On Fri, Oct 11, 2013 at 08:00:41PM +0200, Jérémie Courrèges-Anglas wrote: > Kenneth R Westerback <kwesterb...@rogers.com> writes: > > > On Thu, Oct 10, 2013 at 04:16:31PM -0700, Josh Elsasser wrote: > >> On Thu, Oct 10, 2013 at 09:29:40AM +0200, J?r?mie Courr?ges-Anglas wrote: > >> > > >> > So instead of struggling with clisp, let's just update sbcl first. > >> > Regress tests results and diff below. I'm postponing clisp for now. > >> > > >> > More tests on amd64 / ok? > >> > >> There should be a build-time (and probably runtime) dependency on gmp, > >> or else the gmp contrib should be disabled. I'm not sure if it's right > >> to enforce a runtime dependency when few users will use sb-gmp, but it > >> would also be annoying to start splitting off contribs into separate > >> packages. > > gmp is opened using load-shared-object (dlopen), the build system > doesn't search for it (but we should keep in mind that one day it > might use sb-grovel to detect the size and signedness of gmp objects). > Given this, plus the fact that it is a relatively new and probably > seldom used module, I don't think adding gmp to bdeps/rdeps would add > much value. > > What I should have done, though, is to put gmp in TEST_DEPENDS. > > Josh, do you agree with this?
Will the port build correctly without gmp installed? I would assume that the sb-gmp contrib would fail to build, and then the port bail out when files from the plist are missing. I'm fine with leaving it out of the runtime deps, but think that it should be declared as a build dep if it's really needed then.
