When using TypeAlias, does the interpreter always substitute the type for the TypeAlias?

[hunterhogan]

This is what I see when I use a TypeAlias.

from typing import TypeAlias type alphaKeyword = float betaAssignment: TypeAlias = float chicken: alphaKeyword = 8.9 silky = betaAssignment(21.56) print(type(chicken)) print(type(silky)) <class 'float'> <class 'float'>

I'm all but certain that is working exactly as intended. But I want to make sure I didn't miss something. When I use a TypeAlias, I would like to be able to programmatically get the TypeAlias later.

One way I work around it.

from https://github.com/hunterhogan/mapFolding

from mapFolding import ( Array1DElephino, Array1DLeavesTotal, Array3DLeavesTotal, DatatypeElephino, DatatypeFoldsTotal, DatatypeLeavesTotal, getConnectionGraph, getLeavesTotal, makeDataContainer) import dataclasses @dataclasses.dataclass(slots=True) class MapFoldingState: mapShape: tuple[DatatypeLeavesTotal, ...] = dataclasses.field(init=True, metadata={'elementConstructor': 'DatatypeLeavesTotal'}) groupsOfFolds: DatatypeFoldsTotal = dataclasses.field(default=DatatypeFoldsTotal(0), metadata={'theCountingIdentifier': True}) gap1ndex: DatatypeElephino = DatatypeElephino(0) gap1ndexCeiling: DatatypeElephino = DatatypeElephino(0) indexDimension: DatatypeLeavesTotal = DatatypeLeavesTotal(0) indexLeaf: DatatypeLeavesTotal = DatatypeLeavesTotal(0) indexMiniGap: DatatypeElephino = DatatypeElephino(0) leaf1ndex: DatatypeLeavesTotal = DatatypeLeavesTotal(1) leafConnectee: DatatypeLeavesTotal = DatatypeLeavesTotal(0) dimensionsUnconstrained: DatatypeLeavesTotal = dataclasses.field(default=None, init=True) countDimensionsGapped: Array1DLeavesTotal = dataclasses.field(default=None, init=True, metadata={'dtype': Array1DLeavesTotal.__args__[1].__args__[0]}) gapRangeStart: Array1DElephino = dataclasses.field(default=None, init=True, metadata={'dtype': Array1DElephino.__args__[1].__args__[0]}) gapsWhere: Array1DLeavesTotal = dataclasses.field(default=None, init=True, metadata={'dtype': Array1DLeavesTotal.__args__[1].__args__[0]}) leafAbove: Array1DLeavesTotal = dataclasses.field(default=None, init=True, metadata={'dtype': Array1DLeavesTotal.__args__[1].__args__[0]}) leafBelow: Array1DLeavesTotal = dataclasses.field(default=None, init=True, metadata={'dtype': Array1DLeavesTotal.__args__[1].__args__[0]}) connectionGraph: Array3DLeavesTotal = dataclasses.field(init=False, metadata={'dtype': Array3DLeavesTotal.__args__[1].__args__[0]}) dimensionsTotal: DatatypeLeavesTotal = dataclasses.field(init=False) leavesTotal: DatatypeLeavesTotal = dataclasses.field(init=False) @property def foldsTotal(self) -> DatatypeFoldsTotal: return DatatypeFoldsTotal(self.leavesTotal) * self.groupsOfFolds def __post_init__(self) -> None: self.dimensionsTotal = DatatypeLeavesTotal(len(self.mapShape)) self.leavesTotal = DatatypeLeavesTotal(getLeavesTotal(self.mapShape)) leavesTotalAsInt = int(self.leavesTotal) self.connectionGraph = getConnectionGraph(self.mapShape, leavesTotalAsInt, self.__dataclass_fields__['connectionGraph'].metadata['dtype']) if self.dimensionsUnconstrained is None: self.dimensionsUnconstrained = DatatypeLeavesTotal(int(self.dimensionsTotal)) if self.gapsWhere is None: self.gapsWhere = makeDataContainer(leavesTotalAsInt * leavesTotalAsInt + 1, self.__dataclass_fields__['gapsWhere'].metadata['dtype']) if self.countDimensionsGapped is None: self.countDimensionsGapped = makeDataContainer(leavesTotalAsInt + 1, self.__dataclass_fields__['countDimensionsGapped'].metadata['dtype']) if self.gapRangeStart is None: self.gapRangeStart = makeDataContainer(leavesTotalAsInt + 1, self.__dataclass_fields__['gapRangeStart'].metadata['dtype']) if self.leafAbove is None: self.leafAbove = makeDataContainer(leavesTotalAsInt + 1, self.__dataclass_fields__['leafAbove'].metadata['dtype']) if self.leafBelow is None: self.leafBelow = makeDataContainer(leavesTotalAsInt + 1, self.__dataclass_fields__['leafBelow'].metadata['dtype'])

