0x1.max // exponent expected in hex float
0x1 .max // OK
1.max // OK

What's the ambiguity when it's an hex literal ?

On Tuesday, 5 April 2016 at 19:00:43 UTC, Basile B. wrote:
> 0x1.max // exponent expected in hex float
> 0x1 .max // OK
> 1.max // OK
>
> What's the ambiguity when it's an hex literal ?

https://issues.dlang.org/show_bug.cgi?id=15880

On Tuesday, 5 April 2016 at 19:00:43 UTC, Basile B. wrote:
> 0x1.max // exponent expected in hex float
> 0x1 .max // OK
> 1.max // OK
>
> What's the ambiguity when it's an hex literal ?

It's potentially ambiguous with hexadecimal floating point numbers

0xdeadbeef.p5 // hex float or hex int + method?

dlang.org/spec/lex.html#HexFloat

On Tuesday, 5 April 2016 at 20:56:54 UTC, Alex Parrill wrote:
> On Tuesday, 5 April 2016 at 19:00:43 UTC, Basile B. wrote:
>> 0x1.max // exponent expected in hex float
>> 0x1 .max // OK
>> 1.max // OK
>>
>> What's the ambiguity when it's an hex literal ?
>
> It's potentially ambiguous with hexadecimal floating point numbers
>
> 0xdeadbeef.p5 // hex float or hex int + method?
>
> dlang.org/spec/lex.html#HexFloat

Yes but it's pointless to allow the decimal separator to be followed by the exponent:

void main()
{
    import std.stdio;
    writeln( typeof(0x1p5).stringof ); // double
    writeln( typeof(0x1.p5).stringof ); // double
}

On Tuesday, 5 April 2016 at 21:10:47 UTC, Basile B. wrote:
> On Tuesday, 5 April 2016 at 20:56:54 UTC, Alex Parrill wrote:
>> On Tuesday, 5 April 2016 at 19:00:43 UTC, Basile B. wrote:
>>> 0x1.max // exponent expected in hex float
>>> 0x1 .max // OK
>>> 1.max // OK
>>>
>>> What's the ambiguity when it's an hex literal ?
>>
>> It's potentially ambiguous with hexadecimal floating point numbers
>>
>> 0xdeadbeef.p5 // hex float or hex int + method?
>>
>> dlang.org/spec/lex.html#HexFloat
>
> Yes but it's pointless to allow the decimal separator to be followed by the exponent:
>
> void main()
> {
>     import std.stdio;
>     writeln( typeof(0x1p5).stringof ); // double
>     writeln( typeof(0x1.p5).stringof ); // double
> }

I mean that the rule could be: the decimal separator must be followed by a second group of digits. The second group of digits must be followed by an exponent. The first group of digits can be followed by an exponent.

0x1.0p5 // valid
0xp5 // valid
0x1.p5 // invalid (p is not a hex digit)
0x1.ap5 // valid

On Tuesday, 5 April 2016 at 21:40:59 UTC, Basile B. wrote:
> On Tuesday, 5 April 2016 at 21:10:47 UTC, Basile B. wrote:
>> On Tuesday, 5 April 2016 at 20:56:54 UTC, Alex Parrill wrote:
>>> On Tuesday, 5 April 2016 at 19:00:43 UTC, Basile B. wrote:
>>>> 0x1.max // exponent expected in hex float
>>>> 0x1 .max // OK
>>>> 1.max // OK
>>>>
>>>> What's the ambiguity when it's an hex literal ?
>>>
>>> It's potentially ambiguous with hexadecimal floating point numbers
>>>
>>> 0xdeadbeef.p5 // hex float or hex int + method?
>>>
>>> dlang.org/spec/lex.html#HexFloat
>>
>> Yes but it's pointless to allow the decimal separator to be followed by the exponent:
>>
>> void main()
>> {
>>     import std.stdio;
>>     writeln( typeof(0x1p5).stringof ); // double
>>     writeln( typeof(0x1.p5).stringof ); // double
>> }
>
> I mean that the rule could be: the decimal separator must be followed by a second group of digits. The second group of digits must be followed by an exponent. The first group of digits can be followed by an exponent.
>
> 0x1.0p5 // valid
> 0xp5 // valid
> 0x1.p5 // invalid (p is not a hex digit)
> 0x1.ap5 // valid

Looks like that's how it works for decimal floats; I.e. 1.e5 is an int and property lookup, while 1.0e5 is 100000f. Curiously, 1. Is 1.0.

I agree that floats should be parsed consistently. For now, you can do (0x1).max or typeof(0x1).max.