Universal Pathlib is a Python library that extends the pathlib.Path
API to support a variety of backend filesystems via filesystem_spec
.
Install the latest version of universal_pathlib
with pip or conda. Please note
that while this will install fsspec
as a dependency, for some filesystems, you
have to install additional packages. For example, to use S3, you need to install
s3fs
, or better depend on fsspec[s3]
:
python -m pip install universal_pathlib
conda install -c conda-forge universal_pathlib
Below is a pyproject.toml
based example for adding universal_pathlib
to your
project as a dependency if you want to use it with s3
and http
filesystems:
[project]
name = "myproject"
requires-python = ">=3.8"
dependencies = [
"universal_pathlib>=0.2.5",
"fsspec[s3,http]",
]
See [filesystem_spec/pyproject.toml][fsspec-pyproject-toml] for an overview of the available fsspec extras.
[fsspec-pyproject-toml]: https://github.com/fsspec/filesystem_spec/blob/master/pyproject.toml#L26
# pip install universal_pathlib fsspec[s3]
>>> from upath import UPath
>>>
>>> s3path = UPath("s3://test_bucket") / "example.txt"
>>> s3path.name
example.txt
>>> s3path.stem
example
>>> s3path.suffix
.txt
>>> s3path.exists()
True
>>> s3path.read_text()
'Hello World'
For more examples, see the example notebook here.
file:
and local:
Local filesystemmemory:
Ephemeral filesystem in RAMaz:
, adl:
, abfs:
and abfss:
Azure Storage (requires adlfs
)data:
RFC 2397 style data URLs (requires fsspec>=2023.12.2
)github:
GitHub repository filesystemhttp:
and https:
HTTP(S)-based filesystemhdfs:
Hadoop distributed filesystemgs:
and gcs:
Google Cloud Storage (requires gcsfs
)s3:
and s3a:
AWS S3 (requires s3fs
to be installed)sftp:
and ssh:
SFTP and SSH filesystems (requires paramiko
)smb:
SMB filesystems (requires smbprotocol
)webdav
, webdav+http:
and webdav+https:
WebDAV-based filesystem on top of
HTTP(S) (requires webdav4[fsspec]
)It is likely, that other fsspec-compatible filesystems are supported through the default implementation. But because they are not tested in the universal_pathlib test-suite, correct behavior is not guaranteed. If you encounter any issues with a specific filesystem using the default implementation, please open an issue. We are happy to add support for other filesystems via custom UPath implementations. And of course, contributions for new filesystems are welcome!
The class hierarchy for UPath
implementations and their relation to the stdlib
pathlib
classes are visualized in the following diagram:
flowchart TB
subgraph s0[pathlib]
A---> B
A--> AP
A--> AW
B--> BP
AP---> BP
B--> BW
AW---> BW
end
subgraph s1[upath]
B ---> U
U --> UP
U --> UW
BP --> UP
BW --> UW
U --> UL
U --> US3
U --> UH
U -.-> UO
end
A(PurePath)
AP(PurePosixPath)
AW(PureWindowsPath)
B(Path)
BP(PosixPath)
BW(WindowsPath)
U(UPath)
UP(PosixUPath)
UW(WindowsUPath)
UL(FilePath)
US3(S3Path)
UH(HttpPath)
UO(...Path)
classDef np fill:#f7f7f7,stroke:#2166ac,stroke-width:2px,color:#333
classDef nu fill:#f7f7f7,stroke:#b2182b,stroke-width:2px,color:#333
class A,AP,AW,B,BP,BW,UP,UW np
class U,UL,US3,UH,UO nu
style UO stroke-dasharray: 3 3
style s0 fill:none,stroke:#07b,stroke-width:3px,stroke-dasharray:3,color:#07b
style s1 fill:none,stroke:#d02,stroke-width:3px,stroke-dasharray:3,color:#d02
When instantiating UPath
the returned instance type is determined by the path,
or better said, the "protocol" that was provided to the constructor. The UPath
class will return a registered implementation for the protocol, if available. If
no specialized implementation can be found but the protocol is available through
fsspec
, it will return a UPath
instance and provide filesystem access with a
default implementation. Please note the default implementation can not guarantee
correct behavior for filesystems that are not tested in the test-suite.
If a local path is provided UPath
will return a PosixUPath
or WindowsUPath
instance. These two implementations are 100% compatible with the PosixPath
and
WindowsPath
classes of their specific Python version. They're tested against a
large subset of the CPython pathlib test-suite to ensure compatibility.
If a local urlpath is provided, i.e. a "file://" or "local://" URI, the returned
instance type will be a FilePath
instance. This class is a subclass of UPath
that provides file access via LocalFileSystem
from fsspec
. You can use it to
ensure that all your local file access is done through fsspec
as well.
