[Fedora-packaging] Proposal to limit file deps

[Toshio Kuratomi]

Hey all, I've just added a draft proposal for limiting file dependencies
in packages.  It's a "should" recommendation at this point.

http://www.fedoraproject.org/wiki/PackagingDrafts/FileDeps

= File Deps Guideline =

== Problem ==

Resolving file dependencies can be a bottleneck when running yum.  With
our current repodata format, information on which package owns files
in /etc and the bin directories (/bin /sbin /usr/bin /usr/sbin) are kept
in primary.xml.  All other file deps are kept in filelist.xml.  This
means that anytime yum has to resolve a file dep in /etc or a bin
directory it can just download primary.xml to resolve it.  If the file
dep is outside of those directories, yum has to download and parse a
second xml file, filelist.xml, which is much larger than primary.xml.

Fedora 6 repodata compressed and uncompressed:
||       ||primary.xml||filelist.xml||
||Core   ||1M/7.6M    ||3.1M/39.2M||
||Extras ||4.7M/11.7M ||4M/50.8M||
||Updates||0.4M/4.2M  ||2.1M/27.5M||

The compressed size has direct bearing on the additional time it will
take to download the additional filelist.xml.  The uncompressed size is
one indicator of how much additional time is needed to parse for the
file that satisfies the dependency.  If yum does not need to refresh its
primary.xml and filelist.xml (because it downloaded a copy previously
and the repository has not been updated since), yum won't have to parse
a new file and the filelist info will be quicker to find (as the file
will have been parsed into an sqlite database), however this is the
exception rather than the rule.

If we can avoid having to download and parse filelist.xml, we can
greatly speed up yum's dependency resolution times.

== Proposed Guideline ==

'''SHOULD''': If the package has file dependencies outside
of /etc, /bin, /sbin, /usr/bin, or /usr/sbin consider requiring the
package which provides the file instead of the file itself.

Using file dependencies outside of /etc, /bin, /sbin, /usr/bin,
or /usr/sbin requires yum (and other depsolvers using the repomd format)
to download and parse a large xml file looking for the dependency.
Helping the depsolvers avoid this processing by depending on the package
instead of the file saves our end users a lot of time.

-Toshio
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
URL: <http://listman.redhat.com/archives/fedora-packaging/attachments/20061220/240035b0/attachment.sig>