Code

Merge branch 'jn/gitweb-unspecified-action' into maint-1.7.8
[git.git] / fast-import.c
1 /*
2 (See Documentation/git-fast-import.txt for maintained documentation.)
3 Format of STDIN stream:
5   stream ::= cmd*;
7   cmd ::= new_blob
8         | new_commit
9         | new_tag
10         | reset_branch
11         | checkpoint
12         | progress
13         ;
15   new_blob ::= 'blob' lf
16     mark?
17     file_content;
18   file_content ::= data;
20   new_commit ::= 'commit' sp ref_str lf
21     mark?
22     ('author' (sp name)? sp '<' email '>' sp when lf)?
23     'committer' (sp name)? sp '<' email '>' sp when lf
24     commit_msg
25     ('from' sp committish lf)?
26     ('merge' sp committish lf)*
27     (file_change | ls)*
28     lf?;
29   commit_msg ::= data;
31   ls ::= 'ls' sp '"' quoted(path) '"' lf;
33   file_change ::= file_clr
34     | file_del
35     | file_rnm
36     | file_cpy
37     | file_obm
38     | file_inm;
39   file_clr ::= 'deleteall' lf;
40   file_del ::= 'D' sp path_str lf;
41   file_rnm ::= 'R' sp path_str sp path_str lf;
42   file_cpy ::= 'C' sp path_str sp path_str lf;
43   file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
44   file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
45     data;
46   note_obm ::= 'N' sp (hexsha1 | idnum) sp committish lf;
47   note_inm ::= 'N' sp 'inline' sp committish lf
48     data;
50   new_tag ::= 'tag' sp tag_str lf
51     'from' sp committish lf
52     ('tagger' (sp name)? sp '<' email '>' sp when lf)?
53     tag_msg;
54   tag_msg ::= data;
56   reset_branch ::= 'reset' sp ref_str lf
57     ('from' sp committish lf)?
58     lf?;
60   checkpoint ::= 'checkpoint' lf
61     lf?;
63   progress ::= 'progress' sp not_lf* lf
64     lf?;
66      # note: the first idnum in a stream should be 1 and subsequent
67      # idnums should not have gaps between values as this will cause
68      # the stream parser to reserve space for the gapped values.  An
69      # idnum can be updated in the future to a new object by issuing
70      # a new mark directive with the old idnum.
71      #
72   mark ::= 'mark' sp idnum lf;
73   data ::= (delimited_data | exact_data)
74     lf?;
76     # note: delim may be any string but must not contain lf.
77     # data_line may contain any data but must not be exactly
78     # delim.
79   delimited_data ::= 'data' sp '<<' delim lf
80     (data_line lf)*
81     delim lf;
83      # note: declen indicates the length of binary_data in bytes.
84      # declen does not include the lf preceding the binary data.
85      #
86   exact_data ::= 'data' sp declen lf
87     binary_data;
89      # note: quoted strings are C-style quoting supporting \c for
90      # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
91      # is the signed byte value in octal.  Note that the only
92      # characters which must actually be escaped to protect the
93      # stream formatting is: \, " and LF.  Otherwise these values
94      # are UTF8.
95      #
96   committish  ::= (ref_str | hexsha1 | sha1exp_str | idnum);
97   ref_str     ::= ref;
98   sha1exp_str ::= sha1exp;
99   tag_str     ::= tag;
100   path_str    ::= path    | '"' quoted(path)    '"' ;
101   mode        ::= '100644' | '644'
102                 | '100755' | '755'
103                 | '120000'
104                 ;
106   declen ::= # unsigned 32 bit value, ascii base10 notation;
107   bigint ::= # unsigned integer value, ascii base10 notation;
108   binary_data ::= # file content, not interpreted;
110   when         ::= raw_when | rfc2822_when;
111   raw_when     ::= ts sp tz;
112   rfc2822_when ::= # Valid RFC 2822 date and time;
114   sp ::= # ASCII space character;
115   lf ::= # ASCII newline (LF) character;
117      # note: a colon (':') must precede the numerical value assigned to
118      # an idnum.  This is to distinguish it from a ref or tag name as
119      # GIT does not permit ':' in ref or tag strings.
120      #
121   idnum   ::= ':' bigint;
122   path    ::= # GIT style file path, e.g. "a/b/c";
123   ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
124   tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
125   sha1exp ::= # Any valid GIT SHA1 expression;
126   hexsha1 ::= # SHA1 in hexadecimal format;
128      # note: name and email are UTF8 strings, however name must not
129      # contain '<' or lf and email must not contain any of the
130      # following: '<', '>', lf.
131      #
132   name  ::= # valid GIT author/committer name;
133   email ::= # valid GIT author/committer email;
134   ts    ::= # time since the epoch in seconds, ascii base10 notation;
135   tz    ::= # GIT style timezone;
137      # note: comments, ls and cat requests may appear anywhere
138      # in the input, except within a data command.  Any form
139      # of the data command always escapes the related input
140      # from comment processing.
141      #
142      # In case it is not clear, the '#' that starts the comment
143      # must be the first character on that line (an lf
144      # preceded it).
145      #
147   cat_blob ::= 'cat-blob' sp (hexsha1 | idnum) lf;
148   ls_tree  ::= 'ls' sp (hexsha1 | idnum) sp path_str lf;
150   comment ::= '#' not_lf* lf;
151   not_lf  ::= # Any byte that is not ASCII newline (LF);
152 */
154 #include "builtin.h"
155 #include "cache.h"
156 #include "object.h"
157 #include "blob.h"
158 #include "tree.h"
159 #include "commit.h"
160 #include "delta.h"
161 #include "pack.h"
162 #include "refs.h"
163 #include "csum-file.h"
164 #include "quote.h"
165 #include "exec_cmd.h"
166 #include "dir.h"
168 #define PACK_ID_BITS 16
169 #define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
170 #define DEPTH_BITS 13
171 #define MAX_DEPTH ((1<<DEPTH_BITS)-1)
173 /*
174  * We abuse the setuid bit on directories to mean "do not delta".
175  */
176 #define NO_DELTA S_ISUID
178 struct object_entry {
179         struct pack_idx_entry idx;
180         struct object_entry *next;
181         uint32_t type : TYPE_BITS,
182                 pack_id : PACK_ID_BITS,
183                 depth : DEPTH_BITS;
184 };
186 struct object_entry_pool {
187         struct object_entry_pool *next_pool;
188         struct object_entry *next_free;
189         struct object_entry *end;
190         struct object_entry entries[FLEX_ARRAY]; /* more */
191 };
193 struct mark_set {
194         union {
195                 struct object_entry *marked[1024];
196                 struct mark_set *sets[1024];
197         } data;
198         unsigned int shift;
199 };
201 struct last_object {
202         struct strbuf data;
203         off_t offset;
204         unsigned int depth;
205         unsigned no_swap : 1;
206 };
208 struct mem_pool {
209         struct mem_pool *next_pool;
210         char *next_free;
211         char *end;
212         uintmax_t space[FLEX_ARRAY]; /* more */
213 };
215 struct atom_str {
216         struct atom_str *next_atom;
217         unsigned short str_len;
218         char str_dat[FLEX_ARRAY]; /* more */
219 };
221 struct tree_content;
222 struct tree_entry {
223         struct tree_content *tree;
224         struct atom_str *name;
225         struct tree_entry_ms {
226                 uint16_t mode;
227                 unsigned char sha1[20];
228         } versions[2];
229 };
231 struct tree_content {
232         unsigned int entry_capacity; /* must match avail_tree_content */
233         unsigned int entry_count;
234         unsigned int delta_depth;
235         struct tree_entry *entries[FLEX_ARRAY]; /* more */
236 };
238 struct avail_tree_content {
239         unsigned int entry_capacity; /* must match tree_content */
240         struct avail_tree_content *next_avail;
241 };
243 struct branch {
244         struct branch *table_next_branch;
245         struct branch *active_next_branch;
246         const char *name;
247         struct tree_entry branch_tree;
248         uintmax_t last_commit;
249         uintmax_t num_notes;
250         unsigned active : 1;
251         unsigned pack_id : PACK_ID_BITS;
252         unsigned char sha1[20];
253 };
255 struct tag {
256         struct tag *next_tag;
257         const char *name;
258         unsigned int pack_id;
259         unsigned char sha1[20];
260 };
262 struct hash_list {
263         struct hash_list *next;
264         unsigned char sha1[20];
265 };
267 typedef enum {
268         WHENSPEC_RAW = 1,
269         WHENSPEC_RFC2822,
270         WHENSPEC_NOW
271 } whenspec_type;
273 struct recent_command {
274         struct recent_command *prev;
275         struct recent_command *next;
276         char *buf;
277 };
279 /* Configured limits on output */
280 static unsigned long max_depth = 10;
281 static off_t max_packsize;
282 static int force_update;
283 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
284 static int pack_compression_seen;
286 /* Stats and misc. counters */
287 static uintmax_t alloc_count;
288 static uintmax_t marks_set_count;
289 static uintmax_t object_count_by_type[1 << TYPE_BITS];
290 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
291 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
292 static uintmax_t delta_count_attempts_by_type[1 << TYPE_BITS];
293 static unsigned long object_count;
294 static unsigned long branch_count;
295 static unsigned long branch_load_count;
296 static int failure;
297 static FILE *pack_edges;
298 static unsigned int show_stats = 1;
299 static int global_argc;
300 static const char **global_argv;
302 /* Memory pools */
303 static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
304 static size_t total_allocd;
305 static struct mem_pool *mem_pool;
307 /* Atom management */
308 static unsigned int atom_table_sz = 4451;
309 static unsigned int atom_cnt;
310 static struct atom_str **atom_table;
312 /* The .pack file being generated */
313 static struct pack_idx_option pack_idx_opts;
314 static unsigned int pack_id;
315 static struct sha1file *pack_file;
316 static struct packed_git *pack_data;
317 static struct packed_git **all_packs;
318 static off_t pack_size;
320 /* Table of objects we've written. */
321 static unsigned int object_entry_alloc = 5000;
322 static struct object_entry_pool *blocks;
323 static struct object_entry *object_table[1 << 16];
324 static struct mark_set *marks;
325 static const char *export_marks_file;
326 static const char *import_marks_file;
327 static int import_marks_file_from_stream;
328 static int import_marks_file_ignore_missing;
329 static int relative_marks_paths;
331 /* Our last blob */
332 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
334 /* Tree management */
335 static unsigned int tree_entry_alloc = 1000;
336 static void *avail_tree_entry;
337 static unsigned int avail_tree_table_sz = 100;
338 static struct avail_tree_content **avail_tree_table;
339 static struct strbuf old_tree = STRBUF_INIT;
340 static struct strbuf new_tree = STRBUF_INIT;
342 /* Branch data */
343 static unsigned long max_active_branches = 5;
344 static unsigned long cur_active_branches;
345 static unsigned long branch_table_sz = 1039;
346 static struct branch **branch_table;
347 static struct branch *active_branches;
349 /* Tag data */
350 static struct tag *first_tag;
351 static struct tag *last_tag;
353 /* Input stream parsing */
354 static whenspec_type whenspec = WHENSPEC_RAW;
355 static struct strbuf command_buf = STRBUF_INIT;
356 static int unread_command_buf;
357 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
358 static struct recent_command *cmd_tail = &cmd_hist;
359 static struct recent_command *rc_free;
360 static unsigned int cmd_save = 100;
361 static uintmax_t next_mark;
362 static struct strbuf new_data = STRBUF_INIT;
363 static int seen_data_command;
364 static int require_explicit_termination;
366 /* Signal handling */
367 static volatile sig_atomic_t checkpoint_requested;
369 /* Where to write output of cat-blob commands */
370 static int cat_blob_fd = STDOUT_FILENO;
372 static void parse_argv(void);
373 static void parse_cat_blob(void);
374 static void parse_ls(struct branch *b);
376 static void write_branch_report(FILE *rpt, struct branch *b)
378         fprintf(rpt, "%s:\n", b->name);
380         fprintf(rpt, "  status      :");
381         if (b->active)
382                 fputs(" active", rpt);
383         if (b->branch_tree.tree)
384                 fputs(" loaded", rpt);
385         if (is_null_sha1(b->branch_tree.versions[1].sha1))
386                 fputs(" dirty", rpt);
387         fputc('\n', rpt);
389         fprintf(rpt, "  tip commit  : %s\n", sha1_to_hex(b->sha1));
390         fprintf(rpt, "  old tree    : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
391         fprintf(rpt, "  cur tree    : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
392         fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
394         fputs("  last pack   : ", rpt);
395         if (b->pack_id < MAX_PACK_ID)
396                 fprintf(rpt, "%u", b->pack_id);
397         fputc('\n', rpt);
399         fputc('\n', rpt);
402 static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
404 static void write_crash_report(const char *err)
406         char *loc = git_path("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
407         FILE *rpt = fopen(loc, "w");
408         struct branch *b;
409         unsigned long lu;
410         struct recent_command *rc;
412         if (!rpt) {
413                 error("can't write crash report %s: %s", loc, strerror(errno));
414                 return;
415         }
417         fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
419         fprintf(rpt, "fast-import crash report:\n");
420         fprintf(rpt, "    fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
421         fprintf(rpt, "    parent process     : %"PRIuMAX"\n", (uintmax_t) getppid());
422         fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
423         fputc('\n', rpt);
425         fputs("fatal: ", rpt);
426         fputs(err, rpt);
427         fputc('\n', rpt);
429         fputc('\n', rpt);
430         fputs("Most Recent Commands Before Crash\n", rpt);
431         fputs("---------------------------------\n", rpt);
432         for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
433                 if (rc->next == &cmd_hist)
434                         fputs("* ", rpt);
435                 else
436                         fputs("  ", rpt);
437                 fputs(rc->buf, rpt);
438                 fputc('\n', rpt);
439         }
441         fputc('\n', rpt);
442         fputs("Active Branch LRU\n", rpt);
443         fputs("-----------------\n", rpt);
444         fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
445                 cur_active_branches,
446                 max_active_branches);
447         fputc('\n', rpt);
448         fputs("  pos  clock name\n", rpt);
449         fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
450         for (b = active_branches, lu = 0; b; b = b->active_next_branch)
451                 fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
452                         ++lu, b->last_commit, b->name);
454         fputc('\n', rpt);
455         fputs("Inactive Branches\n", rpt);
