[Mesa-dev] mesademos build system: would one be enough?

[Jose Fonseca]

----- Original Message -----
> On Mon, Nov 21, 2011 at 11:17 AM, Jose Fonseca <jfonseca at vmware.com>
> wrote:
> > Johannes Obermayr's recent patch series remind me of one thing I've
> > been planning to ask here for quite some time:
> >
> > Would anybody oppose dropping automake build system in mesademos
> > for just cmake ?
> >
> > On Mesa there is rationale behind the two different build systems
> > (automake and scons), but for mesademos, there's nothing that
> > automake can do that can't be made with cmake; both generate
> > makefiles giving a similar experience to Linux/Unix developers;
> > cmake is already exclusively used in piglit, so there's
> > substantial expertise/familiarity on mesa developers community
> > already; and cmake has the advantage over automake of supporting
> > native Windows/MSVC builds.
> >
> > So I really don't think there's any need for automake on mesademos
> > anymore.
> >
> > I know mesademo's cmake build system does not yet build some
> > apps/demos, but this is some thing that I could easily and gladly
> > fix if we agreed on deprecating automake over cmake on mesademos.
> 
> I've no real objections to it but the big difference between piglit
> and mesa-demos is that distros package mesa-demos, nobody packages
> piglit really and there is little reason to.
> 
> so dropping automake means all the downstream consumers have to redo
> their packaging to use cmake which can be a bit painful.
> 
> Though I'm sure fedora has lots of simple cmake things to steal
> packaging info from.

Thanks.

Yes, I agree the needs of linux distribution packagers would have to be address too. But my impression is also that cmake is not particularly quirky or hard to package provided minimal care is taken. And some of Johannes Obermayr's patches seem to address some this.

BTW, I think there was some interest in getting piglit to install as normal packages do. The use case, at least for me, is to allow piglit to be built on a build machine, and then tested on a separate test cluster (avoiding the need to build piglit on every node, or share the whole source tree with NFS). But actually packging seems unnecessary to me as well, given the rate at which tests are being added to piglit.

Jose