[PATCH] Fix type-change handling when assigning the status code to filepairs.
[git.git] / diff.c
diff --git a/diff.c b/diff.c
index 5a2aee7..8e3f495 100644 (file)
--- a/diff.c
+++ b/diff.c
@@ -12,9 +12,6 @@ static const char *diff_opts = "-pu";
 static unsigned char null_sha1[20] = { 0, };
 
 static int reverse_diff;
-static int generate_patch;
-static int line_termination = '\n';
-static int inter_name_termination = '\t';
 
 static const char *external_diff(void)
 {
@@ -496,16 +493,27 @@ struct diff_filepair *diff_queue(struct diff_queue_struct *queue,
        dp->one = one;
        dp->two = two;
        dp->score = 0;
-       dp->orig_order = queue->nr;
-       dp->rename_rank = 0;
        diff_q(queue, dp);
        return dp;
 }
 
-static void diff_flush_raw(struct diff_filepair *p)
+static void diff_flush_raw(struct diff_filepair *p,
+                          int line_termination,
+                          int inter_name_termination)
 {
        int two_paths;
        char status[10];
+
+       if (line_termination) {
+               const char *err = "path %s cannot be expressed without -z";
+               if (strchr(p->one->path, line_termination) ||
+                   strchr(p->one->path, inter_name_termination))
+                       die(err, p->one->path);
+               if (strchr(p->two->path, line_termination) ||
+                   strchr(p->two->path, inter_name_termination))
+                       die(err, p->two->path);
+       }
+
        switch (p->status) {
        case 'C': case 'R':
                two_paths = 1;
@@ -566,10 +574,6 @@ static void diff_flush_patch(struct diff_filepair *p)
        const char *name, *other;
        char msg_[PATH_MAX*2+200], *msg;
 
-       /* diffcore_prune() keeps "stay" entries for diff-raw
-        * copy/rename detection, but when we are generating
-        * patches we do not need them.
-        */
        if (diff_unmodified_pair(p))
                return;
 
@@ -585,7 +589,7 @@ static void diff_flush_patch(struct diff_filepair *p)
                        "similarity index %d%%\n"
                        "copy from %s\n"
                        "copy to %s\n",
-                       (int)(0.5 + p->score * 100/MAX_SCORE),
+                       (int)(0.5 + p->score * 100.0/MAX_SCORE),
                        p->one->path, p->two->path);
                msg = msg_;
                break;
@@ -594,7 +598,7 @@ static void diff_flush_patch(struct diff_filepair *p)
                        "similarity index %d%%\n"
                        "rename old %s\n"
                        "rename new %s\n",
-                       (int)(0.5 + p->score * 100/MAX_SCORE),
+                       (int)(0.5 + p->score * 100.0/MAX_SCORE),
                        p->one->path, p->two->path);
                msg = msg_;
                break;
@@ -630,105 +634,160 @@ int diff_needs_to_stay(struct diff_queue_struct *q, int i,
        return 0;
 }
 
-static int diff_used_as_source(struct diff_queue_struct *q, int lim,
-                              struct diff_filespec *it)
+int diff_queue_is_empty(void)
 {
+       struct diff_queue_struct *q = &diff_queued_diff;
        int i;
-       for (i = 0; i < lim; i++) {
-               struct diff_filepair *p = q->queue[i++];
-               if (!strcmp(p->one->path, it->path))
-                       return 1;
-       }
-       return 0;
+       for (i = 0; i < q->nr; i++)
+               if (!diff_unmodified_pair(q->queue[i]))
+                       return 0;
+       return 1;
 }
 