456         fputs("-----------------\n", rpt);
457         for (lu = 0; lu < branch_table_sz; lu++) {
458                 for (b = branch_table[lu]; b; b = b->table_next_branch)
459                         write_branch_report(rpt, b);
460         }
462         if (first_tag) {
463                 struct tag *tg;
464                 fputc('\n', rpt);
465                 fputs("Annotated Tags\n", rpt);
466                 fputs("--------------\n", rpt);
467                 for (tg = first_tag; tg; tg = tg->next_tag) {
468                         fputs(sha1_to_hex(tg->sha1), rpt);
469                         fputc(' ', rpt);
470                         fputs(tg->name, rpt);
471                         fputc('\n', rpt);
472                 }
473         }
475         fputc('\n', rpt);
476         fputs("Marks\n", rpt);
477         fputs("-----\n", rpt);
478         if (export_marks_file)
479                 fprintf(rpt, "  exported to %s\n", export_marks_file);
480         else
481                 dump_marks_helper(rpt, 0, marks);
483         fputc('\n', rpt);
484         fputs("-------------------\n", rpt);
485         fputs("END OF CRASH REPORT\n", rpt);
486         fclose(rpt);
489 static void end_packfile(void);
490 static void unkeep_all_packs(void);
491 static void dump_marks(void);
493 static NORETURN void die_nicely(const char *err, va_list params)
495         static int zombie;
496         char message[2 * PATH_MAX];
498         vsnprintf(message, sizeof(message), err, params);
499         fputs("fatal: ", stderr);
500         fputs(message, stderr);
501         fputc('\n', stderr);
503         if (!zombie) {
504                 zombie = 1;
505                 write_crash_report(message);
506                 end_packfile();
507                 unkeep_all_packs();
508                 dump_marks();
509         }
510         exit(128);
513 #ifndef SIGUSR1 /* Windows, for example */
515 static void set_checkpoint_signal(void)
519 #else
521 static void checkpoint_signal(int signo)
523         checkpoint_requested = 1;
526 static void set_checkpoint_signal(void)
528         struct sigaction sa;
530         memset(&sa, 0, sizeof(sa));
531         sa.sa_handler = checkpoint_signal;
532         sigemptyset(&sa.sa_mask);
533         sa.sa_flags = SA_RESTART;
534         sigaction(SIGUSR1, &sa, NULL);
537 #endif
539 static void alloc_objects(unsigned int cnt)
541         struct object_entry_pool *b;
543         b = xmalloc(sizeof(struct object_entry_pool)
544                 + cnt * sizeof(struct object_entry));
545         b->next_pool = blocks;
546         b->next_free = b->entries;
547         b->end = b->entries + cnt;
548         blocks = b;
549         alloc_count += cnt;
552 static struct object_entry *new_object(unsigned char *sha1)
554         struct object_entry *e;
556         if (blocks->next_free == blocks->end)
557                 alloc_objects(object_entry_alloc);
559         e = blocks->next_free++;
560         hashcpy(e->idx.sha1, sha1);
561         return e;
564 static struct object_entry *find_object(unsigned char *sha1)
566         unsigned int h = sha1[0] << 8 | sha1[1];
567         struct object_entry *e;
568         for (e = object_table[h]; e; e = e->next)
569                 if (!hashcmp(sha1, e->idx.sha1))
570                         return e;
571         return NULL;
574 static struct object_entry *insert_object(unsigned char *sha1)
576         unsigned int h = sha1[0] << 8 | sha1[1];
577         struct object_entry *e = object_table[h];
579         while (e) {
580                 if (!hashcmp(sha1, e->idx.sha1))
581                         return e;
582                 e = e->next;
583         }
585         e = new_object(sha1);
586         e->next = object_table[h];
587         e->idx.offset = 0;
588         object_table[h] = e;
589         return e;
592 static unsigned int hc_str(const char *s, size_t len)
594         unsigned int r = 0;
595         while (len-- > 0)
596                 r = r * 31 + *s++;
597         return r;
600 static void *pool_alloc(size_t len)
602         struct mem_pool *p;
603         void *r;
605         /* round up to a 'uintmax_t' alignment */
606         if (len & (sizeof(uintmax_t) - 1))
607                 len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
609         for (p = mem_pool; p; p = p->next_pool)
610                 if ((p->end - p->next_free >= len))
611                         break;
613         if (!p) {
614                 if (len >= (mem_pool_alloc/2)) {
615                         total_allocd += len;
616                         return xmalloc(len);
617                 }
618                 total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
619                 p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
620                 p->next_pool = mem_pool;
621                 p->next_free = (char *) p->space;
622                 p->end = p->next_free + mem_pool_alloc;
623                 mem_pool = p;
624         }
626         r = p->next_free;
627         p->next_free += len;
628         return r;
631 static void *pool_calloc(size_t count, size_t size)
633         size_t len = count * size;
634         void *r = pool_alloc(len);
635         memset(r, 0, len);
636         return r;
639 static char *pool_strdup(const char *s)
641         char *r = pool_alloc(strlen(s) + 1);
642         strcpy(r, s);
643         return r;
646 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
648         struct mark_set *s = marks;
649         while ((idnum >> s->shift) >= 1024) {
650                 s = pool_calloc(1, sizeof(struct mark_set));
651                 s->shift = marks->shift + 10;
652                 s->data.sets[0] = marks;
653                 marks = s;
654         }
655         while (s->shift) {
656                 uintmax_t i = idnum >> s->shift;
657                 idnum -= i << s->shift;
658                 if (!s->data.sets[i]) {
659                         s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
660                         s->data.sets[i]->shift = s->shift - 10;
661                 }
662                 s = s->data.sets[i];
663         }
664         if (!s->data.marked[idnum])
665                 marks_set_count++;
666         s->data.marked[idnum] = oe;
669 static struct object_entry *find_mark(uintmax_t idnum)
671         uintmax_t orig_idnum = idnum;
672         struct mark_set *s = marks;
673         struct object_entry *oe = NULL;
674         if ((idnum >> s->shift) < 1024) {
675                 while (s && s->shift) {
676                         uintmax_t i = idnum >> s->shift;
677                         idnum -= i << s->shift;
678                         s = s->data.sets[i];
679                 }
680                 if (s)
681                         oe = s->data.marked[idnum];
682         }
683         if (!oe)
684                 die("mark :%" PRIuMAX " not declared", orig_idnum);
685         return oe;
688 static struct atom_str *to_atom(const char *s, unsigned short len)
690         unsigned int hc = hc_str(s, len) % atom_table_sz;
691         struct atom_str *c;
693         for (c = atom_table[hc]; c; c = c->next_atom)
694                 if (c->str_len == len && !strncmp(s, c->str_dat, len))
695                         return c;
697         c = pool_alloc(sizeof(struct atom_str) + len + 1);
698         c->str_len = len;
699         strncpy(c->str_dat, s, len);
700         c->str_dat[len] = 0;
701         c->next_atom = atom_table[hc];
702         atom_table[hc] = c;
703         atom_cnt++;
704         return c;
707 static struct branch *lookup_branch(const char *name)
709         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
710         struct branch *b;
712         for (b = branch_table[hc]; b; b = b->table_next_branch)
713                 if (!strcmp(name, b->name))
714                         return b;
715         return NULL;
718 static struct branch *new_branch(const char *name)
720         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
721         struct branch *b = lookup_branch(name);
723         if (b)
724                 die("Invalid attempt to create duplicate branch: %s", name);
725         if (check_refname_format(name, REFNAME_ALLOW_ONELEVEL))
726                 die("Branch name doesn't conform to GIT standards: %s", name);
728         b = pool_calloc(1, sizeof(struct branch));
729         b->name = pool_strdup(name);
730         b->table_next_branch = branch_table[hc];
731         b->branch_tree.versions[0].mode = S_IFDIR;
732         b->branch_tree.versions[1].mode = S_IFDIR;
733         b->num_notes = 0;
734         b->active = 0;
735         b->pack_id = MAX_PACK_ID;
736         branch_table[hc] = b;
737         branch_count++;
738         return b;
741 static unsigned int hc_entries(unsigned int cnt)
743         cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
744         return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
747 static struct tree_content *new_tree_content(unsigned int cnt)
749         struct avail_tree_content *f, *l = NULL;
750         struct tree_content *t;
751         unsigned int hc = hc_entries(cnt);
753         for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
754                 if (f->entry_capacity >= cnt)
755                         break;
757         if (f) {
758                 if (l)
759                         l->next_avail = f->next_avail;
760                 else
761                         avail_tree_table[hc] = f->next_avail;
762         } else {
763                 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
764                 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
765                 f->entry_capacity = cnt;
766         }
768         t = (struct tree_content*)f;
769         t->entry_count = 0;
770         t->delta_depth = 0;
771         return t;
774 static void release_tree_entry(struct tree_entry *e);
775 static void release_tree_content(struct tree_content *t)
777         struct avail_tree_content *f = (struct avail_tree_content*)t;
778         unsigned int hc = hc_entries(f->entry_capacity);
779         f->next_avail = avail_tree_table[hc];
780         avail_tree_table[hc] = f;
783 static void release_tree_content_recursive(struct tree_content *t)
785         unsigned int i;
786         for (i = 0; i < t->entry_count; i++)
787                 release_tree_entry(t->entries[i]);
788         release_tree_content(t);
791 static struct tree_content *grow_tree_content(
792         struct tree_content *t,
793         int amt)
795         struct tree_content *r = new_tree_content(t->entry_count + amt);
796         r->entry_count = t->entry_count;
797         r->delta_depth = t->delta_depth;
798         memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
799         release_tree_content(t);
800         return r;
803 static struct tree_entry *new_tree_entry(void)
805         struct tree_entry *e;
807         if (!avail_tree_entry) {
808                 unsigned int n = tree_entry_alloc;
809                 total_allocd += n * sizeof(struct tree_entry);
810                 avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
811                 while (n-- > 1) {
812                         *((void**)e) = e + 1;
813                         e++;
814                 }
815                 *((void**)e) = NULL;
816         }
818         e = avail_tree_entry;
819         avail_tree_entry = *((void**)e);
820         return e;
823 static void release_tree_entry(struct tree_entry *e)
825         if (e->tree)
826                 release_tree_content_recursive(e->tree);
827         *((void**)e) = avail_tree_entry;
828         avail_tree_entry = e;
831 static struct tree_content *dup_tree_content(struct tree_content *s)
833         struct tree_content *d;
834         struct tree_entry *a, *b;
835         unsigned int i;
837         if (!s)
838                 return NULL;
839         d = new_tree_content(s->entry_count);
840         for (i = 0; i < s->entry_count; i++) {
841                 a = s->entries[i];
842                 b = new_tree_entry();
843                 memcpy(b, a, sizeof(*a));
844                 if (a->tree && is_null_sha1(b->versions[1].sha1))
845                         b->tree = dup_tree_content(a->tree);
846                 else
847                         b->tree = NULL;
848                 d->entries[i] = b;
849         }
850         d->entry_count = s->entry_count;
851         d->delta_depth = s->delta_depth;
853         return d;
856 static void start_packfile(void)
858         static char tmpfile[PATH_MAX];
859         struct packed_git *p;
860         struct pack_header hdr;
861         int pack_fd;
863         pack_fd = odb_mkstemp(tmpfile, sizeof(tmpfile),
864                               "pack/tmp_pack_XXXXXX");
865         p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
866         strcpy(p->pack_name, tmpfile);
867         p->pack_fd = pack_fd;
868         p->do_not_close = 1;
869         pack_file = sha1fd(pack_fd, p->pack_name);
871         hdr.hdr_signature = htonl(PACK_SIGNATURE);
872         hdr.hdr_version = htonl(2);
873         hdr.hdr_entries = 0;
874         sha1write(pack_file, &hdr, sizeof(hdr));
876         pack_data = p;
877         pack_size = sizeof(hdr);
878         object_count = 0;
880         all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
881         all_packs[pack_id] = p;
884 static const char *create_index(void)
886         const char *tmpfile;
887         struct pack_idx_entry **idx, **c, **last;
888         struct object_entry *e;
889         struct object_entry_pool *o;
891         /* Build the table of object IDs. */
892         idx = xmalloc(object_count * sizeof(*idx));
893         c = idx;
894         for (o = blocks; o; o = o->next_pool)
895                 for (e = o->next_free; e-- != o->entries;)
896                         if (pack_id == e->pack_id)
897                                 *c++ = &e->idx;
898         last = idx + object_count;
899         if (c != last)
900                 die("internal consistency error creating the index");
902         tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts, pack_data->sha1);
903         free(idx);
904         return tmpfile;
907 static char *keep_pack(const char *curr_index_name)
909         static char name[PATH_MAX];
910         static const char *keep_msg = "fast-import";
911         int keep_fd;
913         keep_fd = odb_pack_keep(name, sizeof(name), pack_data->sha1);
914         if (keep_fd < 0)
915                 die_errno("cannot create keep file");
916         write_or_die(keep_fd, keep_msg, strlen(keep_msg));
917         if (close(keep_fd))
918                 die_errno("failed to write keep file");
920         snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
921                  get_object_directory(), sha1_to_hex(pack_data->sha1));
922         if (move_temp_to_file(pack_data->pack_name, name))
923                 die("cannot store pack file");
925         snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
926                  get_object_directory(), sha1_to_hex(pack_data->sha1));
