Code

tests: unset COLUMNS inherited from environment
[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 tmp_file[PATH_MAX];
859         struct packed_git *p;
860         struct pack_header hdr;
861         int pack_fd;
863         pack_fd = odb_mkstemp(tmp_file, sizeof(tmp_file),
864                               "pack/tmp_pack_XXXXXX");
865         p = xcalloc(1, sizeof(*p) + strlen(tmp_file) + 2);
866         strcpy(p->pack_name, tmp_file);
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(struct sha1file_checkpoint *checkpoint)
1148         if (sha1file_truncate(pack_file, checkpoint))
1149                 die_errno("cannot truncate pack to skip duplicate");
1150         pack_size = checkpoint->offset;
1153 static void stream_blob(uintmax_t len, unsigned char *sha1out, uintmax_t mark)
1155         size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1156         unsigned char *in_buf = xmalloc(in_sz);
1157         unsigned char *out_buf = xmalloc(out_sz);
1158         struct object_entry *e;
1159         unsigned char sha1[20];
1160         unsigned long hdrlen;
1161         off_t offset;
1162         git_SHA_CTX c;
1163         git_zstream s;
1164         struct sha1file_checkpoint checkpoint;
1165         int status = Z_OK;
1167         /* Determine if we should auto-checkpoint. */
1168         if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1169                 || (pack_size + 60 + len) < pack_size)
1170                 cycle_packfile();
1172         sha1file_checkpoint(pack_file, &checkpoint);
1173         offset = checkpoint.offset;
1175         hdrlen = snprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1176         if (out_sz <= hdrlen)
1177                 die("impossibly large object header");
1179         git_SHA1_Init(&c);
1180         git_SHA1_Update(&c, out_buf, hdrlen);
1182         crc32_begin(pack_file);
1184         memset(&s, 0, sizeof(s));
1185         git_deflate_init(&s, pack_compression_level);
1187         hdrlen = encode_in_pack_object_header(OBJ_BLOB, len, out_buf);
1188         if (out_sz <= hdrlen)
1189                 die("impossibly large object header");
1191         s.next_out = out_buf + hdrlen;
1192         s.avail_out = out_sz - hdrlen;
1194         while (status != Z_STREAM_END) {
1195                 if (0 < len && !s.avail_in) {
1196                         size_t cnt = in_sz < len ? in_sz : (size_t)len;
1197                         size_t n = fread(in_buf, 1, cnt, stdin);
1198                         if (!n && feof(stdin))
1199                                 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1201                         git_SHA1_Update(&c, in_buf, n);
1202                         s.next_in = in_buf;
1203                         s.avail_in = n;
1204                         len -= n;
1205                 }
1207                 status = git_deflate(&s, len ? 0 : Z_FINISH);
1209                 if (!s.avail_out || status == Z_STREAM_END) {
1210                         size_t n = s.next_out - out_buf;
1211                         sha1write(pack_file, out_buf, n);
1212                         pack_size += n;
1213                         s.next_out = out_buf;
1214                         s.avail_out = out_sz;
1215                 }
1217                 switch (status) {
1218                 case Z_OK:
1219                 case Z_BUF_ERROR:
1220                 case Z_STREAM_END:
1221                         continue;
1222                 default:
1223                         die("unexpected deflate failure: %d", status);
1224                 }
1225         }
1226         git_deflate_end(&s);
1227         git_SHA1_Final(sha1, &c);
1229         if (sha1out)
1230                 hashcpy(sha1out, sha1);
1232         e = insert_object(sha1);
1234         if (mark)
1235                 insert_mark(mark, e);
1237         if (e->idx.offset) {
1238                 duplicate_count_by_type[OBJ_BLOB]++;
1239                 truncate_pack(&checkpoint);
1241         } else if (find_sha1_pack(sha1, packed_git)) {
1242                 e->type = OBJ_BLOB;
1243                 e->pack_id = MAX_PACK_ID;
1244                 e->idx.offset = 1; /* just not zero! */
1245                 duplicate_count_by_type[OBJ_BLOB]++;
1246                 truncate_pack(&checkpoint);
1248         } else {
1249                 e->depth = 0;
1250                 e->type = OBJ_BLOB;
1251                 e->pack_id = pack_id;
1252                 e->idx.offset = offset;
1253                 e->idx.crc32 = crc32_end(pack_file);
1254                 object_count++;
1255                 object_count_by_type[OBJ_BLOB]++;
1256         }
1258         free(in_buf);
1259         free(out_buf);
1262 /* All calls must be guarded by find_object() or find_mark() to
1263  * ensure the 'struct object_entry' passed was written by this
1264  * process instance.  We unpack the entry by the offset, avoiding
1265  * the need for the corresponding .idx file.  This unpacking rule
1266  * works because we only use OBJ_REF_DELTA within the packfiles
1267  * created by fast-import.
1268  *
1269  * oe must not be NULL.  Such an oe usually comes from giving
1270  * an unknown SHA-1 to find_object() or an undefined mark to
1271  * find_mark().  Callers must test for this condition and use
1272  * the standard read_sha1_file() when it happens.
1273  *
1274  * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1275  * find_mark(), where the mark was reloaded from an existing marks
1276  * file and is referencing an object that this fast-import process
1277  * instance did not write out to a packfile.  Callers must test for
1278  * this condition and use read_sha1_file() instead.
1279  */
1280 static void *gfi_unpack_entry(
1281         struct object_entry *oe,
1282         unsigned long *sizep)
1284         enum object_type type;
1285         struct packed_git *p = all_packs[oe->pack_id];
1286         if (p == pack_data && p->pack_size < (pack_size + 20)) {
1287                 /* The object is stored in the packfile we are writing to
1288                  * and we have modified it since the last time we scanned
1289                  * back to read a previously written object.  If an old
1290                  * window covered [p->pack_size, p->pack_size + 20) its
1291                  * data is stale and is not valid.  Closing all windows
1292                  * and updating the packfile length ensures we can read
1293                  * the newly written data.
1294                  */
1295                 close_pack_windows(p);
1296                 sha1flush(pack_file);
1298                 /* We have to offer 20 bytes additional on the end of
1299                  * the packfile as the core unpacker code assumes the
1300                  * footer is present at the file end and must promise
1301                  * at least 20 bytes within any window it maps.  But
1302                  * we don't actually create the footer here.
1303                  */
1304                 p->pack_size = pack_size + 20;
1305         }
1306         return unpack_entry(p, oe->idx.offset, &type, sizep);
1309 static const char *get_mode(const char *str, uint16_t *modep)
1311         unsigned char c;
1312         uint16_t mode = 0;
1314         while ((c = *str++) != ' ') {
1315                 if (c < '0' || c > '7')
1316                         return NULL;
1317                 mode = (mode << 3) + (c - '0');
1318         }
1319         *modep = mode;
1320         return str;
1323 static void load_tree(struct tree_entry *root)
1325         unsigned char *sha1 = root->versions[1].sha1;
1326         struct object_entry *myoe;
1327         struct tree_content *t;
1328         unsigned long size;
1329         char *buf;
1330         const char *c;
1332         root->tree = t = new_tree_content(8);
1333         if (is_null_sha1(sha1))
1334                 return;
1336         myoe = find_object(sha1);
1337         if (myoe && myoe->pack_id != MAX_PACK_ID) {
1338                 if (myoe->type != OBJ_TREE)
1339                         die("Not a tree: %s", sha1_to_hex(sha1));
1340                 t->delta_depth = myoe->depth;
1341                 buf = gfi_unpack_entry(myoe, &size);
1342                 if (!buf)
1343                         die("Can't load tree %s", sha1_to_hex(sha1));
1344         } else {
1345                 enum object_type type;
1346                 buf = read_sha1_file(sha1, &type, &size);
1347                 if (!buf || type != OBJ_TREE)
1348                         die("Can't load tree %s", sha1_to_hex(sha1));
1349         }
1351         c = buf;
1352         while (c != (buf + size)) {
1353                 struct tree_entry *e = new_tree_entry();
1355                 if (t->entry_count == t->entry_capacity)
1356                         root->tree = t = grow_tree_content(t, t->entry_count);
1357                 t->entries[t->entry_count++] = e;
1359                 e->tree = NULL;
1360                 c = get_mode(c, &e->versions[1].mode);
1361                 if (!c)
1362                         die("Corrupt mode in %s", sha1_to_hex(sha1));
1363                 e->versions[0].mode = e->versions[1].mode;
1364                 e->name = to_atom(c, strlen(c));
1365                 c += e->name->str_len + 1;
1366                 hashcpy(e->versions[0].sha1, (unsigned char *)c);
1367                 hashcpy(e->versions[1].sha1, (unsigned char *)c);
1368                 c += 20;
1369         }
1370         free(buf);
1373 static int tecmp0 (const void *_a, const void *_b)
1375         struct tree_entry *a = *((struct tree_entry**)_a);
1376         struct tree_entry *b = *((struct tree_entry**)_b);
1377         return base_name_compare(
1378                 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1379                 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1382 static int tecmp1 (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[1].mode,
1388                 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1391 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1393         size_t maxlen = 0;
1394         unsigned int i;
1396         if (!v)
1397                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1398         else
1399                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1401         for (i = 0; i < t->entry_count; i++) {
1402                 if (t->entries[i]->versions[v].mode)
1403                         maxlen += t->entries[i]->name->str_len + 34;
1404         }
1406         strbuf_reset(b);
1407         strbuf_grow(b, maxlen);
1408         for (i = 0; i < t->entry_count; i++) {
1409                 struct tree_entry *e = t->entries[i];
1410                 if (!e->versions[v].mode)
1411                         continue;
1412                 strbuf_addf(b, "%o %s%c",
1413                         (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1414                         e->name->str_dat, '\0');
1415                 strbuf_add(b, e->versions[v].sha1, 20);
1416         }
1419 static void store_tree(struct tree_entry *root)
1421         struct tree_content *t = root->tree;
1422         unsigned int i, j, del;
1423         struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1424         struct object_entry *le = NULL;
1426         if (!is_null_sha1(root->versions[1].sha1))
1427                 return;
1429         for (i = 0; i < t->entry_count; i++) {
1430                 if (t->entries[i]->tree)
1431                         store_tree(t->entries[i]);
1432         }
1434         if (!(root->versions[0].mode & NO_DELTA))
1435                 le = find_object(root->versions[0].sha1);
1436         if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1437                 mktree(t, 0, &old_tree);
1438                 lo.data = old_tree;
1439                 lo.offset = le->idx.offset;
1440                 lo.depth = t->delta_depth;
1441         }
1443         mktree(t, 1, &new_tree);
1444         store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1446         t->delta_depth = lo.depth;
1447         for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1448                 struct tree_entry *e = t->entries[i];
1449                 if (e->versions[1].mode) {
1450                         e->versions[0].mode = e->versions[1].mode;
1451                         hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1452                         t->entries[j++] = e;
1453                 } else {
1454                         release_tree_entry(e);
1455                         del++;
1456                 }
1457         }
1458         t->entry_count -= del;
1461 static void tree_content_replace(
1462         struct tree_entry *root,
1463         const unsigned char *sha1,
1464         const uint16_t mode,
