Merge branch 'jc/fix-tree-walk'
authorJunio C Hamano <gitster@pobox.com>
Mon, 25 Jan 2010 01:35:58 +0000 (17:35 -0800)
committerJunio C Hamano <gitster@pobox.com>
Mon, 25 Jan 2010 01:35:58 +0000 (17:35 -0800)
* jc/fix-tree-walk:
read-tree --debug-unpack
unpack-trees.c: look ahead in the index
unpack-trees.c: prepare for looking ahead in the index
Aggressive three-way merge: fix D/F case
traverse_trees(): handle D/F conflict case sanely
more D/F conflict tests
tests: move convenience regexp to match object names to test-lib.sh

Conflicts:
builtin-read-tree.c
unpack-trees.c
unpack-trees.h

19 files changed:
builtin-read-tree.c
cache.h
diff-lib.c
diff.c
diff.h
t/diff-lib.sh
t/t1000-read-tree-m-3way.sh
t/t1001-read-tree-m-2way.sh
t/t1002-read-tree-m-u-2way.sh
t/t1012-read-tree-df.sh [new file with mode: 0755]
t/t3100-ls-tree-restrict.sh
t/t3101-ls-tree-dirname.sh
t/t4006-diff-mode.sh
t/t6012-rev-list-simplify.sh
t/t6035-merge-dir-to-symlink.sh
t/test-lib.sh
tree-walk.c
unpack-trees.c
unpack-trees.h
index 5fda9905fcfa5ccbc0f28578cd371314a8a7e5a7..8bdcab11389e8b4facab355f27f0e1926134b86f 100644 (file)
@@ -65,6 +65,34 @@ static int exclude_per_directory_cb(const struct option *opt, const char *arg,
        return 0;
 }
 
+static void debug_stage(const char *label, struct cache_entry *ce,
+                       struct unpack_trees_options *o)
+{
+       printf("%s ", label);
+       if (!ce)
+               printf("(missing)\n");
+       else if (ce == o->df_conflict_entry)
+               printf("(conflict)\n");
+       else
+               printf("%06o #%d %s %.8s\n",
+                      ce->ce_mode, ce_stage(ce), ce->name,
+                      sha1_to_hex(ce->sha1));
+}
+
+static int debug_merge(struct cache_entry **stages, struct unpack_trees_options *o)
+{
+       int i;
+
+       printf("* %d-way merge\n", o->merge_size);
+       debug_stage("index", stages[0], o);
+       for (i = 1; i <= o->merge_size; i++) {
+               char buf[24];
+               sprintf(buf, "ent#%d", i);
+               debug_stage(buf, stages[i], o);
+       }
+       return 0;
+}
+
 static struct lock_file lock_file;
 
 int cmd_read_tree(int argc, const char **argv, const char *unused_prefix)
@@ -101,6 +129,8 @@ int cmd_read_tree(int argc, const char **argv, const char *unused_prefix)
                            "don't check the working tree after merging", 1),
                OPT_SET_INT(0, "no-sparse-checkout", &opts.skip_sparse_checkout,
                            "skip applying sparse checkout filter", 1),
+               OPT_SET_INT(0, "debug-unpack", &opts.debug_unpack,
+                           "debug unpack-trees", 1),
                OPT_END()
        };
 
@@ -169,6 +199,9 @@ int cmd_read_tree(int argc, const char **argv, const char *unused_prefix)
                        opts.head_idx = 1;
        }
 
+       if (opts.debug_unpack)
+               opts.fn = debug_merge;
+
        cache_tree_free(&active_cache_tree);
        for (i = 0; i < nr_trees; i++) {
                struct tree *tree = trees[i];
@@ -178,6 +211,9 @@ int cmd_read_tree(int argc, const char **argv, const char *unused_prefix)
        if (unpack_trees(nr_trees, t, &opts))
                return 128;
 
+       if (opts.debug_unpack)
+               return 0; /* do not write the index out */
+
        /*
         * When reading only one tree (either the most basic form,
         * "-m ent" or "--reset ent" form), we can obtain a fully
diff --git a/cache.h b/cache.h
index 7d5c21e3f1e737504a7f4f8c76781b1552cb77e3..b3370eb41e500b59bdf9d03679e7c17dde59e911 100644 (file)
--- a/cache.h
+++ b/cache.h
@@ -182,6 +182,8 @@ struct cache_entry {
 /* Only remove in work directory, not index */
 #define CE_WT_REMOVE (0x400000)
 
+#define CE_UNPACKED  (0x1000000)
+
 /*
  * Extended on-disk flags
  */
index 23e180eed13230fa234cdcb86db7dbd914abb4a5..ec2e2ac0058a71a58f6e3af57f2bd86520a67779 100644 (file)
@@ -380,21 +380,6 @@ static void do_oneway_diff(struct unpack_trees_options *o,
        show_modified(revs, tree, idx, 1, cached, match_missing);
 }
 
-static inline void skip_same_name(struct cache_entry *ce, struct unpack_trees_options *o)
-{
-       int len = ce_namelen(ce);
-       const struct index_state *index = o->src_index;
-
-       while (o->pos < index->cache_nr) {
-               struct cache_entry *next = index->cache[o->pos];
-               if (len != ce_namelen(next))
-                       break;
-               if (memcmp(ce->name, next->name, len))
-                       break;
-               o->pos++;
-       }
-}
-
 /*
  * The unpack_trees() interface is designed for merging, so
  * the different source entries are designed primarily for
@@ -416,9 +401,6 @@ static int oneway_diff(struct cache_entry **src, struct unpack_trees_options *o)
        struct cache_entry *tree = src[1];
        struct rev_info *revs = o->unpack_data;
 