927         if (move_temp_to_file(curr_index_name, name))
928                 die("cannot store index file");
929         free((void *)curr_index_name);
930         return name;
933 static void unkeep_all_packs(void)
935         static char name[PATH_MAX];
936         int k;
938         for (k = 0; k < pack_id; k++) {
939                 struct packed_git *p = all_packs[k];
940                 snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
941                          get_object_directory(), sha1_to_hex(p->sha1));
942                 unlink_or_warn(name);
943         }
946 static void end_packfile(void)
948         struct packed_git *old_p = pack_data, *new_p;
950         clear_delta_base_cache();
951         if (object_count) {
952                 unsigned char cur_pack_sha1[20];
953                 char *idx_name;
954                 int i;
955                 struct branch *b;
956                 struct tag *t;
958                 close_pack_windows(pack_data);
959                 sha1close(pack_file, cur_pack_sha1, 0);
960                 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
961                                     pack_data->pack_name, object_count,
962                                     cur_pack_sha1, pack_size);
963                 close(pack_data->pack_fd);
964                 idx_name = keep_pack(create_index());
966                 /* Register the packfile with core git's machinery. */
967                 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
968                 if (!new_p)
969                         die("core git rejected index %s", idx_name);
970                 all_packs[pack_id] = new_p;
971                 install_packed_git(new_p);
973                 /* Print the boundary */
974                 if (pack_edges) {
975                         fprintf(pack_edges, "%s:", new_p->pack_name);
976                         for (i = 0; i < branch_table_sz; i++) {
977                                 for (b = branch_table[i]; b; b = b->table_next_branch) {
978                                         if (b->pack_id == pack_id)
979                                                 fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
980                                 }
981                         }
982                         for (t = first_tag; t; t = t->next_tag) {
983                                 if (t->pack_id == pack_id)
984                                         fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
985                         }
986                         fputc('\n', pack_edges);
987                         fflush(pack_edges);
988                 }
990                 pack_id++;
991         }
992         else {
993                 close(old_p->pack_fd);
994                 unlink_or_warn(old_p->pack_name);
995         }
996         free(old_p);
998         /* We can't carry a delta across packfiles. */
999         strbuf_release(&last_blob.data);
1000         last_blob.offset = 0;
1001         last_blob.depth = 0;
1004 static void cycle_packfile(void)
1006         end_packfile();
1007         start_packfile();
1010 static int store_object(
1011         enum object_type type,
1012         struct strbuf *dat,
1013         struct last_object *last,
1014         unsigned char *sha1out,
1015         uintmax_t mark)
1017         void *out, *delta;
1018         struct object_entry *e;
1019         unsigned char hdr[96];
1020         unsigned char sha1[20];
1021         unsigned long hdrlen, deltalen;
1022         git_SHA_CTX c;
1023         git_zstream s;
1025         hdrlen = sprintf((char *)hdr,"%s %lu", typename(type),
1026                 (unsigned long)dat->len) + 1;
1027         git_SHA1_Init(&c);
1028         git_SHA1_Update(&c, hdr, hdrlen);
1029         git_SHA1_Update(&c, dat->buf, dat->len);
1030         git_SHA1_Final(sha1, &c);
1031         if (sha1out)
1032                 hashcpy(sha1out, sha1);
1034         e = insert_object(sha1);
1035         if (mark)
1036                 insert_mark(mark, e);
1037         if (e->idx.offset) {
1038                 duplicate_count_by_type[type]++;
1039                 return 1;
1040         } else if (find_sha1_pack(sha1, packed_git)) {
1041                 e->type = type;
1042                 e->pack_id = MAX_PACK_ID;
1043                 e->idx.offset = 1; /* just not zero! */
1044                 duplicate_count_by_type[type]++;
1045                 return 1;
1046         }
1048         if (last && last->data.buf && last->depth < max_depth && dat->len > 20) {
1049                 delta_count_attempts_by_type[type]++;
1050                 delta = diff_delta(last->data.buf, last->data.len,
1051                         dat->buf, dat->len,
1052                         &deltalen, dat->len - 20);
1053         } else
1054                 delta = NULL;
1056         memset(&s, 0, sizeof(s));
1057         git_deflate_init(&s, pack_compression_level);
1058         if (delta) {
1059                 s.next_in = delta;
1060                 s.avail_in = deltalen;
1061         } else {
1062                 s.next_in = (void *)dat->buf;
1063                 s.avail_in = dat->len;
1064         }
1065         s.avail_out = git_deflate_bound(&s, s.avail_in);
1066         s.next_out = out = xmalloc(s.avail_out);
1067         while (git_deflate(&s, Z_FINISH) == Z_OK)
1068                 ; /* nothing */
1069         git_deflate_end(&s);
1071         /* Determine if we should auto-checkpoint. */
1072         if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1073                 || (pack_size + 60 + s.total_out) < pack_size) {
1075                 /* This new object needs to *not* have the current pack_id. */
1076                 e->pack_id = pack_id + 1;
1077                 cycle_packfile();
1079                 /* We cannot carry a delta into the new pack. */
1080                 if (delta) {
1081                         free(delta);
1082                         delta = NULL;
1084                         memset(&s, 0, sizeof(s));
1085                         git_deflate_init(&s, pack_compression_level);
1086                         s.next_in = (void *)dat->buf;
1087                         s.avail_in = dat->len;
1088                         s.avail_out = git_deflate_bound(&s, s.avail_in);
1089                         s.next_out = out = xrealloc(out, s.avail_out);
1090                         while (git_deflate(&s, Z_FINISH) == Z_OK)
1091                                 ; /* nothing */
1092                         git_deflate_end(&s);
1093                 }
1094         }
1096         e->type = type;
1097         e->pack_id = pack_id;
1098         e->idx.offset = pack_size;
1099         object_count++;
1100         object_count_by_type[type]++;
1102         crc32_begin(pack_file);
1104         if (delta) {
1105                 off_t ofs = e->idx.offset - last->offset;
1106                 unsigned pos = sizeof(hdr) - 1;
1108                 delta_count_by_type[type]++;
1109                 e->depth = last->depth + 1;
1111                 hdrlen = encode_in_pack_object_header(OBJ_OFS_DELTA, deltalen, hdr);
1112                 sha1write(pack_file, hdr, hdrlen);
1113                 pack_size += hdrlen;
1115                 hdr[pos] = ofs & 127;
1116                 while (ofs >>= 7)
1117                         hdr[--pos] = 128 | (--ofs & 127);
1118                 sha1write(pack_file, hdr + pos, sizeof(hdr) - pos);
1119                 pack_size += sizeof(hdr) - pos;
1120         } else {
1121                 e->depth = 0;
1122                 hdrlen = encode_in_pack_object_header(type, dat->len, hdr);
1123                 sha1write(pack_file, hdr, hdrlen);
1124                 pack_size += hdrlen;
1125         }
1127         sha1write(pack_file, out, s.total_out);
1128         pack_size += s.total_out;
1130         e->idx.crc32 = crc32_end(pack_file);
1132         free(out);
1133         free(delta);
1134         if (last) {
1135                 if (last->no_swap) {
1136                         last->data = *dat;
1137                 } else {
1138                         strbuf_swap(&last->data, dat);
1139                 }
1140                 last->offset = e->idx.offset;
1141                 last->depth = e->depth;
1142         }
1143         return 0;
1146 static void truncate_pack(off_t to, git_SHA_CTX *ctx)
1148         if (ftruncate(pack_data->pack_fd, to)
1149          || lseek(pack_data->pack_fd, to, SEEK_SET) != to)
1150                 die_errno("cannot truncate pack to skip duplicate");
1151         pack_size = to;
1153         /* yes this is a layering violation */
1154         pack_file->total = to;
1155         pack_file->offset = 0;
1156         pack_file->ctx = *ctx;
1159 static void stream_blob(uintmax_t len, unsigned char *sha1out, uintmax_t mark)
1161         size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1162         unsigned char *in_buf = xmalloc(in_sz);
1163         unsigned char *out_buf = xmalloc(out_sz);
1164         struct object_entry *e;
1165         unsigned char sha1[20];
1166         unsigned long hdrlen;
1167         off_t offset;
1168         git_SHA_CTX c;
1169         git_SHA_CTX pack_file_ctx;
1170         git_zstream s;
1171         int status = Z_OK;
1173         /* Determine if we should auto-checkpoint. */
1174         if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1175                 || (pack_size + 60 + len) < pack_size)
1176                 cycle_packfile();
1178         offset = pack_size;
1180         /* preserve the pack_file SHA1 ctx in case we have to truncate later */
1181         sha1flush(pack_file);
1182         pack_file_ctx = pack_file->ctx;
1184         hdrlen = snprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1185         if (out_sz <= hdrlen)
1186                 die("impossibly large object header");
1188         git_SHA1_Init(&c);
1189         git_SHA1_Update(&c, out_buf, hdrlen);
1191         crc32_begin(pack_file);
1193         memset(&s, 0, sizeof(s));
1194         git_deflate_init(&s, pack_compression_level);
1196         hdrlen = encode_in_pack_object_header(OBJ_BLOB, len, out_buf);
1197         if (out_sz <= hdrlen)
1198                 die("impossibly large object header");
1200         s.next_out = out_buf + hdrlen;
1201         s.avail_out = out_sz - hdrlen;
1203         while (status != Z_STREAM_END) {
1204                 if (0 < len && !s.avail_in) {
1205                         size_t cnt = in_sz < len ? in_sz : (size_t)len;
1206                         size_t n = fread(in_buf, 1, cnt, stdin);
1207                         if (!n && feof(stdin))
1208                                 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1210                         git_SHA1_Update(&c, in_buf, n);
1211                         s.next_in = in_buf;
1212                         s.avail_in = n;
1213                         len -= n;
1214                 }
1216                 status = git_deflate(&s, len ? 0 : Z_FINISH);
1218                 if (!s.avail_out || status == Z_STREAM_END) {
1219                         size_t n = s.next_out - out_buf;
1220                         sha1write(pack_file, out_buf, n);
1221                         pack_size += n;
1222                         s.next_out = out_buf;
1223                         s.avail_out = out_sz;
1224                 }
1226                 switch (status) {
1227                 case Z_OK:
1228                 case Z_BUF_ERROR:
1229                 case Z_STREAM_END:
1230                         continue;
1231                 default:
1232                         die("unexpected deflate failure: %d", status);
1233                 }
1234         }
1235         git_deflate_end(&s);
1236         git_SHA1_Final(sha1, &c);
1238         if (sha1out)
1239                 hashcpy(sha1out, sha1);
1241         e = insert_object(sha1);
1243         if (mark)
1244                 insert_mark(mark, e);
1246         if (e->idx.offset) {
1247                 duplicate_count_by_type[OBJ_BLOB]++;
1248                 truncate_pack(offset, &pack_file_ctx);
1250         } else if (find_sha1_pack(sha1, packed_git)) {
1251                 e->type = OBJ_BLOB;
1252                 e->pack_id = MAX_PACK_ID;
1253                 e->idx.offset = 1; /* just not zero! */
1254                 duplicate_count_by_type[OBJ_BLOB]++;
1255                 truncate_pack(offset, &pack_file_ctx);
1257         } else {
1258                 e->depth = 0;
1259                 e->type = OBJ_BLOB;
1260                 e->pack_id = pack_id;
1261                 e->idx.offset = offset;
1262                 e->idx.crc32 = crc32_end(pack_file);
1263                 object_count++;
1264                 object_count_by_type[OBJ_BLOB]++;
1265         }
1267         free(in_buf);
1268         free(out_buf);
1271 /* All calls must be guarded by find_object() or find_mark() to
1272  * ensure the 'struct object_entry' passed was written by this
1273  * process instance.  We unpack the entry by the offset, avoiding
1274  * the need for the corresponding .idx file.  This unpacking rule
1275  * works because we only use OBJ_REF_DELTA within the packfiles
1276  * created by fast-import.
1277  *
1278  * oe must not be NULL.  Such an oe usually comes from giving
1279  * an unknown SHA-1 to find_object() or an undefined mark to
1280  * find_mark().  Callers must test for this condition and use
1281  * the standard read_sha1_file() when it happens.
1282  *
1283  * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1284  * find_mark(), where the mark was reloaded from an existing marks
1285  * file and is referencing an object that this fast-import process
1286  * instance did not write out to a packfile.  Callers must test for
1287  * this condition and use read_sha1_file() instead.
1288  */
1289 static void *gfi_unpack_entry(
1290         struct object_entry *oe,
1291         unsigned long *sizep)
1293         enum object_type type;
1294         struct packed_git *p = all_packs[oe->pack_id];
1295         if (p == pack_data && p->pack_size < (pack_size + 20)) {
1296                 /* The object is stored in the packfile we are writing to
1297                  * and we have modified it since the last time we scanned
1298                  * back to read a previously written object.  If an old
1299                  * window covered [p->pack_size, p->pack_size + 20) its
1300                  * data is stale and is not valid.  Closing all windows
1301                  * and updating the packfile length ensures we can read
1302                  * the newly written data.
1303                  */
1304                 close_pack_windows(p);
1305                 sha1flush(pack_file);
1307                 /* We have to offer 20 bytes additional on the end of
1308                  * the packfile as the core unpacker code assumes the
1309                  * footer is present at the file end and must promise
1310                  * at least 20 bytes within any window it maps.  But
1311                  * we don't actually create the footer here.
