git-apply: parse the whole list of patches into memory first
[gitweb.git] / apply.c
diff --git a/apply.c b/apply.c
index 05f921a311e51166f824fb3e3e25c4e71ae712c7..6a5aa6749a6e124d8fb95505037d73e78fae3860 100644 (file)
--- a/apply.c
+++ b/apply.c
 static int merge_patch = 1;
 static const char apply_usage[] = "git-apply <patch>";
 
+/*
+ * 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;
+static char *def_name = NULL;
+
+struct fragment {
+       unsigned long oldpos, oldlines;
+       unsigned long newpos, newlines;
+       const char *patch;
+       int size;
+       struct fragment *next;
+};
+
+struct patch {
+       char *new_name, *old_name;
+       unsigned int old_mode, new_mode;
+       int is_rename, is_copy, is_new, is_delete;
+       struct fragment *fragments;
+       struct patch *next;
+};
 
 #define CHUNKSIZE (8192)
+#define SLOP (16)
 
 static void *read_patch_file(int fd, unsigned long *sizep)
 {
@@ -48,6 +71,15 @@ static void *read_patch_file(int fd, unsigned long *sizep)
                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;
 }
 
@@ -62,48 +94,286 @@ static unsigned long linelen(char *buffer, unsigned long size)
        return len;
 }
 
-static int match_word(const char *line, const char *match)
+static int is_dev_null(const char *str)
+{
+       return !memcmp("/dev/null", str, 9) && isspace(str[9]);
+}
+
+#define TERM_EXIST     1
+#define TERM_SPACE     2
+#define TERM_TAB       4
+
+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;
+
+       /*
+        * Do we want an existing name? Return false and
+        * continue if it's not there.
+        */
+       if (terminate & TERM_EXIST)
+               return cache_name_pos(name, namelen) >= 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;
+
        for (;;) {
-               char c = *match++;
-               if (!c)
+               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)
+{
+       int p_value = 1;
+       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, def_name, 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, def_name, p_value, TERM_EXIST | TERM_SPACE | TERM_TAB);
+               patch->old_name = name;
+       } else {
+               name = find_name(first, def_name, p_value, TERM_EXIST | TERM_SPACE | TERM_TAB);
+               name = find_name(second, name, p_value, TERM_EXIST | 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)
+{
+       int len;
+       const char *name;
+
+       if (!orig_name && !isnull)
+               return find_name(line, NULL, 1, 0);
+
+       name = "/dev/null";
+       len = 9;
+       if (orig_name) {
+               name = orig_name;
+               len = strlen(name);
+               if (isnull)
+                       die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
+       }
+
+       if (*name == '/')
+               goto absolute_path;
+
+       for (;;) {
+               char c = *line++;
+               if (c == '\n')
                        break;
-               if (*line++ != c)
-                       return 0;
+               if (c != '/')
+                       continue;
+absolute_path:
+               if (memcmp(line, name, len) || line[len] != '\n')
+                       break;
+               return orig_name;
        }
-       return *line == ' ';
+       die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, 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;
+       return gitdiff_oldmode(line, patch);
+}
+
+static int gitdiff_newfile(const char *line, struct patch *patch)
+{
+       patch->is_new = 1;
+       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)
+{
+       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;
 }
 
 /* Verify that we recognize the lines following a git header */
-static int parse_git_header(char *line, unsigned int size)
+static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
 {
-       unsigned long offset, len;
+       unsigned long offset;
+
+       /* A git diff has explicit new/delete information, so we don't guess */
+       patch->is_new = 0;
+       patch->is_delete = 0;
+
+       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 from ", gitdiff_renamesrc },
+                       { "rename to ", gitdiff_renamedst },
+                       { "similarity index ", gitdiff_similarity },
+                       { "", gitdiff_unrecognized },
+               };
+               int i;
 
-       for (offset = 0 ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
                len = linelen(line, size);
-               if (!len)
+               if (!len || line[len-1] != '\n')
                        break;
-               if (line[len-1] != '\n')
-                       return -1;
-               if (len < 4)
+               for (i = 0; i < sizeof(optable) / sizeof(optable[0]); 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;
-               if (!memcmp(line, "@@ -", 4))
-                       return offset;
-               if (match_word(line, "new file mode"))
-                       continue;
-               if (match_word(line, "deleted file mode"))
-                       continue;
-               if (match_word(line, "copy"))
-                       continue;
-               if (match_word(line, "rename"))
-                       continue;
-               if (match_word(line, "similarity index"))
-                       continue;
-               break;
+               }
        }
 
