nob 0.5.6

Nested OBject manipulations

nob: the Nested OBject manipulator

JSON is a very popular format for nested data exchange, and Object Relational Mapping (ORM) is a popular method to help developers make sense of large JSON objects, by mapping objects to the data. In some cases however, the nesting can be very deep, and difficult to map with objects. This is where nob can be useful: it offers a simple set of tools to explore and edit any nested data (Python native dicts and lists).

For more, checkout the home page.

Philosophy and trade-offs

Nob is a wrapper around JSON-serializable Python objects that helps in high-level manipulation of the corresponding tree-like structure. So think dicts of dicts and lists, with boolean, integer, floating point and string values as data. Nob strives to be as transparent as possible compared to the native Python object it wraps. Ideally, one should always feel like using Nob is identical to using the underlying Python nested object. Unfortunately, this is not entirely possible, and using Nob will make some of your code slightly more verbose. Most notably, you'll see some [:] symbols cropping up everywhere. This is an important marker:

• If there is no [:], you are manipulating the tree-like structure of the data
• If there is a [:], you are accessing the underlying data (in pure Python)

So, what are the trade-offs? Switching from pure Python to Nob will:

• Add a bunch of [:] to your code
• Add an overhead compared to pure-python nested objects

Why would one ever use Nob? Well, in exchange, you gain:

• Direct and terse access to most data, even in deep complex nested objects
• Read / Write / Copy-Pase of data or full subtrees of the nested object
• Efficient serialization of numpy data in the tree

Let's look at an example. Suppose you're working with this file:

root:
  System:
    Library:
      Frameworks:
        Python.framework:
          Versions:
            2.7:
              bin:
                python2.7

If you wanted the last value here from pure Python, you would have to write:

pure_python['root']['System']['Library']['Frameworks']['Python.framework']['Versions']['2.7']['bin']

With nob, this becomes:

nob_object['bin'][:]

Pretty neat, no? This is in fact slower than the pure Python version since you're performing a search for the 'bin' keyword in the whole tree, but that's the price to pay. Note however a big caveat is you really shouldn't write:

nob_object['root']['System']['Library']['Frameworks']['Python.framework']['Versions']['2.7']['bin'][:]

This will perform a recursive search for 'root' in nob_object, then a new one for 'System' in nob_object['root'], etc... As you can imagine, this might become very slow! If for some reason you must perform long accesses in full, use absolute paths (see below), which do not trigger recursive search:

nob_object['/root/System/Library/Frameworks/Python.framework/Versions/2.7/bin'][:]

Also, remember that you can at any point save a position in the tree as a Python object:

a = nob_object['/root/System/Library/Frameworks/Python.framework/Versions/2.7/bin']
a[:]                     >>> python2.7

With that said, it's now up to you to see if this works for you!

Usage

Instantiation

nob.Nob objects can be instantiated directly from a Python dictionary:

n = Nob({
    'key1': 'val1',
    'key2': {
        'key3': 4,
        'key4': {'key5': 'val2'},
        'key5': [3, 4, 5]
        },
    'key5': 'val3'
    })

To create a Nob from a JSON (or YAML) file, simply read it and feed the data to the constructor:

import json
with open('file.json') as fh:
    n2 = Nob(json.load(fh))

import yaml
with open('file.yml') as fh:
    n3 = Nob(yaml.load(fh))

Similarly, to create a JSON (YAML) file from a tree, you can use:

with open('file.json', 'w') as fh:
    json.dump(n2[:], fh)

with open('file.yml', 'w') as fh:
    yaml.dump(n3[:], fh)

Important Notice: if you try to dump the Nob object, i.e. if you write:

yaml.dump(n3, fh)

this will seem like it works, but in fact yaml will serialize the full Nob object here, making it unreadable as Nob object afterwards. You'll get a cryptic error ending with:

yaml.constructor.ConstructorError: could not determine a constructor for the tag 'tag:yaml.org,2002:python/object:nob.nob.Nob'

TL;DR: don't forget the [:] here!

Basic manipulation

The variable n now holds a Nob, i.e the reference to the actual data. However, for many practical cases it is useful to work with a subset of the nob. nob offers a useful class NobView to this end, that handles identically for the most part as the main Nob, but changes performed on a NobView affect the main Nob instance that it is linked to. In practice, any access to a key of n yields a NobView instance, e.g.:

nv1 = n['/key1']         # NobView(/key1)
nv2 = n['key1']          # NobView(/key1)
nv3 = n.key1             # NobView(/key1)
nv1 == nv2 == nv3        # True

Note that a full path '/key1', as well as a simple key 'key1' are valid identifiers. Simple keys can also be called as attributes, using n.key1.

To access the actual value that is stored in the nested object, simply use the [:] operator:

nv1[:]                   >>> 'val1'
n.key1[:]                >>> 'val1'

To assign a new value to this node, you can do it directly on the NobView instance:

n.key1 = 'new'
nv1[:]                   >>> 'new'
n[:]['key1']             >>> 'new'

Of course, because of how Python variables work, you cannot simply assign the value to nv1, as this would just overwrite it's contents:

nv1 = 'new'
nv1                      >>> 'new'
n[:]['key1']             >>> 'val1'

If you find yourself with a NobView object that you would like to edit directly, you can use the .set method:

nv1 = n.key1
nv1.set('new')
n[:]['key1']             >>> 'new'

Because nested objects can contain both dicts and lists, integers are sometimes needed as keys:

n['/key2/key5/0']        >>> NobView(/key2/key5/0)
n.key2.key5[0]           >>> NobView(/key2/key5/0)
n.key2.key5['0']         >>> NobView(/key2/key5/0)

However, since Python does not support attributes starting with an integer, there is no attribute support for lists. Only key access (full path, integer index or its stringified counterpart) are supported.