1312                  */
1313                 p->pack_size = pack_size + 20;
1314         }
1315         return unpack_entry(p, oe->idx.offset, &type, sizep);
1318 static const char *get_mode(const char *str, uint16_t *modep)
1320         unsigned char c;
1321         uint16_t mode = 0;
1323         while ((c = *str++) != ' ') {
1324                 if (c < '0' || c > '7')
1325                         return NULL;
1326                 mode = (mode << 3) + (c - '0');
1327         }
1328         *modep = mode;
1329         return str;
1332 static void load_tree(struct tree_entry *root)
1334         unsigned char *sha1 = root->versions[1].sha1;
1335         struct object_entry *myoe;
1336         struct tree_content *t;
1337         unsigned long size;
1338         char *buf;
1339         const char *c;
1341         root->tree = t = new_tree_content(8);
1342         if (is_null_sha1(sha1))
1343                 return;
1345         myoe = find_object(sha1);
1346         if (myoe && myoe->pack_id != MAX_PACK_ID) {
1347                 if (myoe->type != OBJ_TREE)
1348                         die("Not a tree: %s", sha1_to_hex(sha1));
1349                 t->delta_depth = myoe->depth;
1350                 buf = gfi_unpack_entry(myoe, &size);
1351                 if (!buf)
1352                         die("Can't load tree %s", sha1_to_hex(sha1));
1353         } else {
1354                 enum object_type type;
1355                 buf = read_sha1_file(sha1, &type, &size);
1356                 if (!buf || type != OBJ_TREE)
1357                         die("Can't load tree %s", sha1_to_hex(sha1));
1358         }
1360         c = buf;
1361         while (c != (buf + size)) {
1362                 struct tree_entry *e = new_tree_entry();
1364                 if (t->entry_count == t->entry_capacity)
1365                         root->tree = t = grow_tree_content(t, t->entry_count);
1366                 t->entries[t->entry_count++] = e;
1368                 e->tree = NULL;
1369                 c = get_mode(c, &e->versions[1].mode);
1370                 if (!c)
1371                         die("Corrupt mode in %s", sha1_to_hex(sha1));
1372                 e->versions[0].mode = e->versions[1].mode;
1373                 e->name = to_atom(c, strlen(c));
1374                 c += e->name->str_len + 1;
1375                 hashcpy(e->versions[0].sha1, (unsigned char *)c);
1376                 hashcpy(e->versions[1].sha1, (unsigned char *)c);
1377                 c += 20;
1378         }
1379         free(buf);
1382 static int tecmp0 (const void *_a, const void *_b)
1384         struct tree_entry *a = *((struct tree_entry**)_a);
1385         struct tree_entry *b = *((struct tree_entry**)_b);
1386         return base_name_compare(
1387                 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1388                 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1391 static int tecmp1 (const void *_a, const void *_b)
1393         struct tree_entry *a = *((struct tree_entry**)_a);
1394         struct tree_entry *b = *((struct tree_entry**)_b);
1395         return base_name_compare(
1396                 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1397                 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1400 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1402         size_t maxlen = 0;
1403         unsigned int i;
1405         if (!v)
1406                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1407         else
1408                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1410         for (i = 0; i < t->entry_count; i++) {
1411                 if (t->entries[i]->versions[v].mode)
1412                         maxlen += t->entries[i]->name->str_len + 34;
1413         }
1415         strbuf_reset(b);
1416         strbuf_grow(b, maxlen);
1417         for (i = 0; i < t->entry_count; i++) {
1418                 struct tree_entry *e = t->entries[i];
1419                 if (!e->versions[v].mode)
1420                         continue;
1421                 strbuf_addf(b, "%o %s%c",
1422                         (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1423                         e->name->str_dat, '\0');
1424                 strbuf_add(b, e->versions[v].sha1, 20);
1425         }
1428 static void store_tree(struct tree_entry *root)
1430         struct tree_content *t = root->tree;
1431         unsigned int i, j, del;
1432         struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1433         struct object_entry *le = NULL;
1435         if (!is_null_sha1(root->versions[1].sha1))
1436                 return;
1438         for (i = 0; i < t->entry_count; i++) {
1439                 if (t->entries[i]->tree)
1440                         store_tree(t->entries[i]);
1441         }
1443         if (!(root->versions[0].mode & NO_DELTA))
1444                 le = find_object(root->versions[0].sha1);
1445         if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1446                 mktree(t, 0, &old_tree);
1447                 lo.data = old_tree;
1448                 lo.offset = le->idx.offset;
1449                 lo.depth = t->delta_depth;
1450         }
1452         mktree(t, 1, &new_tree);
1453         store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1455         t->delta_depth = lo.depth;
1456         for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1457                 struct tree_entry *e = t->entries[i];
1458                 if (e->versions[1].mode) {
1459                         e->versions[0].mode = e->versions[1].mode;
1460                         hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1461                         t->entries[j++] = e;
1462                 } else {
1463                         release_tree_entry(e);
1464                         del++;
1465                 }
1466         }
1467         t->entry_count -= del;
1470 static void tree_content_replace(
1471         struct tree_entry *root,
1472         const unsigned char *sha1,
1473         const uint16_t mode,
1474         struct tree_content *newtree)
1476         if (!S_ISDIR(mode))
1477                 die("Root cannot be a non-directory");
1478         hashclr(root->versions[0].sha1);
1479         hashcpy(root->versions[1].sha1, sha1);
1480         if (root->tree)
1481                 release_tree_content_recursive(root->tree);
1482         root->tree = newtree;
1485 static int tree_content_set(
1486         struct tree_entry *root,
1487         const char *p,
1488         const unsigned char *sha1,
1489         const uint16_t mode,
1490         struct tree_content *subtree)
1492         struct tree_content *t;
1493         const char *slash1;
1494         unsigned int i, n;
1495         struct tree_entry *e;
1497         slash1 = strchr(p, '/');
1498         if (slash1)
1499                 n = slash1 - p;
1500         else
1501                 n = strlen(p);
1502         if (!n)
1503                 die("Empty path component found in input");
1504         if (!slash1 && !S_ISDIR(mode) && subtree)
1505                 die("Non-directories cannot have subtrees");
1507         if (!root->tree)
1508                 load_tree(root);
1509         t = root->tree;
1510         for (i = 0; i < t->entry_count; i++) {
1511                 e = t->entries[i];
1512                 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1513                         if (!slash1) {
1514                                 if (!S_ISDIR(mode)
1515                                                 && e->versions[1].mode == mode
1516                                                 && !hashcmp(e->versions[1].sha1, sha1))
1517                                         return 0;
1518                                 e->versions[1].mode = mode;
1519                                 hashcpy(e->versions[1].sha1, sha1);
1520                                 if (e->tree)
1521                                         release_tree_content_recursive(e->tree);
1522                                 e->tree = subtree;
1524                                 /*
1525                                  * We need to leave e->versions[0].sha1 alone
1526                                  * to avoid modifying the preimage tree used
1527                                  * when writing out the parent directory.
1528                                  * But after replacing the subdir with a
1529                                  * completely different one, it's not a good
1530                                  * delta base any more, and besides, we've
1531                                  * thrown away the tree entries needed to
1532                                  * make a delta against it.
1533                                  *
1534                                  * So let's just explicitly disable deltas
1535                                  * for the subtree.
1536                                  */
1537                                 if (S_ISDIR(e->versions[0].mode))
1538                                         e->versions[0].mode |= NO_DELTA;
1540                                 hashclr(root->versions[1].sha1);
1541                                 return 1;
1542                         }
1543                         if (!S_ISDIR(e->versions[1].mode)) {
1544                                 e->tree = new_tree_content(8);
1545                                 e->versions[1].mode = S_IFDIR;
1546                         }
1547                         if (!e->tree)
1548                                 load_tree(e);
1549                         if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1550                                 hashclr(root->versions[1].sha1);
1551                                 return 1;
1552                         }
1553                         return 0;
1554                 }
1555         }
1557         if (t->entry_count == t->entry_capacity)
1558                 root->tree = t = grow_tree_content(t, t->entry_count);
1559         e = new_tree_entry();
1560         e->name = to_atom(p, n);
1561         e->versions[0].mode = 0;
1562         hashclr(e->versions[0].sha1);
1563         t->entries[t->entry_count++] = e;
1564         if (slash1) {
1565                 e->tree = new_tree_content(8);
1566                 e->versions[1].mode = S_IFDIR;
1567                 tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1568         } else {
1569                 e->tree = subtree;
1570                 e->versions[1].mode = mode;
1571                 hashcpy(e->versions[1].sha1, sha1);
1572         }
1573         hashclr(root->versions[1].sha1);
1574         return 1;
1577 static int tree_content_remove(
1578         struct tree_entry *root,
1579         const char *p,
1580         struct tree_entry *backup_leaf)
1582         struct tree_content *t;
1583         const char *slash1;
1584         unsigned int i, n;
1585         struct tree_entry *e;
1587         slash1 = strchr(p, '/');
1588         if (slash1)
1589                 n = slash1 - p;
1590         else
1591                 n = strlen(p);
1593         if (!root->tree)
1594                 load_tree(root);
1595         t = root->tree;
1596         for (i = 0; i < t->entry_count; i++) {
1597                 e = t->entries[i];
1598                 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1599                         if (slash1 && !S_ISDIR(e->versions[1].mode))
1600                                 /*
1601                                  * If p names a file in some subdirectory, and a
1602                                  * file or symlink matching the name of the
1603                                  * parent directory of p exists, then p cannot
1604                                  * exist and need not be deleted.
1605                                  */
1606                                 return 1;
1607                         if (!slash1 || !S_ISDIR(e->versions[1].mode))
1608                                 goto del_entry;
1609                         if (!e->tree)
1610                                 load_tree(e);
1611                         if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1612                                 for (n = 0; n < e->tree->entry_count; n++) {
1613                                         if (e->tree->entries[n]->versions[1].mode) {
1614                                                 hashclr(root->versions[1].sha1);
1615                                                 return 1;
1616                                         }
1617                                 }
1618                                 backup_leaf = NULL;
1619                                 goto del_entry;
1620                         }
1621                         return 0;
1622                 }
1623         }
1624         return 0;
1626 del_entry:
1627         if (backup_leaf)
1628                 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1629         else if (e->tree)
1630                 release_tree_content_recursive(e->tree);
1631         e->tree = NULL;
1632         e->versions[1].mode = 0;
1633         hashclr(e->versions[1].sha1);
1634         hashclr(root->versions[1].sha1);
1635         return 1;
1638 static int tree_content_get(
1639         struct tree_entry *root,
1640         const char *p,
1641         struct tree_entry *leaf)
1643         struct tree_content *t;
1644         const char *slash1;
1645         unsigned int i, n;
1646         struct tree_entry *e;
1648         slash1 = strchr(p, '/');
1649         if (slash1)
1650                 n = slash1 - p;
1651         else
1652                 n = strlen(p);
1654         if (!root->tree)
1655                 load_tree(root);
1656         t = root->tree;
1657         for (i = 0; i < t->entry_count; i++) {
1658                 e = t->entries[i];
1659                 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1660                         if (!slash1) {
1661                                 memcpy(leaf, e, sizeof(*leaf));
1662                                 if (e->tree && is_null_sha1(e->versions[1].sha1))
1663                                         leaf->tree = dup_tree_content(e->tree);
1664                                 else
1665                                         leaf->tree = NULL;
1666                                 return 1;
1667                         }
1668                         if (!S_ISDIR(e->versions[1].mode))
1669                                 return 0;
1670                         if (!e->tree)
1671                                 load_tree(e);
1672                         return tree_content_get(e, slash1 + 1, leaf);
1673                 }
1674         }
1675         return 0;
1678 static int update_branch(struct branch *b)
1680         static const char *msg = "fast-import";
1681         struct ref_lock *lock;
1682         unsigned char old_sha1[20];
1684         if (is_null_sha1(b->sha1))
1685                 return 0;
1686         if (read_ref(b->name, old_sha1))
1687                 hashclr(old_sha1);
1688         lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1689         if (!lock)
1690                 return error("Unable to lock %s", b->name);
1691         if (!force_update && !is_null_sha1(old_sha1)) {
1692                 struct commit *old_cmit, *new_cmit;
1694                 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1695                 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1696                 if (!old_cmit || !new_cmit) {
1697                         unlock_ref(lock);
1698                         return error("Branch %s is missing commits.", b->name);
1699                 }
1701                 if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1702                         unlock_ref(lock);
1703                         warning("Not updating %s"
1704                                 " (new tip %s does not contain %s)",
1705                                 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1706                         return -1;
1707                 }
1708         }
1709         if (write_ref_sha1(lock, b->sha1, msg) < 0)
1710                 return error("Unable to update %s", b->name);
1711         return 0;
1714 static void dump_branches(void)
1716         unsigned int i;
1717         struct branch *b;
1719         for (i = 0; i < branch_table_sz; i++) {
1720                 for (b = branch_table[i]; b; b = b->table_next_branch)
1721                         failure |= update_branch(b);
1722         }
1725 static void dump_tags(void)
1727         static const char *msg = "fast-import";
1728         struct tag *t;
1729         struct ref_lock *lock;
1730         char ref_name[PATH_MAX];
1732         for (t = first_tag; t; t = t->next_tag) {
1733                 sprintf(ref_name, "tags/%s", t->name);
1734                 lock = lock_ref_sha1(ref_name, NULL);
1735                 if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1736                         failure |= error("Unable to update %s", ref_name);
1737         }
1740 static void dump_marks_helper(FILE *f,
1741         uintmax_t base,
1742         struct mark_set *m)
1744         uintmax_t k;
1745         if (m->shift) {
1746                 for (k = 0; k < 1024; k++) {
1747                         if (m->data.sets[k])
1748                                 dump_marks_helper(f, base + (k << m->shift),
1749                                         m->data.sets[k]);
1750                 }
1751         } else {
1752                 for (k = 0; k < 1024; k++) {
1753                         if (m->data.marked[k])
1754                                 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1755                                         sha1_to_hex(m->data.marked[k]->idx.sha1));
1756                 }
1757         }
1760 static void dump_marks(void)
1762         static struct lock_file mark_lock;
1763         int mark_fd;
1764         FILE *f;
1766         if (!export_marks_file)
1767                 return;
1769         mark_fd = hold_lock_file_for_update(&mark_lock, export_marks_file, 0);
1770         if (mark_fd < 0) {
1771                 failure |= error("Unable to write marks file %s: %s",
1772                         export_marks_file, strerror(errno));
1773                 return;
1774         }
1776         f = fdopen(mark_fd, "w");
1777         if (!f) {
1778                 int saved_errno = errno;
1779                 rollback_lock_file(&mark_lock);
1780                 failure |= error("Unable to write marks file %s: %s",
1781                         export_marks_file, strerror(saved_errno));
1782                 return;
1783         }
1785         /*
1786          * Since the lock file was fdopen()'ed, it should not be close()'ed.
1787          * Assign -1 to the lock file descriptor so that commit_lock_file()
1788          * won't try to close() it.
