nulltype 2.1.5

Null values and sentinels like, but not, None

Helps define ‘null’ values and sentinels parallel to, but different from, None.

None is a great sentinel value and a classic implementation of the null object pattern.

But there are times that you need more than one nullish value to represent different aspects of emptiness. “Nothing there” is logically different from “undefined,” “prohibited,” “end of data” and other kinds of null.

The core function of nulltype is representing emptiness and falsity in a way that doesn’t overload None (or False, 0, {}, [], "", or any of the other possible “there’s nothing here!” values). It helps create designated identifiers with specific meanings such as Passthrough, Prohibited, and Undefined.

Usage

from nulltype import NullType

Empty = NullType('Empty')

# following just to show it's working
assert bool(Empty) == False
assert len(Empty) == 0
assert list(Empty) == []
assert Empty.some_attribute is Empty
assert Empty[22] is Nothing
assert Empty("hey", 12) is Empty

You can create as many custom NullType values as you like. For your convenience, two default values, Null and Nothing, are exported. That way, if you don’t really want to create your own, you can import a pre-constituted null value, such as:

from nulltype import Nothing

Dereferencing

Alternate null types can be particularly useful when parsing data or traversing data structures which might or might not be present. This is common in dealing with the data returned by REST APIs, for instance.

As one example, the documentation for Google’s Gmail API suggests the following code:

threads = gmail_service.users().threads().list(userId='me').execute()
if threads['threads']:
    for thread in threads['threads']:
        print 'Thread ID: %s' % (thread['id'])

There is a lot going on there just to avoid a problematic deference. If instead you have a Nothing null type defined, the code is shorter (and avoids an extra, very transient variable):

results = gmail_service.users().threads().list(userId='me').execute()
for thread in results.get('threads', Nothing):
    print 'Thread ID: %s' % (thread['id'])

Three lines versus four may not seem like a big advantage, but the value increases with the complexity of the task. Many such “if it’s there, then…” constructs are deeply nested when dealing with API results, XML parse trees, and other fundamentally nested information sources. Saving a guard condition on every one of the nesting levels adds up quickly.

While you could almost do this in stock Python, unlike Nothing, None is not iterable. You might use an empty list [] (or an equivalent global such as EMPTYLIST) as the alternative value for the get method. Going by the documentation of many parsers and APIs, however, such uses aren’t broadly idiomatic in today’s Python community. The EMPTYLIST approach also is very specific to routines returning lists, whereas the “go ahead, get it if you can” nulltype model works well for longer chains of access:

results.get("payload", Nothing).get("headers", Nothing)

will return the correct object if it’s there, but Nothing otherwise. And if you then try to test it (e.g. with if or a logical expression) or iterate over it (e.g. with for), it will act as though it’s an empty list, or False–whatever is most useful in a given context. Whether you’re iterating, indexing, dereferencing, calling, or otherwise accessing it, a NullType is unperturbed.

Nothing isn’t nothing. It’s something that will simplify your code.

General Sentinels and Distinguished Values

While nulltype is frequently used to define new kinds of “empty” values, it’s actually more general. Beyond different forms of ‘null’, NullType instances are good general-purpose sentinels or designated values. Instead of the old:

class MySentinel(object):
    pass

Use:

MySentinel = NullType('MySentinel')

That gives you a value with known truthiness properties and a nicer printed representation.

On the off chance you want a sentinel value that is truthy rather than falsey / empty, use NonNullType, a companion to NullType that operates in almost the exact same way, but that evaluates as true.:

from nulltype import NonNullType

Full = NonNullType('Full')

assert bool(Full) is True
assert len(Full) == 1
assert list(Full) == [Full]
assert Full.some_attribute is Full
assert Full[22] is Full
assert Full("hey", 12) is Full

Experience suggests that nullish sentinels are generally adequate and preferable. And the “everything folds back to the same value” nature of even NonNullType gives a somewhat null-like, or at least non-reactive, nature. But if you do want a true-ish sentinel, there it is.

Uniqueness

NullType instances are meant to be singletons, with just one per program. They almost are, though technically multiple NullType instances are reasonable, making it more of a multiton pattern.

The uniqueness of each singleton is currently not enforced, making it a usage convention rather than strict law. With even minimal care, this is a problem roughly 0% of the time.

Notes

• Version 2.1.4 adds wheel packaging.

• Version 2.1.2 switches from BSD to Apache License 2.0 and integrates tox testing with setup.py, and updates testing with Travis CI and pyroma.

• Version 2.1 adds NonNullType, an alternative for truthy sentinels. (Most use cases should still use NullType; “full” sentinels recommended for odd cases only.)

• Version 2.0 starts major upgrade from just Boolean operations being nulled to essentially all sorts of accesses and updates being nulled. It defines two default NullType instances, Null and Nothing. The ability to have anonymous (unnamed) nulls has been removed as superfluous.

• Automated multi-version testing managed with pytest, pytest-cov, and tox. Continuous integration testing with Travis-CI. Packaging linting with pyroma.

  Successfully packaged for, and tested against, all late-model versions of Python: 2.6, 2.7, 3.2, 3.3, 3.4, and 3.5 pre-release (3.5.0b3) as well as PyPy 2.6.0 (based on 2.7.9) and PyPy3 2.4.0 (based on 3.2.5).

• Similar modules include sentinels and null. Of these, I prefer sentinels because it is clearly Python 3 ready, includes a pickle mechanism.

• The author, Jonathan Eunice or @jeunice on Twitter welcomes your comments and suggestions.

Installation

To install or upgrade to the latest version:

pip install -U nulltype

To easy_install under a specific Python version (3.3 in this example):

python3.3 -m easy_install nulltype

(You may need to prefix these with sudo to authorize installation. In environments without super-user privileges, you may want to use pip’s --user option, to install only for a single user, rather than system-wide.)