Code

Merge branch 'master' into lt/apply
authorJunio C Hamano <junkio@cox.net>
Wed, 24 May 2006 21:08:30 +0000 (14:08 -0700)
committerJunio C Hamano <junkio@cox.net>
Wed, 24 May 2006 21:08:30 +0000 (14:08 -0700)
* master: (40 commits)
  Clean up sha1 file writing
  Builtin git-cat-file
  builtin format-patch: squelch content-type for 7-bit ASCII
  CMIT_FMT_EMAIL: Q-encode Subject: and display-name part of From: fields.
  add more informative error messages to git-mktag
  remove the artificial restriction tagsize < 8kb
  git-rebase: use canonical A..B syntax to format-patch
  git-format-patch: now built-in.
  fmt-patch: Support --attach
  fmt-patch: understand old <his> notation
  Teach fmt-patch about --keep-subject
  Teach fmt-patch about --numbered
  fmt-patch: implement -o <dir>
  fmt-patch: output file names to stdout
  Teach fmt-patch to write individual files.
  Use RFC2822 dates from "git fmt-patch".
  git-fmt-patch: thinkofix to show [PATCH] properly.
  rename internal format-patch wip
  Minor tweak on subject line in --pretty=email
  Tentative built-in format-patch.
  ...

1  2 
builtin-apply.c

diff --cc builtin-apply.c
index 0000000000000000000000000000000000000000,4056b9d67bc1c4e50563ceaa0ee68e94873a8e65..69c8e6a6083456cb3e9c6389d49c7723b8854442
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,2300 +1,2317 @@@
+ /*
+  * apply.c
+  *
+  * Copyright (C) Linus Torvalds, 2005
+  *
+  * This applies patches on top of some (arbitrary) version of the SCM.
+  *
+  */
+ #include <fnmatch.h>
+ #include "cache.h"
+ #include "quote.h"
+ #include "blob.h"
+ #include "delta.h"
+ #include "builtin.h"
+ //  --check turns on checking that the working tree matches the
+ //    files that are being modified, but doesn't apply the patch
+ //  --stat does just a diffstat, and doesn't actually apply
+ //  --numstat does numeric diffstat, and doesn't actually apply
+ //  --index-info shows the old and new index info for paths if available.
+ //  --index updates the cache as well.
+ //  --cached updates only the cache without ever touching the working tree.
+ //
+ static const char *prefix;
+ static int prefix_length = -1;
+ static int newfd = -1;
+ static int p_value = 1;
+ static int allow_binary_replacement = 0;
+ static int check_index = 0;
+ static int write_index = 0;
+ static int cached = 0;
+ static int diffstat = 0;
+ static int numstat = 0;
+ static int summary = 0;
+ static int check = 0;
+ static int apply = 1;
+ static int no_add = 0;
+ static int show_index_info = 0;
+ static int line_termination = '\n';
+ static unsigned long p_context = -1;
+ static const char apply_usage[] =
+ "git-apply [--stat] [--numstat] [--summary] [--check] [--index] [--cached] [--apply] [--no-add] [--index-info] [--allow-binary-replacement] [-z] [-pNUM] [-CNUM] [--whitespace=<nowarn|warn|error|error-all|strip>] <patch>...";
+ static enum whitespace_eol {
+       nowarn_whitespace,
+       warn_on_whitespace,
+       error_on_whitespace,
+       strip_whitespace,
+ } new_whitespace = warn_on_whitespace;
+ static int whitespace_error = 0;
+ static int squelch_whitespace_errors = 5;
+ static int applied_after_stripping = 0;
+ static const char *patch_input_file = NULL;
+ static void parse_whitespace_option(const char *option)
+ {
+       if (!option) {
+               new_whitespace = warn_on_whitespace;
+               return;
+       }
+       if (!strcmp(option, "warn")) {
+               new_whitespace = warn_on_whitespace;
+               return;
+       }
+       if (!strcmp(option, "nowarn")) {
+               new_whitespace = nowarn_whitespace;
+               return;
+       }
+       if (!strcmp(option, "error")) {
+               new_whitespace = error_on_whitespace;
+               return;
+       }
+       if (!strcmp(option, "error-all")) {
+               new_whitespace = error_on_whitespace;
+               squelch_whitespace_errors = 0;
+               return;
+       }
+       if (!strcmp(option, "strip")) {
+               new_whitespace = strip_whitespace;
+               return;
+       }
+       die("unrecognized whitespace option '%s'", option);
+ }
+ static void set_default_whitespace_mode(const char *whitespace_option)
+ {
+       if (!whitespace_option && !apply_default_whitespace) {
+               new_whitespace = (apply
+                                 ? warn_on_whitespace
+                                 : nowarn_whitespace);
+       }
+ }
+ /*
+  * For "diff-stat" like behaviour, we keep track of the biggest change
+  * we've seen, and the longest filename. That allows us to do simple
+  * scaling.
+  */
+ static int max_change, max_len;
+ /*
+  * Various "current state", notably line numbers and what
+  * file (and how) we're patching right now.. The "is_xxxx"
+  * things are flags, where -1 means "don't know yet".
+  */
+ static int linenr = 1;
+ struct fragment {
+       unsigned long leading, trailing;
+       unsigned long oldpos, oldlines;
+       unsigned long newpos, newlines;
+       const char *patch;
+       int size;
+       struct fragment *next;
+ };
+ struct patch {
+       char *new_name, *old_name, *def_name;
+       unsigned int old_mode, new_mode;
+       int is_rename, is_copy, is_new, is_delete, is_binary;
+ #define BINARY_DELTA_DEFLATED 1
+ #define BINARY_LITERAL_DEFLATED 2
+       unsigned long deflate_origlen;
+       int lines_added, lines_deleted;
+       int score;
+       struct fragment *fragments;
+       char *result;
+       unsigned long resultsize;
+       char old_sha1_prefix[41];
+       char new_sha1_prefix[41];
+       struct patch *next;
+ };
+ #define CHUNKSIZE (8192)
+ #define SLOP (16)
+ static void *read_patch_file(int fd, unsigned long *sizep)
+ {
+       unsigned long size = 0, alloc = CHUNKSIZE;
+       void *buffer = xmalloc(alloc);
+       for (;;) {
+               int nr = alloc - size;
+               if (nr < 1024) {
+                       alloc += CHUNKSIZE;
+                       buffer = xrealloc(buffer, alloc);
+                       nr = alloc - size;
+               }
+               nr = xread(fd, buffer + size, nr);
+               if (!nr)
+                       break;
+               if (nr < 0)
+                       die("git-apply: read returned %s", strerror(errno));
+               size += nr;
+       }
+       *sizep = size;
+       /*
+        * Make sure that we have some slop in the buffer
+        * so that we can do speculative "memcmp" etc, and
+        * see to it that it is NUL-filled.
+        */
+       if (alloc < size + SLOP)
+               buffer = xrealloc(buffer, size + SLOP);
+       memset(buffer + size, 0, SLOP);
+       return buffer;
+ }
+ static unsigned long linelen(const char *buffer, unsigned long size)
+ {
+       unsigned long len = 0;
+       while (size--) {
+               len++;
+               if (*buffer++ == '\n')
+                       break;
+       }
+       return len;
+ }
+ static int is_dev_null(const char *str)
+ {
+       return !memcmp("/dev/null", str, 9) && isspace(str[9]);
+ }
+ #define TERM_SPACE    1
+ #define TERM_TAB      2
+ static int name_terminate(const char *name, int namelen, int c, int terminate)
+ {
+       if (c == ' ' && !(terminate & TERM_SPACE))
+               return 0;
+       if (c == '\t' && !(terminate & TERM_TAB))
+               return 0;
+       return 1;
+ }
+ static char * find_name(const char *line, char *def, int p_value, int terminate)
+ {
+       int len;
+       const char *start = line;
+       char *name;
+       if (*line == '"') {
+               /* Proposed "new-style" GNU patch/diff format; see
+                * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
+                */
+               name = unquote_c_style(line, NULL);
+               if (name) {
+                       char *cp = name;
+                       while (p_value) {
+                               cp = strchr(name, '/');
+                               if (!cp)
+                                       break;
+                               cp++;
+                               p_value--;
+                       }
+                       if (cp) {
+                               /* name can later be freed, so we need
+                                * to memmove, not just return cp
+                                */
+                               memmove(name, cp, strlen(cp) + 1);
+                               free(def);
+                               return name;
+                       }
+                       else {
+                               free(name);
+                               name = NULL;
+                       }
+               }
+       }
+       for (;;) {
+               char c = *line;
+               if (isspace(c)) {
+                       if (c == '\n')
+                               break;
+                       if (name_terminate(start, line-start, c, terminate))
+                               break;
+               }
+               line++;
+               if (c == '/' && !--p_value)
+                       start = line;
+       }
+       if (!start)
+               return def;
+       len = line - start;
+       if (!len)
+               return def;
+       /*
+        * Generally we prefer the shorter name, especially
+        * if the other one is just a variation of that with
+        * something else tacked on to the end (ie "file.orig"
+        * or "file~").
