[Win64] real constants parsed as 0

[Trass3r]

Lexer::inreal calls Port::strtold to parse the real constants in std.math.

enum real PI =         0x1.921fb54442d18469898cc51701b84p+1L;

This maps to the msvc strtold and just returns 0, also see #747.

[Trass3r]

Looks like hexadecimal strings are only supported in VS14:
http://blogs.msdn.com/b/vcblog/archive/2014/06/18/crt-features-fixes-and-breaking-changes-in-visual-studio-14-ctp1.aspx

[dnadlinger]

Didn't somebody pull the related code out of the MinGW CRT so we could use it? (@kinke maybe?)

[kinke]

I didn't. We've previously used DMD/DMC's strtold() implementation back when real was 80 bits; we'd now actually need a C99-compliant strtod() for real == double. Imho, the best course of action is to continue using MSVC's strtold() but to require MSVC 2014+ as soon as it's released. It simply represents a big leap in terms of portability with its C99 conformance, as the linked MSDN blog shows.
In this case, MS also increased the accuracy when parsing floating points: The old parsing algorithms would consider only up to 17 significant digits from the input string and would discard the rest of the digits. [...] The new implementation considers all present digits and produces the correctly rounded result for all inputs (up to 768 digits in length).

[Trass3r]

Even then we need to add a warning at least. It can't just silently use 0.0
How does dmd deal with invalid float literals?

[dnadlinger]

How would invalid float literals be relevant here?

[Trass3r]

Just wondering if they do error checking on the strtold call or inside.

[dnadlinger]

Invalid literals shouldn't even make it to the strtold call in the lexer.

[Trass3r]

Makes sense.

[redstar]

The compiler part is hopefully fixed with commit 2fdd4fe.

[kinke]

FWIW, although we only support VS 2015+ anymore, this is fixed by #2054.