[patch 0/5] [PATCH CFT] more revlog experiments

[Chris Mason]

On Tuesday 21 March 2006 14:26, Matt Mackall wrote:
> On Tue, Mar 21, 2006 at 10:06:50AM -0800, Bryan O'Sullivan wrote:

> Yeah. I'm much more worried about the interesting new worst cases it
> introduces.
>
> Here's another approach to consider:
>
> Below a certain size (say 64k or 128k), interleave indices and data in
> a single file, still append-only. This reduces the average overhead
> for a single revision from a block and a half of slop and two
> directory entries to half a block and one entry. Our overhead will be
> basically identical to that in the working directory, which should be
> acceptable.

It's an interesting compromise.  I thought you were set on a fixed record 
length .i file next to a variable record .d file.  I'll try this out over the 
weekend.

-chris