+        */
+       if (def) {
+               int deflen = strlen(def);
+               if (deflen < len && !strncmp(start, def, deflen))
+                       return def;
+       }
+       name = xmalloc(len + 1);
+       memcpy(name, start, len);
+       name[len] = 0;
+       free(def);
+       return name;
+ }
+ /*
+  * Get the name etc info from the --/+++ lines of a traditional patch header
+  *
+  * NOTE! This hardcodes "-p1" behaviour in filename detection.
+  *
+  * FIXME! The end-of-filename heuristics are kind of screwy. For existing
+  * files, we can happily check the index for a match, but for creating a
+  * new file we should try to match whatever "patch" does. I have no idea.
+  */
+ static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
+ {
+       char *name;
+       first += 4;     // skip "--- "
+       second += 4;    // skip "+++ "
+       if (is_dev_null(first)) {
+               patch->is_new = 1;
+               patch->is_delete = 0;
+               name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
+               patch->new_name = name;
+       } else if (is_dev_null(second)) {
+               patch->is_new = 0;
+               patch->is_delete = 1;
+               name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
+               patch->old_name = name;
+       } else {
+               name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
+               name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
+               patch->old_name = patch->new_name = name;
+       }
+       if (!name)
+               die("unable to find filename in patch at line %d", linenr);
+ }
+ static int gitdiff_hdrend(const char *line, struct patch *patch)
+ {
+       return -1;
+ }
+ /*
+  * We're anal about diff header consistency, to make
+  * sure that we don't end up having strange ambiguous
+  * patches floating around.
+  *
+  * As a result, gitdiff_{old|new}name() will check
+  * their names against any previous information, just
+  * to make sure..
+  */
+ static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
+ {
+       if (!orig_name && !isnull)
+               return find_name(line, NULL, 1, 0);
+       if (orig_name) {
+               int len;
+               const char *name;
+               char *another;
+               name = orig_name;
+               len = strlen(name);
+               if (isnull)
+                       die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
+               another = find_name(line, NULL, 1, 0);
+               if (!another || memcmp(another, name, len))
+                       die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
+               free(another);
+               return orig_name;
+       }
+       else {
+               /* expect "/dev/null" */
+               if (memcmp("/dev/null", line, 9) || line[9] != '\n')
+                       die("git-apply: bad git-diff - expected /dev/null on line %d", linenr);
+               return NULL;
+       }
+ }
+ static int gitdiff_oldname(const char *line, struct patch *patch)
+ {
+       patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
+       return 0;
+ }
+ static int gitdiff_newname(const char *line, struct patch *patch)
+ {
+       patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
+       return 0;
+ }
+ static int gitdiff_oldmode(const char *line, struct patch *patch)
+ {
+       patch->old_mode = strtoul(line, NULL, 8);
+       return 0;
+ }
+ static int gitdiff_newmode(const char *line, struct patch *patch)
+ {
+       patch->new_mode = strtoul(line, NULL, 8);
+       return 0;
+ }
+ static int gitdiff_delete(const char *line, struct patch *patch)
+ {
+       patch->is_delete = 1;
+       patch->old_name = patch->def_name;
+       return gitdiff_oldmode(line, patch);
+ }
+ static int gitdiff_newfile(const char *line, struct patch *patch)
+ {
+       patch->is_new = 1;
+       patch->new_name = patch->def_name;
+       return gitdiff_newmode(line, patch);
+ }
+ static int gitdiff_copysrc(const char *line, struct patch *patch)
+ {
+       patch->is_copy = 1;
+       patch->old_name = find_name(line, NULL, 0, 0);
+       return 0;
+ }
+ static int gitdiff_copydst(const char *line, struct patch *patch)
+ {
+       patch->is_copy = 1;
+       patch->new_name = find_name(line, NULL, 0, 0);
+       return 0;
+ }
+ static int gitdiff_renamesrc(const char *line, struct patch *patch)
+ {
+       patch->is_rename = 1;
+       patch->old_name = find_name(line, NULL, 0, 0);
+       return 0;
+ }
+ static int gitdiff_renamedst(const char *line, struct patch *patch)
+ {
+       patch->is_rename = 1;
+       patch->new_name = find_name(line, NULL, 0, 0);
+       return 0;
+ }
+ static int gitdiff_similarity(const char *line, struct patch *patch)
+ {
+       if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
+               patch->score = 0;
+       return 0;
+ }
+ static int gitdiff_dissimilarity(const char *line, struct patch *patch)
+ {
+       if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
+               patch->score = 0;
+       return 0;
+ }
+ static int gitdiff_index(const char *line, struct patch *patch)
+ {
+       /* index line is N hexadecimal, "..", N hexadecimal,
+        * and optional space with octal mode.
+        */
+       const char *ptr, *eol;
+       int len;
+       ptr = strchr(line, '.');
+       if (!ptr || ptr[1] != '.' || 40 < ptr - line)
+               return 0;
+       len = ptr - line;
+       memcpy(patch->old_sha1_prefix, line, len);
+       patch->old_sha1_prefix[len] = 0;
+       line = ptr + 2;
+       ptr = strchr(line, ' ');
+       eol = strchr(line, '\n');
+       if (!ptr || eol < ptr)
+               ptr = eol;
+       len = ptr - line;
+       if (40 < len)
+               return 0;
+       memcpy(patch->new_sha1_prefix, line, len);
+       patch->new_sha1_prefix[len] = 0;
+       if (*ptr == ' ')
+               patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
+       return 0;
+ }
+ /*
+  * This is normal for a diff that doesn't change anything: we'll fall through
+  * into the next diff. Tell the parser to break out.
+  */
+ static int gitdiff_unrecognized(const char *line, struct patch *patch)
+ {
+       return -1;
+ }
+ static const char *stop_at_slash(const char *line, int llen)
+ {
+       int i;
+       for (i = 0; i < llen; i++) {
+               int ch = line[i];
+               if (ch == '/')
+                       return line + i;
+       }
+       return NULL;
+ }
+ /* This is to extract the same name that appears on "diff --git"
+  * line.  We do not find and return anything if it is a rename
+  * patch, and it is OK because we will find the name elsewhere.
+  * We need to reliably find name only when it is mode-change only,
+  * creation or deletion of an empty file.  In any of these cases,
+  * both sides are the same name under a/ and b/ respectively.
+  */
+ static char *git_header_name(char *line, int llen)
+ {
+       int len;
+       const char *name;
+       const char *second = NULL;
+       line += strlen("diff --git ");
+       llen -= strlen("diff --git ");
+       if (*line == '"') {
+               const char *cp;
+               char *first = unquote_c_style(line, &second);
+               if (!first)
+                       return NULL;
+               /* advance to the first slash */
+               cp = stop_at_slash(first, strlen(first));
+               if (!cp || cp == first) {
+                       /* we do not accept absolute paths */
+               free_first_and_fail:
+                       free(first);
+                       return NULL;
+               }
+               len = strlen(cp+1);
+               memmove(first, cp+1, len+1); /* including NUL */
+               /* second points at one past closing dq of name.
+                * find the second name.
+                */
+               while ((second < line + llen) && isspace(*second))
+                       second++;
+               if (line + llen <= second)
+                       goto free_first_and_fail;
+               if (*second == '"') {
+                       char *sp = unquote_c_style(second, NULL);
+                       if (!sp)
+                               goto free_first_and_fail;
+                       cp = stop_at_slash(sp, strlen(sp));
+                       if (!cp || cp == sp) {
+                       free_both_and_fail:
+                               free(sp);
+                               goto free_first_and_fail;
+                       }
+                       /* They must match, otherwise ignore */
+                       if (strcmp(cp+1, first))
+                               goto free_both_and_fail;
+                       free(sp);
+                       return first;
+               }
+               /* unquoted second */
+               cp = stop_at_slash(second, line + llen - second);
+               if (!cp || cp == second)
+                       goto free_first_and_fail;
+               cp++;
+               if (line + llen - cp != len + 1 ||
+                   memcmp(first, cp, len))
+                       goto free_first_and_fail;
+               return first;
+       }
+       /* unquoted first name */
+       name = stop_at_slash(line, llen);
+       if (!name || name == line)
+               return NULL;
+       name++;
+       /* since the first name is unquoted, a dq if exists must be
+        * the beginning of the second name.
