Add support for GraalPy (#5380)

* Initial support for GraalPy

* Mark tests that currently fail on GraalPy with xfail

* Add graalpy to CI

* Limit test deps on graalpy to available binary wheels

* Skip cmake test installed_function on GraalPy

CMake won't find libpython on GraalPy, it either fails or silently picks
CPython's libpython.

* Factor out setting function docstrings into a macro

* Try to narrow down skipped tests

.github/workflows/ci.yml

@@ -39,6 +39,7 @@ jobs:
         - 'pypy-3.8'
         - 'pypy-3.9'
         - 'pypy-3.10'
+        - 'graalpy-24.1'
 
         # Items in here will either be added to the build matrix (if not
         # present), or add new keys to an existing matrix element if all the
@@ -67,6 +68,10 @@ jobs:
           # Extra ubuntu latest job
           - runs-on: ubuntu-latest
             python: '3.11'
+        exclude:
+          # The setup-python action currently doesn't have graalpy for windows
+          - python: 'graalpy-24.1'
+            runs-on: 'windows-2022'
 
 
     name: "🐍 ${{ matrix.python }} • ${{ matrix.runs-on }} • x64 ${{ matrix.args }}"

include/pybind11/cast.h

@@ -343,7 +343,7 @@ public:
 #else
             // Alternate approach for CPython: this does the same as the above, but optimized
             // using the CPython API so as to avoid an unneeded attribute lookup.
-            else if (auto *tp_as_number = src.ptr()->ob_type->tp_as_number) {
+            else if (auto *tp_as_number = Py_TYPE(src.ptr())->tp_as_number) {
                 if (PYBIND11_NB_BOOL(tp_as_number)) {
                     res = (*PYBIND11_NB_BOOL(tp_as_number))(src.ptr());
                 }

include/pybind11/detail/common.h

@@ -299,7 +299,7 @@ PYBIND11_WARNING_DISABLE_MSVC(4505)
 #    define PYBIND11_INTERNAL_NUMPY_1_ONLY_DETECTED
 #endif
 
-#if defined(PYPY_VERSION) && !defined(PYBIND11_SIMPLE_GIL_MANAGEMENT)
+#if (defined(PYPY_VERSION) || defined(GRAALVM_PYTHON)) && !defined(PYBIND11_SIMPLE_GIL_MANAGEMENT)
 #    define PYBIND11_SIMPLE_GIL_MANAGEMENT
 #endif
 
@@ -387,6 +387,20 @@ PYBIND11_WARNING_POP
 #define PYBIND11_CONCAT(first, second) first##second
 #define PYBIND11_ENSURE_INTERNALS_READY pybind11::detail::get_internals();
 
+#if !defined(GRAALVM_PYTHON)
+#    define PYBIND11_PYCFUNCTION_GET_DOC(func) ((func)->m_ml->ml_doc)
+#    define PYBIND11_PYCFUNCTION_SET_DOC(func, doc)                                               \
+        do {                                                                                      \
+            (func)->m_ml->ml_doc = (doc);                                                         \
+        } while (0)
+#else
+#    define PYBIND11_PYCFUNCTION_GET_DOC(func) (GraalPyCFunction_GetDoc((PyObject *) (func)))
+#    define PYBIND11_PYCFUNCTION_SET_DOC(func, doc)                                               \
+        do {                                                                                      \
+            GraalPyCFunction_SetDoc((PyObject *) (func), (doc));                                  \
+        } while (0)
+#endif
+
 #define PYBIND11_CHECK_PYTHON_VERSION                                                             \
     {                                                                                             \
         const char *compiled_ver                                                                  \

include/pybind11/detail/internals.h

@@ -454,7 +454,7 @@ inline void translate_local_exception(std::exception_ptr p) {
 
 inline object get_python_state_dict() {
     object state_dict;
-#if PYBIND11_INTERNALS_VERSION <= 4 || defined(PYPY_VERSION)
+#if PYBIND11_INTERNALS_VERSION <= 4 || defined(PYPY_VERSION) || defined(GRAALVM_PYTHON)
     state_dict = reinterpret_borrow<object>(PyEval_GetBuiltins());
 #else
 #    if PY_VERSION_HEX < 0x03090000
@@ -727,7 +727,8 @@ const char *c_str(Args &&...args) {
 }
 
 inline const char *get_function_record_capsule_name() {
-#if PYBIND11_INTERNALS_VERSION > 4
+    // On GraalPy, pointer equality of the names is currently not guaranteed
+#if PYBIND11_INTERNALS_VERSION > 4 && !defined(GRAALVM_PYTHON)
     return get_internals().function_record_capsule_name.c_str();
 #else
     return nullptr;

include/pybind11/detail/type_caster_base.h

@@ -459,7 +459,7 @@ PYBIND11_NOINLINE handle get_object_handle(const void *ptr, const detail::type_i
 }
 
 inline PyThreadState *get_thread_state_unchecked() {
-#if defined(PYPY_VERSION)
+#if defined(PYPY_VERSION) || defined(GRAALVM_PYTHON)
     return PyThreadState_GET();
 #elif PY_VERSION_HEX < 0x030D0000
     return _PyThreadState_UncheckedGet();

include/pybind11/eval.h

@@ -94,18 +94,18 @@ void exec(const char (&s)[N], object global = globals(), object local = object()
     eval<eval_statements>(s, std::move(global), std::move(local));
 }
 