1465         struct tree_content *newtree)
1467         if (!S_ISDIR(mode))
1468                 die("Root cannot be a non-directory");
1469         hashclr(root->versions[0].sha1);
1470         hashcpy(root->versions[1].sha1, sha1);
1471         if (root->tree)
1472                 release_tree_content_recursive(root->tree);
1473         root->tree = newtree;
1476 static int tree_content_set(
1477         struct tree_entry *root,
1478         const char *p,
1479         const unsigned char *sha1,
1480         const uint16_t mode,
1481         struct tree_content *subtree)
1483         struct tree_content *t;
1484         const char *slash1;
1485         unsigned int i, n;
1486         struct tree_entry *e;
1488         slash1 = strchr(p, '/');
1489         if (slash1)
1490                 n = slash1 - p;
1491         else
1492                 n = strlen(p);
1493         if (!n)
1494                 die("Empty path component found in input");
1495         if (!slash1 && !S_ISDIR(mode) && subtree)
1496                 die("Non-directories cannot have subtrees");
1498         if (!root->tree)
1499                 load_tree(root);
1500         t = root->tree;
1501         for (i = 0; i < t->entry_count; i++) {
1502                 e = t->entries[i];
1503                 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1504                         if (!slash1) {
1505                                 if (!S_ISDIR(mode)
1506                                                 && e->versions[1].mode == mode
1507                                                 && !hashcmp(e->versions[1].sha1, sha1))
1508                                         return 0;
1509                                 e->versions[1].mode = mode;
1510                                 hashcpy(e->versions[1].sha1, sha1);
1511                                 if (e->tree)
1512                                         release_tree_content_recursive(e->tree);
1513                                 e->tree = subtree;
1515                                 /*
1516                                  * We need to leave e->versions[0].sha1 alone
1517                                  * to avoid modifying the preimage tree used
1518                                  * when writing out the parent directory.
1519                                  * But after replacing the subdir with a
1520                                  * completely different one, it's not a good
1521                                  * delta base any more, and besides, we've
1522                                  * thrown away the tree entries needed to
1523                                  * make a delta against it.
1524                                  *
1525                                  * So let's just explicitly disable deltas
1526                                  * for the subtree.
1527                                  */
1528                                 if (S_ISDIR(e->versions[0].mode))
1529                                         e->versions[0].mode |= NO_DELTA;
1531                                 hashclr(root->versions[1].sha1);
1532                                 return 1;
1533                         }
1534                         if (!S_ISDIR(e->versions[1].mode)) {
1535                                 e->tree = new_tree_content(8);
1536                                 e->versions[1].mode = S_IFDIR;
1537                         }
1538                         if (!e->tree)
1539                                 load_tree(e);
1540                         if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1541                                 hashclr(root->versions[1].sha1);
1542                                 return 1;
1543                         }
1544                         return 0;
1545                 }
1546         }
1548         if (t->entry_count == t->entry_capacity)
1549                 root->tree = t = grow_tree_content(t, t->entry_count);
1550         e = new_tree_entry();
1551         e->name = to_atom(p, n);
1552         e->versions[0].mode = 0;
1553         hashclr(e->versions[0].sha1);
1554         t->entries[t->entry_count++] = e;
1555         if (slash1) {
1556                 e->tree = new_tree_content(8);
1557                 e->versions[1].mode = S_IFDIR;
1558                 tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1559         } else {
1560                 e->tree = subtree;
1561                 e->versions[1].mode = mode;
1562                 hashcpy(e->versions[1].sha1, sha1);
1563         }
1564         hashclr(root->versions[1].sha1);
1565         return 1;
1568 static int tree_content_remove(
1569         struct tree_entry *root,
1570         const char *p,
1571         struct tree_entry *backup_leaf)
1573         struct tree_content *t;
1574         const char *slash1;
1575         unsigned int i, n;
1576         struct tree_entry *e;
1578         slash1 = strchr(p, '/');
1579         if (slash1)
1580                 n = slash1 - p;
1581         else
1582                 n = strlen(p);
1584         if (!root->tree)
1585                 load_tree(root);
1586         t = root->tree;
1587         for (i = 0; i < t->entry_count; i++) {
1588                 e = t->entries[i];
1589                 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1590                         if (slash1 && !S_ISDIR(e->versions[1].mode))
1591                                 /*
1592                                  * If p names a file in some subdirectory, and a
1593                                  * file or symlink matching the name of the
1594                                  * parent directory of p exists, then p cannot
1595                                  * exist and need not be deleted.
1596                                  */
1597                                 return 1;
1598                         if (!slash1 || !S_ISDIR(e->versions[1].mode))
1599                                 goto del_entry;
1600                         if (!e->tree)
1601                                 load_tree(e);
1602                         if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1603                                 for (n = 0; n < e->tree->entry_count; n++) {
1604                                         if (e->tree->entries[n]->versions[1].mode) {
1605                                                 hashclr(root->versions[1].sha1);
1606                                                 return 1;
1607                                         }
1608                                 }
1609                                 backup_leaf = NULL;
1610                                 goto del_entry;
1611                         }
1612                         return 0;
1613                 }
1614         }
1615         return 0;
1617 del_entry:
1618         if (backup_leaf)
1619                 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1620         else if (e->tree)
1621                 release_tree_content_recursive(e->tree);
1622         e->tree = NULL;
1623         e->versions[1].mode = 0;
1624         hashclr(e->versions[1].sha1);
1625         hashclr(root->versions[1].sha1);
1626         return 1;
1629 static int tree_content_get(
1630         struct tree_entry *root,
1631         const char *p,
1632         struct tree_entry *leaf)
1634         struct tree_content *t;
1635         const char *slash1;
1636         unsigned int i, n;
1637         struct tree_entry *e;
1639         slash1 = strchr(p, '/');
1640         if (slash1)
1641                 n = slash1 - p;
1642         else
1643                 n = strlen(p);
1644         if (!n)
1645                 die("Empty path component found in input");
1647         if (!root->tree)
1648                 load_tree(root);
1649         t = root->tree;
1650         for (i = 0; i < t->entry_count; i++) {
1651                 e = t->entries[i];
1652                 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1653                         if (!slash1) {
1654                                 memcpy(leaf, e, sizeof(*leaf));
1655                                 if (e->tree && is_null_sha1(e->versions[1].sha1))
1656                                         leaf->tree = dup_tree_content(e->tree);
1657                                 else
1658                                         leaf->tree = NULL;
1659                                 return 1;
1660                         }
1661                         if (!S_ISDIR(e->versions[1].mode))
1662                                 return 0;
1663                         if (!e->tree)
1664                                 load_tree(e);
1665                         return tree_content_get(e, slash1 + 1, leaf);
1666                 }
1667         }
1668         return 0;
1671 static int update_branch(struct branch *b)
1673         static const char *msg = "fast-import";
1674         struct ref_lock *lock;
1675         unsigned char old_sha1[20];
1677         if (is_null_sha1(b->sha1))
1678                 return 0;
1679         if (read_ref(b->name, old_sha1))
1680                 hashclr(old_sha1);
1681         lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1682         if (!lock)
1683                 return error("Unable to lock %s", b->name);
1684         if (!force_update && !is_null_sha1(old_sha1)) {
1685                 struct commit *old_cmit, *new_cmit;
1687                 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1688                 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1689                 if (!old_cmit || !new_cmit) {
1690                         unlock_ref(lock);
1691                         return error("Branch %s is missing commits.", b->name);
1692                 }
1694                 if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1695                         unlock_ref(lock);
1696                         warning("Not updating %s"
1697                                 " (new tip %s does not contain %s)",
1698                                 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1699                         return -1;
1700                 }
1701         }
1702         if (write_ref_sha1(lock, b->sha1, msg) < 0)
1703                 return error("Unable to update %s", b->name);
1704         return 0;
1707 static void dump_branches(void)
1709         unsigned int i;
1710         struct branch *b;
1712         for (i = 0; i < branch_table_sz; i++) {
1713                 for (b = branch_table[i]; b; b = b->table_next_branch)
1714                         failure |= update_branch(b);
1715         }
1718 static void dump_tags(void)
1720         static const char *msg = "fast-import";
1721         struct tag *t;
1722         struct ref_lock *lock;
1723         char ref_name[PATH_MAX];
1725         for (t = first_tag; t; t = t->next_tag) {
1726                 sprintf(ref_name, "tags/%s", t->name);
1727                 lock = lock_ref_sha1(ref_name, NULL);
1728                 if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1729                         failure |= error("Unable to update %s", ref_name);
1730         }
1733 static void dump_marks_helper(FILE *f,
1734         uintmax_t base,
1735         struct mark_set *m)
1737         uintmax_t k;
1738         if (m->shift) {
1739                 for (k = 0; k < 1024; k++) {
1740                         if (m->data.sets[k])
1741                                 dump_marks_helper(f, base + (k << m->shift),
1742                                         m->data.sets[k]);
1743                 }
1744         } else {
1745                 for (k = 0; k < 1024; k++) {
1746                         if (m->data.marked[k])
1747                                 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1748                                         sha1_to_hex(m->data.marked[k]->idx.sha1));
1749                 }
1750         }
1753 static void dump_marks(void)
1755         static struct lock_file mark_lock;
1756         int mark_fd;
1757         FILE *f;
1759         if (!export_marks_file)
1760                 return;
1762         mark_fd = hold_lock_file_for_update(&mark_lock, export_marks_file, 0);
1763         if (mark_fd < 0) {
1764                 failure |= error("Unable to write marks file %s: %s",
1765                         export_marks_file, strerror(errno));
1766                 return;
1767         }
1769         f = fdopen(mark_fd, "w");
1770         if (!f) {
1771                 int saved_errno = errno;
1772                 rollback_lock_file(&mark_lock);
1773                 failure |= error("Unable to write marks file %s: %s",
1774                         export_marks_file, strerror(saved_errno));
1775                 return;
1776         }
1778         /*
1779          * Since the lock file was fdopen()'ed, it should not be close()'ed.
