PEP 9999 – Type Manipulation
- Author:
- Michael J. Sullivan <sully at msully.net>, Daniel W. Park <dnwpark at protonmail.com>, Yury Selivanov <yury at edgedb.com>
- Sponsor:
- <name of sponsor>
- PEP-Delegate:
- <PEP delegate’s name>
- Discussions-To:
- Pending
- Status:
- Draft
- Type:
- Standards Track
- Topic:
- Typing
- Created:
- <date created on, in dd-mmm-yyyy format>
- Python-Version:
- 3.15 or 3.16
- Post-History:
- Pending
- Resolution:
- <url>
Table of Contents
- Abstract
- Motivation
- Specification of Needed Preliminaries
- Specification
- Rationale
- Backwards Compatibility
- Security Implications
- How to Teach This
- Reference Implementation
- Rejected Ideas
- Open Issues
- Acknowledgements
- Copyright
Abstract
We propose to add powerful type-level type introspection and type construction facilities to the type system, inspired in large part by TypeScript’s conditional and mapped types, but adapted to the quite different conditions of Python typing.
Motivation
Python has a gradual type system, but at the heart of it is a fairly conventional and tame static type system (apart from untagged union types and type narrowing, which are common in gradual type systems but not in traditional static ones). In Python as a language, on the other hand, it is not unusual to perform complex metaprogramming, especially at the library layer.
Typically, type safety is lost when doing these sorts of things. Some
libraries come with custom mypy plugins, and a special-case
@dataclass_transform decorator was added specifically to cover the
case of dataclass-like transformations (PEP 681).
Examples: pydantic/fastapi, dataclasses, sqlalchemy
Prisma-style ORMs
Prisma, a popular ORM for TypeScript, allows writing queries like (adapted from this example):
const user = await prisma.user.findMany({
select: {
name: true,
email: true,
posts: true,
},
});
for which the inferred type will be something like:
{
email: string;
name: string | null;
posts: {
id: number;
title: string;
content: string | null;
authorId: number | null;
}[];
}[]
Here, the output type is a combination of both existing information
about the type of prisma.user and the type of the argument to
findMany. It returns an array of objects containing the properties
of user that were requested; one of the requested elements,
posts, is a “relation” referencing another model; it has all of
its properties fetched but not its relations.
We would like to be able to do something similar in Python, perhaps with a schema defined like:
class Comment:
id: Property[int]
name: Property[str]
poster: Link[User]
class Post:
id: Property[int]
title: Property[str]
content: Property[str]
comments: MultiLink[Comment]
author: Link[Comment]
class User:
id: Property[int]
name: Property[str]
email: Property[str]
posts: Link[Post]
(In Prisma, a code generator generates type definitions based on a prisma schema in its own custom format; you could imagine something similar here, or that the definitions were hand written)
and a call like:
db.select(
User,
name=True,
email=True,
posts=True,
)
which would have return type list[<User>] where:
class <User>:
name: str
email: str
posts: list[<Post>]
class <Post>
id: int
title: str
content: str
Unlike the FastAPI-style example above, we probably don’t have too much need for runtime introspection of the types here, which is good: inferring the type of a function is much less likely to be feasible.
Implementation
This will take something of a tutorial approach in discussing the implementation, and explain the features being used as we use them. More details were appear in the specification section.
First, to support the annotations we saw above, we have a collection of dummy classes with generic types.
class Pointer[T]:
pass
class Property[T](Pointer[T]):
pass
class Link[T](Pointer[T]):
pass
class SingleLink[T](Link[T]):
pass
class MultiLink[T](Link[T]):
pass
The select method is where we start seeing new things.
The **kwargs: Unpack[K] is part of this proposal, and allows
inferring a TypedDict from keyword args.
Attrs[K] extracts Member types corresponding to every
type-annotated attribute of K, while calling NewProtocol with
Member arguments constructs a new structural type.
GetName is a getter operator that fetches the name of a Member
as a literal type–all of these mechanisms lean very heavily on literal types.
