[Patch, Fortran] [RFC] PR fortran/36746 / missing part from PR 32095?

[Daniel Kraft]

Tobias Burnus wrote:
> Hi Daniel,
> 
> Daniel Kraft wrote:
>> On the other hand, however, code that gets now
>> rejected is:
>>
>> IMPLICIT TYPE(t) (x)
>> DIMENSION x(:)
>> INTEGER :: x ! 'x' already has a type (TYPE(t))!
>>
>> Is code like that legal?
> 
> I believe so - at least I could not find anything in the standard which
> implies otherwise; additionally, NAG f95, g95, ifort, sunf95, gfortran
> and openf95 compile the program without warning/error.
> It also makes sense since in the dimension statement the type of "x"
> does not matter at all.
> 
> That is quite different from 32095. For specification expressions (cf.
> 7.1.6.2, quoted in PR 32095) this is different: Here, the complete type
> spec might be crutial.
> 
> Thus, the patch is unfortunately NOT OK.

No problem, I already suspected so :)  While I would have liked this 
solution if it was legal, it shouldn't be too hard to find a 
conservative one.  Stay tuned for another try ;)

Thanks for the comments,
Daniel