Re: NEW: lang/flang

On Sun Dec 31, 2017 at 10:10:55AM -0500, Brian Callahan wrote:
>
> On 12/31/17 09:43, Rafael Sadowski wrote:
> > On Sat Dec 30, 2017 at 10:12:23PM -0500, Brian Callahan wrote:
> > > On 12/23/17 00:00, Brian Callahan wrote:
> > > > On 12/15/17 23:43, Brian Callahan wrote:
> > > > > On 12/10/17 12:27, Brian Callahan wrote:
> > > > > > Hi ports --
> > > > > >
> > > > > > Attached is a new port, lang/flang. Flang is a Fortran compiler
> > > > > > targeting LLVM.
> > > > > >
> > > > > > Supposedly Flang is on its way to becoming part of LLVM, though
> > > > > > there's no timetable for that.
> > > > > >
> > > > > > Flang is divided into two ports, the driver and the runtime. The
> > > > > > driver is a modified version of clang. I chose to only install
> > > > > > the flang driver binary from the driver port. The CMake build
> > > > > > system wants to install the entirety of the clang utils and
> > > > > > libraries and headers, but that means having three copies of
> > > > > > some things (1 in base, 1 from devel/llvm, 1 from flang), and
> > > > > > that seemed crazy. We only need the flang driver binary anyway.
> > > > > >
> > > > > > In order to link binaries with flang, you must remember to
> > > > > > include -L${LOCALBASE}/lib -lflang when linking the binary.
> > > > > >
> > > > > > I chose to use flang's built-in ompstubs library for OpenMP
> > > > > > "support." And by "support" I mean that the ompstubs library is
> > > > > > just stubbed out omp routines that do nothing but allow flang to
> > > > > > build. According to upstream, !amd64 archs use the ompstubs
> > > > > > library anyway. It would be nice to eventually move to real
> > > > > > OpenMP support, but that can be for a later day.
> > > > > >
> > > > > > Only tested on amd64 because I do not (yet) have an arm64
> > > > > > machine. Flang only supports those two platforms. I will add the
> > > > > > needed arm64 bits once I have my arm64 machine. Flang
> > > > > > successfully builds a bunch of F90/F95 programs I have and
> > > > > > passes a significant chunk of the NIST F77 test suite.
> > > > > >
> > > > > > OK?
> > > > > >
> > > > > > ~Brian
> > > > > >
> > > > > Ping. Tarball with latest version attached.
> > > > >
> > > > > OK?
> > > > >
> > > > > ~Brian
> > > > >
> > > > Anyone?
> > > >
> > > > Latest tarball attached.
> > > >
> > > > ~Brian
> > > >
> > > Weekly bump with the latest tarball attached :)
> > >
> > > ~Brian
> > >
> > Not much to grumble about.
> >
> > * Tab(s) after PERMIT_PACKAGE_CDROM
>
> Fixed.
>
> > * Some comments in the patches would be very helpful.
>
> Comments added, but only to the flang/flang part. The patches directory in
> flang/driver are direct copies of the clang patches from the devel/llvm
> port. I'll have to keep doing that going forward for ease of maintenance, I
> think.

Thanks, could we also add a comment in Makefile? So that no one forgets
to synchronize the patches with devel/llvm.

>
> >
> > Only build test on amd64. Is there any run test I could do?
>
> flang/flang is a test of itself, which is nice. Otherwise, your favorite F95
> programs?

Is there any clue?

$ flang hello.f90
/usr/bin/ld: cannot find -lflangmain
flang: error: linker command failed with exit code 1 (use -v to see invocation)