amber-developers: Re: AMBER: amber10 installation on sgi_mips

From: David A. Case <>
Date: Thu, 9 Oct 2008 12:47:55 -0400

On Thu, Oct 09, 2008, Roberto Gomperts wrote:
> Immediately I encounter several problems, starting with how configure_at
> generates config.h for irix_cc. For example it defines a "ranlib"
> function not present in irix.

will fix.

> Also no fortran compiler is set.

Probably correct as is: the *only* fortran compilers known to work with mopac
are g77 and gfortran. If neither are present, the install process should just
skip compiling mopac. Of course, it is possible that SGI's f90 would also
work, but I don't know that anyone has tested this.

I keep hoping we can deep-six the mopac code soon...Ross??

> reduce/Makefile uses a CURRENT_DIR variable that has not been set
> reduce/reduce_src/Makefile has a bug in defining the OBJLIST:
> OBJLIST = GraphToHoldScores.o reduce.o CTab.o ElementInfo.o StdResH.o
> ResBlk.o AtomConn.o \
> AtomPositions.o DotSph.o Mover.o RotMethyl.o RotDonor.o \
> FlipMemo.o CliqueList.o AtomDescr.o PDBrec.o MoveableNode.o \

Will fix, but this and other parts of AmberTools (sleap & gleap) really
require the GNU versions of make and lex. I'm wondering if it is just not
simpler to say: "you have to install GNU tools to use this software": does
the number of IRIX and Solaris users justify the efforts to keep our code
compatiable with "legacy" Unix?

> So a continuation "\" in the last line.

will fix.

> I fix this and then I get into actual compilation errors.
> For example:
> ./nab -c
> cc -c -woff 1116,1167,1174,1552 -Wl,-dont_warn_unused -DSYSV
> -DBINTRAJ -o dump.o dump.c
> cc -c -woff 1116,1167,1174,1552 -Wl,-dont_warn_unused -DSYSV
> -O0 -DSYSV -DBINTRAJ -Dlex embed.c
> cc-1028 cc: ERROR File = lex.dg_options.c, Line = 45
> The expression used must have a constant value.
> FILE *dgoin = {stdin}, *dgoout = {nabout};

This is actually odd, since we do test nab installation on an (old?) sgi_mips
IRIX machine. But it could be that I had installed flex (=GNU lex) on that
machine, which I no longer have access to.

> I suppose that I am a bit surprised by all of this as I have not
> problems building for x86_64 and ia64 with the Intel compilers.

These are not compiler issues, I think. Rather, it's the old Unix versions of
make and lex that are only partially compatible with flex that are at fault

Received on Sun Oct 12 2008 - 05:10:04 PDT
Custom Search