[Bill Schnurr (bschnurr)]

Description

Preserve abstract-class instantiation diagnostics through proven identity decorators while allowing decorators that genuinely replace a class to return concrete results.

How you figured out what to do

I traced decorated class evaluation and abstract-use diagnostics through decorators.ts, typeEvaluator.ts, and the checker. A generic decorator returning type[T] obscured the original abstract class, but type signatures alone were insufficient because replacement decorators can have the same apparent shape.

Implementation

Record class identity preservation only when the applied decorator can be proven from its body and declarations to return its class argument unchanged. Recover the underlying class for abstract-use diagnostics only in that proven case; replacements, casts, reassignments, and function-decorated lookalikes retain existing behavior.

Screenshots or GIFs are not applicable because this change affects analyzer or language-service behavior and has no visual UI.

Testing

Added checker coverage and local/imported samples for abstract protocols and ABCs, stacked decorators, concrete implementations, replacement returns, casts, reassignments, and false-positive boundaries. A 565-test checker/evaluator selection passed.

Addresses

Fixes #11411

[github-actions]

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

[Heejae Chang (heejaechang)]

🔴 Pyright CLI QA — b576ee46 — verdict: red

Automated CLI checks compared this change with its base revision.
9 focused check(s) completed.

Needs review: reportAbstractUsage newly reported

The PR reports reportAbstractUsage for this example; the base does not. Is this change intended?

Reproduction and observed results

Create these files in an empty directory:

main.py

from abc import ABC, abstractmethod
from typing import reveal_type

class Base(ABC):
    @abstractmethod
    def value(self) -> int: ...

def identity(cls: type[Base]) -> type[Base]:
    return cls

@identity
class Selected(Base):
    pass

def construct() -> Base:
    return Selected()

class Concrete(Base):
    def value(self) -> int:
        return 42

Selected = Concrete
result = construct()
assert result.value() == 42
reveal_type(result)

pyrightconfig.json

{"include":["."],"pythonVersion":"3.13","typeCheckingMode":"basic"}

From that directory, run each built Pyright revision:

node <pyright-cli-entrypoint> --outputjson -p .

Base fa9b200906fadc093e862ce62ec1afa985c7233d

{
  "exit_code": 0,
  "diagnostics": [
    {
      "file": "main.py",
      "severity": "information",
      "rule": null,
      "message": "Type of \"result\" is \"Base\"",
      "range": {
        "start": [
          24,
          12
        ],
        "end": [
          24,
          18
        ]
      }
    }
  ]
}

PR b576ee46f40098c4df2bbfc424efc2e8667b2a92

{
  "exit_code": 1,
  "diagnostics": [
    {
      "file": "main.py",
      "severity": "error",
      "rule": "reportAbstractUsage",
      "message": "Cannot instantiate abstract class \"Selected\"\n  \"Base.value\" is not implemented",
      "range": {
        "start": [
          15,
          11
        ],
        "end": [
          15,
          21
        ]
      }
    },
    {
      "file": "main.py",
      "severity": "information",
      "rule": null,
      "message": "Type of \"result\" is \"Base\"",
      "range": {
        "start": [
          24,
          12
        ],
        "end": [
          24,
          18
        ]
      }
    }
  ]
}

Diagnostic ranges use zero-based line and character positions.

Needs review: reportAbstractUsage newly reported

The PR reports reportAbstractUsage for this example; the base does not. Is this change intended?

Reproduction and observed results

Create these files in an empty directory:

main.py

from abc import ABC, abstractmethod
from typing import assert_type

class Base(ABC):
    @abstractmethod
    def value(self) -> int: ...

class Concrete(Base):
    def value(self) -> int:
        return 42

class Decorator:
    def decorate(self, cls: type[Base]) -> type[Base]:
        return cls

class Materializer(Decorator):
    def decorate(self, cls: type[Base]) -> type[Base]:
        return Concrete

def get_decorator() -> Decorator:
    return Materializer()

@get_decorator().decorate
class Replaced(Base):
    pass

result = Replaced()
assert_type(result, Base)
assert result.value() == 42

pyrightconfig.json

{"include":["."],"pythonVersion":"3.13","typeCheckingMode":"basic"}

From that directory, run each built Pyright revision:

node <pyright-cli-entrypoint> --outputjson -p .

