[Mageia-dev] bug 2317 revisited: --update option should behave like --search-media

[andre999]

AL13N a écrit :
>> On 21/06/12 22:01, AL13N wrote:
>>      
> [...]
>    
>> All this assumes that backport media will be treated as a normal update
>> media. That is certainly not my impression. My impression of backports
>> are being able to install a new blender for example, not having a system
>> where backports are just another update media and replace everything
>> available. The QA task for that scenario would be ridiculously huge. If
>> you want to have backports which go any further than backports testing
>> then you seriously need to rethink this idea.
>>      
> [...]
>    
>> The aim of fixing this bug is to reduce the complexity and extra
>> workload of working around it for QA. This assumption and solution
>> actually has the opposite effect, dramatically increasing the complexity
>> and workload. As I've explained, that is simply not possible if we want
>> to release timely updates.
>>
>> I hope this makes the situation clearer. There is a workable solution
>> but I'm afraid it isn't this one, for the reasons given above.
>>      
> No offense, but i think it didn't make myself clear and as a result i
> think you are not understanding this properly.
>
> my proposal is actually to make sure QA only needs to test twice for each
> package (both updates and backports).
>
> "My impression of backports are being able to install a new blender for
> example"
>
> this exact idea that you have, will make QA testing unworkable. let me try
> to explain:
>
> suppose that only blender and firefox and gimp and java is backported. any
> kind of combination would have to be tested to be able to support
> backports:
> - testing backports blender on a system without backports
> - testing backports blender on a system with backports and only firefox
> installed from backports
> - testing backports blender on a system with backports and only gimp
> installed from backports
> - testing backports blender on a system with backports and only java
> installed from backports
> - testing backports blender on a system with backports and both firefox
> and gimp installed from backports
> - testing backports blender on a system with backports and both firefox
> and java installed from backports
> - testing backports blender on a system with backports and both gimp and
> java installed from backports
> - testing backports blender on a system with backports and firefox and
> gimp and java installed from backports
>
> This for each arch: thus 16 tests.
>    

This analysis makes absolutely no sense.
All "cherry-picking" backports means is that a user can choose to 
install only one or several backport packages, just as a user can 
install only one or several optional release packages, or one or several 
proposed updates.
Do you really think that QA tests release blender with/without firefox 
installed, with/without gimp installed, etc ?  Considering that there 
are a lot more than 5 optional packages in a release, that would make an 
incredible number of tests.

> This amount of tests is a direct result of trying to support backports
> when you can have any single backported package installed, that you want.
>
> you'd have to test this because in case of new dependencies, it could even
> conflict during installation!!!
>
> and the biggest problem is that the same problem exists when having an
> update that has a new dependency. Thus, the same tests should be done for
> updates as well.
>
> all of this, just to support backports being cherry-picked.
>    

Just think what is meant by "cherry-picked".

> I'd rather have unsupported backports.
>
> My proposal (B2) is a compromise that has only supporting backports if you
> use it for everything, and has only 2 tests per package. THE SAME AS WE DO
> NOW!
>    

Which is all that is ever needed.

> to repeat: i'm trying to propose a solution that makes QA have NO INCREASE
> of workload.
>    

It does increase the total workload, as each backport package has to be 
tested in the release to which it applies.  But only one test per 
architecture.
Don't forget that backports will be leaf packages (or a set of related 
packages on which nothing else is dependant), so they will be simpler to 
test.

> the only extra point, is that for validating:
>
> right now, you're asking if it's tested for both i586 and x86_64.
>
> for B2, this is still the same, except that i586 should have backports
> disabled and x86_64 have backports enabled.
>
> I hope this is clearer now
>    

-- 
André