-       if (idx && ce_stage(idx))
-               skip_same_name(idx, o);
-
        /*
         * Unpack-trees generates a DF/conflict entry if
         * there was a directory in the index and a tree
@@ -464,6 +446,7 @@ int run_diff_index(struct rev_info *revs, int cached)
                exit(128);
 
        diff_set_mnemonic_prefix(&revs->diffopt, "c/", cached ? "i/" : "w/");
+       diffcore_fix_diff_index(&revs->diffopt);
        diffcore_std(&revs->diffopt);
        diff_flush(&revs->diffopt);
        return 0;
diff --git a/diff.c b/diff.c
index 160dbfd7186ebaa34764a40dc7ba7d2338244a73..f130a367dcfe8668d09a0e4a55f96d6b66251db9 100644 (file)
--- a/diff.c
+++ b/diff.c
@@ -3678,6 +3678,23 @@ static void diffcore_skip_stat_unmatch(struct diff_options *diffopt)
        *q = outq;
 }
 
+static int diffnamecmp(const void *a_, const void *b_)
+{
+       const struct diff_filepair *a = *((const struct diff_filepair **)a_);
+       const struct diff_filepair *b = *((const struct diff_filepair **)b_);
+       const char *name_a, *name_b;
+
+       name_a = a->one ? a->one->path : a->two->path;
+       name_b = b->one ? b->one->path : b->two->path;
+       return strcmp(name_a, name_b);
+}
+
+void diffcore_fix_diff_index(struct diff_options *options)
+{
+       struct diff_queue_struct *q = &diff_queued_diff;
+       qsort(q->queue, q->nr, sizeof(q->queue[0]), diffnamecmp);
+}
+
 void diffcore_std(struct diff_options *options)
 {
        if (options->skip_stat_unmatch)
diff --git a/diff.h b/diff.h
index 968a8dce95b93d76e299c070ffa3587c31fcc965..2ef3341fb0852fc8958fa5c5eacab69ee68c0ad9 100644 (file)
--- a/diff.h
+++ b/diff.h
@@ -210,6 +210,7 @@ extern int diff_setup_done(struct diff_options *);
 #define DIFF_PICKAXE_REGEX     2
 
 extern void diffcore_std(struct diff_options *);
+extern void diffcore_fix_diff_index(struct diff_options *);
 
 #define COMMON_DIFF_OPTIONS_HELP \
 "\ncommon diff options:\n" \
index 4bddeb591ecc17ec532164d0d6cf1ad1a54eb996..75a35fcd0699940aea51534b5fe954847bed7bb9 100644 (file)
@@ -1,7 +1,5 @@
 :
 
-_x40='[0-9a-f][0-9a-f][0-9a-f][0-9a-f][0-9a-f]'
-_x40="$_x40$_x40$_x40$_x40$_x40$_x40$_x40$_x40"
 sanitize_diff_raw='/^:/s/ '"$_x40"' '"$_x40"' \([A-Z]\)[0-9]*  / X X \1#       /'
 compare_diff_raw () {
     # When heuristics are improved, the score numbers would change.
index 22ba7a5442c587f4536aad5668df43661231de56..4f171722d9e1dda7668f54b4104267eb70d91c67 100755 (executable)
@@ -126,9 +126,6 @@ cat >expected <<\EOF
 100644 X 0     Z/NN
 EOF
 
-_x40='[0-9a-f][0-9a-f][0-9a-f][0-9a-f][0-9a-f]'
-_x40="$_x40$_x40$_x40$_x40$_x40$_x40$_x40$_x40"
-
 check_result () {
     git ls-files --stage | sed -e 's/ '"$_x40"' / X /' >current &&
     test_cmp expected current
index c2d408b46120ef9d962272b652a68a83ba0852a5..6327d205cb89b00e77502031e3b56b4b895cc32d 100755 (executable)
@@ -26,8 +26,6 @@ read_tree_twoway () {
     git read-tree -m "$1" "$2" && git ls-files --stage
 }
 
-_x40='[0-9a-f][0-9a-f][0-9a-f][0-9a-f][0-9a-f]'
-_x40="$_x40$_x40$_x40$_x40$_x40$_x40$_x40$_x40"
 compare_change () {
        sed -n >current \
            -e '/^--- /d; /^+++ /d; /^@@ /d;' \
index 5e40cec530df07a8b7c088d31b28ac2d39abdc1b..0241329a08af0be642131ffe4b619231d9ea3844 100755 (executable)
@@ -10,8 +10,6 @@ This is identical to t1001, but uses -u to update the work tree as well.
 '
 . ./test-lib.sh
 
