[PATCH 3 of 3] revlog: read index data using mmap

[Jun Wu]

Excerpts from Augie Fackler's message of 2016-10-04 12:58:42 -0400:
> On Tue, Oct 04, 2016 at 06:28:54PM +0200, Pierre-Yves David wrote:
> > Can we get some number for small and large repository here ?
> >
> > I remember Matt making arguments against the use of mmap in multiple
> > occurrence in the past. I can find the details again, but I think this was
> > about the amount of kernel roundtrip involved compared to a plain read.
> 
> We could probably do the benchmark effectively with just some sort of
> debugrevlog construct, on some large and small indicies, rather than
> on some entire repositories.

I think the case we worry about is the case that we need to read the entire
file, and the pattern of reading matters - what happens if we are reading
the file sequentially, or randomly?

This depends on the kernel implementation. I will get some numbers on Linux
and OS X.