The public class interface of UPath
extends pathlib.Path
via attributes that
simplify interaction with filesystem_spec
. Think of the UPath
class in terms
of the following code:
from pathlib import Path
from typing import Any, Mapping
from fsspec import AbstractFileSystem
class UPath(Path):
# the real implementation is more complex, but this is the general idea
@property
def protocol(self) -> str:
"""The fsspec protocol for the path."""
@property
def storage_options(self) -> Mapping[str, Any]:
"""The fsspec storage options for the path."""
@property
def path(self) -> str:
"""The path that a fsspec filesystem can use."""
@property
def fs(self) -> AbstractFileSystem:
"""The cached fsspec filesystem instance for the path."""
These attributes are used to provide a public interface to move from the UPath
instance to more fsspec specific code:
from upath import UPath
from fsspec import filesystem
p = UPath("s3://bucket/file.txt", anon=True)
fs = filesystem(p.protocol, **p.storage_options) # equivalent to p.fs
with fs.open(p.path) as f:
data = f.read()
If you want to create your own UPath implementations, there are multiple ways to customize your subclass behavior. Here are a few things to keep in mind when you create your own UPath implementation:
upath.registry
, and subclassingWhen instantiating UPath(...)
the UPath.__new__()
method determines the path
protocol and returns a registered implementation for the protocol, if available.
The registered implementations are mapped in the upath.registry
module. When a
protocol is not registered, universal_pathlib
checks if the protocol is mapped
to an fsspec
filesystem. If so, it returns an instance of UPath
and provides
filesystem access through the default implementation. The protocol is determined
by either looking at the URI scheme of the first argument to the constructor, or
by using the protocol
keyword argument:
from upath import UPath
from upath.implementations.cloud import S3Path
from upath.implementations.memory import MemoryPath
p0 = UPath("s3://bucket/file.txt")
assert p0.protocol == "s3"
assert type(p0) is S3Path
assert isinstance(p0, UPath)
p1 = UPath("/some/path/file.txt", protocol="memory")
assert p1.protocol == "memory"
assert type(p1) is MemoryPath
assert isinstance(p1, UPath)
# the ftp filesystem current has no custom UPath implementation and is not
# tested in the universal_pathlib test-suite. Therefore, the default UPath
# implementation is returned, and a warning is emitted on instantiation.
p2 = UPath("ftp://ftp.ncbi.nih.gov/snp/archive")
assert p2.protocol == "ftp"
assert type(p2) is UPath
This has some implications for custom UPath subclasses. We'll go through the two main cases where you might want to create a custom UPath implementation:
Let's say you would like to add a new implementation of your "myproto" protocol.
You already built a custom AbstractFileSystem implementation for "myproto" which
you have registered through fsspec.registry
. In some cases it is possible that
the custom filesystem class already works with UPath
's default implementation,
and you don't need to necessarily create a custom UPath implementation:
import fsspec.registry
from fsspec.spec import AbstractFileSystem
class MyProtoFileSystem(AbstractFileSystem):
protocol = ("myproto",)
... # your custom implementation
fsspec.registry.register_implementation("myproto", MyProtoFileSystem)
from upath import UPath
p = UPath("myproto:///my/proto/path")
assert type(p) is UPath
assert p.protocol == "myproto"
assert isinstance(p.fs, MyProtoFileSystem)
Sometimes the default implementation isn't sufficient and some method(s) have to
be overridden to provide correct behavior. In this case, create a custom UPath
implementation:
from upath import UPath
class MyProtoPath(UPath):
def mkdir(self, mode=0o777, parents=False, exist_ok=False):
something = {...: ...} # fixes to make MyProtoFileSystem.mkdir work
self.fs.mkdir(self.path, **something)
def path(self):
path = super().path
if path.startswith("/"):
return path[1:] # MyProtoFileSystem needs the path without "/"
return path
If you use your implementation directly via MyProtoPath("myproto:///a/b")
, you
can use this implementation already as is. If you want a call to UPath(...)
to
return your custom implementation when the detected protocol is "myproto"
, you
need to register your implementation. The next section explains your options.
Also note: In case you develop a custom UPath
implementation, please feel free
to open an issue to discuss integrating it in universal_pathlib
.
You can register your custom UPath implementation dynamically from Python:
# for example: mymodule/submodule.py
from upath import UPath
from upath.registry import register_implementation
class MyProtoPath(UPath):
... # your custom implementation
register_implementation("myproto", MyProtoPath)
If you distribute your implementation in your own Python package, you can inform
universal_pathlib
about your implementation via the entry_points
mechanism:
# pyproject.toml
[project.entry-points."universal_pathlib.implementations"]
myproto = "my_module.submodule:MyPath"
# setup.cfg
[options.entry_points]
universal_pathlib.implementations =
myproto = my_module.submodule:MyPath
Chose the method that fits your use-case best. If you have questions, open a new
issue in the universal_pathlib
repository. We are happy to help you!
