Allow nativeparse to parse source code directly

misc/dump-ast.py

@@ -19,7 +19,7 @@ def dump(fname: str, python_version: tuple[int, int], quiet: bool = False) -> No
     options.python_version = python_version
     with open(fname, "rb") as f:
         s = f.read()
-        tree = parse(s, fname, None, errors=Errors(options), options=options, file_exists=True)
+        tree = parse(s, fname, None, errors=Errors(options), options=options)
         if not quiet:
             print(tree)
 

mypy/build.py

@@ -1287,13 +1287,12 @@ def parse_file(
 
         Raise CompileError if there is a parse error.
         """
-        file_exists = self.fscache.exists(path, real_only=True)
         t0 = time.time()
         if raw_data:
             # If possible, deserialize from known binary data instead of parsing from scratch.
             tree = load_from_raw(path, id, raw_data, self.errors, options)
         else:
-            tree = parse(source, path, id, self.errors, options=options, file_exists=file_exists)
+            tree = parse(source, path, id, self.errors, options=options)
         tree._fullname = id
         if self.stats_enabled:
             with self.stats_lock:

mypy/checkstrformat.py

@@ -587,7 +587,6 @@ def apply_field_accessors(
             module=None,
             options=self.chk.options,
             errors=temp_errors,
-            file_exists=False,
             eager=True,
         )
         if temp_errors.is_errors():

mypy/nativeparse.py

@@ -182,7 +182,10 @@ def add_error(
 
 
 def native_parse(
-    filename: str, options: Options, skip_function_bodies: bool = False
+    filename: str,
+    options: Options,
+    source: str | bytes | None = None,
+    skip_function_bodies: bool = False,
 ) -> tuple[MypyFile, list[ParseError], TypeIgnores]:
     """Parse a Python file using the native Rust-based parser.
 
@@ -211,7 +214,7 @@ def native_parse(
         uses_template_strings,
         source_hash,
         mypy_comments,
-    ) = parse_to_binary_ast(filename, options, skip_function_bodies)
+    ) = parse_to_binary_ast(filename, options, source, skip_function_bodies)
     node = MypyFile([], [])
     node.path = filename
     node.raw_data = FileRawData(
@@ -248,7 +251,10 @@ def read_statements(state: State, data: ReadBuffer, n: int) -> list[Statement]:
 
 
 def parse_to_binary_ast(
-    filename: str, options: Options, skip_function_bodies: bool = False
+    filename: str,
+    options: Options,
+    source: str | bytes | None = None,
+    skip_function_bodies: bool = False,
 ) -> tuple[bytes, list[ParseError], TypeIgnores, bytes, bool, bool, str, list[tuple[int, str]]]:
     # This is a horrible hack to work around a mypyc bug where imported
     # module may be not ready in a thread sometimes.
@@ -259,6 +265,7 @@ def parse_to_binary_ast(
             raise ImportError("Cannot import ast_serialize")
     ast_bytes, errors, ignores, import_bytes, ast_data = ast_serialize.parse(
         filename,
+        source,
         skip_function_bodies=skip_function_bodies,
         python_version=options.python_version,
         platform=options.platform,

mypy/parse.py

@@ -17,7 +17,6 @@ def parse(
     module: str | None,
     errors: Errors,
     options: Options,
-    file_exists: bool,
     eager: bool = False,
 ) -> MypyFile:
     """Parse a source file, without doing any semantic analysis.
@@ -29,25 +28,21 @@ def parse(
     the parse errors, use eager=True.
     """
     if options.native_parser:
-        # Native parser only works with actual files on disk
-        # Fall back to fastparse for in-memory source or non-existent files
-        if file_exists:
-            import mypy.nativeparse
-
-            ignore_errors = options.ignore_errors or fnam in errors.ignored_files
-            # If errors are ignored, we can drop many function bodies to speed up type checking.
-            strip_function_bodies = ignore_errors and not options.preserve_asts
-            tree, _, _ = mypy.nativeparse.native_parse(
-                fnam, options, skip_function_bodies=strip_function_bodies
-            )
-            # Set is_stub based on file extension
-            tree.is_stub = fnam.endswith(".pyi")
-            # Note: tree.imports is populated directly by load_from_raw() with deserialized
-            # import metadata, so we don't need to collect imports via AST traversal
-            if eager and tree.raw_data is not None:
-                tree = load_from_raw(fnam, module, tree.raw_data, errors, options)
-            return tree
-        # Fall through to fastparse for non-existent files
+        import mypy.nativeparse
+
+        ignore_errors = options.ignore_errors or fnam in errors.ignored_files
+        # If errors are ignored, we can drop many function bodies to speed up type checking.
+        strip_function_bodies = ignore_errors and not options.preserve_asts
+        tree, _, _ = mypy.nativeparse.native_parse(
+            fnam, options, source, skip_function_bodies=strip_function_bodies
+        )
+        # Set is_stub based on file extension
+        tree.is_stub = fnam.endswith(".pyi")
+        # Note: tree.imports is populated directly by load_from_raw() with deserialized
+        # import metadata, so we don't need to collect imports via AST traversal
+        if eager and tree.raw_data is not None:
+            tree = load_from_raw(fnam, module, tree.raw_data, errors, options)
+        return tree
 
