On Sun, Feb 13, 2011 at 12:42:57PM +0100, Bernhard R. Link wrote:
> * Roger Leigh <rleigh@codelibre.net> [110212 21:58]:
> Allowing things to build in a non-artificial environment is simply an
> important part of being a good free software citizen. We as packagers do
> not like it if upstream has an arcane build system that mostly only
> works on their build servers, so we should also allow out users to
> get things builds.
>
> > The former situation is simple, robust and maintainable. But the
> > latter, it's a virtually intractable problem, and given the lack of
> > concern about it up to now, it's not a major worry for most people,
> > and from a cost/benefit POV it doesn't look practical.
>
> Nobody claims it is easy.
>
> I think there is even a measurable part of people producing binary-only
> "freeware" mostly because they do not want to be bothered by people with
> problems compiling their source.
>
> But it is an essential part of doing it right, so we should try to do
> our best and not just give up early.
I am by no means saying we shouldn't do it. I'm just saying that:
• having the buildds using a clean minimal environment is good for
the consistency of officially built and distributed packages.
In this situation, we specify *exactly* what is required, and
this is possible to get exactly right.
• adding build conflicts to ensure it will build on any arbitrary
system with a random selection of installed packages will always be
on a "best effort" basis:
· there are too many combinations to test
· as the source packages and the rest of the system changes, the
conflicts that are added may become outdated, and without
proper testing will bitrot
· as the system changes, new conflicts may be required
· we can not guarantee that the package will be identical to those
built in a clean+minimal environment; there are just too many
variables, especially when features are autodetected
I would have thought that the current situation, where developers add
conflicts as they find such things on their own systems, or others
report them as they find them, is quite sufficient. But supporting
much more than this is not practical. I am not suggesting at all that
packages which don't build on more than minimal setups are not broken,
just that we can't guarantee a successful and/or reproducible build on
every random system out there.
Regards,
Roger
--
.''`. Roger Leigh
: :' : Debian GNU/Linux http://people.debian.org/~rleigh/
`. `' Printing on GNU/Linux? http://gutenprint.sourceforge.net/
`- GPG Public Key: 0x25BFB848 Please GPG sign your mail.
Attachment:
signature.asc
Description: Digital signature