from __future__ import co_annotations

In the prototype, the semantics presented in this PEP are gated with:

from __future__ import co_annotations

__co_annotations__

Python supports runtime metadata for annotations for three different types: function, classes, and modules. The basic approach to implement this PEP is much the same for all three with only minor variations.

With this PEP, each of these types adds a new attribute, __co_annotations__. __co_annotations__ is a function: it takes no arguments, and must return either None or a dict (or subclass of dict). It adds the following semantics:

• __co_annotations__ is always set, and may contain either None or a callable.
• __co_annotations__ cannot be deleted.
• __annotations__ and __co_annotations__ can’t both be set to a useful value simultaneously:
  • If you set __annotations__ to a dict, this also sets __co_annotations__ to None.
  • If you set __co_annotations__ to a callable, this also deletes __annotations__

Internally, __co_annotations__ is a “data descriptor”, where functions are called whenever user code gets, sets, or deletes the attribute. In all three cases, the object has separate internal storage for the current value of the __co_annotations__ attribute.

__annotations__ is also as a data descriptor, with its own separate internal storage for its internal value. The code implementing the “get” for __annotations__ works something like this:

if (the internal value is set)
    return the internal annotations dict
if (__co_annotations__ is not None)
    call the __co_annotations__ function
    if the result is a dict:
        store the result as the internal value
        set __co_annotations__ to None
        return the internal value
do whatever this object does when there are no annotations

Unbound code objects

When Python code defines one of these three objects with annotations, the Python compiler generates a separate code object which builds and returns the appropriate annotations dict. Wherever possible, the “annotation code object” is then stored unbound as the internal value of __co_annotations__; it is then bound on demand when the user asks for __annotations__.

This is a useful optimization for both speed and memory consumption. Python processes rarely examine annotations at runtime. Therefore, pre-binding these code objects to function objects would usually be a waste of resources.

When is this optimization not possible?

• When an annotation function contains references to free variables, in the current function or in an outer function.
• When an annotation function is defined on a method (a function defined inside a class) and the annotations possibly refer directly to class variables.

Note that user code isn’t permitted to directly access these unbound code objects. If the user “gets” the value of __co_annotations__, and the internal value of __co_annotations__ is an unbound code object, it immediately binds the code object, and the resulting function object is stored as the new value of __co_annotations__ and returned.

(However, these unbound code objects are stored in the .pyc file. So a determined user could examine them should that be necessary for some reason.)

Function Annotations

When compiling a function, the CPython bytecode compiler visits the annotations for the function all in one place, starting with compiler_visit_annotations() in compile.c. If there are any annotations, they create the scope for the annotations function on demand, and compiler_visit_annotations() assembles it.

The code object is passed in place of the annotations dict for the MAKE_FUNCTION bytecode instruction. MAKE_FUNCTION supports a new bit in its oparg bitfield, 0x10, which tells it to expect a co_annotations code object on the stack. The bitfields for annotations (0x04) and co_annotations (0x10) are mutually exclusive.

When binding an unbound annotation code object, a function will use its own __globals__ as the new function’s globals.

One quirk of Python: you can’t actually remove the annotations from a function object. If you delete the __annotations__ attribute of a function, then get its __annotations__ member, it will create an empty dict and use that as its __annotations__. The implementation of this PEP maintains this quirk for backwards compatibility.

Class Annotations

When compiling a class body, the compiler maintains two scopes: one for the normal class body code, and one for annotations. (This is facilitated by four new functions: compiler.c adds compiler_push_scope() and compiler_pop_scope(), and symtable.c adds symtable_push_scope() and symtable_pop_scope().) Once the code generator reaches the end of the class body, but before it generates the bytecode for the class body, it assembles the bytecode for __co_annotations__, then assigns that to __co_annotations__ using STORE_NAME.

It also sets a new __globals__ attribute. Currently it does this by calling globals() and storing the result. (Surely there’s a more elegant way to find the class’s globals–but this was good enough for the prototype.) When binding an unbound annotation code object, a class will use the value of this __globals__ attribute. When the class drops its reference to the unbound code object–either because it has bound it to a function, or because __annotations__ has been explicitly set–it also deletes its __globals__ attribute.

