[RFC] extending pathspec support to submodules

From: Brandon Williams <bmwill@google.com>
Date: Sep 14, 2016, 23:57 UTC
Message-ID: <1473897473-154528-1-git-send-email-bmwill@google.com>

--- I've been trying to think through how we could potentially add pathspec support for --recurse-submodule options (for builtins like ls-files or grep down the line). This is something that could be useful if the user supply's a pathspec that could match to a file in a submodule. We could match the submodule to the pathspec and then fork the process to recursively run the command on the submodule which can be passed a modified pathspec.

For example with a pathspec 'sub/dir/a', where sub is a submodule in the root directory of the supermodule's repo, we could match 'sub' to that spec and then recursively call the git command with a pathspec of 'dir/a'. The child process would then have the responsibility of matching 'dir/a' to files in its repo.

Does this seem like a reasonable feature to add? And if so are how is my initial approach at solving the problem?

One idea I had was to add a submodule match flag in order to perform special matching just in the --recurse-submodules cases since we'll want somethings to match here that wouldn't normally match.

@@ -283,6 +284,29 @@ static int match_pathspec_item(const struct pathspec_item *item, int prefix,
 			 item->nowildcard_len - prefix))
 		return MATCHED_FNMATCH;
 
+	/*
+	 * Preform some checks to see if "name" is a super set of the pathspec
+	 */
+	if (flags & DO_MATCH_SUBMODULE) {
+		struct strbuf buf = STRBUF_INIT;
+		strbuf_addstr(&buf, name);
+		strbuf_addch(&buf, '/');
+		/*
+		 * Check if the name is a prefix of the pathspec
+		 */
+		if ((item->match[namelen] == '/') &&
+		    !ps_strncmp(item, match, name, namelen))
+			return MATCHED_RECURSIVELY;
+		/*
+		 * Check if the name wildmatches to the pathspec
+		 */
+		if (!wildmatch(item->match, buf.buf,
+			       WM_PREFIX |
+			       (item->magic & PATHSPEC_ICASE ? WM_CASEFOLD : 0),
+			       NULL));
+		    return MATCHED_FNMATCH;
+	}
+
 	return 0;
 }
 
One of the main difficulties I was having is figuring out how wildmatching
should be applied in this case.  What I believe we want is the ability for the
whole name of the submodule to match a prefix of the pathspec pattern.  To do
this I was thinking of adding a flag to do prefix matching to the wildmatch
function like so: 


diff --git a/wildmatch.c b/wildmatch.c
index 57c8765..f1e1725 100644
--- a/wildmatch.c
+++ b/wildmatch.c
@@ -60,8 +60,12 @@ static int dowild(const uchar *p, const uchar *text, unsigned int flags)
 	for ( ; (p_ch = *p) != '\0'; text++, p++) {
 		int matched, match_slash, negated;
 		uchar t_ch, prev_ch;
-		if ((t_ch = *text) == '\0' && p_ch != '*')
-			return WM_ABORT_ALL;
+		if ((t_ch = *text) == '\0' && p_ch != '*') {
+			if ((flags & WM_PREFIX) && (*(p-1) == '/'))
+				return WM_MATCH;
+			else
+				return WM_ABORT_ALL;
+		}
 		if ((flags & WM_CASEFOLD) && ISUPPER(t_ch))
 			t_ch = tolower(t_ch);
 		if ((flags & WM_CASEFOLD) && ISUPPER(p_ch))
diff --git a/wildmatch.h b/wildmatch.h
index 4090c8f..490db51 100644
--- a/wildmatch.h
+++ b/wildmatch.h
@@ -3,6 +3,7 @@
 
 #define WM_CASEFOLD 1
 #define WM_PATHNAME 2
+#define WM_PREFIX 4
 
 #define WM_ABORT_MALFORMED 2
 #define WM_NOMATCH 1

-- 
Any comments or thoughts on this would be appreciated.

Thanks,
Brandon