Some keywords are reserved, due to the inner workings of Nob. To access a key that has a name equal to a reserved keyword, use item access (n['key'] but not n.key). To view reserved keywords, use:

n.reserved()             >>> ['_MutableMapping__marker', '_abc_cache', '_abc_negative_cache', '_abc_negative_cache_version', '_abc_registry', '_data', '_find_all', '_find_unique', '_getitem_slice', '_raw_data', '_root', '_tree', 'clear', 'copy', 'find', 'get', 'items', 'keys', 'np_deserialize', 'np_serialize', 'paths', 'pop', 'popitem', 'reserved', 'root', 'setdefault', 'update', 'val', 'values']

Manipulation summary

The tl;dr of nob manipulation is summarized by 3 rules:

1. n['/path/to/key'] will always work
2. n['key'] will work if key is unambiguous
3. n.key will work if key is unambiguous and key is not a reserved keyword, and key is a legal python attribute (no spaces, doesn't start with a number, no dots...)

So you can use a Nob like a nested dictionary at all times (method 1.). Methods 2 and 3 enable fast access except when they don't apply.

Smart key access

In a simple nested dictionary, the access to 'key1' would be simply done with:

nested_dict['key1']

If you are looking for e.g. key3, you would need to write:

nested_dict['key2']['key3']

For deep nested objects however, this can be a chore, and become very difficult to read. nob helps you here by supplying a smart method for finding unique keys:

n['key3']                >>> NobView(/key2/key3)
n.key3                   >>> NobView(/key2/key3)

Note that attribute access t.key3 behaves like simple key access t['key3']. This has some implications when the key is not unique in the tree. Let's say e.g. we wish to access key5. Let's try using attribute access:

n.key5                   >>> KeyError: Identifier key5 yielded 3 results instead of 1

Oops! Because key5 is not unique (it appears 3 times in the tree), t.key5 is not specific, and nob wouldn't know which one to return. In this instance, we have several possibilities, depending on which key5 we are looking for:

n.key4.key5              >>> NobView(/key2/key4/key5)
n.key2['/key5']          >>> NobView(/key2/key5)
n['/key5']               >>> NobView(/key5)

There is a bit to unpack here:

• The first key5 is unique in the NobView t.key4 (and key4 is itself unique), so t.key4.key5 finds it correctly.
• The second is complex: key2 is unique, but key5 is still not unique to t.key2. There is not much advantage compared to a full path access t['/key2/key5'].
• The last cannot be resolved using keys in its path, because there are none. The only solution is to use a full path.

Other tree tools

Paths: any Nob (or NobView) object can introspect itself to find all its valid paths:

n.paths                  >>> [Path('/'),
                              Path('/key1'),
                              Path('/key2'),
                              Path('/key2/key3'),
                              Path('/key2/key4'),
                              Path('/key2/key4/key5'),
                              Path('/key2/key5'),
                              Path('/key2/key5/0'),
                              Path('/key2/key5/1'),
                              Path('/key2/key5/2'),
                              Path('/key5')]

Find: in order to easily search in this path list, the .find method is available:

n.find('key5')           >>> [Path('/key2/key4/key5'),
                              Path('/key2/key5'),
                              Path('/key5')]

The elements of these lists are not strings, but Path objects, as described below.

Iterable: any tree or tree view is also iterable, yielding its children:

[nv for nv in n.key2]    >>> [NobView(/key2/key3),
                              NobView(/key2/key4),
                              NobView(/key2/key5)]

Copy: to make an independant copy of a tree, use its .copy() method:

n_cop = n.copy()
n == n_cop               >>> True
n_cop.key1 = 'new_val'
n == n_cop               >>> False

A new standalone tree can also be produced from any tree view:

n_cop = n.key2.copy()
n_cop == n.key2          >>> True
n_cop.key3 = 5
n_cop == n.key2          >>> False

Numpy specifics

If you end up with numpy arrays in your tree, you are no longer JSON compatible. You can remediate this by using the np.ndarray.tolist() method, but this can lead to a very long JSON file. To help you with this, Nob offers the np_serialize method, which efficiently rewrites all numpy arrays as binary strings using the internal np.save function. You can even compress these using the standard zip algorithm by passing the compress=True argument. The result can be written directly to disc as a JSON or YAML file:

n.np_serialize()
# OR
n.np_serialize(compress=True)

with open('file.json', 'w') as fh:
    json.dump(n[:], fh)
# OR
with open('file.yml', 'w') as fh:
    yaml.dump(n[:], fh)

To read it back, use the opposite function np_deserialize:

with open('file.json') as fh:
    n = Nob(json.load(fh))
# OR
with open('file.yml') as fh:
    n = Nob(yaml.load(fh))
n.np_deserialize()

And that's it, your original Nob has been recreated.

Path

All paths are stored internally using the nob.Path class. Paths are full (w.r.t. their Nob or NobView), and are in essence a list of the keys constituting the nested address. They can however be viewed equivalently as a unix-type path string with / separators. Here are some examples

p1 = Path(['key1'])
p1                       >>> Path(/key1)
p2 = Path('/key1/key2')
p2                       >>> Path(/key1/key2)
p1 / 'key3'              >>> Path(/key1/key3)
p2.parent                >>> Path(/key1)
p2.parent == p1          >>> True
'key2' in p2             >>> True
[k for k in p2]          >>> ['key1', 'key2']
p2[-1]                   >>> 'key2'
len(p2)                  >>> 2

These can be helpful to manipulate paths yourself, as any full access with a string to a Nob or NobView object also accepts a Path object. So say you are accessing the keys in list_of_keys at one position, but that thet also exist elsewhere in the tree. You could use e.g.:

root = Path('/path/to/root/of/keys')
[n[root / key] for key in list_of_keys]