Re: [PATCH v5 11/35] delete_ref_loose(): don't muck around in the lock_file's filename

From: Jonathan Nieder <jrnieder@gmail.com>
Date: Sep 16, 2014, 21:11 UTC
Message-ID: <20140916211140.GL29050@google.com>
In-Reply-To: <1410896036-12750-12-git-send-email-mhagger@alum.mit.edu>

Michael Haggerty wrote:

Show 6 quoted lines

> It's bad manners. Especially since there could be a signal during the
> call to unlink_or_warn(), in which case the signal handler will see
> the wrong filename and delete the reference file, leaving the lockfile
> behind.
>
> So make our own copy to work with.

Nice. Could this be a helper, to help others avoid a PATH_MAX sized buffer too?

[...]

> --- a/refs.c
> +++ b/refs.c
> @@ -2551,12 +2551,15 @@ int repack_without_refs(const char **refnames, int n, struct strbuf *err)
>  static int delete_ref_loose(struct ref_lock *lock, int flag)

[...]

Show 7 quoted lines

> +		/*
> +		 * loose.  The loose file name is the same as the
> +		 * lockfile name, minus ".lock":
> +		 */
> +		char *loose_filename = xmemdupz(
> +				lock->lk->filename,
> +				strlen(lock->lk->filename) - LOCK_SUFFIX_LEN);

Ronnie mentioned that this would misbehave in the !lock->lk->filename[0] case, which shouldn't come up because lock_ref_sha1_basic dies on error but seems worth futureproofing against. Maybe something like

	assert(lock->lk->filename[0]);

would be enough to make the assumption obvious for people reading.

Otherwise looks good.

Thanks, Jonathan