]>
Commit | Line | Data |
---|---|---|
c1bb9350 LT |
1 | /* |
2 | * apply.c | |
3 | * | |
4 | * Copyright (C) Linus Torvalds, 2005 | |
5 | * | |
6 | * This applies patches on top of some (arbitrary) version of the SCM. | |
7 | * | |
8 | * NOTE! It does all its work in the index file, and only cares about | |
9 | * the files in the working directory if you tell it to "merge" the | |
10 | * patch apply. | |
11 | * | |
12 | * Even when merging it always takes the source from the index, and | |
13 | * uses the working tree as a "branch" for a 3-way merge. | |
14 | */ | |
15 | #include <ctype.h> | |
16 | ||
17 | #include "cache.h" | |
18 | ||
19 | // We default to the merge behaviour, since that's what most people would | |
20 | // expect | |
21 | static int merge_patch = 1; | |
22 | static const char apply_usage[] = "git-apply <patch>"; | |
23 | ||
a4acb0eb LT |
24 | /* |
25 | * Various "current state", notably line numbers and what | |
26 | * file (and how) we're patching right now.. The "is_xxxx" | |
27 | * things are flags, where -1 means "don't know yet". | |
28 | */ | |
46979f56 | 29 | static int linenr = 1; |
a4acb0eb LT |
30 | static int old_mode, new_mode; |
31 | static char *old_name, *new_name, *def_name; | |
32 | static int is_rename, is_copy, is_new, is_delete; | |
46979f56 | 33 | |
c1bb9350 | 34 | #define CHUNKSIZE (8192) |
a4acb0eb | 35 | #define SLOP (16) |
c1bb9350 LT |
36 | |
37 | static void *read_patch_file(int fd, unsigned long *sizep) | |
38 | { | |
39 | unsigned long size = 0, alloc = CHUNKSIZE; | |
40 | void *buffer = xmalloc(alloc); | |
41 | ||
42 | for (;;) { | |
43 | int nr = alloc - size; | |
44 | if (nr < 1024) { | |
45 | alloc += CHUNKSIZE; | |
46 | buffer = xrealloc(buffer, alloc); | |
47 | nr = alloc - size; | |
48 | } | |
49 | nr = read(fd, buffer + size, nr); | |
50 | if (!nr) | |
51 | break; | |
52 | if (nr < 0) { | |
53 | if (errno == EAGAIN) | |
54 | continue; | |
55 | die("git-apply: read returned %s", strerror(errno)); | |
56 | } | |
57 | size += nr; | |
58 | } | |
59 | *sizep = size; | |
a4acb0eb LT |
60 | |
61 | /* | |
62 | * Make sure that we have some slop in the buffer | |
63 | * so that we can do speculative "memcmp" etc, and | |
64 | * see to it that it is NUL-filled. | |
65 | */ | |
66 | if (alloc < size + SLOP) | |
67 | buffer = xrealloc(buffer, size + SLOP); | |
68 | memset(buffer + size, 0, SLOP); | |
c1bb9350 LT |
69 | return buffer; |
70 | } | |
71 | ||
72 | static unsigned long linelen(char *buffer, unsigned long size) | |
73 | { | |
74 | unsigned long len = 0; | |
75 | while (size--) { | |
76 | len++; | |
77 | if (*buffer++ == '\n') | |
78 | break; | |
79 | } | |
80 | return len; | |
81 | } | |
82 | ||
a4acb0eb LT |
83 | static int is_dev_null(const char *str) |
84 | { | |
85 | return !memcmp("/dev/null", str, 9) && isspace(str[9]); | |
86 | } | |
87 | ||
88 | static char * find_name(const char *line, char *def, int p_value) | |
c1bb9350 | 89 | { |
a4acb0eb LT |
90 | int len; |
91 | const char *start = line; | |
92 | char *name; | |
93 | ||
c1bb9350 | 94 | for (;;) { |
a4acb0eb LT |
95 | char c = *line; |
96 | if (isspace(c)) | |
c1bb9350 | 97 | break; |
a4acb0eb LT |
98 | line++; |
99 | if (c == '/' && !--p_value) | |
100 | start = line; | |
101 | } | |
102 | if (!start) | |
103 | return def; | |
104 | len = line - start; | |
105 | if (!len) | |
106 | return def; | |