+        */
+       for (second = name; second < line + llen; second++) {
+               if (*second == '"') {
+                       const char *cp = second;
+                       const char *np;
+                       char *sp = unquote_c_style(second, NULL);
+                       if (!sp)
+                               return NULL;
+                       np = stop_at_slash(sp, strlen(sp));
+                       if (!np || np == sp) {
+                       free_second_and_fail:
+                               free(sp);
+                               return NULL;
+                       }
+                       np++;
+                       len = strlen(np);
+                       if (len < cp - name &&
+                           !strncmp(np, name, len) &&
+                           isspace(name[len])) {
+                               /* Good */
+                               memmove(sp, np, len + 1);
+                               return sp;
+                       }
+                       goto free_second_and_fail;
+               }
+       }
+       /*
+        * Accept a name only if it shows up twice, exactly the same
+        * form.
+        */
+       for (len = 0 ; ; len++) {
+               char c = name[len];
+               switch (c) {
+               default:
+                       continue;
+               case '\n':
+                       return NULL;
+               case '\t': case ' ':
+                       second = name+len;
+                       for (;;) {
+                               char c = *second++;
+                               if (c == '\n')
+                                       return NULL;
+                               if (c == '/')
+                                       break;
+                       }
+                       if (second[len] == '\n' && !memcmp(name, second, len)) {
+                               char *ret = xmalloc(len + 1);
+                               memcpy(ret, name, len);
+                               ret[len] = 0;
+                               return ret;
+                       }
+               }
+       }
+       return NULL;
+ }
+ /* Verify that we recognize the lines following a git header */
+ static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
+ {
+       unsigned long offset;
+       /* A git diff has explicit new/delete information, so we don't guess */
+       patch->is_new = 0;
+       patch->is_delete = 0;
+       /*
+        * Some things may not have the old name in the
+        * rest of the headers anywhere (pure mode changes,
+        * or removing or adding empty files), so we get
+        * the default name from the header.
+        */
+       patch->def_name = git_header_name(line, len);
+       line += len;
+       size -= len;
+       linenr++;
+       for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
+               static const struct opentry {
+                       const char *str;
+                       int (*fn)(const char *, struct patch *);
+               } optable[] = {
+                       { "@@ -", gitdiff_hdrend },
+                       { "--- ", gitdiff_oldname },
+                       { "+++ ", gitdiff_newname },
+                       { "old mode ", gitdiff_oldmode },
+                       { "new mode ", gitdiff_newmode },
+                       { "deleted file mode ", gitdiff_delete },
+                       { "new file mode ", gitdiff_newfile },
+                       { "copy from ", gitdiff_copysrc },
+                       { "copy to ", gitdiff_copydst },
+                       { "rename old ", gitdiff_renamesrc },
+                       { "rename new ", gitdiff_renamedst },
+                       { "rename from ", gitdiff_renamesrc },
+                       { "rename to ", gitdiff_renamedst },
+                       { "similarity index ", gitdiff_similarity },
+                       { "dissimilarity index ", gitdiff_dissimilarity },
+                       { "index ", gitdiff_index },
+                       { "", gitdiff_unrecognized },
+               };
+               int i;
+               len = linelen(line, size);
+               if (!len || line[len-1] != '\n')
+                       break;
+               for (i = 0; i < ARRAY_SIZE(optable); i++) {
+                       const struct opentry *p = optable + i;
+                       int oplen = strlen(p->str);
+                       if (len < oplen || memcmp(p->str, line, oplen))
+                               continue;
+                       if (p->fn(line + oplen, patch) < 0)
+                               return offset;
+                       break;
+               }
+       }
+       return offset;
+ }
+ static int parse_num(const char *line, unsigned long *p)
+ {
+       char *ptr;
+       if (!isdigit(*line))
+               return 0;
+       *p = strtoul(line, &ptr, 10);
+       return ptr - line;
+ }
+ static int parse_range(const char *line, int len, int offset, const char *expect,
+                       unsigned long *p1, unsigned long *p2)
+ {
+       int digits, ex;
+       if (offset < 0 || offset >= len)
+               return -1;
+       line += offset;
+       len -= offset;
+       digits = parse_num(line, p1);
+       if (!digits)
+               return -1;
+       offset += digits;
+       line += digits;
+       len -= digits;
+       *p2 = 1;
+       if (*line == ',') {
+               digits = parse_num(line+1, p2);
+               if (!digits)
+                       return -1;
+               offset += digits+1;
+               line += digits+1;
+               len -= digits+1;
+       }
+       ex = strlen(expect);
+       if (ex > len)
+               return -1;
+       if (memcmp(line, expect, ex))
+               return -1;
+       return offset + ex;
+ }
+ /*
+  * Parse a unified diff fragment header of the
+  * form "@@ -a,b +c,d @@"
+  */
+ static int parse_fragment_header(char *line, int len, struct fragment *fragment)
+ {
+       int offset;
+       if (!len || line[len-1] != '\n')
+               return -1;
+       /* Figure out the number of lines in a fragment */
+       offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
+       offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
+       return offset;
+ }
+ static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
+ {
+       unsigned long offset, len;
+       patch->is_rename = patch->is_copy = 0;
+       patch->is_new = patch->is_delete = -1;
+       patch->old_mode = patch->new_mode = 0;
+       patch->old_name = patch->new_name = NULL;
+       for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
+               unsigned long nextlen;
+               len = linelen(line, size);
+               if (!len)
+                       break;
+               /* Testing this early allows us to take a few shortcuts.. */
+               if (len < 6)
+                       continue;
+               /*
+                * Make sure we don't find any unconnected patch fragmants.
+                * That's a sign that we didn't find a header, and that a
+                * patch has become corrupted/broken up.
+                */
+               if (!memcmp("@@ -", line, 4)) {
+                       struct fragment dummy;
+                       if (parse_fragment_header(line, len, &dummy) < 0)
+                               continue;
+                       error("patch fragment without header at line %d: %.*s", linenr, (int)len-1, line);
+               }
+               if (size < len + 6)
+                       break;
+               /*
+                * Git patch? It might not have a real patch, just a rename
+                * or mode change, so we handle that specially
+                */
+               if (!memcmp("diff --git ", line, 11)) {
+                       int git_hdr_len = parse_git_header(line, len, size, patch);
+                       if (git_hdr_len <= len)
+                               continue;
+                       if (!patch->old_name && !patch->new_name) {
+                               if (!patch->def_name)
+                                       die("git diff header lacks filename information (line %d)", linenr);
+                               patch->old_name = patch->new_name = patch->def_name;
+                       }
+                       *hdrsize = git_hdr_len;
+                       return offset;
+               }
+               /** --- followed by +++ ? */
+               if (memcmp("--- ", line,  4) || memcmp("+++ ", line + len, 4))
+                       continue;
+               /*
+                * We only accept unified patches, so we want it to
+                * at least have "@@ -a,b +c,d @@\n", which is 14 chars
+                * minimum
+                */
+               nextlen = linelen(line + len, size - len);
+               if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
+                       continue;
+               /* Ok, we'll consider it a patch */
+               parse_traditional_patch(line, line+len, patch);
+               *hdrsize = len + nextlen;
+               linenr += 2;
+               return offset;
+       }
+       return -1;
+ }
+ /*
+  * Parse a unified diff. Note that this really needs
+  * to parse each fragment separately, since the only
+  * way to know the difference between a "---" that is
+  * part of a patch, and a "---" that starts the next
+  * patch is to look at the line counts..
+  */
+ static int parse_fragment(char *line, unsigned long size, struct patch *patch, struct fragment *fragment)
+ {
+       int added, deleted;
+       int len = linelen(line, size), offset;
+       unsigned long oldlines, newlines;
+       unsigned long leading, trailing;
+       offset = parse_fragment_header(line, len, fragment);
+       if (offset < 0)
+               return -1;
+       oldlines = fragment->oldlines;
+       newlines = fragment->newlines;
+       leading = 0;
+       trailing = 0;
+       if (patch->is_new < 0) {
+               patch->is_new =  !oldlines;
+               if (!oldlines)
+                       patch->old_name = NULL;
+       }
+       if (patch->is_delete < 0) {
+               patch->is_delete = !newlines;
+               if (!newlines)
+                       patch->new_name = NULL;
+       }
+       if (patch->is_new && oldlines)
+               return error("new file depends on old contents");
+       if (patch->is_delete != !newlines) {
+               if (newlines)
+                       return error("deleted file still has contents");
+               fprintf(stderr, "** warning: file %s becomes empty but is not deleted\n", patch->new_name);
+       }
+       /* Parse the thing.. */
+       line += len;
+       size -= len;
+       linenr++;
+       added = deleted = 0;
+       for (offset = len; size > 0; offset += len, size -= len, line += len, linenr++) {
+               if (!oldlines && !newlines)
+                       break;
+               len = linelen(line, size);
+               if (!len || line[len-1] != '\n')
+                       return -1;
+               switch (*line) {
+               default:
+                       return -1;
+               case ' ':
+                       oldlines--;
+                       newlines--;
+                       if (!deleted && !added)
+                               leading++;
+                       trailing++;
+                       break;
+               case '-':
+                       deleted++;
+                       oldlines--;
+                       trailing = 0;
+                       break;
+               case '+':
+                       /*
+                        * We know len is at least two, since we have a '+' and
+                        * we checked that the last character was a '\n' above.
