docs: add docs on extending zarr 3

docs/user-guide/arrays.rst

@@ -196,7 +196,7 @@ algorithm (compression level 3) internally within Blosc, and with the
 bit-shuffle filter applied.
 
 When using a compressor, it can be useful to get some diagnostics on the
-compression ratio. Zarr arrays provide the :property:`zarr.Array.info` property
+compression ratio. Zarr arrays provide the :attr:`zarr.Array.info` property
 which can be used to print useful diagnostics, e.g.:
 
 .. ipython:: python
@@ -212,7 +212,7 @@ prints additional diagnostics, e.g.:
 
 .. note::
    :func:`zarr.Array.info_complete` will inspect the underlying store and may
-   be slow for large arrays. Use :property:`zarr.Array.info` if detailed storage
+   be slow for large arrays. Use :attr:`zarr.Array.info` if detailed storage
    statistics are not needed.
 
 If you don't specify a compressor, by default Zarr uses the Blosc

docs/user-guide/extending.rst

@@ -0,0 +1,91 @@
+
+Extending Zarr
+==============
+
+Zarr-Python 3 was designed to be extensible. This means that you can extend
+the library by writing custom classes and plugins. Currently, Zarr can be extended
+in the following ways:
+
+Custom codecs
+-------------
+
+.. note::
+    This section explains how custom codecs can be created for Zarr version 3 data. For Zarr
+    version 2, codecs should subclass the
+    `numcodecs.abc.Codec <https://numcodecs.readthedocs.io/en/stable/abc.html#numcodecs.abc.Codec>`_
+    base class and register through
+    `numcodecs.registry.register_codec <https://numcodecs.readthedocs.io/en/stable/registry.html#numcodecs.registry.register_codec>`_.
+
+There are three types of codecs in Zarr:
+- array-to-array
+- array-to-bytes
+- bytes-to-bytes
+
+Array-to-array codecs are used to transform the array data before serializing
+to bytes. Examples include delta encoding or scaling codecs. Array-to-bytes codecs are used
+for serializing the array data to bytes. In Zarr, the main codec to use for numeric arrays
+is the :class:`zarr.codecs.BytesCodec`. Bytes-to-bytes codecs transform the serialized bytestreams
+of the array data. Examples include compression codecs, such as
+:class:`zarr.codecs.GzipCodec`, :class:`zarr.codecs.BloscCodec` or
+:class:`zarr.codecs.ZstdCodec`, and codecs that add a checksum to the bytestream, such as
+:class:`zarr.codecs.Crc32cCodec`.
+
+Custom codecs for Zarr are implemented by subclassing the relevant base class, see
+:class:`zarr.abc.codec.ArrayArrayCodec`, :class:`zarr.abc.codec.ArrayBytesCodec` and
+:class:`zarr.abc.codec.BytesBytesCodec`. Most custom codecs should implemented the
+``_encode_single`` and ``_decode_single`` methods. These methods operate on single chunks
+of the array data. Alternatively, custom codecs can implement the ``encode`` and ``decode``
+methods, which operate on batches of chunks, in case the codec is intended to implement
+its own batch processing.
+
+Custom codecs should also implement the following methods:
+
+- ``compute_encoded_size``, which returns the byte size of the encoded data given the byte
+  size of the original data. It should raise ``NotImplementedError`` for codecs with
+  variable-sized outputs, such as compression codecs.
+- ``validate`` (optional), which can be used to check that the codec metadata is compatible with the
+  array metadata. It should raise errors if not.
+- ``resolve_metadata`` (optional), which is important for codecs that change the shape,
+  dtype or fill value of a chunk.
+- ``evolve_from_array_spec`` (optional), which can be useful for automatically filling in
+  codec configuration metadata from the array metadata.
+
+To use custom codecs in Zarr, they need to be registered using the
+`entrypoint mechanism <https://packaging.python.org/en/latest/specifications/entry-points/>`_.
+Commonly, entrypoints are declared in the ``pyproject.toml`` of your package under the
+``[project.entry-points."zarr.codecs"]`` section. Zarr will automatically discover and
+load all codecs registered with the entrypoint mechanism from imported modules.
+
+.. code-block:: toml
+
+    [project.entry-points."zarr.codecs"]
+    "custompackage.fancy_codec" = "custompackage:FancyCodec"
+
+New codecs need to have their own unique identifier. To avoid naming collisions, it is
+strongly recommended to prefix the codec identifier with a unique name. For example,
+the codecs from ``numcodecs`` are prefixed with ``numcodecs.``, e.g. ``numcodecs.delta``.
+
+.. note::
+    Note that the extension mechanism for the Zarr version 3 is still under development.
+    Requirements for custom codecs including the choice of codec identifiers might
+    change in the future.
+
+It is also possible to register codecs as replacements for existing codecs. This might be
+useful for providing specialized implementations, such as GPU-based codecs. In case of
+multiple codecs, the :mod:`zarr.core.config` mechanism can be used to select the preferred
+implementation.
+
+Custom stores
+-------------
+
+Coming soon.
+
+Custom array buffers
+--------------------
+
+Coming soon.
+
+Other extensions
+----------------
+
+In the future, Zarr will support writing custom custom data types and chunk grids.

docs/user-guide/index.rst

@@ -24,7 +24,8 @@ Advanced Topics
 
     performance
     consolidated_metadata
+    extending
+
 
 .. Coming soon
     async
-    extending

pyproject.toml

@@ -139,8 +139,8 @@ numpy = ["1.25", "2.1"]
 features = ["gpu"]
 
 [tool.hatch.envs.test.scripts]
-run-coverage = "pytest --cov-config=pyproject.toml --cov=pkg --cov=tests"
-run-coverage-gpu = "pip install cupy-cuda12x && pytest -m gpu --cov-config=pyproject.toml --cov=pkg --cov=tests"
+run-coverage = "pytest --cov-config=pyproject.toml --cov=pkg --cov=src"
+run-coverage-gpu = "pip install cupy-cuda12x && pytest -m gpu --cov-config=pyproject.toml --cov=pkg --cov=src"
 run = "run-coverage --no-cov"
 run-verbose = "run-coverage --verbose"
 run-mypy = "mypy src"
@@ -160,7 +160,7 @@ numpy = ["1.25", "2.1"]
 version = ["minimal"]
 
 [tool.hatch.envs.gputest.scripts]
-run-coverage = "pytest -m gpu --cov-config=pyproject.toml --cov=pkg --cov=tests"
+run-coverage = "pytest -m gpu --cov-config=pyproject.toml --cov=pkg --cov=src"
 run = "run-coverage --no-cov"
 run-verbose = "run-coverage --verbose"
 run-mypy = "mypy src"

src/zarr/storage/local.py

@@ -189,6 +189,18 @@ async def set_partial_values(
         await concurrent_map(args, asyncio.to_thread, limit=None)  # TODO: fix limit
 
     async def delete(self, key: str) -> None:
+        """
+        Remove a key from the store.
+
+        Parameters
+        ----------
+        key : str
+
+        Notes
+        -----
+        If ``key`` is a directory within this store, the entire directory
+        at ``store.root / key`` is deleted.
+        """
         # docstring inherited
         self._check_writable()
         path = self.root / key