Excessive package interdependency

[Tyler larson]

On Thu, 2003-12-18 at 10:44, Jef Spaleta wrote:
> Tyler larson wrote:
> > This is obviously wrong, but is still the only option we currently have.
> > I think Aurelien's suggestion about a "Recommends" or "Suggests" list in
> > addition to "Requires" would make a lot more sense
> 
> Doing anything like Suggests/Recommends correctly would take a HELL of a
> lot more package testing in order to do correctly. How does a packager
> really know that a linked library is not a required dependancy to
> prevent the program from running in most/all usage cases?
> 
> ldd /usr/bin/mc
> 
> How does the packager know which of those libraries are not true
> dependancies? I think we are talking about an order of magnitude more
> testing to get Suggests/Recommends to work reliably correctly.
> 
> -jef

If it's a linked library, it's not optional (even if it *could* be
optional under a different build environment or in specific user
settings). What I'm talking about is the practice of taking truly
independent packages and creating an artificial dependency (just in the
rpm spec) simply so that the setup tools will install the second package
when the first package is selected (solely for additional
functionality). What I'm suggesting won't require any additional
testing. You're not *removing* existing dependencies, you're just not
adding artificial ones.

Excessive dependencies where the software actually *is* dependent on too
many other packages can't be solved simply by changing the package
management paradigm. You need to work that part out in the code first.
Such software would need to be changed to determine what extensions
exist at runtime, rather than at compile time.