buildstream/storage/_casbaseddirectory.py

@@ -38,6 +38,8 @@ from .._exceptions import BstError
 from .directory import Directory, VirtualDirectoryError
 from ._filebaseddirectory import FileBasedDirectory
 from ..utils import FileListResult, safe_copy, list_relative_paths
+from ..utils import FileListResult, safe_copy, list_relative_paths, _relative_symlink_target
+from .._artifactcache.cascache import CASCache
 
 
 class IndexEntry():
@@ -51,6 +53,20 @@ class IndexEntry():
         self.modified = modified
 
 
+class ResolutionException(Exception):
+    """ Superclass of all exceptions that can be raised by CasBasedDirectory._resolve. Should not be exposed externally."""
+    pass
+
+class InfiniteSymlinkException(ResolutionException):
+    """ Raised when an infinite symlink loop is found. """
+    pass
+
+class AbsoluteSymlinkException(ResolutionException):
+    """Raised if we try to follow an absolute symlink (i.e. one whose
+    target starts with the path separator) and we have disallowed
+    following such symlinks. """
+    pass
+
 # CasBasedDirectory intentionally doesn't call its superclass constuctor,
 # which is meant to be unimplemented.
 # pylint: disable=super-init-not-called
@@ -176,21 +192,26 @@ class CasBasedDirectory(Directory):
         filenode.is_executable = is_executable
         self.index[filename] = IndexEntry(filenode, modified=(filename in self.index))
 
-    def _add_new_link(self, basename, filename):
-        existing_link = self._find_pb2_entry(filename)
+    def _copy_link_from_filesystem(self, basename, filename):
+        self._add_new_link_direct(filename, os.readlink(os.path.join(basename, filename)))
+
+    def _add_new_link_direct(self, name, target):
+        existing_link = self._find_pb2_entry(name)
         if existing_link:
             symlinknode = existing_link
         else:
             symlinknode = self.pb2_directory.symlinks.add()
-        symlinknode.name = filename
+        assert(isinstance(symlinknode, remote_execution_pb2.SymlinkNode))
+        symlinknode.name = name
         # A symlink node has no digest.
-        symlinknode.target = os.readlink(os.path.join(basename, filename))
-        self.index[filename] = IndexEntry(symlinknode, modified=(existing_link is not None))
+        symlinknode.target = target
+        self.index[name] = IndexEntry(symlinknode, modified=(existing_link is not None))
 
     def delete_entry(self, name):
         for collection in [self.pb2_directory.files, self.pb2_directory.symlinks, self.pb2_directory.directories]:
-            if name in collection:
-                collection.remove(name)
+            for thing in collection:
+                if thing.name == name:
+                    collection.remove(thing)
         if name in self.index:
             del self.index[name]
 
@@ -231,9 +252,13 @@ class CasBasedDirectory(Directory):
             if isinstance(entry, CasBasedDirectory):
                 return entry.descend(subdirectory_spec[1:], create)
             else:
+                # May be a symlink
+                target = self._resolve(subdirectory_spec[0], force_create=create)
+                if isinstance(target, CasBasedDirectory):
+                    return target
                 error = "Cannot descend into {}, which is a '{}' in the directory {}"
                 raise VirtualDirectoryError(error.format(subdirectory_spec[0],
-                                                         type(entry).__name__,
+                                                         type(self.index[subdirectory_spec[0]].pb_object).__name__,
                                                          self))
         else:
             if create:
@@ -254,7 +279,7 @@ class CasBasedDirectory(Directory):
         else:
             return self
 
-    def _resolve_symlink_or_directory(self, name):
+    def _force_resolve(self, name):
         """Used only by _import_files_from_directory. Tries to resolve a
         directory name or symlink name. 'name' must be an entry in this
         directory. It must be a single symlink or directory name, not a path
@@ -267,23 +292,128 @@ class CasBasedDirectory(Directory):
         as a directory as long as it's within this directory tree.
         """
 
