Re: [PATCH] Fix FMT= with CHARACTER arrays (PR fortran/39865, take 3)

[Walter Spector <w6ws at earthlink dot net>]

Tim Prince wrote:
> ...
> I checked the manual for the f66 compiler I learned on, and nowhere is the
> fmt= specifier mentioned.  Nor does that F77 appendix C mention the
> possibility of using fmt= with Hollerith.  Thus my belief that fmt= was
> not in the standard until f77, when Hollerith had already been eliminated.

I am not sure why the silly comment when I am being sincere in trying
to clarify the situation.  As you well know, FMT= is simply an optional
keyword, yes, introduced at f77, and there should be no difference in how
things work with/without it.  (Outside the obvious positional flexibility
when keywords are used.)

Thinking about it a bit more, one other problem with extending a compiler
to support scalar Hollerith formats is that it would cause an ambiguity
with 'assigned formats':

      integer myfmt
      ...
      assign 100 to myfmt
      write (6,myfmt)
      ...
  100 format (12hHello world!)

Since 'assign' is used with scalar integers, this would be a different
case than supporting array elements as an extension.

IMHO, compilers shouldn't support any extensions in the Hollerith
department (e.g., Hollerith constants in arith expressions) unless
some sort of under-documented compiler option is specified.

Walter