+                        * That is, an addition of an empty line would check
+                        * the '+' here.  Sneaky...
+                        */
+                       if ((new_whitespace != nowarn_whitespace) &&
+                           isspace(line[len-2])) {
+                               whitespace_error++;
+                               if (squelch_whitespace_errors &&
+                                   squelch_whitespace_errors <
+                                   whitespace_error)
+                                       ;
+                               else {
+                                       fprintf(stderr, "Adds trailing whitespace.\n%s:%d:%.*s\n",
+                                               patch_input_file,
+                                               linenr, len-2, line+1);
+                               }
+                       }
+                       added++;
+                       newlines--;
+                       trailing = 0;
+                       break;
+                 /* We allow "\ No newline at end of file". Depending
+                  * on locale settings when the patch was produced we
+                  * don't know what this line looks like. The only
+                  * thing we do know is that it begins with "\ ".
+                * Checking for 12 is just for sanity check -- any
+                * l10n of "\ No newline..." is at least that long.
+                */
+               case '\\':
+                       if (len < 12 || memcmp(line, "\\ ", 2))
+                               return -1;
+                       break;
+               }
+       }
+       if (oldlines || newlines)
+               return -1;
+       fragment->leading = leading;
+       fragment->trailing = trailing;
+       /* If a fragment ends with an incomplete line, we failed to include
+        * it in the above loop because we hit oldlines == newlines == 0
+        * before seeing it.
+        */
+       if (12 < size && !memcmp(line, "\\ ", 2))
+               offset += linelen(line, size);
+       patch->lines_added += added;
+       patch->lines_deleted += deleted;
+       return offset;
+ }
+ static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
+ {
+       unsigned long offset = 0;
+       struct fragment **fragp = &patch->fragments;
+       while (size > 4 && !memcmp(line, "@@ -", 4)) {
+               struct fragment *fragment;
+               int len;
+               fragment = xcalloc(1, sizeof(*fragment));
+               len = parse_fragment(line, size, patch, fragment);
+               if (len <= 0)
+                       die("corrupt patch at line %d", linenr);
+               fragment->patch = line;
+               fragment->size = len;
+               *fragp = fragment;
+               fragp = &fragment->next;
+               offset += len;
+               line += len;
+               size -= len;
+       }
+       return offset;
+ }
+ static inline int metadata_changes(struct patch *patch)
+ {
+       return  patch->is_rename > 0 ||
+               patch->is_copy > 0 ||
+               patch->is_new > 0 ||
+               patch->is_delete ||
+               (patch->old_mode && patch->new_mode &&
+                patch->old_mode != patch->new_mode);
+ }
+ static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
+ {
+       /* We have read "GIT binary patch\n"; what follows is a line
+        * that says the patch method (currently, either "deflated
+        * literal" or "deflated delta") and the length of data before
+        * deflating; a sequence of 'length-byte' followed by base-85
+        * encoded data follows.
+        *
+        * Each 5-byte sequence of base-85 encodes up to 4 bytes,
+        * and we would limit the patch line to 66 characters,
+        * so one line can fit up to 13 groups that would decode
+        * to 52 bytes max.  The length byte 'A'-'Z' corresponds
+        * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
+        * The end of binary is signalled with an empty line.
+        */
+       int llen, used;
+       struct fragment *fragment;
+       char *data = NULL;
+       patch->fragments = fragment = xcalloc(1, sizeof(*fragment));
+       /* Grab the type of patch */
+       llen = linelen(buffer, size);
+       used = llen;
+       linenr++;
+       if (!strncmp(buffer, "delta ", 6)) {
+               patch->is_binary = BINARY_DELTA_DEFLATED;
+               patch->deflate_origlen = strtoul(buffer + 6, NULL, 10);
+       }
+       else if (!strncmp(buffer, "literal ", 8)) {
+               patch->is_binary = BINARY_LITERAL_DEFLATED;
+               patch->deflate_origlen = strtoul(buffer + 8, NULL, 10);
+       }
+       else
+               return error("unrecognized binary patch at line %d: %.*s",
+                            linenr-1, llen-1, buffer);
+       buffer += llen;
+       while (1) {
+               int byte_length, max_byte_length, newsize;
+               llen = linelen(buffer, size);
+               used += llen;
+               linenr++;
+               if (llen == 1)
+                       break;
+               /* Minimum line is "A00000\n" which is 7-byte long,
+                * and the line length must be multiple of 5 plus 2.
+                */
+               if ((llen < 7) || (llen-2) % 5)
+                       goto corrupt;
+               max_byte_length = (llen - 2) / 5 * 4;
+               byte_length = *buffer;
+               if ('A' <= byte_length && byte_length <= 'Z')
+                       byte_length = byte_length - 'A' + 1;
+               else if ('a' <= byte_length && byte_length <= 'z')
+                       byte_length = byte_length - 'a' + 27;
+               else
+                       goto corrupt;
+               /* if the input length was not multiple of 4, we would
+                * have filler at the end but the filler should never
+                * exceed 3 bytes
+                */
+               if (max_byte_length < byte_length ||
+                   byte_length <= max_byte_length - 4)
+                       goto corrupt;
+               newsize = fragment->size + byte_length;
+               data = xrealloc(data, newsize);
+               if (decode_85(data + fragment->size,
+                             buffer + 1,
+                             byte_length))
+                       goto corrupt;
+               fragment->size = newsize;
+               buffer += llen;
+               size -= llen;
+       }
+       fragment->patch = data;
+       return used;
+  corrupt:
+       return error("corrupt binary patch at line %d: %.*s",
+                    linenr-1, llen-1, buffer);
+ }
+ static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
+ {
+       int hdrsize, patchsize;
+       int offset = find_header(buffer, size, &hdrsize, patch);
+       if (offset < 0)
+               return offset;
+       patchsize = parse_single_patch(buffer + offset + hdrsize, size - offset - hdrsize, patch);
+       if (!patchsize) {
+               static const char *binhdr[] = {
+                       "Binary files ",
+                       "Files ",
+                       NULL,
+               };
+               static const char git_binary[] = "GIT binary patch\n";
+               int i;
+               int hd = hdrsize + offset;
+               unsigned long llen = linelen(buffer + hd, size - hd);
+               if (llen == sizeof(git_binary) - 1 &&
+                   !memcmp(git_binary, buffer + hd, llen)) {
+                       int used;
+                       linenr++;
+                       used = parse_binary(buffer + hd + llen,
+                                           size - hd - llen, patch);
+                       if (used)
+                               patchsize = used + llen;
+                       else
+                               patchsize = 0;
+               }
+               else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
+                       for (i = 0; binhdr[i]; i++) {
+                               int len = strlen(binhdr[i]);
+                               if (len < size - hd &&
+                                   !memcmp(binhdr[i], buffer + hd, len)) {
+                                       linenr++;
+                                       patch->is_binary = 1;
+                                       patchsize = llen;
+                                       break;
+                               }
+                       }
+               }
+               /* Empty patch cannot be applied if:
+                * - it is a binary patch and we do not do binary_replace, or
+                * - text patch without metadata change
+                */
+               if ((apply || check) &&
+                   (patch->is_binary
+                    ? !allow_binary_replacement
+                    : !metadata_changes(patch)))
+                       die("patch with only garbage at line %d", linenr);
+       }
+       return offset + hdrsize + patchsize;
+ }
+ static const char pluses[] = "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
+ static const char minuses[]= "----------------------------------------------------------------------";
+ static void show_stats(struct patch *patch)
+ {
+       const char *prefix = "";
+       char *name = patch->new_name;
+       char *qname = NULL;
+       int len, max, add, del, total;
+       if (!name)
+               name = patch->old_name;
+       if (0 < (len = quote_c_style(name, NULL, NULL, 0))) {
+               qname = xmalloc(len + 1);
+               quote_c_style(name, qname, NULL, 0);
+               name = qname;
+       }
+       /*
+        * "scale" the filename
+        */
+       len = strlen(name);
+       max = max_len;
+       if (max > 50)
+               max = 50;
+       if (len > max) {
+               char *slash;
+               prefix = "...";
+               max -= 3;
+               name += len - max;
+               slash = strchr(name, '/');
+               if (slash)
