repr changes

[alimanfoo]

This PR has some work on array and group repr, resolves #83, resolves #115, resolves #132.

[alimanfoo]

I've simplified the array and group repr to a minimal output consistent with h5py. To still get diagnostic information on an array when you want it there is a new info() method on the array class. E.g.:

>>> import zarr
>>> g = zarr.group()
>>> g
<zarr group '/' (0 arrays, 0 groups)>
>>> g.create_group('foo/bar')
<zarr group '/foo/bar' (0 arrays, 0 groups)>
>>> g
<zarr group '/' (0 arrays, 1 groups)>
>>> g.create_dataset('qux', shape=1000, chunks=100, dtype='i4', filters=[zarr.Delta('i4')], compression='gzip')
<zarr array '/qux': shape (1000,), type '<i4'>
>>> g
<zarr group '/' (1 arrays, 1 groups)>
>>> z = g['qux']
>>> z
<zarr array '/qux': shape (1000,), type '<i4'>
>>> print(z.info())
Name                   : /qux
Type                   : zarr.core.Array
Data type              : int32
Shape                  : (1000,)
Chunk shape            : (100,)
Order                  : C
Filter [0]             : Delta(dtype='<i4')
Compressor             : Zlib(level=1)
Store type             : zarr.storage.DictStore
No. bytes              : 4000 (3.9K)
No. bytes stored       : 350
Storage ratio          : 11.43
No. chunks initialized : 0/10

>>> z[:] = 42
>>> print(z.info())
Name                   : /qux
Type                   : zarr.core.Array
Data type              : int32
Shape                  : (1000,)
Chunk shape            : (100,)
Order                  : C
Filter [0]             : Delta(dtype='<i4')
Compressor             : Zlib(level=1)
Store type             : zarr.storage.DictStore
No. bytes              : 4000 (3.9K)
No. bytes stored       : 490
Storage ratio          : 8.16
No. chunks initialized : 10/10

cc @jakirkham, comments welcome.

[alimanfoo]

I've made some further changes to also simplify the group repr and avoid any potentially expensive computations when generating reprs. New examples:

    >>> import zarr
    >>> root = zarr.group()
    >>> foo = root.create_group('foo')
    >>> bar = foo.zeros('bar', shape=1000000, chunks=100000)
    >>> bar[:] = 42
    >>> root
    <zarr.hierarchy.Group '/'>
    >>> root.info
    Name        : /
    Type        : zarr.hierarchy.Group
    Read-only   : False
    Store type  : zarr.storage.DictStore
    No. members : 1
    No. arrays  : 0
    No. groups  : 1
    Groups      : foo

    >>> foo
    <zarr.hierarchy.Group '/foo'>
    >>> foo.info
    Name        : /foo
    Type        : zarr.hierarchy.Group
    Read-only   : False
    Store type  : zarr.storage.DictStore
    No. members : 1
    No. arrays  : 1
    No. groups  : 0
    Arrays      : bar

    >>> bar
    <zarr.core.Array '/foo/bar' (1000000,) float64>
    >>> bar.info
    Name               : /foo/bar
    Type               : zarr.core.Array
    Data type          : float64
    Shape              : (1000000,)
    Chunk shape        : (100000,)
    Order              : C
    Read-only          : False
    Compressor         : Blosc(cname='lz4', clevel=5, shuffle=SHUFFLE, blocksize=0)
    Store type         : zarr.storage.DictStore
    No. bytes          : 8000000 (7.6M)
    No. bytes stored   : 38482 (37.6K)
    Storage ratio      : 207.9
    Chunks initialized : 10/10

[alimanfoo]

Test coverage should be back up. Merging soon.

[jakirkham]

Sorry coming to this late and have only skimmed the changes. Generally I like the new repr. It's compact, simple, and clean. Adding everything else to info is nice. Also like that info is much more orderly.

The only comment I might have is whether we want to be listing groups through info or whether we should be using the tree-work to handle that somehow.

[alimanfoo]

No problem, thanks for comments. FWIW I think it's worth including a listing of groups and arrays in group info, it's potentially more compact than viewing as a tree, and so provides a convenient way to get an overview of first level members. The tree view is then useful to browse beyond immediate children.

[jakirkham]

So I missed this change the other day. Part of the problem with dropping coverage is we may miss Python 2 specific branches. Am suggesting we revert this one line in PR ( https://github.com/alimanfoo/zarr/pull/169 ). Though it would be good to know the motivation for this change in case there is a better solution.