-#if defined(PYPY_VERSION)
+#if defined(PYPY_VERSION) || defined(GRAALVM_PYTHON)
 template <eval_mode mode = eval_statements>
 object eval_file(str, object, object) {
-    pybind11_fail("eval_file not supported in PyPy3. Use eval");
+    pybind11_fail("eval_file not supported in this interpreter. Use eval");
 }
 template <eval_mode mode = eval_statements>
 object eval_file(str, object) {
-    pybind11_fail("eval_file not supported in PyPy3. Use eval");
+    pybind11_fail("eval_file not supported in this interpreter. Use eval");
 }
 template <eval_mode mode = eval_statements>
 object eval_file(str) {
-    pybind11_fail("eval_file not supported in PyPy3. Use eval");
+    pybind11_fail("eval_file not supported in this interpreter. Use eval");
 }
 #else
 template <eval_mode mode = eval_statements>

include/pybind11/pybind11.h

@@ -573,8 +573,7 @@ protected:
                 // chain.
                 chain_start = rec;
                 rec->next = chain;
-                auto rec_capsule
+                auto rec_capsule = reinterpret_borrow<capsule>(PyCFunction_GET_SELF(m_ptr));
-                    = reinterpret_borrow<capsule>(((PyCFunctionObject *) m_ptr)->m_self);
                 rec_capsule.set_pointer(unique_rec.release());
                 guarded_strdup.release();
             } else {
@@ -634,12 +633,11 @@ protected:
             }
         }
 
-        /* Install docstring */
         auto *func = (PyCFunctionObject *) m_ptr;
-        std::free(const_cast<char *>(func->m_ml->ml_doc));
         // Install docstring if it's non-empty (when at least one option is enabled)
-        func->m_ml->ml_doc
+        auto *doc = signatures.empty() ? nullptr : PYBIND11_COMPAT_STRDUP(signatures.c_str());
-            = signatures.empty() ? nullptr : PYBIND11_COMPAT_STRDUP(signatures.c_str());
+        std::free(const_cast<char *>(PYBIND11_PYCFUNCTION_GET_DOC(func)));
+        PYBIND11_PYCFUNCTION_SET_DOC(func, doc);
 
         if (rec->is_method) {
             m_ptr = PYBIND11_INSTANCE_METHOD_NEW(m_ptr, rec->scope.ptr());
@@ -2780,8 +2778,8 @@ get_type_override(const void *this_ptr, const type_info *this_type, const char *
     }
 
     /* Don't call dispatch code if invoked from overridden function.
-       Unfortunately this doesn't work on PyPy. */
+       Unfortunately this doesn't work on PyPy and GraalPy. */
-#if !defined(PYPY_VERSION)
+#if !defined(PYPY_VERSION) && !defined(GRAALVM_PYTHON)
 #    if PY_VERSION_HEX >= 0x03090000
     PyFrameObject *frame = PyThreadState_GetFrame(PyThreadState_Get());
     if (frame != nullptr) {

include/pybind11/pytypes.h

@@ -643,7 +643,7 @@ struct error_fetch_and_normalize {
 
         bool have_trace = false;
         if (m_trace) {
-#if !defined(PYPY_VERSION)
+#if !defined(PYPY_VERSION) && !defined(GRAALVM_PYTHON)
             auto *tb = reinterpret_cast<PyTracebackObject *>(m_trace.ptr());
 
             // Get the deepest trace possible.
@@ -1356,7 +1356,7 @@ inline bool PyUnicode_Check_Permissive(PyObject *o) {
 #    define PYBIND11_STR_CHECK_FUN PyUnicode_Check
 #endif
 
-inline bool PyStaticMethod_Check(PyObject *o) { return o->ob_type == &PyStaticMethod_Type; }
+inline bool PyStaticMethod_Check(PyObject *o) { return Py_TYPE(o) == &PyStaticMethod_Type; }
 
 class kwargs_proxy : public handle {
 public:

tests/conftest.py

@@ -28,8 +28,8 @@ except Exception:
 
 @pytest.fixture(scope="session", autouse=True)
 def use_multiprocessing_forkserver_on_linux():
-    if sys.platform != "linux":
+    if sys.platform != "linux" or sys.implementation.name == "graalpy":
-        # The default on Windows and macOS is "spawn": If it's not broken, don't fix it.
+        # The default on Windows, macOS and GraalPy is "spawn": If it's not broken, don't fix it.
         return
 
     # Full background: https://github.com/pybind/pybind11/issues/4105#issuecomment-1301004592

tests/env.py

@@ -12,6 +12,7 @@ WIN = sys.platform.startswith("win32") or sys.platform.startswith("cygwin")
 
