mypy_test.py: Move type-checking of our tests and scripts into a different test (#8587)

AlexWaygood · web-flow · commit 875f0ca7fcc6 · 2022-08-22T17:16:03.000+01:00
diff --git a/.github/workflows/tests.yml b/.github/workflows/tests.yml
@@ -61,7 +61,7 @@ jobs:
       - run: ./tests/pytype_test.py --print-stderr
 
   mypy:
-    name: Run mypy against typeshed
+    name: Test the stubs with mypy
     runs-on: ubuntu-latest
     strategy:
       matrix:
@@ -77,7 +77,7 @@ jobs:
       - run: ./tests/mypy_test.py --platform=${{ matrix.platform }} --python-version=${{ matrix.python-version }}
 
   pyright:
-    name: Run pyright against typeshed
+    name: Test the stubs with pyright
     runs-on: ubuntu-latest
     strategy:
       matrix:
diff --git a/.github/workflows/typecheck_typeshed_code.yml b/.github/workflows/typecheck_typeshed_code.yml
@@ -0,0 +1,36 @@
+name: Typecheck-typeshed-code
+
+on:
+  workflow_dispatch:
+  push:
+    branches:
+      - main
+      - master
+  pull_request:
+    paths:
+      - 'scripts/**'
+      - 'tests/**'
+      - '.github/workflows/typecheck_typeshed_code.yml'
+      - 'requirements-tests.txt'
+
+permissions:
+  contents: read
+
+env:
+  PIP_DISABLE_PIP_VERSION_CHECK: 1
+
+jobs:
+  mypy:
+    name: Run mypy against the scripts and tests directories
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        platform: ["linux", "win32"]
+      fail-fast: false
+    steps:
+      - uses: actions/checkout@v3
+      - uses: actions/setup-python@v4
+        with:
+          python-version: 3.9
+      - run: pip install -r requirements-tests.txt
+      - run: python ./tests/typecheck_typeshed.py --platform=${{ matrix.platform }}
diff --git a/tests/README.md b/tests/README.md
@@ -1,16 +1,18 @@
 This directory contains several tests:
 - `tests/mypy_test.py`
-tests typeshed with [mypy](https://github.com/python/mypy/)
-- `tests/pytype_test.py` tests typeshed with
+tests the stubs with [mypy](https://github.com/python/mypy/)
+- `tests/pytype_test.py` tests the stubs with
 [pytype](https://github.com/google/pytype/).
-- `tests/pyright_test.py` tests typeshed with
+- `tests/pyright_test.py` tests the stubs with
 [pyright](https://github.com/microsoft/pyright).
 - `tests/check_consistent.py` checks certain files in typeshed remain
 consistent with each other.
 - `tests/stubtest_stdlib.py` checks standard library stubs against the
 objects at runtime.
 - `tests/stubtest_third_party.py` checks third-party stubs against the
 objects at runtime.
+- `tests/typecheck_typeshed.py` runs mypy against typeshed's own code
+in the `tests` and `scripts` directories.
 
 To run the tests, follow the [setup instructions](../CONTRIBUTING.md#preparing-the-environment)
 in the `CONTRIBUTING.md` document. In particular, we recommend running with Python 3.9+.
@@ -22,10 +24,9 @@ Run using:
 (.venv3)$ python3 tests/mypy_test.py
 ```
 
-The test has four parts. Each part uses mypy with slightly different configuration options:
+The test has three parts. Each part uses mypy with slightly different configuration options:
 - Running mypy on the stdlib stubs
 - Running mypy on the third-party stubs
-- Running mypy `--strict` on the scripts in the `tests` directory
 - Running mypy `--strict` on the regression tests in the `test_cases` directory.
 
 When running mypy on the stubs, this test is shallow — it verifies that all stubs can be
@@ -118,3 +119,14 @@ check on the command line:
 For each distribution, stubtest ignores definitions listed in a `@tests/stubtest_allowlist.txt` file,
 relative to the distribution. Additional packages that are needed to run stubtest for a
 distribution can be added to `@tests/requirements-stubtest.txt`.
