[Yum-devel] Sqlite improvements

[seth vidal]

> Just a quick follow up that I have implemented a first version of
> filelists.xml now. Only one drawback: its very slow at the moment,
> maybe this can be resolved (partialy) by using the same trick that rpm
> uses to store file info in its internal database: splitting filenames
> up into two different tables: a directory and a filename table and to
> join these via a key. I'll try if this speeds things up a bit, if not
> maybe we should consider a hybrid aproach that only stores primary
> metadata in sqlite and uses general pickles for filelists metadata.
> 
> Maybe splitting filenames into file and directory might also be an
> idea for the current pickle files, I guess it should reduce their size
> drastically (on my machine filelists.xml.gz pickle for fedora-updates
> is 31 Megabytes(!))

The problem I've seen with reading in the filelists.xml is the number of
nodes in the xml, not actually with the data itself. So I'm not sure
splitting into dir/file is going to be _that_ helpful, but it's worth a
shot.

-sv