[PATCH 4/4] Test in git-init-db if the filemode can be trusted

[PATCH 4/4] Test in git-init-db if the filemode can be trusted

[Johannes Schindelin]

... and if not, write an appropriate .git/config. Of course, that happens 
only if no config file was yet created (by a template or a hook).

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>

---

	This might be a bit controversial, as we probably need a way to 
	write to .git/config anyway. Until there are such functions, I'll
	live with this patch.

 init-db.c |   37 +++++++++++++++++++++++++++++++++++++
 1 files changed, 37 insertions(+), 0 deletions(-)

applies-to: 047019cf0b103c8350c6b4da7836b15795c69498
64ca59712d766e8264f9208d70b8f5c811dfd3ea
diff --git a/init-db.c b/init-db.c
index 2c27e18..b242fbf 100644
--- a/init-db.c
+++ b/init-db.c
@@ -196,6 +196,43 @@ static void create_default_files(const c
 	}
 	path[len] = 0;
 	copy_templates(path, len, template_path);
+
+	/*
+	 * Find out if we can trust the executable bit.
+	 */
+	safe_create_dir(path);
+	strcpy(path + len, "config");
+	if (access(path, R_OK) < 0) {
+		static const char contents[] =
+			"#\n"
+			"# This is the config file\n"
+			"#\n"
+			"\n"
+			"; core variables\n"
+			"[core]\n"
+			"	; Don't trust file modes\n"
+			"	filemode = false\n"
+			"\n";
+		FILE *config = fopen(path, "w");
+		struct stat st;
+
+		if (!config)
+			die("Can not write to %s?", path);
+
+		fwrite(contents, sizeof(contents)-1, 1, config);
+
+		fclose(config);
+
+		if (!lstat(path, &st)) {
+			struct stat st2;
+			if (!chmod(path, st.st_mode ^ S_IXUSR) &&
+					!lstat(path, &st2) &&
+					st.st_mode != st2.st_mode)
+				unlink(path);
+			else
+				fprintf(stderr, "Ignoring file modes\n");
+		}
+	}
 }
 
 static const char init_db_usage[] =
---
0.99.8.GIT

Re: [PATCH 4/4] Test in git-init-db if the filemode can be trusted

[Junio C Hamano]

Johannes Schindelin <Johannes.Schindelin@gmx.de> writes:

> ... and if not, write an appropriate .git/config. Of course, that happens 
> only if no config file was yet created (by a template or a hook).
>
> Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
>
> ---
>
> 	This might be a bit controversial, as we probably need a way to 
> 	write to .git/config anyway. Until there are such functions, I'll
> 	live with this patch.

Hmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmm.  Yes, we discussed something
like this earlier.  It probably is a good usability patch, but
at the same time it does not feel quite right, but I cannot
explain why.  Probably because I have not suffered from FAT for
some time, lucky me, I guess.