Cache directory collection when multiple file args share a parent

changelog/15187.bugfix.rst

@@ -0,0 +1,3 @@
+Directory collection is now cached when multiple file path arguments share a
+parent directory (unless ``--keep-duplicates`` is set), avoiding repeated
+directory scans during collection.

src/_pytest/main.py

@@ -992,11 +992,21 @@ def collect(self) -> Iterator[nodes.Item | nodes.Collector]:
                 else:
                     # For backward compat, files given directly multiple
                     # times on the command line should not be deduplicated.
-                    handle_dupes = not (
+                    # That intent applies to File nodes (handled in genitems);
+                    # Directory/Package parents must still use the collection
+                    # cache, otherwise N file args from the same directory
+                    # re-walk the directory N times (#15187). Keep the cache
+                    # off under --keep-duplicates so a file named twice still
+                    # gets two Module nodes.
+                    file_arg_leaf = (
                         len(matchparts) == 1
                         and isinstance(matchparts[0], Path)
                         and matchparts[0].is_file()
                     )
+                    handle_dupes = (
+                        isinstance(matchnode, nodes.Directory)
+                        and not self.config.option.keepduplicates
+                    ) or not file_arg_leaf
                     rep, duplicate = self._collect_one_node(matchnode, handle_dupes)
                     if not duplicate and not rep.passed:
                         # Report collection failures here to avoid failing to

testing/test_collection.py

@@ -2807,3 +2807,32 @@ def test1(x):
             "Duplicates:     a",
         ]
     )
+
+
+def test_directory_collection_cached_for_multiple_file_args(pytester: Pytester) -> None:
+    """File args from one directory must not re-collect that directory each time.
+
+    See https://github.com/pytest-dev/pytest/issues/15187
+    """
+    calls: dict[str, int] = {}
+
+    class Plugin:
+        def pytest_collect_file(self, file_path: Path) -> None:
+            calls[file_path.name] = calls.get(file_path.name, 0) + 1
+
+    for i in range(5):
+        pytester.makepyfile(**{f"test_{i:04d}": "def test_x():\n    pass\n"})
+
+    files = [pytester.path / f"test_{i:04d}.py" for i in range(5)]
+    result = pytester.runpytest(
+        "--collect-only",
+        "-q",
+        "-p",
+        "no:cacheprovider",
+        *map(str, files),
+        plugins=[Plugin()],
+    )
+    assert result.ret == 0
+    # Each file offered once (plus possibly non-test files; assert the repeated ones)
+    for i in range(5):
+        assert calls.get(f"test_{i:04d}.py", 0) == 1, calls