math/big: nil *Int appears to not roundtrip with MarshalText and UnmarshalText

[mvdan]

https://go.dev/play/p/uV8Vr7trqBv

https://pkg.go.dev/encoding#TextUnmarshaler says:

UnmarshalText must be able to decode the form generated by MarshalText.

MarshalText clearly supports a nil *big.Int value, marshaling it as <nil>, but UnmarshalText rejects it. I think this goes against the TextUnmarshaler docs.

I'm not arguing that marshaling and unmarshaling a nil big int makes any sense, for what it's worth. Just that I always assumed, like the docs say, that roundtripping a value via marshal and then unmarshal was a safe thing to do. If marshaling allows for a nil pointer, unmarshal should support it too - although I'm not sure what it could possibly do, given that it can't set a *big.Int receiver to nil. Perhaps set it to its zero value?

Note that MarshalJSON and UnmarshalJSON don't have this problem. They both support null, and the unmarshal is simply a no-op. Presumably we could do the same in UnmarshalText.

[mknyszek]

CC @griesemer @golang/security via https://dev.golang.org/owners

[Jorropo]

FWIW GobDecode does zero on nil values:

go/src/math/big/intmarsh.go

Lines 33 to 38 in 9552a11

	func (z *Int) GobDecode(buf []byte) error {
	if len(buf) == 0 {
	// Other side sent a nil or default value.
	*z = Int{}
	return nil
	}

[Jorropo]

But not for json where here it does nothing:

go/src/math/big/intmarsh.go

Lines 77 to 83 in 9552a11

	func (z *Int) UnmarshalJSON(text []byte) error {
	// Ignore null, like in the main JSON package.
	if string(text) == "null" {
	return nil
	}
	return z.UnmarshalText(text)
	}

[mvdan]

I would have expected them all to zero out the receiver, although I imagine that a non-nil and zero *Int does not encode or stringify the same way as a nil *Int. Still, at least as far as encoding/json goes, a null generally decodes by zeroing rather than doing nothing.