+        return self._resolve(name, force_create=True)
+
+    def _is_followable(self, name):
+        """ Returns true if this is a directory or symlink to a valid directory. """
+        if name not in self.index:
+            return False
         if isinstance(self.index[name].buildstream_object, Directory):
-            return self.index[name].buildstream_object
-        # OK then, it's a symlink
-        symlink = self._find_pb2_entry(name)
+            return True
+        try:
+            target = self._resolve(name)
+        except InfiniteSymlinkException:
+            return False
+        return isinstance(target, CasBasedDirectory) or target is None
+        # TODO: But why return True if it's None (broken link/circular loop)? Surely that is against the docstring.
+
+    def _resolve(self, name, absolute_symlinks_resolve=True, force_create=False, seen_objects=None):
+        """Resolves any name to an object. If the name points to a symlink in
+        this directory, it returns the thing it points to,
+        recursively.
+
+        Returns a CasBasedDirectory, FileNode or None. None indicates
+        either that 'none' does not exist in this directory, or is a
+        symlink chain which points to a nonexistent name (broken
+        symlink).
+
+
+        Raises:
+        - InfiniteSymlinkException if 'name' points to an infinite symlink loop.
+        - AbsoluteSymlinkException if 'name' points to an absolute symlink and absolute_symlinks_resolve is False.
+
+        If force_create is on, this will attempt to create directories to make symlinks and directories resolve.
+        If force_create is off, this will never alter this directory.
+
+        """
+
+        if name not in self.index:
+            return None
+
+        # First check if it's a normal object and return that
+        index_entry = self.index[name]
+        if isinstance(index_entry.buildstream_object, Directory):
+            return index_entry.buildstream_object
+        elif isinstance(index_entry.pb_object, remote_execution_pb2.FileNode):
+            return index_entry.pb_object
+
+        assert isinstance(index_entry.pb_object, remote_execution_pb2.SymlinkNode)
+
+        if seen_objects is None:
+            seen_objects = [index_entry.pb_object]
+        else:
+            if index_entry.pb_object in seen_objects:
+                # Infinite symlink loop detected
+                raise InfiniteSymlinkException()
+
+        symlink = index_entry.pb_object
+        components = symlink.target.split(CasBasedDirectory._pb2_path_sep)
+
         absolute = symlink.target.startswith(CasBasedDirectory._pb2_absolute_path_prefix)
         if absolute:
-            root = self.find_root()
+            if absolute_symlinks_resolve:
+                start_directory = self.find_root()
+                # Discard the first empty element
+                components.pop(0)
+            else:
+                # Unresolvable absolute symlink
+                raise AbsoluteSymlinkException()
         else:
-            root = self
-        directory = root
-        components = symlink.target.split(CasBasedDirectory._pb2_path_sep)
-        for c in components:
-            if c == "..":
-                directory = directory.parent
+            start_directory = self
+
+        directory = start_directory
+        while True:
+            if not components:
+                # We ran out of path elements and ended up in a directory
+                return directory
+            c = components.pop(0)
+            if c == ".":
+                pass
+            elif c == "..":
+                if directory.parent is not None:
+                    directory = directory.parent
+                # If directory.parent *is* None, this is an attempt to access
+                # '..' from the root, which is valid under POSIX; it just
+                # returns the root.
+            elif c in directory.index:
+                # Recursive resolve and continue
+                try:
+                    f = directory._resolve(c, absolute_symlinks_resolve, seen_objects=seen_objects)
+                except ResolutionException:
+                    f = None
+                if isinstance(f, CasBasedDirectory):
+                    directory = f
+                elif isinstance(f, remote_execution_pb2.FileNode):
+                    if components:
+                        # We have components still to resolve, but one of the path components
+                        # is a file.
+                        if force_create:
+                            self.delete_entry(c)
+                            directory = directory.descend(c, create=True)
+                        else:
+                            return f
+                            # TODO: Why return f? We've got
+                            # components left and hit a file; this
+                            # should be an error.
+
+                            # errormsg = "Reached a file called {} while trying to resolve a symlink; cannot proceed"
+                            # raise VirtualDirectoryError(errormsg.format(c))
+                    else:
+                        # It's a file, and there's no path components left, so just return that.
+                        return f
+                else:
+                    # f was not found
+                    if force_create:
+                        directory = directory.descend(c, create=True)
+                    else:
+                        return None
             else:
-                directory = directory.descend(c, create=True)
-        return directory
+                # c is not in our index
+                if force_create:
+                    directory = directory.descend(c, create=True)
+                else:
+                    return None
+        # You can only exit the while loop with a return, or exception, so you shouldn't be here.
 