 CPYTHON = platform.python_implementation() == "CPython"
 PYPY = platform.python_implementation() == "PyPy"
+GRAALPY = sys.implementation.name == "graalpy"
 PY_GIL_DISABLED = bool(sysconfig.get_config_var("Py_GIL_DISABLED"))
 
 

tests/requirements.txt

@@ -2,11 +2,12 @@
 build~=1.0; python_version>="3.8"
 numpy~=1.23.0; python_version=="3.8" and platform_python_implementation=="PyPy"
 numpy~=1.25.0; python_version=="3.9" and platform_python_implementation=='PyPy'
-numpy~=1.21.5; platform_python_implementation!="PyPy" and python_version>="3.8" and python_version<"3.10"
+numpy~=1.26.0; platform_python_implementation=="GraalVM" and sys_platform=="linux"
-numpy~=1.22.2; platform_python_implementation!="PyPy" and python_version=="3.10"
+numpy~=1.21.5; platform_python_implementation!="PyPy" and platform_python_implementation!="GraalVM" and python_version>="3.8" and python_version<"3.10"
-numpy~=1.26.0; platform_python_implementation!="PyPy" and python_version>="3.11" and python_version<"3.13"
+numpy~=1.22.2; platform_python_implementation!="PyPy" and platform_python_implementation!="GraalVM" and python_version=="3.10"
+numpy~=1.26.0; platform_python_implementation!="PyPy" and platform_python_implementation!="GraalVM" and python_version>="3.11" and python_version<"3.13"
 pytest~=7.0
 pytest-timeout
-scipy~=1.5.4; platform_python_implementation!="PyPy" and python_version<"3.10"
+scipy~=1.5.4; platform_python_implementation!="PyPy" and platform_python_implementation!="GraalVM" and python_version<"3.10"
-scipy~=1.8.0; platform_python_implementation!="PyPy" and python_version=="3.10" and sys_platform!='win32'
+scipy~=1.8.0; platform_python_implementation!="PyPy" and platform_python_implementation!="GraalVM" and python_version=="3.10" and sys_platform!='win32'
-scipy~=1.11.1; platform_python_implementation!="PyPy" and python_version>="3.11" and python_version<"3.13" and sys_platform!='win32'
+scipy~=1.11.1; platform_python_implementation!="PyPy" and platform_python_implementation!="GraalVM" and python_version>="3.11" and python_version<"3.13" and sys_platform!='win32'

tests/test_buffers.py

@@ -82,6 +82,8 @@ def test_from_python():
         for j in range(m4.cols()):
             assert m3[i, j] == m4[i, j]
 
+    if env.GRAALPY:
+        pytest.skip("ConstructorStats is incompatible with GraalPy.")
     cstats = ConstructorStats.get(m.Matrix)
     assert cstats.alive() == 1
     del m3, m4
@@ -118,6 +120,8 @@ def test_to_python():
     mat2[2, 3] = 5
     assert mat2[2, 3] == 5
 
+    if env.GRAALPY:
+        pytest.skip("ConstructorStats is incompatible with GraalPy.")
     cstats = ConstructorStats.get(m.Matrix)
     assert cstats.alive() == 1
     del mat

tests/test_call_policies.cpp

@@ -95,8 +95,8 @@ TEST_SUBMODULE(call_policies, m) {
         },
         py::call_guard<DependentGuard, CustomGuard>());
 
-#if !defined(PYPY_VERSION)
+#if !defined(PYPY_VERSION) && !defined(GRAALVM_PYTHON)
-    // `py::call_guard<py::gil_scoped_release>()` should work in PyPy as well,
+    // `py::call_guard<py::gil_scoped_release>()` should work in PyPy/GraalPy as well,
     // but it's unclear how to test it without `PyGILState_GetThisThreadState`.
     auto report_gil_status = []() {
         auto is_gil_held = false;

tests/test_call_policies.py

@@ -8,6 +8,7 @@ from pybind11_tests import call_policies as m
 
 
 @pytest.mark.xfail("env.PYPY", reason="sometimes comes out 1 off on PyPy", strict=False)
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_keep_alive_argument(capture):
     n_inst = ConstructorStats.detail_reg_inst()
     with capture:
@@ -60,6 +61,7 @@ def test_keep_alive_argument(capture):
     assert str(excinfo.value) == "Could not activate keep_alive!"
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_keep_alive_return_value(capture):
     n_inst = ConstructorStats.detail_reg_inst()
     with capture:
@@ -118,6 +120,7 @@ def test_keep_alive_return_value(capture):
 
 # https://foss.heptapod.net/pypy/pypy/-/issues/2447
 @pytest.mark.xfail("env.PYPY", reason="_PyObject_GetDictPtr is unimplemented")
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_alive_gc(capture):
     n_inst = ConstructorStats.detail_reg_inst()
     p = m.ParentGC()
@@ -137,6 +140,7 @@ def test_alive_gc(capture):
     )
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_alive_gc_derived(capture):
     class Derived(m.Parent):
         pass
@@ -159,6 +163,7 @@ def test_alive_gc_derived(capture):
     )
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_alive_gc_multi_derived(capture):
     class Derived(m.Parent, m.Child):
         def __init__(self):

tests/test_callbacks.cpp

@@ -270,7 +270,7 @@ TEST_SUBMODULE(callbacks, m) {
     m.add_object("custom_function", PyCFunction_New(custom_def, rec_capsule.ptr()));
 
