------- Comment #3 from David dot Duffy at qimr dot edu dot au 2007-04-14 01:03 ------- Subject: Re: Array size declaration depended on order of declaration of variable containing size
On Sat, 13 Apr 2007, tobi at gcc dot gnu dot org wrote: > > > ------- Comment #2 from tobi at gcc dot gnu dot org 2007-04-13 15:22 ------- > (In reply to comment #0) >> GNU Fortran (GCC) 4.3.0 20070412 (experimental) >> Linux 2.4.20-20030701 #2 SMP >> >> use ped_class >> type (ped_data) :: dataset >> integer, dimension(dataset%maxsiz) :: nobs >> >> works but >> >> use ped_class >> integer, dimension(dataset%maxsiz) :: nobs >> type (ped_data) :: dataset >> >> doesn't. > > If I understand you correctly, what you're trying to do is invalid. You may > only reference previously declared objects in data object declarations. In > your second example dataset is referenced before it is declared. > > Please provide a complete testcase, and if the problem is indeed the order of > declarations please tell us where you think I'm wrong. > Hi. Please find a cutdown example attached. In this form, it compiles successfully with gfortran, ifort and g95. If line 251 is commented out and line 256 uncommented, only g95 compiles it successfully. Since dataset, nobs and relid are all subroutine arguments, it seems plausible to me that the order of declaration should be irrelevant, even though it require more work by the compiler. I haven't looked at the Fortran Standard (and probably couldn't work out what it was saying anyway ;)) to see if there is a defined behaviour. Cheers, David Duffy. ------- Comment #4 from David dot Duffy at qimr dot edu dot au 2007-04-14 01:03 ------- Created an attachment (id=13364) --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=13364&action=view) -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31560