     def _check_replacement(self, name, path_prefix, fileListResult):
         """ Checks whether 'name' exists, and if so, whether we can overwrite it.
@@ -297,6 +427,7 @@ class CasBasedDirectory(Directory):
             return True
         if (isinstance(existing_entry,
                        (remote_execution_pb2.FileNode, remote_execution_pb2.SymlinkNode))):
+            self.delete_entry(name)
             fileListResult.overwritten.append(relative_pathname)
             return True
         elif isinstance(existing_entry, remote_execution_pb2.DirectoryNode):
@@ -314,23 +445,29 @@ class CasBasedDirectory(Directory):
                        .format(name, type(existing_entry)))
         return False  # In case asserts are disabled
 
-    def _import_directory_recursively(self, directory_name, source_directory, remaining_path, path_prefix):
-        """ _import_directory_recursively and _import_files_from_directory will be called alternately
-        as a directory tree is descended. """
-        if directory_name in self.index:
-            subdir = self._resolve_symlink_or_directory(directory_name)
-        else:
-            subdir = self._add_directory(directory_name)
-        new_path_prefix = os.path.join(path_prefix, directory_name)
-        subdir_result = subdir._import_files_from_directory(os.path.join(source_directory, directory_name),
-                                                            [os.path.sep.join(remaining_path)],
-                                                            path_prefix=new_path_prefix)
-        return subdir_result
-
     def _import_files_from_directory(self, source_directory, files, path_prefix=""):
-        """ Imports files from a traditional directory """
+        """ Imports files from a traditional directory. """
+
+        def _import_directory_recursively(directory_name, source_directory, remaining_path, path_prefix):
+            """ _import_directory_recursively and _import_files_from_directory will be called alternately
+            as a directory tree is descended. """
+            if directory_name in self.index:
+                if self._is_followable(directory_name):
+                    subdir = self._force_resolve(directory_name)
+                else:
+                    self.delete_entry(directory_name)
+                    subdir = self._add_directory(directory_name)
+                    result.overwritten.append(relative_pathname)
+            else:
+                subdir = self._add_directory(directory_name)
+            new_path_prefix = os.path.join(path_prefix, directory_name)
+            subdir_result = subdir._import_files_from_directory(os.path.join(source_directory, directory_name),
+                                                                [os.path.sep.join(remaining_path)],
+                                                                path_prefix=new_path_prefix)
+            return subdir_result
+
         result = FileListResult()
-        for entry in sorted(files):
+        for entry in files:
             split_path = entry.split(os.path.sep)
             # The actual file on the FS we're importing
             import_file = os.path.join(source_directory, entry)
@@ -338,14 +475,18 @@ class CasBasedDirectory(Directory):
             relative_pathname = os.path.join(path_prefix, entry)
             if len(split_path) > 1:
                 directory_name = split_path[0]
-                # Hand this off to the importer for that subdir. This will only do one file -
-                # a better way would be to hand off all the files in this subdir at once.
-                subdir_result = self._import_directory_recursively(directory_name, source_directory,
-                                                                   split_path[1:], path_prefix)
+                # Hand this off to the importer for that subdir.
+
+                # It would be advantageous to batch these together by
+                # directory_name. However, we can't do it out of
+                # order, since importing symlinks affects the results
+                # of other imports.
+                subdir_result = _import_directory_recursively(directory_name, source_directory,
+                                                              split_path[1:], path_prefix)
                 result.combine(subdir_result)
             elif os.path.islink(import_file):
                 if self._check_replacement(entry, path_prefix, result):
-                    self._add_new_link(source_directory, entry)
+                    self._copy_link_from_filesystem(source_directory, entry)
                     result.files_written.append(relative_pathname)
             elif os.path.isdir(import_file):
                 # A plain directory which already exists isn't a problem; just ignore it.
@@ -357,6 +498,93 @@ class CasBasedDirectory(Directory):
                     result.files_written.append(relative_pathname)
         return result
 