-_x40='[0-9a-f][0-9a-f][0-9a-f][0-9a-f][0-9a-f]'
-_x40="$_x40$_x40$_x40$_x40$_x40$_x40$_x40$_x40"
 compare_change () {
        sed >current \
            -e '1{/^diff --git /d;}' \
diff --git a/t/t1012-read-tree-df.sh b/t/t1012-read-tree-df.sh
new file mode 100755 (executable)
index 0000000..9811d46
--- /dev/null
@@ -0,0 +1,102 @@
+#!/bin/sh
+
+test_description='read-tree D/F conflict corner cases'
+
+. ./test-lib.sh
+
+maketree () {
+       (
+               rm -f .git/index .git/index.lock &&
+               git clean -d -f -f -q -x &&
+               name="$1" &&
+               shift &&
+               for it
+               do
+                       path=$(expr "$it" : '\([^:]*\)') &&
+                       mkdir -p $(dirname "$path") &&
+                       echo "$it" >"$path" &&
+                       git update-index --add "$path" || exit
+               done &&
+               git tag "$name" $(git write-tree)
+       )
+}
+
+settree () {
+       rm -f .git/index .git/index.lock &&
+       git clean -d -f -f -q -x &&
+       git read-tree "$1" &&
+       git checkout-index -f -q -u -a &&
+       git update-index --refresh
+}
+
+checkindex () {
+       git ls-files -s |
+       sed "s|^[0-7][0-7]* $_x40 \([0-3]\)     |\1 |" >current &&
+       cat >expect &&
+       test_cmp expect current
+}
+
+test_expect_success setup '
+       maketree O-000 a/b-2/c/d a/b/c/d a/x &&
+       maketree A-000 a/b-2/c/d a/b/c/d a/x &&
+       maketree A-001 a/b-2/c/d a/b/c/d a/b/c/e a/x &&
+       maketree B-000 a/b-2/c/d a/b     a/x &&
+
+       maketree O-010 t-0     t/1  t/2 t=3 &&
+       maketree A-010 t-0 t            t=3 &&
+       maketree B-010         t/1:     t=3: &&
+
+       maketree O-020 ds/dma/ioat.c ds/dma/ioat_dca.c &&
+       maketree A-020 ds/dma/ioat/Makefile ds/dma/ioat/registers.h &&
+       :
+'
+
+test_expect_success '3-way (1)' '
+       settree A-000 &&
+       git read-tree -m -u O-000 A-000 B-000 &&
+       checkindex <<-EOF
+       3 a/b
+       0 a/b-2/c/d
+       1 a/b/c/d
+       2 a/b/c/d
+       0 a/x
+       EOF
+'
+
+test_expect_success '3-way (2)' '
+       settree A-001 &&
+       git read-tree -m -u O-000 A-001 B-000 &&
+       checkindex <<-EOF
+       3 a/b
+       0 a/b-2/c/d
+       1 a/b/c/d
+       2 a/b/c/d
+       2 a/b/c/e
+       0 a/x
+       EOF
+'
+
+test_expect_success '3-way (3)' '
+       settree A-010 &&
+       git read-tree -m -u O-010 A-010 B-010 &&
+       checkindex <<-EOF
+       2 t
+       1 t-0
+       2 t-0
+       1 t/1
+       3 t/1
+       1 t/2
+       0 t=3
+       EOF
+'
+
+test_expect_success '2-way (1)' '
+       settree O-020 &&
+       git read-tree -m -u O-020 A-020 &&
+       checkindex <<-EOF
+       0 ds/dma/ioat/Makefile
+       0 ds/dma/ioat/registers.h
+       EOF
+'
+
+test_done
index ee60d03fe8a582100e9df5511f6fea8900bcc543..eee0d344d24623ad9928366729c8efaea699fcd7 100755 (executable)
@@ -43,8 +43,6 @@ test_expect_success \
      tree=`git write-tree` &&
      echo $tree'
 
-_x40='[0-9a-f][0-9a-f][0-9a-f][0-9a-f][0-9a-f]'
-_x40="$_x40$_x40$_x40$_x40$_x40$_x40$_x40$_x40"
 test_output () {
     sed -e "s/ $_x40   / X     /" <current >check
     test_cmp expected check
index 8be9fb411276e3c416374a813d691ff0455a8746..06654c6f82fb6cfe955b91bcbd4a5fa064c73e70 100755 (executable)
@@ -39,8 +39,6 @@ test_expect_success \
      tree=`git write-tree` &&
      echo $tree'
 
-_x05='[0-9a-f][0-9a-f][0-9a-f][0-9a-f][0-9a-f]'
-_x40="$_x05$_x05$_x05$_x05$_x05$_x05$_x05$_x05"
 test_output () {
     sed -e "s/ $_x40   / X     /" <current >check
     test_cmp expected check
index 8c1b81e248bced2ccb5e4ff0067996462e89deb8..ff8c2f7532aa6fd748dbbdc0322384576f3a35ed 100755 (executable)
@@ -20,8 +20,6 @@ test_expect_success \
     'test_chmod +x rezrov &&
      git diff-index $tree >current'
 
-_x40='[0-9a-f][0-9a-f][0-9a-f][0-9a-f][0-9a-f]'
-_x40="$_x40$_x40$_x40$_x40$_x40$_x40$_x40$_x40"
 sed -e 's/\(:100644 100755\) \('"$_x40"'\) \2 /\1 X X /' <current >check
 echo ":100644 100755 X X M     rezrov" >expected
 
index 510bb9679f29c10645c902caea020651a797aa2f..af34a1e81711ca183680c9882353bd786b55e79f 100755 (executable)
@@ -8,9 +8,6 @@ note () {
        git tag "$1"
 }
 
-_x40='[0-9a-f][0-9a-f][0-9a-f][0-9a-f][0-9a-f]'
-_x40="$_x40$_x40$_x40$_x40$_x40$_x40$_x40$_x40"
-
 unnote () {
        git name-rev --tags --stdin | sed -e "s|$_x40 (tags/\([^)]*\)) |\1 |g"
 }
index 5b96fb0b37b8397342796fd3b9fec94c91063ba4..d1b22871c40674582e6999804102afb95283b653 100755 (executable)
@@ -48,7 +48,7 @@ test_expect_success 'setup for merge test' '
        git tag baseline
 '
 
-test_expect_success 'do not lose a/b-2/c/d in merge (resolve)' '
+test_expect_failure 'do not lose a/b-2/c/d in merge (resolve)' '
        git reset --hard &&
        git checkout baseline^0 &&
        git merge -s resolve master &&
@@ -74,7 +74,7 @@ test_expect_success 'setup a merge where dir a/b-2 changed to symlink' '
        git tag test2
 '
 
-test_expect_failure 'merge should not have conflicts (resolve)' '
+test_expect_success 'merge should not have conflicts (resolve)' '
        git reset --hard &&
        git checkout baseline^0 &&
        git merge -s resolve test2 &&
index c1476f9a2391396a201a919025963b82f7e14217..baa4093073564e081dcf6bc692910257663c9b03 100644 (file)
@@ -74,6 +74,12 @@ case $(echo $GIT_TRACE |tr "[A-Z]" "[a-z]") in
                ;;
 esac
 
+# Convenience
+#
+# A regexp to match 5 and 40 hexdigits
+_x05='[0-9a-f][0-9a-f][0-9a-f][0-9a-f][0-9a-f]'
+_x40="$_x05$_x05$_x05$_x05$_x05$_x05$_x05$_x05"
+
 # Each test should start with something like this, after copyright notices:
 #
 # test_description='Description of this test...
index 02e2aed7737207225f1b96eed774a1b75dd6d8d9..08796c23228fbfb6eb255c479f0196768ee83b27 100644 (file)
@@ -60,13 +60,6 @@ void *fill_tree_descriptor(struct tree_desc *desc, const unsigned char *sha1)
        return buf;
 }
 
