As an exercise in D programming, I'm writing a D lexer from scratch, based on the online specs posted on DPLO. I'm running into what looks like a discrepancy between the specs and compiler behaviour (I'm using gdc-4.6.2):

Spec says:
	DecimalFloat:
	    LeadingDecimal .
	    LeadingDecimal . DecimalDigits
	    DecimalDigits . DecimalDigitsNoSingleUS DecimalExponent
	    . DecimalInteger
	    . DecimalInteger DecimalExponent
	    LeadingDecimal DecimalExponent

Based on the 3rd condition above, things like "123._e12" should be illegal (you can't have only a '_' between '.' and 'e').  However, the compiler accepts "123._e12".

Furthermore, the spec says:

	DecimalExponent
	    DecimalExponentStart DecimalDigitsNoSingleUS

This means "123.e_2" should be accepted, but the compiler says:

	test.d:6: exponent expected

So the question is: am I reading the specs wrong, or is this a compiler bug, or a spec bug?  What *should* the floating literal syntax be?

Does it make sense to treat '_' essentially as a null string when seen inside a floating literal? If so, "123._e12" should be accepted, and so should "123.e_2", in which case *both* the spec and the compiler are wrong.

Also, are multiple consecutive underscores permitted? I.e., is "1__000__000.0___0" legal?


T

-- 
First Rule of History: History doesn't repeat itself -- historians merely repeat each other.

On 02/11/2012 01:51 AM, H. S. Teoh wrote:
> As an exercise in D programming, I'm writing a D lexer from scratch,
> based on the online specs posted on DPLO. I'm running into what looks
> like a discrepancy between the specs and compiler behaviour (I'm using
> gdc-4.6.2):
>
> Spec says:
> 	DecimalFloat:
> 	    LeadingDecimal .
> 	    LeadingDecimal . DecimalDigits
> 	    DecimalDigits . DecimalDigitsNoSingleUS DecimalExponent
> 	    . DecimalInteger
> 	    . DecimalInteger DecimalExponent
> 	    LeadingDecimal DecimalExponent
>
> Based on the 3rd condition above, things like "123._e12" should be
> illegal (you can't have only a '_' between '.' and 'e').  However, the
> compiler accepts "123._e12".
>
> Furthermore, the spec says:
>
> 	DecimalExponent
> 	    DecimalExponentStart DecimalDigitsNoSingleUS
>
> This means "123.e_2" should be accepted, but the compiler says:
>
> 	test.d:6: exponent expected
>
> So the question is: am I reading the specs wrong, or is this a compiler
> bug, or a spec bug?  What *should* the floating literal syntax be?
>

Spec bug. What you are observing is in accordance with TDPL. (However, I think it would be better to reject the first example too.)

> Does it make sense to treat '_' essentially as a null string when seen
> inside a floating literal? If so, "123._e12" should be accepted, and so
> should "123.e_2", in which case *both* the spec and the compiler are
> wrong.
>
> Also, are multiple consecutive underscores permitted? I.e., is
> "1__000__000.0___0" legal?
>

Yes.