1780          * Assign -1 to the lock file descriptor so that commit_lock_file()
1781          * won't try to close() it.
1782          */
1783         mark_lock.fd = -1;
1785         dump_marks_helper(f, 0, marks);
1786         if (ferror(f) || fclose(f)) {
1787                 int saved_errno = errno;
1788                 rollback_lock_file(&mark_lock);
1789                 failure |= error("Unable to write marks file %s: %s",
1790                         export_marks_file, strerror(saved_errno));
1791                 return;
1792         }
1794         if (commit_lock_file(&mark_lock)) {
1795                 int saved_errno = errno;
1796                 rollback_lock_file(&mark_lock);
1797                 failure |= error("Unable to commit marks file %s: %s",
1798                         export_marks_file, strerror(saved_errno));
1799                 return;
1800         }
1803 static void read_marks(void)
1805         char line[512];
1806         FILE *f = fopen(import_marks_file, "r");
1807         if (f)
1808                 ;
1809         else if (import_marks_file_ignore_missing && errno == ENOENT)
1810                 return; /* Marks file does not exist */
1811         else
1812                 die_errno("cannot read '%s'", import_marks_file);
1813         while (fgets(line, sizeof(line), f)) {
1814                 uintmax_t mark;
1815                 char *end;
1816                 unsigned char sha1[20];
1817                 struct object_entry *e;
1819                 end = strchr(line, '\n');
1820                 if (line[0] != ':' || !end)
1821                         die("corrupt mark line: %s", line);
1822                 *end = 0;
1823                 mark = strtoumax(line + 1, &end, 10);
1824                 if (!mark || end == line + 1
1825                         || *end != ' ' || get_sha1(end + 1, sha1))
1826                         die("corrupt mark line: %s", line);
1827                 e = find_object(sha1);
1828                 if (!e) {
1829                         enum object_type type = sha1_object_info(sha1, NULL);
1830                         if (type < 0)
1831                                 die("object not found: %s", sha1_to_hex(sha1));
1832                         e = insert_object(sha1);
1833                         e->type = type;
1834                         e->pack_id = MAX_PACK_ID;
1835                         e->idx.offset = 1; /* just not zero! */
1836                 }
1837                 insert_mark(mark, e);
1838         }
1839         fclose(f);
1843 static int read_next_command(void)
1845         static int stdin_eof = 0;
1847         if (stdin_eof) {
1848                 unread_command_buf = 0;
1849                 return EOF;
1850         }
1852         for (;;) {
1853                 if (unread_command_buf) {
1854                         unread_command_buf = 0;
1855                 } else {
1856                         struct recent_command *rc;
1858                         strbuf_detach(&command_buf, NULL);
1859                         stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1860                         if (stdin_eof)
1861                                 return EOF;
1863                         if (!seen_data_command
1864                                 && prefixcmp(command_buf.buf, "feature ")
1865                                 && prefixcmp(command_buf.buf, "option ")) {
1866                                 parse_argv();
1867                         }
1869                         rc = rc_free;
1870                         if (rc)
1871                                 rc_free = rc->next;
1872                         else {
1873                                 rc = cmd_hist.next;
1874                                 cmd_hist.next = rc->next;
1875                                 cmd_hist.next->prev = &cmd_hist;
1876                                 free(rc->buf);
1877                         }
1879                         rc->buf = command_buf.buf;
1880                         rc->prev = cmd_tail;
1881                         rc->next = cmd_hist.prev;
1882                         rc->prev->next = rc;
1883                         cmd_tail = rc;
1884                 }
1885                 if (!prefixcmp(command_buf.buf, "cat-blob ")) {
1886                         parse_cat_blob();
1887                         continue;
1888                 }
1889                 if (command_buf.buf[0] == '#')
1890                         continue;
1891                 return 0;
1892         }
1895 static void skip_optional_lf(void)
1897         int term_char = fgetc(stdin);
1898         if (term_char != '\n' && term_char != EOF)
1899                 ungetc(term_char, stdin);
1902 static void parse_mark(void)
1904         if (!prefixcmp(command_buf.buf, "mark :")) {
1905                 next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1906                 read_next_command();
1907         }
1908         else
1909                 next_mark = 0;
1912 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1914         strbuf_reset(sb);
1916         if (prefixcmp(command_buf.buf, "data "))
1917                 die("Expected 'data n' command, found: %s", command_buf.buf);
1919         if (!prefixcmp(command_buf.buf + 5, "<<")) {
1920                 char *term = xstrdup(command_buf.buf + 5 + 2);
1921                 size_t term_len = command_buf.len - 5 - 2;
1923                 strbuf_detach(&command_buf, NULL);
1924                 for (;;) {
1925                         if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1926                                 die("EOF in data (terminator '%s' not found)", term);
1927                         if (term_len == command_buf.len
1928                                 && !strcmp(term, command_buf.buf))
1929                                 break;
1930                         strbuf_addbuf(sb, &command_buf);
1931                         strbuf_addch(sb, '\n');
1932                 }
1933                 free(term);
1934         }
1935         else {
1936                 uintmax_t len = strtoumax(command_buf.buf + 5, NULL, 10);
1937                 size_t n = 0, length = (size_t)len;
1939                 if (limit && limit < len) {
1940                         *len_res = len;
1941                         return 0;
1942                 }
1943                 if (length < len)
1944                         die("data is too large to use in this context");
1946                 while (n < length) {
1947                         size_t s = strbuf_fread(sb, length - n, stdin);
1948                         if (!s && feof(stdin))
1949                                 die("EOF in data (%lu bytes remaining)",
1950                                         (unsigned long)(length - n));
1951                         n += s;
1952                 }
1953         }
1955         skip_optional_lf();
1956         return 1;
1959 static int validate_raw_date(const char *src, char *result, int maxlen)
1961         const char *orig_src = src;
1962         char *endp;
1963         unsigned long num;
1965         errno = 0;
1967         num = strtoul(src, &endp, 10);
1968         /* NEEDSWORK: perhaps check for reasonable values? */
1969         if (errno || endp == src || *endp != ' ')
1970                 return -1;
1972         src = endp + 1;
1973         if (*src != '-' && *src != '+')
1974                 return -1;
1976         num = strtoul(src + 1, &endp, 10);
1977         if (errno || endp == src + 1 || *endp || (endp - orig_src) >= maxlen ||
1978             1400 < num)
1979                 return -1;
1981         strcpy(result, orig_src);
1982         return 0;
1985 static char *parse_ident(const char *buf)
1987         const char *ltgt;
1988         size_t name_len;
1989         char *ident;
1991         /* ensure there is a space delimiter even if there is no name */
1992         if (*buf == '<')
1993                 --buf;
1995         ltgt = buf + strcspn(buf, "<>");
1996         if (*ltgt != '<')
1997                 die("Missing < in ident string: %s", buf);
1998         if (ltgt != buf && ltgt[-1] != ' ')
1999                 die("Missing space before < in ident string: %s", buf);
2000         ltgt = ltgt + 1 + strcspn(ltgt + 1, "<>");
2001         if (*ltgt != '>')
2002                 die("Missing > in ident string: %s", buf);
2003         ltgt++;
2004         if (*ltgt != ' ')
2005                 die("Missing space after > in ident string: %s", buf);
2006         ltgt++;
2007         name_len = ltgt - buf;
2008         ident = xmalloc(name_len + 24);
2009         strncpy(ident, buf, name_len);
2011         switch (whenspec) {
2012         case WHENSPEC_RAW:
2013                 if (validate_raw_date(ltgt, ident + name_len, 24) < 0)
2014                         die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
2015                 break;
2016         case WHENSPEC_RFC2822:
2017                 if (parse_date(ltgt, ident + name_len, 24) < 0)
2018                         die("Invalid rfc2822 date \"%s\" in ident: %s", ltgt, buf);
2019                 break;
2020         case WHENSPEC_NOW:
2021                 if (strcmp("now", ltgt))
2022                         die("Date in ident must be 'now': %s", buf);
2023                 datestamp(ident + name_len, 24);
2024                 break;
2025         }
2027         return ident;
2030 static void parse_and_store_blob(
2031         struct last_object *last,
2032         unsigned char *sha1out,
2033         uintmax_t mark)
2035         static struct strbuf buf = STRBUF_INIT;
2036         uintmax_t len;
2038         if (parse_data(&buf, big_file_threshold, &len))
2039                 store_object(OBJ_BLOB, &buf, last, sha1out, mark);
2040         else {
2041                 if (last) {
2042                         strbuf_release(&last->data);
2043                         last->offset = 0;
2044                         last->depth = 0;
2045                 }
