Invalid test cases for runtime.Marshaler

[movsb]

🐛 Bug Report

Taking address of any element in this slice produces the same value (because the element is an empty struct):

grpc-gateway/runtime/marshaler_registry_test.go

Lines 30 to 52 in 855204b

	var marshalers [3]dummyMarshaler
	specs := []struct {
	opt runtime.ServeMuxOption
	wantIn runtime.Marshaler
	wantOut runtime.Marshaler
	}{
	{
	opt: runtime.WithMarshalerOption(runtime.MIMEWildcard, &marshalers[0]),
	wantIn: &marshalers[0],
	wantOut: &marshalers[0],
	},
	{
	opt: runtime.WithMarshalerOption("application/x-in", &marshalers[1]),
	wantIn: &marshalers[1],
	wantOut: &marshalers[0],
	},
	{
	opt: runtime.WithMarshalerOption("application/x-out", &marshalers[2]),
	wantIn: &marshalers[1],
	wantOut: &marshalers[2],
	},
	}

So, these test cases almost won't work as expected.

As per the last line of golang spec:

A struct or array type has size zero if it contains no fields (or elements, respectively) that have a size greater than zero.

Two distinct zero-size variables may have the same address in memory.

To Reproduce

See the code at playground:

package main

import (
	"fmt"
)

type S struct {
}

type T struct {
	b bool
}

func main() {
	var s [3]S
	var t [3]T

	fmt.Printf("%p,%p,%p\n", &s[0], &s[1], &s[2])
	fmt.Printf("%p,%p,%p\n", &t[0], &t[1], &t[2])
}

Outputs:

0x58fd18,0x58fd18,0x58fd18
0xc00009400b,0xc00009400c,0xc00009400d

Your Environment

Any.

[johanbrandhorst]

Hi, thanks for the bug report! Do you have any ideas for how we could fix this? I think this is a pretty old part of the code.

[yugui]

Thank you for reporting this bug, @movsb. That was my fault.
Fortunately the actual implementation of marshaler registry was implemented as expected.

Do you have any ideas for how we could fix this? I think this is a pretty old part of the code.

Giving a non-zero size to the type could solve the issue. (#1502).
I found that the test spec I wrote was completely wrong and that the zero-length type has been hiding it. So I also corrected the spec.

[movsb]

Hi, @yugui , thanks for the correction.

Fortunately the actual implementation of marshaler registry was implemented as expected.

Unfortunately, I found this bug when I was trying to open another issue ( #1505 ) 😅