gh-39615: Add warnings.warn() skip_file_prefixes support (#100840)

`warnings.warn()` gains the ability to skip stack frames based on code
filename prefix rather than only a numeric `stacklevel=` via a new
`skip_file_prefixes=` keyword argument.

Author: gpshead

Doc/library/warnings.rst

@@ -396,7 +396,7 @@ Available Functions
 -------------------
 
 
-.. function:: warn(message, category=None, stacklevel=1, source=None)
+.. function:: warn(message, category=None, stacklevel=1, source=None, \*, skip_file_prefixes=None)
 
    Issue a warning, or maybe ignore it or raise an exception.  The *category*
    argument, if given, must be a :ref:`warning category class <warning-categories>`; it
@@ -407,19 +407,49 @@ Available Functions
    :ref:`warnings filter <warning-filter>`.  The *stacklevel* argument can be used by wrapper
    functions written in Python, like this::
 
-      def deprecation(message):
+      def deprecated_api(message):
           warnings.warn(message, DeprecationWarning, stacklevel=2)
 
-   This makes the warning refer to :func:`deprecation`'s caller, rather than to the
-   source of :func:`deprecation` itself (since the latter would defeat the purpose
-   of the warning message).
+   This makes the warning refer to ``deprecated_api``'s caller, rather than to
+   the source of ``deprecated_api`` itself (since the latter would defeat the
+   purpose of the warning message).
+
+   The *skip_file_prefixes* keyword argument can be used to indicate which
+   stack frames are ignored when counting stack levels. This can be useful when
+   you want the warning to always appear at call sites outside of a package
+   when a constant *stacklevel* does not fit all call paths or is otherwise
+   challenging to maintain. If supplied, it must be a tuple of strings. When
+   prefixes are supplied, stacklevel is implicitly overridden to be ``max(2,
+   stacklevel)``. To cause a warning to be attributed to the caller from
+   outside of the current package you might write::
+
+      # example/lower.py
+      _warn_skips = (os.path.dirname(__file__),)
+
+      def one_way(r_luxury_yacht=None, t_wobbler_mangrove=None):
+          if r_luxury_yacht:
+              warnings.warn("Please migrate to t_wobbler_mangrove=.",
+                            skip_file_prefixes=_warn_skips)
+
+      # example/higher.py
+      from . import lower
+
+      def another_way(**kw):
+          lower.one_way(**kw)
+
+   This makes the warning refer to both the ``example.lower.one_way()`` and
+   ``package.higher.another_way()`` call sites only from calling code living
+   outside of ``example`` package.
 
    *source*, if supplied, is the destroyed object which emitted a
    :exc:`ResourceWarning`.
 
    .. versionchanged:: 3.6
       Added *source* parameter.
 
+   .. versionchanged:: 3.12
+      Added *skip_file_prefixes*.
+
 
 .. function:: warn_explicit(message, category, filename, lineno, module=None, registry=None, module_globals=None, source=None)
 

Include/internal/pycore_global_objects_fini_generated.h

@@ -637,6 +637,7 @@ struct _Py_global_strings {
         STRUCT_FOR_ID(signed)
         STRUCT_FOR_ID(size)
         STRUCT_FOR_ID(sizehint)
+        STRUCT_FOR_ID(skip_file_prefixes)
         STRUCT_FOR_ID(sleep)
         STRUCT_FOR_ID(sock)
         STRUCT_FOR_ID(sort)

Include/internal/pycore_global_strings.h

@@ -12,6 +12,7 @@
 from test.support import warnings_helper
 from test.support.script_helper import assert_python_ok, assert_python_failure
 
+from test.test_warnings.data import package_helper
 from test.test_warnings.data import stacklevel as warning_tests
 
 import warnings as original_warnings
@@ -472,6 +473,42 @@ def test_stacklevel_import(self):
                 self.assertEqual(len(w), 1)
                 self.assertEqual(w[0].filename, __file__)
 
+    def test_skip_file_prefixes(self):
+        with warnings_state(self.module):
+            with original_warnings.catch_warnings(record=True,
+                    module=self.module) as w:
+                self.module.simplefilter('always')
+
+                # Warning never attributed to the data/ package.
+                package_helper.inner_api(
+                        "inner_api", stacklevel=2,
+                        warnings_module=warning_tests.warnings)
+                self.assertEqual(w[-1].filename, __file__)
+                warning_tests.package("package api", stacklevel=2)
+                self.assertEqual(w[-1].filename, __file__)
+                self.assertEqual(w[-2].filename, w[-1].filename)
+                # Low stacklevels are overridden to 2 behavior.
+                warning_tests.package("package api 1", stacklevel=1)
+                self.assertEqual(w[-1].filename, __file__)
+                warning_tests.package("package api 0", stacklevel=0)
+                self.assertEqual(w[-1].filename, __file__)
+                warning_tests.package("package api -99", stacklevel=-99)
+                self.assertEqual(w[-1].filename, __file__)
+
+                # The stacklevel still goes up out of the package.
+                warning_tests.package("prefix02", stacklevel=3)
+                self.assertIn("unittest", w[-1].filename)
+
+    def test_skip_file_prefixes_type_errors(self):
+        with warnings_state(self.module):
+            warn = warning_tests.warnings.warn
+            with self.assertRaises(TypeError):
+                warn("msg", skip_file_prefixes=[])
+            with self.assertRaises(TypeError):
+                warn("msg", skip_file_prefixes=(b"bytes",))
+            with self.assertRaises(TypeError):
+                warn("msg", skip_file_prefixes="a sequence of strs")
+
     def test_exec_filename(self):
         filename = "<warnings-test>"
         codeobj = compile(("import warnings\n"
@@ -895,7 +932,7 @@ def test_formatwarning(self):
         message = "msg"
         category = Warning
         file_name = os.path.splitext(warning_tests.__file__)[0] + '.py'
-        line_num = 3
+        line_num = 5
         file_line = linecache.getline(file_name, line_num).strip()
         format = "%s:%s: %s: %s\n  %s\n"
         expect = format % (file_name, line_num, category.__name__, message,