     if options.transform_source is not None:
         source = options.transform_source(source)

mypy/stubgen.py

@@ -1745,13 +1745,7 @@ def parse_source_file(mod: StubSource, mypy_options: MypyOptions) -> None:
     source = mypy.util.decode_python_encoding(data)
     errors = Errors(mypy_options)
     mod.ast = mypy.parse.parse(
-        source,
-        fnam=mod.path,
-        module=mod.module,
-        errors=errors,
-        options=mypy_options,
-        file_exists=True,
-        eager=True,
+        source, fnam=mod.path, module=mod.module, errors=errors, options=mypy_options, eager=True
     )
     mod.ast._fullname = mod.module
     if errors.is_blockers():

mypy/test/test_nativeparse.py

@@ -98,7 +98,7 @@ def test_parser(testcase: DataDrivenTestCase) -> None:
 
     try:
         with temp_source(source) as fnam:
-            node, errors, type_ignores = native_parse(fnam, options, skip_function_bodies)
+            node, errors, type_ignores = native_parse(fnam, options, None, skip_function_bodies)
             errors += load_tree(node, options)
             node.path = "main"
             a = node.str_with_options(options).split("\n")
@@ -234,7 +234,7 @@ def format_reachable_imports(node: MypyFile) -> list[str]:
 
 @unittest.skipUnless(has_nativeparse, "nativeparse not available")
 class TestNativeParserBinaryFormat(unittest.TestCase):
-    def test_trivial_binary_data(self) -> None:
+    def _assert_trivial_binary_data(self, b: bytes, /) -> None:
         # A quick sanity check to ensure the serialized data looks as expected. Only covers
         # a few AST nodes.
 
@@ -250,9 +250,9 @@ def locs(start_line: int, start_column: int, end_line: int, end_column: int) ->
                 int_enc(end_column - start_column),
             ]
 
-        with temp_source("print('hello')") as fnam:
-            b, _, _, _, _, _, _, _ = parse_to_binary_ast(fnam, Options())
-            assert list(b) == (
+        self.assertEqual(
+            list(b),
+            (
                 [LITERAL_INT, 22, nodes.EXPR_STMT, nodes.CALL_EXPR]
                 + [nodes.NAME_EXPR, LITERAL_STR]
                 + [int_enc(5)]
@@ -269,7 +269,25 @@ def locs(start_line: int, start_column: int, end_line: int, end_column: int) ->
                 + [LIST_GEN, 22, LITERAL_NONE]
                 + locs(1, 0, 1, 14)
                 + [END_TAG, END_TAG]
-            )
+            ),
+        )
+
+    def test_trivial_binary_data_from_file(self) -> None:
+        with temp_source("print('hello')") as fnam:
+            b, _, _, _, _, _, _, _ = parse_to_binary_ast(fnam, Options())
+            self._assert_trivial_binary_data(b)
+
+    def test_trivial_binary_data_from_string_source(self) -> None:
+        b, _, _, _, _, _, _, _ = parse_to_binary_ast("", Options(), "print('hello')")
+        self._assert_trivial_binary_data(b)
+
+    def test_trivial_binary_data_from_bytes_source(self) -> None:
+        b, _, _, _, _, _, _, _ = parse_to_binary_ast("", Options(), b"print('hello')")
+        self._assert_trivial_binary_data(b)
+
+    def test_invalid_bytes_raises(self) -> None:
+        with self.assertRaises(UnicodeDecodeError):
+            parse_to_binary_ast("", Options(), b"\xff")
 
 
 @contextlib.contextmanager

mypy/test/testparse.py

@@ -66,7 +66,6 @@ def test_parser(testcase: DataDrivenTestCase) -> None:
             module="__main__",
             errors=errors,
             options=options,
-            file_exists=False,
             eager=True,
         )
         if errors.is_errors():
@@ -108,7 +107,6 @@ def test_parse_error(testcase: DataDrivenTestCase) -> None:
             "__main__",
             errors=errors,
             options=options,
-            file_exists=False,
             eager=True,
         )
         if errors.is_errors():