Summary
On ty 0.0.43, checking a nested dictionary/list literal against a recursive union of discriminated TypedDict types takes several seconds. The minimal reproducer here takes ~2.1s on my machine; the real application code this was based off of takes ~5.5s to type check. This seems possibly related to #2026, but does not use Pydantic.
Reproducer
from typing import Literal, TypedDict
class A(TypedDict): type: Literal["a"]; children: list["Node"]
class B(TypedDict): type: Literal["b"]; children: list["Node"]
class C(TypedDict): type: Literal["c"]; children: list["Node"]
class D(TypedDict): type: Literal["d"]; children: list["Node"]
class E(TypedDict): type: Literal["e"]; children: list["Node"]
class F(TypedDict): type: Literal["f"]; children: list["Node"]
class G(TypedDict): type: Literal["g"]; children: list["Node"]
class H(TypedDict): type: Literal["h"]; children: list["Node"]
class I(TypedDict): type: Literal["i"]; children: list["Node"]
class J(TypedDict): type: Literal["j"]; children: list["Node"]
class K(TypedDict): type: Literal["k"]; children: list["Node"]
class Leaf(TypedDict): type: Literal["leaf"]; text: str
type Node = A | B | C | D | E | F | G | H | I | J | K | Leaf
value: list[Node] = [
{"type": "a", "children": [
{"type": "b", "children": [
{"type": "c", "children": [{"type": "leaf", "text": "x"}]},
{"type": "d", "children": [{"type": "leaf", "text": "y"}]},
]},
{"type": "e", "children": [
{"type": "f", "children": [{"type": "leaf", "text": "z"}]},
{"type": "g", "children": [{"type": "leaf", "text": "w"}]},
]},
]},
]Performance
The time taken scales with the number of node variants, even if value itself doesn't change. This seems to have gotten slightly slower since 0.0.38.
| Node variants |
ty 0.0.38 |
ty 0.0.43 |
| 12 |
0.58s |
2.18s |
| 16 |
1.90s |
10.24s |
| 20 |
4.16s |
26.14s |
| 24 |
8.36s |
>60s timeout |
Version
ty 0.0.43
Summary
On
ty 0.0.43, checking a nested dictionary/list literal against a recursive union of discriminatedTypedDicttypes takes several seconds. The minimal reproducer here takes ~2.1s on my machine; the real application code this was based off of takes ~5.5s to type check. This seems possibly related to #2026, but does not use Pydantic.Reproducer
Performance
The time taken scales with the number of node variants, even if
valueitself doesn't change. This seems to have gotten slightly slower since0.0.38.Version
ty 0.0.43