2046                 stream_blob(len, sha1out, mark);
2047                 skip_optional_lf();
2048         }
2051 static void parse_new_blob(void)
2053         read_next_command();
2054         parse_mark();
2055         parse_and_store_blob(&last_blob, NULL, next_mark);
2058 static void unload_one_branch(void)
2060         while (cur_active_branches
2061                 && cur_active_branches >= max_active_branches) {
2062                 uintmax_t min_commit = ULONG_MAX;
2063                 struct branch *e, *l = NULL, *p = NULL;
2065                 for (e = active_branches; e; e = e->active_next_branch) {
2066                         if (e->last_commit < min_commit) {
2067                                 p = l;
2068                                 min_commit = e->last_commit;
2069                         }
2070                         l = e;
2071                 }
2073                 if (p) {
2074                         e = p->active_next_branch;
2075                         p->active_next_branch = e->active_next_branch;
2076                 } else {
2077                         e = active_branches;
2078                         active_branches = e->active_next_branch;
2079                 }
2080                 e->active = 0;
2081                 e->active_next_branch = NULL;
2082                 if (e->branch_tree.tree) {
2083                         release_tree_content_recursive(e->branch_tree.tree);
2084                         e->branch_tree.tree = NULL;
2085                 }
2086                 cur_active_branches--;
2087         }
2090 static void load_branch(struct branch *b)
2092         load_tree(&b->branch_tree);
2093         if (!b->active) {
2094                 b->active = 1;
2095                 b->active_next_branch = active_branches;
2096                 active_branches = b;
2097                 cur_active_branches++;
2098                 branch_load_count++;
2099         }
2102 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2104         unsigned char fanout = 0;
2105         while ((num_notes >>= 8))
2106                 fanout++;
2107         return fanout;
2110 static void construct_path_with_fanout(const char *hex_sha1,
2111                 unsigned char fanout, char *path)
2113         unsigned int i = 0, j = 0;
2114         if (fanout >= 20)
2115                 die("Too large fanout (%u)", fanout);
2116         while (fanout) {
2117                 path[i++] = hex_sha1[j++];
2118                 path[i++] = hex_sha1[j++];
2119                 path[i++] = '/';
2120                 fanout--;
2121         }
2122         memcpy(path + i, hex_sha1 + j, 40 - j);
2123         path[i + 40 - j] = '\0';
2126 static uintmax_t do_change_note_fanout(
2127                 struct tree_entry *orig_root, struct tree_entry *root,
2128                 char *hex_sha1, unsigned int hex_sha1_len,
2129                 char *fullpath, unsigned int fullpath_len,
2130                 unsigned char fanout)
2132         struct tree_content *t = root->tree;
2133         struct tree_entry *e, leaf;
2134         unsigned int i, tmp_hex_sha1_len, tmp_fullpath_len;
2135         uintmax_t num_notes = 0;
2136         unsigned char sha1[20];
2137         char realpath[60];
2139         for (i = 0; t && i < t->entry_count; i++) {
2140                 e = t->entries[i];
2141                 tmp_hex_sha1_len = hex_sha1_len + e->name->str_len;
2142                 tmp_fullpath_len = fullpath_len;
2144                 /*
2145                  * We're interested in EITHER existing note entries (entries
2146                  * with exactly 40 hex chars in path, not including directory
2147                  * separators), OR directory entries that may contain note
2148                  * entries (with < 40 hex chars in path).
2149                  * Also, each path component in a note entry must be a multiple
2150                  * of 2 chars.
2151                  */
2152                 if (!e->versions[1].mode ||
2153                     tmp_hex_sha1_len > 40 ||
2154                     e->name->str_len % 2)
2155                         continue;
2157                 /* This _may_ be a note entry, or a subdir containing notes */
2158                 memcpy(hex_sha1 + hex_sha1_len, e->name->str_dat,
2159                        e->name->str_len);
2160                 if (tmp_fullpath_len)
2161                         fullpath[tmp_fullpath_len++] = '/';
2162                 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2163                        e->name->str_len);
2164                 tmp_fullpath_len += e->name->str_len;
2165                 fullpath[tmp_fullpath_len] = '\0';
2167                 if (tmp_hex_sha1_len == 40 && !get_sha1_hex(hex_sha1, sha1)) {
2168                         /* This is a note entry */
2169                         if (fanout == 0xff) {
2170                                 /* Counting mode, no rename */
2171                                 num_notes++;
2172                                 continue;
2173                         }
2174                         construct_path_with_fanout(hex_sha1, fanout, realpath);
2175                         if (!strcmp(fullpath, realpath)) {
2176                                 /* Note entry is in correct location */
2177                                 num_notes++;
2178                                 continue;
2179                         }
2181                         /* Rename fullpath to realpath */
2182                         if (!tree_content_remove(orig_root, fullpath, &leaf))
2183                                 die("Failed to remove path %s", fullpath);
2184                         tree_content_set(orig_root, realpath,
2185                                 leaf.versions[1].sha1,
2186                                 leaf.versions[1].mode,
2187                                 leaf.tree);
2188                 } else if (S_ISDIR(e->versions[1].mode)) {
2189                         /* This is a subdir that may contain note entries */
2190                         if (!e->tree)
2191                                 load_tree(e);
2192                         num_notes += do_change_note_fanout(orig_root, e,
2193                                 hex_sha1, tmp_hex_sha1_len,
2194                                 fullpath, tmp_fullpath_len, fanout);
2195                 }
2197                 /* The above may have reallocated the current tree_content */
2198                 t = root->tree;
2199         }
2200         return num_notes;
2203 static uintmax_t change_note_fanout(struct tree_entry *root,
2204                 unsigned char fanout)
2206         char hex_sha1[40], path[60];
2207         return do_change_note_fanout(root, root, hex_sha1, 0, path, 0, fanout);
2210 static void file_change_m(struct branch *b)
2212         const char *p = command_buf.buf + 2;
2213         static struct strbuf uq = STRBUF_INIT;
2214         const char *endp;
2215         struct object_entry *oe = oe;
2216         unsigned char sha1[20];
2217         uint16_t mode, inline_data = 0;
2219         p = get_mode(p, &mode);
2220         if (!p)
2221                 die("Corrupt mode: %s", command_buf.buf);
2222         switch (mode) {
2223         case 0644:
2224         case 0755:
2225                 mode |= S_IFREG;
2226         case S_IFREG | 0644:
2227         case S_IFREG | 0755:
2228         case S_IFLNK:
2229         case S_IFDIR:
2230         case S_IFGITLINK:
2231                 /* ok */
2232                 break;
2233         default:
2234                 die("Corrupt mode: %s", command_buf.buf);
2235         }
2237         if (*p == ':') {
2238                 char *x;
2239                 oe = find_mark(strtoumax(p + 1, &x, 10));
2240                 hashcpy(sha1, oe->idx.sha1);
2241                 p = x;
2242         } else if (!prefixcmp(p, "inline")) {
2243                 inline_data = 1;
2244                 p += 6;
2245         } else {
2246                 if (get_sha1_hex(p, sha1))
2247                         die("Invalid SHA1: %s", command_buf.buf);
2248                 oe = find_object(sha1);
2249                 p += 40;
2250         }
2251         if (*p++ != ' ')
2252                 die("Missing space after SHA1: %s", command_buf.buf);
2254         strbuf_reset(&uq);
2255         if (!unquote_c_style(&uq, p, &endp)) {
2256                 if (*endp)
2257                         die("Garbage after path in: %s", command_buf.buf);
2258                 p = uq.buf;
2259         }
2261         /* Git does not track empty, non-toplevel directories. */
2262         if (S_ISDIR(mode) && !memcmp(sha1, EMPTY_TREE_SHA1_BIN, 20) && *p) {
2263                 tree_content_remove(&b->branch_tree, p, NULL);
2264                 return;
2265         }
2267         if (S_ISGITLINK(mode)) {
2268                 if (inline_data)
2269                         die("Git links cannot be specified 'inline': %s",
2270                                 command_buf.buf);
2271                 else if (oe) {
2272                         if (oe->type != OBJ_COMMIT)
2273                                 die("Not a commit (actually a %s): %s",
2274                                         typename(oe->type), command_buf.buf);
2275                 }
2276                 /*
2277                  * Accept the sha1 without checking; it expected to be in
2278                  * another repository.
2279                  */
2280         } else if (inline_data) {
2281                 if (S_ISDIR(mode))
2282                         die("Directories cannot be specified 'inline': %s",
2283                                 command_buf.buf);
2284                 if (p != uq.buf) {
2285                         strbuf_addstr(&uq, p);
2286                         p = uq.buf;
2287                 }
2288                 read_next_command();
2289                 parse_and_store_blob(&last_blob, sha1, 0);
2290         } else {
2291                 enum object_type expected = S_ISDIR(mode) ?
