[Yum-devel] random todos

[James Antill]

On Tue, 2010-06-01 at 21:50 +0300, Ville Skyttä wrote:
> The other nagging thing I have on my "list" is that Provides versioning should 
> trump shortest name when deciding best packages for a dep.  For example if 
> foob provides bar = 1.0 and foobar provides bar = 2.0 and something asks for 
> bar, it should resolve to foobar (because bar = 2.0 is "better" than bar = 
> 1.0) and not foob (like it currently does, because foob is shorter than 
> foobar).  One real world case for this in Fedora and friends is various java 
> (JDK) packages.  I believe I've suggested/reported this earlier but to my 
> surprise it turned out somewhat controversial.

 While it's not a bad idea in theory, the problem is the packages aren't
built to use that in practise ... so you have problems like:

java-1.5.0-gcj-devel provides java-devel = 1.5.0

java-1.6.0-openjdk-devel provides java-devel = 1:1.6.0

...or in other words they both provide "GENERIC-PROVIDE = %{epoch}:
%{version}", which means we shouldn't do a version comparison as it's
the same as doing a version comparison on the providing packages ==
meaningless, IMO.
 Of course it's kind of moot because some stuff¹ requires gcj (via.
java-gcj-compat) and some stuff² requires openjdk (via.³ java >=
1:1.6.0) so my desktop gets both non-devel things anyway.


¹ Including openoffice

² Including eclipse

³ And, yeh, using epoch here is bad too IMNSHO.