Re: RFC: Flat directory for notes, or fan-out? Both!
- From
Johannes Schindelin <johannes.schindelin@gmx.de>
- Date
- Feb 11, 2009, 23:05 UTC
- Message-ID
- <alpine.DEB.1.00.0902120002241.10279@pacific.mpi-cbg.de>
- In-Reply-To
- <7vljscd695.fsf@gitster.siamese.dyndns.org>
Hi,
On Wed, 11 Feb 2009, Junio C Hamano wrote:
Show 19 quoted lines
> Johannes Schindelin <Johannes.Schindelin@gmx.de> writes: > > > Even the merging should not pose any problem at all; we need a custom > > merge driver anyway, and there is no reason whatsoever why we should > > not just teach the merge driver to remove the slashes before comparing > > the filie names. > > Once you start talking about "remove the slashes", you are assuming that > the custom merge algorithm must look at *all the paths* in the two trees > being merged, and it is a sign that your thinking is so trapped in the > inefficient way the current merge-recursive and unpack-trees based merge > works, and cannot think about the possibility that there could be more > efficient way to do merges. Not very good. > > If you have a fixed boundary and if most subtrees are the same between > two notes during a merge, we can do the same optimization as we do for > two input "diff-tree" codepath. If the top of a subtree matches, we do > not even have to look at their subtree. But that is true only if you do > not remove the slashes and allow a random hierarchy.
Well, I think in the case of notes, we have to optimize for random access of dozens of blobs rather than for merging. I was well aware that the merging is more expensive when the hierarchy is not defined a priori.
Ciao, Dscho