GetAttr gets the type of an attribute from a class.
def select[ModelT, K: BaseTypedDict](
typ: type[ModelT],
/,
**kwargs: Unpack[K],
) -> list[
NewProtocol[
*[
Member[
GetName[c],
ConvertField[GetAttr[ModelT, GetName[c]]],
]
for c in Iter[Attrs[K]]
]
]
]: ...
ConvertField is our first type helper, and it is a conditional type alias, which decides between two types based on a (limited) subtype-ish check.
In ConvertField, we wish to drop the Property or Link
annotation and produce the underlying type, as well as, for links,
producing a new target type containing only properties and wrapping
MultiLink in a list.
type ConvertField[T] = (
AdjustLink[PropsOnly[PointerArg[T]], T] if IsSub[T, Link] else PointerArg[T]
)
PointerArg gets the type argument to Pointer or a subclass.
GetArg[T, Base, I] is one of the core primitives; it fetches the
index I type argument to Base from a type T, if T
inherits from Base.
(The subtleties of this will be discussed later; in this case, it just
grabs the argument to a Pointer).
type PointerArg[T: Pointer] = GetArg[T, Pointer, Literal[0]]
AdjustLink sticks a list around MultiLink, using features
we’ve discussed already.
type AdjustLink[Tgt, LinkTy] = list[Tgt] if IsSub[LinkTy, MultiLink] else Tgt
And the final helper, PropsOnly[T], generates a new type that
contains all the Property attributes of T.
type PropsOnly[T] = list[
NewProtocol[
*[
Member[GetName[p], PointerArg[GetType[p]]]
for p in Iter[Attrs[T]]
if IsSub[GetType[p], Property]
]
]
]
The full test is in our test suite.
Automatically deriving FastAPI CRUD models
In the FastAPI tutorial, they show how to
build CRUD endpoints for a simple Hero type. At its heart is a
series of class definitions used both to define the database interface
and to perform validation/filtering of the data in the endpoint:
class HeroBase(SQLModel):
name: str = Field(index=True)
age: int | None = Field(default=None, index=True)
class Hero(HeroBase, table=True):
id: int | None = Field(default=None, primary_key=True)
secret_name: str
class HeroPublic(HeroBase):
id: int
class HeroCreate(HeroBase):
secret_name: str
class HeroUpdate(HeroBase):
name: str | None = None
age: int | None = None
secret_name: str | None = None
The HeroPublic type is used as the return types of the read
endpoint (and is validated while being output, including having extra
fields stripped), while HeroCreate and HeroUpdate serve as
input types (automatically converted from JSON and validated based on
the types, using Pydantic).
Despite all multiple types and duplication here, mechanical rules could be written for deriving these types:
- Public should include all non-“hidden” fields, and the primary key should be made non-optional
- Create should include all fields except the primary key
- Update should include all fields except the primary key, but they should all be made optional and given a default value
With the definition of appropriate helpers, this proposal would allow writing:
class Hero(NewSQLModel, table=True):
id: int | None = Field(default=None, primary_key=True)
name: str = Field(index=True)
age: int | None = Field(default=None, index=True)
secret_name: str = Field(hidden=True)
type HeroPublic = Public[Hero]
type HeroCreate = Create[Hero]
type HeroUpdate = Update[Hero]
Those types, evaluated, would look something like:
class HeroPublic:
id: int
name: str
age: int | None
class HeroCreate:
name: str
age: int | None = None
secret_name: str
class HeroUpdate:
name: str | None = None
age: int | None = None
secret_name: str | None = None
While the implementation of Public, Create, and Update
(presented in the next subsection) are certainly more complex than
duplicating code would be, they perform quite mechanical operations
and could be included in the framework library.
A notable feature of this use case is that it depends on performing runtime evaluation of the type annotations. FastAPI uses the Pydantic models to validate and convert to/from JSON for both input and output from endpoints.
