[PATCH v2] Support "core.excludesfile = ~/.gitignore"

The config variable core.excludesfile is parsed to substitute ~ and ~user with
getpw entries.

Signed-off-by: Karl Chen <redacted>
---
 config.c |   41 +++++++++++++++++++++++++++++++++++++++--
 1 files changed, 39 insertions(+), 2 deletions(-)


Based on the discussion it sounds like there are complications to
supporting relative paths (due to worktree config), and "$HOME"
(when generalized, due to bootstrapping issues with $GIT_*).

Since ~ and ~user are orthogonal to these, can I suggest going
forward with this, without blocking on those two?

I have reworked the patch to use getpw to support ~user.  $HOME
can eventually be supported via $ENVVARs.

diff --git a/config.c b/config.c
index 53f04a0..6a83c64 100644
--- a/config.c
+++ b/config.c

@@ -334,6 +334,42 @@ int git_config_string(const char **dest, const char *var, const char *value)
 	return 0;
 }
 
+/*
+ * Expand ~ and ~user.  Returns a newly malloced string.  (If input does not
+ * start with "~", equivalent to xstrdup.)
+ */
+static char *expand_userdir(const char *value) {
+	if (value[0] == '~') {
+		struct passwd *pw;
+		char *expanded_dir;
+		const char *slash = strchr(value+1, '/');
+		const char *after_username = slash ? slash : value+strlen(value);
+		if (after_username == value+1) {
+			pw = getpwuid(getuid());
+			if (!pw) die("You don't exist!");
+		} else {
+			char save = *after_username;
+			*(char*)after_username = '\0';
+			pw = getpwnam(value+1);
+			if (!pw) die("No such user: '%s'", value+1);
+			*(char*)after_username = save;
+		}
+		expanded_dir = xmalloc(strlen(pw->pw_dir) + strlen(after_username) + 1);
+		strcpy(expanded_dir, pw->pw_dir);
+		strcat(expanded_dir, after_username);
+		return expanded_dir;
+	} else {
+		return xstrdup(value);
+	}
+}
+
+int git_config_userdir(const char **dest, const char *var, const char *value) {
+	if (!value)
+		return config_error_nonbool(var);
+	*dest = expand_userdir(value);
+	return 0;
+}
+
 static int git_default_core_config(const char *var, const char *value)
 {
 	/* This needs a better name */

@@ -456,8 +492,9 @@ static int git_default_core_config(const char *var, const char *value)
 	if (!strcmp(var, "core.editor"))
 		return git_config_string(&editor_program, var, value);
 
-	if (!strcmp(var, "core.excludesfile"))
-		return git_config_string(&excludes_file, var, value);
+	if (!strcmp(var, "core.excludesfile")) {
+		return git_config_userdir(&excludes_file, var, value);
+	}
 
 	if (!strcmp(var, "core.whitespace")) {
 		if (!value)

-- 
1.5.6.2

Karl Chen schrieb:

+/*
+ * Expand ~ and ~user.  Returns a newly malloced string.  (If input does not
+ * start with "~", equivalent to xstrdup.)
+ */
+static char *expand_userdir(const char *value) {

There is user_path() in path.c that does the same thing.

Watch your style: The opening brace of functions is on the next line.

-- Hannes

On Mon, Aug 25, 2008 at 12:07:15PM -0700, Karl Chen wrote:

Based on the discussion it sounds like there are complications to
supporting relative paths (due to worktree config), and "$HOME"
(when generalized, due to bootstrapping issues with $GIT_*).

I think that is fine for now. One other simple possibility would be to
expand _just_ $HOME, and then if we later decided to do all environment
variables it would naturally encompass that. However, we might want to
support "~" then anyway, so I think doing "~" first is fine.

However, there are two problems with the patch:

  1. It should probably re-use path.c:user_path, as Johannes mentioned.

  2. There is no documentation update.

Also, are there any other config variables which would benefit from this
substitution (I can't think of any off-hand, but there are quite a few I
don't use).

-Peff