Re: [RFH] bug in unpack_trees

From: Linus Torvalds <torvalds@linux-foundation.org>
Date: Mar 5, 2008, 15:56 UTC
Message-ID: <alpine.LFD.1.00.0803050750400.12253@woody.linux-foundation.org>
In-Reply-To: <alpine.LNX.1.00.0803050130190.19665@iabervon.org>

On Wed, 5 Mar 2008, Daniel Barkalow wrote:

> 
> The thing that's hopeless isn't including the index; it's including the 
> index that's simultaneously being regenerated.

Yeah. I was thinking about just putting the result in a new index. It's *usually* what the user wants anyway. The whole complexity with updating the old index is really nasty.

There are other complexities there, but the index one is the worst.

When doing a stupid try at using "traverse_trees()" (which in itself was not that easy - traverse_trees() is a fundamentally simpler walker and _different_ enough to not match well), one of the bigger issues is that traverse_trees() wants to do the directories in a separate phase from the files (becasue they sort differently), and that coupled with the fact that we do a kind of "read-modify-write" on the index makes it all really ugly.

I'm still working on it, but it's nastier than I was hoping for. Maybe you can come up with a better solution.

		Linus