Implementation
We have a more fully-worked example in our test
suite, but here is a possible implementation of just Public:
# Extract the default type from an Init field.
# If it is a Field, then we try pulling out the "default" field,
# otherwise we return the type itself.
type GetDefault[Init] = (
GetFieldItem[Init, Literal["default"]] if IsSub[Init, Field] else Init
)
# Create takes everything but the primary key and preserves defaults
type Create[T] = NewProtocol[
*[
Member[GetName[p], GetType[p], GetQuals[p], GetDefault[GetInit[p]]]
for p in Iter[Attrs[T]]
if not IsSub[
Literal[True], GetFieldItem[GetInit[p], Literal["primary_key"]]
]
]
]
The Create type alias creates a new type (via NewProtocol) by
iterating over the attributes of the original type. It has access to
names, types, qualifiers, and the literal types of initializers (in
part through new facilities to handle the extremely common
= Field(...) like pattern used here.
Here, we filter out attributes that have primary_key=True in their
Field as well as extracting default arguments (which may be either
from a default argument to a field or specified directly as an
initializer).
dataclasses-style method generation
We would additionally like to be able to generate method signatures
based on the attributes of an object. The most well-known example of
this is probably generating __init__ methods for dataclasses,
which we present a simplified example of. (In our test suites, this is
merged with the FastAPI-style example above, but it need not be).
This kind of pattern is widespread enough that PEP 681 was created to represent a lowest-common denominator subset of what existing libraries do.
Implementation
# Generate the Member field for __init__ for a class
type InitFnType[T] = Member[
Literal["__init__"],
Callable[
[
Param[Literal["self"], Self],
*[
Param[
GetName[p],
GetType[p],
# All arguments are keyword-only
# It takes a default if a default is specified in the class
Literal["keyword"]
if IsSub[
GetDefault[GetInit[p]],
Never,
]
else Literal["keyword", "default"],
]
for p in Iter[Attrs[T]]
],
],
None,
],
Literal["ClassVar"],
]
type AddInit[T] = NewProtocol[
InitFnType[T],
*[x for x in Iter[Members[T]]],
]
Specification of Needed Preliminaries
(Some content is still in spec-draft.rst).
We have two subproposals that are necessary to get mileage out of the main part of this proposal.
Unpack of typevars for **kwargs
A minor proposal that could be split out maybe:
Supporting Unpack of typevars for **kwargs:
def f[K: BaseTypedDict](**kwargs: Unpack[K]) -> K:
return kwargs
Here BaseTypedDict is defined as:
class BaseTypedDict(typing.TypedDict):
pass
But any typeddict would be allowed there. (Or, maybe we should allow dict?)
This is basically a combination of
“PEP 692 – Using TypedDict for more precise **kwargs typing”
and the behavior of Unpack for *args
from “PEP 646 – Variadic Generics”.
This is potentially moderately useful on its own but is being done to
support processing **kwargs with type level computation.
—
Extended Callables, take 2
We introduce a Param type the contains all the information about a function param:
class Param[N: str | None, T, Q: ParamQuals = typing.Never]:
pass
ParamQuals = typing.Literal["*", "**", "default", "keyword"]
type PosParam[N: str | None, T] = Param[N, T, Literal["positional"]]
type PosDefaultParam[N: str | None, T] = Param[N, T, Literal["positional", "default"]]
type DefaultParam[N: str, T] = Param[N, T, Literal["default"]]
type NamedParam[N: str, T] = Param[N, T, Literal["keyword"]]
type NamedDefaultParam[N: str, T] = Param[N, T, Literal["keyword", "default"]]
type ArgsParam[T] = Param[Literal[None], T, Literal["*"]]
type KwargsParam[T] = Param[Literal[None], T, Literal["**"]]
And then, we can represent the type of a function like:
def func(
a: int,
/,
b: int,
c: int = 0,
*args: int,
d: int,
e: int = 0,
**kwargs: int
) -> int:
...
as (we are omiting the Literal in places):
Callable[
[
Param["a", int, "positional"],
Param["b", int],
Param["c", int, "default"],
Param[None, int, "*"],
Param["d", int, "keyword"],
Param["e", int, Literal["default", "keyword"]],
Param[None, int, "**"],
],
int,
]
or, using the type abbreviations we provide:
Callable[
[
PosParam["a", int],
Param["b", int],
DefaultParam["c", int,
ArgsParam[int, "*"],
NamedParam["d", int],
NamedDefaultParam["e", int],
KwargsParam[int],
],
int,
]
(Rationale discussed below.)