1789          */
1790         mark_lock.fd = -1;
1792         dump_marks_helper(f, 0, marks);
1793         if (ferror(f) || fclose(f)) {
1794                 int saved_errno = errno;
1795                 rollback_lock_file(&mark_lock);
1796                 failure |= error("Unable to write marks file %s: %s",
1797                         export_marks_file, strerror(saved_errno));
1798                 return;
1799         }
1801         if (commit_lock_file(&mark_lock)) {
1802                 int saved_errno = errno;
1803                 rollback_lock_file(&mark_lock);
1804                 failure |= error("Unable to commit marks file %s: %s",
1805                         export_marks_file, strerror(saved_errno));
1806                 return;
1807         }
1810 static void read_marks(void)
1812         char line[512];
1813         FILE *f = fopen(import_marks_file, "r");
1814         if (f)
1815                 ;
1816         else if (import_marks_file_ignore_missing && errno == ENOENT)
1817                 return; /* Marks file does not exist */
1818         else
1819                 die_errno("cannot read '%s'", import_marks_file);
1820         while (fgets(line, sizeof(line), f)) {
1821                 uintmax_t mark;
1822                 char *end;
1823                 unsigned char sha1[20];
1824                 struct object_entry *e;
1826                 end = strchr(line, '\n');
1827                 if (line[0] != ':' || !end)
1828                         die("corrupt mark line: %s", line);
1829                 *end = 0;
1830                 mark = strtoumax(line + 1, &end, 10);
1831                 if (!mark || end == line + 1
1832                         || *end != ' ' || get_sha1(end + 1, sha1))
1833                         die("corrupt mark line: %s", line);
1834                 e = find_object(sha1);
1835                 if (!e) {
1836                         enum object_type type = sha1_object_info(sha1, NULL);
1837                         if (type < 0)
1838                                 die("object not found: %s", sha1_to_hex(sha1));
1839                         e = insert_object(sha1);
1840                         e->type = type;
1841                         e->pack_id = MAX_PACK_ID;
1842                         e->idx.offset = 1; /* just not zero! */
1843                 }
1844                 insert_mark(mark, e);
1845         }
1846         fclose(f);
1850 static int read_next_command(void)
1852         static int stdin_eof = 0;
1854         if (stdin_eof) {
1855                 unread_command_buf = 0;
1856                 return EOF;
1857         }
1859         for (;;) {
1860                 if (unread_command_buf) {
1861                         unread_command_buf = 0;
1862                 } else {
1863                         struct recent_command *rc;
1865                         strbuf_detach(&command_buf, NULL);
1866                         stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1867                         if (stdin_eof)
1868                                 return EOF;
1870                         if (!seen_data_command
1871                                 && prefixcmp(command_buf.buf, "feature ")
1872                                 && prefixcmp(command_buf.buf, "option ")) {
1873                                 parse_argv();
1874                         }
1876                         rc = rc_free;
1877                         if (rc)
1878                                 rc_free = rc->next;
1879                         else {
1880                                 rc = cmd_hist.next;
1881                                 cmd_hist.next = rc->next;
1882                                 cmd_hist.next->prev = &cmd_hist;
1883                                 free(rc->buf);
1884                         }
1886                         rc->buf = command_buf.buf;
1887                         rc->prev = cmd_tail;
1888                         rc->next = cmd_hist.prev;
1889                         rc->prev->next = rc;
1890                         cmd_tail = rc;
1891                 }
1892                 if (!prefixcmp(command_buf.buf, "cat-blob ")) {
1893                         parse_cat_blob();
1894                         continue;
1895                 }
1896                 if (command_buf.buf[0] == '#')
1897                         continue;
1898                 return 0;
1899         }
1902 static void skip_optional_lf(void)
1904         int term_char = fgetc(stdin);
1905         if (term_char != '\n' && term_char != EOF)
1906                 ungetc(term_char, stdin);
1909 static void parse_mark(void)
1911         if (!prefixcmp(command_buf.buf, "mark :")) {
1912                 next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1913                 read_next_command();
1914         }
1915         else
1916                 next_mark = 0;
1919 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1921         strbuf_reset(sb);
1923         if (prefixcmp(command_buf.buf, "data "))
1924                 die("Expected 'data n' command, found: %s", command_buf.buf);
1926         if (!prefixcmp(command_buf.buf + 5, "<<")) {
1927                 char *term = xstrdup(command_buf.buf + 5 + 2);
1928                 size_t term_len = command_buf.len - 5 - 2;
1930                 strbuf_detach(&command_buf, NULL);
1931                 for (;;) {
1932                         if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1933                                 die("EOF in data (terminator '%s' not found)", term);
1934                         if (term_len == command_buf.len
1935                                 && !strcmp(term, command_buf.buf))
1936                                 break;
1937                         strbuf_addbuf(sb, &command_buf);
1938                         strbuf_addch(sb, '\n');
1939                 }
1940                 free(term);
1941         }
1942         else {
1943                 uintmax_t len = strtoumax(command_buf.buf + 5, NULL, 10);
1944                 size_t n = 0, length = (size_t)len;
1946                 if (limit && limit < len) {
1947                         *len_res = len;
1948                         return 0;
1949                 }
1950                 if (length < len)
1951                         die("data is too large to use in this context");
1953                 while (n < length) {
1954                         size_t s = strbuf_fread(sb, length - n, stdin);
1955                         if (!s && feof(stdin))
1956                                 die("EOF in data (%lu bytes remaining)",
1957                                         (unsigned long)(length - n));
1958                         n += s;
1959                 }
1960         }
1962         skip_optional_lf();
1963         return 1;
1966 static int validate_raw_date(const char *src, char *result, int maxlen)
1968         const char *orig_src = src;
1969         char *endp;
1970         unsigned long num;
1972         errno = 0;
1974         num = strtoul(src, &endp, 10);
1975         /* NEEDSWORK: perhaps check for reasonable values? */
1976         if (errno || endp == src || *endp != ' ')
1977                 return -1;
1979         src = endp + 1;
1980         if (*src != '-' && *src != '+')
1981                 return -1;
1983         num = strtoul(src + 1, &endp, 10);
1984         if (errno || endp == src + 1 || *endp || (endp - orig_src) >= maxlen ||
1985             1400 < num)
1986                 return -1;
1988         strcpy(result, orig_src);
1989         return 0;
1992 static char *parse_ident(const char *buf)
1994         const char *ltgt;
1995         size_t name_len;
1996         char *ident;
1998         /* ensure there is a space delimiter even if there is no name */
1999         if (*buf == '<')
2000                 --buf;
2002         ltgt = buf + strcspn(buf, "<>");
2003         if (*ltgt != '<')
2004                 die("Missing < in ident string: %s", buf);
2005         if (ltgt != buf && ltgt[-1] != ' ')
2006                 die("Missing space before < in ident string: %s", buf);
2007         ltgt = ltgt + 1 + strcspn(ltgt + 1, "<>");
2008         if (*ltgt != '>')
2009                 die("Missing > in ident string: %s", buf);
2010         ltgt++;
2011         if (*ltgt != ' ')
2012                 die("Missing space after > in ident string: %s", buf);
2013         ltgt++;
2014         name_len = ltgt - buf;
2015         ident = xmalloc(name_len + 24);
2016         strncpy(ident, buf, name_len);
2018         switch (whenspec) {
2019         case WHENSPEC_RAW:
2020                 if (validate_raw_date(ltgt, ident + name_len, 24) < 0)
2021                         die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
2022                 break;
2023         case WHENSPEC_RFC2822:
2024                 if (parse_date(ltgt, ident + name_len, 24) < 0)
2025                         die("Invalid rfc2822 date \"%s\" in ident: %s", ltgt, buf);
2026                 break;
2027         case WHENSPEC_NOW:
2028                 if (strcmp("now", ltgt))
2029                         die("Date in ident must be 'now': %s", buf);
2030                 datestamp(ident + name_len, 24);
2031                 break;
2032         }
2034         return ident;
2037 static void parse_and_store_blob(
2038         struct last_object *last,
2039         unsigned char *sha1out,
2040         uintmax_t mark)
2042         static struct strbuf buf = STRBUF_INIT;
2043         uintmax_t len;
2045         if (parse_data(&buf, big_file_threshold, &len))
2046                 store_object(OBJ_BLOB, &buf, last, sha1out, mark);
2047         else {
2048                 if (last) {
2049                         strbuf_release(&last->data);
2050                         last->offset = 0;
2051                         last->depth = 0;
2052                 }
2053                 stream_blob(len, sha1out, mark);
2054                 skip_optional_lf();
2055         }
2058 static void parse_new_blob(void)
2060         read_next_command();
2061         parse_mark();
2062         parse_and_store_blob(&last_blob, NULL, next_mark);
2065 static void unload_one_branch(void)
2067         while (cur_active_branches
2068                 && cur_active_branches >= max_active_branches) {
2069                 uintmax_t min_commit = ULONG_MAX;
2070                 struct branch *e, *l = NULL, *p = NULL;
2072                 for (e = active_branches; e; e = e->active_next_branch) {
2073                         if (e->last_commit < min_commit) {
2074                                 p = l;
2075                                 min_commit = e->last_commit;
2076                         }
2077                         l = e;
2078                 }
2080                 if (p) {
2081                         e = p->active_next_branch;
2082                         p->active_next_branch = e->active_next_branch;
2083                 } else {
2084                         e = active_branches;
2085                         active_branches = e->active_next_branch;
2086                 }
2087                 e->active = 0;
2088                 e->active_next_branch = NULL;
2089                 if (e->branch_tree.tree) {
2090                         release_tree_content_recursive(e->branch_tree.tree);
2091                         e->branch_tree.tree = NULL;
2092                 }
2093                 cur_active_branches--;
2094         }
2097 static void load_branch(struct branch *b)
2099         load_tree(&b->branch_tree);
2100         if (!b->active) {
2101                 b->active = 1;
2102                 b->active_next_branch = active_branches;
2103                 active_branches = b;
2104                 cur_active_branches++;
2105                 branch_load_count++;
2106         }
2109 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2111         unsigned char fanout = 0;
2112         while ((num_notes >>= 8))
2113                 fanout++;
2114         return fanout;
2117 static void construct_path_with_fanout(const char *hex_sha1,
2118                 unsigned char fanout, char *path)
2120         unsigned int i = 0, j = 0;
2121         if (fanout >= 20)
2122                 die("Too large fanout (%u)", fanout);
2123         while (fanout) {
2124                 path[i++] = hex_sha1[j++];
2125                 path[i++] = hex_sha1[j++];
2126                 path[i++] = '/';
2127                 fanout--;
2128         }
2129         memcpy(path + i, hex_sha1 + j, 40 - j);
2130         path[i + 40 - j] = '\0';
2133 static uintmax_t do_change_note_fanout(
2134                 struct tree_entry *orig_root, struct tree_entry *root,
2135                 char *hex_sha1, unsigned int hex_sha1_len,
2136                 char *fullpath, unsigned int fullpath_len,
2137                 unsigned char fanout)
2139         struct tree_content *t = root->tree;
2140         struct tree_entry *e, leaf;
2141         unsigned int i, tmp_hex_sha1_len, tmp_fullpath_len;
2142         uintmax_t num_notes = 0;
2143         unsigned char sha1[20];
2144         char realpath[60];
2146         for (i = 0; t && i < t->entry_count; i++) {
2147                 e = t->entries[i];
2148                 tmp_hex_sha1_len = hex_sha1_len + e->name->str_len;
2149                 tmp_fullpath_len = fullpath_len;
2151                 /*
2152                  * We're interested in EITHER existing note entries (entries
2153                  * with exactly 40 hex chars in path, not including directory
2154                  * separators), OR directory entries that may contain note
2155                  * entries (with < 40 hex chars in path).
2156                  * Also, each path component in a note entry must be a multiple
2157                  * of 2 chars.
2158                  */
2159                 if (!e->versions[1].mode ||
2160                     tmp_hex_sha1_len > 40 ||
2161                     e->name->str_len % 2)
2162                         continue;
2164                 /* This _may_ be a note entry, or a subdir containing notes */
2165                 memcpy(hex_sha1 + hex_sha1_len, e->name->str_dat,
2166                        e->name->str_len);
2167                 if (tmp_fullpath_len)
2168                         fullpath[tmp_fullpath_len++] = '/';
2169                 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2170                        e->name->str_len);
2171                 tmp_fullpath_len += e->name->str_len;
2172                 fullpath[tmp_fullpath_len] = '\0';
2174                 if (tmp_hex_sha1_len == 40 && !get_sha1_hex(hex_sha1, sha1)) {
2175                         /* This is a note entry */
2176                         if (fanout == 0xff) {
2177                                 /* Counting mode, no rename */
2178                                 num_notes++;
2179                                 continue;
2180                         }
2181                         construct_path_with_fanout(hex_sha1, fanout, realpath);
2182                         if (!strcmp(fullpath, realpath)) {
2183                                 /* Note entry is in correct location */
2184                                 num_notes++;
2185                                 continue;
2186                         }
2188                         /* Rename fullpath to realpath */
2189                         if (!tree_content_remove(orig_root, fullpath, &leaf))
2190                                 die("Failed to remove path %s", fullpath);
2191                         tree_content_set(orig_root, realpath,
2192                                 leaf.versions[1].sha1,
2193                                 leaf.versions[1].mode,
2194                                 leaf.tree);
2195                 } else if (S_ISDIR(e->versions[1].mode)) {
2196                         /* This is a subdir that may contain note entries */
2197                         if (!e->tree)
2198                                 load_tree(e);
2199                         num_notes += do_change_note_fanout(orig_root, e,
2200                                 hex_sha1, tmp_hex_sha1_len,
2201                                 fullpath, tmp_fullpath_len, fanout);
2202                 }
2204                 /* The above may have reallocated the current tree_content */
2205                 t = root->tree;
2206         }
2207         return num_notes;
2210 static uintmax_t change_note_fanout(struct tree_entry *root,
2211                 unsigned char fanout)
2213         char hex_sha1[40], path[60];
2214         return do_change_note_fanout(root, root, hex_sha1, 0, path, 0, fanout);
2217 static void file_change_m(struct branch *b)
2219         const char *p = command_buf.buf + 2;
2220         static struct strbuf uq = STRBUF_INIT;
2221         const char *endp;
2222         struct object_entry *oe = oe;
2223         unsigned char sha1[20];
2224         uint16_t mode, inline_data = 0;
2226         p = get_mode(p, &mode);
2227         if (!p)
2228                 die("Corrupt mode: %s", command_buf.buf);
2229         switch (mode) {
2230         case 0644:
2231         case 0755:
2232                 mode |= S_IFREG;
2233         case S_IFREG | 0644:
2234         case S_IFREG | 0755:
2235         case S_IFLNK:
2236         case S_IFDIR:
2237         case S_IFGITLINK:
2238                 /* ok */
2239                 break;
2240         default:
2241                 die("Corrupt mode: %s", command_buf.buf);
2242         }
2244         if (*p == ':') {
2245                 char *x;
2246                 oe = find_mark(strtoumax(p + 1, &x, 10));
2247                 hashcpy(sha1, oe->idx.sha1);
2248                 p = x;
2249         } else if (!prefixcmp(p, "inline")) {
2250                 inline_data = 1;
2251                 p += 6;
2252         } else {
2253                 if (get_sha1_hex(p, sha1))
2254                         die("Invalid SHA1: %s", command_buf.buf);
2255                 oe = find_object(sha1);
2256                 p += 40;
2257         }
2258         if (*p++ != ' ')
2259                 die("Missing space after SHA1: %s", command_buf.buf);
2261         strbuf_reset(&uq);
2262         if (!unquote_c_style(&uq, p, &endp)) {
2263                 if (*endp)
2264                         die("Garbage after path in: %s", command_buf.buf);
2265                 p = uq.buf;
2266         }
2268         /* Git does not track empty, non-toplevel directories. */
2269         if (S_ISDIR(mode) && !memcmp(sha1, EMPTY_TREE_SHA1_BIN, 20) && *p) {
2270                 tree_content_remove(&b->branch_tree, p, NULL);
2271                 return;
2272         }
2274         if (S_ISGITLINK(mode)) {
2275                 if (inline_data)
2276                         die("Git links cannot be specified 'inline': %s",
2277                                 command_buf.buf);
2278                 else if (oe) {
2279                         if (oe->type != OBJ_COMMIT)
2280                                 die("Not a commit (actually a %s): %s",
2281                                         typename(oe->type), command_buf.buf);
2282                 }
2283                 /*
2284                  * Accept the sha1 without checking; it expected to be in
2285                  * another repository.