+
+## typecheck\_typeshed.py
+
+Run using
+```
+(.venv3)$ python3 tests/typecheck_typeshed.py
+```
+
+This is a small wrapper script that uses mypy to typecheck typeshed's own code in the
+`scripts` and `tests` directories. Run `python tests/typecheck_typeshed.py --help` for
+information on the various configuration options.
diff --git a/tests/mypy_test.py b/tests/mypy_test.py
@@ -29,11 +29,14 @@
 from colors import colored, print_error, print_success_msg
 
 SUPPORTED_VERSIONS = [(3, 11), (3, 10), (3, 9), (3, 8), (3, 7)]
-SUPPORTED_PLATFORMS = frozenset({"linux", "win32", "darwin"})
-TYPESHED_DIRECTORIES = frozenset({"stdlib", "stubs", "tests", "test_cases", "scripts"})
+SUPPORTED_PLATFORMS = ("linux", "win32", "darwin")
+TYPESHED_DIRECTORIES = frozenset({"stdlib", "stubs", "test_cases"})
 
+ReturnCode: TypeAlias = int
 MajorVersion: TypeAlias = int
 MinorVersion: TypeAlias = int
+MinVersion: TypeAlias = tuple[MajorVersion, MinorVersion]
+MaxVersion: TypeAlias = tuple[MajorVersion, MinorVersion]
 Platform: TypeAlias = Annotated[str, "Must be one of the entries in SUPPORTED_PLATFORMS"]
 Directory: TypeAlias = Annotated[str, "Must be one of the entries in TYPESHED_DIRECTORIES"]
 
@@ -122,8 +125,6 @@ def match(fn: str, args: TestConfig) -> bool:
 
 
 _VERSION_LINE_RE = re.compile(r"^([a-zA-Z_][a-zA-Z0-9_.]*): ([23]\.\d{1,2})-([23]\.\d{1,2})?$")
-MinVersion: TypeAlias = tuple[MajorVersion, MinorVersion]
-MaxVersion: TypeAlias = tuple[MajorVersion, MinorVersion]
 
 
 def parse_versions(fname: StrPath) -> dict[str, tuple[MinVersion, MaxVersion]]:
@@ -210,7 +211,7 @@ def add_configuration(configurations: list[MypyDistConf], distribution: str) ->
         configurations.append(MypyDistConf(module_name, values.copy()))
 
 
-def run_mypy(args: TestConfig, configurations: list[MypyDistConf], files: list[str], *, custom_typeshed: bool = False) -> int:
+def run_mypy(args: TestConfig, configurations: list[MypyDistConf], files: list[str]) -> ReturnCode:
     try:
         from mypy.api import run as mypy_run
     except ImportError:
@@ -225,7 +226,7 @@ def run_mypy(args: TestConfig, configurations: list[MypyDistConf], files: list[s
                 temp.write(f"{k} = {v}\n")
         temp.flush()
 
-        flags = get_mypy_flags(args, temp.name, custom_typeshed=custom_typeshed)
+        flags = get_mypy_flags(args, temp.name)
         mypy_args = [*flags, *files]
         if args.verbose:
             print("running mypy", " ".join(mypy_args))
@@ -253,9 +254,6 @@ def run_mypy(args: TestConfig, configurations: list[MypyDistConf], files: list[s
         return exit_code
 
 
-ReturnCode: TypeAlias = int
-
-
 def run_mypy_as_subprocess(directory: StrPath, flags: Iterable[str]) -> ReturnCode:
     result = subprocess.run([sys.executable, "-m", "mypy", directory, *flags], capture_output=True)
     stdout, stderr = result.stdout, result.stderr
@@ -267,14 +265,7 @@ def run_mypy_as_subprocess(directory: StrPath, flags: Iterable[str]) -> ReturnCo
 
 
 def get_mypy_flags(
-    args: TestConfig,
-    temp_name: str | None,
-    *,
-    custom_typeshed: bool = False,
-    strict: bool = False,
-    test_suite_run: bool = False,
-    enforce_error_codes: bool = True,
-    ignore_missing_imports: bool = False,
+    args: TestConfig, temp_name: str | None, *, strict: bool = False, enforce_error_codes: bool = True
 ) -> list[str]:
     flags = [
         "--python-version",
@@ -285,27 +276,18 @@ def get_mypy_flags(
         "--no-error-summary",
         "--platform",
         args.platform,
+        "--no-site-packages",