107 | ||
108 | /* | |
109 | * Generally we prefer the shorter name, especially | |
110 | * if the other one is just a variation of that with | |
111 | * something else tacked on to the end (ie "file.orig" | |
112 | * or "file~"). | |
113 | */ | |
114 | if (def) { | |
115 | int deflen = strlen(def); | |
116 | if (deflen < len && !strncmp(start, def, deflen)) | |
117 | return def; | |
c1bb9350 | 118 | } |
a4acb0eb LT |
119 | |
120 | name = xmalloc(len + 1); | |
121 | memcpy(name, start, len); | |
122 | name[len] = 0; | |
123 | free(def); | |
124 | return name; | |
125 | } | |
126 | ||
127 | /* | |
128 | * Get the name etc info from the --/+++ lines of a traditional patch header | |
129 | * | |
130 | * NOTE! This hardcodes "-p1" behaviour in filename detection. | |
131 | */ | |
132 | static int parse_traditional_patch(const char *first, const char *second) | |
133 | { | |
134 | int p_value = 1; | |
135 | char *name; | |
136 | ||
137 | first += 4; // skip "--- " | |
138 | second += 4; // skip "+++ " | |
139 | if (is_dev_null(first)) { | |
140 | is_new = 1; | |
141 | name = find_name(second, def_name, p_value); | |
142 | } else if (is_dev_null(second)) { | |
143 | is_delete = 1; | |
144 | name = find_name(first, def_name, p_value); | |
145 | } else { | |
146 | name = find_name(first, def_name, p_value); | |
147 | name = find_name(second, name, p_value); | |
148 | } | |
149 | if (!name) | |
150 | die("unable to find filename in patch at line %d", linenr); | |
151 | old_name = name; | |
152 | new_name = name; | |
153 | } | |
154 | ||
155 | static int gitdiff_hdrend(const char *line) | |
156 | { | |
157 | return -1; | |
158 | } | |
159 | ||
160 | static int gitdiff_oldname(const char *line) | |
161 | { | |
162 | if (!old_name) | |
163 | old_name = find_name(line, NULL, 1); | |
164 | return 0; | |
165 | } | |
166 | ||
167 | static int gitdiff_newname(const char *line) | |
168 | { | |
169 | if (!new_name) | |
170 | new_name = find_name(line, NULL, 1); | |
171 | return 0; | |
172 | } | |
173 | ||
174 | static int gitdiff_oldmode(const char *line) | |
175 | { | |
176 | old_mode = strtoul(line, NULL, 8); | |
177 | return 0; | |
178 | } | |
179 | ||
180 | static int gitdiff_newmode(const char *line) | |
181 | { | |
182 | new_mode = strtoul(line, NULL, 8); | |
183 | return 0; | |
184 | } | |
185 | ||
186 | static int gitdiff_delete(const char *line) | |
187 | { | |
188 | is_delete = 1; | |
189 | return gitdiff_oldmode(line); | |
190 | } | |
191 | ||
192 | static int gitdiff_newfile(const char *line) | |
193 | { | |
194 | is_new = 1; | |
195 | return gitdiff_newmode(line); | |
196 | } | |
197 | ||
198 | static int gitdiff_copysrc(const char *line) | |
199 | { | |
200 | is_copy = 1; | |
201 | old_name = find_name(line, NULL, 0); | |
202 | return 0; | |
203 | } | |
204 | ||
205 | static int gitdiff_copydst(const char *line) | |
206 | { | |
207 | is_copy = 1; | |
208 | new_name = find_name(line, NULL, 0); | |
209 | return 0; | |
210 | } | |
211 | ||
212 | static int gitdiff_renamesrc(const char *line) | |
213 | { | |
214 | is_rename = 1; | |
215 | old_name = find_name(line, NULL, 0); | |
216 | return 0; | |
217 | } | |
218 | ||
219 | static int gitdiff_renamedst(const char *line) | |
220 | { | |
221 | is_rename = 1; | |
222 | new_name = find_name(line, NULL, 0); | |
223 | return 0; | |
224 | } | |
225 | ||
