From:
[email protected]
On Wed, Aug 04, 2004 at 09:16:12PM +0200, Chris Niekel wrote:
Hi,
Hello!
I've tested 0.9.23-1 as well, and although memory-usage seems to
increase slower, it still went past 500MB. I had set a limit there (to
avoid random things getting killed) (ulimit -v 512000).
du -ks /etc yields:
33076 /etc
This is a known issue, and I'm pretty much resigned to living with it.
Most darcs operations use about 10 times the size of the data involved
(this is obviously only crudely true). The initial record for some reason seems to take something like twice what I'd expect by the x10 rule, and I haven't been able to track down why--memory usage debugging in a lazy garbage-collected language isn't easy.
This is a problem I've taken a shot at, spent a week or so on, and then
given up on. I'd love for someone else to take a shot at it, but am not terribly optimistic, and there's no point spending too long working on the memory usage of the initial record, since a darcs get of the recorded patch will still take ten times the size of the patch since it'll have to hold
the parsed patch in memory.
If I can do more for testing, don't hesitate to contact me.
Well, I'd guess that the memory usage for this record ought to be somewhere around 600 to 700 megabytes. If your machine can handle that (it should be pretty virtual-memory-friendly), if you could verify that it'll run with
that much memory at its limit, that would be nice.
As far as workarounds go, you add and record one subdirectory at a time.
This produces (obviously) smaller patches.
--
David Roundy
http://www.abridgegame.org
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (GNU/Linux)
iD8DBQFBG0kGQ6uZI9PVvOwRAtuaAKDMLHRP7Okr/5XGVAZndL+1OGoj0wCfVlZM GSkMAJ/9cbj7Igd4thywzWU=
=2P/C
-----END PGP SIGNATURE-----
--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)