     // This test requires a new ABI version to pass
-#if PYBIND11_INTERNALS_VERSION > 4
+#if PYBIND11_INTERNALS_VERSION > 4 && !defined(GRAALVM_PYTHON)
     // rec_capsule with nullptr name
     py::capsule rec_capsule2(std::malloc(1), [](void *data) { std::free(data); });
     m.add_object("custom_function2", PyCFunction_New(custom_def, rec_capsule2.ptr()));

tests/test_callbacks.py

@@ -90,6 +90,7 @@ def test_keyword_args_and_generalized_unpacking():
     )
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_lambda_closure_cleanup():
     m.test_lambda_closure_cleanup()
     cstats = m.payload_cstats()
@@ -98,6 +99,7 @@ def test_lambda_closure_cleanup():
     assert cstats.move_constructions >= 1
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_cpp_callable_cleanup():
     alive_counts = m.test_cpp_callable_cleanup()
     assert alive_counts == [0, 1, 2, 1, 2, 1, 0]

tests/test_class.py

@@ -361,7 +361,7 @@ def test_brace_initialization():
     assert b.vec == [123, 456]
 
 
-@pytest.mark.xfail("env.PYPY")
+@pytest.mark.xfail("env.PYPY or env.GRAALPY")
 def test_class_refcount():
     """Instances must correctly increase/decrease the reference count of their types (#1029)"""
     from sys import getrefcount

tests/test_cmake_build/CMakeLists.txt

@@ -55,8 +55,10 @@ possibly_uninitialized(PYTHON_MODULE_EXTENSION Python_INTERPRETER_ID)
 
 pybind11_add_build_test(subdirectory_function)
 pybind11_add_build_test(subdirectory_target)
-if("${PYTHON_MODULE_EXTENSION}" MATCHES "pypy" OR "${Python_INTERPRETER_ID}" STREQUAL "PyPy")
+if("${PYTHON_MODULE_EXTENSION}" MATCHES "pypy"
-  message(STATUS "Skipping embed test on PyPy")
+   OR "${Python_INTERPRETER_ID}" STREQUAL "PyPy"
+   OR "${PYTHON_MODULE_EXTENSION}" MATCHES "graalpy")
+  message(STATUS "Skipping embed test on PyPy or GraalPy")
 else()
   pybind11_add_build_test(subdirectory_embed)
 endif()
@@ -66,10 +68,14 @@ if(PYBIND11_INSTALL)
     mock_install ${CMAKE_COMMAND} "-DCMAKE_INSTALL_PREFIX=${pybind11_BINARY_DIR}/mock_install" -P
                  "${pybind11_BINARY_DIR}/cmake_install.cmake")
 
+  if(NOT "${PYTHON_MODULE_EXTENSION}" MATCHES "graalpy")
     pybind11_add_build_test(installed_function INSTALL)
+  endif()
   pybind11_add_build_test(installed_target INSTALL)
-  if(NOT ("${PYTHON_MODULE_EXTENSION}" MATCHES "pypy" OR "${Python_INTERPRETER_ID}" STREQUAL "PyPy"
+  if(NOT
-         ))
+     ("${PYTHON_MODULE_EXTENSION}" MATCHES "pypy"
+      OR "${Python_INTERPRETER_ID}" STREQUAL "PyPy"
+      OR "${PYTHON_MODULE_EXTENSION}" MATCHES "graalpy"))
     pybind11_add_build_test(installed_embed INSTALL)
   endif()
 endif()

tests/test_cpp_conduit.py

@@ -7,6 +7,7 @@ import exo_planet_pybind11
 import home_planet_very_lonely_traveler
 import pytest
 
+import env
 from pybind11_tests import cpp_conduit as home_planet
 
 
@@ -27,7 +28,10 @@ def test_call_cpp_conduit_success():
         home_planet.cpp_type_info_capsule_Traveler,
         b"raw_pointer_ephemeral",
     )
-    assert cap.__class__.__name__ == "PyCapsule"
+    assert cap.__class__.__name__ == "PyCapsule" or (
+        # Note: this will become unnecessary in the next GraalPy release
+        env.GRAALPY and cap.__class__.__name__ == "capsule"
+    )
 
 
 def test_call_cpp_conduit_platform_abi_id_mismatch():

tests/test_custom_type_setup.py

@@ -34,7 +34,7 @@ def gc_tester():
 
 
 # PyPy does not seem to reliably garbage collect.
-@pytest.mark.skipif("env.PYPY")
+@pytest.mark.skipif("env.PYPY or env.GRAALPY")
 def test_self_cycle(gc_tester):
     obj = m.OwnsPythonObjects()
     obj.value = obj
@@ -42,7 +42,7 @@ def test_self_cycle(gc_tester):
 