+    def _files_in_subdir(sorted_files, dirname):
+        """Filters sorted_files and returns only the ones which have
+           'dirname' as a prefix, with that prefix removed.
+
+        """
+        if not dirname.endswith(os.path.sep):
+            dirname += os.path.sep
+        return [f[len(dirname):] for f in sorted_files if f.startswith(dirname)]
+
+    def _partial_import_cas_into_cas(self, source_directory, files, path_prefix="", file_list_required=True):
+        """ Import only the files and symlinks listed in 'files' from source_directory to this one.
+        Args:
+           source_directory (:class:`.CasBasedDirectory`): The directory to import from
+           files ([str]): List of pathnames to import.
+           path_prefix (str): Prefix used to add entries to the file list result.
+           file_list_required: Whether to update the file list while processing.
+        """
+        result = FileListResult()
+        processed_directories = set()
+        for f in files:
+            fullname = os.path.join(path_prefix, f)
+            components = f.split(os.path.sep)
+            if len(components) > 1:
+                # We are importing a thing which is in a subdirectory. We may have already seen this dirname
+                # for a previous file.
+                dirname = components[0]
+                if dirname not in processed_directories:
+                    # Now strip off the first directory name and import files recursively.
+                    subcomponents = CasBasedDirectory._files_in_subdir(files, dirname)
+                    # We will fail at this point if there is a file or symlink to file called 'dirname'.
+                    if dirname in self.index:
+                        resolved_component = self._resolve(dirname, force_create=True)
+                        if isinstance(resolved_component, remote_execution_pb2.FileNode):
+                            self.delete_entry(dirname)
+                            result.overwritten.append(f)
+                            dest_subdir = self.descend(dirname, create=True)
+                        else:
+                            dest_subdir = resolved_component
+                    else:
+                        dest_subdir = self.descend(dirname, create=True)
+                    src_subdir = source_directory.descend(dirname)
+                    import_result = dest_subdir._partial_import_cas_into_cas(src_subdir, subcomponents,
+                                                                             path_prefix=fullname,
+                                                                             file_list_required=file_list_required)
+                    result.combine(import_result)
+                processed_directories.add(dirname)
+            elif isinstance(source_directory.index[f].buildstream_object, CasBasedDirectory):
+                # The thing in the input file list is a directory on
+                # its own. In which case, replace any existing file,
+                # or symlink to file with the new, blank directory -
+                # if it's neither of those things, or doesn't exist,
+                # then just create the dir.
+                if f in self.index:
+                    x = self._resolve(f)
+                    if x is None:
+                        # If we're importing a blank directory, and the target has a broken symlink, then do nothing.
+                        pass
+                    elif isinstance(x, remote_execution_pb2.FileNode):
+                        # Files with the same name, or symlinks to files, get removed.
+                        pass
+                    else:
+                        # There's either a symlink (valid or not) or existing directory with this name, so do nothing.
+                        pass
+                else:
+                    self.descend(f, create=True)
+            else:
+                # We're importing a file or symlink - replace anything with the same name.
+                importable = self._check_replacement(f, path_prefix, result)
+                if importable:
+                    item = source_directory.index[f].pb_object
+                    if isinstance(item, remote_execution_pb2.FileNode):
+                        filenode = self.pb2_directory.files.add(digest=item.digest, name=f,
+                                                                is_executable=item.is_executable)
+                        self.index[f] = IndexEntry(filenode, modified=(fullname in result.overwritten))
+                    else:
+                        assert(isinstance(item, remote_execution_pb2.SymlinkNode))
+                        self._add_new_link_direct(name=f, target=item.target)
+        return result
+
+    def _import_cas_into_cas(self, source_directory, files=None):
+        """ A full import is significantly quicker than a partial import, because we can just
+        replace one directory with another's hash, without doing any recursion.
+        """
+
+        # You must pass a list into _partial_import (not a generator)
+        return self._partial_import_cas_into_cas(source_directory, list(files))
+
     def import_files(self, external_pathspec, *, files=None,
                      report_written=True, update_utimes=False,
                      can_link=False):
@@ -378,28 +606,27 @@ class CasBasedDirectory(Directory):
 
