*: replace syscall with unix

[vrothberg]

Some parts of the code required some more work than pure substitutions
as the unix interfaces expect a unix.Stat_t, so we couldn't just use
os.FileInfo's Sys() interface to get actual metadata.

Note: this PR bases on and extends https://github.com/openSUSE/umoci/pull/128
Closes #128

[cyphar]

You need to go fmt your code.

[vrothberg]

So, if we use os.RemoveAll(path) here, the CI succeeds. Would this be an alternative here, @cyphar?

[cyphar]

The failure happens because of this change. In particular, err == unix.ENOTDIR will always be false. This is what I was referring to:

// isNotExist tells you if err is an error that implies that either the path
// accessed does not exist (or path components don't exist).
func isNotExist(err error) bool {
	if os.IsNotExist(errors.Cause(err)) {
		return true
	}
	// Check that it's not actually an ENOTDIR.
	perr, ok := errors.Cause(err).(*os.PathError)
	if !ok {
		return false
	}
	errno := perr.Err
	return errno == syscall.ENOTDIR || errno == syscall.ENOENT
}

(And then just swap syscall for unix). However, I was thinking about this some more and maybe we should do this (we keep os.IsNotExist to avoid possible cases where the stdlib changes in some way).

// isNotExist tells you if err is an error that implies that either the path
// accessed does not exist (or path components don't exist).
func isNotExist(err error) bool {
	if os.IsNotExist(errors.Cause(err)) {
		return true
	}

	// Check that it's not actually an ENOTDIR in disguise. This is technically
	// a dup of doing os.IsNotExist, but it doesn't hurt to be thorough.
	var errno error
	switch err := errors.Cause(err).(type) {
	case *os.PathError:
		errno = err.Err
	case *os.LinkError:
		errno = err.Err
	case *os.SyscallError:
		errno = err.Err
	}
	return errno == syscall.ENOTDIR || errno == syscall.ENOENT
}

Also, make sure you don't remove errors.Causes. It's a no-op for non-wrapped error messages, but for wrapper error messages it makes sure that things like os.IsNotExist actually do the right thing.

[vrothberg]

After some CI hiccups, it has finally passed \o/

[cyphar]

Yeah, the CI hiccups were due to openSUSE repos being broken yesterday. I'll go through a review later today.

[vrothberg]

I rebased and also fixed a copy-paste error in utime_linux_test.go that caused random fails since the the nano seconds of atime has been used for the mtime timestamp.

[cyphar]

LGTM