 
 # PyPy does not seem to reliably garbage collect.
-@pytest.mark.skipif("env.PYPY")
+@pytest.mark.skipif("env.PYPY or env.GRAALPY")
 def test_indirect_cycle(gc_tester):
     obj = m.OwnsPythonObjects()
     obj_list = [obj]

tests/test_eigen_matrix.py

@@ -2,6 +2,7 @@ from __future__ import annotations
 
 import pytest
 
+import env  # noqa: F401
 from pybind11_tests import ConstructorStats
 
 np = pytest.importorskip("numpy")
@@ -409,6 +410,7 @@ def assert_keeps_alive(cl, method, *args):
     assert cstats.alive() == start_with
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_eigen_keepalive():
     a = m.ReturnTester()
     cstats = ConstructorStats.get(m.ReturnTester)

tests/test_eigen_tensor.py

@@ -4,6 +4,8 @@ import sys
 
 import pytest
 
+import env  # noqa: F401
+
 np = pytest.importorskip("numpy")
 eigen_tensor = pytest.importorskip("pybind11_tests.eigen_tensor")
 submodules = [eigen_tensor.c_style, eigen_tensor.f_style]
@@ -61,6 +63,7 @@ def assert_equal_tensor_ref(mat, writeable=True, modified=None):
 
 @pytest.mark.parametrize("m", submodules)
 @pytest.mark.parametrize("member_name", ["member", "member_view"])
+@pytest.mark.skipif("env.GRAALPY", reason="Different refcounting mechanism")
 def test_reference_internal(m, member_name):
     if not hasattr(sys, "getrefcount"):
         pytest.skip("No reference counting")

tests/test_embed/CMakeLists.txt

@@ -1,8 +1,10 @@
 possibly_uninitialized(PYTHON_MODULE_EXTENSION Python_INTERPRETER_ID)
 
-if("${PYTHON_MODULE_EXTENSION}" MATCHES "pypy" OR "${Python_INTERPRETER_ID}" STREQUAL "PyPy")
+if("${PYTHON_MODULE_EXTENSION}" MATCHES "pypy"
-  message(STATUS "Skipping embed test on PyPy")
+   OR "${Python_INTERPRETER_ID}" STREQUAL "PyPy"
-  add_custom_target(cpptest) # Dummy target on PyPy. Embedding is not supported.
+   OR "${PYTHON_MODULE_EXTENSION}" MATCHES "graalpy")
+  message(STATUS "Skipping embed test on PyPy or GraalPy")
+  add_custom_target(cpptest) # Dummy target on PyPy or GraalPy. Embedding is not supported.
   set(_suppress_unused_variable_warning "${DOWNLOAD_CATCH}")
   return()
 endif()

tests/test_enum.py

@@ -3,9 +3,11 @@ from __future__ import annotations
 
 import pytest
 
+import env  # noqa: F401
 from pybind11_tests import enums as m
 
 
+@pytest.mark.xfail("env.GRAALPY", reason="TODO should get fixed on GraalPy side")
 def test_unscoped_enum():
     assert str(m.UnscopedEnum.EOne) == "UnscopedEnum.EOne"
     assert str(m.UnscopedEnum.ETwo) == "UnscopedEnum.ETwo"
@@ -193,6 +195,7 @@ def test_implicit_conversion():
     assert repr(x) == "{<EMode.EFirstMode: 1>: 3, <EMode.ESecondMode: 2>: 4}"
 
 
+@pytest.mark.xfail("env.GRAALPY", reason="TODO should get fixed on GraalPy side")
 def test_binary_operators():
     assert int(m.Flags.Read) == 4
     assert int(m.Flags.Write) == 2

tests/test_eval.py

@@ -19,7 +19,7 @@ def test_evals(capture):
     assert m.test_eval_failure()
 
 
-@pytest.mark.xfail("env.PYPY", raises=RuntimeError)
+@pytest.mark.xfail("env.PYPY or env.GRAALPY", raises=RuntimeError)
 def test_eval_file():
     filename = os.path.join(os.path.dirname(__file__), "test_eval_call.py")
     assert m.test_eval_file(filename)

tests/test_exceptions.py

@@ -201,6 +201,7 @@ def test_custom(msg):
     assert msg(excinfo.value) == "[PythonMyException7]: abc"
 
 
+@pytest.mark.xfail("env.GRAALPY", reason="TODO should get fixed on GraalPy side")
 def test_nested_throws(capture):
     """Tests nested (e.g. C++ -> Python -> C++) exception handling"""
 
@@ -369,6 +370,7 @@ def _test_flaky_exception_failure_point_init_py_3_12():
     "env.PYPY and sys.version_info[:2] < (3, 12)",
     reason="PyErr_NormalizeException Segmentation fault",
 )
+@pytest.mark.xfail("env.GRAALPY", reason="TODO should be fixed on GraalPy side")
 def test_flaky_exception_failure_point_init():
     if sys.version_info[:2] < (3, 12):
         _test_flaky_exception_failure_point_init_before_py_3_12()
@@ -376,6 +378,7 @@ def test_flaky_exception_failure_point_init():
         _test_flaky_exception_failure_point_init_py_3_12()
 