226 | static int gitdiff_similarity(const char *line) | |
227 | { | |
228 | return 0; | |
c1bb9350 LT |
229 | } |
230 | ||
231 | /* Verify that we recognize the lines following a git header */ | |
a4acb0eb | 232 | static int parse_git_header(char *line, int len, unsigned int size) |
c1bb9350 | 233 | { |
a4acb0eb LT |
234 | unsigned long offset; |
235 | ||
236 | /* A git diff has explicit new/delete information, so we don't guess */ | |
237 | is_new = 0; | |
238 | is_delete = 0; | |
239 | ||
240 | line += len; | |
241 | size -= len; | |
242 | linenr++; | |
243 | for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) { | |
244 | static const struct opentry { | |
245 | const char *str; | |
246 | int (*fn)(const char *); | |
247 | } optable[] = { | |
248 | { "@@ -", gitdiff_hdrend }, | |
249 | { "--- ", gitdiff_oldname }, | |
250 | { "+++ ", gitdiff_newname }, | |
251 | { "old mode ", gitdiff_oldmode }, | |
252 | { "new mode ", gitdiff_newmode }, | |
253 | { "deleted file mode ", gitdiff_delete }, | |
254 | { "new file mode ", gitdiff_newfile }, | |
255 | { "copy from ", gitdiff_copysrc }, | |
256 | { "copy to ", gitdiff_copydst }, | |
257 | { "rename from ", gitdiff_renamesrc }, | |
258 | { "rename to ", gitdiff_renamedst }, | |
259 | { "similarity index ", gitdiff_similarity }, | |
260 | }; | |
261 | int i; | |
c1bb9350 | 262 | |
c1bb9350 | 263 | len = linelen(line, size); |
a4acb0eb | 264 | if (!len || line[len-1] != '\n') |
c1bb9350 | 265 | break; |
a4acb0eb LT |
266 | for (i = 0; i < sizeof(optable) / sizeof(optable[0]); i++) { |
267 | const struct opentry *p = optable + i; | |
268 | int oplen = strlen(p->str); | |
269 | if (len < oplen || memcmp(p->str, line, oplen)) | |
270 | continue; | |
271 | if (p->fn(line + oplen) < 0) | |
272 | return offset; | |
273 | } | |
c1bb9350 LT |
274 | } |
275 | ||
a4acb0eb | 276 | return offset; |
c1bb9350 LT |
277 | } |
278 | ||
46979f56 LT |
279 | static int parse_num(const char *line, int len, int offset, const char *expect, unsigned long *p) |
280 | { | |
281 | char *ptr; | |
282 | int digits, ex; | |
283 | ||
284 | if (offset < 0 || offset >= len) | |
285 | return -1; | |
286 | line += offset; | |
287 | len -= offset; | |
288 | ||
289 | if (!isdigit(*line)) | |
290 | return -1; | |
291 | *p = strtoul(line, &ptr, 10); | |
292 | ||
293 | digits = ptr - line; | |
294 | ||
295 | offset += digits; | |
296 | line += digits; | |
297 | len -= digits; | |
298 | ||
299 | ex = strlen(expect); | |
300 | if (ex > len) | |
301 | return -1; | |
302 | if (memcmp(line, expect, ex)) | |
303 | return -1; | |
304 | ||
305 | return offset + ex; | |
306 | } | |
307 | ||
308 | /* | |
309 | * Parse a unified diff fragment header of the | |
310 | * form "@@ -a,b +c,d @@" | |
311 | */ | |
312 | static int parse_fragment_header(char *line, int len, unsigned long *pos) | |
313 | { | |
314 | int offset; | |
315 | ||
316 | if (!len || line[len-1] != '\n') | |
317 | return -1; | |
318 | ||
319 | /* Figure out the number of lines in a fragment */ | |
320 | offset = parse_num(line, len, 4, ",", pos); | |
321 | offset = parse_num(line, len, offset, " +", pos+1); | |
322 | offset = parse_num(line, len, offset, ",", pos+2); | |
323 | offset = parse_num(line, len, offset, " @@", pos+3); | |
324 | ||
325 | return offset; | |
326 | } | |
327 | ||
c1bb9350 LT |
328 | static int find_header(char *line, unsigned long size, int *hdrsize) |
