proposal: errors: add AsOf to the API

[phenpessoa]

With the introduction of generics, I think it makes sense to have an AsOf function in the errors package that will return the wanted error.
The API would look like this:

func AsOf[E error](err error) (E, bool)

A simple implementation of this func could be

func AsOf[E error](err error) (E, bool) {
	e := new(E)
	ok := As(err, e)
	return *e, ok
}

But, with generics, we can have this be implemented without reflection and with an allocation free path

func AsOf[E error](err error) (E, bool) {
	var ptrToE *E
	for err != nil {
		if e, ok := err.(E); ok {
			return e, true
		}
		if x, ok := err.(interface{ As(any) bool }); ok {
			if ptrToE == nil {
				ptrToE = new(E)
			}
			if x.As(ptrToE) {
				return *ptrToE, true
			}
		}
		err = Unwrap(err)
	}
	var zero E
	return zero, false
}

cc: @Jorropo who actually wrote the second implementation of AsOf

---

Benchmarks

type MyError struct{}

func (me *MyError) Error() string {
	return "my error"
}

var sink *MyError

func AsOf[E error](err error) (E, bool) {
	e := new(E)
	ok := errors.As(err, e)
	return *e, ok
}

func BenchmarkAsOf(b *testing.B) {
	var err error = &MyError{}

	b.ReportAllocs()
	for i := 0; i < b.N; i++ {
		sink, _ = AsOf[*MyError](err)
	}
}

func AsOf2[E error](err error) (E, bool) {
	var ptrToE *E
	for err != nil {
		if e, ok := err.(E); ok {
			return e, true
		}
		if x, ok := err.(interface{ As(any) bool }); ok {
			if ptrToE == nil {
				ptrToE = new(E)
			}
			if x.As(ptrToE) {
				return *ptrToE, true
			}
		}
		err = errors.Unwrap(err)
	}
	var zero E
	return zero, false
}

func BenchmarkAsOf2(b *testing.B) {
	var err error = &MyError{}

	b.ReportAllocs()
	for i := 0; i < b.N; i++ {
		sink, _ = AsOf2[*MyError](err)
	}
}

Results:

First implementation:
goos: windows
goarch: amd64
pkg: test
cpu: Intel(R) Core(TM) i9-9900K CPU @ 3.60GHz
BenchmarkAsOf-16    	16641680	        70.14 ns/op	       8 B/op	       1 allocs/op
BenchmarkAsOf-16    	16340714	        70.77 ns/op	       8 B/op	       1 allocs/op
BenchmarkAsOf-16    	16931168	        71.64 ns/op	       8 B/op	       1 allocs/op
BenchmarkAsOf-16    	16693258	        71.01 ns/op	       8 B/op	       1 allocs/op
BenchmarkAsOf-16    	16718749	        70.24 ns/op	       8 B/op	       1 allocs/op
PASS
ok  	test	6.458s



Second implementation:
goos: windows
goarch: amd64
pkg: test
cpu: Intel(R) Core(TM) i9-9900K CPU @ 3.60GHz
BenchmarkAsOf2-16    	604845721	         1.940 ns/op	       0 B/op	       0 allocs/op
BenchmarkAsOf2-16    	618726052	         1.925 ns/op	       0 B/op	       0 allocs/op
BenchmarkAsOf2-16    	621167846	         1.928 ns/op	       0 B/op	       0 allocs/op
BenchmarkAsOf2-16    	608969202	         1.962 ns/op	       0 B/op	       0 allocs/op
BenchmarkAsOf2-16    	617731360	         1.905 ns/op	       0 B/op	       0 allocs/op
PASS
ok  	test	7.138s

[Jorropo]

I think it's important to point out that errors.As can be quite confusing to use, it has many ways it can panic at runtime all related to feeding a wrong type in the any argument: https://cs.opensource.google/go/go/+/refs/tags/go1.19.3:src/errors/wrap.go;l=79-90
This solution instead replace it with fully compile-time safe solution, by using the type checker and generic constraints.

[fzipp]

The proposed name is strange. "as of" is a temporal preposition as in "as of January 2022".

[phenpessoa]

@fzipp im open to suggestions for the name, as I’m not a fan of it either.

But the idea of using the Of suffix came from here: #48287

[fzipp]

Yes, in my opinion, before adding generic variants of existing functions and types to the standard library, a general approach to this topic (#48287) should be decided first.

[seankhliao]

Duplicate of #51945

[phenpessoa]

@seankhliao the proposed implementation is not the same
Should we suggest this implementation on that issue?

[Jorropo]

@pedro-pessoa proposals are for API changes and some observable behaviour side effects.
My implementation and #51945 ones have identical API and behaviour this is then not concerned by the proposal process. (if #51945 is accepted I would be allowed to submit a CL with my implementation)