proposal: os: extend Root FS with interface io/fs.SubFS

[dolmen]

Proposal Details

Extend type os.Root by implementing interface io/fs.SubFS.

The implementation is almost trivial as os.Root already has an OpenRoot method:

func (rfs *rootFS) Sub(dir string) (fs.FS, error) {
    if !isValidRootFSPath(dir) {
        return nil, &PathError{Op: "sub", Path: dir, Err: ErrInvalid}
    }
    r, err := (*Root)(rfs).OpenRoot(dir)
    if err != nil {
        return nil, err
    }
    return r.FS(), nil
}

var _ fs.SubFS = (*rootFS)(nil)

The documentation of os.Root.FS() will be extend to link to fs.SubFS.

[gabyhelp]

Related Issues

• proposal: os: add FS() (filesystem simply using os functions) #47803 (closed)

_{(Emoji vote if this was helpful or unhelpful; more detailed feedback welcome in this discussion.)}

[ianlancetaylor]

CC @neild

[neild]

This seems like an obvious and useful addition to os.Root. I support this proposal.

[dolmen]

I now notice that an os.Root has to be closed with Root.Close.

Root.OpenRoot opens a separate Root whose lifetime must be handled independently from the parent. So a SubFS implemented using Root.OpenRoot would have to be closed, but this is beyond the SubFS contract.

In the current state (without this proposal), those who want a SubFS of an os.Root can use io/fs.Sub. That way the lifetime of the Sub stays attached to its parent, and the responsibility of closing it stays attached to the one who opened the original Root and called its FS method. So everything is fine now.

For this reason I think that this proposal is a bad idea and should be rejected.

[dolmen]

(CL 655796 improves documentation about os.Root.Close call responsibility)

[aclements]

So a SubFS implemented using Root.OpenRoot would have to be closed, but this is beyond the SubFS contract.

Alas, this does sound like a fatal problem.

[aclements]

Based on the discussion above, this proposal seems like a likely decline.
— aclements for the proposal review group