Re: OS checks with configuration time DLOPEN_NEEDS_LIBDL

Le 30/09/15 16:38, Richard PALO a écrit :
> Le 22/09/15 18:51, Norbert Thiebaud a écrit :
> > On Tue, Sep 22, 2015 at 6:31 AM, Richard PALO <richard@netbsd.org> wrote:
> > > Since GUIBASE was replaced with the equivalent OS checks, there seems
> > > to be a bit of abandon in keeping multiple OSes up to snuff.
> > >
> > > in https://gerrit.libreoffice.org/#/c/18591/  I took our base in pkgsrc and
> > > given the biggest difference in unix systems is the need for -ldl or not for dlopen,
> > >
> > > I added a configuration check for DLOPEN_NEEDS_LIBDL and updated a number of the existing
> > > OS checks to check simultaneously for, among others, %BSD and SOLARIS in addition to LINUX.
> > >
> > > It would be nice to get some review on this approach.
> >
> > Jenkins got us a first review. the patch break on windows because it seems that
> > AC_SEARCH_LIBS([dlopen],
> > trigger a search of gcc...
> >
> > is there any chance that is relevant for windows ? if not the 'if
> > test' that out of configure ofr cygwin case (and prolly for macosx
> > case as at bes it will work , but it can pick up accidentally some
> > crap if the mac has some extra stuff installed)
> >
> > that being said couldn't this be done without putting all that logic
> > in the 'client' makefile.
> >
> > and just have $(if $(DLOPEN_NEEDS_LIBDL), -ldl) logic in gbuild and
> > define a gb_LIBDL defined to -ldl or empty (most likely in
> > RepositoryExternal.mk or in platform/* not sure on top of my head
> > which is the cleanest.
> >
> > and then have for example
> >
> > $(eval $(call gb_Executable_add_libs,gengal,\
> >     $(gb_LIBDL) \
> >     -lpthread \
> >
> > (and yes that could be extended to the few place that do the same kind
> > of exercise with lpthread... and there is maybe an even better way but
> > one step at the time...)
> >
> >
> > also you may want to split the part of the patch that touch
> > Repositoty.mk as that seems somewhat orthogonal with the -ldl stuff
> >
> > Norbert
>
> Well, personally I'm of the opinion that configure is the typical place for this type
> of check, in order to leave more pertinent things in platform...
>
> I added an `if test $_OS != "WINNT"` guard, I believe that should be ok.
> As far as -lpthread, I guess I agree...  I can add that to this now as well,
> or it could be a follow up patch.
>
> gerrit updated, if it's possible to launch a jenkins run.

I'd like to advance, one way or another on this topic.

I can add an addition guard to avoid MACOSX if need be but I doubt the use
of a non-configuration parameter buried in platform settings.

It would be a pity to leave things as they are as most BSD and SunOS currently need to 
seriously patch things in their ports.

-- 
Richard PALO