788 – Compiler rejects hex floats in the format: HexPrefix HexDigits . HexDigits(opt) with binary-exponent-part required

Spec: http://www.digitalmars.com/d/lex.html#floatliteral
Program:
void main()
{
 float f1 = 0x0.;
 float f2 = 0x1.5;
 float f3 = 0x.5;
}
Compiler output:
C:\...\bugs>dmd hexfloat.d
hexfloat.d(3): binary-exponent-part required
hexfloat.d(4): binary-exponent-part required
hexfloat.d(5): binary-exponent-part required

There is also another problem with HexFloat:
0x10f
Is this float (0x10)f, or double (0x10f)?
If, binary-exponent-part is really required, then spec is outdated, and there is no problem:
0x10p0f // float
0x10fp0 // double
"Hexadecimal floats are preceded with a 0x and the exponent is a p or P followed by a decimal number serving as the exponent of 2."
so lexer grammar should be:
HexFloat:
 HexPrefix HexDigits . HexDigits HexExponent
 HexPrefix . HexDigits HexExponent
 HexPrefix HexDigits HexExponent

The hex exponent is required. I'll fix the doc.

Fixed DMD 1.005

Commit pushed to master at https://github.com/D-Programming-Language/druntime
https://github.com/D-Programming-Language/druntime/commit/fda62efe0f526e8272dfcc47a4cc7c68f5419569
Ignore real pad in core.internal.convert unittest
Change unittest to ignore any pad bytes in a floating point type because
those bits are undefined. Fixes test part of ldc issue #788.

Commit pushed to stable at https://github.com/D-Programming-Language/druntime
https://github.com/D-Programming-Language/druntime/commit/fda62efe0f526e8272dfcc47a4cc7c68f5419569
Ignore real pad in core.internal.convert unittest