RFC legacy support for array constructor with different string lengths

[Daniel Kraft]

Tobias Burnus wrote:
> Hi,
> 
> when reading 
> http://groups.google.com/group/comp.lang.fortran/browse_thread/thread/75fa01278753793e 
> , I saw that
> 
>  character(len=14), dimension(5), parameter :: var_names &
>    & = ['YEAR', 'COMPCODE', 'RECEPTYPE', 'NCOLS', 'NROWS']
> 
> is now rejected with gfortran 4.4:
> 
> Error: Different CHARACTER lengths (4/8) in array constructor at (1)
> 
> With gfortran 4.2 and 4.3 it is accepted by default (with no warning 
> with -Wall/-pedantic) but rejected with -std=f2003.
> 
> 
> The proper solution is to use Fortran 2003's
> 
>  character(len=14), dimension(5), parameter :: &
>       var_names = [character(len=14) :: &
>       'YEAR', 'COMPCODE', 'RECEPTYPE', 'NCOLS', 'NROWS']
> 
> or for Fortran 90/95: To pad the strings manually to equal length (and 
> to use (/ ... /) instead of [...]).
> 
> * * *
> 
> Question: Some other compilers, e.g. ifort and openf95, allow the first 
> example with default option, with -stand f03 ifort print a warning. Does 
> it make sense to think about supporting it as legacy extension (e.g. 
> combined with printing a default warning)? Or should one simply reject 
> it - as NAG f95, g95 and sunf95 do?

For reference: http://gcc.gnu.org/ml/fortran/2008-05/msg00216.html

I guess this was 'introduced' with my patch there for PR 36112.

And in http://gcc.gnu.org/ml/fortran/2008-05/msg00218.html, FX is 
clearly of the opinion to reject such code.  I think it depends on how 
common code like that is.  If there seems not to be too much of a 
problem with legacy code, I'd keep the behaviour we have now.  If there 
is and you determine to allow such code, I'll implement this into my 
bounds-checking fix.

Cheers,
Daniel


-- 
Done:     Arc-Bar-Cav-Sam-Val-Wiz, Dwa-Elf-Gno-Hum-Orc, Law-Neu-Cha, Fem-Mal
To go:    Hea-Kni-Mon-Pri-Ran-Rog-Tou