Float parsing should be capable to parse min and max values

[hirschenberger]

The from_str function for float values should be capable to parse it's MIN_VALUE and MAX_VALUE to a valid float number.

fn main() {
  println!("{}", from_str::<f32>("-3.40282346e+38"));
  println!("{}", from_str::<f32>("-340282368002860660002286082464244022240"));
  println!("{}", -3.40282346e+38f32);
  println!("{}", std::f32::MIN_VALUE.to_str());
}

Output:

Some(-inf)
Some(-inf)
-340282368002860660002286082464244022240
-340282368002860660002286082464244022240

Blocking #10934

[hirschenberger]

Works in C++11

#include <limits>
#include <iostream>
#include <string>

int main()
{
  std::cout << std::stof(std::to_string(std::numeric_limits<float>::max())) << std::endl;
}

[hirschenberger]

This is related to issue #7588 where an extra check was added if the num's range was overflowing. Unfortunately it was realized by undoing the last multiplication by a division and checking the values with !=. This fails for floatingpoint types due to precision issues (a epsilon test would be required) and is not necessary because floats don't overflow.

https://github.com/mozilla/rust/blob/master/src/libstd/num/strconv.rs#L642

I have a solution by passing another parameter to from_str_bytes_common named may_overflow which indicates if the target num can overflow or not and skipping the test for floats.

Perhaps a more elegant solution would be the Introduction of a new Overflow trait for all nums.

Or just implement CheckedAdd and CheckedSub for floats to avoid the whole undoing.

[steveklabnik]

Triage: while a related lint got merged, I'm not seeing it when I compile the sample:

fn main() {
      println!("{:?}", "-3.40282346e+38".parse::<f32>());
        println!("{:?}", "-340282368002860660002286082464244022240".parse::<f32>());
          println!("{}", -3.40282346e+38f32);
}