math: MaxFloat32/64 and SmallestNonzeroFloat32/64 value comments misleading

[extemporalgenome]

What version of Go are you using (go version)?

Playground 1.20 as of Fri Jul 7 15:17:35 UTC 2023.

Does this issue reproduce with the latest release?

Yes.

What operating system and processor architecture are you using (go env)?

Playground

What did you do?

https://go.dev/play/p/HZ_Nr7FpU9d

strconv.FormatFloat(math.MaxFloat64, 'e', -1, 64)

What did you expect to see?

1.79769313486231570814527423731704356798070e+308

(since this longer value is listed in the doc comment for math.MaxFloat64)

What did you see instead?

1.7976931348623157e+308

This applies to MaxFloat32, MaxFloat64, SmallestNonzeroFloat32, and SmallestNonzeroFloat64.

Since all of these cases are rational numbers defined in terms of the type itself, I expect their doc comments to have exactly the same value that strconv.FormatFloat would produce with -1 precision.

Note that C# (link) and Java (link) also express the constant in terms of the value that strconv.FormatFloat prints with -1 precision.

[extemporalgenome]

Further request for clarification: float64(math.MaxFloat32) != float64(float32(math.MaxFloat32))

https://go.dev/play/p/feXsT2QSRIU

Why doesn't math.MaxFloat32 exactly represent the maximum finite value that can be stored in a float32? It may happen to produce the maximum-finite bit-pattern when stored in a float32, but is it still reasonable to claim that this is the maximum finite value that a float32 holds when it's not exactly that same finite value?

[randall77]

From strconv.FormatFloat:

The special precision -1 uses the smallest number of digits necessary such that ParseFloat will return f exactly.

So the "actual" you're generating here is not the actual value contained in the float. It is the shortest possible string that parses to that float.

fmt.Println I think also does not print with arbitrary precision. (It prints with "default format", whatever that means.)

If you print with %.0f it will give you the exact value. The resulting value is what is listed in comments for float32, and probably would be listed for float64 only it is too long.

	f32 := float32(math.MaxFloat32)
	fmt.Printf("f32:   %.0f\n", f32)
	fmt.Printf("f32    %x\n", math.Float32bits(f32))
	f64 := float64(math.MaxFloat64)
	fmt.Printf("f64:   %.0f\n", f64)
	fmt.Printf("f64    %x\n", math.Float64bits(f64))

[extemporalgenome]

Ah. I misunderstood. Thanks for the explanation!