On 7/3/26 08:46, Jerry D wrote:
Hi all,
As I have been looking at PR103367, I was questioning again about the
warnings we give for reals as array indexes and as DO variables.
While I am at it here, should we make those hard errors and only allow
with a warning with -std=legacy. Will we end up breaking a bunch of old
codes by doing this?
There are two issues here. The first one is array indices. The
Fortran standard has never allowed REAL types.
F66, 5.1.3.3:
A subscript expression is written as one the following constructs:
c*v+k, c*v-k, c*v, v+k, v-k, v, k
where c and k are integer constants and v is an integer variable
reference.
F77, 5.4.2 Subscript Expression
A subscription expression is an integer expression.
F90, 6.2.2 Array elements and array sections
...
R617 subscript is scalar-int-expr
R618 section-subscript is subscript
or subscript-triplet
or vector-subscript
R619 subscript-triplet is [ subscript ] : [ subscript ] [ : stride ]
R620 stride is scalar-int-expr
R621 vector-subscript is int-expr
IMHO, gfortran should issue a hard error for a non-integer index.
The second issue is a poor choice from F77.
F66, 7.1.2.8:
DO n i = m1, m2, m3
or
DO n i = m1, m2
...
(2) i is an integer variable, ...
(3) m1 ..., m2 ..., m3 ..., are each either an integer constant or
integer variable reference.
F77, 11.10 DO statement
The form of a DO statment is
DO s [,] i = e1, e2 [, e3]
...
i is the name of an integer, real or double precision variable ...
e1, e2, and e3 are each an integer, real or double precision expression
The F90 standard committee saw the error in the above.
F90, 8.1.4.1.1 Form of the block DO construct
...
R821 loop-control is [ , ] do-variable = scalar-numeric-expr ,
scalar-numeric-expr [ , scalar-numeric-expr ]
or ...
R822 do-variable is scalar-variable
Constraint: The do-variable must be a named scalar variable of
type integer, ***default real, or double precision real.***
Constraint: Each scalar-numeric-expr in loop-control must be of
type integer, ***default real, or double precision real.***
The *** ... *** denotes the Obsolescent lower case font.
B.2 Obsolescent features
...
(2) Real and double precision DO control variables and DO loop
control expressions -- use integer (8.1.4.1)
The committee moved this (mis)feature to the Deleted Features
list in Fortran 95.
So, gfortran should accept REAL DO control variables and
control expressions with -std=legacy. Issuing a warning
would seem to be appropriate even with the -std=legacy
option.
--
steve
