Fix String#to_f out of range

[straight-shoota]

String#to_f generally does not accept values representing infinity or NaN. It would allow them if the first character is whitespace (and whitespace is enabled), them.

String#to_f returns infinity for values outside of the data type's range. That's incorrect, it should be an error.

This patch fixes that and makes sure it fails when strtod/strtof return a non-finite value.

We should probably consider an option for parsing infinity and NaN. But for now this is just a simple bug fix.

[bcardiff]

Can we add specs that NaN INF without whitespaces also return nil?

Parsing those values should also consider localization. I agree to leave that out for now.

[straight-shoota]

Seems I misinterpreted the condition. Infinity and NaN values are always accepted if whitespace: true (the default). So this fix is probably not good. But I really can't make any sense why toggling whitespace should have an effect on infinity and NaN.

[straight-shoota]

The main bug is handling of values outside of range. That should fail instead of returning infinity and should be high priority for correctness.

The minimal fix for that would be to recognize if infinity value returned from strtod/strtof is actually infinity or just represents over- or underflow.

[straight-shoota]

Despite #10367 this only checks for ascii whitespace because strtod/strtof only support ascii whitespace.

[bcardiff]

String#to_f returns infinity for values outside of the data type's range.

You mean this as in before this PR, right? Because

Before this PR: Float64::MAX.to_s.to_f32 # => Infinity
After this PR: Float64::MAX.to_s.to_f32 # => Unhandled exception: Invalid Float32: 1.7976931348623157e+308 (ArgumentError)

[straight-shoota]

You mean this as in before this PR, right?

Yeah, the edit was probably somewhat disruptive to the message. I tried to fix that.