Absolute BucketPath is passed to the bucket API with leading slash #329

Summary

The docstring of BucketPath.__init__ says a path may be given in absolute form, e.g. '/dir1/...', and is still relative to the bucket. Only part of BucketPath follows this:

• exists(), is_file(), is_dir(), iterdir(), walk() and rm()'s existence check use _get_relative_posix(), which strips the leading /.
• read(), write() and rm() pass str(self._path) to the bucket API, including the leading /.
• as_udf_path() joins the absolute path onto the bucket's udf_path, which discards the bucket root.

Reproduction

Tested with exasol-bucketfs 2.3.0 on Linux.

import exasol.bucketfs as bfs
from exasol.bucketfs._buckets import MountedBucket
from exasol.bucketfs._shared import _build_url

bucket = MountedBucket(base_path="/tmp/bucket")
path = bfs.path.BucketPath("/dir/a.txt", bucket)

print(path.as_udf_path())                  # '/dir/a.txt'
print(bucket.root / str(path))             # '/dir/a.txt', used by MountedBucket.upload
print(_build_url("http://localhost:2580", "default", str(path)))
                                           # 'http://localhost:2580/default//dir/a.txt'

Consequences

• MountedBucket: write() creates /dir/a.txt on the local file system, outside the bucket's base path; read() and rm() also act outside the bucket.
• On-prem Bucket: the URL contains a double slash, .../default//dir/a.txt.
• as_udf_path() returns /dir/a.txt instead of /buckets/bfsdefault/default/dir/a.txt, which is not a valid path inside a UDF.
• exists() etc. look up dir/a.txt, so they disagree with where write() put the data.

Expected

An absolute BucketPath behaves like the corresponding relative one: read(), write(), rm() and as_udf_path() use the path without the leading /.

Context

Found while fixing #322. That fix leaves this behaviour unchanged, as it is not specific to Windows.