+                       name = slash;
+       }
+       len = max;
+       /*
+        * scale the add/delete
+        */
+       max = max_change;
+       if (max + len > 70)
+               max = 70 - len;
+       add = patch->lines_added;
+       del = patch->lines_deleted;
+       total = add + del;
+       if (max_change > 0) {
+               total = (total * max + max_change / 2) / max_change;
+               add = (add * max + max_change / 2) / max_change;
+               del = total - add;
+       }
+       if (patch->is_binary)
+               printf(" %s%-*s |  Bin\n", prefix, len, name);
+       else
+               printf(" %s%-*s |%5d %.*s%.*s\n", prefix,
+                      len, name, patch->lines_added + patch->lines_deleted,
+                      add, pluses, del, minuses);
+       if (qname)
+               free(qname);
+ }
+ static int read_old_data(struct stat *st, const char *path, void *buf, unsigned long size)
+ {
+       int fd;
+       unsigned long got;
+       switch (st->st_mode & S_IFMT) {
+       case S_IFLNK:
+               return readlink(path, buf, size);
+       case S_IFREG:
+               fd = open(path, O_RDONLY);
+               if (fd < 0)
+                       return error("unable to open %s", path);
+               got = 0;
+               for (;;) {
+                       int ret = xread(fd, buf + got, size - got);
+                       if (ret <= 0)
+                               break;
+                       got += ret;
+               }
+               close(fd);
+               return got;
+       default:
+               return -1;
+       }
+ }
+ static int find_offset(const char *buf, unsigned long size, const char *fragment, unsigned long fragsize, int line, int *lines)
+ {
+       int i;
+       unsigned long start, backwards, forwards;
+       if (fragsize > size)
+               return -1;
+       start = 0;
+       if (line > 1) {
+               unsigned long offset = 0;
+               i = line-1;
+               while (offset + fragsize <= size) {
+                       if (buf[offset++] == '\n') {
+                               start = offset;
+                               if (!--i)
+                                       break;
+                       }
+               }
+       }
+       /* Exact line number? */
+       if (!memcmp(buf + start, fragment, fragsize))
+               return start;
+       /*
+        * There's probably some smart way to do this, but I'll leave
+        * that to the smart and beautiful people. I'm simple and stupid.
+        */
+       backwards = start;
+       forwards = start;
+       for (i = 0; ; i++) {
+               unsigned long try;
+               int n;
+               /* "backward" */
+               if (i & 1) {
+                       if (!backwards) {
+                               if (forwards + fragsize > size)
+                                       break;
+                               continue;
+                       }
+                       do {
+                               --backwards;
+                       } while (backwards && buf[backwards-1] != '\n');
+                       try = backwards;
+               } else {
+                       while (forwards + fragsize <= size) {
+                               if (buf[forwards++] == '\n')
+                                       break;
+                       }
+                       try = forwards;
+               }
+               if (try + fragsize > size)
+                       continue;
+               if (memcmp(buf + try, fragment, fragsize))
+                       continue;
+               n = (i >> 1)+1;
+               if (i & 1)
+                       n = -n;
+               *lines = n;
+               return try;
+       }
+       /*
+        * We should start searching forward and backward.
+        */
+       return -1;
+ }
+ static void remove_first_line(const char **rbuf, int *rsize)
+ {
+       const char *buf = *rbuf;
+       int size = *rsize;
+       unsigned long offset;
+       offset = 0;
+       while (offset <= size) {
+               if (buf[offset++] == '\n')
+                       break;
+       }
+       *rsize = size - offset;
+       *rbuf = buf + offset;
+ }
+ static void remove_last_line(const char **rbuf, int *rsize)
+ {
+       const char *buf = *rbuf;
+       int size = *rsize;
+       unsigned long offset;
+       offset = size - 1;
+       while (offset > 0) {
+               if (buf[--offset] == '\n')
+                       break;
+       }
+       *rsize = offset + 1;
+ }
+ struct buffer_desc {
+       char *buffer;
+       unsigned long size;
+       unsigned long alloc;
+ };
+ static int apply_line(char *output, const char *patch, int plen)
+ {
+       /* plen is number of bytes to be copied from patch,
+        * starting at patch+1 (patch[0] is '+').  Typically
+        * patch[plen] is '\n'.
+        */
+       int add_nl_to_tail = 0;
+       if ((new_whitespace == strip_whitespace) &&
+           1 < plen && isspace(patch[plen-1])) {
+               if (patch[plen] == '\n')
+                       add_nl_to_tail = 1;
+               plen--;
+               while (0 < plen && isspace(patch[plen]))
+                       plen--;
+               applied_after_stripping++;
+       }
+       memcpy(output, patch + 1, plen);
+       if (add_nl_to_tail)
+               output[plen++] = '\n';
+       return plen;
+ }
+ static int apply_one_fragment(struct buffer_desc *desc, struct fragment *frag)
+ {
++      int match_beginning, match_end;
+       char *buf = desc->buffer;
+       const char *patch = frag->patch;
+       int offset, size = frag->size;
+       char *old = xmalloc(size);
+       char *new = xmalloc(size);
+       const char *oldlines, *newlines;
+       int oldsize = 0, newsize = 0;
+       unsigned long leading, trailing;
+       int pos, lines;
+       while (size > 0) {
+               int len = linelen(patch, size);
+               int plen;
+               if (!len)
+                       break;
+               /*
+                * "plen" is how much of the line we should use for
+                * the actual patch data. Normally we just remove the
+                * first character on the line, but if the line is
+                * followed by "\ No newline", then we also remove the
+                * last one (which is the newline, of course).
+                */
+               plen = len-1;
+               if (len < size && patch[len] == '\\')
+                       plen--;
+               switch (*patch) {
+               case ' ':
+               case '-':
+                       memcpy(old + oldsize, patch + 1, plen);
+                       oldsize += plen;
+                       if (*patch == '-')
+                               break;
+               /* Fall-through for ' ' */
+               case '+':
+                       if (*patch != '+' || !no_add)
+                               newsize += apply_line(new + newsize, patch,
+                                                     plen);
+                       break;
+               case '@': case '\\':
+                       /* Ignore it, we already handled it */
+                       break;
+               default:
+                       return -1;
+               }
+               patch += len;
+               size -= len;
+       }
+ #ifdef NO_ACCURATE_DIFF
+       if (oldsize > 0 && old[oldsize - 1] == '\n' &&
+                       newsize > 0 && new[newsize - 1] == '\n') {
+               oldsize--;
+               newsize--;
+       }
+ #endif
+       oldlines = old;
+       newlines = new;
+       leading = frag->leading;
+       trailing = frag->trailing;
++
++      /*
++       * If we don't have any leading/trailing data in the patch,
++       * we want it to match at the beginning/end of the file.
++       */
++      match_beginning = !leading && (frag->oldpos == 1);
++      match_end = !trailing;
++
+       lines = 0;
+       pos = frag->newpos;
+       for (;;) {
+               offset = find_offset(buf, desc->size, oldlines, oldsize, pos, &lines);
++              if (match_end && offset + oldsize != desc->size)
++                      offset = -1;
++              if (match_beginning && offset)
++                      offset = -1;
+               if (offset >= 0) {
+                       int diff = newsize - oldsize;
+                       unsigned long size = desc->size + diff;
+                       unsigned long alloc = desc->alloc;
+                       /* Warn if it was necessary to reduce the number
+                        * of context lines.
+                        */
+                       if ((leading != frag->leading) || (trailing != frag->trailing))
+                               fprintf(stderr, "Context reduced to (%ld/%ld) to apply fragment at %d\n",
+                                       leading, trailing, pos + lines);
+                       if (size > alloc) {
+                               alloc = size + 8192;
+                               desc->alloc = alloc;
+                               buf = xrealloc(buf, alloc);
+                               desc->buffer = buf;
+                       }
+                       desc->size = size;
+                       memmove(buf + offset + newsize, buf + offset + oldsize, size - offset - newsize);
+                       memcpy(buf + offset, newlines, newsize);
+                       offset = 0;
+                       break;
+               }
+               /* Am I at my context limits? */
+               if ((leading <= p_context) && (trailing <= p_context))
+                       break;
++              if (match_beginning || match_end) {
++                      match_beginning = match_end = 0;
++                      continue;
++              }
+               /* Reduce the number of context lines
+                * Reduce both leading and trailing if they are equal
+                * otherwise just reduce the larger context.
