(j3.2006) a question on interp F08/0049

[Malcolm Cohen]

Aleks Donev wrote:
>On 11/15/10 03:26, Malcolm Cohen wrote:
>> Don't confuse this with the COELEMENTAL attribute!  Of course coelemental 
>> functions cannot have variable type parameters at this time.
>If Malcolm meant this as a joke

Yes.  An obviously horrible idea.

Jim Xia wrote:
> the original wording in F08 seems to prevent this on purpose.

That's not an accurate characterisation of what happened here.  The so-called 
original wording was not "originally" in F2008 but was put there by the flawed 
answer to interp F03/0119, see F08/0049 which begins

  "Interpretation F03/0119, which became part of the Fortran 2008 standard, ..."

followed by an explanation of why F03/0119 is flawed.

> I'd not be surprised to see a further interp trying to reverse this interp 
> answer again.

F08/0049 does not "reverse" F03/0119!

The edit in F08/0049 does effectively replace the edit in F03/0119, but please 
note that the problematic example in F03/0119, viz example3, which F03/0119 says 
"was intended to be prohibited", remains prohibited by F08/0049.  (And example1 
remains conforming, and example2 was already permitted in Fortran 2008 and 
remains permitted by F08/0049.)

You can call it a "partial reversal" if you like - but to the extent that it is 
a reversal, it is a reversal of a case that was not considered by F03/0119, i.e. 
an unfortunate consequence of the edits but without any justification in 
F03/0119 for that effect.  The justification in F03/0119 is that we did not 
intend to allow "elemental functions to produce arrays whose elements have 
different length type parameters", and F08/0049 agrees with that justification.

You can certainly doubt that we yet have the wording right, since we got it 
wrong in F03/0119, but we absolutely did not intend to introduce an 
incompatibility with Fortran 95.

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