Re: OPEN statement binary equivalence with F2003

[Tobias Burnus <burnus at net-b dot de>]

Mike Toews wrote:
> First, I cannot seem to find any documentation on the OPEN statement
> with the current version of the onlinedocs. Instead, I only see useful
> fragments, e.g. [1]. Is the OPEN statement documented for gfortran
> anywhere?

The online documentation tries to document the supported command line 
arguments plus the most important compiler specific features (such as 
extensions). The latter also includes the documentation of all intrinsic 
procedures - both the compiler/vendor extensions and the ones of the 
Fortran standard.

However, OPEN is a statement and currently statements are not 
documented; this might change in the future but it is a lot of work. You 
are essentially asking for a full language reference. But the best 
definite source of the Fortran language is the Fortran standard. See 
http://gcc.gnu.org/wiki/GFortranStandards - or in a more readable form, 
in several Fortran text books.

The only exception are compiler-specific extensions, which are hopefully 
mostly documented in gfortran's documentation.

> A topic that needs some documentation is the vendor-dependant and
> non-Standard FORM='BINARY',

Well, I do hope that those compilers have documented it. Admittedly, one 
could also add it to "Extensions not implemented in GNU Fortran", 
http://gcc.gnu.org/onlinedocs/gfortran/Extensions-not-implemented-in-GNU-Fortran.html


> which I see in source code like MODFLOW,[2] which is developed by the USGS with Intel compilers with default FORM='BINARY' (although this is customisable with openspec.ini). Up until now, I had assumed that gfortran was unable to read/write a similar data structure to the programs produced by other compilers that supported FORM='BINARY', and I was silently awaiting this support with a future release.
>
> I thought the only supported binary-like mode was OPEN(...,ACCESS='SEQUENTIAL',FORM='UNFORMATTED'),

First, sometimes it helps to ask - because then someone could have told 
you to try "stream".

Secondly, besides sequential and Fortran 2003's stream there is also 
"direct". "stream" mimics C's I/O while "sequential" writes some special 
information to quickly jump forth to the next record or back to the 
previous one.

Compilers added some vendor extensions to allow writing code in the C 
way - and users used them because the Fortran standard only added it in 
Fortran 2003. In general, vendor extensions should be avoided as they 
are compiler-specific and not well defined.

> On a related note, perhaps gfortran should support FORM='BINARY' by
> resetting the OPEN statement parameters with the F2003 equivalent, and
> possibly emitting a warning to the user.

Maybe. In any case, using vendor is discouraged and implementing new 
features (and fixing bugs) cost time - and Fortran-standard features 
have a higher precedence that adding vendor extensions, especially if a 
Fortran-standard replacement exists.

In any case, contributions* to GCC/gfortran are welcome - especially as 
the compiler is mostly developed by unpaid volunteers. (* documentation, 
feature implementation, bug fixes but also bug reports.)

Tobias