Re: -frecord-marker

[Keith Bennett <keith at mcs dot st-and dot ac dot uk>]

On Tue, Jul 25, 2006, Steve Kargl wrote:
> Which compiler vendor guarantees that a library you compile
> with their compiler will work flawlessly with a competitors
> compiler?  First and foremost, a library generated with gfortran
> needs to work with gfortran.  You may use -frecord-marker in
> a very safe consistent way, but you can't say how anyone
> else may abuse this option.   We, the gfortran developers,
> need to consider all of the pros and cons, and come up with
> a good technical design.

I wasn't asking for it to work flawlessly. I was asking for it to be
moderately compatible, in the same way that they have all made an
effort to be moderately compatible with each other.

Sure, spending time over the details is important and it is worth
exploring all the options to find a method that is as safe as possible.
I'm just a little concerned that many of the suggestions don't actually
solve the problem.

> I understand what you want.  Unfortunately, we aren't writing 
> "Bennett Fortran".  Adrain's patch may be the Right Thing to Do.

I don't think that I'm the only user that wants it. In fact I know
leagues of other fortran users who don't currently use gfortran for
exactly this reason.

> However, someone needs to be the Devil's Advocate to look for 
> problems and issues with the design.  The possibility of multiple
> different recorder markers written into a single files needs
> a solution.  That solution may be to prohibit this behavior such
> that the recorder marker of a pre-existing file will override
> any requested conflicting marker.

Sure. If detecting the format of the existing record-marker is easy
enough to implement then that may well be worth considering. Actually,
I would think that this is relatively simple since there are two identical
markers for each record. Impossible to be sure that your guess is correct
without reading the whole file, though.

If no totally safe method can be found, I personally think that usefulness
is slightly more important than safety or strict adherence to standards.
Otherwise Cray pointers would never have been accepted.

Keith.