Followup to FESCO meeting: firefox dependancy tracking.

[Ralf Corsepius]

On Thu, 2007-01-04 at 22:17 -0900, Jeff Spaleta wrote:
> As per the FESCO meeting item list, and irc discussion, here is my
> humble attempt to identify via repoquery what packages are currently
> prone to library dependancy breakage without being noticed by the
> automated scripts which just look for rpm autogenerated library
> dependancies.
> 
> these are packages which have a requirement on a library from firefox,
> but do not explicitly require firefox or gecko-libs:
> 
> epiphany-extensions-0:2.16.1-1.i386
> gnome-chemistry-utils-mozplugin-0:0.6.3-4.fc6.i386
> openvrml-gtkplug-0:0.16.3-1.fc6.i386
> openvrml-mozilla-plugin-0:0.16.3-1.fc6.i386
> 
> If you just look at packages which do not use a versioned firefox dep
> you also get:
> 
> devhelp-0:0.12-9.fc6.i386
> epiphany-0:2.16.2-1.fc6.i386
> galeon-0:2.0.3-4.fc6.1.i386
> gtkmozembedmm-0:1.4.2.cvs20060817-7.fc6.i386
> libswt3-gtk2-1:3.2.1-23.fc6.i386
> openvrml-0:0.16.3-1.fc6.i386
> yelp-0:2.16.0-11.fc6.i386
> 
> 
> The only packages which use a versioned firefox requirements are:
> 
> gnome-python2-gtkmozembed-0:2.14.2-6.fc6.i386
> liferea-0:1.0.26-2.fc6.i386
Hmm?

# rpm -q --requires  openvrml | grep firefox
firefox = 1.5.0.9

> My suggestion is that all packages which end up requiring a library
> from firefox should use a versioned dependancy as long as firefox
> continues to keep its libraries in a versioned directory tree (
> currently  /usr/lib/firefox-1.5.0.9/ ). 
ACK.


> Comments? Should I start filing bugs against these packages to get
> versioned firefox requires added to their specfiles ?
Probably. It actually depends on what a package needs a versioned
firefox dep for. 

In some cases, it's a library search path (Some packages use firefox
libs as system libraries, but they are out of ld.so's search path, some
explicitly dlopen them), in some cases it's a directory name, in some
cases it's a particular version of a firefox library (Firefox libs lack
proper SONAMES and properly versioned API).

> Should we look at making this sort of thing part of the review process
> that should be checked for?
Yes, .. better have firefox fixed. IMO it's "plain broken".

Ralf