cvs commit: ports/lang Makefile ports/lang/g95 Makefile distinfo pkg-descr pkg-plist ports/lang/g95/files patch-fibheap.c patch-mf-impl.h

Steve Kargl sgk at troutmask.apl.washington.edu
Mon Feb 2 08:04:00 PST 2004


On Mon, Feb 02, 2004 at 10:25:17PM +0900, Nakata Maho wrote:
> Message-ID: <20040202041956.GA85518 at troutmask.apl.washington.edu> ?$B$N5-;v$K$F
> Steve Kargl <sgk at troutmask.apl.washington.edu> ?$B$5$s$O=q$+$l$^$7$? :
> 
> > 
> > Please back this out.  There are two different compilers
> > named g95 and gfortran, respectively.  g95 is the name
> > used by Andy Vaught for his compiler, which you cannot
> > compile of FreeBSD.  gfortran is the name of the Fortran
> > 95 compiler that is in the tree-ssa branch of the GCC
> > cvs source tree.  I believe that the gfortran developers
> > would prefer to avoid the confusion between g95 and
> > gfortran.
> 
> Really? I extracted
> http://g95.sourceforge.net/g95-x86-linux.tgz

This is a linux binary only distribution of g95.  AFAIK,
Andy creates a new binary nightly, but only releases updates
to his GPL source code on a very sporadic basis.

> from Andy Vaught's site and found a directory named
> g95-install/lib/gcc-lib/i686-pc-linux-gnu/3.5-tree-ssa.
> so I thought that both two compilers are integrated.
> Anyway I will ask him about it.

It is a political problem caused by Andy.  Andy opened
his source tree up for contributions from others.  Paul
Brook and Steven Bosscher wrote all the initial code to
interface Andy's frontend to the tree-ssa gcc backend.
Paul and Steven were working and improving g95 too quickly
for Andy, so he suddenly removed their access to the g95
source code.   Paul, Steven, and Toon Moene relucantly
forked Andy's code by importing the last available version
into the tree-ssa branch.  Paul, Steven, and Toon also
decided to rename the compiler to gfortran because they
intend to support the Fortran 77, 90, 95, and the upoming
Fortran 2003 standard and they wanted to avoid confusion
with Andy's compiler.

> > Also, I build and run 250000 LOC of Fortran 77 and
> > Fortran 90 through gfortran several times a week without
> > patching the source.  Why are patch-fibheap.c and
> > patch-mf-impl.h necessary?  Have you filed a PR with
> > GCC?
> No.
> 
> Just compiled and didn't pass compilations. this is
> very ad hoc patch. I'll add --disable-libmudflap and maybe
> I can remove patch-mf-impl.h. 

Most likely, Andy's tree-ssa branch is out of date and if you're
building a tree-ssa gcc for FreeBSD to use with Andy's binaries,
then you're mixing FreeBSD and Linux ELF formats.  Try doing

cvs -d :pserver:anoncvs at gcc.gnu.org:/cvs/gcc login
cvs -qz9 -d :pserver:anoncvs at gcc.gnu.org:/cvs/gcc co \
    -r tree-ssa-20020619-branch gcc

../gcc/configure --prefix=${LOCALBASE} --disable-libmudflap
 --enable-languages=c,f95 --with-gmp=${LOCALBASE}

> > Finally, the gfortran run time library uses the C99
> > round[f](3) math function.  FreeBSD does not have
> > this function implemented.  See the FreeBSD-standards
> > mailing list for details.
> 
> Then what should I do?
> 

See the FreeBSD-standards mailing list for my quick and dirty
implementations of round[f](3).

-- 
Steve


More information about the cvs-ports mailing list