[OpenAFS-devel] Re: adding "make check" to build slaves

[Derrick Brashear]

On Thu, Apr 12, 2012 at 8:20 PM, Russ Allbery <rra@stanford.edu> wrote:
> Jason Edgecombe <jason@rampaginggeek.com> writes:
>> On 04/12/2012 02:34 PM, Russ Allbery wrote:
>
>>> runtests itself assumes that something else will handle this, since for
>>> all of my other projects I use libtool. =A0OpenAFS is somewhat unique
>>> among projects using that test driver at the moment in its reinvention
>>> of that wheel. =A0(For reasons that I understand; libtool has its bugs
>>> too, of course. =A0But this is something that it normally handles
>>> reasonably well with current versions.)
>
>> is there a reason why we shouldn't use libtool? licensing perhaps?
>
> I believe that we should use both Libtool and Automake, but the amount of
> work required for a transition is immense, and will also require working
> with the upstreams of both projects to resolve bugs. =A0OpenAFS is more
> broadly portable than a lot of free software, and has some interesting an=
d
> special build system challenges (such as the separate build of kernel
> modules).
>
> I've looked at doing this incrementally a few times, as have other people=
,
> but it's a very difficult change to make incrementally, and the amount of
> effort required is daunting. =A0It's generally far easier to make
> incremental improvements to our current ad hoc arrangement of shell
> scripts with portability testing on the platforms we know we care about.

Yeah. I did the first autoconf support while on a train to Boston, and
because it wasn't
overly ambitious, it was doable. Incremental automake is much harder.

Personally, I hate automake, and I find libtool's .lo pollution to be
tedious but at least
bearable. Sadly, all the other options suck differently, including the
status quo.

Derrick