(j3.2006) paper J3/14-174

[Malcolm Cohen]

>Paper J3/14-174 proposes to allow a file to be
>connected to more than one unit simultaneously.

Actually it is proposing to stop forbidding it, and to allow a processor to 
provide the facility to a standard-conforming program.

>How would the effects of I/O operations on those
>units be interleaved?

Not.

>  For example, if a file on
>magnetic tape is connected to more than one unit
>and the program does a rewind on one of those
>units and then does a write on another, would
>the rewind happen for both units before the
>write is done?

You've got a file on magtape?

Your OS supports multiple readers for a magtape?  Scary.

>  Would a close operation on one
>of the units, possibly with STATUS='DELETE',
>also cause the other units to become
>disconnected?

No.

The Sun compiler already implements this feature (but maybe not for magtapes), 
and acts as explained above, at least when reading.

Even without this new feature, it's possible for file operations on separate 
files to interfere with one another, so the possibility of interference is not a 
priori a good reason for Fortran to continue to forbid use of a feature provided 
by the OS.

In particular, multiple readers for a single file is a long-solved problem 
implemented by virtually every OS.  Of course not every file can necessarily be 
so shared, but that is adequately covered by "processor dependent".

Cheers,
-- 
................................Malcolm Cohen, Nihon NAG, Tokyo.