Serialization of float minus zero should include the minus sign.

[SimonSapin]

Test case: println!("{}", -0_f64), expected result -0, actual result: 0.

[Diggsey]

While that's clearly desirable for the Show trait, it's not obvious that fmt::String should have that behaviour?

[SimonSapin]

I don’t see why not. +0 and -0 are different values in IEEE 754.

[Diggsey]

Right, but as per the recent RFC, the String trait (being renamed to Display) is intended for user-facing output. For users, negative zero makes no sense, as they're unlikely to even be aware of IEEE 754.

The Show trait (being renamed to Debug) is what's used for debugging, so that should distinguish the two.

[SimonSapin]

shrug

[steveklabnik]

@aturon what do you think?

[SimonSapin]

Even if it's not the default, this behaviour should be available somehow. At the moment I have to do silly things like if value == 0.0 && value.is_negative().

[aturon]

I agree with @Diggsey that doing this in Debug output but not Display output is probably the right way to go (and means that @SimonSapin can get the behavior he wants by writing {:?} instead.)

[rprichard]

println!("{:+}", -0_f64) prints +0, but the {:+?} format is going to print -0. The formatters for scientific notation ('e' and 'E') also have this behavior where they omit the minus sign for negative zero, and emit a plus sign when a sign is explicitly requested, but there's no way to get scientific notation and the correct IEEE-754 sign.

I think I'd prefer that we be consistent and transparent, and print -0 for negative zero even for Display:

• It's consistent with most of the other languages I've tested, including C, Java, Haskell, OCaml, Python, Ruby, and Lua. On the other hand, I did find exceptions -- JavaScript has negative zero but omits the '-'. C# and Go have negative zero and can print +0, though C# hides the '+' by default.
• There is also NaN, which is even less readable to a non-technical audience than -0. A displayed NaN usually indicates a bug in the code, IMO.

Regardless, if the new behavior is desired, then we should test for the edge cases I've described.

Correction: C# does not print +0 for negative zero. Instead, it omits the sign. Its ToString method allows for specifying three formats, one each for positive, negative, and zero.

[rprichard]

I noticed that this test case currently prints -0. I think it will continue doing so with this PR:

println!("{}", -1e-9);

[hanna-kruppe]

@rprichard Yes, my PR does not (well, should not) change behavior for small negative numbers. The existing code is pretty naive w.r.t. rounding (and probably rather inexact at the rounding it does perform), but this is a separate issue.

As for scientific notation, I'm not sure if this is desired: like Display, it's probably more commonly used for user-facing output, right? If it's desired, I could add it pretty easily.