-void diffcore_prune(void)
+#if DIFF_DEBUG
+void diff_debug_filespec(struct diff_filespec *s, int x, const char *one)
 {
-       /*
-        * Although rename/copy detection wants to have "no-change"
-        * entries fed into them, the downstream do not need to see
-        * them, unless we had rename/copy for the same path earlier.
-        * This function removes such entries.
-        *
-        * The applications that use rename/copy should:
-        *
-        * (1) feed change and "no-change" entries via diff_queue().
-        * (2) call diffcore_rename, and any other future diffcore_xxx
-        *     that would benefit by still having "no-change" entries.
-        * (3) call diffcore_prune
-        * (4) call other diffcore_xxx that do not need to see
-        *     "no-change" entries.
-        * (5) call diff_flush().
-        */
-       struct diff_queue_struct *q = &diff_queued_diff;
-       struct diff_queue_struct outq;
-       int i;
+       fprintf(stderr, "queue[%d] %s (%s) %s %06o %s\n",
+               x, one ? : "",
+               s->path,
+               DIFF_FILE_VALID(s) ? "valid" : "invalid",
+               s->mode,
+               s->sha1_valid ? sha1_to_hex(s->sha1) : "");
+       fprintf(stderr, "queue[%d] %s size %lu flags %d\n",
+               x, one ? : "",
+               s->size, s->xfrm_flags);
+}
 
-       outq.queue = NULL;
-       outq.nr = outq.alloc = 0;
+void diff_debug_filepair(const struct diff_filepair *p, int i)
+{
+       diff_debug_filespec(p->one, i, "one");
+       diff_debug_filespec(p->two, i, "two");
+       fprintf(stderr, "score %d, status %c\n",
+               p->score, p->status ? : '?');
+}
 
+void diff_debug_queue(const char *msg, struct diff_queue_struct *q)
+{
+       int i;
+       if (msg)
+               fprintf(stderr, "%s\n", msg);
+       fprintf(stderr, "q->nr = %d\n", q->nr);
        for (i = 0; i < q->nr; i++) {
                struct diff_filepair *p = q->queue[i];
-               if (!diff_unmodified_pair(p) ||
-                   diff_used_as_source(q, i, p->one))
-                       diff_q(&outq, p);
-               else
-                       free(p);
+               diff_debug_filepair(p, i);
        }
-       free(q->queue);
-       *q = outq;
-       return;
 }
+#endif
 
-int diff_queue_is_empty(void)
+static void diff_resolve_rename_copy(void)
 {
+       int i, j;
+       struct diff_filepair *p, *pp;
        struct diff_queue_struct *q = &diff_queued_diff;
-       return q->nr == 0;
+
+       /* This should not depend on the ordering of things. */
+
+       diff_debug_queue("resolve-rename-copy", q);
+
+       for (i = 0; i < q->nr; i++) {
+               p = q->queue[i];
+               p->status = 0; /* undecided */
+               if (DIFF_PAIR_UNMERGED(p))
+                       p->status = 'U';
+               else if (!DIFF_FILE_VALID((p)->one))
+                       p->status = 'N';
+               else if (!DIFF_FILE_VALID((p)->two)) {
+                       /* Deletion record should be omitted if there
+                        * are rename/copy entries using this one as
+                        * the source.  Then we can say one of them
+                        * is a rename and the rest are copies.
+                        */
+                       p->status = 'D';
+                       for (j = 0; j < q->nr; j++) {
+                               pp = q->queue[j];
+                               if (!strcmp(pp->one->path, p->one->path) &&
+                                   strcmp(pp->one->path, pp->two->path)) {
+                                       p->status = 'X';
+                                       break;
+                               }
+                       }
+               }
+               else if (DIFF_PAIR_TYPE_CHANGED(p))
+                       p->status = 'T';
+
+               /* from this point on, we are dealing with a pair
+                * whose both sides are valid and of the same type, i.e.
+                * either in-place edit or rename/copy edit.
+                */
+               else if (strcmp(p->one->path, p->two->path)) {
+                       /* See if there is somebody else anywhere that
+                        * will keep the path (either modified or
+                        * unmodified).  If so, we have to be a copy,
+                        * not a rename.  In addition, if there is
+                        * some other rename or copy that comes later
+                        * than us that uses the same source, we
+                        * have to be a copy, not a rename.
+                        */
+                       for (j = 0; j < q->nr; j++) {
+                               pp = q->queue[j];
+                               if (strcmp(pp->one->path, p->one->path))
+                                       continue;
+                               if (!strcmp(pp->one->path, pp->two->path)) {
+                                       if (DIFF_FILE_VALID(pp->two)) {
+                                               /* non-delete */
+                                               p->status = 'C';
+                                               break;
+                                       }
+                                       continue;
+                               }
+                               /* pp is a rename/copy ... */
+                               if (i < j) {
+                                       /* ... and comes later than us */
+                                       p->status = 'C';
+                                       break;
+                               }
+                       }
+                       if (!p->status)
+                               p->status = 'R';
+               }
+               else if (memcmp(p->one->sha1, p->two->sha1, 20))
+                       p->status = 'M';
+               else
+                       /* this is a "no-change" entry */
+                       p->status = 'X';
+       }
+       diff_debug_queue("resolve-rename-copy done", q);
 }
 
