Re: Help fixing parser for negative exponents to **

[Steve Kargl <sgk at troutmask dot apl dot washington dot edu>]

On Mon, May 24, 2004 at 01:17:41PM -0600, Roger Sayle wrote:
> 
> Unfortunately, I don't have access to the official grammars for
> fortran95 and fortran2003, but I suspect that there's a bug in
> gfortran's parser for processing power expressions.  Currently,
> we fail to parse unary minus (and unary plus) expressions
> immediately following an exponentiation/power operator.
> 
>        X = 2.0
>        Y = X**-1.5D0
>        PRINT 10, Y
> 10     FORMAT (F8.5)
>        END
> 
>        X = 2.0
>        Y = X**+1.5D0
>        PRINT 10, Y
> 10     FORMAT (F8.5)
>        END
> 
>        X = 2.0
>        Y = 1.5
>        Z = X**-Y
>        PRINT 10, Z
> 10     FORMAT (F8.5)
>        END
> 
> 
> I believe these are all valid, and compile fine with g77.  They might
> however be a "common" extension, so I thought I'd check with this list
> first.  The examples in gfortran.fortran-torture/execute/power.f90 all
> use parenthesis around negative exponents.

These are all illegal with respect to the standard.  From page 136 of
the draft standard, Note 7.31, we have

   As a consequence of the general form (7.1.1), only the first add-operand
   of a level-2-expr may be preceded by the identity (+) or negation (-)
   operator.  These formation rules do not permit expressions containing
   two consecutive numeric operators, such as A ** -B or A + -B.   However,
   expressions such as A ** (-B) and A + (-B) are permitted.  The rules do
   allow a binary operator or an intrinsic unary operator to be followed
   by a defined unary operator, such as:

The above discussion concerns operator precedence in 7.1.1.

You can get a draft of the upcoming standard Fortran 200x at
http://www.j3-fortran.org/.  You want ISO_IEC_1539-1.pdf.  From
my reading of comp.lang.fortran, where many members of the 
standards committee provide advice, ISO_IEC_1539-1.pdf should
be nearly identical to forthcoming ratified standard. 

I also have PDF versions of the F66 and F77 standards, but
I don't remember where I found them on the net.

> Rather than just file a bug report, I'd like some advice on how best
> to fix this problem.  It looks as though the parser in matchexp.c
> is based upon some official/formal description, using terms like
> "level 1" expression and "level 2" expression.

Yep.  See page 117 and following pages in ISO_IEC_1539-1.pdf.

However, you are correct that this is a common extension, and
if g77 supports this construct, then we need to provide for it.
gfortran should be backward compatible with g77.

-- 
Steve