+                */
+               if (leading >= trailing) {
+                       remove_first_line(&oldlines, &oldsize);
+                       remove_first_line(&newlines, &newsize);
+                       pos--;
+                       leading--;
+               }
+               if (trailing > leading) {
+                       remove_last_line(&oldlines, &oldsize);
+                       remove_last_line(&newlines, &newsize);
+                       trailing--;
+               }
+       }
+       free(old);
+       free(new);
+       return offset;
+ }
+ static char *inflate_it(const void *data, unsigned long size,
+                       unsigned long inflated_size)
+ {
+       z_stream stream;
+       void *out;
+       int st;
+       memset(&stream, 0, sizeof(stream));
+       stream.next_in = (unsigned char *)data;
+       stream.avail_in = size;
+       stream.next_out = out = xmalloc(inflated_size);
+       stream.avail_out = inflated_size;
+       inflateInit(&stream);
+       st = inflate(&stream, Z_FINISH);
+       if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
+               free(out);
+               return NULL;
+       }
+       return out;
+ }
+ static int apply_binary_fragment(struct buffer_desc *desc, struct patch *patch)
+ {
+       unsigned long dst_size;
+       struct fragment *fragment = patch->fragments;
+       void *data;
+       void *result;
+       data = inflate_it(fragment->patch, fragment->size,
+                         patch->deflate_origlen);
+       if (!data)
+               return error("corrupt patch data");
+       switch (patch->is_binary) {
+       case BINARY_DELTA_DEFLATED:
+               result = patch_delta(desc->buffer, desc->size,
+                                    data,
+                                    patch->deflate_origlen,
+                                    &dst_size);
+               free(desc->buffer);
+               desc->buffer = result;
+               free(data);
+               break;
+       case BINARY_LITERAL_DEFLATED:
+               free(desc->buffer);
+               desc->buffer = data;
+               dst_size = patch->deflate_origlen;
+               break;
+       }
+       if (!desc->buffer)
+               return -1;
+       desc->size = desc->alloc = dst_size;
+       return 0;
+ }
+ static int apply_binary(struct buffer_desc *desc, struct patch *patch)
+ {
+       const char *name = patch->old_name ? patch->old_name : patch->new_name;
+       unsigned char sha1[20];
+       unsigned char hdr[50];
+       int hdrlen;
+       if (!allow_binary_replacement)
+               return error("cannot apply binary patch to '%s' "
+                            "without --allow-binary-replacement",
+                            name);
+       /* For safety, we require patch index line to contain
+        * full 40-byte textual SHA1 for old and new, at least for now.
+        */
+       if (strlen(patch->old_sha1_prefix) != 40 ||
+           strlen(patch->new_sha1_prefix) != 40 ||
+           get_sha1_hex(patch->old_sha1_prefix, sha1) ||
+           get_sha1_hex(patch->new_sha1_prefix, sha1))
+               return error("cannot apply binary patch to '%s' "
+                            "without full index line", name);
+       if (patch->old_name) {
+               /* See if the old one matches what the patch
+                * applies to.
+                */
+               write_sha1_file_prepare(desc->buffer, desc->size,
+                                       blob_type, sha1, hdr, &hdrlen);
+               if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
+                       return error("the patch applies to '%s' (%s), "
+                                    "which does not match the "
+                                    "current contents.",
+                                    name, sha1_to_hex(sha1));
+       }
+       else {
+               /* Otherwise, the old one must be empty. */
+               if (desc->size)
+                       return error("the patch applies to an empty "
+                                    "'%s' but it is not empty", name);
+       }
+       get_sha1_hex(patch->new_sha1_prefix, sha1);
+       if (!memcmp(sha1, null_sha1, 20)) {
+               free(desc->buffer);
+               desc->alloc = desc->size = 0;
+               desc->buffer = NULL;
+               return 0; /* deletion patch */
+       }
+       if (has_sha1_file(sha1)) {
+               /* We already have the postimage */
+               char type[10];
+               unsigned long size;
+               free(desc->buffer);
+               desc->buffer = read_sha1_file(sha1, type, &size);
+               if (!desc->buffer)
+                       return error("the necessary postimage %s for "
+                                    "'%s' cannot be read",
+                                    patch->new_sha1_prefix, name);
+               desc->alloc = desc->size = size;
+       }
+       else {
+               /* We have verified desc matches the preimage;
+                * apply the patch data to it, which is stored
+                * in the patch->fragments->{patch,size}.
+                */
+               if (apply_binary_fragment(desc, patch))
+                       return error("binary patch does not apply to '%s'",
+                                    name);
+               /* verify that the result matches */
+               write_sha1_file_prepare(desc->buffer, desc->size, blob_type,
+                                       sha1, hdr, &hdrlen);
+               if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
+                       return error("binary patch to '%s' creates incorrect result", name);
+       }
+       return 0;
+ }
+ static int apply_fragments(struct buffer_desc *desc, struct patch *patch)
+ {
+       struct fragment *frag = patch->fragments;
+       const char *name = patch->old_name ? patch->old_name : patch->new_name;
+       if (patch->is_binary)
+               return apply_binary(desc, patch);
+       while (frag) {
+               if (apply_one_fragment(desc, frag) < 0)
+                       return error("patch failed: %s:%ld",
+                                    name, frag->oldpos);
+               frag = frag->next;
+       }
+       return 0;
+ }
+ static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
+ {
+       char *buf;
+       unsigned long size, alloc;
+       struct buffer_desc desc;
+       size = 0;
+       alloc = 0;
+       buf = NULL;
+       if (cached) {
+               if (ce) {
+                       char type[20];
+                       buf = read_sha1_file(ce->sha1, type, &size);
+                       if (!buf)
+                               return error("read of %s failed",
+                                            patch->old_name);
+                       alloc = size;
+               }
+       }
+       else if (patch->old_name) {
+               size = st->st_size;
+               alloc = size + 8192;
+               buf = xmalloc(alloc);
+               if (read_old_data(st, patch->old_name, buf, alloc) != size)
+                       return error("read of %s failed", patch->old_name);
+       }
+       desc.size = size;
+       desc.alloc = alloc;
+       desc.buffer = buf;
+       if (apply_fragments(&desc, patch) < 0)
+               return -1;
+       patch->result = desc.buffer;
+       patch->resultsize = desc.size;
+       if (patch->is_delete && patch->resultsize)
+               return error("removal patch leaves file contents");
+       return 0;
+ }
+ static int check_patch(struct patch *patch)
+ {
+       struct stat st;
+       const char *old_name = patch->old_name;
+       const char *new_name = patch->new_name;
+       const char *name = old_name ? old_name : new_name;
+       struct cache_entry *ce = NULL;
+       if (old_name) {
+               int changed = 0;
+               int stat_ret = 0;
+               unsigned st_mode = 0;
+               if (!cached)
+                       stat_ret = lstat(old_name, &st);
+               if (check_index) {
+                       int pos = cache_name_pos(old_name, strlen(old_name));
+                       if (pos < 0)
+                               return error("%s: does not exist in index",
+                                            old_name);
+                       ce = active_cache[pos];
+                       if (stat_ret < 0) {
+                               struct checkout costate;
+                               if (errno != ENOENT)
+                                       return error("%s: %s", old_name,
+                                                    strerror(errno));
+                               /* checkout */
+                               costate.base_dir = "";
+                               costate.base_dir_len = 0;
+                               costate.force = 0;
+                               costate.quiet = 0;
+                               costate.not_new = 0;
+                               costate.refresh_cache = 1;
+                               if (checkout_entry(ce,
+                                                  &costate,
+                                                  NULL) ||
+                                   lstat(old_name, &st))
+                                       return -1;
+                       }
+                       if (!cached)
+                               changed = ce_match_stat(ce, &st, 1);
+                       if (changed)
+                               return error("%s: does not match index",
+                                            old_name);
+                       if (cached)
+                               st_mode = ntohl(ce->ce_mode);
+               }
+               else if (stat_ret < 0)
+                       return error("%s: %s", old_name, strerror(errno));
+               if (!cached)
+                       st_mode = ntohl(create_ce_mode(st.st_mode));
+               if (patch->is_new < 0)
+                       patch->is_new = 0;
+               if (!patch->old_mode)
+                       patch->old_mode = st_mode;
+               if ((st_mode ^ patch->old_mode) & S_IFMT)
+                       return error("%s: wrong type", old_name);
+               if (st_mode != patch->old_mode)
+                       fprintf(stderr, "warning: %s has type %o, expected %o\n",
+                               old_name, st_mode, patch->old_mode);
+       }
+       if (new_name && (patch->is_new | patch->is_rename | patch->is_copy)) {
+               if (check_index && cache_name_pos(new_name, strlen(new_name)) >= 0)
+                       return error("%s: already exists in index", new_name);
+               if (!cached) {
+                       if (!lstat(new_name, &st))
+                               return error("%s: already exists in working directory", new_name);
+                       if (errno != ENOENT)
+                               return error("%s: %s", new_name, strerror(errno));
+               }
+               if (!patch->new_mode) {
+                       if (patch->is_new)
+                               patch->new_mode = S_IFREG | 0644;
+                       else
+                               patch->new_mode = patch->old_mode;
+               }
+       }
+       if (new_name && old_name) {
+               int same = !strcmp(old_name, new_name);
+               if (!patch->new_mode)
+                       patch->new_mode = patch->old_mode;
+               if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
+                       return error("new mode (%o) of %s does not match old mode (%o)%s%s",
+                               patch->new_mode, new_name, patch->old_mode,
+                               same ? "" : " of ", same ? "" : old_name);
+       }
+       if (apply_data(patch, &st, ce) < 0)
+               return error("%s: patch does not apply", name);
+       return 0;
+ }
+ static int check_patch_list(struct patch *patch)
+ {
+       int error = 0;
+       for (;patch ; patch = patch->next)
+               error |= check_patch(patch);
+       return error;
+ }
+ static inline int is_null_sha1(const unsigned char *sha1)
+ {
+       return !memcmp(sha1, null_sha1, 20);
+ }
+ static void show_index_list(struct patch *list)
+ {
+       struct patch *patch;
+       /* Once we start supporting the reverse patch, it may be
+        * worth showing the new sha1 prefix, but until then...