 
+@pytest.mark.xfail("env.GRAALPY", reason="TODO should be fixed on GraalPy side")
 def test_flaky_exception_failure_point_str():
     what, py_err_set_after_what = m.error_already_set_what(
         FlakyException, ("failure_point_str",)

tests/test_factory_constructors.py

@@ -4,11 +4,13 @@ import re
 
 import pytest
 
+import env  # noqa: F401
 from pybind11_tests import ConstructorStats
 from pybind11_tests import factory_constructors as m
 from pybind11_tests.factory_constructors import tag
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_init_factory_basic():
     """Tests py::init_factory() wrapper around various ways of returning the object"""
 
@@ -102,6 +104,7 @@ def test_init_factory_signature(msg):
     )
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_init_factory_casting():
     """Tests py::init_factory() wrapper with various upcasting and downcasting returns"""
 
@@ -150,6 +153,7 @@ def test_init_factory_casting():
     ]
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_init_factory_alias():
     """Tests py::init_factory() wrapper with value conversions and alias types"""
 
@@ -220,6 +224,7 @@ def test_init_factory_alias():
     ]
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_init_factory_dual():
     """Tests init factory functions with dual main/alias factory functions"""
     from pybind11_tests.factory_constructors import TestFactory7
@@ -302,6 +307,7 @@ def test_init_factory_dual():
     ]
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_no_placement_new(capture):
     """Prior to 2.2, `py::init<...>` relied on the type supporting placement
     new; this tests a class without placement new support."""
@@ -350,6 +356,7 @@ def strip_comments(s):
     return re.sub(r"\s+#.*", "", s)
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_reallocation_a(capture, msg):
     """When the constructor is overloaded, previous overloads can require a preallocated value.
     This test makes sure that such preallocated values only happen when they might be necessary,
@@ -372,6 +379,7 @@ def test_reallocation_a(capture, msg):
     )
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_reallocation_b(capture, msg):
     with capture:
         create_and_destroy(1.5)
@@ -388,6 +396,7 @@ def test_reallocation_b(capture, msg):
     )
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_reallocation_c(capture, msg):
     with capture:
         create_and_destroy(2, 3)
@@ -402,6 +411,7 @@ def test_reallocation_c(capture, msg):
     )
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_reallocation_d(capture, msg):
     with capture:
         create_and_destroy(2.5, 3)
@@ -417,6 +427,7 @@ def test_reallocation_d(capture, msg):
     )
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_reallocation_e(capture, msg):
     with capture:
         create_and_destroy(3.5, 4.5)
@@ -432,6 +443,7 @@ def test_reallocation_e(capture, msg):
     )
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_reallocation_f(capture, msg):
     with capture:
         create_and_destroy(4, 0.5)
@@ -448,6 +460,7 @@ def test_reallocation_f(capture, msg):
     )
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_reallocation_g(capture, msg):
     with capture:
         create_and_destroy(5, "hi")

tests/test_gil_scoped.py

@@ -211,6 +211,10 @@ def _run_in_threads(test_fn, num_threads, parallel):
 
 @pytest.mark.skipif(sys.platform.startswith("emscripten"), reason="Requires threads")
 @pytest.mark.parametrize("test_fn", ALL_BASIC_TESTS_PLUS_INTENTIONAL_DEADLOCK)
+@pytest.mark.skipif(
+    "env.GRAALPY",
+    reason="GraalPy transiently complains about unfinished threads at process exit",
+)
 def test_run_in_process_one_thread(test_fn):
     """Makes sure there is no GIL deadlock when running in a thread.
 
@@ -221,6 +225,10 @@ def test_run_in_process_one_thread(test_fn):
 
 @pytest.mark.skipif(sys.platform.startswith("emscripten"), reason="Requires threads")
 @pytest.mark.parametrize("test_fn", ALL_BASIC_TESTS_PLUS_INTENTIONAL_DEADLOCK)
+@pytest.mark.skipif(
+    "env.GRAALPY",
+    reason="GraalPy transiently complains about unfinished threads at process exit",
+)
 def test_run_in_process_multiple_threads_parallel(test_fn):
     """Makes sure there is no GIL deadlock when running in a thread multiple times in parallel.
 
@@ -231,6 +239,10 @@ def test_run_in_process_multiple_threads_parallel(test_fn):
 
 @pytest.mark.skipif(sys.platform.startswith("emscripten"), reason="Requires threads")
 @pytest.mark.parametrize("test_fn", ALL_BASIC_TESTS_PLUS_INTENTIONAL_DEADLOCK)
+@pytest.mark.skipif(
+    "env.GRAALPY",
+    reason="GraalPy transiently complains about unfinished threads at process exit",
+)
 def test_run_in_process_multiple_threads_sequential(test_fn):
     """Makes sure there is no GIL deadlock when running in a thread multiple times sequentially.
 