2292                                                 OBJ_TREE: OBJ_BLOB;
2293                 enum object_type type = oe ? oe->type :
2294                                         sha1_object_info(sha1, NULL);
2295                 if (type < 0)
2296                         die("%s not found: %s",
2297                                         S_ISDIR(mode) ?  "Tree" : "Blob",
2298                                         command_buf.buf);
2299                 if (type != expected)
2300                         die("Not a %s (actually a %s): %s",
2301                                 typename(expected), typename(type),
2302                                 command_buf.buf);
2303         }
2305         if (!*p) {
2306                 tree_content_replace(&b->branch_tree, sha1, mode, NULL);
2307                 return;
2308         }
2309         tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
2312 static void file_change_d(struct branch *b)
2314         const char *p = command_buf.buf + 2;
2315         static struct strbuf uq = STRBUF_INIT;
2316         const char *endp;
2318         strbuf_reset(&uq);
2319         if (!unquote_c_style(&uq, p, &endp)) {
2320                 if (*endp)
2321                         die("Garbage after path in: %s", command_buf.buf);
2322                 p = uq.buf;
2323         }
2324         tree_content_remove(&b->branch_tree, p, NULL);
2327 static void file_change_cr(struct branch *b, int rename)
2329         const char *s, *d;
2330         static struct strbuf s_uq = STRBUF_INIT;
2331         static struct strbuf d_uq = STRBUF_INIT;
2332         const char *endp;
2333         struct tree_entry leaf;
2335         s = command_buf.buf + 2;
2336         strbuf_reset(&s_uq);
2337         if (!unquote_c_style(&s_uq, s, &endp)) {
2338                 if (*endp != ' ')
2339                         die("Missing space after source: %s", command_buf.buf);
2340         } else {
2341                 endp = strchr(s, ' ');
2342                 if (!endp)
2343                         die("Missing space after source: %s", command_buf.buf);
2344                 strbuf_add(&s_uq, s, endp - s);
2345         }
2346         s = s_uq.buf;
2348         endp++;
2349         if (!*endp)
2350                 die("Missing dest: %s", command_buf.buf);
2352         d = endp;
2353         strbuf_reset(&d_uq);
2354         if (!unquote_c_style(&d_uq, d, &endp)) {
2355                 if (*endp)
2356                         die("Garbage after dest in: %s", command_buf.buf);
2357                 d = d_uq.buf;
2358         }
2360         memset(&leaf, 0, sizeof(leaf));
2361         if (rename)
2362                 tree_content_remove(&b->branch_tree, s, &leaf);
2363         else
2364                 tree_content_get(&b->branch_tree, s, &leaf);
2365         if (!leaf.versions[1].mode)
2366                 die("Path %s not in branch", s);
2367         if (!*d) {      /* C "path/to/subdir" "" */
2368                 tree_content_replace(&b->branch_tree,
2369                         leaf.versions[1].sha1,
2370                         leaf.versions[1].mode,
2371                         leaf.tree);
2372                 return;
2373         }
2374         tree_content_set(&b->branch_tree, d,
2375                 leaf.versions[1].sha1,
2376                 leaf.versions[1].mode,
2377                 leaf.tree);
2380 static void note_change_n(struct branch *b, unsigned char *old_fanout)
2382         const char *p = command_buf.buf + 2;
2383         static struct strbuf uq = STRBUF_INIT;
2384         struct object_entry *oe = oe;
2385         struct branch *s;
2386         unsigned char sha1[20], commit_sha1[20];
2387         char path[60];
2388         uint16_t inline_data = 0;
2389         unsigned char new_fanout;
2391         /*
2392          * When loading a branch, we don't traverse its tree to count the real
2393          * number of notes (too expensive to do this for all non-note refs).
2394          * This means that recently loaded notes refs might incorrectly have
2395          * b->num_notes == 0, and consequently, old_fanout might be wrong.
2396          *
2397          * Fix this by traversing the tree and counting the number of notes
2398          * when b->num_notes == 0. If the notes tree is truly empty, the
2399          * calculation should not take long.
2400          */
2401         if (b->num_notes == 0 && *old_fanout == 0) {
2402                 /* Invoke change_note_fanout() in "counting mode". */
2403                 b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2404                 *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2405         }
2407         /* Now parse the notemodify command. */
2408         /* <dataref> or 'inline' */
2409         if (*p == ':') {
2410                 char *x;
2411                 oe = find_mark(strtoumax(p + 1, &x, 10));
2412                 hashcpy(sha1, oe->idx.sha1);
2413                 p = x;
2414         } else if (!prefixcmp(p, "inline")) {
2415                 inline_data = 1;
2416                 p += 6;
2417         } else {
2418                 if (get_sha1_hex(p, sha1))
2419                         die("Invalid SHA1: %s", command_buf.buf);
2420                 oe = find_object(sha1);
2421                 p += 40;
2422         }
2423         if (*p++ != ' ')
2424                 die("Missing space after SHA1: %s", command_buf.buf);
2426         /* <committish> */
2427         s = lookup_branch(p);
2428         if (s) {
2429                 if (is_null_sha1(s->sha1))
2430                         die("Can't add a note on empty branch.");
2431                 hashcpy(commit_sha1, s->sha1);
2432         } else if (*p == ':') {
2433                 uintmax_t commit_mark = strtoumax(p + 1, NULL, 10);
2434                 struct object_entry *commit_oe = find_mark(commit_mark);
2435                 if (commit_oe->type != OBJ_COMMIT)
2436                         die("Mark :%" PRIuMAX " not a commit", commit_mark);
2437                 hashcpy(commit_sha1, commit_oe->idx.sha1);
2438         } else if (!get_sha1(p, commit_sha1)) {
2439                 unsigned long size;
2440                 char *buf = read_object_with_reference(commit_sha1,
2441                         commit_type, &size, commit_sha1);
2442                 if (!buf || size < 46)
2443                         die("Not a valid commit: %s", p);
2444                 free(buf);
2445         } else
2446                 die("Invalid ref name or SHA1 expression: %s", p);
2448         if (inline_data) {
2449                 if (p != uq.buf) {
2450                         strbuf_addstr(&uq, p);
2451                         p = uq.buf;
2452                 }
2453                 read_next_command();
2454                 parse_and_store_blob(&last_blob, sha1, 0);
2455         } else if (oe) {
2456                 if (oe->type != OBJ_BLOB)
2457                         die("Not a blob (actually a %s): %s",
2458                                 typename(oe->type), command_buf.buf);
2459         } else if (!is_null_sha1(sha1)) {
2460                 enum object_type type = sha1_object_info(sha1, NULL);
2461                 if (type < 0)
2462                         die("Blob not found: %s", command_buf.buf);
2463                 if (type != OBJ_BLOB)
2464                         die("Not a blob (actually a %s): %s",
2465                             typename(type), command_buf.buf);
2466         }
2468         construct_path_with_fanout(sha1_to_hex(commit_sha1), *old_fanout, path);
2469         if (tree_content_remove(&b->branch_tree, path, NULL))
2470                 b->num_notes--;
2472         if (is_null_sha1(sha1))
2473                 return; /* nothing to insert */
2475         b->num_notes++;
2476         new_fanout = convert_num_notes_to_fanout(b->num_notes);
2477         construct_path_with_fanout(sha1_to_hex(commit_sha1), new_fanout, path);
2478         tree_content_set(&b->branch_tree, path, sha1, S_IFREG | 0644, NULL);
2481 static void file_change_deleteall(struct branch *b)
2483         release_tree_content_recursive(b->branch_tree.tree);
2484         hashclr(b->branch_tree.versions[0].sha1);
2485         hashclr(b->branch_tree.versions[1].sha1);
2486         load_tree(&b->branch_tree);
2487         b->num_notes = 0;
2490 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2492         if (!buf || size < 46)
2493                 die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2494         if (memcmp("tree ", buf, 5)
2495                 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2496                 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2497         hashcpy(b->branch_tree.versions[0].sha1,
2498                 b->branch_tree.versions[1].sha1);
2501 static void parse_from_existing(struct branch *b)
2503         if (is_null_sha1(b->sha1)) {
2504                 hashclr(b->branch_tree.versions[0].sha1);
2505                 hashclr(b->branch_tree.versions[1].sha1);
2506         } else {
2507                 unsigned long size;
2508                 char *buf;
2510                 buf = read_object_with_reference(b->sha1,
2511                         commit_type, &size, b->sha1);
2512                 parse_from_commit(b, buf, size);
2513                 free(buf);
2514         }
2517 static int parse_from(struct branch *b)
2519         const char *from;
2520         struct branch *s;
2522         if (prefixcmp(command_buf.buf, "from "))
2523                 return 0;
2525         if (b->branch_tree.tree) {
2526                 release_tree_content_recursive(b->branch_tree.tree);
2527                 b->branch_tree.tree = NULL;
2528         }
2530         from = strchr(command_buf.buf, ' ') + 1;
2531         s = lookup_branch(from);
2532         if (b == s)
2533                 die("Can't create a branch from itself: %s", b->name);
2534         else if (s) {
2535                 unsigned char *t = s->branch_tree.versions[1].sha1;
2536                 hashcpy(b->sha1, s->sha1);
2537                 hashcpy(b->branch_tree.versions[0].sha1, t);
2538                 hashcpy(b->branch_tree.versions[1].sha1, t);
2539         } else if (*from == ':') {
2540                 uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2541                 struct object_entry *oe = find_mark(idnum);
2542                 if (oe->type != OBJ_COMMIT)
2543                         die("Mark :%" PRIuMAX " not a commit", idnum);
2544                 hashcpy(b->sha1, oe->idx.sha1);
2545                 if (oe->pack_id != MAX_PACK_ID) {
2546                         unsigned long size;
2547                         char *buf = gfi_unpack_entry(oe, &size);
2548                         parse_from_commit(b, buf, size);
2549                         free(buf);
2550                 } else