Specification
As was visible in the examples above, we introduce a few new syntactic
forms of valid types, but much of the power comes from type level
operators that will be defined in the typing module.
Grammar specification of the extensions to the type language
Note first that no changes to the Python grammar are being proposed, only to the grammar of what Python expressions are considered as valid types.
(It’s also slightly imprecise to call this a grammar: where operator
names are mentioned directly, like IsSub, they require that name
to be imported, and it could also be used qualified as
typing.IsSub or imported as a different name.)
<type> = ...
# Type booleans are all valid types too
| <type-bool>
# Conditional types
| <type> if <type-bool> else <type>
# Types with variadic arguments can have
# *[... for t in ...] arguments
| <ident>[<variadic-type-arg> +]
# Type conditional checks are boolean compositions of
# "subtype checking" and boolean Literal type checking.
<type-bool> =
IsSub[<type>, <type>]
| Bool[<type>]
| not <type-bool>
| <type-bool> and <type-bool>
| <type-bool> or <type-bool>
# Do we want these next two? Maybe not.
| Any[<variadic-type-arg> +]
| All[<variadic-type-arg> +]
<variadic-type-arg> =
<type> ,
| * <type-for-iter> ,
<type-for> = [ <type> <type-for-iter>+ <type-for-if>* ]
<type-for-iter> =
# Iterate over a tuple type
for <var> in Iter[<type>]
<type-for-if> =
if <type-bool>
TODO: explain conditional types and iteration
Type operators
In some sections below we write things like Literal[int]] to mean
“a literal that is of type int”. I don’t think I’m really
proposing to add that as a notion, but we could.
Boolean types
IsSub[T, S]: What we would want is that it returns a boolean literal type indicating whetherTis a subtype ofS. To support runtime checking, we probably need something weaker.
Basic operators
GetArg[T, Base, Idx: Literal[int]]: returns the type argument numberIdxtoTwhen interpreted asBase, orNeverif it cannot be. (That is, if we haveclass A(B[C]): ..., thenGetArg[A, B, 0] == CwhileGetArg[A, A, 0] == Never).N.B: Unfortunately
Basemust be a proper class, not a protocol. So, for example,GetArg[Ty, Iterable, 0]]to get the type of something iterable won’t work. This is because we can’t do protocol checks at runtime in general. Special forms unfortunately require some special handling: the arguments list of aCallablewill be packed in a tuple, and a...will becomeSpecialFormEllipsis.GetArgs[T, Base]: returns a tuple containing all of the type arguments ofTwhen interpreted asBase, orNeverif it cannot be.GetAttr[T, S: Literal[str]]: Extract the type of the member namedSfrom the classT.Length[T: tuple]- get the length of a tuple as an int literal (orLiteral[None]if it is unbounded)
All of the operators in this section are “lifted” over union types.
Union processing
FromUnion[T]: returns a tuple containing all of the union elements, or a 1-ary tuple containing T if it is not a union.
Object inspection
Members[T]: produces atupleofMembertypes describing the members (attributes and methods) of classT.In order to allow typechecking time and runtime evaluation coincide more closely, only members with explicit type annotations are included.
Attrs[T]: likeMembers[T]but only returns attributes (not methods).Member[N: Literal[str], T, Q: MemberQuals, Init, D]:Member, is a simple type, not an operator, that is used to describe members of classes. Its type parameters encode the information about each member.Nis the name, as a literal string typeTis the typeQis a union of qualifiers (seeMemberQualsbelow)Initis the literal type of the attribute initializer in the class (see InitField)Dis the defining class of the member. (That is, which class the member is inherited from.)
MemberQuals = Literal['ClassVar', 'Final']-MemberQualsis the type of “qualifiers” that can apply to a member; currentlyClassVarandFinal
Methods are returned as callables using the new Param based
extended callables, and carrying the ClassVar
qualifier. staticmethod and classmethod will return
staticmethod and classmethod types, which are subscriptable as
of 3.14.
