Re: Testing migration doesn't check build-depends?
Andreas Barth wrote:
> * Luk Claes (email@example.com) [080511 19:09]:
>> Adeodato Simó wrote:
>>> * Steve Langasek [Sat, 10 May 2008 23:03:25 -0700]:
>>>> That wouldn't be a good idea unless we first get support for keeping
>>>> old library packages around in testing to allow asynchronous transitions.
>>>> Otherwise, testing transitions would become far more brittle than they
>>>> currently are.
>>> Oh, point. But I don't think there were plans to make the current
>>> britney support build-depends, and britney2 *already* supports keeping
>>> old library packages around.
>> I don't understand why dependency checking plus build dependency
>> checking would need support for keeping old library packages while
>> dependency checking without build dependency checking doesn't need it?
> Eh, it's so:
> A transition are a couple of packages (up to a few hundered sometimes)
> that are glued together and can only go to testing at the same point in
> time. Issue with that is that all of them need to be ready at one
> britney run.
> Now, there are proposals that reduce the glue possibilities (or: making
> the glue clowds smaller), and there are some that increase the glue
> possibilities. One can also use another word for that: Some make testing
> migrations easier, some make them harder.
> In Steves and my opinion, there are already enough (or even: too many)
> large transitions which are already hard enough, so we want to avoid
> changes making migrations harder unless there are (at least) the same
> amount of easier things in place compared to now.
Ok, so it's probably preferrable to also implement something that makes
transitions easier while implementing the check on build dependencies,
though it's not absolutely needed from a technical point of view :-)