Re: Is the included code standard conforming?

[Brooks Moses <bmoses at stanford dot edu>]

Dominique Dhumieres wrote:
> From the first sentence of
>
> 13.1 Classes of intrinsic procedures
> ...
> If an intrinsic function is used as an actual argument to a procedure, its
> specific name shall be used and it may be referenced in the called
> procedure only with scalar arguments.  If an intrinsic function does not
> have a specific name, it shall not be used as an actual argument
> (12.4.1.3). ...
>
> can I infer that the following code is not standard conforming?
[...]
> program specifics
>
>   intrinsic log
>   intrinsic log10
>
>   call test_r (log, 2.0, log(2.0))
>   call test_r (log10, 2.0, log10(2.0))
>   call test_d (log, 2.0d0, log(2.0d0))
>   call test_d (log10, 2.0d0, log10(2.0d0))
>
> end program

I would agree that this is not standard-conforming, yes.

> Should this be detected with -std=f95? (yes I now understand that the 
> "shall" is for the user and not for the compiler).

I believe that it should be detected and an error thrown, regardless of 
option.

Unless I am quite mistaken, it is essentially impossible for a compiler 
to generate correct code in this case, at least if the program and 
subroutines are compiled separately.  The main program does not have an 
explicit interface for test_r and test_d, and so the compiler has no way 
of resolving the generic to a specific prior to passing it.  However, 
when test_r and test_d are compiled, the compiler has no way of knowing 
whether or not the passed function needs to be resolved from a generic 
to a specific (nor, most likely, any mechanism for resolving it).

- Brooks