2286                  */
2287         } else if (inline_data) {
2288                 if (S_ISDIR(mode))
2289                         die("Directories cannot be specified 'inline': %s",
2290                                 command_buf.buf);
2291                 if (p != uq.buf) {
2292                         strbuf_addstr(&uq, p);
2293                         p = uq.buf;
2294                 }
2295                 read_next_command();
2296                 parse_and_store_blob(&last_blob, sha1, 0);
2297         } else {
2298                 enum object_type expected = S_ISDIR(mode) ?
2299                                                 OBJ_TREE: OBJ_BLOB;
2300                 enum object_type type = oe ? oe->type :
2301                                         sha1_object_info(sha1, NULL);
2302                 if (type < 0)
2303                         die("%s not found: %s",
2304                                         S_ISDIR(mode) ?  "Tree" : "Blob",
2305                                         command_buf.buf);
2306                 if (type != expected)
2307                         die("Not a %s (actually a %s): %s",
2308                                 typename(expected), typename(type),
2309                                 command_buf.buf);
2310         }
2312         if (!*p) {
2313                 tree_content_replace(&b->branch_tree, sha1, mode, NULL);
2314                 return;
2315         }
2316         tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
2319 static void file_change_d(struct branch *b)
2321         const char *p = command_buf.buf + 2;
2322         static struct strbuf uq = STRBUF_INIT;
2323         const char *endp;
2325         strbuf_reset(&uq);
2326         if (!unquote_c_style(&uq, p, &endp)) {
2327                 if (*endp)
2328                         die("Garbage after path in: %s", command_buf.buf);
2329                 p = uq.buf;
2330         }
2331         tree_content_remove(&b->branch_tree, p, NULL);
2334 static void file_change_cr(struct branch *b, int rename)
2336         const char *s, *d;
2337         static struct strbuf s_uq = STRBUF_INIT;
2338         static struct strbuf d_uq = STRBUF_INIT;
2339         const char *endp;
2340         struct tree_entry leaf;
2342         s = command_buf.buf + 2;
2343         strbuf_reset(&s_uq);
2344         if (!unquote_c_style(&s_uq, s, &endp)) {
2345                 if (*endp != ' ')
2346                         die("Missing space after source: %s", command_buf.buf);
2347         } else {
2348                 endp = strchr(s, ' ');
2349                 if (!endp)
2350                         die("Missing space after source: %s", command_buf.buf);
2351                 strbuf_add(&s_uq, s, endp - s);
2352         }
2353         s = s_uq.buf;
2355         endp++;
2356         if (!*endp)
2357                 die("Missing dest: %s", command_buf.buf);
2359         d = endp;
2360         strbuf_reset(&d_uq);
2361         if (!unquote_c_style(&d_uq, d, &endp)) {
2362                 if (*endp)
2363                         die("Garbage after dest in: %s", command_buf.buf);
2364                 d = d_uq.buf;
2365         }
2367         memset(&leaf, 0, sizeof(leaf));
2368         if (rename)
2369                 tree_content_remove(&b->branch_tree, s, &leaf);
2370         else
2371                 tree_content_get(&b->branch_tree, s, &leaf);
2372         if (!leaf.versions[1].mode)
2373                 die("Path %s not in branch", s);
2374         if (!*d) {      /* C "path/to/subdir" "" */
2375                 tree_content_replace(&b->branch_tree,
2376                         leaf.versions[1].sha1,
2377                         leaf.versions[1].mode,
2378                         leaf.tree);
2379                 return;
2380         }
2381         tree_content_set(&b->branch_tree, d,
2382                 leaf.versions[1].sha1,
2383                 leaf.versions[1].mode,
2384                 leaf.tree);
2387 static void note_change_n(struct branch *b, unsigned char *old_fanout)
2389         const char *p = command_buf.buf + 2;
2390         static struct strbuf uq = STRBUF_INIT;
2391         struct object_entry *oe = oe;
2392         struct branch *s;
2393         unsigned char sha1[20], commit_sha1[20];
2394         char path[60];
2395         uint16_t inline_data = 0;
2396         unsigned char new_fanout;
2398         /*
2399          * When loading a branch, we don't traverse its tree to count the real
2400          * number of notes (too expensive to do this for all non-note refs).
2401          * This means that recently loaded notes refs might incorrectly have
2402          * b->num_notes == 0, and consequently, old_fanout might be wrong.
2403          *
2404          * Fix this by traversing the tree and counting the number of notes
2405          * when b->num_notes == 0. If the notes tree is truly empty, the
2406          * calculation should not take long.
2407          */
2408         if (b->num_notes == 0 && *old_fanout == 0) {
2409                 /* Invoke change_note_fanout() in "counting mode". */
2410                 b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2411                 *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2412         }
2414         /* Now parse the notemodify command. */
2415         /* <dataref> or 'inline' */
2416         if (*p == ':') {
2417                 char *x;
2418                 oe = find_mark(strtoumax(p + 1, &x, 10));
2419                 hashcpy(sha1, oe->idx.sha1);
2420                 p = x;
2421         } else if (!prefixcmp(p, "inline")) {
2422                 inline_data = 1;
2423                 p += 6;
2424         } else {
2425                 if (get_sha1_hex(p, sha1))
2426                         die("Invalid SHA1: %s", command_buf.buf);
2427                 oe = find_object(sha1);
2428                 p += 40;
2429         }
2430         if (*p++ != ' ')
2431                 die("Missing space after SHA1: %s", command_buf.buf);
2433         /* <committish> */
2434         s = lookup_branch(p);
2435         if (s) {
2436                 if (is_null_sha1(s->sha1))
2437                         die("Can't add a note on empty branch.");
2438                 hashcpy(commit_sha1, s->sha1);
2439         } else if (*p == ':') {
2440                 uintmax_t commit_mark = strtoumax(p + 1, NULL, 10);
2441                 struct object_entry *commit_oe = find_mark(commit_mark);
2442                 if (commit_oe->type != OBJ_COMMIT)
2443                         die("Mark :%" PRIuMAX " not a commit", commit_mark);
2444                 hashcpy(commit_sha1, commit_oe->idx.sha1);
2445         } else if (!get_sha1(p, commit_sha1)) {
2446                 unsigned long size;
2447                 char *buf = read_object_with_reference(commit_sha1,
2448                         commit_type, &size, commit_sha1);
2449                 if (!buf || size < 46)
2450                         die("Not a valid commit: %s", p);
2451                 free(buf);
2452         } else
2453                 die("Invalid ref name or SHA1 expression: %s", p);
2455         if (inline_data) {
2456                 if (p != uq.buf) {
2457                         strbuf_addstr(&uq, p);
2458                         p = uq.buf;
2459                 }
2460                 read_next_command();
2461                 parse_and_store_blob(&last_blob, sha1, 0);
2462         } else if (oe) {
2463                 if (oe->type != OBJ_BLOB)
2464                         die("Not a blob (actually a %s): %s",
2465                                 typename(oe->type), command_buf.buf);
2466         } else if (!is_null_sha1(sha1)) {
2467                 enum object_type type = sha1_object_info(sha1, NULL);
2468                 if (type < 0)
2469                         die("Blob not found: %s", command_buf.buf);
2470                 if (type != OBJ_BLOB)
2471                         die("Not a blob (actually a %s): %s",
2472                             typename(type), command_buf.buf);
2473         }
2475         construct_path_with_fanout(sha1_to_hex(commit_sha1), *old_fanout, path);
2476         if (tree_content_remove(&b->branch_tree, path, NULL))
2477                 b->num_notes--;
2479         if (is_null_sha1(sha1))
2480                 return; /* nothing to insert */
2482         b->num_notes++;
2483         new_fanout = convert_num_notes_to_fanout(b->num_notes);
2484         construct_path_with_fanout(sha1_to_hex(commit_sha1), new_fanout, path);
2485         tree_content_set(&b->branch_tree, path, sha1, S_IFREG | 0644, NULL);
2488 static void file_change_deleteall(struct branch *b)
2490         release_tree_content_recursive(b->branch_tree.tree);
2491         hashclr(b->branch_tree.versions[0].sha1);
2492         hashclr(b->branch_tree.versions[1].sha1);
2493         load_tree(&b->branch_tree);
2494         b->num_notes = 0;
2497 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2499         if (!buf || size < 46)
2500                 die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2501         if (memcmp("tree ", buf, 5)
2502                 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2503                 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2504         hashcpy(b->branch_tree.versions[0].sha1,
2505                 b->branch_tree.versions[1].sha1);
2508 static void parse_from_existing(struct branch *b)
2510         if (is_null_sha1(b->sha1)) {
2511                 hashclr(b->branch_tree.versions[0].sha1);
2512                 hashclr(b->branch_tree.versions[1].sha1);
2513         } else {
2514                 unsigned long size;
2515                 char *buf;
2517                 buf = read_object_with_reference(b->sha1,
2518                         commit_type, &size, b->sha1);
2519                 parse_from_commit(b, buf, size);
2520                 free(buf);
2521         }
2524 static int parse_from(struct branch *b)
2526         const char *from;
2527         struct branch *s;
2529         if (prefixcmp(command_buf.buf, "from "))
2530                 return 0;
2532         if (b->branch_tree.tree) {
2533                 release_tree_content_recursive(b->branch_tree.tree);
2534                 b->branch_tree.tree = NULL;
2535         }
2537         from = strchr(command_buf.buf, ' ') + 1;
2538         s = lookup_branch(from);
2539         if (b == s)
2540                 die("Can't create a branch from itself: %s", b->name);
2541         else if (s) {
2542                 unsigned char *t = s->branch_tree.versions[1].sha1;
2543                 hashcpy(b->sha1, s->sha1);
2544                 hashcpy(b->branch_tree.versions[0].sha1, t);
2545                 hashcpy(b->branch_tree.versions[1].sha1, t);
2546         } else if (*from == ':') {
2547                 uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2548                 struct object_entry *oe = find_mark(idnum);
2549                 if (oe->type != OBJ_COMMIT)
2550                         die("Mark :%" PRIuMAX " not a commit", idnum);
2551                 hashcpy(b->sha1, oe->idx.sha1);
2552                 if (oe->pack_id != MAX_PACK_ID) {
2553                         unsigned long size;
2554                         char *buf = gfi_unpack_entry(oe, &size);
2555                         parse_from_commit(b, buf, size);
2556                         free(buf);
2557                 } else
2558                         parse_from_existing(b);
2559         } else if (!get_sha1(from, b->sha1))
2560                 parse_from_existing(b);
2561         else
2562                 die("Invalid ref name or SHA1 expression: %s", from);
2564         read_next_command();
2565         return 1;
2568 static struct hash_list *parse_merge(unsigned int *count)
2570         struct hash_list *list = NULL, *n, *e = e;
2571         const char *from;
2572         struct branch *s;
2574         *count = 0;
2575         while (!prefixcmp(command_buf.buf, "merge ")) {
2576                 from = strchr(command_buf.buf, ' ') + 1;
2577                 n = xmalloc(sizeof(*n));
2578                 s = lookup_branch(from);
2579                 if (s)
2580                         hashcpy(n->sha1, s->sha1);
2581                 else if (*from == ':') {
2582                         uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2583                         struct object_entry *oe = find_mark(idnum);
2584                         if (oe->type != OBJ_COMMIT)
2585                                 die("Mark :%" PRIuMAX " not a commit", idnum);
2586                         hashcpy(n->sha1, oe->idx.sha1);
2587                 } else if (!get_sha1(from, n->sha1)) {
2588                         unsigned long size;
2589                         char *buf = read_object_with_reference(n->sha1,
2590                                 commit_type, &size, n->sha1);
2591                         if (!buf || size < 46)
2592                                 die("Not a valid commit: %s", from);
2593                         free(buf);
2594                 } else
2595                         die("Invalid ref name or SHA1 expression: %s", from);
2597                 n->next = NULL;
2598                 if (list)
2599                         e->next = n;
2600                 else
2601                         list = n;
2602                 e = n;
2603                 (*count)++;
2604                 read_next_command();
2605         }
2606         return list;
2609 static void parse_new_commit(void)
2611         static struct strbuf msg = STRBUF_INIT;
2612         struct branch *b;
2613         char *sp;
2614         char *author = NULL;
2615         char *committer = NULL;
2616         struct hash_list *merge_list = NULL;
2617         unsigned int merge_count;
2618         unsigned char prev_fanout, new_fanout;
2620         /* Obtain the branch name from the rest of our command */
2621         sp = strchr(command_buf.buf, ' ') + 1;
2622         b = lookup_branch(sp);
2623         if (!b)
2624                 b = new_branch(sp);
2626         read_next_command();
2627         parse_mark();
2628         if (!prefixcmp(command_buf.buf, "author ")) {
2629                 author = parse_ident(command_buf.buf + 7);
2630                 read_next_command();
2631         }
2632         if (!prefixcmp(command_buf.buf, "committer ")) {
2633                 committer = parse_ident(command_buf.buf + 10);
2634                 read_next_command();
2635         }
2636         if (!committer)
2637                 die("Expected committer but didn't get one");
2638         parse_data(&msg, 0, NULL);
2639         read_next_command();
2640         parse_from(b);
2641         merge_list = parse_merge(&merge_count);
2643         /* ensure the branch is active/loaded */
2644         if (!b->branch_tree.tree || !max_active_branches) {
2645                 unload_one_branch();
2646                 load_branch(b);
2647         }
2649         prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2651         /* file_change* */
