Do not treat match value patterns as isinstance checks

[sterliakov]

As discovered by @randolf-scholz in #20142, mypy treats value patterns in match statement in a completely wrong way.

From PEP 622:

Literal pattern uses equality with literal on the right hand side, so that in the above example number == 0 and then possibly number == 1, etc will be evaluated.

Existing tests for the feature are invalid: for example,

foo: object

match foo:
    case 1:
        reveal_type(foo)

must reveal object, and test testMatchValuePatternNarrows asserts the opposite. Here's a runtime example:

>>> class A:
...     def __eq__(self,o): return True
...     
>>> match A():
...     case 1:
...         print("eq")
...         
eq

I have updated the existing tests accordingly.

The idea is that value patterns are essentially equivalent to if foo == SomeValue checks, not isinstance checks modelled by conditional_types_wit_intersection.

The original implementation was introduced in #10191.

[sterliakov]

I'm frankly not sure what part of the core team to ping here - reviewers of the original PR, @JukkaL? @randolf-scholz please also take a look if you have time, IIRC some recent PRs of yours were related to pattern matching?

Primer finds nothing, apparently match isn't that popular in wild...

[randolf-scholz]

Primer finds nothing, apparently match isn't that popular in wild...

Well, many popular projects still support 3.9, but match-case requires at least 3.10, so that's not surprising.

[randolf-scholz]

One thing that might be worth testing is if it interacts correctly with unions types and union patterns, e.g.

from typing import Literal

def test1(x: Literal[1,2,3]) -> None:
    match x:
        case 1:
            reveal_type(x)
        case other:
            reveal_type(x)

def test2(x: Literal[1,2,3]) -> None:
    match x:
        case 1:
            reveal_type(x)
        case 2:
            reveal_type(x)
        case 3:
            reveal_type(x)
        case other:
            assert_never(x)

def test3(x: Literal[1,2,3]) -> None:
    match x:
        case 1 | 3:
            reveal_type(x)
        case other:
            reveal_type(x)

and possibly the same with enum arguments / enum patterns.

[sterliakov]

I don't want to add any enum tests here, they won't be representative due to #19594 - I'd prefer to add match equivalents to tests there once this PR is merged. Unions are more relevant, thanks!

[github-actions]

According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