+        "--custom-typeshed-dir",
+        os.path.dirname(os.path.dirname(__file__)),
     ]
     if strict:
         flags.append("--strict")
     else:
         flags.extend(["--no-implicit-optional", "--disallow-untyped-decorators", "--disallow-any-generics", "--strict-equality"])
     if temp_name is not None:
         flags.extend(["--config-file", temp_name])
-    if custom_typeshed:
-        # Setting custom typeshed dir prevents mypy from falling back to its bundled
-        # typeshed in case of stub deletions
-        flags.extend(["--custom-typeshed-dir", os.path.dirname(os.path.dirname(__file__))])
-    if test_suite_run:
-        flags.append("--namespace-packages")
-        if args.platform == "win32":
-            flags.extend(["--exclude", "tests/pytype_test.py"])
-    else:
-        flags.append("--no-site-packages")
     if enforce_error_codes:
         flags.extend(["--enable-error-code", "ignore-without-code"])
-    if ignore_missing_imports:
-        flags.append("--ignore-missing-imports")
     return flags
 
 
@@ -389,8 +371,8 @@ def test_stdlib(code: int, args: TestConfig) -> TestResults:
 
     if files:
         print(f"Testing stdlib ({len(files)} files)...")
-        print("Running mypy " + " ".join(get_mypy_flags(args, "/tmp/...", custom_typeshed=True)))
-        this_code = run_mypy(args, [], files, custom_typeshed=True)
+        print("Running mypy " + " ".join(get_mypy_flags(args, "/tmp/...")))
+        this_code = run_mypy(args, [], files)
         code = max(code, this_code)
 
     return TestResults(code, len(files))
@@ -414,44 +396,10 @@ def test_third_party_stubs(code: int, args: TestConfig) -> TestResults:
     return TestResults(code, files_checked)
 
 
-def test_the_test_scripts(code: int, args: TestConfig) -> TestResults:
-    files_to_test = list(Path("tests").rglob("*.py"))
-    if args.platform == "win32":
-        files_to_test.remove(Path("tests/pytype_test.py"))
-    num_test_files_to_test = len(files_to_test)
-    flags = get_mypy_flags(args, None, strict=True, test_suite_run=True)
-    print(f"Testing the test suite ({num_test_files_to_test} files)...")
-    print("Running mypy " + " ".join(flags))
-    if args.dry_run:
-        this_code = 0
-    else:
-        this_code = run_mypy_as_subprocess("tests", flags)
-    if not this_code:
-        print_success_msg()
-    code = max(code, this_code)
-    return TestResults(code, num_test_files_to_test)
-
-
-def test_scripts_directory(code: int, args: TestConfig) -> TestResults:
-    files_to_test = list(Path("scripts").rglob("*.py"))
-    num_test_files_to_test = len(files_to_test)
-    flags = get_mypy_flags(args, None, strict=True, ignore_missing_imports=True)
-    print(f"Testing the scripts directory ({num_test_files_to_test} files)...")
-    print("Running mypy " + " ".join(flags))
-    if args.dry_run:
-        this_code = 0
-    else:
-        this_code = run_mypy_as_subprocess("scripts", flags)
-    if not this_code:
-        print_success_msg()
-    code = max(code, this_code)
-    return TestResults(code, num_test_files_to_test)
-
-
 def test_the_test_cases(code: int, args: TestConfig) -> TestResults:
     test_case_files = list(map(str, Path("test_cases").rglob("*.py")))
     num_test_case_files = len(test_case_files)
-    flags = get_mypy_flags(args, None, strict=True, custom_typeshed=True, enforce_error_codes=False)
+    flags = get_mypy_flags(args, None, strict=True, enforce_error_codes=False)
     print(f"Running mypy on the test_cases directory ({num_test_case_files} files)...")
