Re: A question about update-excuses (was Re: testing is broken)

[Philip Blundell <philb@gnu.org>]

>No, that doesn't follow.  
>
>You can either (a) load up the buildd machines for a limited period of time
>(until they 'catch up') and then start checking build-deps properly again,
>or (b) have one buildd machine (say, the i386 one) do full checking of
>build-deps, and the rest can be lax.   There are probably other options.

Yeah, maybe.  I'm sceptical that the first option would buy much -- it would 
certainly give a short-term improvement and allow more packages to go 
into testing today, but they'd rapidly become outdated again as soon as we 
started requiring proper build-depends.  Having just one buildd do checking also 
wouldn't do us much good: even one architecture falling behind will keep the 
binaries out of testing, and you'd end up concentrating the burden of 
reporting these bugs on the administrators of just that one autobuilder.

Fundamentally, the problem is that the maintainers don't actually use the 
source dependencies directly, and they don't have any easy way to verify that 
they're correct.  I'd guess that something like 5-10% of the packages being 
uploaded have some kind of build-depends problem.  Often it takes several 
iterations to get them right, and new versions of other packages can require 
them to change again ("xpm4g-dev" being a prime example; I got so fed up of 
filing bugs on that one that I did end up giving in and hacking the build 
daemon to work around it.)

p.