2551                         parse_from_existing(b);
2552         } else if (!get_sha1(from, b->sha1))
2553                 parse_from_existing(b);
2554         else
2555                 die("Invalid ref name or SHA1 expression: %s", from);
2557         read_next_command();
2558         return 1;
2561 static struct hash_list *parse_merge(unsigned int *count)
2563         struct hash_list *list = NULL, *n, *e = e;
2564         const char *from;
2565         struct branch *s;
2567         *count = 0;
2568         while (!prefixcmp(command_buf.buf, "merge ")) {
2569                 from = strchr(command_buf.buf, ' ') + 1;
2570                 n = xmalloc(sizeof(*n));
2571                 s = lookup_branch(from);
2572                 if (s)
2573                         hashcpy(n->sha1, s->sha1);
2574                 else if (*from == ':') {
2575                         uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2576                         struct object_entry *oe = find_mark(idnum);
2577                         if (oe->type != OBJ_COMMIT)
2578                                 die("Mark :%" PRIuMAX " not a commit", idnum);
2579                         hashcpy(n->sha1, oe->idx.sha1);
2580                 } else if (!get_sha1(from, n->sha1)) {
2581                         unsigned long size;
2582                         char *buf = read_object_with_reference(n->sha1,
2583                                 commit_type, &size, n->sha1);
2584                         if (!buf || size < 46)
2585                                 die("Not a valid commit: %s", from);
2586                         free(buf);
2587                 } else
2588                         die("Invalid ref name or SHA1 expression: %s", from);
2590                 n->next = NULL;
2591                 if (list)
2592                         e->next = n;
2593                 else
2594                         list = n;
2595                 e = n;
2596                 (*count)++;
2597                 read_next_command();
2598         }
2599         return list;
2602 static void parse_new_commit(void)
2604         static struct strbuf msg = STRBUF_INIT;
2605         struct branch *b;
2606         char *sp;
2607         char *author = NULL;
2608         char *committer = NULL;
2609         struct hash_list *merge_list = NULL;
2610         unsigned int merge_count;
2611         unsigned char prev_fanout, new_fanout;
2613         /* Obtain the branch name from the rest of our command */
2614         sp = strchr(command_buf.buf, ' ') + 1;
2615         b = lookup_branch(sp);
2616         if (!b)
2617                 b = new_branch(sp);
2619         read_next_command();
2620         parse_mark();
2621         if (!prefixcmp(command_buf.buf, "author ")) {
2622                 author = parse_ident(command_buf.buf + 7);
2623                 read_next_command();
2624         }
2625         if (!prefixcmp(command_buf.buf, "committer ")) {
2626                 committer = parse_ident(command_buf.buf + 10);
2627                 read_next_command();
2628         }
2629         if (!committer)
2630                 die("Expected committer but didn't get one");
2631         parse_data(&msg, 0, NULL);
2632         read_next_command();
2633         parse_from(b);
2634         merge_list = parse_merge(&merge_count);
2636         /* ensure the branch is active/loaded */
2637         if (!b->branch_tree.tree || !max_active_branches) {
2638                 unload_one_branch();
2639                 load_branch(b);
2640         }
2642         prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2644         /* file_change* */
2645         while (command_buf.len > 0) {
2646                 if (!prefixcmp(command_buf.buf, "M "))
2647                         file_change_m(b);
2648                 else if (!prefixcmp(command_buf.buf, "D "))
2649                         file_change_d(b);
2650                 else if (!prefixcmp(command_buf.buf, "R "))
2651                         file_change_cr(b, 1);
2652                 else if (!prefixcmp(command_buf.buf, "C "))
2653                         file_change_cr(b, 0);
2654                 else if (!prefixcmp(command_buf.buf, "N "))
2655                         note_change_n(b, &prev_fanout);
2656                 else if (!strcmp("deleteall", command_buf.buf))
2657                         file_change_deleteall(b);
2658                 else if (!prefixcmp(command_buf.buf, "ls "))
2659                         parse_ls(b);
2660                 else {
2661                         unread_command_buf = 1;
2662                         break;
2663                 }
2664                 if (read_next_command() == EOF)
2665                         break;
2666         }
2668         new_fanout = convert_num_notes_to_fanout(b->num_notes);
2669         if (new_fanout != prev_fanout)
2670                 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2672         /* build the tree and the commit */
2673         store_tree(&b->branch_tree);
2674         hashcpy(b->branch_tree.versions[0].sha1,
2675                 b->branch_tree.versions[1].sha1);
2677         strbuf_reset(&new_data);
2678         strbuf_addf(&new_data, "tree %s\n",
2679                 sha1_to_hex(b->branch_tree.versions[1].sha1));
2680         if (!is_null_sha1(b->sha1))
2681                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2682         while (merge_list) {
2683                 struct hash_list *next = merge_list->next;
2684                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2685                 free(merge_list);
2686                 merge_list = next;
2687         }
2688         strbuf_addf(&new_data,
2689                 "author %s\n"
2690                 "committer %s\n"
2691                 "\n",
2692                 author ? author : committer, committer);
2693         strbuf_addbuf(&new_data, &msg);
2694         free(author);
2695         free(committer);
2697         if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2698                 b->pack_id = pack_id;
2699         b->last_commit = object_count_by_type[OBJ_COMMIT];
2702 static void parse_new_tag(void)
2704         static struct strbuf msg = STRBUF_INIT;
2705         char *sp;
2706         const char *from;
2707         char *tagger;
2708         struct branch *s;
2709         struct tag *t;
2710         uintmax_t from_mark = 0;
2711         unsigned char sha1[20];
2712         enum object_type type;
2714         /* Obtain the new tag name from the rest of our command */
2715         sp = strchr(command_buf.buf, ' ') + 1;
2716         t = pool_alloc(sizeof(struct tag));
2717         memset(t, 0, sizeof(struct tag));
2718         t->name = pool_strdup(sp);
2719         if (last_tag)
2720                 last_tag->next_tag = t;
2721         else
2722                 first_tag = t;
2723         last_tag = t;
2724         read_next_command();
2726         /* from ... */
2727         if (prefixcmp(command_buf.buf, "from "))
2728                 die("Expected from command, got %s", command_buf.buf);
2729         from = strchr(command_buf.buf, ' ') + 1;
2730         s = lookup_branch(from);
2731         if (s) {
2732                 if (is_null_sha1(s->sha1))
2733                         die("Can't tag an empty branch.");
2734                 hashcpy(sha1, s->sha1);
2735                 type = OBJ_COMMIT;
2736         } else if (*from == ':') {
2737                 struct object_entry *oe;
2738                 from_mark = strtoumax(from + 1, NULL, 10);
2739                 oe = find_mark(from_mark);
2740                 type = oe->type;
2741                 hashcpy(sha1, oe->idx.sha1);
2742         } else if (!get_sha1(from, sha1)) {
2743                 struct object_entry *oe = find_object(sha1);
2744                 if (!oe) {
2745                         type = sha1_object_info(sha1, NULL);
2746                         if (type < 0)
2747                                 die("Not a valid object: %s", from);
2748                 } else
2749                         type = oe->type;
2750         } else
2751                 die("Invalid ref name or SHA1 expression: %s", from);
2752         read_next_command();
2754         /* tagger ... */
2755         if (!prefixcmp(command_buf.buf, "tagger ")) {
2756                 tagger = parse_ident(command_buf.buf + 7);
2757                 read_next_command();
2758         } else
2759                 tagger = NULL;
2761         /* tag payload/message */
2762         parse_data(&msg, 0, NULL);
2764         /* build the tag object */
2765         strbuf_reset(&new_data);
2767         strbuf_addf(&new_data,
2768                     "object %s\n"
2769                     "type %s\n"
2770                     "tag %s\n",
2771                     sha1_to_hex(sha1), typename(type), t->name);
2772         if (tagger)
2773                 strbuf_addf(&new_data,
2774                             "tagger %s\n", tagger);
2775         strbuf_addch(&new_data, '\n');
2776         strbuf_addbuf(&new_data, &msg);
2777         free(tagger);
2779         if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2780                 t->pack_id = MAX_PACK_ID;
2781         else
2782                 t->pack_id = pack_id;
2785 static void parse_reset_branch(void)
2787         struct branch *b;
2788         char *sp;
2790         /* Obtain the branch name from the rest of our command */
2791         sp = strchr(command_buf.buf, ' ') + 1;
2792         b = lookup_branch(sp);
2793         if (b) {
2794                 hashclr(b->sha1);
2795                 hashclr(b->branch_tree.versions[0].sha1);
2796                 hashclr(b->branch_tree.versions[1].sha1);
2797                 if (b->branch_tree.tree) {
2798                         release_tree_content_recursive(b->branch_tree.tree);
2799                         b->branch_tree.tree = NULL;
2800                 }
2801         }
2802         else
2803                 b = new_branch(sp);
2804         read_next_command();
2805         parse_from(b);
2806         if (command_buf.len > 0)
2807                 unread_command_buf = 1;
2810 static void cat_blob_write(const char *buf, unsigned long size)
2812         if (write_in_full(cat_blob_fd, buf, size) != size)
2813                 die_errno("Write to frontend failed");
2816 static void cat_blob(struct object_entry *oe, unsigned char sha1[20])
2818         struct strbuf line = STRBUF_INIT;
2819         unsigned long size;
2820         enum object_type type = 0;
2821         char *buf;
2823         if (!oe || oe->pack_id == MAX_PACK_ID) {
2824                 buf = read_sha1_file(sha1, &type, &size);
2825         } else {
2826                 type = oe->type;
2827                 buf = gfi_unpack_entry(oe, &size);
2828         }
2830         /*
2831          * Output based on batch_one_object() from cat-file.c.