+        */
+       for (patch = list; patch; patch = patch->next) {
+               const unsigned char *sha1_ptr;
+               unsigned char sha1[20];
+               const char *name;
+               name = patch->old_name ? patch->old_name : patch->new_name;
+               if (patch->is_new)
+                       sha1_ptr = null_sha1;
+               else if (get_sha1(patch->old_sha1_prefix, sha1))
+                       die("sha1 information is lacking or useless (%s).",
+                           name);
+               else
+                       sha1_ptr = sha1;
+               printf("%06o %s ",patch->old_mode, sha1_to_hex(sha1_ptr));
+               if (line_termination && quote_c_style(name, NULL, NULL, 0))
+                       quote_c_style(name, NULL, stdout, 0);
+               else
+                       fputs(name, stdout);
+               putchar(line_termination);
+       }
+ }
+ static void stat_patch_list(struct patch *patch)
+ {
+       int files, adds, dels;
+       for (files = adds = dels = 0 ; patch ; patch = patch->next) {
+               files++;
+               adds += patch->lines_added;
+               dels += patch->lines_deleted;
+               show_stats(patch);
+       }
+       printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
+ }
+ static void numstat_patch_list(struct patch *patch)
+ {
+       for ( ; patch; patch = patch->next) {
+               const char *name;
+               name = patch->new_name ? patch->new_name : patch->old_name;
+               printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
+               if (line_termination && quote_c_style(name, NULL, NULL, 0))
+                       quote_c_style(name, NULL, stdout, 0);
+               else
+                       fputs(name, stdout);
+               putchar('\n');
+       }
+ }
+ static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
+ {
+       if (mode)
+               printf(" %s mode %06o %s\n", newdelete, mode, name);
+       else
+               printf(" %s %s\n", newdelete, name);
+ }
+ static void show_mode_change(struct patch *p, int show_name)
+ {
+       if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
+               if (show_name)
+                       printf(" mode change %06o => %06o %s\n",
+                              p->old_mode, p->new_mode, p->new_name);
+               else
+                       printf(" mode change %06o => %06o\n",
+                              p->old_mode, p->new_mode);
+       }
+ }
+ static void show_rename_copy(struct patch *p)
+ {
+       const char *renamecopy = p->is_rename ? "rename" : "copy";
+       const char *old, *new;
+       /* Find common prefix */
+       old = p->old_name;
+       new = p->new_name;
+       while (1) {
+               const char *slash_old, *slash_new;
+               slash_old = strchr(old, '/');
+               slash_new = strchr(new, '/');
+               if (!slash_old ||
+                   !slash_new ||
+                   slash_old - old != slash_new - new ||
+                   memcmp(old, new, slash_new - new))
+                       break;
+               old = slash_old + 1;
+               new = slash_new + 1;
+       }
+       /* p->old_name thru old is the common prefix, and old and new
+        * through the end of names are renames
+        */
+       if (old != p->old_name)
+               printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
+                      (int)(old - p->old_name), p->old_name,
+                      old, new, p->score);
+       else
+               printf(" %s %s => %s (%d%%)\n", renamecopy,
+                      p->old_name, p->new_name, p->score);
+       show_mode_change(p, 0);
+ }
+ static void summary_patch_list(struct patch *patch)
+ {
+       struct patch *p;
+       for (p = patch; p; p = p->next) {
+               if (p->is_new)
+                       show_file_mode_name("create", p->new_mode, p->new_name);
+               else if (p->is_delete)
+                       show_file_mode_name("delete", p->old_mode, p->old_name);
+               else {
+                       if (p->is_rename || p->is_copy)
+                               show_rename_copy(p);
+                       else {
+                               if (p->score) {
+                                       printf(" rewrite %s (%d%%)\n",
+                                              p->new_name, p->score);
+                                       show_mode_change(p, 0);
+                               }
+                               else
+                                       show_mode_change(p, 1);
+                       }
+               }
+       }
+ }
+ static void patch_stats(struct patch *patch)
+ {
+       int lines = patch->lines_added + patch->lines_deleted;
+       if (lines > max_change)
+               max_change = lines;
+       if (patch->old_name) {
+               int len = quote_c_style(patch->old_name, NULL, NULL, 0);
+               if (!len)
+                       len = strlen(patch->old_name);
+               if (len > max_len)
+                       max_len = len;
+       }
+       if (patch->new_name) {
+               int len = quote_c_style(patch->new_name, NULL, NULL, 0);
+               if (!len)
+                       len = strlen(patch->new_name);
+               if (len > max_len)
+                       max_len = len;
+       }
+ }
+ static void remove_file(struct patch *patch)
+ {
+       if (write_index) {
+               if (remove_file_from_cache(patch->old_name) < 0)
+                       die("unable to remove %s from index", patch->old_name);
+       }
+       if (!cached)
+               unlink(patch->old_name);
+ }
+ static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
+ {
+       struct stat st;
+       struct cache_entry *ce;
+       int namelen = strlen(path);
+       unsigned ce_size = cache_entry_size(namelen);
+       if (!write_index)
+               return;
+       ce = xcalloc(1, ce_size);
+       memcpy(ce->name, path, namelen);
+       ce->ce_mode = create_ce_mode(mode);
+       ce->ce_flags = htons(namelen);
+       if (!cached) {
+               if (lstat(path, &st) < 0)
+                       die("unable to stat newly created file %s", path);
+               fill_stat_cache_info(ce, &st);
+       }
+       if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
+               die("unable to create backing store for newly created file %s", path);
+       if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
+               die("unable to add cache entry for %s", path);
+ }
+ static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
+ {
+       int fd;
+       if (S_ISLNK(mode))
+               return symlink(buf, path);
+       fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
+       if (fd < 0)
+               return -1;
+       while (size) {
+               int written = xwrite(fd, buf, size);
+               if (written < 0)
+                       die("writing file %s: %s", path, strerror(errno));
+               if (!written)
+                       die("out of space writing file %s", path);
+               buf += written;
+               size -= written;
+       }
+       if (close(fd) < 0)
+               die("closing file %s: %s", path, strerror(errno));
+       return 0;
+ }
+ /*
+  * We optimistically assume that the directories exist,
+  * which is true 99% of the time anyway. If they don't,
+  * we create them and try again.
