Re: [Gnu-arch-users] [PATCH] arch speedups on big trees

[Chris Mason]

On Fri, 2003-12-19 at 18:08, Tom Lord wrote:

[ issues with the speedup hack ]
> 
> Meanwhile, as your own profiling suggests, the bulk of the speed-up
> from mason's optimization can be achieved by other, more obviously
> correct means.

Possibly.  Using tla replay as a benchmark doesn't do any inode sig
stuff, which is where many of the fixes you suggested for Andrew were
aimed (or did I misread?).  

For arch to really scale, we need to limit the operations that start
with 'read entire source tree'.  I had a printf at the start and end of
each inventory scan, it's surprising how often it gets done.

There are places where it is entirely justified, but many others where
we only need to work on a small subset of the tree.

Would it make sense to add a reverse mapping from id to file name?  It
should help solve these corner cases, and we should be able to insert
the code for it into the same places that maintain the id->filename
mapping now.

-chris