encoding/json: clarify what happens when multiple items match the same field

[robpike]

This test I added to encoding/json

func TestUnmarshalCase(t *testing.T) {
	var v struct {
		Dist string `json:"dist"`
	}

	Unmarshal([]byte(`{"dist": "ok", "Dist": "KO"}`), &v)
	t.Logf("%+v", v)

	Unmarshal([]byte(`{"Dist": "ok", "dist": "KO"}`), &v)
	t.Logf("%+v", v)
}

prints

    x_test.go:13: {Dist:KO}
    x_test.go:16: {Dist:KO}

The comment says Unmarshal "prefers" exact matches, but if there are no words about what happens if multiple fields match.

It's clearly an odd case, but clarity would be nice. I expected the exact match to win always.

[cagedmantis]

/cc @rsc @mvdan

[mvdan]

I'm well aware of this problem, unfortunately. See #14750, which this issue is likely a duplicate of. My attempt at fixing this without incurring a performance or memory cost is https://go-review.googlesource.com/c/go/+/224079, which has been stuck in review limbo for a little while :)

[dsnet]

At the core of this problem is whether we want to favor a streaming decode or whether we want to buffer all the members of a JSON object in order to perform an exact match. The former has significantly better performance, while the latter probably matches user expectation better.

[cristaloleg]

Kindly ping on this. Looks like this will not be available in 1.17 too already.

@mvdan looks like PR is ready, there is 1 question is 1% perf acceptable, am I right?

[mvdan]

The PR isn't ready - there are conflicts, and some edge cases that Joe brought up require extra work.