Re: Upcoming FTPMaster meeting
* Lars Wirzenius <liw@liw.fi> [110212 20:40]:
> On la, 2011-02-12 at 20:22 +0100, Bernhard R. Link wrote:
> > If the packages used are only ever built in unnatural virgin
> > environments, there is basically no testing if building them on
> > a real user machine works. And things not tested usually just stop
> > working after some time...
>
> Right now, such testing of such build environments is pretty haphazard
> already. Developers presumably build things with debuild on their
> development machines, but that's a very small test.
Testing is not good, and I think it got a lot worse in recent years.
New developers are nowadays told to only build packages in artificial
environments, buildds test this less and less (in the past, they mostly
only installed new packages and only deinstalled for conflicts, now the
mayority or even all deinstall as much as they can).
> If we wanted to be serious about this, it would be nice for someone to
> set up a maximal build chroot: something with as many packages installed
> as possible.
That is a nice test, but also another additional test. Apart from
problems with automatically installing packages (many packages not
usually being build-depended on tend to want to start things or get
stuff configured or things like that), you can never test all possible
combinations. While DD unstable machines usually sould have the
common combinations of packages installed, that a user may also have.
> On the other hand, I don't see that this is all that important. Most
> packages will build fine in a dirty environment, and if there's trouble,
> users can report problems, and then they can get fixed.
Users or porters or non-maintainer DDs usually do not look at packages
without a reason, but because of other problems. Making them run into
unrelated problems they might even need some time to properly recognize
as the cause for their problems is the worst time this problem can show
up.
Bernhard R. Link
Reply to: