seth vidal wrote:
On Tue, 2007-12-18 at 13:39 +0100, Florian Festi wrote:
OK, after some more tests I found some more tweaks to reduce both runtime
and memory usage. I'd like to postpone the LRU cache stuff a bit but get the
sqlitesack changes and the additional tweaks in fast.
They include:
* SqliteSack
* Use pkgKey instead of pkgId
* Generate only one package object per package (Now fixed)
* Load only minimal tags per default.
* Create YumAvailablePackageSqlite._files and ._checksums on demand only
All of the above looks reasonable. I'm a little worried about api
artifacts and responsiveness in file-resolutions from the change to
_files but this all seems reasonable.
I changed Package.hash to ._hash to avoid API clutter. Responsiveness in
file-resolutions shouldn't be affected by the ._files change. Filelists
aren't loaded anyway. Or are you talking about the searchFiles implementation?
* Package
* Make Package.pkgtup a real tuple
How much testing has this portion of the patch seen? I ask b/c I
remembered trying to switch away from the pkgtup property before. IIRC,
the wheels fall off when you use the objects in interesting ways.
I really would like to try this as it saves a lot of time and memory. For a
F8 full install it saves 120000 tuples. Anyone else have an idea what to
check to find possible problems?
Florian
_______________________________________________
Yum-devel mailing list
[email protected]
https://lists.dulug.duke.edu/mailman/listinfo/yum-devel
