[Yum-devel] Obsoletes Test Cases

[Florian Festi]

Hi!

Same test environment, same packages, same variations, new game: Obsoletes

Packages available in i368, x86_64 and noarch:
  * Installed packages: zsh-1-1.arch - iarch
  * Obsoleting packages (obsoletes: zsh): zsh-ng-0.3-1.arch - oarch

Additional there is
  * A package requiring obsoleting package(s) (requires: zsh-ng): 
superzippy-3.5-3.noarch - ro

Each test calls runOperation with 3 parameters:
  * Command list
  * Packages installed in the rpmdb
  * Packages available in the repository

There is always one "yum update" test case that should handle the obsoletes 
and then we have one test cases where the obsoleting package(s) are required 
by the to be installed package ro (aka superzippy aka "Requires Obsolete").

testObsoletenoarchToMultiarch - works
     runOperation(['update'], [inoarch], [oi386, ox86_64])
     oi386, ox86_64 # ox86_64 only would be an alternative policy
testObsoletenoarchToMultiarchForDependency
     runOperation(['install', 'superzippy'], [inoarch], [p.oi386, p.ox86_64, 
p.ro])
     #ox86_64, ro
     inoarch, ox86_64, ro
     # likely to run into file conflicts, additional test case indicates 
that "yum update" doesn't fix that problem as the obsolete is already 
installed - yes, this is the "right behavior"

testObsoletei386ToMultiarch - works
     runOperation(['update'], [ii386], [oi386, ox86_64])
     oi386
testObsoletei386ToMultiarchForDependency
     runOperation(['install', 'superzippy'], [ii386], [oi386, ox86_64, ro])
     #ox86_64, ro
     ii386, ox86_64, ro
     # similar to the case above - see below
testObsoletei386ToMultiarchForDependencyFix # try if it goes away with "yum 
update"
     runOperation(['update'], [ii386, ox86_64, ro], [oi386, ox86_64])
     ii386, ox86_64, ro
     # yes, the obsolete is already installed - not removing ii386 is the 
right thing here - but that doesn't help us...

testObsoletex86_64ToMultiarch - works
     runOperation(['update'], [ix86_64], [oi386, ox86_64])
     ox86_64
testObsoletex86_64ToMultiarchForDependency
     runOperation(['install', 'superzippy'], [ix86_64], [oi386, ox86_64, ro])
     #ox86_64, ro
     ix86_64, ox86_64, ro
     # ...

testObsoleteMultiarchToMultiarch - works
     runOperation(['update'], [ii386, ix86_64], [oi386, ox86_64])
     oi386, ox86_64
testObsoleteMultiarchToMultiarchForDependency
     runOperation(['install', 'superzippy'], [ii386, ix86_64], [oi386, 
ox86_64, ro])
     #oi386, ox86_64, ro
     ii386, ix86_64, ox86_64, ro
     # ...

testObsoleteMultiarchTonoarch - works
     runOperation(['update'], [ii386, ix86_64], [onoarch])
     onoarch
testObsoleteMultiarchTonoarchForDependency
     runOperation(['install', 'superzippy'], [ii386, ix86_64], [onoarch, ro])
     #onoarch, ro
     ii386, ix86_64, onoarch, ro
     # ...

Looks like packages added for dependency are not checked as possible 
obsoletes. This explains all of the problems seen above. Normal obsoletes 
seam to work quite well. Unfortunately the (correctly implemented!) 
obsoletes semantic doesn't clean up the problems created by the installs for 
dependencies.

Opinions?

Florian