I have extracted files into directories starting at /usr/local/g95/
and obtained all the files listed in the g95 manual. The executable
compiler is supposed to be in
/usr/local/g95/g95-install/bin/
with the name: i686-pc-linux-gnu-g95
and there is a file with this name in that location. I added the
symbolic link at /usr/bin/ as recommended.
However when I try to execute the i686-pc-linux-gnu-g95 I get
an error
bash: /usr/bin/g95: /lib/ld-linux.so.2: bad ELF interpreter: No such file or directory

I think the g95 may be a 32 bit app and I need to install some 32 bit
libraries but would rather not do this if there is a better solution.

Welcome to LQ.

# yum install glibc.i686


Are you sure that a 32bits fortran compiler can be of any use in a 64bits OS ?


.

Based on your recommendation, I went ahead and installed the additional 32 bit package. Now I get a new
error:

g95 Hello.f90
/tmp/ccfBpywF.s: Assembler messages:
/tmp/ccfBpywF.s:11: Error: suffix or operands invalid for `push'

The code is just:
program hello
print *,"Hello World!"
end program hello

Concerning the running of a 32-bit compiler on a 64-bit OS, there should not be a problem as long as we do not use the
extended address space available from a 64-bit system and as long as we do not use 64-bit integers. Obviously there
are some difficulties in setting up the compiler without the assistance of an rpm process. Any ideas about our new
error (which is far from self explanatory) would be welcome.

Suggest : A 64bits G95 http://www.g95.org/downloads.shtml
> g95-x86_64.deb
( # 'alien -r g95-x86_64.deb' creates : g95-20100616-2.x86_64.rpm )


Or may be the 64bits "f95 compiler" for your OS.
That's gfortran. Package name = gcc-gfortran


.

Thanks again for the help. The gfortran package was in fact already installed on our machine and appears to function
correctly. What we do not yet know is whether we can effectively us a 64-bit compiler. Our problem is a bunch of
legacy code (probably not a rare situation). Our binary records of observations have been managed using a method
that is generally considered poor code practice -- we just pack a number of different quantities into arrays then
pull out the parts as integers, reals or character strings depending on a mapping that is known to us. The mapping
is organized through a set of equivalence statements. What we will need to figure out is the flags necessary to have
the parts extracted as if the mapping were done using 32-bit quantities -- the reals especially. I think the integers
can be declared the same as before, possibly just from an implicit statement that applies to all the code. I am
not so sure about the reals.

Our code began in the late 1960's and has been ported forward a number of times.

Any comments about these word size questions would be welcome.