Base fa9b200906fadc093e862ce62ec1afa985c7233d

{
  "exit_code": 0,
  "diagnostics": []
}

PR b576ee46f40098c4df2bbfc424efc2e8667b2a92

{
  "exit_code": 1,
  "diagnostics": [
    {
      "file": "main.py",
      "severity": "error",
      "rule": "reportAbstractUsage",
      "message": "Cannot instantiate abstract class \"Replaced\"\n  \"Base.value\" is not implemented",
      "range": {
        "start": [
          26,
          9
        ],
        "end": [
          26,
          19
        ]
      }
    }
  ]
}

Diagnostic ranges use zero-based line and character positions.

[Bill Schnurr (bschnurr)]

/benchmark

[github-actions]

Compared candidate e715724e6a7bce1a012e6699cd90b3be075b1430 against its first parent 9bf95be56744123da06567463df1eb5d6fe81896.

View the full release history charts with the base and PR results.

Execution time history

Peak memory history

Download the chart artifact.

Type checker benchmark

🟢 No performance regressions detected.

Regression threshold: 20.0%
Statistic: median
Variance guard: >1.0s time and >100.0 MB memory

Package	Checker	Files checked	Time	Time delta	Peak memory	Memory delta	Status
ansible	pyright	583	18.219s	+1.4%	1082.0 MB	-0.5%	🟢 Pass
click	pyright	17	2.674s	+0.2%	355.0 MB	-0.1%	🟢 Pass
homeassistant	pyright	9850	78.149s	+0.0%	6060.9 MB	-0.4%	🟢 Pass
numpy	pyright	356	33.698s	+0.8%	1339.1 MB	-1.4%	🟢 Pass
pandas	pyright	1459	130.364s	-1.0%	3518.1 MB	-3.6%	🟢 Pass
pytest	pyright	243	12.422s	+0.6%	843.6 MB	-0.1%	🟢 Pass
requests	pyright	19	2.348s	-0.2%	342.7 MB	-1.0%	🟢 Pass
torch	pyright	2400	135.585s	-1.1%	4796.6 MB	-3.7%	🟢 Pass
transformers	pyright	2901	84.187s	-2.9%	4966.3 MB	-1.0%	🟢 Pass

Pyright stats

Package	Parsed/bound	Checked	Find	Read	Tokenize	Parse	Imports	Bind	Check	Cycles
ansible	989	583	0.040s	0.087s	0.390s	0.740s	0.203s	0.997s	14.713s	0.000s
click	111	17	0.000s	0.013s	0.103s	0.180s	0.030s	0.237s	1.817s	0.000s
homeassistant	11147	9850	0.213s	0.503s	1.767s	2.997s	1.360s	4.030s	65.750s	0.000s
numpy	603	356	0.010s	0.057s	0.343s	0.720s	0.250s	0.807s	30.937s	0.000s
pandas	1895	1459	0.060s	0.213s	1.140s	2.460s	0.283s	2.483s	123.190s	0.000s
pytest	542	243	0.020s	0.053s	0.347s	0.537s	0.110s	0.723s	10.210s	0.000s
requests	168	19	0.000s	0.017s	0.120s	0.223s	0.057s	0.267s	1.360s	0.000s
torch	3079	2400	0.140s	0.377s	2.010s	3.637s	0.597s	4.407s	122.887s	0.000s
transformers	3535	2901	0.060s	0.303s	1.497s	2.430s	0.670s	6.317s	74.093s	0.000s