topic Non-optimal error message for unsupported character kind in Intel® Fortran Compiler

Non-optimal error message for unsupported character kind

Harald — Thu, 13 Sep 2012 21:14:04 GMT

Hello,

the compiler's error message could by more helpful for the following example:

integer, parameter :: iso = SELECTED_CHAR_KIND ("ISO_10646") ! unsupported
character(kind=iso) , parameter :: c = CHAR (32,KIND=iso)
print *, 'selected_char_kind ("ISO_10646") =', iso
end

ifort 13.0 produces:

ifort_character_kinds.f90(2): error #6684: This is an incorrect value for a kind type parameter in this context. [ISO]
character(kind=iso) , parameter :: c = CHAR (32,KIND=iso)
-----------------^
ifort_character_kinds.f90(2): error #6684: This is an incorrect value for a kind type parameter in this context.
character(kind=iso) , parameter :: c = CHAR (32,KIND=iso)
-------------------------------------------------------^
compilation aborted for ifort_character_kinds.f90 (code 1)

Commenting out the offending line 2 and running the program shows that

the encoding is not supported and thus iso==-1. It would be helpful if the error message

would quote the actual invalid value.

Regards,

Harald

Hi Harald

Heinz_B_Intel — Mon, 17 Sep 2012 12:26:28 GMT

Hi Harald yes - agree. This should be reported in a better way. I will submit file a case to improve the message. Heinz

For tracking purpose:

Heinz_B_Intel — Mon, 17 Sep 2012 13:12:30 GMT

For tracking purpose: Submitted as case DPD200236373 I will inform you here as soon as this gets implemented.

Compiler 14.0 now correctly

Heinz_B_Intel — Tue, 24 Sep 2013 10:59:20 GMT

Compiler 14.0 now correctly displays an error message about the unsupported ISO character type:

iso_char.f90(1): error #6104: ISO_10646 character type is not yet supported. ['ISO_10646']

I'm closing this thread.

The compiler shouldn't be

IanH — Tue, 24 Sep 2013 20:22:59 GMT

The compiler shouldn't be issuing an error for the SELECTED_CHAR_KIND intrinsic - its behaviour in the face of an unsupported kind is well defined by the standard - it needs to return -1. An error prevents the compiler from being able to execute a conforming program, such as:

[fortran] INTEGER, PARAMETER :: kind = SELECTED_CHAR_KIND ("ISO_10646")
IF (kind /= 0) THEN
PRINT "('ISO_10646 characters supported.')"
ELSE
PRINT "('ISO_10646 characters not supported.')"
END IF
END[/fortran]

Ian

Heinz_B_Intel — Tue, 24 Sep 2013 21:24:24 GMT

Ian

thanks for the feedback. I will forward to engineering. Our documentation is not clear here either: While it excludes ISO_10646 as input:

(Input) Must be scalar and of type default character. Its value must be 'DEFAULT' or 'ASCII'

it defines the return value as you describe:

The result value is 1 if NAME has the value 'DEFAULT' or 'ASCII'; otherwise, the result value is -1

I agree - we should allow the input and return -1 until it is implemented.

Make it spit out a warning

IanH — Tue, 24 Sep 2013 21:45:50 GMT

Make it spit out a warning instead, then everyone is happy.

Hi Ian

Heinz_B_Intel — Wed, 25 Sep 2013 13:58:37 GMT

Hi Ian

I talked to engineering: There are a few additional issues related to this intrinsics. We will fix them now. I filed a new case DPD200248361 to keep track of this and will add an update here as soon as it got corrected. I requested too to convert the error message to a WARNING

BTW: In your sample code the comparison should be ( == 1) of ( /= -1) but it is clear what you mean :-)

Heinz