Re: [Patch, fortran] PR25018 - ICE in parameter initialization - and PR25029 - assumed size pointer assignments

[Paul Thomas <paulthomas2 at wanadoo dot fr>]

Thank you Steve and Paul,

I will incorporate your comments.

In fact, I want to withdraw the second half of this patch, if that is 
acceptable to you, because I am most of the way to sorting out all the 
restrictions on assumed size arrays. 

As well as PR25029, there are PR20868 and PR20870, together with a 
number of incipient bugs that a read of the standard rought to light.   
With 75% of a complete patch in place, I found that 
gfortran.dg/pr15140.f90 possesses a no-no; it contains an array 
reference to an assumed size array without the last upper bound. Ifort 
and Lahey agree.

function M(NAMES)
 CHARACTER*(*) NAMES(*)
 if (any(names.ne."asdfg")) call abort  ! <<< I will add the required 
bound to fix it.
 m = LEN(NAMES(1))
END function M

character(5) :: c(2)
c = "asdfg"
if(m(c).ne.5) call abort()
end

> I don't think it's necessary to meantion the standard at all, that's implied 
> by the fact we're generating an error. It's hardly likely that the user 
> thinks we're enforcing the C standard :-)
> "Assumed length variable '%s' not permitted in constant expression at $L"
What I was trying to do there was to indicate a difference between a 
warning/error that arises from application of the standard, from those 
that are plain wrong in any circumstances.  Would you prefer  
"Extension: assumed length variable '%s' in constant expression at $L"; 
a style, which is used elsewhere in gfortran?

Cheers

Paul