Once you thoroughly test your custom UPath implementation, it's likely that some
methods need to be overridden to provide correct behavior compared to stdlib
's
pathlib.Path
class. The most common issue is that for certain edge cases, your
implementation is not raising the same exceptions compared to the pathlib.Path
class. Or that the UPath.path
property needs some prefix removed or added.
class MyProtoPath(UPath):
@property
def path(self) -> str:
if p := self.path.startswith("/"):
p = p[1:]
return p
def mkdir(self, mode=0o777, parents=False, exist_ok=False):
if some_edge_case:
raise FileExistsError(str(self))
super().mkdir(mode=mode, parents=parents, exist_ok=exist_ok)
def is_file(self):
return self.fs.isfile(self.path, myproto_option=123)
It's possible that you might want to extract additional storage options from the
user provided arguments to you constructor. You can provide a custom classmethod
for _parse_storage_options
:
import os
class MyProtoPath(UPath):
@classmethod
def _parse_storage_options(
cls, urlpath: str, protocol: str, storage_options: Mapping[str, Any]
) -> dict[str, Any]:
if "SOME_VAR" in os.environ:
storage_options["some_var"] = os.environ["SOME_VAR"]
storage_options["my_proto_caching"] = True
storage_options["extra"] = get_setting_from_path(urlpath)
return storage_options
To have more control over fsspec filesystem instantiation you can write a custom
_fs_factory
classmethod:
class MyProtoPath(UPath):
@classmethod
def _fs_factory(
cls, urlpath: str, protocol: str, storage_options: Mapping[str, Any]
) -> AbstractFileSystem:
myfs = ... # custom code that creates a AbstractFileSystem instance
return myfs
In special cases you need to take more control over how the init args are parsed
for your custom subclass. You can override __init__
or the UPath
classmethod
_transform_init_args
. The latter handles pickling of your custom subclass in a
better way in case you modify storage options or the protocol.
class MyProtoPath(UPath):
@classmethod
def _transform_init_args(
cls,
args: tuple[str | os.PathLike, ...],
protocol: str,
storage_options: dict[str, Any],
) -> tuple[tuple[str | os.PathLike, ...], str, dict[str, Any]]:
# check the cloud, http or webdav implementations for examples
...
return args, protocol, storage_options
There are cases for which you want to disable the protocol dispatch mechanism of
the UPath.__new__
constructor. For example if you want to extend the class API
of your UPath
implementation, and use it as the base class for other, directly
instantiated subclasses. Together with other customization options this can be a
useful feature. Please be aware that in this case all protocols are handled with
the default implementation in UPath. Please always feel free to open an issue in
the issue tracker to discuss your use case. We're happy to help with finding the
most maintainable solution.
class ExtraUPath(UPath):
_protocol_dispatch = False # disable the registry return an ExtraUPath
def some_extra_method(self) -> str:
return "hello world"
assert ExtraUPath("s3://bucket/file.txt").some_extra_method() == "hello world"
UPath's internal implementation is likely going to change with larger changes in
CPython's stdlib pathlib
landing in the next Python versions (3.13
, 3.14
).
To reduce the problems for user code, when these changes are landing in UPath
,
there have been some significant changes in v0.2.0
. This migration guide tries
to help migrating code that extensively relies on private implementation details
of the UPath
class of versions v0.1.x
to the new and better supported public
interface of v0.2.0
v0.2.0
If you implemented a custom accessor subclass, it is now recommended to override
the corresponding UPath
methods in your subclass directly:
# OLD: v0.1.x
from upath.core import UPath, _FSSpecAccessor
class MyAccessor(_FSSpecAccessor):
def exists(self, path, **kwargs):
# custom code
return path.fs.exists(self._format_path(path), **kwargs)
def touch(self, path, **kwargs):
# custom
return path.fs.touch(self._format_path(path), **kwargs)
class MyPath(UPath):
_default_accessor = MyAccessor
# NEW: v0.2.0+
from upath import UPath
class MyPath(UPath):
def exists(self, *, follow_symlinks=True):
kwargs = {} # custom code
return self.fs.exists(self.path, **kwargs)
def touch(self, mode=0o666, exist_ok=True):
kwargs = {} # custom code
self.fs.touch(self.path, **kwargs)
__init__
methodIf you implemented a custom __init__
method for your accessor subclass usually
the intention is to customize how the fsspec filesystem instance is created. You
have two options to recreate this with the new implementation. Chose one or both
dependent on the level of control you need.
