RFR: 8069345: (fs) FileTreeWalker throws NotDirectoryException on file junction points [v2]

[Brian Burkhalter]

On Thu, 17 Oct 2024 06:09:42 GMT, Alan Bateman <alanb at openjdk.org> wrote:

> The JDK supports sym links on Windows and adding support for other unusual junctions types has the potential to introduce inconsistencies and weirdness that are outside the scope of the APIs.

There appear to be two types of junctions for reparse points with tag `IO_REPARSE_TAG_MOUNT_POINT`: directory junctions and volume junctions. The reparse tag is set on a directory which is empty. Both of these reparse points contain a target string which represents a path.

For directory junctions, the path is an absolute path beginning with a drive letter such as

C:\Users\username\directory\subdirectory

This path is intended to be to a directory, but there is nothing preventing storing the path to a file or a symbolic link.

For volume junctions, the path uses a GUID, for example, like this

\?\Volume{blah-foo-bar-gus-blah}\


Presently, the JDK _appears_ already to support "true" directory junctions, i.e., where the target path begins with a drive letter and is an absolute path to a directory. The two new tests proposed in this request verify at least a portion of this behavior. All but two sub-tests of these tests already pass with no source changes.

> [...] this is not a change that we should rush into.

Agreed. Note that the only actual implementation changes being proposed here are

1. `Files.isSymbolicLink` would return `true` for directory junctions.
2. `Files.readSymbolicLink` would return the target of the mount point.

It is incomplete in the current form of this request, but I think tighter checks should be performed to ensure that the target does not represent a volume mount point, nor a regular file, nor a symbolic link.

-------------

PR Comment: https://git.openjdk.org/jdk/pull/21555#issuecomment-2420886803