         can_link (bool): Ignored, since hard links do not have any meaning within CAS.
         """
-        if isinstance(external_pathspec, FileBasedDirectory):
-            source_directory = external_pathspec._get_underlying_directory()
-        elif isinstance(external_pathspec, CasBasedDirectory):
-            # TODO: This transfers from one CAS to another via the
-            # filesystem, which is very inefficient. Alter this so it
-            # transfers refs across directly.
-            with tempfile.TemporaryDirectory(prefix="roundtrip") as tmpdir:
-                external_pathspec.export_files(tmpdir)
-                if files is None:
-                    files = list_relative_paths(tmpdir)
-                result = self._import_files_from_directory(tmpdir, files=files)
-            return result
-        else:
-            source_directory = external_pathspec
 
         if files is None:
-            files = list_relative_paths(source_directory)
+            if isinstance(external_pathspec, str):
+                files = list_relative_paths(external_pathspec)
+            else:
+                assert isinstance(external_pathspec, Directory)
+                files = external_pathspec.list_relative_paths()
+
+        if isinstance(external_pathspec, FileBasedDirectory):
+            source_directory = external_pathspec.get_underlying_directory()
+            result = self._import_files_from_directory(source_directory, files=files)
+        elif isinstance(external_pathspec, str):
+            source_directory = external_pathspec
+            result = self._import_files_from_directory(source_directory, files=files)
+        else:
+            assert isinstance(external_pathspec, CasBasedDirectory)
+            result = self._import_cas_into_cas(external_pathspec, files=files)
 
         # TODO: No notice is taken of report_written, update_utimes or can_link.
         # Current behaviour is to fully populate the report, which is inefficient,
         # but still correct.
-        result = self._import_files_from_directory(source_directory, files=files)
 
         # We need to recalculate and store the hashes of all directories both
         # up and down the tree; we have changed our directory by importing files
@@ -526,7 +753,7 @@ class CasBasedDirectory(Directory):
                 filelist.append(k)
         return filelist
 
-    def list_relative_paths(self):
+    def list_relative_paths(self, relpath=""):
         """Provide a list of all relative paths.
 
         NOTE: This list is not in the same order as utils.list_relative_paths.
@@ -534,13 +761,32 @@ class CasBasedDirectory(Directory):
         Return value: List(str) - list of all paths
         """
 
-        filelist = []
-        for (k, v) in self.index.items():
-            if isinstance(v.buildstream_object, CasBasedDirectory):
-                filelist.extend([k + os.path.sep + x for x in v.buildstream_object.list_relative_paths()])
-            elif isinstance(v.pb_object, remote_execution_pb2.FileNode):
-                filelist.append(k)
-        return filelist
+        symlink_list = filter(lambda i: isinstance(i[1].pb_object, remote_execution_pb2.SymlinkNode),
+                              self.index.items())
+        file_list = list(filter(lambda i: isinstance(i[1].pb_object, remote_execution_pb2.FileNode),
+                           self.index.items()))
+        directory_list = filter(lambda i: isinstance(i[1].buildstream_object, CasBasedDirectory),
+                                self.index.items())
+
+        # We need to mimic the behaviour of os.walk, in which symlinks
+        # to directories count as directories and symlinks to file or
+        # broken symlinks count as files. os.walk doesn't follow
+        # symlinks, so we don't recurse.
+        for (k, v) in sorted(symlink_list):
+            target = self._resolve(k, absolute_symlinks_resolve=True)
+            if isinstance(target, CasBasedDirectory):
+                yield os.path.join(relpath, k)
+            else:
+                file_list.append((k, v))
+
+        if file_list == [] and relpath != "":
+            yield relpath
+        else:
+            for (k, v) in sorted(file_list):
+                yield os.path.join(relpath, k)
+
+        for (k, v) in sorted(directory_list):
+            yield from v.buildstream_object.list_relative_paths(relpath=os.path.join(relpath, k))
 
     def recalculate_hash(self):
         """ Recalcuates the hash for this directory and store the results in

tests/storage/virtual_directory_import.py

+import os
+import pytest
+import random
+import copy
+import tempfile
+from tests.testutils import cli
+
+
+from buildstream.storage._casbaseddirectory import CasBasedDirectory
+from buildstream.storage._filebaseddirectory import FileBasedDirectory
+from buildstream._artifactcache import ArtifactCache
+from buildstream._artifactcache.cascache import CASCache
+from buildstream import utils
+
+
+class FakeContext():
+    def __init__(self):
+        self.config_cache_quota = "65536"
+
+    def get_projects(self):
+        return []
+
+# This is a set of example file system contents. The test attempts to import
+# each on top of each other to test importing works consistently.
+# Each tuple is defined as (<filename>, <type>, <content>). Type can be
+# 'F' (file), 'S' (symlink) or 'D' (directory) with content being the contents
+# for a file or the destination for a symlink.
+root_filesets = [
+    # Arbitrary test sets
+    [('a/b/c/textfile1', 'F', 'This is textfile 1\n')],
+    [('a/b/c/textfile1', 'F', 'This is the replacement textfile 1\n')],
+    [('a/b/d', 'D', '')],
+    [('a/b/c', 'S', '/a/b/d')],
+    [('a/b/d', 'S', '/a/b/c')],
+    [('a/b/d', 'D', ''), ('a/b/c', 'S', '/a/b/d')],
+    [('a/b/c', 'D', ''), ('a/b/d', 'S', '/a/b/c')],
+    [('a/b', 'F', 'This is textfile 1\n')],
+    [('a/b/c', 'F', 'This is textfile 1\n')],
+    [('a/b/c', 'D', '')]
+]
+
+empty_hash_ref = "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"
+RANDOM_SEED = 69105
+
+
+def generate_import_roots(rootno, directory):
+    rootname = "root{}".format(rootno)
+    rootdir = os.path.join(directory, "content", rootname)
+
+    for (path, typesymbol, content) in root_filesets[rootno - 1]:
+        if typesymbol == 'F':
+            (dirnames, filename) = os.path.split(path)
+            os.makedirs(os.path.join(rootdir, dirnames), exist_ok=True)
+            with open(os.path.join(rootdir, dirnames, filename), "wt") as f:
+                f.write(content)
+        elif typesymbol == 'D':
+            os.makedirs(os.path.join(rootdir, path), exist_ok=True)
+        elif typesymbol == 'S':
+            (dirnames, filename) = os.path.split(path)
+            os.makedirs(os.path.join(rootdir, dirnames), exist_ok=True)
+            os.symlink(content, os.path.join(rootdir, path))
+
+
+def generate_random_root(rootno, directory):
+    random.seed(RANDOM_SEED + rootno)
+    rootname = "root{}".format(rootno)
+    rootdir = os.path.join(directory, "content", rootname)
+    things = []
+    locations = ['.']
+    os.makedirs(rootdir)
+    for i in range(0, 100):
+        location = random.choice(locations)
+        thingname = "node{}".format(i)
+        thing = random.choice(['dir', 'link', 'file'])
+        target = os.path.join(rootdir, location, thingname)
+        description = thing
+        if thing == 'dir':
+            os.makedirs(target)
+            locations.append(os.path.join(location, thingname))
+        elif thing == 'file':
+            with open(target, "wt") as f:
+                f.write("This is node {}\n".format(i))
+        elif thing == 'link':
+            # TODO: Make some relative symlinks
+            if random.randint(1, 3) == 1 or len(things) == 0:
+                os.symlink("/broken", target)
+                description = "symlink pointing to /broken"
+            else:
+                symlink_destination = random.choice(things)
+                os.symlink(symlink_destination, target)
+                description = "symlink pointing to {}".format(symlink_destination)
+        things.append(os.path.join(location, thingname))
+
+
+def file_contents(path):