TODO: What do we do about decorators in general, at runtime… This seems pretty cursed. We can probably sometimes evaluate them, if there are annotations at runtime, but in general that would require full subtype checking, which we can’t do.
We also have helpers for extracting the fields of Members; they
are all definable in terms of GetArg. (Some of them are shared
with Param, discussed below.)
GetName[T: Member | Param]GetType[T: Member | Param]GetQuals[T: Member | Param]GetInit[T: Member]GetDefiner[T: Member]
Object creation
NewProtocol[*Ps: Member]NewProtocolWithBases[Bases, Ps: tuple[Member]]- A variant that allows specifying bases too. (UNIMPLEMENTED) - OR MAYBE SHOULD NOT EXISTNewTypedDict[*Ps: Member]– TODO: Needs fleshing out; will work similarly toNewProtocolbut has different flags
InitField
We want to be able to support transforming types based on
dataclasses/attrs/pydantic style field descriptors. In order to do
that, we need to be able to consume things like calls to Field.
Our strategy for this is to introduce a new type
InitField[KwargDict] that collects arguments defined by a
KwargDict: TypedDict:
class InitField[KwargDict: BaseTypedDict]:
def __init__(self, **kwargs: typing.Unpack[KwargDict]) -> None:
...
def _get_kwargs(self) -> KwargDict:
...
When InitField or (more likely) a subtype of it is instantiated
inside a class body, we infer a more specific type for it, based on
Literal types for all the arguments passed.
So if we write:
class A:
foo: int = InitField(default=0)
then we would infer the type InitField[TypedDict('...', {'default':
Literal[0]})] for the initializer, and that would be made available
as the Init field of the Member.
Annotated
This could maybe be dropped?
Libraries like FastAPI use annotations heavily, and we would like to be able to use annotations to drive type-level computation decision making.
We understand that this may be controversial, as currently Annotated
may be fully ignored by typecheckers. The operations proposed are:
GetAnnotations[T]- Fetch the annotations of a potentially Annotated type, as Literals. Examples:GetAnnotations[Annotated[int, 'xxx']] = Literal['xxx'] GetAnnotations[Annotated[int, 'xxx', 5]] = Literal['xxx', 5] GetAnnotations[int] = Never
DropAnnotations[T]- Drop the annotations of a potentially Annotated type. Examples:DropAnnotations[Annotated[int, 'xxx']] = int DropAnnotations[Annotated[int, 'xxx', 5]] = int DropAnnotations[int] = int
Callable inspection and creation
Callable types always have their arguments exposed in the extended
Callable format discussed above.
The names, type, and qualifiers share getter operations with
Member.
TODO: Should we make GetInit be literal types of default parameter
values too?
Generic Callable
GenericCallable[Vs, Ty]: A generic callable.Vsare a tuple type of unbound type variables andTyshould be aCallable,staticmethod, orclassmethodthat has access to the variables inVs
This is kind of unsatisfying but we at least need some way to return existing generic methods and put them back into a new protocol.
String manipulation
String manipulation operations for string Literal types.
We can put more in, but this is what typescript has.
Slice and Concat are a poor man’s literal template.
We can actually implement the case functions in terms of them and a
bunch of conditionals, but shouldn’t (especially if we want it to work
for all unicode!).
Slice[S: Literal[str] | tuple, Start: Literal[int | None], End: Literal[int | None]]: Slices astror a tuple type.Concat[S1: Literal[str], S2: Literal[str]]: concatenate two stringsUppercase[S: Literal[str]]: uppercase a string literalLowercase[S: Literal[str]]: lowercase a string literalCapitalize[S: Literal[str]]: capitalize a string literalUncapitalize[S: Literal[str]]: uncapitalize a string literal
All of the operators in this section are “lifted” over union types.
Raise error
RaiseError[S: Literal[str]]: If this type needs to be evaluated to determine some actual type, generate a type error with the provided message.
Update class
TODO: This is kind of sketchy but it is I think needed for defining
base classes and type decorators that do dataclass like things.
