-frecord-marker

[Janne Blomqvist]

On Fri, Jul 21, 2006 at 08:23:03AM +0200, Thomas Koenig wrote:
> For records of length less than 2**31-1 bytes, Intel does the
> same as just about anybody else, use four-byte record markers.
> gfortran deviated from that for good reasons, but we don't (yet)
> have enough "clout" in the market to force other compiler vendors
> to add compatibility to what we do.
> 
> Interoperability _is_ important for a sizable number of users,
> who, for a number of good reasons, won't switch over to gfortran
> completely (which would be too much to expect right now, really).
> 
> There are also good reasons why these users might want more
> fine-grained control than what we currently offer (just one option,
> to be set when the main Fortran program compiles).

I think the problem is that we can't make everybody happy. Instead, an
effort to please everyone will result in gfortran gettin bogged down
in a morass of mutually incompatible and inconsistent linker options,
compile-time options, environment variables, non-standard extensions
and whatnot.

> The real question is how to implement this.  I can understand that
> you and Jerry don't want to add Yet Another Extension to the OPEN
> statement.

Personally, I'd be happy if we had a single record marker format (be
it the current 64-bit marker or the Intel method), and directed users
to use F2003 stream (or heaven forbid, C) to read/write other record
marker formats. Everybody has slightly different needs, we can't meet
them all by adding yet another "feature".

-- 
Janne Blomqvist
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 187 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/fortran/attachments/20060721/c878b4f7/attachment.sig>