2832          */
2833         if (type <= 0) {
2834                 strbuf_reset(&line);
2835                 strbuf_addf(&line, "%s missing\n", sha1_to_hex(sha1));
2836                 cat_blob_write(line.buf, line.len);
2837                 strbuf_release(&line);
2838                 free(buf);
2839                 return;
2840         }
2841         if (!buf)
2842                 die("Can't read object %s", sha1_to_hex(sha1));
2843         if (type != OBJ_BLOB)
2844                 die("Object %s is a %s but a blob was expected.",
2845                     sha1_to_hex(sha1), typename(type));
2846         strbuf_reset(&line);
2847         strbuf_addf(&line, "%s %s %lu\n", sha1_to_hex(sha1),
2848                                                 typename(type), size);
2849         cat_blob_write(line.buf, line.len);
2850         strbuf_release(&line);
2851         cat_blob_write(buf, size);
2852         cat_blob_write("\n", 1);
2853         if (oe && oe->pack_id == pack_id) {
2854                 last_blob.offset = oe->idx.offset;
2855                 strbuf_attach(&last_blob.data, buf, size, size);
2856                 last_blob.depth = oe->depth;
2857         } else
2858                 free(buf);
2861 static void parse_cat_blob(void)
2863         const char *p;
2864         struct object_entry *oe = oe;
2865         unsigned char sha1[20];
2867         /* cat-blob SP <object> LF */
2868         p = command_buf.buf + strlen("cat-blob ");
2869         if (*p == ':') {
2870                 char *x;
2871                 oe = find_mark(strtoumax(p + 1, &x, 10));
2872                 if (x == p + 1)
2873                         die("Invalid mark: %s", command_buf.buf);
2874                 if (!oe)
2875                         die("Unknown mark: %s", command_buf.buf);
2876                 if (*x)
2877                         die("Garbage after mark: %s", command_buf.buf);
2878                 hashcpy(sha1, oe->idx.sha1);
2879         } else {
2880                 if (get_sha1_hex(p, sha1))
2881                         die("Invalid SHA1: %s", command_buf.buf);
2882                 if (p[40])
2883                         die("Garbage after SHA1: %s", command_buf.buf);
2884                 oe = find_object(sha1);
2885         }
2887         cat_blob(oe, sha1);
2890 static struct object_entry *dereference(struct object_entry *oe,
2891                                         unsigned char sha1[20])
2893         unsigned long size;
2894         char *buf = NULL;
2895         if (!oe) {
2896                 enum object_type type = sha1_object_info(sha1, NULL);
2897                 if (type < 0)
2898                         die("object not found: %s", sha1_to_hex(sha1));
2899                 /* cache it! */
2900                 oe = insert_object(sha1);
2901                 oe->type = type;
2902                 oe->pack_id = MAX_PACK_ID;
2903                 oe->idx.offset = 1;
2904         }
2905         switch (oe->type) {
2906         case OBJ_TREE:  /* easy case. */
2907                 return oe;
2908         case OBJ_COMMIT:
2909         case OBJ_TAG:
2910                 break;
2911         default:
2912                 die("Not a treeish: %s", command_buf.buf);
2913         }
2915         if (oe->pack_id != MAX_PACK_ID) {       /* in a pack being written */
2916                 buf = gfi_unpack_entry(oe, &size);
2917         } else {
2918                 enum object_type unused;
2919                 buf = read_sha1_file(sha1, &unused, &size);
2920         }
2921         if (!buf)
2922                 die("Can't load object %s", sha1_to_hex(sha1));
2924         /* Peel one layer. */
2925         switch (oe->type) {
2926         case OBJ_TAG:
2927                 if (size < 40 + strlen("object ") ||
2928                     get_sha1_hex(buf + strlen("object "), sha1))
2929                         die("Invalid SHA1 in tag: %s", command_buf.buf);
2930                 break;
2931         case OBJ_COMMIT:
2932                 if (size < 40 + strlen("tree ") ||
2933                     get_sha1_hex(buf + strlen("tree "), sha1))
2934                         die("Invalid SHA1 in commit: %s", command_buf.buf);
2935         }
2937         free(buf);
2938         return find_object(sha1);
2941 static struct object_entry *parse_treeish_dataref(const char **p)
2943         unsigned char sha1[20];
2944         struct object_entry *e;
2946         if (**p == ':') {       /* <mark> */
2947                 char *endptr;
2948                 e = find_mark(strtoumax(*p + 1, &endptr, 10));
2949                 if (endptr == *p + 1)
2950                         die("Invalid mark: %s", command_buf.buf);
2951                 if (!e)
2952                         die("Unknown mark: %s", command_buf.buf);
2953                 *p = endptr;
2954                 hashcpy(sha1, e->idx.sha1);
2955         } else {        /* <sha1> */
2956                 if (get_sha1_hex(*p, sha1))
2957                         die("Invalid SHA1: %s", command_buf.buf);
2958                 e = find_object(sha1);
2959                 *p += 40;
2960         }
2962         while (!e || e->type != OBJ_TREE)
2963                 e = dereference(e, sha1);
2964         return e;
2967 static void print_ls(int mode, const unsigned char *sha1, const char *path)
2969         static struct strbuf line = STRBUF_INIT;
2971         /* See show_tree(). */
2972         const char *type =
2973                 S_ISGITLINK(mode) ? commit_type :
2974                 S_ISDIR(mode) ? tree_type :
2975                 blob_type;
2977         if (!mode) {
2978                 /* missing SP path LF */
2979                 strbuf_reset(&line);
2980                 strbuf_addstr(&line, "missing ");
2981                 quote_c_style(path, &line, NULL, 0);
2982                 strbuf_addch(&line, '\n');
2983         } else {
2984                 /* mode SP type SP object_name TAB path LF */
2985                 strbuf_reset(&line);
2986                 strbuf_addf(&line, "%06o %s %s\t",
2987                                 mode & ~NO_DELTA, type, sha1_to_hex(sha1));
2988                 quote_c_style(path, &line, NULL, 0);
2989                 strbuf_addch(&line, '\n');
2990         }
2991         cat_blob_write(line.buf, line.len);
2994 static void parse_ls(struct branch *b)
2996         const char *p;
2997         struct tree_entry *root = NULL;
2998         struct tree_entry leaf = {NULL};
3000         /* ls SP (<treeish> SP)? <path> */
3001         p = command_buf.buf + strlen("ls ");
3002         if (*p == '"') {
3003                 if (!b)
3004                         die("Not in a commit: %s", command_buf.buf);
3005                 root = &b->branch_tree;
3006         } else {
3007                 struct object_entry *e = parse_treeish_dataref(&p);
3008                 root = new_tree_entry();
3009                 hashcpy(root->versions[1].sha1, e->idx.sha1);
3010                 load_tree(root);
3011                 if (*p++ != ' ')
3012                         die("Missing space after tree-ish: %s", command_buf.buf);
3013         }
3014         if (*p == '"') {
3015                 static struct strbuf uq = STRBUF_INIT;
3016                 const char *endp;
3017                 strbuf_reset(&uq);
3018                 if (unquote_c_style(&uq, p, &endp))
3019                         die("Invalid path: %s", command_buf.buf);
3020                 if (*endp)
3021                         die("Garbage after path in: %s", command_buf.buf);
3022                 p = uq.buf;
3023         }
3024         tree_content_get(root, p, &leaf);
3025         /*
3026          * A directory in preparation would have a sha1 of zero
3027          * until it is saved.  Save, for simplicity.