329 | { | |
330 | unsigned long offset, len; | |
331 | ||
a4acb0eb LT |
332 | is_rename = is_copy = 0; |
333 | is_new = is_delete = -1; | |
881b0765 | 334 | old_mode = new_mode = 0; |
a4acb0eb | 335 | def_name = old_name = new_name = NULL; |
46979f56 | 336 | for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) { |
c1bb9350 LT |
337 | unsigned long nextlen; |
338 | ||
339 | len = linelen(line, size); | |
340 | if (!len) | |
341 | break; | |
342 | ||
343 | /* Testing this early allows us to take a few shortcuts.. */ | |
344 | if (len < 6) | |
345 | continue; | |
46979f56 LT |
346 | |
347 | /* | |
348 | * Make sure we don't find any unconnected patch fragmants. | |
349 | * That's a sign that we didn't find a header, and that a | |
350 | * patch has become corrupted/broken up. | |
351 | */ | |
352 | if (!memcmp("@@ -", line, 4)) { | |
353 | unsigned long pos[4]; | |
354 | if (parse_fragment_header(line, len, pos) < 0) | |
355 | continue; | |
356 | error("patch fragment without header at line %d: %.*s", linenr, len-1, line); | |
357 | } | |
358 | ||
c1bb9350 LT |
359 | if (size < len + 6) |
360 | break; | |
361 | ||
362 | /* | |
363 | * Git patch? It might not have a real patch, just a rename | |
364 | * or mode change, so we handle that specially | |
365 | */ | |
366 | if (!memcmp("diff --git ", line, 11)) { | |
a4acb0eb | 367 | int git_hdr_len = parse_git_header(line, len, size); |
c1bb9350 LT |
368 | if (git_hdr_len < 0) |
369 | continue; | |
370 | ||
a4acb0eb | 371 | *hdrsize = git_hdr_len; |
c1bb9350 LT |
372 | return offset; |
373 | } | |
374 | ||
375 | /** --- followed by +++ ? */ | |
376 | if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4)) | |
377 | continue; | |
378 | ||
379 | /* | |
380 | * We only accept unified patches, so we want it to | |
381 | * at least have "@@ -a,b +c,d @@\n", which is 14 chars | |
382 | * minimum | |
383 | */ | |
384 | nextlen = linelen(line + len, size - len); | |
385 | if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4)) | |
386 | continue; | |
387 | ||
388 | /* Ok, we'll consider it a patch */ | |
a4acb0eb | 389 | parse_traditional_patch(line, line+len); |
c1bb9350 | 390 | *hdrsize = len + nextlen; |
46979f56 | 391 | linenr += 2; |
c1bb9350 LT |
392 | return offset; |
393 | } | |
394 | return -1; | |
395 | } | |
396 | ||
c1bb9350 LT |
397 | /* |
398 | * Parse a unified diff. Note that this really needs | |
399 | * to parse each fragment separately, since the only | |
400 | * way to know the difference between a "---" that is | |
401 | * part of a patch, and a "---" that starts the next | |
402 | * patch is to look at the line counts.. | |
403 | */ | |
404 | static int apply_fragment(char *line, unsigned long size) | |
405 | { | |
406 | int len = linelen(line, size), offset; | |
46979f56 | 407 | unsigned long pos[4], oldlines, newlines; |
c1bb9350 | 408 | |
46979f56 | 409 | offset = parse_fragment_header(line, len, pos); |
c1bb9350 LT |
410 | if (offset < 0) |
411 | return -1; | |
46979f56 LT |
412 | oldlines = pos[1]; |
413 | newlines = pos[3]; | |
c1bb9350 | 414 | |
a4acb0eb LT |
415 | if (is_new < 0 && (pos[0] || oldlines)) |
416 | is_new = 0; | |
417 | if (is_delete < 0 && (pos[1] || newlines)) | |
418 | is_delete = 0; | |
419 | ||
c1bb9350 LT |
420 | /* Parse the thing.. */ |
421 | line += len; | |
422 | size -= len; | |