As discussed above, examination or modification of __annotations__ from within the class body is no longer supported. Also, any flow control (if or try blocks) around declarations of members with annotations is unsupported.

If you delete the __annotations__ attribute of a class, then get its __annotations__ member, it will return the annotations dict of the first base class with annotations set. If no base classes have annotations set, it will raise AttributeError.

Although it’s an implementation-specific detail, currently classes store the internal value of __co_annotations__ in their tp_dict under the same name.

Module Annotations

Module annotations work much the same as class annotations. The main difference is, a module uses its own dict as the __globals__ when binding the function.

If you delete the __annotations__ attribute of a class, then get its __annotations__ member, the module will raise AttributeError.

Annotations With Closures

It’s possible to write annotations that refer to free variables, and even free variables that have yet to be defined. For example:

from __future__ import co_annotations

def outer():
    def middle():
        def inner(a:mytype, b:mytype2): pass
        mytype = str
        return inner
    mytype2 = int
    return middle()

fn = outer()
print(fn.__annotations__)

At the time fn is set, inner.__co_annotations__() hasn’t been run. So it has to retain a reference to the future definitions of mytype and mytype2 if it is to correctly evaluate its annotations.

If an annotation function refers to a local variable from the current function scope, or a free variable from an enclosing function scope–if, in CPython, the annotation function code object contains one or more LOAD_DEREF opcodes–then the annotation code object is bound at definition time with references to these variables. LOAD_DEREF instructions require the annotation function to be bound with special run-time information (in CPython, a freevars array). Rather than store that separately and use that to later lazy-bind the function object, the current implementation simply early-binds the function object.

Note that, since the annotation function inner.__co_annotations__() is defined while parsing outer(), from Python’s perspective the annotation function is a “nested function”. So “local variable inside the ‘current’ function” and “free variable from an enclosing function” are, from the perspective of the annotation function, the same thing.

Annotations That Refer To Class Variables

It’s possible to write annotations that refer to class variables, and even class variables that haven’t yet been defined. For example:

from __future__ import co_annotations

class C:
    def method(a:mytype): pass
    mytype = str

print(C.method.__annotations__)

Internally, annotation functions are defined as a new type of “block” in CPython’s symbol table called an AnnotationBlock. An AnnotationBlock is almost identical to a FunctionBlock. It differs in that it’s permitted to see names from an enclosing class scope. (Again: annotation functions are functions, and they’re defined inside the same scope as the thing they’re being defined on. So in the above example, the annotation function for C.method() is defined inside C.)

If it’s possible that an annotation function refers to class variables–if all these conditions are true:

• The annotation function is being defined inside a class scope.
• The generated code for the annotation function has at least one LOAD_NAME instruction.

Then the annotation function is bound at the time it’s set on the class/function, and this binding includes a reference to the class dict. The class dict is pushed on the stack, and the MAKE_FUNCTION bytecode instruction takes a new second bitfield (0x20) indicating that it should consume that stack argument and store it as __locals__ on the newly created function object.

Then, at the time the function is executed, the f_locals field of the frame object is set to the function’s __locals__, if set. This permits LOAD_NAME opcodes to work normally, which means the code generated for annotation functions is nearly identical to that generated for conventional Python functions.

Interactive REPL Shell

Everything works the same inside Python’s interactive REPL shell, except for module annotations in the interactive module (__main__) itself. Since that module is never “finished”, there’s no specific point where we can compile the __co_annotations__ function.

For the sake of simplicity, in this case we forego delayed evaluation. Module-level annotations in the REPL shell will continue to work exactly as they do today, evaluating immediately and setting the result directly inside the __annotations__ dict.

(It might be possible to support delayed evaluation here. But it gets complicated quickly, and for a nearly-non-existent use case.)

