CodingGuidelines: add Python coding guidelines
[gitweb.git] / tree-walk.c
index af871c50d1a4a2e4311228a82d6ac4ab52a660b5..6e30ef9d048c62c11a92aa5b0ee6df2d227776e6 100644 (file)
@@ -572,6 +572,54 @@ static int match_dir_prefix(const char *base,
        return 0;
 }
 
+/*
+ * Perform matching on the leading non-wildcard part of
+ * pathspec. item->nowildcard_len must be greater than zero. Return
+ * non-zero if base is matched.
+ */
+static int match_wildcard_base(const struct pathspec_item *item,
+                              const char *base, int baselen,
+                              int *matched)
+{
+       const char *match = item->match;
+       /* the wildcard part is not considered in this function */
+       int matchlen = item->nowildcard_len;
+
+       if (baselen) {
+               int dirlen;
+               /*
+                * Return early if base is longer than the
+                * non-wildcard part but it does not match.
+                */
+               if (baselen >= matchlen) {
+                       *matched = matchlen;
+                       return !strncmp(base, match, matchlen);
+               }
+
+               dirlen = matchlen;
+               while (dirlen && match[dirlen - 1] != '/')
+                       dirlen--;
+
+               /*
+                * Return early if base is shorter than the
+                * non-wildcard part but it does not match. Note that
+                * base ends with '/' so we are sure it really matches
+                * directory
+                */
+               if (strncmp(base, match, baselen))
+                       return 0;
+               *matched = baselen;
+       } else
+               *matched = 0;
+       /*
+        * we could have checked entry against the non-wildcard part
+        * that is not in base and does similar never_interesting
+        * optimization as in match_entry. For now just be happy with
+        * base comparison.
+        */
+       return entry_interesting;
+}
+
 /*
  * Is a tree entry interesting given the pathspec we have?
  *
@@ -602,7 +650,7 @@ enum interesting tree_entry_interesting(const struct name_entry *entry,
                const struct pathspec_item *item = ps->items+i;
                const char *match = item->match;
                const char *base_str = base->buf + base_offset;
-               int matchlen = item->len;
+               int matchlen = item->len, matched = 0;
 
                if (baselen >= matchlen) {
                        /* If it doesn't match, move along... */
@@ -627,7 +675,9 @@ enum interesting tree_entry_interesting(const struct name_entry *entry,
                                return entry_interesting;
 
                        if (item->nowildcard_len < item->len) {
-                               if (!fnmatch(match + baselen, entry->path, 0))
+                               if (!git_fnmatch(match + baselen, entry->path,
+                                                item->flags & PATHSPEC_ONESTAR ? GFNM_ONESTAR : 0,
+                                                item->nowildcard_len - baselen))
                                        return entry_interesting;
 
                                /*
@@ -645,14 +695,31 @@ enum interesting tree_entry_interesting(const struct name_entry *entry,
                if (item->nowildcard_len == item->len)
                        continue;
 
+               if (item->nowildcard_len &&
+                   !match_wildcard_base(item, base_str, baselen, &matched))
+                       return entry_not_interesting;
+
                /*
                 * Concatenate base and entry->path into one and do
                 * fnmatch() on it.
+                *
+                * While we could avoid concatenation in certain cases
+                * [1], which saves a memcpy and potentially a
+                * realloc, it turns out not worth it. Measurement on
+                * linux-2.6 does not show any clear improvements,
+                * partly because of the nowildcard_len optimization
+                * in git_fnmatch(). Avoid micro-optimizations here.
+                *
+                * [1] if match_wildcard_base() says the base
+                * directory is already matched, we only need to match
+                * the rest, which is shorter so _in theory_ faster.
                 */
 
                strbuf_add(base, entry->path, pathlen);
 
-               if (!fnmatch(match, base->buf + base_offset, 0)) {
+               if (!git_fnmatch(match, base->buf + base_offset,
+                                item->flags & PATHSPEC_ONESTAR ? GFNM_ONESTAR : 0,
+                                item->nowildcard_len)) {
                        strbuf_setlen(base, base_offset + baselen);
                        return entry_interesting;
                }