-frecord-marker

[Thomas Koenig]

On Tue, Jul 18, 2006 at 01:53:26PM +0100, Adrian Umpleby wrote:

> Well, after creating a short test, it's clear it's not a runtime option (which is what the online
> manual and man page claims).

Quite a number of people have used this option, and you're the
first to read the documentation that way.  Please feel free to
propose a documentation patch if you feel the documentation
is incomplete or misleading.  For a larger patch, you'll need
to get a copyright assignment to the FSF in place.

> So:
> 
> gfortran -frecord-marker=4 -c mymain.f
> 
> does produce a different object to:
> 
> gfortran -frecord-marker=8 -c mymain.f

Correct.  This behavior is shared by, for example, the -std= (where they
apply to different library behaviors) or the -fconvert flag.

If you use the -fdump-tree-original option, you'll see what actually
happens.

> Dead annoying!

If this feature is so important for you, please submit a patch.
The note above about the FSF paperwork applies here, as well.

(From a previous mail).
> > It only works if the main routine is written in Fortran. I've been meaning
> > to bring this up for a while because it makes it impossible to write
> > libraries using gfortran which are then called by C. (something I do all
> > the time at work).

This is doable with the right magic incantation.  -fdump-tree-original
is your friend here.

> It also makes it impossible to write a program that could test for and read either format.

Testing for the format is very hard.  As for reading/writing, you
can check the "CONVERT" option to OPEN for how it's done.