-static int entry_compare(struct name_entry *a, struct name_entry *b)
-{
-       return df_name_compare(
-                       a->path, tree_entry_len(a->path, a->sha1), a->mode,
-                       b->path, tree_entry_len(b->path, b->sha1), b->mode);
-}
-
 static void entry_clear(struct name_entry *a)
 {
        memset(a, 0, sizeof(*a));
@@ -138,66 +131,264 @@ char *make_traverse_path(char *path, const struct traverse_info *info, const str
        return path;
 }
 
+struct tree_desc_skip {
+       struct tree_desc_skip *prev;
+       const void *ptr;
+};
+
+struct tree_desc_x {
+       struct tree_desc d;
+       struct tree_desc_skip *skip;
+};
+
+static int name_compare(const char *a, int a_len,
+                       const char *b, int b_len)
+{
+       int len = (a_len < b_len) ? a_len : b_len;
+       int cmp = memcmp(a, b, len);
+       if (cmp)
+               return cmp;
+       return (a_len - b_len);
+}
+
+static int check_entry_match(const char *a, int a_len, const char *b, int b_len)
+{
+       /*
+        * The caller wants to pick *a* from a tree or nothing.
+        * We are looking at *b* in a tree.
+        *
+        * (0) If a and b are the same name, we are trivially happy.
+        *
+        * There are three possibilities where *a* could be hiding
+        * behind *b*.
+        *
+        * (1) *a* == "t",   *b* == "ab"  i.e. *b* sorts earlier than *a* no
+        *                                matter what.
+        * (2) *a* == "t",   *b* == "t-2" and "t" is a subtree in the tree;
+        * (3) *a* == "t-2", *b* == "t"   and "t-2" is a blob in the tree.
+        *
+        * Otherwise we know *a* won't appear in the tree without
+        * scanning further.
+        */
+
+       int cmp = name_compare(a, a_len, b, b_len);
+
+       /* Most common case first -- reading sync'd trees */
+       if (!cmp)
+               return cmp;
+
+       if (0 < cmp) {
+               /* a comes after b; it does not matter if it is case (3)
+               if (b_len < a_len && !memcmp(a, b, b_len) && a[b_len] < '/')
+                       return 1;
+               */
+               return 1; /* keep looking */
+       }
+
+       /* b comes after a; are we looking at case (2)? */
+       if (a_len < b_len && !memcmp(a, b, a_len) && b[a_len] < '/')
+               return 1; /* keep looking */
+
+       return -1; /* a cannot appear in the tree */
+}
+
+/*
+ * From the extended tree_desc, extract the first name entry, while
+ * paying attention to the candidate "first" name.  Most importantly,
+ * when looking for an entry, if there are entries that sorts earlier
+ * in the tree object representation than that name, skip them and
+ * process the named entry first.  We will remember that we haven't
+ * processed the first entry yet, and in the later call skip the
+ * entry we processed early when update_extended_entry() is called.
+ *
+ * E.g. if the underlying tree object has these entries:
+ *
+ *    blob    "t-1"
+ *    blob    "t-2"
+ *    tree    "t"
+ *    blob    "t=1"
+ *
+ * and the "first" asks for "t", remember that we still need to
+ * process "t-1" and "t-2" but extract "t".  After processing the
+ * entry "t" from this call, the caller will let us know by calling
+ * update_extended_entry() that we can remember "t" has been processed
+ * already.
+ */
+
+static void extended_entry_extract(struct tree_desc_x *t,
+                                  struct name_entry *a,
+                                  const char *first,
+                                  int first_len)
+{
+       const char *path;
+       int len;
+       struct tree_desc probe;
+       struct tree_desc_skip *skip;
+
+       /*
+        * Extract the first entry from the tree_desc, but skip the
+        * ones that we already returned in earlier rounds.
+        */
+       while (1) {
+               if (!t->d.size) {
+                       entry_clear(a);
+                       break; /* not found */
+               }
+               entry_extract(&t->d, a);
+               for (skip = t->skip; skip; skip = skip->prev)
+                       if (a->path == skip->ptr)
+                               break; /* found */
+               if (!skip)
+                       break;
+               /* We have processed this entry already. */
+               update_tree_entry(&t->d);
+       }
+
+       if (!first || !a->path)
+               return;
+
+       /*
+        * The caller wants "first" from this tree, or nothing.
+        */
+       path = a->path;
+       len = tree_entry_len(a->path, a->sha1);
+       switch (check_entry_match(first, first_len, path, len)) {
+       case -1:
+               entry_clear(a);
+       case 0:
+               return;
+       default:
+               break;
+       }
+
+       /*
+        * We need to look-ahead -- we suspect that a subtree whose
+        * name is "first" may be hiding behind the current entry "path".
+        */
+       probe = t->d;
+       while (probe.size) {
+               entry_extract(&probe, a);
+               path = a->path;
+               len = tree_entry_len(a->path, a->sha1);
+               switch (check_entry_match(first, first_len, path, len)) {
+               case -1:
+                       entry_clear(a);
+               case 0:
+                       return;
+               default:
+                       update_tree_entry(&probe);
+                       break;
+               }
+               /* keep looking */
+       }
+       entry_clear(a);
+}
+
+static void update_extended_entry(struct tree_desc_x *t, struct name_entry *a)
+{
+       if (t->d.entry.path == a->path) {
+               update_tree_entry(&t->d);
+       } else {
+               /* we have returned this entry early */
+               struct tree_desc_skip *skip = xmalloc(sizeof(*skip));
+               skip->ptr = a->path;
+               skip->prev = t->skip;
+               t->skip = skip;
+       }
+}
+
+static void free_extended_entry(struct tree_desc_x *t)
+{
+       struct tree_desc_skip *p, *s;
+
+       for (s = t->skip; s; s = p) {
+               p = s->prev;
+               free(s);
+       }
+}
+
 int traverse_trees(int n, struct tree_desc *t, struct traverse_info *info)
 {
        int ret = 0;
        struct name_entry *entry = xmalloc(n*sizeof(*entry));
+       int i;
+       struct tree_desc_x *tx = xcalloc(n, sizeof(*tx));
+
+       for (i = 0; i < n; i++)
+               tx[i].d = t[i];
 
        for (;;) {
-               unsigned long mask = 0;
-               unsigned long dirmask = 0;
-               int i, last;
+               unsigned long mask, dirmask;
+               const char *first = NULL;
+               int first_len = 0;
+               struct name_entry *e;
+               int len;
 
