[darcs-users] antimemoization--call to large repo users!

Ivan Stankovic pokemon at fly.srk.fer.hr
Sat Sep 11 20:54:28 UTC 2004


On Sat, Sep 11, 2004 at 07:50:39AM -0400, David Roundy wrote:
> Thanks! I had one other tester (offlist) who saw dramatic improvements.  It
> looks like I shouldn't enable antimemoization by default.
> 
> I can see why the cpu time goes up by a factor of two, since every file is
> parsed for newlines twice, but I'm surprised that the peak memory usage
> didn't go down.  I guess I need to understand when the peak is happening to
> see if it can be alleviated.  I'm still not terribly fluent in the
> antimemoization framework.

I think darcs without antimemoization works quite well, in the usual scenarios.
For large repositories there's really only one solution: avoid parsing in the
first place.  I'm not talking only about memory issues here; 16 minutes for an
initial record of a 217 MB tree may not look too much, but compared to the time
it takes 'cp -R' to complete (~40 seconds) sure suggests that darcs does much 
more than is really required.

Someone posted a hack to quickly import a tree into darcs earlier on this
list.  It isn't the best and nicest way to go around and import trees, but
it serves its purpose.  I would certainly welcome such a thing in darcs,
appropriately implemented.  This probably wouldn't be a pre-1.0 feature though.

-- 
Ivan Stankovic, pokemon at fly.srk.fer.hr




More information about the darcs-users mailing list