Am 03.07.26 um 18:56 schrieb Steve Kargl:
IMHO, gfortran should issue a hard error for a non-integer index.
I looked at what other compilers are doing, at godbolt, with
standard options.
gfortran accepts with a warning.
flang: rejects
ifort/ifx: accepts silently
nvfortran: accepts with warning
so this is mixed.
I would prefer issuing an error without -std=legacy, and
always issuing a warning with that option. Why?
There are safety-relevant application fields where code changes
need to go through a lengthy and expensive recertification process.
People using such code would then be pushed towards using a compiler
family that accepts this kind of thing, that is ifort/ifx, and that
is not platform-independent (and not gfortran).
But users who are under no such restrictions should be pointed
towards doing the right thing as strongly as possible, without
breaking things for the case above.
For this, so my preference would be the same as yours for
real DO control variables
> 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.
Side remark: I understand why REAL index variables were removed,
because of code like
do a=0,1,0.1
where the number of iterations is indeterminate. However, if written
like
do a=0,1.05,0.1
there is no ambiguity. But that particular ship has sailed.
Best regards
Thomas