2652         while (command_buf.len > 0) {
2653                 if (!prefixcmp(command_buf.buf, "M "))
2654                         file_change_m(b);
2655                 else if (!prefixcmp(command_buf.buf, "D "))
2656                         file_change_d(b);
2657                 else if (!prefixcmp(command_buf.buf, "R "))
2658                         file_change_cr(b, 1);
2659                 else if (!prefixcmp(command_buf.buf, "C "))
2660                         file_change_cr(b, 0);
2661                 else if (!prefixcmp(command_buf.buf, "N "))
2662                         note_change_n(b, &prev_fanout);
2663                 else if (!strcmp("deleteall", command_buf.buf))
2664                         file_change_deleteall(b);
2665                 else if (!prefixcmp(command_buf.buf, "ls "))
2666                         parse_ls(b);
2667                 else {
2668                         unread_command_buf = 1;
2669                         break;
2670                 }
2671                 if (read_next_command() == EOF)
2672                         break;
2673         }
2675         new_fanout = convert_num_notes_to_fanout(b->num_notes);
2676         if (new_fanout != prev_fanout)
2677                 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2679         /* build the tree and the commit */
2680         store_tree(&b->branch_tree);
2681         hashcpy(b->branch_tree.versions[0].sha1,
2682                 b->branch_tree.versions[1].sha1);
2684         strbuf_reset(&new_data);
2685         strbuf_addf(&new_data, "tree %s\n",
2686                 sha1_to_hex(b->branch_tree.versions[1].sha1));
2687         if (!is_null_sha1(b->sha1))
2688                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2689         while (merge_list) {
2690                 struct hash_list *next = merge_list->next;
2691                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2692                 free(merge_list);
2693                 merge_list = next;
2694         }
2695         strbuf_addf(&new_data,
2696                 "author %s\n"
2697                 "committer %s\n"
2698                 "\n",
2699                 author ? author : committer, committer);
2700         strbuf_addbuf(&new_data, &msg);
2701         free(author);
2702         free(committer);
2704         if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2705                 b->pack_id = pack_id;
2706         b->last_commit = object_count_by_type[OBJ_COMMIT];
2709 static void parse_new_tag(void)
2711         static struct strbuf msg = STRBUF_INIT;
2712         char *sp;
2713         const char *from;
2714         char *tagger;
2715         struct branch *s;
2716         struct tag *t;
2717         uintmax_t from_mark = 0;
2718         unsigned char sha1[20];
2719         enum object_type type;
2721         /* Obtain the new tag name from the rest of our command */
2722         sp = strchr(command_buf.buf, ' ') + 1;
2723         t = pool_alloc(sizeof(struct tag));
2724         t->next_tag = NULL;
2725         t->name = pool_strdup(sp);
2726         if (last_tag)
2727                 last_tag->next_tag = t;
2728         else
2729                 first_tag = t;
2730         last_tag = t;
2731         read_next_command();
2733         /* from ... */
2734         if (prefixcmp(command_buf.buf, "from "))
2735                 die("Expected from command, got %s", command_buf.buf);
2736         from = strchr(command_buf.buf, ' ') + 1;
2737         s = lookup_branch(from);
2738         if (s) {
2739                 if (is_null_sha1(s->sha1))
2740                         die("Can't tag an empty branch.");
2741                 hashcpy(sha1, s->sha1);
2742                 type = OBJ_COMMIT;
2743         } else if (*from == ':') {
2744                 struct object_entry *oe;
2745                 from_mark = strtoumax(from + 1, NULL, 10);
2746                 oe = find_mark(from_mark);
2747                 type = oe->type;
2748                 hashcpy(sha1, oe->idx.sha1);
2749         } else if (!get_sha1(from, sha1)) {
2750                 struct object_entry *oe = find_object(sha1);
2751                 if (!oe) {
2752                         type = sha1_object_info(sha1, NULL);
2753                         if (type < 0)
2754                                 die("Not a valid object: %s", from);
2755                 } else
2756                         type = oe->type;
2757         } else
2758                 die("Invalid ref name or SHA1 expression: %s", from);
2759         read_next_command();
2761         /* tagger ... */
2762         if (!prefixcmp(command_buf.buf, "tagger ")) {
2763                 tagger = parse_ident(command_buf.buf + 7);
2764                 read_next_command();
2765         } else
2766                 tagger = NULL;
2768         /* tag payload/message */
2769         parse_data(&msg, 0, NULL);
2771         /* build the tag object */
2772         strbuf_reset(&new_data);
2774         strbuf_addf(&new_data,
2775                     "object %s\n"
2776                     "type %s\n"
2777                     "tag %s\n",
2778                     sha1_to_hex(sha1), typename(type), t->name);
2779         if (tagger)
2780                 strbuf_addf(&new_data,
2781                             "tagger %s\n", tagger);
2782         strbuf_addch(&new_data, '\n');
2783         strbuf_addbuf(&new_data, &msg);
2784         free(tagger);
2786         if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2787                 t->pack_id = MAX_PACK_ID;
2788         else
2789                 t->pack_id = pack_id;
2792 static void parse_reset_branch(void)
2794         struct branch *b;
2795         char *sp;
2797         /* Obtain the branch name from the rest of our command */
2798         sp = strchr(command_buf.buf, ' ') + 1;
2799         b = lookup_branch(sp);
2800         if (b) {
2801                 hashclr(b->sha1);
2802                 hashclr(b->branch_tree.versions[0].sha1);
2803                 hashclr(b->branch_tree.versions[1].sha1);
2804                 if (b->branch_tree.tree) {
2805                         release_tree_content_recursive(b->branch_tree.tree);
2806                         b->branch_tree.tree = NULL;
2807                 }
2808         }
2809         else
2810                 b = new_branch(sp);
2811         read_next_command();
2812         parse_from(b);
2813         if (command_buf.len > 0)
2814                 unread_command_buf = 1;
2817 static void cat_blob_write(const char *buf, unsigned long size)
2819         if (write_in_full(cat_blob_fd, buf, size) != size)
2820                 die_errno("Write to frontend failed");
2823 static void cat_blob(struct object_entry *oe, unsigned char sha1[20])
2825         struct strbuf line = STRBUF_INIT;
2826         unsigned long size;
2827         enum object_type type = 0;
2828         char *buf;
2830         if (!oe || oe->pack_id == MAX_PACK_ID) {
2831                 buf = read_sha1_file(sha1, &type, &size);
2832         } else {
2833                 type = oe->type;
2834                 buf = gfi_unpack_entry(oe, &size);
2835         }
2837         /*
2838          * Output based on batch_one_object() from cat-file.c.
2839          */
2840         if (type <= 0) {
2841                 strbuf_reset(&line);
2842                 strbuf_addf(&line, "%s missing\n", sha1_to_hex(sha1));
2843                 cat_blob_write(line.buf, line.len);
2844                 strbuf_release(&line);
2845                 free(buf);
2846                 return;
2847         }
2848         if (!buf)
2849                 die("Can't read object %s", sha1_to_hex(sha1));
2850         if (type != OBJ_BLOB)
2851                 die("Object %s is a %s but a blob was expected.",
2852                     sha1_to_hex(sha1), typename(type));
2853         strbuf_reset(&line);
2854         strbuf_addf(&line, "%s %s %lu\n", sha1_to_hex(sha1),
2855                                                 typename(type), size);
2856         cat_blob_write(line.buf, line.len);
2857         strbuf_release(&line);
2858         cat_blob_write(buf, size);
2859         cat_blob_write("\n", 1);
2860         if (oe && oe->pack_id == pack_id) {
2861                 last_blob.offset = oe->idx.offset;
2862                 strbuf_attach(&last_blob.data, buf, size, size);
2863                 last_blob.depth = oe->depth;
2864         } else
2865                 free(buf);
2868 static void parse_cat_blob(void)
2870         const char *p;
2871         struct object_entry *oe = oe;
2872         unsigned char sha1[20];
2874         /* cat-blob SP <object> LF */
2875         p = command_buf.buf + strlen("cat-blob ");
2876         if (*p == ':') {
2877                 char *x;
2878                 oe = find_mark(strtoumax(p + 1, &x, 10));
2879                 if (x == p + 1)
2880                         die("Invalid mark: %s", command_buf.buf);
2881                 if (!oe)
2882                         die("Unknown mark: %s", command_buf.buf);
2883                 if (*x)
2884                         die("Garbage after mark: %s", command_buf.buf);
2885                 hashcpy(sha1, oe->idx.sha1);
2886         } else {
2887                 if (get_sha1_hex(p, sha1))
2888                         die("Invalid SHA1: %s", command_buf.buf);
2889                 if (p[40])
2890                         die("Garbage after SHA1: %s", command_buf.buf);
2891                 oe = find_object(sha1);
2892         }
2894         cat_blob(oe, sha1);
2897 static struct object_entry *dereference(struct object_entry *oe,
2898                                         unsigned char sha1[20])
2900         unsigned long size;
2901         char *buf = NULL;
2902         if (!oe) {
2903                 enum object_type type = sha1_object_info(sha1, NULL);
2904                 if (type < 0)
2905                         die("object not found: %s", sha1_to_hex(sha1));
2906                 /* cache it! */
2907                 oe = insert_object(sha1);
2908                 oe->type = type;
2909                 oe->pack_id = MAX_PACK_ID;
2910                 oe->idx.offset = 1;
2911         }
2912         switch (oe->type) {
2913         case OBJ_TREE:  /* easy case. */
2914                 return oe;
2915         case OBJ_COMMIT:
2916         case OBJ_TAG:
2917                 break;
2918         default:
2919                 die("Not a treeish: %s", command_buf.buf);
2920         }
2922         if (oe->pack_id != MAX_PACK_ID) {       /* in a pack being written */
2923                 buf = gfi_unpack_entry(oe, &size);
2924         } else {
2925                 enum object_type unused;
2926                 buf = read_sha1_file(sha1, &unused, &size);
2927         }
2928         if (!buf)
2929                 die("Can't load object %s", sha1_to_hex(sha1));
2931         /* Peel one layer. */
2932         switch (oe->type) {
2933         case OBJ_TAG:
2934                 if (size < 40 + strlen("object ") ||
2935                     get_sha1_hex(buf + strlen("object "), sha1))
2936                         die("Invalid SHA1 in tag: %s", command_buf.buf);
2937                 break;
2938         case OBJ_COMMIT:
2939                 if (size < 40 + strlen("tree ") ||
2940                     get_sha1_hex(buf + strlen("tree "), sha1))
2941                         die("Invalid SHA1 in commit: %s", command_buf.buf);
2942         }
2944         free(buf);
2945         return find_object(sha1);
2948 static struct object_entry *parse_treeish_dataref(const char **p)
2950         unsigned char sha1[20];
2951         struct object_entry *e;
2953         if (**p == ':') {       /* <mark> */
2954                 char *endptr;
2955                 e = find_mark(strtoumax(*p + 1, &endptr, 10));
2956                 if (endptr == *p + 1)
2957                         die("Invalid mark: %s", command_buf.buf);
2958                 if (!e)
2959                         die("Unknown mark: %s", command_buf.buf);
2960                 *p = endptr;
2961                 hashcpy(sha1, e->idx.sha1);
2962         } else {        /* <sha1> */
2963                 if (get_sha1_hex(*p, sha1))
2964                         die("Invalid SHA1: %s", command_buf.buf);
2965                 e = find_object(sha1);
2966                 *p += 40;
2967         }
2969         while (!e || e->type != OBJ_TREE)
2970                 e = dereference(e, sha1);
2971         return e;
2974 static void print_ls(int mode, const unsigned char *sha1, const char *path)
2976         static struct strbuf line = STRBUF_INIT;
2978         /* See show_tree(). */
2979         const char *type =
2980                 S_ISGITLINK(mode) ? commit_type :
2981                 S_ISDIR(mode) ? tree_type :
2982                 blob_type;
2984         if (!mode) {
2985                 /* missing SP path LF */
2986                 strbuf_reset(&line);
2987                 strbuf_addstr(&line, "missing ");
2988                 quote_c_style(path, &line, NULL, 0);
2989                 strbuf_addch(&line, '\n');
2990         } else {
2991                 /* mode SP type SP object_name TAB path LF */
2992                 strbuf_reset(&line);
2993                 strbuf_addf(&line, "%06o %s %s\t",
2994                                 mode & ~NO_DELTA, type, sha1_to_hex(sha1));
2995                 quote_c_style(path, &line, NULL, 0);
2996                 strbuf_addch(&line, '\n');
2997         }
2998         cat_blob_write(line.buf, line.len);
3001 static void parse_ls(struct branch *b)
3003         const char *p;
3004         struct tree_entry *root = NULL;
3005         struct tree_entry leaf = {NULL};
3007         /* ls SP (<treeish> SP)? <path> */
3008         p = command_buf.buf + strlen("ls ");
3009         if (*p == '"') {
3010                 if (!b)
3011                         die("Not in a commit: %s", command_buf.buf);
3012                 root = &b->branch_tree;
3013         } else {
3014                 struct object_entry *e = parse_treeish_dataref(&p);
3015                 root = new_tree_entry();
3016                 hashcpy(root->versions[1].sha1, e->idx.sha1);
3017                 load_tree(root);
3018                 if (*p++ != ' ')
3019                         die("Missing space after tree-ish: %s", command_buf.buf);
3020         }
3021         if (*p == '"') {
3022                 static struct strbuf uq = STRBUF_INIT;
3023                 const char *endp;
3024                 strbuf_reset(&uq);
3025                 if (unquote_c_style(&uq, p, &endp))
3026                         die("Invalid path: %s", command_buf.buf);
3027                 if (*endp)
3028                         die("Garbage after path in: %s", command_buf.buf);
3029                 p = uq.buf;
3030         }
3031         tree_content_get(root, p, &leaf);
3032         /*
3033          * A directory in preparation would have a sha1 of zero
3034          * until it is saved.  Save, for simplicity.