Annotations On Local Variables Inside Functions

Python supports syntax for local variable annotations inside functions. However, these annotations have no runtime effect–they’re discarded at compile-time. Therefore, this PEP doesn’t need to do anything to support them, the same as stock semantics and PEP 563.

Performance Comparison

Performance with this PEP should be favorable, when compared with either stock behavior or PEP 563. In general, resources are only consumed on demand—“you only pay for what you use”.

There are three scenarios to consider:

• the runtime cost when annotations aren’t defined,
• the runtime cost when annotations are defined but not referenced, and
• the runtime cost when annotations are defined and referenced.

We’ll examine each of these scenarios in the context of all three semantics for annotations: stock, PEP 563, and this PEP.

When there are no annotations, all three semantics have the same runtime cost: zero. No annotations dict is created and no code is generated for it. This requires no runtime processor time and consumes no memory.

When annotations are defined but not referenced, the runtime cost of Python with this PEP should be roughly equal to or slightly better than PEP 563 semantics, and slightly better than “stock” Python semantics. The specifics depend on the object being annotated:

• With stock semantics, the annotations dict is always built, and set as an attribute of the object being annotated.
• In PEP 563 semantics, for function objects, a single constant (a tuple) is set as an attribute of the function. For class and module objects, the annotations dict is always built and set as an attribute of the class or module.
• With this PEP, a single object is set as an attribute of the object being annotated. Most often, this object is a constant (a code object). In cases where the annotation refers to local variables or class variables, the code object will be bound to a function object, and the function object is set as the attribute of the object being annotated.

When annotations are both defined and referenced, code using this PEP should be much faster than code using PEP 563 semantics, and equivalent to or slightly improved over original Python semantics. PEP 563 semantics requires invoking eval() for every value inside an annotations dict, which is enormously slow. And, as already mentioned, this PEP generates measurably more efficient bytecode for class and module annotations than stock semantics; for function annotations, this PEP and stock semantics should be roughly equivalent.

Memory use should also be comparable in all three scenarios across all three semantic contexts. In the first and third scenarios, memory usage should be roughly equivalent in all cases. In the second scenario, when annotations are defined but not referenced, using this PEP’s semantics will mean the function/class/module will store one unused code object (possibly bound to an unused function object); with the other two semantics, they’ll store one unused dictionary (or constant tuple).

Bytecode Comparison

The bytecode generated for annotations functions with this PEP uses the efficient BUILD_CONST_KEY_MAP opcode to build the dict for all annotatable objects: functions, classes, and modules.

Stock semantics also uses BUILD_CONST_KEY_MAP bytecode for function annotations. PEP 563 has an even more efficient method for building annotations dicts on functions, leveraging the fact that its annotations dicts only contain strings for both keys and values. At compile-time it constructs a tuple containing pairs of keys and values at compile-time, then at runtime it converts that tuple into a dict on demand. This is a faster technique than either stock semantics or this PEP can employ, because in those two cases annotations dicts can contain Python values of any type. Of course, this performance win is negated if the annotations are examined, due to the overhead of eval().

For class and module annotations, both stock semantics and PEP 563 generate a longer and slightly-less-efficient stanza of bytecode, creating the dict and setting the annotations individually.

Circular Imports

There is one unfortunately-common scenario where PEP 563 currently provides a better experience, and it has to do with large code bases, with circular dependencies and imports, that examine their annotations at run-time.

PEP 563 permitted defining and examining invalid expressions as annotations. Its implementation requires annotations to be legal Python expressions, which it then converts into strings at compile-time. But legal Python expressions may not be computable at runtime, if for example the expression references a name that isn’t defined. This is a problem for stringized annotations if they’re evaluated, e.g. with typing.get_type_hints(). But any stringized annotation may be examined harmlessly at any time–as long as you don’t evaluate it, and only examine it as a string.

