Thomas Wolff <towo <at> towo.net> writes:
> Sorry that I take this up once more (after promising <end:of>), but I
> had this additional idea after seeing your point about being strictly
> consistent with the POSIX pathname namespace:
>
> So what about using "/" as a delimiter? If "foo" is a file, "foo/bar" is
> not a legal pathname in POSIX, so it could be used to access the "bar"
> fork of "foo" without causing real harm.
NO - a thousand times no. Using / in file names, but not as a directory, is
just ASKING to break everything ever written, and penalize speed of interfaces
that could care less about this.
But, you _could_ borrow a leaf from Solaris, and support and implementation:
openat(open("foo",flags), "bar", flags)
as a way to open the "bar" stream of the "foo" fd, aka "foo:bar" in windows
terms. In other words, open("foo/bar") MUST fail, because foo is not a
directory, but openat(fd_of_foo,"bar") is an extension allowed by POSIX (just
because we currently fail with ENOTDIR in that situation doesn't mean we have
to); and by using the *at interfaces, we could isolate the performance penalty
to just the situations where the fd is not a directory fd.
You would also want to consider implementing opendir2 (borrowing from BSD
heritage; there, opendir2 exists to allow the user to select whether whiteout
entries in a union mount will be ignored), and adding a new DTF_* bit that
allows opening a file to traverse its alternate streams, instead of the normal
opening a directory to traverse its contents.
But I won't be writing the patches to do that.
--
Eric Blake
--
Problem reports: http://cygwin.com/problems.html
FAQ: http://cygwin.com/faq/
Documentation: http://cygwin.com/docs.html
Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple
