pack-objects: improve path grouping heuristics.
[git.git] / merge-tree.c
index 00ae4e8..9dcaab7 100644 (file)
 #include "cache.h"
+#include "tree-walk.h"
 
-struct tree_entry {
-       unsigned mode;
-       unsigned char *sha1;
-       char *path;
-       struct tree_entry *next;
-};
+static const char merge_tree_usage[] = "git-merge-tree <base-tree> <branch1> <branch2>";
+static int resolve_directories = 1;
 
-static struct tree_entry *read_tree(unsigned char *sha1)
+static void merge_trees(struct tree_desc t[3], const char *base);
+
+/* An empty entry never compares same, not even to another empty entry */
+static int same_entry(struct name_entry *a, struct name_entry *b)
 {
-       char type[20];
-       unsigned long size;
-       void *buf = read_sha1_file(sha1, type, &size);
-       struct tree_entry *ret = NULL, **tp = &ret;
-
-       if (!buf || strcmp(type, "tree"))
-               die("unable to read 'tree' object %s", sha1_to_hex(sha1));
-       while (size) {
-               int len = strlen(buf)+1;
-               struct tree_entry * entry = malloc(sizeof(struct tree_entry));
-               if (size < len + 20 || sscanf(buf, "%o", &entry->mode) != 1)
-                       die("corrupt 'tree' object %s", sha1_to_hex(sha1));
-               entry->path = strchr(buf, ' ')+1;
-               entry->sha1 = buf + len;
-               entry->next = NULL;
-               *tp = entry;
-               tp = &entry->next;
-               len += 20;
-               buf += len;
-               size -= len;
-       }
-       return ret;
+       return  a->sha1 &&
+               b->sha1 &&
+               !memcmp(a->sha1, b->sha1, 20) &&
+               a->mode == b->mode;
 }
 
-static void show(const struct tree_entry *a, const char *path)
+static const char *sha1_to_hex_zero(const unsigned char *sha1)
 {
-       printf("select %o %s %s%c", a->mode, sha1_to_hex(a->sha1), path, 0);
+       if (sha1)
+               return sha1_to_hex(sha1);
+       return "0000000000000000000000000000000000000000";
 }
 
-static void merge(const struct tree_entry *a, const struct tree_entry *b, const struct tree_entry *c, const char *path)
+static void resolve(const char *base, struct name_entry *branch1, struct name_entry *result)
 {
-       char hex_a[60], hex_b[60], hex_c[60];
-       strcpy(hex_a, sha1_to_hex(a->sha1));
-       strcpy(hex_b, sha1_to_hex(b->sha1));
-       strcpy(hex_c, sha1_to_hex(c->sha1));
-       printf("merge %o->%o,%o %s->%s,%s %s%c",
-               a->mode, b->mode, c->mode,
-               hex_a, hex_b, hex_c, path, 0);
+       /* If it's already branch1, don't bother showing it */
+       if (!branch1)
+               return;
+
+       printf("0 %06o->%06o %s->%s %s%s\n",
+               branch1->mode, result->mode,
+               sha1_to_hex_zero(branch1->sha1),
+               sha1_to_hex_zero(result->sha1),
+               base, result->path);
 }
 
-static int same(const struct tree_entry *a, const struct tree_entry *b)
+static int unresolved_directory(const char *base, struct name_entry n[3])
 {
-       return a->mode == b->mode && !memcmp(a->sha1, b->sha1, 20);
+       int baselen;
+       char *newbase;
+       struct name_entry *p;
+       struct tree_desc t[3];
+       void *buf0, *buf1, *buf2;
+
+       if (!resolve_directories)
+               return 0;
+       p = n;
+       if (!p->mode) {
+               p++;
+               if (!p->mode)
+                       p++;
+       }
+       if (!S_ISDIR(p->mode))
+               return 0;
+       baselen = strlen(base);
+       newbase = xmalloc(baselen + p->pathlen + 2);
+       memcpy(newbase, base, baselen);
+       memcpy(newbase + baselen, p->path, p->pathlen);
+       memcpy(newbase + baselen + p->pathlen, "/", 2);
+
+       buf0 = fill_tree_descriptor(t+0, n[0].sha1);
+       buf1 = fill_tree_descriptor(t+1, n[1].sha1);
+       buf2 = fill_tree_descriptor(t+2, n[2].sha1);
+       merge_trees(t, newbase);
+
+       free(buf0);
+       free(buf1);
+       free(buf2);
+       free(newbase);
+       return 1;
 }
 