-               last = -1;
                for (i = 0; i < n; i++) {
-                       if (!t[i].size)
+                       e = entry + i;
+                       extended_entry_extract(tx + i, e, NULL, 0);
+               }
+
+               /*
+                * A tree may have "t-2" at the current location even
+                * though it may have "t" that is a subtree behind it,
+                * and another tree may return "t".  We want to grab
+                * all "t" from all trees to match in such a case.
+                */
+               for (i = 0; i < n; i++) {
+                       e = entry + i;
+                       if (!e->path)
                                continue;
-                       entry_extract(t+i, entry+i);
-                       if (last >= 0) {
-                               int cmp = entry_compare(entry+i, entry+last);
-
-                               /*
-                                * Is the new name bigger than the old one?
-                                * Ignore it
-                                */
-                               if (cmp > 0)
+                       len = tree_entry_len(e->path, e->sha1);
+                       if (!first) {
+                               first = e->path;
+                               first_len = len;
+                               continue;
+                       }
+                       if (name_compare(e->path, len, first, first_len) < 0) {
+                               first = e->path;
+                               first_len = len;
+                       }
+               }
+
+               if (first) {
+                       for (i = 0; i < n; i++) {
+                               e = entry + i;
+                               extended_entry_extract(tx + i, e, first, first_len);
+                               /* Cull the ones that are not the earliest */
+                               if (!e->path)
                                        continue;
-                               /*
-                                * Is the new name smaller than the old one?
-                                * Ignore all old ones
-                                */
-                               if (cmp < 0)
-                                       mask = 0;
+                               len = tree_entry_len(e->path, e->sha1);
+                               if (name_compare(e->path, len, first, first_len))
+                                       entry_clear(e);
                        }
+               }
+
+               /* Now we have in entry[i] the earliest name from the trees */
+               mask = 0;
+               dirmask = 0;
+               for (i = 0; i < n; i++) {
+                       if (!entry[i].path)
+                               continue;
                        mask |= 1ul << i;
                        if (S_ISDIR(entry[i].mode))
                                dirmask |= 1ul << i;
-                       last = i;
                }
                if (!mask)
                        break;
-               dirmask &= mask;
-
-               /*
-                * Clear all the unused name-entries.
-                */
-               for (i = 0; i < n; i++) {
-                       if (mask & (1ul << i))
-                               continue;
-                       entry_clear(entry + i);
-               }
                ret = info->fn(n, mask, dirmask, entry, info);
                if (ret < 0)
                        break;
-               if (ret)
-                       mask &= ret;
+               mask &= ret;
                ret = 0;
-               for (i = 0; i < n; i++) {
+               for (i = 0; i < n; i++)
                        if (mask & (1ul << i))
-                               update_tree_entry(t + i);
-               }
+                               update_extended_entry(tx + i, entry + i);
        }
        free(entry);
+       for (i = 0; i < n; i++)
+               free_extended_entry(tx + i);
+       free(tx);
        return ret;
 }
 
index 0ddbef3e6355da40a1293a6a7c49b4bc5d75b842..75f54cac97f62ddaad736c2cd582cc6cdeaaebfa 100644 (file)
@@ -198,23 +198,142 @@ static inline int call_unpack_fn(struct cache_entry **src, struct unpack_trees_o
        return ret;
 }
 
-static int unpack_index_entry(struct cache_entry *ce, struct unpack_trees_options *o)
+static void mark_ce_used(struct cache_entry *ce, struct unpack_trees_options *o)
+{
+       ce->ce_flags |= CE_UNPACKED;
+
+       if (o->cache_bottom < o->src_index->cache_nr &&
+           o->src_index->cache[o->cache_bottom] == ce) {
+               int bottom = o->cache_bottom;
+               while (bottom < o->src_index->cache_nr &&
+                      o->src_index->cache[bottom]->ce_flags & CE_UNPACKED)
+                       bottom++;
+               o->cache_bottom = bottom;
+       }
+}
+
+static void mark_all_ce_unused(struct index_state *index)
+{
+       int i;
+       for (i = 0; i < index->cache_nr; i++)
+               index->cache[i]->ce_flags &= ~CE_UNPACKED;
+}
+
+static int locate_in_src_index(struct cache_entry *ce,
+                              struct unpack_trees_options *o)
+{
+       struct index_state *index = o->src_index;
+       int len = ce_namelen(ce);
+       int pos = index_name_pos(index, ce->name, len);
+       if (pos < 0)
+               pos = -1 - pos;
+       return pos;
+}
+
+/*
+ * We call unpack_index_entry() with an unmerged cache entry
+ * only in diff-index, and it wants a single callback.  Skip
+ * the other unmerged entry with the same name.
+ */
+static void mark_ce_used_same_name(struct cache_entry *ce,
+                                  struct unpack_trees_options *o)
+{
+       struct index_state *index = o->src_index;
+       int len = ce_namelen(ce);
+       int pos;
+
+       for (pos = locate_in_src_index(ce, o); pos < index->cache_nr; pos++) {
+               struct cache_entry *next = index->cache[pos];
+               if (len != ce_namelen(next) ||
+                   memcmp(ce->name, next->name, len))
+                       break;
+               mark_ce_used(next, o);
+       }
+}
+
+static struct cache_entry *next_cache_entry(struct unpack_trees_options *o)
+{
+       const struct index_state *index = o->src_index;
+       int pos = o->cache_bottom;
+
+       while (pos < index->cache_nr) {
+               struct cache_entry *ce = index->cache[pos];
+               if (!(ce->ce_flags & CE_UNPACKED))
+                       return ce;
+               pos++;
+       }
+       return NULL;
+}
+
+static void add_same_unmerged(struct cache_entry *ce,
+                             struct unpack_trees_options *o)
+{
+       struct index_state *index = o->src_index;
+       int len = ce_namelen(ce);
+       int pos = index_name_pos(index, ce->name, len);
+
+       if (0 <= pos)
+               die("programming error in a caller of mark_ce_used_same_name");
+       for (pos = -pos - 1; pos < index->cache_nr; pos++) {
+               struct cache_entry *next = index->cache[pos];
+               if (len != ce_namelen(next) ||
+                   memcmp(ce->name, next->name, len))
+                       break;
+               add_entry(o, next, 0, 0);
+               mark_ce_used(next, o);
+       }
+}
+
+static int unpack_index_entry(struct cache_entry *ce,
+                             struct unpack_trees_options *o)
 {
        struct cache_entry *src[5] = { ce, NULL, };
+       int ret;
 
