL) | |
217 | return GIT_ENOMEM; | |
218 | ||
219 | if (new_size > self->size) { | |
220 | memset(&self->nodes[self->size], 0x0, (new_size - self->size) * sizeof(trie_node)); | |
221 | } | |
222 | ||
223 | self->size = new_size; | |
224 | return GIT_SUCCESS; | |
225 | } | |
226 | ||
227 | static trie_node *push_leaf(git_oid_shorten *os, node_index idx, int push_at, const char *oid) | |
228 | { | |
229 | trie_node *node, *leaf; | |
230 | node_index idx_leaf; | |
231 | ||
232 | if (os->node_count >= os->size) { | |
233 | if (resize_trie(os, os->size * 2) < GIT_SUCCESS) | |
234 | return NULL; | |
235 | } | |
236 | ||
237 | idx_leaf = (node_index)os->node_count++; | |
238 | ||
239 | if (os->node_count == SHRT_MAX) | |
240 | os->full = 1; | |
241 | ||
242 | node = &os->nodes[idx]; | |
243 | node->children[push_at] = -idx_leaf; | |
244 | ||
245 | leaf = &os->nodes[idx_leaf]; | |
246 | leaf->tail = oid; | |
247 | ||
248 | return node; | |
249 | } | |
250 | ||
251 | git_oid_shorten *git_oid_shorten_new(size_t min_length) | |
252 | { | |
253 | git_oid_shorten *os; | |
254 | ||
255 | os = git__malloc(sizeof(git_oid_shorten)); | |
256 | if (os == NULL) | |
257 | return NULL; | |
258 | ||
259 | memset(os, 0x0, sizeof(git_oid_shorten)); | |
260 | ||
261 | if (resize_trie(os, 16) < GIT_SUCCESS) { | |
262 | free(os); | |
263 | return NULL; | |
264 | } | |
265 | ||
266 | os->node_count = 1; | |
267 | os->min_length = min_length; | |
268 | ||
269 | return os; | |
270 | } | |
271 | ||
272 | void git_oid_shorten_free(git_oid_shorten *os) | |
273 | { | |
274 | free(os->nodes); | |
275 | free(os); | |
276 | } | |
277 | ||
278 | ||
279 | /* | |
280 | * What wizardry is this? | |
281 | * | |
282 | * This is just a memory-optimized trie: basically a very fancy | |
283 | * 16-ary tree, which is used to store the prefixes of the OID | |
284 | * strings. | |
285 | * | |
286 | * Read more: http://en.wikipedia.org/wiki/Trie | |
287 | * | |
288 | * Magic that happens in this method: | |
289 | * | |
290 | * - Each node in the trie is an union, so it can work both as | |
291 | * a normal node, or as a leaf. | |
292 | * | |
293 | * - Each normal node points to 16 children (one for each possible | |
294 | * character in the oid). This is *not* stored in an array of | |
295 | * pointers, because in a 64-bit arch this would be sucking | |
296 | * 16*sizeof(void*) = 128 bytes of memory per node, which is fucking | |
297 | * insane. What we do is store Node Indexes, and use these indexes | |
298 | * to look up each node in the om->index array. These indexes are | |
299 | * signed shorts, so this limits the amount of unique OIDs that | |
300 | * fit in the structure to about 20000 (assuming a more or less uniform | |
301 | * distribution). | |
302 | * | |
303 | * - All the nodes in om->index array are stored contiguously in | |
304 | * memory, and each of them is 32 bytes, so we fit 2x nodes per | |
305 | * cache line. Convenient for speed. | |
306 | * | |
307 | * - To differentiate the leafs from the normal nodes, we store all | |
308 | * the indexes towards a leaf as a negative index (indexes to normal | |
309 | * nodes are positives). When we find that one of the children for | |
310 | * a node has a negative value, that means it's going to be a leaf. | |
311 | * This reduces the amount of indexes we have by two, but also reduces | |
312 | * the size of each node by 1-4 bytes (the amount we would need to | |
313 | * add a `is_leaf` field): this is good because it allows the nodes | |
314 | * to fit cleanly in cache lines. | |
315 | * | |
316 | * - Once we reach an empty children, instead of continuing to insert | |
317 | * new nodes for each remaining character of the OID, we store a pointer | |
318 | * to the tail in the leaf; if the leaf is reached again, we turn it | |
319 | * into a normal node and use the tail to create a new leaf. | |
320 | * | |
321 | * This is a pretty good balance between performance and memory usage. | |
322 | */ | |
323 | int git_oid_shorten_add(git_oid_shorten *os, const char *text_oid) | |
324 | { | |
325 | int i, is_leaf; | |
326 | node_index idx; | |
327 | ||
328 | if (os->full) | |
329 | return GIT_ENOMEM; | |
330 | ||
331 | if (text_oid == NULL) | |
332 | return os->min_length; | |
333 | ||
334 | idx = 0; | |
335 | is_leaf = 0; | |
336 | ||
337 | for (i = 0; i < GIT_OID_HEXSZ; ++i) { | |
338 | int c = from_hex[(int)text_oid[i]]; | |
339 | trie_node *node; | |
340 | ||
341 | if (c == -1) | |
342 | return git__throw(GIT_ENOTOID, "Failed to shorten OID. Invalid hex value"); | |
343 | ||
344 | node = &os->nodes[idx]; | |
345 | ||
346 | if (is_leaf) { | |
347 | const char *tail; | |
348 | ||
349 | tail = node->tail; | |
350 | node->tail = NULL; | |
351 | ||
352 | node = push_leaf(os, idx, from_hex[(int)tail[0]], &tail[1]); | |
353 | if (node == NULL) | |
354 | return GIT_ENOMEM; | |
355 | } | |
356 | ||
357 | if (node->children[c] == 0) { | |
358 | if (push_leaf(os, idx, c, &text_oid[i + 1]) == NULL) | |
359 | return GIT_ENOMEM; | |
360 | break; | |
361 | } | |
362 | ||
363 | idx = node->children[c]; | |
364 | is_leaf = 0; | |
365 | ||
366 | if (idx < 0) { | |
367 | node->children[c] = idx = -idx; | |
368 | is_leaf = 1; | |
369 | } | |
370 | } | |
371 | ||
372 | if (++i > os->min_length) | |
373 | os->min_length = i; | |
374 | ||
375 | return os->min_length; | |
376 | } | |
377 |