-static void merge_entry(const struct tree_entry *src, const struct tree_entry *dst1, const struct tree_entry *dst2)
+static void unresolved(const char *base, struct name_entry n[3])
 {
-       static unsigned char nullsha1[20];
-       static const struct tree_entry none = { 0, nullsha1, "", NULL };
-       const char *path = NULL;
-       const struct tree_entry *a, *b, *c;
-
-       a = &none;
-       b = &none;
-       c = &none;
-       if (src) { a = src; path = src->path; }
-       if (dst1) { b = dst1; path = dst1->path; }
-       if (dst2) { c = dst2; path = dst2->path; }
-       if (same(b, c)) {
-               show(b, path);
+       if (unresolved_directory(base, n))
                return;
+       if (n[0].sha1)
+               printf("1 %06o %s %s%s\n", n[0].mode, sha1_to_hex(n[0].sha1), base, n[0].path);
+       if (n[1].sha1)
+               printf("2 %06o %s %s%s\n", n[1].mode, sha1_to_hex(n[1].sha1), base, n[1].path);
+       if (n[2].sha1)
+               printf("3 %06o %s %s%s\n", n[2].mode, sha1_to_hex(n[2].sha1), base, n[2].path);
+}
+
+/*
+ * Merge two trees together (t[1] and t[2]), using a common base (t[0])
+ * as the origin.
+ *
+ * This walks the (sorted) trees in lock-step, checking every possible
+ * name. Note that directories automatically sort differently from other
+ * files (see "base_name_compare"), so you'll never see file/directory
+ * conflicts, because they won't ever compare the same.
+ *
+ * IOW, if a directory changes to a filename, it will automatically be
+ * seen as the directory going away, and the filename being created.
+ *
+ * Think of this as a three-way diff.
+ *
+ * The output will be either:
+ *  - successful merge
+ *      "0 mode sha1 filename"
+ *    NOTE NOTE NOTE! FIXME! We really really need to walk the index
+ *    in parallel with this too!
+ *
+ *  - conflict:
+ *     "1 mode sha1 filename"
+ *     "2 mode sha1 filename"
+ *     "3 mode sha1 filename"
+ *    where not all of the 1/2/3 lines may exist, of course.
+ *
+ * The successful merge rules are the same as for the three-way merge
+ * in git-read-tree.
+ */
+static void threeway_callback(int n, unsigned long mask, struct name_entry *entry, const char *base)
+{
+       /* Same in both? */
+       if (same_entry(entry+1, entry+2)) {
+               if (entry[0].sha1) {
+                       resolve(base, NULL, entry+1);
+                       return;
+               }
        }
-       if (same(a, b)) {
-               show(c, path);
-               return;
+
+       if (same_entry(entry+0, entry+1)) {
+               if (entry[2].sha1 && !S_ISDIR(entry[2].mode)) {
+                       resolve(base, entry+1, entry+2);
+                       return;
+               }
        }
-       if (same(a, c)) {
-               show(b, path);
-               return;
+
+       if (same_entry(entry+0, entry+2)) {
+               if (entry[1].sha1 && !S_ISDIR(entry[1].mode)) {
+                       resolve(base, NULL, entry+1);
+                       return;
+               }
        }
-       merge(a, b, c, path);
+
+       unresolved(base, entry);
 }
 
-/* For two entries, select the smaller one, clear the bigger one */
-static void smaller(struct tree_entry **ap, struct tree_entry **bp)
+static void merge_trees(struct tree_desc t[3], const char *base)
 {
-       struct tree_entry *a = *ap, *b = *bp;
-       if (a && b) {
-               int cmp = cache_name_compare(a->path, strlen(a->path), b->path, strlen(b->path));
-               if (cmp) {
-                       if (cmp < 0)
-                               *bp = NULL;
-                       else
-                               *ap = NULL;
-               }
-       }
+       traverse_trees(3, t, base, threeway_callback);
 }
 
-static void merge_tree(struct tree_entry *src, struct tree_entry *dst1, struct tree_entry *dst2)
+static void *get_tree_descriptor(struct tree_desc *desc, const char *rev)
 {
-       while (src || dst1 || dst2) {
-               struct tree_entry *a, *b, *c;
-               a = src;
-               b = dst1;
-               c = dst2;
-               smaller(&a,&b);
-               smaller(&a,&c);
-               smaller(&b,&c);
-               if (a) src = a->next;
-               if (b) dst1 = b->next;
-               if (c) dst2 = c->next;
-               merge_entry(a,b,c);
-       }
+       unsigned char sha1[20];
+       void *buf;
+
+       if (get_sha1(rev, sha1))
+               die("unknown rev %s", rev);
+       buf = fill_tree_descriptor(desc, sha1);
+       if (!buf)
+               die("%s is not a tree", rev);
+       return buf;
 }
 
 int main(int argc, char **argv)
 {
-       unsigned char src[20], dst1[20], dst2[20];
-
-       if (argc != 4 ||
-           get_sha1_hex(argv[1], src) ||
-           get_sha1_hex(argv[2], dst1) ||
-           get_sha1_hex(argv[3], dst2))
-               usage("merge-tree <src> <dst1> <dst2>");
-       merge_tree(read_tree(src), read_tree(dst1), read_tree(dst2));
+       struct tree_desc t[3];
+       void *buf1, *buf2, *buf3;
+
+       if (argc < 4)
+               usage(merge_tree_usage);
+
+       buf1 = get_tree_descriptor(t+0, argv[1]);
+       buf2 = get_tree_descriptor(t+1, argv[2]);
+       buf3 = get_tree_descriptor(t+2, argv[3]);
+       merge_trees(t, "");
+       free(buf1);
+       free(buf2);
+       free(buf3);
        return 0;
 }