io/fs: should fs.Sub(fsys, "no such directory") return a file system that fails fstest.TestFS?

[dchapes]

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

$ go version
go version go1.16.2 freebsd/amd64

Does this issue reproduce with the latest release?

If you mean latest released Go, then yes.
If you mean tip, I'm not sure, but I expect so.

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

go env Output

$ go env
[output trimmed]
GO111MODULE=""
GOARCH="amd64"
GOHOSTARCH="amd64"
GOHOSTOS="freebsd"
GOOS="freebsd"
GOVERSION="go1.16.2"
CGO_ENABLED="1"

What did you do?

    fs, err := fs.Sub(someFS, "noSuchDir")
    err = fstest.TestFS(fs)

https://play.golang.org/p/eXsqSUYHWQs

What did you expect to see?

No error returned.

What did you see instead?

It fails due to lack of an empty root directory:

    prog.go:18: TestFS found errors:
        .: Open: open .: file does not exist

This may very well be a "works as intended" issue, and if so I apologise for the noise.

FYI: I ran into this while experimenting with implementing the opposite of fs.Sub, a Prefix(fs.FS, string) (fs.FS, error).
I provided a prefixFS.Sub as an optimisation where it returns another prefixFS with the prefix adjusted or calls fs.Sub on it's underlying file system to avoid chaining multiple wrappers. I ended up also implementing an EmptyFS that I return when the sub-dir isn't within my prefix in order to keep fstest.TestFS happy.

I imagine others implementing fs.SubFS and trying to test their code would appreciate not having to make their own EmptyFS just to satisfy fstest.TestFS. Perhaps an empty file system implementation could be provided or fstest could be tweaked with respect to what it accepts as a completely emtpy file system.

[cherrymui]

cc @rsc
It may be intended. It would be good if someone could confirm.

@dchapes I think you can implement EmptyFS with fstest.MapFS with an empty map.

[dchapes]

@cherrymui

@dchapes I think you can implement EmptyFS with fstest.MapFS with an empty map.

D'oh, you are indeed correct. Thanks for your comment!

I just verified that var fsys fstest.MapFS does indeed work correctly with fstest.TestFS as it implements a "." directory (as does var fsys embed.FS BTW). For some reason I had thought it wouldn't do that without even checking it. Oh well, I was just doing this all as an exercise with how the io/fs interfaces work in practice so my short time doing an EmptyFS wasn't wasted. If an empty file system turns out useful in practice I'd probably prefer one that didn't depend on somthing under testing/… anyway.

[seankhliao]

I think like #44310, we should consider this working as intended.