@@ -241,6 +253,10 @@ def test_run_in_process_multiple_threads_sequential(test_fn):
 
 @pytest.mark.skipif(sys.platform.startswith("emscripten"), reason="Requires threads")
 @pytest.mark.parametrize("test_fn", ALL_BASIC_TESTS_PLUS_INTENTIONAL_DEADLOCK)
+@pytest.mark.skipif(
+    "env.GRAALPY",
+    reason="GraalPy transiently complains about unfinished threads at process exit",
+)
 def test_run_in_process_direct(test_fn):
     """Makes sure there is no GIL deadlock when using processes.
 

tests/test_iostream.py

@@ -6,8 +6,14 @@ from io import StringIO
 
 import pytest
 
+import env  # noqa: F401
 from pybind11_tests import iostream as m
 
+pytestmark = pytest.mark.skipif(
+    "env.GRAALPY",
+    reason="Delayed prints from finalizers from other tests can end up in the output",
+)
+
 
 def test_captured(capsys):
     msg = "I've been redirected to Python, I hope!"

tests/test_kwargs_and_defaults.py

@@ -2,6 +2,7 @@ from __future__ import annotations
 
 import pytest
 
+import env  # noqa: F401
 from pybind11_tests import kwargs_and_defaults as m
 
 
@@ -383,6 +384,7 @@ def test_signatures():
     )
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Different refcounting mechanism")
 def test_args_refcount():
     """Issue/PR #1216 - py::args elements get double-inc_ref()ed when combined with regular
     arguments"""

tests/test_methods_and_attributes.py

@@ -4,7 +4,7 @@ import sys
 
 import pytest
 
-import env  # noqa: F401
+import env
 from pybind11_tests import ConstructorStats
 from pybind11_tests import methods_and_attributes as m
 
@@ -68,6 +68,9 @@ def test_methods_and_attributes():
     instance1.value = 100
     assert str(instance1) == "ExampleMandA[value=100]"
 
+    if env.GRAALPY:
+        pytest.skip("ConstructorStats is incompatible with GraalPy.")
+
     cstats = ConstructorStats.get(m.ExampleMandA)
     assert cstats.alive() == 2
     del instance1, instance2
@@ -316,6 +319,8 @@ def test_dynamic_attributes():
         instance.__dict__ = []
     assert str(excinfo.value) == "__dict__ must be set to a dictionary, not a 'list'"
 
+    if env.GRAALPY:
+        pytest.skip("ConstructorStats is incompatible with GraalPy.")
     cstats = ConstructorStats.get(m.DynamicClass)
     assert cstats.alive() == 1
     del instance
@@ -337,6 +342,7 @@ def test_dynamic_attributes():
 
 # https://foss.heptapod.net/pypy/pypy/-/issues/2447
 @pytest.mark.xfail("env.PYPY")
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_cyclic_gc():
     # One object references itself
     instance = m.DynamicClass()

tests/test_modules.py

@@ -39,6 +39,9 @@ def test_reference_internal():
     assert str(b.get_a2()) == "A[43]"
     assert str(b.a2) == "A[43]"
 
+    if env.GRAALPY:
+        pytest.skip("ConstructorStats is incompatible with GraalPy.")
+
     astats, bstats = ConstructorStats.get(ms.A), ConstructorStats.get(ms.B)
     assert astats.alive() == 2
     assert bstats.alive() == 1
@@ -97,9 +100,9 @@ def test_def_submodule_failures():
     sm = m.def_submodule(m, b"ScratchSubModuleName")  # Using bytes to show it works.
     assert sm.__name__ == m.__name__ + "." + "ScratchSubModuleName"
     malformed_utf8 = b"\x80"
-    if env.PYPY:
+    if env.PYPY or env.GRAALPY:
         # It is not worth the effort finding a trigger for a failure when running with PyPy.
-        pytest.skip("Sufficiently exercised on platforms other than PyPy.")
+        pytest.skip("Sufficiently exercised on platforms other than PyPy/GraalPy.")
     else:
         # Meant to trigger PyModule_GetName() failure:
         sm_name_orig = sm.__name__

tests/test_multiple_inheritance.py

@@ -2,7 +2,7 @@ from __future__ import annotations
 
 import pytest
 
-import env  # noqa: F401
+import env
 from pybind11_tests import ConstructorStats
 from pybind11_tests import multiple_inheritance as m
 
@@ -279,6 +279,7 @@ def test_mi_unaligned_base():
 
     c = m.I801C()
     d = m.I801D()
+    if not env.GRAALPY:
         # + 4 below because we have the two instances, and each instance has offset base I801B2
         assert ConstructorStats.detail_reg_inst() == n_inst + 4
     b1c = m.i801b1_c(c)
@@ -290,6 +291,9 @@ def test_mi_unaligned_base():
     b2d = m.i801b2_d(d)
     assert b2d is d
 