     print("Running mypy " + " ".join(flags))
     if args.dry_run:
@@ -481,21 +429,6 @@ def test_typeshed(code: int, args: TestConfig) -> TestResults:
         files_checked_this_version += third_party_files_checked
         print()
 
-    if args.minor >= 9:
-        # Run mypy against our own test suite and the scripts directory
-        #
-        # Skip this on earlier Python versions,
-        # as we're using new syntax and new functions in some test files
-        if "tests" in args.directories:
-            code, test_script_files_checked = test_the_test_scripts(code, args)
-            files_checked_this_version += test_script_files_checked
-            print()
-
-        if "scripts" in args.directories:
-            code, script_files_checked = test_scripts_directory(code, args)
-            files_checked_this_version += script_files_checked
-            print()
-
     if "test_cases" in args.directories:
         code, test_case_files_checked = test_the_test_cases(code, args)
         files_checked_this_version += test_case_files_checked
diff --git a/tests/typecheck_typeshed.py b/tests/typecheck_typeshed.py
@@ -0,0 +1,102 @@
+#!/usr/bin/env python3
+"""Run mypy on the "tests" and "scripts" directories."""
+from __future__ import annotations
+
+import argparse
+import subprocess
+import sys
+from itertools import product
+from typing_extensions import TypeAlias
+
+from colors import colored, print_error
+
+ReturnCode: TypeAlias = int
+
+SUPPORTED_PLATFORMS = ("linux", "darwin", "win32")
+SUPPORTED_VERSIONS = ("3.11", "3.10", "3.9")
+DIRECTORIES_TO_TEST = ("scripts", "tests")
+
+parser = argparse.ArgumentParser(description="Run mypy on typeshed's own code in the `scripts` and `tests` directories.")
+parser.add_argument(
+    "--platform",
+    choices=SUPPORTED_PLATFORMS,
+    nargs="*",
+    action="extend",
+    help="Run mypy for certain OS platforms (defaults to sys.platform)",
+)
+parser.add_argument(
+    "-p",
+    "--python-version",
+    choices=SUPPORTED_VERSIONS,
+    nargs="*",
+    action="extend",
+    help="Run mypy for certain Python versions (defaults to sys.version_info[:2])",
+)
+parser.add_argument(
+    "-d",
+    "--dir",
+    choices=DIRECTORIES_TO_TEST,
+    nargs="*",
+    action="extend",
+    help=f"Test only these top-level typeshed directories (defaults to {DIRECTORIES_TO_TEST!r})",
+)
+
+
+def run_mypy_as_subprocess(directory: str, platform: str, version: str) -> ReturnCode:
+    command = [
+        sys.executable,
+        "-m",
+        "mypy",
+        directory,
+        "--platform",
+        platform,
+        "--python-version",
+        version,
+        "--strict",
+        "--show-traceback",
+        "--show-error-codes",
+        "--no-error-summary",
+        "--enable-error-code",
+        "ignore-without-code",
+        "--namespace-packages",
+    ]
+    if directory == "tests":
+        if platform == "win32":
+            command.extend(["--exclude", "tests/pytype_test.py"])
+    else:
+        command.append("--ignore-missing-imports")
+    result = subprocess.run(command, capture_output=True)
+    stdout, stderr = result.stdout, result.stderr
+    if stderr:
+        print_error(stderr.decode())
+    if stdout:
+        print_error(stdout.decode())
+    return result.returncode
+
+
+def main() -> ReturnCode:
+    args = parser.parse_args()
+    directories = args.dir or DIRECTORIES_TO_TEST
+    platforms = args.platform or [sys.platform]
+    versions = args.python_version or [f"3.{sys.version_info[1]}"]
+
+    code = 0
+
+    for directory, platform, version in product(directories, platforms, versions):
+        print(f'Running "mypy --platform {platform} --python-version {version}" on the "{directory}" directory...')
+        code = max(code, run_mypy_as_subprocess(directory, platform, version))
+
+    if code:
+        print_error("Test completed with errors")
+    else:
+        print(colored("Test completed successfully!", "green"))
+    return code
+
+
+if __name__ == "__main__":
+    try:
+        code = main()
+    except KeyboardInterrupt:
+        print_error("\n\n!!!\nTest aborted due to KeyboardInterrupt\n!!!")
+        code = 1
+    raise SystemExit(code)