Some large organizations have code bases that unfortunately have circular dependency problems with their annotations–class A has methods annotated with class B, but class B has methods annotated with class A–that can be difficult to resolve. Since PEP 563 stringizes their annotations, it allows them to leave these circular dependencies in place, and they can sidestep the circular import problem by never importing the module that defines the types used in the annotations. Their annotations can no longer be evaluated, but this appears not to be a concern in practice. They can then examine the stringized form of the annotations at runtime and this seems to be sufficient for their needs.

This PEP allows for many of the same behaviors. Annotations must be legal Python expressions, which are compiled into a function at compile-time. And if the code never examines an annotation, it won’t have any runtime effect, so here too annotations can harmlessly refer to undefined names. (It’s exactly like defining a function that refers to undefined names–then never calling that function. Until you call the function, nothing bad will happen.)

But examining an annotation when this PEP is active means evaluating it, which means the names evaluated in that expression must be defined. An undefined name will throw a NameError in an annotation function, just as it would with a stringized annotation passed in to typing.get_type_hints(), and just like any other context in Python where an expression is evaluated.

In discussions we have yet to find a solution to this problem that makes all the participants in the conversation happy. There are various avenues to explore here:

• One workaround is to continue to stringize one’s annotations, either by hand or done automatically by the Python compiler (as it does today with from __future__ import annotations). This might mean preserving Python’s current stringizing annotations going forward, although leaving it turned off by default, only available by explicit request (though likely with a different mechanism than from __future__ import annotations).
• Another possible workaround involves importing the circularly-dependent modules separately, then externally adding (“monkey-patching”) their dependencies to each other after the modules are loaded. As long as the modules don’t examine their annotations until after they are completely loaded, this should work fine and be maintainable with a minimum of effort.
• A third and more radical approach would be to change the semantics of annotations so that they don’t raise a NameError when an unknown name is evaluated, but instead create some sort of proxy “reference” object.
• Of course, even if we do deprecate PEP 563, it will be several releases before the functionality is removed, giving us several years in which to research and innovate new solutions for this problem.

In any case, the participants of the discussion agree that this PEP should still move forward, even as this issue remains currently unresolved [1].

cls.__globals__ and fn.__locals__

Is it permissible to add the __globals__ reference to class objects as proposed here? It’s not clear why this hasn’t already been done; PEP 563 could have made use of class globals, but instead made do with looking up classes inside sys.modules. Python seems strangely allergic to adding a __globals__ reference to class objects.

If adding __globals__ to class objects is indeed a bad idea (for reasons I don’t know), here are two alternatives as to how classes could get a reference to their globals for the implementation of this PEP:

• The generate code for a class could bind its annotations code object to a function at the time the class is bound, rather than waiting for __annotations__ to be referenced, making them an exception to the rule (even though “special cases aren’t special enough to break the rules”). This would result in a small additional runtime cost when annotations were defined but not referenced on class objects. Honestly I’m more worried about the lack of symmetry in semantics. (But I wouldn’t want to pre-bind all annotations code objects, as that would become much more costly for function objects, even as annotations are rarely used at runtime.)
• Use the class’s __module__ attribute to look up its module by name in sys.modules. This is what PEP 563 advises. While this is passable for userspace or library code, it seems like a little bit of a code smell for this to be defined semantics baked into the language itself.

Also, the prototype gets globals for class objects by calling globals() then storing the result. I’m sure there’s a much faster way to do this, I just didn’t know what it was when I was prototyping. I’m sure we can revise this to something much faster and much more sanitary. I’d prefer to make it completely internal anyway, and not make it visible to the user (via this new __globals__ attribute). There’s possibly already a good place to put it anyway–ht_module.

Similarly, this PEP adds one new dunder member to functions, classes, and modules (__co_annotations__), and a second new dunder member to functions (__locals__). This might be considered excessive.

Bikeshedding the name

During most of the development of this PEP, user code actually could see the raw annotation code objects. __co_annotations__ could only be set to a code object; functions and other callables weren’t permitted. In that context the name co_annotations makes a lot of sense. But with this last-minute pivot where __co_annotations__ now presents itself as a callable, perhaps the name of the attribute and the name of the from __future__ import needs a re-think.