3028          */
3029         if (S_ISDIR(leaf.versions[1].mode))
3030                 store_tree(&leaf);
3032         print_ls(leaf.versions[1].mode, leaf.versions[1].sha1, p);
3033         if (leaf.tree)
3034                 release_tree_content_recursive(leaf.tree);
3035         if (!b || root != &b->branch_tree)
3036                 release_tree_entry(root);
3039 static void checkpoint(void)
3041         checkpoint_requested = 0;
3042         if (object_count) {
3043                 cycle_packfile();
3044                 dump_branches();
3045                 dump_tags();
3046                 dump_marks();
3047         }
3050 static void parse_checkpoint(void)
3052         checkpoint_requested = 1;
3053         skip_optional_lf();
3056 static void parse_progress(void)
3058         fwrite(command_buf.buf, 1, command_buf.len, stdout);
3059         fputc('\n', stdout);
3060         fflush(stdout);
3061         skip_optional_lf();
3064 static char* make_fast_import_path(const char *path)
3066         struct strbuf abs_path = STRBUF_INIT;
3068         if (!relative_marks_paths || is_absolute_path(path))
3069                 return xstrdup(path);
3070         strbuf_addf(&abs_path, "%s/info/fast-import/%s", get_git_dir(), path);
3071         return strbuf_detach(&abs_path, NULL);
3074 static void option_import_marks(const char *marks,
3075                                         int from_stream, int ignore_missing)
3077         if (import_marks_file) {
3078                 if (from_stream)
3079                         die("Only one import-marks command allowed per stream");
3081                 /* read previous mark file */
3082                 if(!import_marks_file_from_stream)
3083                         read_marks();
3084         }
3086         import_marks_file = make_fast_import_path(marks);
3087         safe_create_leading_directories_const(import_marks_file);
3088         import_marks_file_from_stream = from_stream;
3089         import_marks_file_ignore_missing = ignore_missing;
3092 static void option_date_format(const char *fmt)
3094         if (!strcmp(fmt, "raw"))
3095                 whenspec = WHENSPEC_RAW;
3096         else if (!strcmp(fmt, "rfc2822"))
3097                 whenspec = WHENSPEC_RFC2822;
3098         else if (!strcmp(fmt, "now"))
3099                 whenspec = WHENSPEC_NOW;
3100         else
3101                 die("unknown --date-format argument %s", fmt);
3104 static unsigned long ulong_arg(const char *option, const char *arg)
3106         char *endptr;
3107         unsigned long rv = strtoul(arg, &endptr, 0);
3108         if (strchr(arg, '-') || endptr == arg || *endptr)
3109                 die("%s: argument must be a non-negative integer", option);
3110         return rv;
3113 static void option_depth(const char *depth)
3115         max_depth = ulong_arg("--depth", depth);
3116         if (max_depth > MAX_DEPTH)
3117                 die("--depth cannot exceed %u", MAX_DEPTH);
3120 static void option_active_branches(const char *branches)
3122         max_active_branches = ulong_arg("--active-branches", branches);
3125 static void option_export_marks(const char *marks)
3127         export_marks_file = make_fast_import_path(marks);
3128         safe_create_leading_directories_const(export_marks_file);
3131 static void option_cat_blob_fd(const char *fd)
3133         unsigned long n = ulong_arg("--cat-blob-fd", fd);
3134         if (n > (unsigned long) INT_MAX)
3135                 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3136         cat_blob_fd = (int) n;
3139 static void option_export_pack_edges(const char *edges)
3141         if (pack_edges)
3142                 fclose(pack_edges);
3143         pack_edges = fopen(edges, "a");
3144         if (!pack_edges)
3145                 die_errno("Cannot open '%s'", edges);
3148 static int parse_one_option(const char *option)
3150         if (!prefixcmp(option, "max-pack-size=")) {
3151                 unsigned long v;
3152                 if (!git_parse_ulong(option + 14, &v))
3153                         return 0;
3154                 if (v < 8192) {
3155                         warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3156                         v *= 1024 * 1024;
3157                 } else if (v < 1024 * 1024) {
3158                         warning("minimum max-pack-size is 1 MiB");
3159                         v = 1024 * 1024;
3160                 }
3161                 max_packsize = v;
3162         } else if (!prefixcmp(option, "big-file-threshold=")) {
3163                 unsigned long v;
3164                 if (!git_parse_ulong(option + 19, &v))
3165                         return 0;
3166                 big_file_threshold = v;
3167         } else if (!prefixcmp(option, "depth=")) {
3168                 option_depth(option + 6);
3169         } else if (!prefixcmp(option, "active-branches=")) {
3170                 option_active_branches(option + 16);
3171         } else if (!prefixcmp(option, "export-pack-edges=")) {
3172                 option_export_pack_edges(option + 18);
3173         } else if (!prefixcmp(option, "quiet")) {
3174                 show_stats = 0;
3175         } else if (!prefixcmp(option, "stats")) {
3176                 show_stats = 1;
3177         } else {
3178                 return 0;
3179         }
3181         return 1;
3184 static int parse_one_feature(const char *feature, int from_stream)
3186         if (!prefixcmp(feature, "date-format=")) {
3187                 option_date_format(feature + 12);
3188         } else if (!prefixcmp(feature, "import-marks=")) {
3189                 option_import_marks(feature + 13, from_stream, 0);
3190         } else if (!prefixcmp(feature, "import-marks-if-exists=")) {
3191                 option_import_marks(feature + strlen("import-marks-if-exists="),
3192                                         from_stream, 1);
3193         } else if (!prefixcmp(feature, "export-marks=")) {
3194                 option_export_marks(feature + 13);
3195         } else if (!strcmp(feature, "cat-blob")) {
3196                 ; /* Don't die - this feature is supported */
3197         } else if (!strcmp(feature, "relative-marks")) {
3198                 relative_marks_paths = 1;
3199         } else if (!strcmp(feature, "no-relative-marks")) {
3200                 relative_marks_paths = 0;
3201         } else if (!strcmp(feature, "done")) {
3202                 require_explicit_termination = 1;
3203         } else if (!strcmp(feature, "force")) {
3204                 force_update = 1;
3205         } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3206                 ; /* do nothing; we have the feature */
3207         } else {
3208                 return 0;
3209         }
3211         return 1;
3214 static void parse_feature(void)
3216         char *feature = command_buf.buf + 8;
3218         if (seen_data_command)
3219                 die("Got feature command '%s' after data command", feature);
3221         if (parse_one_feature(feature, 1))
3222                 return;
3224         die("This version of fast-import does not support feature %s.", feature);
3227 static void parse_option(void)
3229         char *option = command_buf.buf + 11;
3231         if (seen_data_command)
3232                 die("Got option command '%s' after data command", option);
3234         if (parse_one_option(option))
3235                 return;
3237         die("This version of fast-import does not support option: %s", option);
3240 static int git_pack_config(const char *k, const char *v, void *cb)
3242         if (!strcmp(k, "pack.depth")) {
3243                 max_depth = git_config_int(k, v);
3244                 if (max_depth > MAX_DEPTH)
3245                         max_depth = MAX_DEPTH;
3246                 return 0;
3247         }
3248         if (!strcmp(k, "pack.compression")) {
3249                 int level = git_config_int(k, v);
3250                 if (level == -1)
3251                         level = Z_DEFAULT_COMPRESSION;
3252                 else if (level < 0 || level > Z_BEST_COMPRESSION)
3253                         die("bad pack compression level %d", level);
3254                 pack_compression_level = level;
3255                 pack_compression_seen = 1;
3256                 return 0;
3257         }
3258         if (!strcmp(k, "pack.indexversion")) {
3259                 pack_idx_opts.version = git_config_int(k, v);
3260                 if (pack_idx_opts.version > 2)
3261                         die("bad pack.indexversion=%"PRIu32,
3262                             pack_idx_opts.version);
3263                 return 0;
3264         }
3265         if (!strcmp(k, "pack.packsizelimit")) {
3266                 max_packsize = git_config_ulong(k, v);
3267                 return 0;
3268         }
3269         return git_default_config(k, v, cb);
3272 static const char fast_import_usage[] =
3273 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3275 static void parse_argv(void)
3277         unsigned int i;
3279         for (i = 1; i < global_argc; i++) {
3280                 const char *a = global_argv[i];
3282                 if (*a != '-' || !strcmp(a, "--"))
3283                         break;
3285                 if (parse_one_option(a + 2))
3286                         continue;
3288                 if (parse_one_feature(a + 2, 0))
3289                         continue;
3291                 if (!prefixcmp(a + 2, "cat-blob-fd=")) {
3292                         option_cat_blob_fd(a + 2 + strlen("cat-blob-fd="));
3293                         continue;
3294                 }
3296                 die("unknown option %s", a);
3297         }
3298         if (i != global_argc)
3299                 usage(fast_import_usage);
3301         seen_data_command = 1;
3302         if (import_marks_file)
3303                 read_marks();
3306 int main(int argc, const char **argv)
3308         unsigned int i;
3310         git_extract_argv0_path(argv[0]);
3312         git_setup_gettext();
3314         if (argc == 2 && !strcmp(argv[1], "-h"))
3315                 usage(fast_import_usage);
3317         setup_git_directory();
3318         reset_pack_idx_option(&pack_idx_opts);
3319         git_config(git_pack_config, NULL);
3320         if (!pack_compression_seen && core_compression_seen)
3321                 pack_compression_level = core_compression_level;
3323         alloc_objects(object_entry_alloc);
3324         strbuf_init(&command_buf, 0);
3325         atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3326         branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3327         avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3328         marks = pool_calloc(1, sizeof(struct mark_set));
3330         global_argc = argc;
3331         global_argv = argv;
3333         rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
3334         for (i = 0; i < (cmd_save - 1); i++)
3335                 rc_free[i].next = &rc_free[i + 1];
3336         rc_free[cmd_save - 1].next = NULL;
3338         prepare_packed_git();
3339         start_packfile();
3340         set_die_routine(die_nicely);
3341         set_checkpoint_signal();
3342         while (read_next_command() != EOF) {
3343                 if (!strcmp("blob", command_buf.buf))
3344                         parse_new_blob();
3345                 else if (!prefixcmp(command_buf.buf, "ls "))
3346                         parse_ls(NULL);
3347                 else if (!prefixcmp(command_buf.buf, "commit "))
3348                         parse_new_commit();
3349                 else if (!prefixcmp(command_buf.buf, "tag "))
3350                         parse_new_tag();
3351                 else if (!prefixcmp(command_buf.buf, "reset "))
3352                         parse_reset_branch();
3353                 else if (!strcmp("checkpoint", command_buf.buf))
3354                         parse_checkpoint();
3355                 else if (!strcmp("done", command_buf.buf))
3356                         break;
3357                 else if (!prefixcmp(command_buf.buf, "progress "))
3358                         parse_progress();
3359                 else if (!prefixcmp(command_buf.buf, "feature "))
3360                         parse_feature();
3361                 else if (!prefixcmp(command_buf.buf, "option git "))
3362                         parse_option();
3363                 else if (!prefixcmp(command_buf.buf, "option "))
3364                         /* ignore non-git options*/;
3365                 else
3366                         die("Unsupported command: %s", command_buf.buf);
3368                 if (checkpoint_requested)
3369                         checkpoint();
3370         }
3372         /* argv hasn't been parsed yet, do so */
3373         if (!seen_data_command)
3374                 parse_argv();
3376         if (require_explicit_termination && feof(stdin))
3377                 die("stream ends early");
3379         end_packfile();
3381         dump_branches();
3382         dump_tags();
3383         unkeep_all_packs();
3384         dump_marks();
3386         if (pack_edges)
3387                 fclose(pack_edges);
3389         if (show_stats) {
3390                 uintmax_t total_count = 0, duplicate_count = 0;
3391                 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3392                         total_count += object_count_by_type[i];
3393                 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3394                         duplicate_count += duplicate_count_by_type[i];
3396                 fprintf(stderr, "%s statistics:\n", argv[0]);
3397                 fprintf(stderr, "---------------------------------------------------------------------\n");
3398                 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3399                 fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
3400                 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]);
3401                 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]);
3402                 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]);
3403                 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]);
3404                 fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
3405                 fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3406                 fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
3407                 fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
3408                 fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
3409                 fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3410                 fprintf(stderr, "---------------------------------------------------------------------\n");
3411                 pack_report();
3412                 fprintf(stderr, "---------------------------------------------------------------------\n");
3413                 fprintf(stderr, "\n");
3414         }
3416         return failure ? 1 : 0;