Re: policy request: 'make uninstall' should work

[Rodney Dawes <dobey novell com>]

On Mon, 2006-04-10 at 12:49 +0100, Ross Burton wrote:
> On Mon, 2006-04-10 at 13:39 +0200, Ruben Vermeersch wrote:
> > On Sun, 2006-04-09 at 20:59 -0700, Nigel Tao wrote:
> > > On 4/9/06, Joseph E. Sacco, Ph.D. <joseph_sacco comcast net> wrote:
> > > > support 'make uninstall'
> > > 
> > > Should this be an upcoming Gnome Goal??
> > 
> > That's pretty hard to make a goal of it, as this can break over time. A
> > tinderbox doing nothing but continuous 'make distcheck's would be of
> > more use, it could automatically notify the maintainer when the
> > build/(un)install is broken.
> 
> In an ideal world this isn't required as everyone should be releasing
> tarballs with make distcheck anyway.  At the least it saves foolish
> "tarball doesn't compile as foo.c is missing" bugs. :)

Unless foo.c is a Makefile conditional and is not getting disted without
the conditional enabled, and the person making the tarball hasn't
enabled that feature. This can be particularly problematic when people
write code where either one of a set of features may be supported, but
not all features in the set, during any one build. The distcheck itself
won't find these issues.

Perhaps we really need a set of build system guidelines, like the HIG,
and other guides, but for building packages. We could also have a
hello-world type package that includes examples of all the cases, and
how to deal with them.

-- dobey