-       /* We want either a patch _or_ something real */
-       return offset ? :-1;
+       return offset;
 }
 
 static int parse_num(const char *line, int len, int offset, const char *expect, unsigned long *p)
@@ -139,7 +409,7 @@ static int parse_num(const char *line, int len, int offset, const char *expect,
  * Parse a unified diff fragment header of the
  * form "@@ -a,b +c,d @@"
  */
-static int parse_fragment_header(char *line, int len, unsigned long *pos)
+static int parse_fragment_header(char *line, int len, struct fragment *fragment)
 {
        int offset;
 
@@ -147,18 +417,22 @@ static int parse_fragment_header(char *line, int len, unsigned long *pos)
                return -1;
 
        /* Figure out the number of lines in a fragment */
-       offset = parse_num(line, len, 4, ",", pos);
-       offset = parse_num(line, len, offset, " +", pos+1);
-       offset = parse_num(line, len, offset, ",", pos+2);
-       offset = parse_num(line, len, offset, " @@", pos+3);
+       offset = parse_num(line, len, 4, ",", &fragment->oldpos);
+       offset = parse_num(line, len, offset, " +", &fragment->oldlines);
+       offset = parse_num(line, len, offset, ",", &fragment->newpos);
+       offset = parse_num(line, len, offset, " @@", &fragment->newlines);
 
        return offset;
 }
 
-static int find_header(char *line, unsigned long size, int *hdrsize)
+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;
 
@@ -176,8 +450,8 @@ static int find_header(char *line, unsigned long size, int *hdrsize)
                 * patch has become corrupted/broken up.
                 */
                if (!memcmp("@@ -", line, 4)) {
-                       unsigned long pos[4];
-                       if (parse_fragment_header(line, len, pos) < 0)
+                       struct fragment dummy;
+                       if (parse_fragment_header(line, len, &dummy) < 0)
                                continue;
                        error("patch fragment without header at line %d: %.*s", linenr, len-1, line);
                }
@@ -190,11 +464,11 @@ static int find_header(char *line, unsigned long size, int *hdrsize)
                 * or mode change, so we handle that specially
                 */
                if (!memcmp("diff --git ", line, 11)) {
-                       int git_hdr_len = parse_git_header(line + len, size - len);
+                       int git_hdr_len = parse_git_header(line, len, size, patch);
                        if (git_hdr_len < 0)
                                continue;
 
-                       *hdrsize = len + git_hdr_len;
+                       *hdrsize = git_hdr_len;
                        return offset;
                }
 
@@ -212,6 +486,7 @@ static int find_header(char *line, unsigned long size, int *hdrsize)
                        continue;
 
                /* Ok, we'll consider it a patch */
+               parse_traditional_patch(line, line+len, patch);
                *hdrsize = len + nextlen;
                linenr += 2;
                return offset;
@@ -226,16 +501,21 @@ static int find_header(char *line, unsigned long size, int *hdrsize)
  * part of a patch, and a "---" that starts the next
  * patch is to look at the line counts..
  */
-static int apply_fragment(char *line, unsigned long size)
+static int parse_fragment(char *line, unsigned long size, struct patch *patch, struct fragment *fragment)
 {
        int len = linelen(line, size), offset;
        unsigned long pos[4], oldlines, newlines;
 
-       offset = parse_fragment_header(line, len, pos);
+       offset = parse_fragment_header(line, len, fragment);
        if (offset < 0)
                return -1;
-       oldlines = pos[1];
-       newlines = pos[3];
+       oldlines = fragment->oldlines;
+       newlines = fragment->newlines;
+
+       if (patch->is_new < 0 && (pos[0] || oldlines))
+               patch->is_new = 0;
+       if (patch->is_delete < 0 && (pos[1] || newlines))
+               patch->is_delete = 0;
 
        /* Parse the thing.. */
        line += len;
@@ -265,16 +545,26 @@ static int apply_fragment(char *line, unsigned long size)
        return offset;
 }
 
-static int apply_single_patch(char *line, unsigned long size)
+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)) {
-               int len = apply_fragment(line, size);
+               struct fragment *fragment;
+               int len;
+
+               fragment = xmalloc(sizeof(*fragment));
+               memset(fragment, 0, sizeof(*fragment));
+               len = parse_fragment(line, size, patch, fragment);
                if (len <= 0)
                        die("corrupt patch at line %d", linenr);
 
-printf("applying fragment:\n%.*s\n\n", len, line);
+               fragment->patch = line;
+               fragment->size = len;
+
+               *fragp = fragment;
+               fragp = &fragment->next;
 
                offset += len;
                line += len;
@@ -283,39 +573,82 @@ printf("applying fragment:\n%.*s\n\n", len, line);
        return offset;
 }
 
-static int apply_chunk(char *buffer, unsigned long size)
+static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
 {
        int hdrsize, patchsize;
-       int offset = find_header(buffer, size, &hdrsize);
-       char *header, *patch;
+       int offset = find_header(buffer, size, &hdrsize, patch);
 
        if (offset < 0)
                return offset;
-       header = buffer + offset;
-
-printf("Found header:\n%.*s\n\n", hdrsize, header);
 
-       patch = header + hdrsize;
-       patchsize = apply_single_patch(patch, size - offset - hdrsize);
+       patchsize = parse_single_patch(buffer + offset + hdrsize, size - offset - hdrsize, patch);
 
        return offset + hdrsize + patchsize;
 }
 
+static void apply_patch_list(struct patch *patch)
+{
+       if (!patch)
+               die("no patch found");
+       do {
+               const char *old_name = patch->old_name;
+               const char *new_name = patch->new_name;
+               struct fragment *frag;
+
+               if (old_name) {
+                       if (cache_name_pos(old_name, strlen(old_name)) < 0)
+                               die("file %s does not exist", old_name);
+                       if (patch->is_new < 0)
+                               patch->is_new = 0;
+               }
+               if (new_name && (patch->is_new | patch->is_rename | patch->is_copy)) {
+                       if (cache_name_pos(new_name, strlen(new_name)) >= 0)
+                               die("file %s already exists", new_name);
+               }
+
+               printf("Applying patch to %s\n", new_name);
+               printf("  new=%d delete=%d copy=%d rename=%d\n",
+                       patch->is_new, patch->is_delete, patch->is_copy, patch->is_rename);
+               if (patch->old_mode != patch->new_mode)
+                       printf("  %o->%o\n", patch->old_mode, patch->new_mode);
+               frag = patch->fragments;
+               while (frag) {
+                       printf("Fragment %lu,%lu -> %lu,%lu\n%.*s",
+                               frag->oldpos, frag->oldlines,
+                               frag->newpos, frag->newlines,
+                               frag->size, frag->patch);
+                       frag = frag->next;
+               }
+
+       } while ((patch = patch->next) != NULL);
+}
+
 static int apply_patch(int fd)
 {
        unsigned long offset, size;
        char *buffer = read_patch_file(fd, &size);
+       struct patch *list = NULL, **listp = &list;
 
        if (!buffer)
                return -1;
        offset = 0;
        while (size > 0) {
-               int nr = apply_chunk(buffer + offset, size);
+               struct patch *patch;
+               int nr;
+
+               patch = xmalloc(sizeof(*patch));
+               memset(patch, 0, sizeof(*patch));
+               nr = parse_chunk(buffer + offset, size, patch);
                if (nr < 0)
                        break;
+               *listp = patch;
+               listp = &patch->next;
                offset += nr;
                size -= nr;
        }
+
+       apply_patch_list(list);
+
        free(buffer);
        return 0;
 }
@@ -323,6 +656,7 @@ static int apply_patch(int fd)
 int main(int argc, char **argv)
 {
        int i;
+       int read_stdin = 1;
 
        if (read_cache() < 0)
                die("unable to read index file");
@@ -333,6 +667,7 @@ int main(int argc, char **argv)
 
                if (!strcmp(arg, "-")) {
                        apply_patch(0);
+                       read_stdin = 0;
                        continue;
                }
                if (!strcmp(arg, "--no-merge")) {
@@ -342,8 +677,11 @@ int main(int argc, char **argv)
                fd = open(arg, O_RDONLY);
                if (fd < 0)
                        usage(apply_usage);
+               read_stdin = 0;
                apply_patch(fd);
                close(fd);
        }
+       if (read_stdin)
+               apply_patch(0);
        return 0;
 }