# OLD: v0.1.x
import fsspec
from upath.core import UPath, _FSSpecAccessor
class MyAccessor(_FSSpecAccessor):
def __init__(self, parsed_url: SplitResult | None, **kwargs: Any) -> None:
# custom code
protocol = ...
storage_options = ...
self._fs = fsspec.filesystem(protocol, storage_options)
class MyPath(UPath):
_default_accessor = MyAccessor
# NEW: v0.2.0+
from upath import UPath
class MyPath(UPath):
@classmethod
def _parse_storage_options(
cls, urlpath: str, protocol: str, storage_options: Mapping[str, Any]
) -> dict[str, Any]:
# custom code to change storage_options
storage_options = ...
return storage_options
@classmethod
def _fs_factory(
cls, urlpath: str, protocol: str, storage_options: Mapping[str, Any]
) -> AbstractFileSystem:
# custom code to instantiate fsspec filesystem
protocol = ...
storage_options = ... # note changes to storage_options here won't
# show up in MyPath().storage_options
return fsspec.filesystem(protocol, **storage_options)
._accessor
The _accessor
attribute and the _FSSpecAccessor
class is deprecated. In case
you need direct access to the underlying filesystem, just access UPath().fs
.
# OLD: v0.1.x
from upath.core import UPath
class MyPath(UPath):
def mkdir(self, mode=0o777, parents=False, exist_ok=False):
self._accessor.mkdir(...) # custom access to the underlying fs...
# NEW: v0.2.0+
from upath import UPath
class MyPath(UPath):
def mkdir(self, mode=0o777, parents=False, exist_ok=False):
self.fs.mkdir(...)
._path
, ._kwargs
, ._drv
, ._root
, ._parts
If you access one of the listed private attributes directly, move your code over to the following public versions:
deprecated | v0.2.0+ |
---|---|
UPath()._path |
UPath().path |
UPath()._kwargs |
UPath().storage_options |
UPath()._drv |
UPath().drive |
UPath()._root |
UPath().root |
UPath()._parts |
UPath().parts |
._url
Be aware that the ._url
attribute will likely be deprecated once UPath()
has
support for uri fragments and uri query parameters through a public api. In case
you are interested in contributing this functionality, please open an issue!
_from_parts
, _parse_args
, _format_parsed_parts
If your code is currently calling any of the three above listed classmethods, it
relies on functionality based on the implementation of pathlib
in Python up to
3.11
. In universal_pathlib
we vendor code that allows the UPath()
class to
be based on the 3.12
implementation of pathlib.Path
alone. Usually, usage of
those classmethods occurs when copying some code of the internal implementations
of methods of the UPath
0.1.4
classes.
_format_parsed_parts
methods in v0.2.0
, try overriding
UPath().path
and/or UPath().with_segments()
._from_parts
and _parse_args
classmethods can now be implemented via
the _transform_init_args
method or via more functionality in the new flavour
class. Please open an issue for discussion in case you have this use case._URIFlavour
classesThe _URIFlavour
class was removed from universal_pathlib
and the new flavour
class for fsspec filesystem path operations now lives in upath._flavour
. As of
now the internal FSSpecFlavour is experimental. In a future Python version, it's
likely that a flavour or flavour-like base class will become public, that allows
us to base our internal implementation on. Until then, if you find yourself in a
situation where a custom path flavour would solve your problem, please feel free
to open an issue for discussion. We're happy to find a maintainable solution.
.parse_parts()
, .casefold()
, .join_parsed_parts()
of ._flavour
These methods of the ._flavour
attribute of pathlib.Path()
and UPath()
are
specific to pathlib
of Python versions up to 3.11
. UPath()
is now based on
the 3.12
implementation of pathlib.Path
. Please refer to the implementations
of the upath._flavour
submodule to see how you could avoid using them.
Some issues in UPath
's behavior with specific fsspec filesystems are fixed via
installation of a newer version of its upstream dependencies. Below you can find
a list of known issues and their solutions. We attempt to keep this list updated
whenever we encounter more:
fsspec
fixed glob behavior when handling **
patterns in fsspec>=2023.9.0
gcsfs>=2022.7.1
fsspec>=2022.5.0
fsspec>=2024.2.0
to correctly return True
for directoriesContributions are very welcome. To learn more, see the Contributor Guide.
Distributed under the terms of the MIT license, universal_pathlib is free and open source software.
If you encounter any problems, or if you create your own implementations and run into limitations, please file an issue with a detailed description. We are always happy to help with any problems you might encounter.