Re: [PATCH 1/3] Add read_cache_from() and discard_cache()

From: Junio C Hamano <junkio@cox.net>
Date: Jun 30, 2006, 16:44 UTC
Message-ID: <7v3bdmk2zj.fsf@assigned-by-dhcp.cox.net>
In-Reply-To: <Pine.LNX.4.63.0606301643150.29667@wbgn013.biozentrum.uni-wuerzburg.de>

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

Show 14 quoted lines

> +int discard_cache()
> +{
> +	int ret;
> +	
> +	if (cache_mmap == NULL)
> +		return 0;
> +	ret = munmap(cache_mmap, cache_mmap_size);
> +	cache_mmap = NULL;
> +	cache_mmap_size = 0;
> +	active_nr = active_cache_changed = 0;
> +	/* no need to throw away allocated active_cache */
> +	return ret;
> +}
> +

I haven't been following the details of the patches in this thread while they are being cooked actively, but two things to look out for are:

 - I am guessing you run discard_cache() because you want to
   read in a new cache (or start from a clean slate).  I am not
   sure what you are doing with the old cache tree data
   structure.  If you are starting from a clean slate
   (i.e. there is no read_cache_from() after you call
   discard_cache), you would probably need to discard the old
   cache tree; otherwise your next write-tree may produce an
   incorrect index file.  If you keep the old one and later
   swap it in, the problem might be even more severe.

 - index_timestamp is left as the old value in this patch when
   you switch cache using read_cache_from() directly.  I have a
   suspicion you may be bitten by "Racy Git" problem, especially
   because the operations are supposed to happen quickly thanks
   to the effort of you two ;-) increasing the risks that the
   file timestamp of the working tree file and the cached entry
   match.