(j3.2006) Bug or interp bait?

[Malcolm Cohen]

>Two processors say this is invalid:
>
>  subroutine S ( A )
>    character(len=*), intent(in) :: A
>    integer, parameter :: ESize = ( storage_size(a) + 7 ) / 8

I would hope so too, and would expect it of all compilers that had STORAGE_SIZE.

>One might argue "Sure, this is invalid because a scalar with len=* might
>have a storage_size that is different from an element of an array with
>len=*, because of alignment requirements.  Therefore, the 'properties
>inquired about' (7.1.12(4)(b)) are assumed."

Huh?  STORAGE_SIZE(A)==LEN(A)*STORAGE_SIZE('A'), this is required by normative 
text.  This is rather obviously not constant, but the very epitome of "assumed".

...
>But those processors are also unhappy with
>
>    integer, parameter :: ESize = ( storage_size(a(1:1)) + 7 ) / 8
>
>Was this intended to be invalid?

I certainly did not intend to allow it.  Also it is definitely invalid if 
LEN(A)==0 at runtime, so I would not recommend it anyway.

>If so, how does one get the storage size of a character scalar dummy
>argument?  Must one create a len=1 scalar?

If you want the storage-size of something that does not have a constant storage 
size, you can ask about it in an executable statement.  Not being constant it is 
of course not suitable for a constant expression!

If you want the storage size of a single character, why not just ask for that? 
STORAGE_SIZE('A') for example, is a lot easier to read than all this convoluted 
stuff.

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