UpdateClass[*Ps: Member]: A special form that updates an existing nominal class with new members (possibly overriding old ones, or removing them by making them have typeNever).This can only be used in the return type of a type decorator or as the return type of
__init_subclass__.
One snag here: it introduces type-evaluation-order dependence; if the
UpdateClass return type for some __init_subclass__ inspects
some unrelated class’s Members , and that class also has an
__init_subclass__, then the results might depend on what order they
are evaluated.
This does actually exactly mirror a potential runtime evaluation-order dependence, though.
Lifting over Unions
Many of the builtin operations are “lifted” over Union.
For example:
Concat[Literal['a'] | Literal['b'], Literal['c'] | Literal['d']] = (
Literal['ac'] | Literal['ad'] | Literal['bc'] | Literal['bd']
)
TODO: EXPLAIN
Runtime evaluation support
Rationale
Extended Callables
We need extended callable support, in order to inspect and produce callables via type-level computation. mypy supports extended callables but they are deprecated in favor of callback protocols.
Unfortunately callback protocols don’t work well for type level computation. (They probably could be made to work, but it would require a separate facility for creating and introspecting methods, which wouldn’t be any simpler.)
- I am proposing a fully new extended callable syntax because:
- The
mypy_extensionsfunctions are full no-ops, and we need real runtime objects - They use parentheses and not brackets, which really goes against the philosophy here.
- We can make an API that more nicely matches what we are going to
do for inspecting members (We could introduce extended callables that
closely mimic the
mypy_extensionsversion though, if something new is a non starter)
- The
Backwards Compatibility
[Describe potential impact and severity on pre-existing code.]
Security Implications
None are expected.
How to Teach This
I think some inspiration can be taken from how TypeScript teaches their equivalent features.
(Though not complete inspiration—some important subtleties of things like mapped types are unmentioned in current documentation (“homomorphic mappings”).)
Reference Implementation
[Link to any existing implementation and details about its state, e.g. proof-of-concept.]
Rejected Ideas
Renounce all cares of runtime evaluation
This would have a lot of simplifying features.
We wouldn’t need to worry about making IsSub be checkable at
runtime,
XXX
Support TypeScript style pattern matching in subtype checking
This would almost certainly only be possible if we also decide not to care about runtime evaluation, as above.
Use type operators for conditional and iteration
- Instead of writing:
tt if tb else tf*[tres for T in Iter[ttuple]]
- we could use type operator forms like:
Cond[tb, tt, tf]UnpackMap[ttuple, lambda T: tres]- or
UnpackMap[ttuple, T, tres]whereTmust be a declaredTypeVar
Boolean operations would likewise become operators (Not, And,
etc).
The advantage of this is that constructing a type annotation never needs to do non-trivial computation, and thus we don’t need runtime hooks to support evaluating them.
It would also mean that it would be much easier to extract the raw
type annotation. (The lambda form would still be somewhat fiddly.
The non-lambda form would be trivial to extract, but requiring the
declaration of a TypeVar goes against the grain of recent
changes.)
Another advantage is not needing any notion of a special
<type-bool> class of types.
The disadvantage is that is that the syntax seems a lot worse. Supporting filtering while mapping would make it even more bad (maybe an extra argument for a filter?).
We can explore other options too if needed.
Make the type-level operations more “strictly-typed”
This proposal is less “strictly-typed” than typescript (strictly-kinded, maybe?).
Typescript has better typechecking at the alias definition site:
For P[K], K needs to have keyof P…
We could do potentially better but it would require more meachinery.
KeyOf[T]- literal keys ofTMember[T], when statically checking a type alias, could be treated as having some type liketuple[Member[KeyOf[T], object, str, ..., ...], ...]GetAttr[T, S: KeyOf[T]]- but this isn’t supported yet. TS supports it.- We would also need to do context sensitive type bound inference
Open Issues
- Should we support building new nominal types??
- What invalid operations should be errors and what should return
Never?
[Any points that are still being decided/discussed.]
Acknowledgements
Jukka Lehtosalo
[Thank anyone who has helped with the PEP.]
Copyright
This document is placed in the public domain or under the CC0-1.0-Universal license, whichever is more permissive.
Source: https://github.com/python/peps/blob/main/peps/pep-9999.rst
Last modified: GMT