-       o->pos++;
+       mark_ce_used(ce, o);
        if (ce_stage(ce)) {
                if (o->skip_unmerged) {
                        add_entry(o, ce, 0, 0);
                        return 0;
                }
        }
-       return call_unpack_fn(src, o);
+       ret = call_unpack_fn(src, o);
+       if (ce_stage(ce))
+               mark_ce_used_same_name(ce, o);
+       return ret;
+}
+
+static int find_cache_pos(struct traverse_info *, const struct name_entry *);
+
+static void restore_cache_bottom(struct traverse_info *info, int bottom)
+{
+       struct unpack_trees_options *o = info->data;
+
+       if (o->diff_index_cached)
+               return;
+       o->cache_bottom = bottom;
+}
+
+static int switch_cache_bottom(struct traverse_info *info)
+{
+       struct unpack_trees_options *o = info->data;
+       int ret, pos;
+
+       if (o->diff_index_cached)
+               return 0;
+       ret = o->cache_bottom;
+       pos = find_cache_pos(info->prev, &info->name);
+
+       if (pos < -1)
+               o->cache_bottom = -2 - pos;
+       else if (pos < 0)
+               o->cache_bottom = o->src_index->cache_nr;
+       return ret;
 }
 
 static int traverse_trees_recursive(int n, unsigned long dirmask, unsigned long df_conflicts, struct name_entry *names, struct traverse_info *info)
 {
-       int i;
+       int i, ret, bottom;
        struct tree_desc t[MAX_UNPACK_TREES];
        struct traverse_info newinfo;
        struct name_entry *p;
@@ -235,7 +354,11 @@ static int traverse_trees_recursive(int n, unsigned long dirmask, unsigned long
                        sha1 = names[i].sha1;
                fill_tree_descriptor(t+i, sha1);
        }
-       return traverse_trees(n, t, &newinfo);
+
+       bottom = switch_cache_bottom(&newinfo);
+       ret = traverse_trees(n, t, &newinfo);
+       restore_cache_bottom(&newinfo, bottom);
+       return ret;
 }
 
 /*
@@ -284,6 +407,20 @@ static int compare_entry(const struct cache_entry *ce, const struct traverse_inf
        return ce_namelen(ce) > traverse_path_len(info, n);
 }
 
+static int ce_in_traverse_path(const struct cache_entry *ce,
+                              const struct traverse_info *info)
+{
+       if (!info->prev)
+               return 1;
+       if (do_compare_entry(ce, info->prev, &info->name))
+               return 0;
+       /*
+        * If ce (blob) is the same name as the path (which is a tree
+        * we will be descending into), it won't be inside it.
+        */
+       return (info->pathlen < ce_namelen(ce));
+}
+
 static struct cache_entry *create_ce_entry(const struct traverse_info *info, const struct name_entry *n, int stage)
 {
        int len = traverse_path_len(info, n);
@@ -360,6 +497,114 @@ static int unpack_failed(struct unpack_trees_options *o, const char *message)
        return -1;
 }
 