+    with open(path, "r") as f:
+        result = f.read()
+    return result
+
+
+def file_contents_are(path, contents):
+    return file_contents(path) == contents
+
+
+def create_new_casdir(root_number, fake_context, tmpdir):
+    d = CasBasedDirectory(fake_context)
+    d.import_files(os.path.join(tmpdir, "content", "root{}".format(root_number)))
+    assert d.ref.hash != empty_hash_ref
+    return d
+
+
+def create_new_filedir(root_number, tmpdir):
+    root = os.path.join(tmpdir, "vdir")
+    os.makedirs(root)
+    d = FileBasedDirectory(root)
+    d.import_files(os.path.join(tmpdir, "content", "root{}".format(root_number)))
+    return d
+
+
+def combinations(integer_range):
+    for x in integer_range:
+        for y in integer_range:
+            yield (x, y)
+
+
+def resolve_symlinks(path, root):
+    """ A function to resolve symlinks inside 'path' components apart from the last one.
+        For example, resolve_symlinks('/a/b/c/d', '/a/b')
+        will return '/a/b/f/d' if /a/b/c is a symlink to /a/b/f. The final component of
+        'path' is not resolved, because we typically want to inspect the symlink found
+        at that path, not its target.
+
+    """
+    components = path.split(os.path.sep)
+    location = root
+    for i in range(0, len(components) - 1):
+        location = os.path.join(location, components[i])
+        if os.path.islink(location):
+            # Resolve the link, add on all the remaining components
+            target = os.path.join(os.readlink(location))
+            tail = os.path.sep.join(components[i + 1:])
+
+            if target.startswith(os.path.sep):
+                # Absolute link - relative to root
+                location = os.path.join(root, target, tail)
+            else:
+                # Relative link - relative to symlink location
+                location = os.path.join(location, target)
+            return resolve_symlinks(location, root)
+    # If we got here, no symlinks were found. Add on the final component and return.
+    location = os.path.join(location, components[-1])
+    return location
+
+
+def directory_not_empty(path):
+    return os.listdir(path)
+
+
+def _import_test(tmpdir, original, overlay, generator_function, verify_contents=False):
+    fake_context = FakeContext()
+    fake_context.artifactdir = tmpdir
+    print("Creating CAS Cache with artifact dir {}".format(tmpdir))
+    fake_context.artifactcache = CASCache(fake_context)
+    # Create some fake content
+    generator_function(original, tmpdir)
+    if original != overlay:
+        generator_function(overlay, tmpdir)
+
+    d = create_new_casdir(original, fake_context, tmpdir)
+
+    duplicate_cas = create_new_casdir(original, fake_context, tmpdir)
+
+    assert duplicate_cas.ref.hash == d.ref.hash
+
+    d2 = create_new_casdir(overlay, fake_context, tmpdir)
+    print("Importing dir {} into {}".format(overlay, original))
+    d.import_files(d2)
+    export_dir = os.path.join(tmpdir, "output")
+    roundtrip_dir = os.path.join(tmpdir, "roundtrip")
+    d2.export_files(roundtrip_dir)
+    d.export_files(export_dir)
+
+    if verify_contents:
+        for item in root_filesets[overlay - 1]:
+            (path, typename, content) = item
+            realpath = resolve_symlinks(path, export_dir)
+            if typename == 'F':
+                if os.path.isdir(realpath) and directory_not_empty(realpath):
+                    # The file should not have overwritten the directory in this case.
+                    pass
+                else:
+                    assert os.path.isfile(realpath), "{} did not exist in the combined virtual directory".format(path)
+                    assert file_contents_are(realpath, content)
+            elif typename == 'S':
