Re: [PATCH] Per-path attribute based hunk header selection.

Johannes Schindelin [off-list ref] writes:

quoted

quoted

+	if (!strcmp(ident, "tex"))
+		return "^\\\\\\(sub\\)*section{";

It is always easier, and will never require C skills, to put this into the 
config. With Junio's current version:

	echo '*.tex funcname=tex' >> .gitattributes
	echo '[funcname] tex = ^\(\\\(sub\)*section{.*\)' >> .git/config

The problem is of course that the backslashes have to be escaped _both_ in 
C and in regexps.

I think giving a reasonable set of basic language support as
built-in patterns is important for usability.  Otherwise the
users end up needing to have them in their $HOME/.gitconfig.

I am not sure if Java and LaTeX qualify as the first two most
important cases, but they are what we already have
demonstrated.  How about doing something like this?


-- >8 --
diff.c: make built-in hunk header pattern a separate table

This would hopefully make it easier to maintain.  Initially we
would have "java" and "tex" defined, as they are the only ones
we already have.

Signed-off-by: Junio C Hamano <redacted>
---

diff --git a/diff.c b/diff.c
index b8473f5..cd6b0c4 100644
--- a/diff.c
+++ b/diff.c

@@ -1216,9 +1216,22 @@ static const char *funcname_pattern(const char *ident)
 	return NULL;
 }
 
+static struct builtin_funcname_pattern {
+	const char *name;
+	const char *pattern;
+} builtin_funcname_pattern[] = {
+	{ "java", "!^[ 	]*\\(catch\\|do\\|for\\|if\\|instanceof\\|"
+			"new\\|return\\|switch\\|throw\\|while\\)\n"
+			"^[ 	]*\\(\\([ 	]*"
+			"[A-Za-z_][A-Za-z_0-9]*\\)\\{2,\\}"
+			"[ 	]*([^;]*$\\)" },
+	{ "tex", "^\\(\\\\\\(sub\\)*section{.*\\)$" },
+};
+
 static const char *diff_funcname_pattern(struct diff_filespec *one)
 {
 	const char *ident, *pattern;
+	int i;
 
 	diff_filespec_check_attr(one);
 	ident = one->funcname_pattern_ident;

@@ -1240,12 +1253,9 @@ static const char *diff_funcname_pattern(struct diff_filespec *one)
 	 * And define built-in fallback patterns here.  Note that
 	 * these can be overriden by the user's config settings.
 	 */
-	if (!strcmp(ident, "java"))
-		return "!^[ 	]*\\(catch\\|do\\|for\\|if\\|instanceof\\|"
-			"new\\|return\\|switch\\|throw\\|while\\)\n"
-			"^[ 	]*\\(\\([ 	]*"
-			"[A-Za-z_][A-Za-z_0-9]*\\)\\{2,\\}"
-			"[ 	]*([^;]*$\\)";
+	for (i = 0; i < ARRAY_SIZE(builtin_funcname_pattern); i++)
+		if (!strcmp(ident, builtin_funcname_pattern[i].name))
+			return builtin_funcname_pattern[i].pattern;
 
 	return NULL;
 }

Hi,

On Sun, 8 Jul 2007, Junio C Hamano wrote:

Johannes Schindelin [off-list ref] writes:

quoted

quoted

quoted

+	if (!strcmp(ident, "tex"))
+		return "^\\\\\\(sub\\)*section{";

It is always easier, and will never require C skills, to put this into the 
config. With Junio's current version:

	echo '*.tex funcname=tex' >> .gitattributes
	echo '[funcname] tex = ^\(\\\(sub\)*section{.*\)' >> .git/config

The problem is of course that the backslashes have to be escaped _both_ in 
C and in regexps.

I think giving a reasonable set of basic language support as
built-in patterns is important for usability.  Otherwise the
users end up needing to have them in their $HOME/.gitconfig.

I agree.

I am not sure if Java and LaTeX qualify as the first two most important 
cases, but they are what we already have demonstrated.

Java and LaTeX are the two languages I use, for which the inbuilt funcname 
default is absolutely unusable.  Given that Java is the only language with 
a notable git-related code base, for which the default fails, Java is the 
prime (if not the only) reason I wanted this whole patch series in the 
first place.

How about doing something like this?

I do not see how a table in diff.c could be easier to maintain than a 
small part in a small function in diff.c.  It would make sense, though, to 
move the table to xdiff-interface.[ch].

Ciao,
Dscho