+/* NEEDSWORK: give this a better name and share with tree-walk.c */
+static int name_compare(const char *a, int a_len,
+                       const char *b, int b_len)
+{
+       int len = (a_len < b_len) ? a_len : b_len;
+       int cmp = memcmp(a, b, len);
+       if (cmp)
+               return cmp;
+       return (a_len - b_len);
+}
+
+/*
+ * The tree traversal is looking at name p.  If we have a matching entry,
+ * return it.  If name p is a directory in the index, do not return
+ * anything, as we will want to match it when the traversal descends into
+ * the directory.
+ */
+static int find_cache_pos(struct traverse_info *info,
+                         const struct name_entry *p)
+{
+       int pos;
+       struct unpack_trees_options *o = info->data;
+       struct index_state *index = o->src_index;
+       int pfxlen = info->pathlen;
+       int p_len = tree_entry_len(p->path, p->sha1);
+
+       for (pos = o->cache_bottom; pos < index->cache_nr; pos++) {
+               struct cache_entry *ce = index->cache[pos];
+               const char *ce_name, *ce_slash;
+               int cmp, ce_len;
+
+               if (!ce_in_traverse_path(ce, info))
+                       continue;
+               if (ce->ce_flags & CE_UNPACKED)
+                       continue;
+               ce_name = ce->name + pfxlen;
+               ce_slash = strchr(ce_name, '/');
+               if (ce_slash)
+                       ce_len = ce_slash - ce_name;
+               else
+                       ce_len = ce_namelen(ce) - pfxlen;
+               cmp = name_compare(p->path, p_len, ce_name, ce_len);
+               /*
+                * Exact match; if we have a directory we need to
+                * delay returning it.
+                */
+               if (!cmp)
+                       return ce_slash ? -2 - pos : pos;
+               if (0 < cmp)
+                       continue; /* keep looking */
+               /*
+                * ce_name sorts after p->path; could it be that we
+                * have files under p->path directory in the index?
+                * E.g.  ce_name == "t-i", and p->path == "t"; we may
+                * have "t/a" in the index.
+                */
+               if (p_len < ce_len && !memcmp(ce_name, p->path, p_len) &&
+                   ce_name[p_len] < '/')
+                       continue; /* keep looking */
+               break;
+       }
+       return -1;
+}
+
+static struct cache_entry *find_cache_entry(struct traverse_info *info,
+                                           const struct name_entry *p)
+{
+       int pos = find_cache_pos(info, p);
+       struct unpack_trees_options *o = info->data;
+
+       if (0 <= pos)
+               return o->src_index->cache[pos];
+       else
+               return NULL;
+}
+
+static void debug_path(struct traverse_info *info)
+{
+       if (info->prev) {
+               debug_path(info->prev);
+               if (*info->prev->name.path)
+                       putchar('/');
+       }
+       printf("%s", info->name.path);
+}
+
+static void debug_name_entry(int i, struct name_entry *n)
+{
+       printf("ent#%d %06o %s\n", i,
+              n->path ? n->mode : 0,
+              n->path ? n->path : "(missing)");
+}
+
+static void debug_unpack_callback(int n,
+                                 unsigned long mask,
+                                 unsigned long dirmask,
+                                 struct name_entry *names,
+                                 struct traverse_info *info)
+{
+       int i;
+       printf("* unpack mask %lu, dirmask %lu, cnt %d ",
+              mask, dirmask, n);
+       debug_path(info);
+       putchar('\n');
+       for (i = 0; i < n; i++)
+               debug_name_entry(i, names + i);
+}
+
 static int unpack_callback(int n, unsigned long mask, unsigned long dirmask, struct name_entry *names, struct traverse_info *info)
 {
        struct cache_entry *src[MAX_UNPACK_TREES + 1] = { NULL, };
@@ -370,25 +615,38 @@ static int unpack_callback(int n, unsigned long mask, unsigned long dirmask, str
        while (!p->mode)
                p++;
 
+       if (o->debug_unpack)
+               debug_unpack_callback(n, mask, dirmask, names, info);
+
        /* Are we supposed to look at the index too? */
        if (o->merge) {
-               while (o->pos < o->src_index->cache_nr) {
-                       struct cache_entry *ce = o->src_index->cache[o->pos];
-                       int cmp = compare_entry(ce, info, p);
+               while (1) {
+                       int cmp;
+                       struct cache_entry *ce;
+
+                       if (o->diff_index_cached)
+                               ce = next_cache_entry(o);
+                       else
+                               ce = find_cache_entry(info, p);
+
+                       if (!ce)
+                               break;
+                       cmp = compare_entry(ce, info, p);
                        if (cmp < 0) {
                                if (unpack_index_entry(ce, o) < 0)
                                        return unpack_failed(o, NULL);
                                continue;
                        }
                        if (!cmp) {
-                               o->pos++;
                                if (ce_stage(ce)) {
                                        /*
-                                        * If we skip unmerged index entries, we'll skip this
-                                        * entry *and* the tree entries associated with it!
+                                        * If we skip unmerged index
+                                        * entries, we'll skip this
+                                        * entry *and* the tree
+                                        * entries associated with it!
                                         */
                                        if (o->skip_unmerged) {
-                                               add_entry(o, ce, 0, 0);
+                                               add_same_unmerged(ce, o);
                                                return mask;
                                        }
                                }
@@ -401,6 +659,13 @@ static int unpack_callback(int n, unsigned long mask, unsigned long dirmask, str
        if (unpack_nondirectories(n, mask, dirmask, src, names, info) < 0)
                return -1;
 
+       if (src[0]) {
+               if (ce_stage(src[0]))
+                       mark_ce_used_same_name(src[0], o);
+               else
+                       mark_ce_used(src[0], o);
+       }
+
        /* Now handle any directories.. */
        if (dirmask) {
                unsigned long conflicts = mask & ~dirmask;
@@ -417,11 +682,13 @@ static int unpack_callback(int n, unsigned long mask, unsigned long dirmask, str
                        matches = cache_tree_matches_traversal(o->src_index->cache_tree,
                                                               names, info);
                        /*
-                        * Everything under the name matches.  Adjust o->pos to
-                        * skip the entire hierarchy.
+                        * Everything under the name matches; skip the
+                        * entire hierarchy.  diff_index_cached codepath
+                        * special cases D/F conflicts in such a way that
+                        * it does not do any look-ahead, so this is safe.
                         */
                        if (matches) {
-                               o->pos += matches;
+                               o->cache_bottom += matches;
                                return mask;
                        }
                }
@@ -465,11 +732,10 @@ int unpack_trees(unsigned len, struct tree_desc *t, struct unpack_trees_options
 
        memset(&o->result, 0, sizeof(o->result));
        o->result.initialized = 1;
-       if (o->src_index) {
-               o->result.timestamp.sec = o->src_index->timestamp.sec;
-               o->result.timestamp.nsec = o->src_index->timestamp.nsec;
-       }
+       o->result.timestamp.sec = o->src_index->timestamp.sec;
+       o->result.timestamp.nsec = o->src_index->timestamp.nsec;
        o->merge_size = len;
+       mark_all_ce_unused(o->src_index);
 
        if (!dfc)
                dfc = xcalloc(1, cache_entry_size(0));
@@ -483,22 +749,38 @@ int unpack_trees(unsigned len, struct tree_desc *t, struct unpack_trees_options
                info.fn = unpack_callback;
                info.data = o;
 