 void diff_flush(int diff_output_style, int resolve_rename_copy)
 {
        struct diff_queue_struct *q = &diff_queued_diff;
        int i;
+       int line_termination = '\n';
+       int inter_name_termination = '\t';
 
-       generate_patch = 0;
-       switch (diff_output_style) {
-       case DIFF_FORMAT_HUMAN:
-               line_termination = '\n';
-               inter_name_termination = '\t';
-               break;
-       case DIFF_FORMAT_MACHINE:
+       if (diff_output_style == DIFF_FORMAT_MACHINE)
                line_termination = inter_name_termination = 0;
-               break;
-       case DIFF_FORMAT_PATCH:
-               generate_patch = 1;
-               break;
-       }
+       if (resolve_rename_copy)
+               diff_resolve_rename_copy();
+
        for (i = 0; i < q->nr; i++) {
                struct diff_filepair *p = q->queue[i];
-               if (resolve_rename_copy) {
-                       if (DIFF_PAIR_UNMERGED(p))
-                               p->status = 'U';
-                       else if (!DIFF_FILE_VALID((p)->one))
-                               p->status = 'N';
-                       else if (!DIFF_FILE_VALID((p)->two))
-                               p->status = 'D';
-                       else if (strcmp(p->one->path, p->two->path)) {
-                               /* This is rename or copy.  Which one is it? */
-                               if (diff_needs_to_stay(q, i+1, p->one))
-                                       p->status = 'C';
-                               else
-                                       p->status = 'R';
-                       }
-                       else
-                               p->status = 'M';
-               }
-               if (generate_patch)
+               if (p->status == 'X')
+                       continue;
+               if (p->status == 0)
+                       die("internal error in diff-resolve-rename-copy");
+               switch (diff_output_style) {
+               case DIFF_FORMAT_PATCH:
                        diff_flush_patch(p);
-               else
-                       diff_flush_raw(p);
+                       break;
+               case DIFF_FORMAT_HUMAN:
+               case DIFF_FORMAT_MACHINE:
+                       diff_flush_raw(p, line_termination,
+                                      inter_name_termination);
+                       break;
+               }
        }
-
        for (i = 0; i < q->nr; i++) {
                struct diff_filepair *p = q->queue[i];
                diff_free_filespec_data(p->one);
@@ -755,9 +814,9 @@ void diff_addremove(int addremove, unsigned mode,
         * with something like '=' or '*' (I haven't decided
         * which but should not make any difference).
         * Feeding the same new and old to diff_change() 
-        * also has the same effect.  diffcore_prune() should
-        * be used to filter uninteresting ones out before the
-        * final output happens.
+        * also has the same effect.
+        * Before the final output happens, they are pruned after
+        * merged into rename/copy pairs as appropriate.
         */
        if (reverse_diff)
                addremove = (addremove == '+' ? '-' :