Is `character :: foo(N)*N` obsolescent?

Help
1

We’re just trying to puzzle this out in fortitude.

Looking at the 2023 standard on deleted and obsolescent features, we have this clause:

B.3.8 CHARACTER* form of CHARACTER declaration
In addition to the CHARACTER*char-length form introduced in Fortran 77, Fortran 90 provided the CHARACTER([ LEN = ] type-param-value) form. The older form (CHARACTER*char-length) is redundant.

If we look at the rule for Character type specifier:

R721 char-selector  is length-selector
                    or ( LEN = type-param-value ,
                         KIND = scalar-int-constant-expr )
                    or ( type-param-value ,
                         [ KIND = ] scalar-int-constant-expr )
                    or ( KIND = scalar-int-constant-expr
                         [ , LEN =type-param-value ] )

R722 length-selector is ( [ LEN = ] type-param-value )
                     or * char-length [ , ]
R723 char-length     is ( type-param-value )
                     or int-literal-constant

and:

R701 type-param-value is scalar-int-expr
                      or *
                      or :

so we can have * char-length with char-length as (*) or 5 for instance, and they’re both forms of the obsolete rule. You also can’t tell from the text above, but in the standard * char-length in R722 is in small case, indicating that it’s obsolescent.

Sometimes in old code you might see character foo(N)*N. Is this actually covered under R721?
Looking further through the standard, there’s this bit for entity-decl:

R803 entity-decl  is  object-name [ ( array-spec ) ]
                        [ lbracket coarray-spec rbracket ]
                        [ * char-length ] [ initialization ]
                      or function-name [ * char-length ]

with a comment:

C804 (R803) If the entity is not of type character, * char-length shall not appear.

Here * char-length is not in small case and there’s no reference to R721 or char-selector, and I think a really strict reading of B.3.8 means that it’s not obsolescent?! I guess because this might be the only way to have multiple character variables with different lengths declared on the same line, similar to how (array-spec) in R803 is how you can have different length arrays declared on the same line, but this feels surprising to me.

Looking at how different compilers treat the different forms:

subroutine foo(a, b, c)
    implicit none
    character*8 :: a
    character*(*) :: b
    character c(4)*4
end subroutine foo

Only gfortran and lfortran warn about any of them, and then they both only warn about the first form. I can see how potentially the last form isn’t technically obsolescent (but should be?), but my reading is definitely that the second form is and should be a warning. I could be missing some compiler flags for flang and ifx.

Any language lawyers able to give some insight into what’s going on here? I think the standard could be a bit more explicit either way!

2

No. R721 applies to optional things that can follow the token CHARACTER. In your example, there is no optional char-selector after CHARACTER.

A and B are declared with the obsolescent form, you can literally see the * after CHARACTER. C is declared with a non-obsolescent form, there is no * after CHARACTER.

The NAG compiler warns about obsolescence of A and B declarations, I believe.

1 Like
3

Thanks @themos, that was helpful!

Ok, good to know that B definitely is obsolescent.

I think my issue with C is that the obsolescent clause B.3.8 doesn’t explicitly refer to any rules, so a loose (and in my opinion, not unreasonable) reading could be that it means that it applies to * char-length appearing anywhere in a character declaration. My understanding is that character foo(4)*4 is R803 with object-name (array-spec) * char-length. But you’re saying that B.3.8 only applies to the R721 form and not to R803?

4

image
image1309×186 38.2 KB

R803 does not involve CHARACTER*, so this cannot apply to it. I mean, they do say CHARACTER* three times in three lines. If they meant *char-length they would have written so.
image
image1276×166 27.3 KB

5

Thanks! To be fair, they do write *char-length twice as well! :smiley:

1 Like