-               if (traverse_trees(len, t, &info) < 0) {
-                       ret = unpack_failed(o, NULL);
-                       goto done;
+               if (o->prefix) {
+                       /*
+                        * Unpack existing index entries that sort before the
+                        * prefix the tree is spliced into.  Note that o->merge
+                        * is always true in this case.
+                        */
+                       while (1) {
+                               struct cache_entry *ce = next_cache_entry(o);
+                               if (!ce)
+                                       break;
+                               if (ce_in_traverse_path(ce, &info))
+                                       break;
+                               if (unpack_index_entry(ce, o) < 0)
+                                       goto return_failed;
+                       }
                }
+
+               if (traverse_trees(len, t, &info) < 0)
+                       goto return_failed;
        }
 
        /* Any left-over entries in the index? */
        if (o->merge) {
-               while (o->pos < o->src_index->cache_nr) {
-                       struct cache_entry *ce = o->src_index->cache[o->pos];
-                       if (unpack_index_entry(ce, o) < 0) {
-                               ret = unpack_failed(o, NULL);
-                               goto done;
-                       }
+               while (1) {
+                       struct cache_entry *ce = next_cache_entry(o);
+                       if (!ce)
+                               break;
+                       if (unpack_index_entry(ce, o) < 0)
+                               goto return_failed;
                }
        }
+       mark_all_ce_unused(o->src_index);
 
        if (o->trivial_merges_only && o->nontrivial_merge) {
                ret = unpack_failed(o, "Merge requires file-level merging");
@@ -543,6 +825,11 @@ int unpack_trees(unsigned len, struct tree_desc *t, struct unpack_trees_options
                free(el.excludes);
 
        return ret;
+
+return_failed:
+       mark_all_ce_unused(o->src_index);
+       ret = unpack_failed(o, NULL);
+       goto done;
 }
 
 /* Here come the merge functions */
@@ -661,7 +948,9 @@ static int verify_clean_subdirectory(struct cache_entry *ce, const char *action,
         * in that directory.
         */
        namelen = strlen(ce->name);
-       for (i = o->pos; i < o->src_index->cache_nr; i++) {
+       for (i = locate_in_src_index(ce, o);
+            i < o->src_index->cache_nr;
+            i++) {
                struct cache_entry *ce2 = o->src_index->cache[i];
                int len = ce_namelen(ce2);
                if (len < namelen ||
@@ -669,12 +958,14 @@ static int verify_clean_subdirectory(struct cache_entry *ce, const char *action,
                    ce2->name[namelen] != '/')
                        break;
                /*
-                * ce2->name is an entry in the subdirectory.
+                * ce2->name is an entry in the subdirectory to be
+                * removed.
                 */
                if (!ce_stage(ce2)) {
                        if (verify_uptodate(ce2, o))
                                return -1;
                        add_entry(o, ce2, CE_REMOVE, 0);
+                       mark_ce_used(ce2, o);
                }
                cnt++;
        }
@@ -731,7 +1022,6 @@ static int verify_absent_1(struct cache_entry *ce, const char *action,
                return 0;
 
        if (!lstat(ce->name, &st)) {
-               int ret;
                int dtype = ce_to_dtype(ce);
                struct cache_entry *result;
 
@@ -759,28 +1049,8 @@ static int verify_absent_1(struct cache_entry *ce, const char *action,
                         * files that are in "foo/" we would lose
                         * them.
                         */
-                       ret = verify_clean_subdirectory(ce, action, o);
-                       if (ret < 0)
-                               return ret;
-
-                       /*
-                        * If this removed entries from the index,
-                        * what that means is:
-                        *
-                        * (1) the caller unpack_callback() saw path/foo
-                        * in the index, and it has not removed it because
-                        * it thinks it is handling 'path' as blob with
-                        * D/F conflict;
-                        * (2) we will return "ok, we placed a merged entry
-                        * in the index" which would cause o->pos to be
-                        * incremented by one;
-                        * (3) however, original o->pos now has 'path/foo'
-                        * marked with "to be removed".
-                        *
-                        * We need to increment it by the number of
-                        * deleted entries here.
-                        */
-                       o->pos += ret;
+                       if (verify_clean_subdirectory(ce, action, o) < 0)
+                               return -1;
                        return 0;
                }
 
@@ -927,7 +1197,8 @@ int threeway_merge(struct cache_entry **stages, struct unpack_trees_options *o)
                remote = NULL;
        }
 
-       /* First, if there's a #16 situation, note that to prevent #13
+       /*
+        * First, if there's a #16 situation, note that to prevent #13
         * and #14.
         */
        if (!same(remote, head)) {
@@ -941,7 +1212,8 @@ int threeway_merge(struct cache_entry **stages, struct unpack_trees_options *o)
                }
        }
 
-       /* We start with cases where the index is allowed to match
+       /*
+        * We start with cases where the index is allowed to match
         * something other than the head: #14(ALT) and #2ALT, where it
         * is permitted to match the result instead.
         */
@@ -971,12 +1243,13 @@ int threeway_merge(struct cache_entry **stages, struct unpack_trees_options *o)
        if (!head && !remote && any_anc_missing)
                return 0;
 
-       /* Under the new "aggressive" rule, we resolve mostly trivial
+       /*
+        * Under the "aggressive" rule, we resolve mostly trivial
         * cases that we historically had git-merge-one-file resolve.
         */
        if (o->aggressive) {
-               int head_deleted = !head && !df_conflict_head;
-               int remote_deleted = !remote && !df_conflict_remote;
+               int head_deleted = !head;
+               int remote_deleted = !remote;
                struct cache_entry *ce = NULL;
 
                if (index)
index 95ff36c824a14a24234459ad552306fb1b5dabf1..ef70eab39025fcdaccda059692ae447a13fa0aeb 100644 (file)
@@ -31,10 +31,11 @@ struct unpack_trees_options {
                     skip_unmerged,
                     initial_checkout,
                     diff_index_cached,
+                    debug_unpack,
                     skip_sparse_checkout,
                     gently;
        const char *prefix;
-       int pos;
+       int cache_bottom;
        struct dir_struct *dir;
        merge_fn_t fn;
        struct unpack_trees_error_msgs msgs;