46979f56 LT |
423 | linenr++; |
424 | for (offset = len; size > 0; offset += len, size -= len, line += len, linenr++) { | |
c1bb9350 LT |
425 | if (!oldlines && !newlines) |
426 | break; | |
427 | len = linelen(line, size); | |
428 | if (!len || line[len-1] != '\n') | |
429 | return -1; | |
430 | switch (*line) { | |
431 | default: | |
432 | return -1; | |
433 | case ' ': | |
434 | oldlines--; | |
435 | newlines--; | |
436 | break; | |
437 | case '-': | |
438 | oldlines--; | |
439 | break; | |
440 | case '+': | |
441 | newlines--; | |
442 | break; | |
443 | } | |
444 | } | |
445 | return offset; | |
446 | } | |
447 | ||
448 | static int apply_single_patch(char *line, unsigned long size) | |
449 | { | |
450 | unsigned long offset = 0; | |
451 | ||
452 | while (size > 4 && !memcmp(line, "@@ -", 4)) { | |
453 | int len = apply_fragment(line, size); | |
454 | if (len <= 0) | |
46979f56 | 455 | die("corrupt patch at line %d", linenr); |
c1bb9350 LT |
456 | |
457 | printf("applying fragment:\n%.*s\n\n", len, line); | |
458 | ||
459 | offset += len; | |
460 | line += len; | |
461 | size -= len; | |
462 | } | |
463 | return offset; | |
464 | } | |
465 | ||
466 | static int apply_chunk(char *buffer, unsigned long size) | |
467 | { | |
468 | int hdrsize, patchsize; | |
469 | int offset = find_header(buffer, size, &hdrsize); | |
470 | char *header, *patch; | |
471 | ||
472 | if (offset < 0) | |
473 | return offset; | |
474 | header = buffer + offset; | |
475 | ||
476 | printf("Found header:\n%.*s\n\n", hdrsize, header); | |
a4acb0eb LT |
477 | printf("Rename: %d\n", is_rename); |
478 | printf("Copy: %d\n", is_copy); | |
479 | printf("New: %d\n", is_new); | |
480 | printf("Delete: %d\n", is_delete); | |
481 | printf("Mode: %o->%o\n", old_mode, new_mode); | |
482 | printf("Name: '%s'->'%s'\n", old_name, new_name); | |
c1bb9350 LT |
483 | |
484 | patch = header + hdrsize; | |
485 | patchsize = apply_single_patch(patch, size - offset - hdrsize); | |
486 | ||
487 | return offset + hdrsize + patchsize; | |
488 | } | |
489 | ||
490 | static int apply_patch(int fd) | |
491 | { | |
492 | unsigned long offset, size; | |
493 | char *buffer = read_patch_file(fd, &size); | |
494 | ||
495 | if (!buffer) | |
496 | return -1; | |
497 | offset = 0; | |
498 | while (size > 0) { | |
499 | int nr = apply_chunk(buffer + offset, size); | |
500 | if (nr < 0) | |
501 | break; | |
502 | offset += nr; | |
503 | size -= nr; | |
504 | } | |
505 | free(buffer); | |
506 | return 0; | |
507 | } | |
508 | ||
509 | int main(int argc, char **argv) | |
510 | { | |
511 | int i; | |
4dfdbe10 | 512 | int read_stdin = 1; |
c1bb9350 LT |
513 | |
514 | if (read_cache() < 0) | |
515 | die("unable to read index file"); | |
516 | ||
517 | for (i = 1; i < argc; i++) { | |
518 | const char *arg = argv[i]; | |
519 | int fd; | |
520 | ||
521 | if (!strcmp(arg, "-")) { | |
522 | apply_patch(0); | |
4dfdbe10 | 523 | read_stdin = 0; |
c1bb9350 LT |
524 | continue; |
525 | } | |
526 | if (!strcmp(arg, "--no-merge")) { | |
527 | merge_patch = 0; | |
528 | continue; | |
529 | } | |
530 | fd = open(arg, O_RDONLY); | |
531 | if (fd < 0) | |
532 | usage(apply_usage); | |
4dfdbe10 | 533 | read_stdin = 0; |
c1bb9350 LT |
534 | apply_patch(fd); |
535 | close(fd); | |
536 | } | |
4dfdbe10 LT |
537 | if (read_stdin) |
538 | apply_patch(0); | |
c1bb9350 LT |
539 | return 0; |
540 | } |