float::round doesn't

[jesse99]

Doesn't work as expected that is. Reasoning by analogy with the int modules I would expect that float::round(5.8) returns 6.0, but instead you get: error: mismatched types: expected core::libc::types::os::arch::c95::c_double but found float.

Looking at the code I expect that more that just round has this problem.

[brson]

This is because c_double is a typedef for f64, and f64 and float are not type compatible (they used to be). Most of the float functions are just reexports of bindings to the C math libraries.

The easiest fix is to write wrapper functions in core::float that cast to the right type (will be a noop).

Beyond that it might be nice to think about how to get rid of the duplication between float/f32/f64.

[catamorphism]

3e9c95e fixes the specific problem with round. @brson, I'm not sure if the other issues you pointed out still apply or not.

[catamorphism]

(sorry, clicked close by mistake)

[alexcrichton]

The specific code in question appears to have been fixed in d9c54f8, although
it turns out that std::float::round == std::f64::round always, so
std::float::round(2.3f) is actually a failure. I think this should be a
separate issue, however (one that probably wants removal of a number of
top-level functions in the numeric modules.

I believe the preferred method is now 3.2f.round(). This won't work for
literals (3.2.round() has multiple methods applicable in scope), but it would
work for any floating type which is inferred somewhere else (to be expected).

I'm going to close this saying that it's been fixed, but if others feel
differently I can definitely reopen.