[Daraan]

It's not entirely clear to me what exactly you want to achieve. I give you the broad rundown

The usage of var: TypeAlias = expression tells your type-checker to interpret the assignment as a type expression. One of the most important cases is to distinguish strings vs type expressions written as strings.

Foo = "int | str" # interpreted as string Bar: TypeAlias = "int | str" # interpreted as a Union type expression

The most crucial runtime difference is that the type statement creates a TypeAliasType, while : TypeAlias is basically just like any other variable assignment.

print(Bar is "int | str") # true type Baz = "int | str" print(Baz is "int | str") # false; but equivalent 

---

That means in your case betaAssignment is identical to float, it is a variable to which you assigned float. alphaAssignment is only typing equivalent to float, programmatically it is a different type.
For that reason if you want to know programmatically if silky is a betaAssignment or a float you cannot do that as you cannot distinguish between betaAssignment(21.56) and float(21.56). You either need to parse your code similar to what type-checkers do or add silky: your_type and check the annotations of the module or class.

From your dataclass example I am not entirely sure what you want to achieve, or what to focus on. Save the dtype as metadata? Or do you want to avoid this and get Array1DLeavesTotal from the gapsWhere: Array1DLeavesTotal = whatever assignment?

Generally a good was is to use typing.get_type_hints or typing_extensions.get_annotations

from __future__ import annotations from dataclasses import dataclass from typing import TypeAlias, get_type_hints # or use typing_extensions from typing_extensions import get_annotations # 3.14 upcomming annotationlib / inspect.get_annotations type alphaKeyword = float betaAssignment: TypeAlias = float @dataclass class Data: a: alphaKeyword b: betaAssignment @dataclass class Data2(Data): c: float d: "float" # Instance vs class difference: print(get_type_hints(Data2)) # {'a': alphaKeyword, 'b': float, 'c': float, 'd': float} print(get_type_hints(Data2(1, 2, 3, 4))) # {'c': float, , 'd': float} print(get_annotations(Data)) # {'a': alphaKeyword, 'b': float} print(get_annotations(Data2)) # {'c': <class 'float'>, 'd': 'float'} 

Here you see with get_type_hints you will notice betaAssignment is not present as it is a float, also notice that the string "float" was evaluated, for more details about the necessary namespaces refer to the documentation. With get_annotations you get the string back.

Note that this behavior changes when from __future__ import annotations is used as the type hints are then stringified for the annotations:
Your output will be:

# get_type_hints evaluated: {'a': alphaKeyword, 'b': <class 'float'>, 'c': <class 'float'>, 'd': <class 'float'>} {'c': <class 'float'>, 'd': <class 'float'>} # get_annotations as strings {'a': 'alphaKeyword', 'b': 'betaAssignment'} {'c': 'float', 'd': "'float'"}

Similarly you can also apply it to modules:

# foo.py a: int b: float type C = "int | str" D = "int | str" c: "C" d: D # no annotation for cases below: I = int i = I(2) from typing import NewType I2 = NewType("I2", int) i2 = I2(3)

import foo from typing_extensions import get_type_hints, get_annotations get_type_hints(foo) # {'a': int, 'b': float, 'c': C, 'd': int | str} get_annotations(foo) {'a': int, 'b': float, 'c': 'C', 'd': 'int | str'}

I hope that gave you some insights and the tools you need to achieve what you need.

[hunterhogan]

Thank you! That is a fantastic answer.

I hope that gave you some insights and the tools you need to achieve what you need.

Yes, for sure. This will make things far easier than I thought possible. I'm kind of in shock right now. Wow, thank you!