3035          */
3036         if (S_ISDIR(leaf.versions[1].mode))
3037                 store_tree(&leaf);
3039         print_ls(leaf.versions[1].mode, leaf.versions[1].sha1, p);
3040         if (!b || root != &b->branch_tree)
3041                 release_tree_entry(root);
3044 static void checkpoint(void)
3046         checkpoint_requested = 0;
3047         if (object_count) {
3048                 cycle_packfile();
3049                 dump_branches();
3050                 dump_tags();
3051                 dump_marks();
3052         }
3055 static void parse_checkpoint(void)
3057         checkpoint_requested = 1;
3058         skip_optional_lf();
3061 static void parse_progress(void)
3063         fwrite(command_buf.buf, 1, command_buf.len, stdout);
3064         fputc('\n', stdout);
3065         fflush(stdout);
3066         skip_optional_lf();
3069 static char* make_fast_import_path(const char *path)
3071         struct strbuf abs_path = STRBUF_INIT;
3073         if (!relative_marks_paths || is_absolute_path(path))
3074                 return xstrdup(path);
3075         strbuf_addf(&abs_path, "%s/info/fast-import/%s", get_git_dir(), path);
3076         return strbuf_detach(&abs_path, NULL);
3079 static void option_import_marks(const char *marks,
3080                                         int from_stream, int ignore_missing)
3082         if (import_marks_file) {
3083                 if (from_stream)
3084                         die("Only one import-marks command allowed per stream");
3086                 /* read previous mark file */
3087                 if(!import_marks_file_from_stream)
3088                         read_marks();
3089         }
3091         import_marks_file = make_fast_import_path(marks);
3092         safe_create_leading_directories_const(import_marks_file);
3093         import_marks_file_from_stream = from_stream;
3094         import_marks_file_ignore_missing = ignore_missing;
3097 static void option_date_format(const char *fmt)
3099         if (!strcmp(fmt, "raw"))
3100                 whenspec = WHENSPEC_RAW;
3101         else if (!strcmp(fmt, "rfc2822"))
3102                 whenspec = WHENSPEC_RFC2822;
3103         else if (!strcmp(fmt, "now"))
3104                 whenspec = WHENSPEC_NOW;
3105         else
3106                 die("unknown --date-format argument %s", fmt);
3109 static unsigned long ulong_arg(const char *option, const char *arg)
3111         char *endptr;
3112         unsigned long rv = strtoul(arg, &endptr, 0);
3113         if (strchr(arg, '-') || endptr == arg || *endptr)
3114                 die("%s: argument must be a non-negative integer", option);
3115         return rv;
3118 static void option_depth(const char *depth)
3120         max_depth = ulong_arg("--depth", depth);
3121         if (max_depth > MAX_DEPTH)
3122                 die("--depth cannot exceed %u", MAX_DEPTH);
3125 static void option_active_branches(const char *branches)
3127         max_active_branches = ulong_arg("--active-branches", branches);
3130 static void option_export_marks(const char *marks)
3132         export_marks_file = make_fast_import_path(marks);
3133         safe_create_leading_directories_const(export_marks_file);
3136 static void option_cat_blob_fd(const char *fd)
3138         unsigned long n = ulong_arg("--cat-blob-fd", fd);
3139         if (n > (unsigned long) INT_MAX)
3140                 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3141         cat_blob_fd = (int) n;
3144 static void option_export_pack_edges(const char *edges)
3146         if (pack_edges)
3147                 fclose(pack_edges);
3148         pack_edges = fopen(edges, "a");
3149         if (!pack_edges)
3150                 die_errno("Cannot open '%s'", edges);
3153 static int parse_one_option(const char *option)
3155         if (!prefixcmp(option, "max-pack-size=")) {
3156                 unsigned long v;
3157                 if (!git_parse_ulong(option + 14, &v))
3158                         return 0;
3159                 if (v < 8192) {
3160                         warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3161                         v *= 1024 * 1024;
3162                 } else if (v < 1024 * 1024) {
3163                         warning("minimum max-pack-size is 1 MiB");
3164                         v = 1024 * 1024;
3165                 }
3166                 max_packsize = v;
3167         } else if (!prefixcmp(option, "big-file-threshold=")) {
3168                 unsigned long v;
3169                 if (!git_parse_ulong(option + 19, &v))
3170                         return 0;
3171                 big_file_threshold = v;
3172         } else if (!prefixcmp(option, "depth=")) {
3173                 option_depth(option + 6);
3174         } else if (!prefixcmp(option, "active-branches=")) {
3175                 option_active_branches(option + 16);
3176         } else if (!prefixcmp(option, "export-pack-edges=")) {
3177                 option_export_pack_edges(option + 18);
3178         } else if (!prefixcmp(option, "quiet")) {
3179                 show_stats = 0;
3180         } else if (!prefixcmp(option, "stats")) {
3181                 show_stats = 1;
3182         } else {
3183                 return 0;
3184         }
3186         return 1;
3189 static int parse_one_feature(const char *feature, int from_stream)
3191         if (!prefixcmp(feature, "date-format=")) {
3192                 option_date_format(feature + 12);
3193         } else if (!prefixcmp(feature, "import-marks=")) {
3194                 option_import_marks(feature + 13, from_stream, 0);
3195         } else if (!prefixcmp(feature, "import-marks-if-exists=")) {
3196                 option_import_marks(feature + strlen("import-marks-if-exists="),
3197                                         from_stream, 1);
3198         } else if (!prefixcmp(feature, "export-marks=")) {
3199                 option_export_marks(feature + 13);
3200         } else if (!strcmp(feature, "cat-blob")) {
3201                 ; /* Don't die - this feature is supported */
3202         } else if (!strcmp(feature, "relative-marks")) {
3203                 relative_marks_paths = 1;
3204         } else if (!strcmp(feature, "no-relative-marks")) {
3205                 relative_marks_paths = 0;
3206         } else if (!strcmp(feature, "done")) {
3207                 require_explicit_termination = 1;
3208         } else if (!strcmp(feature, "force")) {
3209                 force_update = 1;
3210         } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3211                 ; /* do nothing; we have the feature */
3212         } else {
3213                 return 0;
3214         }
3216         return 1;
3219 static void parse_feature(void)
3221         char *feature = command_buf.buf + 8;
3223         if (seen_data_command)
3224                 die("Got feature command '%s' after data command", feature);
3226         if (parse_one_feature(feature, 1))
3227                 return;
3229         die("This version of fast-import does not support feature %s.", feature);
3232 static void parse_option(void)
3234         char *option = command_buf.buf + 11;
3236         if (seen_data_command)
3237                 die("Got option command '%s' after data command", option);
3239         if (parse_one_option(option))
3240                 return;
3242         die("This version of fast-import does not support option: %s", option);
3245 static int git_pack_config(const char *k, const char *v, void *cb)
3247         if (!strcmp(k, "pack.depth")) {
3248                 max_depth = git_config_int(k, v);
3249                 if (max_depth > MAX_DEPTH)
3250                         max_depth = MAX_DEPTH;
3251                 return 0;
3252         }
3253         if (!strcmp(k, "pack.compression")) {
3254                 int level = git_config_int(k, v);
3255                 if (level == -1)
3256                         level = Z_DEFAULT_COMPRESSION;
3257                 else if (level < 0 || level > Z_BEST_COMPRESSION)
3258                         die("bad pack compression level %d", level);
3259                 pack_compression_level = level;
3260                 pack_compression_seen = 1;
3261                 return 0;
3262         }
3263         if (!strcmp(k, "pack.indexversion")) {
3264                 pack_idx_opts.version = git_config_int(k, v);
3265                 if (pack_idx_opts.version > 2)
3266                         die("bad pack.indexversion=%"PRIu32,
3267                             pack_idx_opts.version);
3268                 return 0;
3269         }
3270         if (!strcmp(k, "pack.packsizelimit")) {
3271                 max_packsize = git_config_ulong(k, v);
3272                 return 0;
3273         }
3274         return git_default_config(k, v, cb);
3277 static const char fast_import_usage[] =
3278 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3280 static void parse_argv(void)
3282         unsigned int i;
3284         for (i = 1; i < global_argc; i++) {
3285                 const char *a = global_argv[i];
3287                 if (*a != '-' || !strcmp(a, "--"))
3288                         break;
3290                 if (parse_one_option(a + 2))
3291                         continue;
3293                 if (parse_one_feature(a + 2, 0))
3294                         continue;
3296                 if (!prefixcmp(a + 2, "cat-blob-fd=")) {
3297                         option_cat_blob_fd(a + 2 + strlen("cat-blob-fd="));
3298                         continue;
3299                 }
3301                 die("unknown option %s", a);
3302         }
3303         if (i != global_argc)
3304                 usage(fast_import_usage);
3306         seen_data_command = 1;
3307         if (import_marks_file)
3308                 read_marks();
3311 int main(int argc, const char **argv)
3313         unsigned int i;
3315         git_extract_argv0_path(argv[0]);
3317         if (argc == 2 && !strcmp(argv[1], "-h"))
3318                 usage(fast_import_usage);
3320         setup_git_directory();
3321         reset_pack_idx_option(&pack_idx_opts);
3322         git_config(git_pack_config, NULL);
3323         if (!pack_compression_seen && core_compression_seen)
3324                 pack_compression_level = core_compression_level;
3326         alloc_objects(object_entry_alloc);
3327         strbuf_init(&command_buf, 0);
3328         atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3329         branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3330         avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3331         marks = pool_calloc(1, sizeof(struct mark_set));
3333         global_argc = argc;
3334         global_argv = argv;
3336         rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
3337         for (i = 0; i < (cmd_save - 1); i++)
3338                 rc_free[i].next = &rc_free[i + 1];
3339         rc_free[cmd_save - 1].next = NULL;
3341         prepare_packed_git();
3342         start_packfile();
3343         set_die_routine(die_nicely);
3344         set_checkpoint_signal();
3345         while (read_next_command() != EOF) {
3346                 if (!strcmp("blob", command_buf.buf))
3347                         parse_new_blob();
3348                 else if (!prefixcmp(command_buf.buf, "ls "))
3349                         parse_ls(NULL);
3350                 else if (!prefixcmp(command_buf.buf, "commit "))
3351                         parse_new_commit();
3352                 else if (!prefixcmp(command_buf.buf, "tag "))
3353                         parse_new_tag();
3354                 else if (!prefixcmp(command_buf.buf, "reset "))
3355                         parse_reset_branch();
3356                 else if (!strcmp("checkpoint", command_buf.buf))
3357                         parse_checkpoint();
3358                 else if (!strcmp("done", command_buf.buf))
3359                         break;
3360                 else if (!prefixcmp(command_buf.buf, "progress "))
3361                         parse_progress();
3362                 else if (!prefixcmp(command_buf.buf, "feature "))
3363                         parse_feature();
3364                 else if (!prefixcmp(command_buf.buf, "option git "))
3365                         parse_option();
3366                 else if (!prefixcmp(command_buf.buf, "option "))
3367                         /* ignore non-git options*/;
3368                 else
3369                         die("Unsupported command: %s", command_buf.buf);
3371                 if (checkpoint_requested)
3372                         checkpoint();
3373         }
3375         /* argv hasn't been parsed yet, do so */
3376         if (!seen_data_command)
3377                 parse_argv();
3379         if (require_explicit_termination && feof(stdin))
3380                 die("stream ends early");
3382         end_packfile();
3384         dump_branches();
3385         dump_tags();
3386         unkeep_all_packs();
3387         dump_marks();
3389         if (pack_edges)
3390                 fclose(pack_edges);
3392         if (show_stats) {
3393                 uintmax_t total_count = 0, duplicate_count = 0;
3394                 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3395                         total_count += object_count_by_type[i];
3396                 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3397                         duplicate_count += duplicate_count_by_type[i];
3399                 fprintf(stderr, "%s statistics:\n", argv[0]);
3400                 fprintf(stderr, "---------------------------------------------------------------------\n");
3401                 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3402                 fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
3403                 fprintf(stderr, "      blobs  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB], delta_count_by_type[OBJ_BLOB], delta_count_attempts_by_type[OBJ_BLOB]);
3404                 fprintf(stderr, "      trees  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE], delta_count_by_type[OBJ_TREE], delta_count_attempts_by_type[OBJ_TREE]);
3405                 fprintf(stderr, "      commits:   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT], delta_count_by_type[OBJ_COMMIT], delta_count_attempts_by_type[OBJ_COMMIT]);
3406                 fprintf(stderr, "      tags   :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG], delta_count_by_type[OBJ_TAG], delta_count_attempts_by_type[OBJ_TAG]);
3407                 fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
3408                 fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3409                 fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
3410                 fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
3411                 fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
3412                 fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3413                 fprintf(stderr, "---------------------------------------------------------------------\n");
3414                 pack_report();
3415                 fprintf(stderr, "---------------------------------------------------------------------\n");
3416                 fprintf(stderr, "\n");
3417         }
3419         return failure ? 1 : 0;