What are variable annotations in Python 3.6?

Question

Python 3.6 is about to be released. PEP 494 -- Python 3.6 Release Schedule mentions the end of December, so I went through What's New in Python 3.6 to see they mention the variable annotations:

PEP 484 introduced standard for type annotations of function parameters, a.k.a. type hints. This PEP adds syntax to Python for annotating the types of variables including class variables and instance variables:

primes: List[int] = []

captain: str  # Note: no initial value!

class Starship:
     stats: Dict[str, int] = {}

Just as for function annotations, the Python interpreter does not attach any particular meaning to variable annotations and only stores them in a special attribute __annotations__ of a class or module. In contrast to variable declarations in statically typed languages, the goal of annotation syntax is to provide an easy way to specify structured type metadata for third party tools and libraries via the abstract syntax tree and the __annotations__ attribute.

So from what I read they are part of the type hints coming from Python 3.5, described in What are Type hints in Python 3.5.

I follow the captain: str and class Starship example, but not sure about the last one: How does primes: List[int] = [] explain? Is it defining an empty list that will just allow integers?

Answer 1

What are variable annotations?

Variable annotations are just the next step from # type comments, as they were defined in PEP 484; the rationale behind this change is highlighted in the respective section of PEP 526.

So, instead of hinting the type with:

primes = []  # type: List[int]

New syntax was introduced to allow for directly annotating the type with an assignment of the form:

primes: List[int] = []

which, as @Martijn pointed out, denotes a list of integers by using types available in typing and initializing it to an empty list.

What changes does it bring?

The first change introduced was new syntax that allows you to annotate a name with a type, either standalone after the : character or optionally annotate while also assigning a value to it:

annotated_assignment_stmt ::=  augtarget ":" expression ["=" expression]

So the example in question:

   primes: List[int] = [ ]
#    ^        ^         ^
#  augtarget  |         |
#         expression    |
#                  expression (optionally initialize to empty list)

Additional changes were also introduced along with the new syntax; modules and classes now have an __annotations__ attribute (as functions have had since PEP 3107 -- Function Annotations) in which the type metadata is attached:

from typing import get_type_hints  # grabs __annotations__

Now __main__.__annotations__ holds the declared types:

>>> from typing import List, get_type_hints
>>> primes: List[int] = []
>>> captain: str
>>> import __main__
>>> get_type_hints(__main__)
{'primes': typing.List<~T>[int]}

captain won't currently show up through get_type_hints because get_type_hints only returns types that can also be accessed on a module; i.e., it needs a value first:

>>> captain = "Picard"
>>> get_type_hints(__main__)
{'primes': typing.List<~T>[int], 'captain': <class 'str'>}

_{Using print(__annotations__) will show 'captain': <class 'str'> but you really shouldn't be accessing __annotations__ directly.}

Similarly, for classes:

>>> get_type_hints(Starship)
ChainMap({'stats': typing.Dict<~KT, ~VT>[str, int]}, {})

Where a ChainMap is used to grab the annotations for a given class (located in the first mapping) and all annotations defined in the base classes found in its mro (consequent mappings, {} for object).

Along with the new syntax, a new ClassVar type has been added to denote class variables. Yup, stats in your example is actually an instance variable, not a ClassVar.

Will I be forced to use it?

As with type hints from PEP 484, these are completely optional and are of main use for type checking tools (and whatever else you can build based on this information). It is to be provisional when the stable version of Python 3.6 is released so small tweaks might be added in the future.