This is the mail archive of the cygwin mailing list for the Cygwin project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

RE: Fortran Compiler Error CMBFAST

Dave Korn wrote:
>   We *need* to see the actual command line.

That was in the original post (sorry, I should have made
sure it was included in the text when I CC'ed you...):

> f77 -O2   -c -o jlgen.o jlgen.F

The command was being executed from the same directory
as jlgen.F, which also contains the include file that
it cannot find. Adding -I. to the command line fixes
the problem.

The OP didn't provide enough information about where
the files resided, but I downloaded the CMBFAST package
myself and verified that "./configure && make" results
in the same error. I also verified that it builds if
you manually compile with "-I." (or add -I. to FFLAGS
in the makefile).

> Is everyone in this thread aware of the effects of
> -I- [...]

I assume that's not relevant since -I- is not used.

> It needs to be pointed out that adding "-I." doesn't
> help any if you aren't building in the source dir but
> a parallel object dir...

True. This was a workaround for this specific case. In
general, you would want to use -I$(SRCDIR) or something
along those lines, although technically I don't think
you should need to do either. :-)


I just recreated the problem with some minimal code,
and got some more information in the process:

Create a file "foo.for" that contains one line:
        include ''

Create an empty file named in the same

Copy "foo.for" to "foo.f", "foo.FOR", and "foo.F" and
try to compile all four of them from that directory:

$ f77 -c foo.for    # runs OK
$ f77 -c foo.f      # runs OK
$ f77 -c foo.FOR    # runs OK
$ f77 -c foo.F
           include ''
Unable to open INCLUDE file `' at (^)
$ _

So it seems that the rules by which the preprocessor
looks in the current directory and/or the directory
containing the source file does not apply to files
named *.F for some reason. It's possible that this
was intentional, although I don't get this behavior
on my Linux box (which granted is using GCC version
3.2.3, so it could have changed since then...).


Unsubscribe info:
Problem reports:

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]