Have a BaseStore class ?

[Carreau]

Thi is a suggestion to have a BaseStore class or ABC, that store implementations can inherit from to have basic functionality.

Problem description

Saying that stores are "just" MutableMapping may be insufficient, a number of places call methods on stores that are not available in MutableMapping, which:

• Make static type Checking fails if one uses MutableMapping as a type annotation.
• Would provide a default methods, like close() method that is no-op, and would avoid the need to check for method existence.
• Avoid reimplementation of boilerplate :def __len__(self): return len(self.keys()), still allowing overwrite.

[jakirkham]

Also issue ( #573 ) and issue ( #349 ) are related.

[jakirkham]

One thing that might be nice is to have a store that wraps a minimal MutableMapping with the extra bits of this ABC. That way users still have a path for using a minimal MutableMapping in this new framework.

[joshmoore]

see https://github.com/manzt/napari-dzi-zarr/blob/master/napari_dzi_zarr/store.py from @manzt for a possible consumer of a base class.

[Carreau]

there seem to be a number of place where the test assumes that stores can do things quite differently than what I was expecting and where dict, pass the test, but MemoryStore does not.

Typically things like

def test_getsize():
    ...

    store = # Put dict() or MemoryStore() here
    store['boo'] = None
    assert -1 == getsize(store)

Will pass for dict, but MemoryStore will refuse to store None.

or

from zarr.storage import MemoryStore, init_array
from zarr import Array

store = # Put dict() or MemoryStore() here
init_array(store, path="foo/bar", shape=100, chunks=10)
z =  Array(
    store,
    path="foo/bar"
)

z[:] = 42
z.store['foo'] = b'bar'
z[:] = 43

Where MemoryStore will be quite unhappy because z.store['foo'] = b'bar' will mess-up the hierarchy, but dict won't call "requires parent" so works fine.

So i'm unsure which is the "correct" behavior.

[jakirkham]

Done in PR ( #789 ) and preceding work