+                if os.path.isdir(realpath) and directory_not_empty(realpath):
+                    # The symlink should not have overwritten the directory in this case.
+                    pass
+                else:
+                    assert os.path.islink(realpath)
+                    assert os.readlink(realpath) == content
+            elif typename == 'D':
+                # We can't do any more tests than this because it
+                # depends on things present in the original. Blank
+                # directories here will be ignored and the original
+                # left in place.
+                assert os.path.lexists(realpath)
+
+    # Now do the same thing with filebaseddirectories and check the contents match
+
+    files = list(utils.list_relative_paths(roundtrip_dir))
+    print("Importing from filesystem: filelist is: {}".format(files))
+    duplicate_cas._import_files_from_directory(roundtrip_dir, files=files)
+    duplicate_cas._recalculate_recursing_down()
+    if duplicate_cas.parent:
+        duplicate_cas.parent._recalculate_recursing_up(duplicate_cas)
+        print("Result of direct import: {}".format(duplicate_cas.show_files_recursive()))
+
+    assert duplicate_cas.ref.hash == d.ref.hash
+
+
+@pytest.mark.parametrize("original,overlay", combinations(range(1, len(root_filesets) + 1)))
+def test_fixed_cas_import(cli, tmpdir, original, overlay):
+    _import_test(tmpdir, original, overlay, generate_import_roots, verify_contents=True)
+
+
+@pytest.mark.parametrize("original,overlay", combinations(range(1, 11)))
+def test_random_cas_import_fast(cli, tmpdir, original, overlay):
+    _import_test(tmpdir, original, overlay, generate_random_root, verify_contents=False)
+
+
+def _listing_test(tmpdir, root, generator_function):
+    fake_context = FakeContext()
+    fake_context.artifactdir = tmpdir
+    print("Creating CAS Cache with artifact dir {}".format(tmpdir))
+    fake_context.artifactcache = CASCache(fake_context)
+    # Create some fake content
+    generator_function(root, tmpdir)
+
+    d = create_new_filedir(root, tmpdir)
+    filelist = list(d.list_relative_paths())
+
+    d2 = create_new_casdir(root, fake_context, tmpdir)
+    filelist2 = list(d2.list_relative_paths())
+
+    print("filelist for root {} via FileBasedDirectory:".format(root))
+    print("{}".format(filelist))
+    print("filelist for root {} via CasBasedDirectory:".format(root))
+    print("{}".format(filelist2))
+    assert filelist == filelist2
+
+
+@pytest.mark.parametrize("root", range(1, 11))
+def test_random_directory_listing(cli, tmpdir, root):
+    _listing_test(tmpdir, root, generate_random_root)
+
+
+@pytest.mark.parametrize("root", [1, 2, 3, 4, 5])
+def test_fixed_directory_listing(cli, tmpdir, root):
+    _listing_test(tmpdir, root, generate_import_roots)
+
+
+def main():
+    for i in range(1, 6):
+        with tempfile.TemporaryDirectory(prefix="/home/jimmacarthur/.cache/buildstream/cas") as tmpdirname:
+            test_fixed_directory_listing(None, tmpdirname, i)
+
+    for i in range(1, 11):
+        with tempfile.TemporaryDirectory(prefix="/home/jimmacarthur/.cache/buildstream/cas") as tmpdirname:
+            test_random_directory_listing(None, tmpdirname, i)
+
+    for i in range(1, 21):
+        for j in range(1, 21):
+            with tempfile.TemporaryDirectory(prefix="/home/jimmacarthur/.cache/buildstream/cas") as tmpdirname:
+                test_random_cas_import_fast(None, tmpdirname, i, j)
+
+    for i in range(1, len(root_filesets) + 1):
+        for j in range(1, len(root_filesets) + 1):
+            with tempfile.TemporaryDirectory(prefix="/home/jimmacarthur/.cache/buildstream/cas") as tmpdirname:
+                test_fixed_cas_import(None, tmpdirname, i, j)
+
+
+if __name__ == "__main__":
+    main()