+  */
+ static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
+ {
+       if (cached)
+               return;
+       if (!try_create_file(path, mode, buf, size))
+               return;
+       if (errno == ENOENT) {
+               if (safe_create_leading_directories(path))
+                       return;
+               if (!try_create_file(path, mode, buf, size))
+                       return;
+       }
+       if (errno == EEXIST) {
+               unsigned int nr = getpid();
+               for (;;) {
+                       const char *newpath;
+                       newpath = mkpath("%s~%u", path, nr);
+                       if (!try_create_file(newpath, mode, buf, size)) {
+                               if (!rename(newpath, path))
+                                       return;
+                               unlink(newpath);
+                               break;
+                       }
+                       if (errno != EEXIST)
+                               break;
+                       ++nr;
+               }
+       }
+       die("unable to write file %s mode %o", path, mode);
+ }
+ static void create_file(struct patch *patch)
+ {
+       char *path = patch->new_name;
+       unsigned mode = patch->new_mode;
+       unsigned long size = patch->resultsize;
+       char *buf = patch->result;
+       if (!mode)
+               mode = S_IFREG | 0644;
+       create_one_file(path, mode, buf, size); 
+       add_index_file(path, mode, buf, size);
+ }
+ static void write_out_one_result(struct patch *patch)
+ {
+       if (patch->is_delete > 0) {
+               remove_file(patch);
+               return;
+       }
+       if (patch->is_new > 0 || patch->is_copy) {
+               create_file(patch);
+               return;
+       }
+       /*
+        * Rename or modification boils down to the same
+        * thing: remove the old, write the new
+        */
+       remove_file(patch);
+       create_file(patch);
+ }
+ static void write_out_results(struct patch *list, int skipped_patch)
+ {
+       if (!list && !skipped_patch)
+               die("No changes");
+       while (list) {
+               write_out_one_result(list);
+               list = list->next;
+       }
+ }
+ static struct cache_file cache_file;
+ static struct excludes {
+       struct excludes *next;
+       const char *path;
+ } *excludes;
+ static int use_patch(struct patch *p)
+ {
+       const char *pathname = p->new_name ? p->new_name : p->old_name;
+       struct excludes *x = excludes;
+       while (x) {
+               if (fnmatch(x->path, pathname, 0) == 0)
+                       return 0;
+               x = x->next;
+       }
+       if (0 < prefix_length) {
+               int pathlen = strlen(pathname);
+               if (pathlen <= prefix_length ||
+                   memcmp(prefix, pathname, prefix_length))
+                       return 0;
+       }
+       return 1;
+ }
+ static int apply_patch(int fd, const char *filename)
+ {
+       unsigned long offset, size;
+       char *buffer = read_patch_file(fd, &size);
+       struct patch *list = NULL, **listp = &list;
+       int skipped_patch = 0;
+       patch_input_file = filename;
+       if (!buffer)
+               return -1;
+       offset = 0;
+       while (size > 0) {
+               struct patch *patch;
+               int nr;
+               patch = xcalloc(1, sizeof(*patch));
+               nr = parse_chunk(buffer + offset, size, patch);
+               if (nr < 0)
+                       break;
+               if (use_patch(patch)) {
+                       patch_stats(patch);
+                       *listp = patch;
+                       listp = &patch->next;
+               } else {
+                       /* perhaps free it a bit better? */
+                       free(patch);
+                       skipped_patch++;
+               }
+               offset += nr;
+               size -= nr;
+       }
+       if (whitespace_error && (new_whitespace == error_on_whitespace))
+               apply = 0;
+       write_index = check_index && apply;
+       if (write_index && newfd < 0)
+               newfd = hold_index_file_for_update(&cache_file, get_index_file());
+       if (check_index) {
+               if (read_cache() < 0)
+                       die("unable to read index file");
+       }
+       if ((check || apply) && check_patch_list(list) < 0)
+               exit(1);
+       if (apply)
+               write_out_results(list, skipped_patch);
+       if (show_index_info)
+               show_index_list(list);
+       if (diffstat)
+               stat_patch_list(list);
+       if (numstat)
+               numstat_patch_list(list);
+       if (summary)
+               summary_patch_list(list);
+       free(buffer);
+       return 0;
+ }
+ static int git_apply_config(const char *var, const char *value)
+ {
+       if (!strcmp(var, "apply.whitespace")) {
+               apply_default_whitespace = strdup(value);
+               return 0;
+       }
+       return git_default_config(var, value);
+ }
+ int cmd_apply(int argc, const char **argv, char **envp)
+ {
+       int i;
+       int read_stdin = 1;
+       const char *whitespace_option = NULL;
+       for (i = 1; i < argc; i++) {
+               const char *arg = argv[i];
+               char *end;
+               int fd;
+               if (!strcmp(arg, "-")) {
+                       apply_patch(0, "<stdin>");
+                       read_stdin = 0;
+                       continue;
+               }
+               if (!strncmp(arg, "--exclude=", 10)) {
+                       struct excludes *x = xmalloc(sizeof(*x));
+                       x->path = arg + 10;
+                       x->next = excludes;
+                       excludes = x;
+                       continue;
+               }
+               if (!strncmp(arg, "-p", 2)) {
+                       p_value = atoi(arg + 2);
+                       continue;
+               }
+               if (!strcmp(arg, "--no-add")) {
+                       no_add = 1;
+                       continue;
+               }
+               if (!strcmp(arg, "--stat")) {
+                       apply = 0;
+                       diffstat = 1;
+                       continue;
+               }
+               if (!strcmp(arg, "--allow-binary-replacement") ||
+                   !strcmp(arg, "--binary")) {
+                       allow_binary_replacement = 1;
+                       continue;
+               }
+               if (!strcmp(arg, "--numstat")) {
+                       apply = 0;
+                       numstat = 1;
+                       continue;
+               }
+               if (!strcmp(arg, "--summary")) {
+                       apply = 0;
+                       summary = 1;
+                       continue;
+               }
+               if (!strcmp(arg, "--check")) {
+                       apply = 0;
+                       check = 1;
+                       continue;
+               }
+               if (!strcmp(arg, "--index")) {
+                       check_index = 1;
+                       continue;
+               }
+               if (!strcmp(arg, "--cached")) {
+                       check_index = 1;
+                       cached = 1;
+                       continue;
+               }
+               if (!strcmp(arg, "--apply")) {
+                       apply = 1;
+                       continue;
+               }
+               if (!strcmp(arg, "--index-info")) {
+                       apply = 0;
+                       show_index_info = 1;
+                       continue;
+               }
+               if (!strcmp(arg, "-z")) {
+                       line_termination = 0;
+                       continue;
+               }
+               if (!strncmp(arg, "-C", 2)) {
+                       p_context = strtoul(arg + 2, &end, 0);
+                       if (*end != '\0')
+                               die("unrecognized context count '%s'", arg + 2);
+                       continue;
+               }
+               if (!strncmp(arg, "--whitespace=", 13)) {
+                       whitespace_option = arg + 13;
+                       parse_whitespace_option(arg + 13);
+                       continue;
+               }
+               if (check_index && prefix_length < 0) {
+                       prefix = setup_git_directory();
+                       prefix_length = prefix ? strlen(prefix) : 0;
+                       git_config(git_apply_config);
+                       if (!whitespace_option && apply_default_whitespace)
+                               parse_whitespace_option(apply_default_whitespace);
+               }
+               if (0 < prefix_length)
+                       arg = prefix_filename(prefix, prefix_length, arg);
+               fd = open(arg, O_RDONLY);
+               if (fd < 0)
+                       usage(apply_usage);
+               read_stdin = 0;
+               set_default_whitespace_mode(whitespace_option);
+               apply_patch(fd, arg);
+               close(fd);
+       }
+       set_default_whitespace_mode(whitespace_option);
+       if (read_stdin)
+               apply_patch(0, "<stdin>");
+       if (whitespace_error) {
+               if (squelch_whitespace_errors &&
+                   squelch_whitespace_errors < whitespace_error) {
+                       int squelched =
+                               whitespace_error - squelch_whitespace_errors;
+                       fprintf(stderr, "warning: squelched %d whitespace error%s\n",
+                               squelched,
+                               squelched == 1 ? "" : "s");
+               }
+               if (new_whitespace == error_on_whitespace)
+                       die("%d line%s add%s trailing whitespaces.",
+                           whitespace_error,
+                           whitespace_error == 1 ? "" : "s",
+                           whitespace_error == 1 ? "s" : "");
+               if (applied_after_stripping)
+                       fprintf(stderr, "warning: %d line%s applied after"
+                               " stripping trailing whitespaces.\n",
+                               applied_after_stripping,
+                               applied_after_stripping == 1 ? "" : "s");
+               else if (whitespace_error)
+                       fprintf(stderr, "warning: %d line%s add%s trailing"
+                               " whitespaces.\n",
+                               whitespace_error,
+                               whitespace_error == 1 ? "" : "s",
+                               whitespace_error == 1 ? "s" : "");
+       }
+       if (write_index) {
+               if (write_cache(newfd, active_cache, active_nr) ||
+                   commit_index_file(&cache_file))
+                       die("Unable to write new cachefile");
+       }
+       return 0;
+ }