Include/internal/pycore_runtime_init_generated.h

@@ -0,0 +1,10 @@
+# helper to the helper for testing skip_file_prefixes.
+
+import os
+
+package_path = os.path.dirname(__file__)
+
+def inner_api(message, *, stacklevel, warnings_module):
+    warnings_module.warn(
+            message, stacklevel=stacklevel,
+            skip_file_prefixes=(package_path,))

Include/internal/pycore_unicodeobject_generated.h

@@ -1,9 +1,15 @@
-# Helper module for testing the skipmodules argument of warnings.warn()
+# Helper module for testing stacklevel and skip_file_prefixes arguments
+# of warnings.warn()
 
 import warnings
+from test.test_warnings.data import package_helper
 
 def outer(message, stacklevel=1):
     inner(message, stacklevel)
 
 def inner(message, stacklevel=1):
     warnings.warn(message, stacklevel=stacklevel)
+
+def package(message, *, stacklevel):
+    package_helper.inner_api(message, stacklevel=stacklevel,
+                             warnings_module=warnings)

Lib/test/test_warnings/__init__.py

@@ -269,22 +269,32 @@ def _getcategory(category):
     return cat
 
 
+def _is_internal_filename(filename):
+    return 'importlib' in filename and '_bootstrap' in filename
+
+
+def _is_filename_to_skip(filename, skip_file_prefixes):
+    return any(filename.startswith(prefix) for prefix in skip_file_prefixes)
+
+
 def _is_internal_frame(frame):
     """Signal whether the frame is an internal CPython implementation detail."""
-    filename = frame.f_code.co_filename
-    return 'importlib' in filename and '_bootstrap' in filename
+    return _is_internal_filename(frame.f_code.co_filename)
 
 
-def _next_external_frame(frame):
-    """Find the next frame that doesn't involve CPython internals."""
+def _next_external_frame(frame, skip_file_prefixes):
+    """Find the next frame that doesn't involve Python or user internals."""
     frame = frame.f_back
-    while frame is not None and _is_internal_frame(frame):
+    while frame is not None and (
+            _is_internal_filename(filename := frame.f_code.co_filename) or
+            _is_filename_to_skip(filename, skip_file_prefixes)):
         frame = frame.f_back
     return frame
 
 
 # Code typically replaced by _warnings
-def warn(message, category=None, stacklevel=1, source=None):
+def warn(message, category=None, stacklevel=1, source=None,
+         *, skip_file_prefixes=()):
     """Issue a warning, or maybe ignore it or raise an exception."""
     # Check if message is already a Warning object
     if isinstance(message, Warning):
@@ -295,6 +305,11 @@ def warn(message, category=None, stacklevel=1, source=None):
     if not (isinstance(category, type) and issubclass(category, Warning)):
         raise TypeError("category must be a Warning subclass, "
                         "not '{:s}'".format(type(category).__name__))
+    if not isinstance(skip_file_prefixes, tuple):
+        # The C version demands a tuple for implementation performance.
+        raise TypeError('skip_file_prefixes must be a tuple of strs.')
+    if skip_file_prefixes:
+        stacklevel = max(2, stacklevel)
     # Get context information
     try:
         if stacklevel <= 1 or _is_internal_frame(sys._getframe(1)):
@@ -305,7 +320,7 @@ def warn(message, category=None, stacklevel=1, source=None):
             frame = sys._getframe(1)
             # Look for one frame less since the above line starts us off.
             for x in range(stacklevel-1):
-                frame = _next_external_frame(frame)
+                frame = _next_external_frame(frame, skip_file_prefixes)
                 if frame is None:
                     raise ValueError
     except ValueError:

Lib/test/test_warnings/data/package_helper.py

@@ -0,0 +1,3 @@
+:func:`warnings.warn` now has the ability to skip stack frames based on code
+filename prefix rather than only a numeric ``stacklevel`` via the new
+``skip_file_prefixes`` keyword argument.