+    if env.GRAALPY:
+        pytest.skip("ConstructorStats is incompatible with GraalPy.")
+
     assert ConstructorStats.detail_reg_inst() == n_inst + 4  # no extra instances
     del c, b1c, b2c
     assert ConstructorStats.detail_reg_inst() == n_inst + 2
@@ -312,6 +316,7 @@ def test_mi_base_return():
     assert d1.a == 1
     assert d1.b == 2
 
+    if not env.GRAALPY:
         assert ConstructorStats.detail_reg_inst() == n_inst + 4
 
     c2 = m.i801c_b2()
@@ -324,6 +329,7 @@ def test_mi_base_return():
     assert d2.a == 1
     assert d2.b == 2
 
+    if not env.GRAALPY:
         assert ConstructorStats.detail_reg_inst() == n_inst + 8
 
         del c2

tests/test_numpy_array.py

@@ -242,6 +242,7 @@ def test_wrap():
     assert_references(a1m, a2, a1)
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_numpy_view(capture):
     with capture:
         ac = m.ArrayClass()
@@ -465,7 +466,7 @@ def test_array_resize():
     assert b.shape == (8, 8)
 
 
-@pytest.mark.xfail("env.PYPY")
+@pytest.mark.xfail("env.PYPY or env.GRAALPY")
 def test_array_create_and_resize():
     a = m.create_and_resize(2)
     assert a.size == 4

tests/test_operator_overloading.py

@@ -2,10 +2,12 @@ from __future__ import annotations
 
 import pytest
 
+import env  # noqa: F401
 from pybind11_tests import ConstructorStats
 from pybind11_tests import operators as m
 
 
+@pytest.mark.xfail("env.GRAALPY", reason="TODO should get fixed on GraalPy side")
 def test_operator_overloading():
     v1 = m.Vector2(1, 2)
     v2 = m.Vector(3, -1)
@@ -83,6 +85,7 @@ def test_operator_overloading():
     assert cstats.move_assignments == 0
 
 
+@pytest.mark.xfail("env.GRAALPY", reason="TODO should get fixed on GraalPy side")
 def test_operators_notimplemented():
     """#393: need to return NotSupported to ensure correct arithmetic operator behavior"""
 

tests/test_pickling.py

@@ -20,7 +20,7 @@ def test_pickle_simple_callable():
         # all C Python versions.
         with pytest.raises(TypeError) as excinfo:
             pickle.dumps(m.simple_callable)
-        assert re.search("can.*t pickle .*PyCapsule.* object", str(excinfo.value))
+        assert re.search("can.*t pickle .*[Cc]apsule.* object", str(excinfo.value))
 
 
 @pytest.mark.parametrize("cls_name", ["Pickleable", "PickleableNew"])

tests/test_pytypes.py

@@ -262,6 +262,7 @@ def test_str(doc):
         m.str_from_std_string_input,
     ],
 )
+@pytest.mark.xfail("env.GRAALPY", reason="TODO should be fixed on GraalPy side")
 def test_surrogate_pairs_unicode_error(func):
     input_str = "\ud83d\ude4f".encode("utf-8", "surrogatepass")
     with pytest.raises(UnicodeDecodeError):
@@ -420,6 +421,7 @@ def test_accessor_moves():
         pytest.skip("Not defined: PYBIND11_HANDLE_REF_DEBUG")
 
 
+@pytest.mark.xfail("env.GRAALPY", reason="TODO should be fixed on GraalPy side")
 def test_constructors():
     """C++ default and converting constructors are equivalent to type calls in Python"""
     types = [bytes, bytearray, str, bool, int, float, tuple, list, dict, set]
@@ -712,6 +714,7 @@ def test_pass_bytes_or_unicode_to_string_types():
             m.pass_to_pybind11_str(malformed_utf8)
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 @pytest.mark.parametrize(
     ("create_weakref", "create_weakref_with_callback"),
     [
@@ -765,7 +768,10 @@ def test_weakref_err(create_weakref, has_callback):
 
     ob = C()
     # Should raise TypeError on CPython
-    with pytest.raises(TypeError) if not env.PYPY else contextlib.nullcontext():
+    cm = pytest.raises(TypeError)
+    if env.PYPY or env.GRAALPY:
+        cm = contextlib.nullcontext()
+    with cm:
         _ = create_weakref(ob, callback) if has_callback else create_weakref(ob)
 
 

tests/test_sequences_and_iterators.py

@@ -3,6 +3,7 @@ from __future__ import annotations
 import pytest
 from pytest import approx  # noqa: PT013
 
+import env
 from pybind11_tests import ConstructorStats
 from pybind11_tests import sequences_and_iterators as m
 
@@ -110,6 +111,7 @@ def test_sequence():
     cstats = ConstructorStats.get(m.Sequence)
 
     s = m.Sequence(5)
+    if not env.GRAALPY:
         assert cstats.values() == ["of size", "5"]
 
     assert "Sequence" in repr(s)
@@ -123,15 +125,18 @@ def test_sequence():
     assert s[3] == approx(56.78, rel=1e-05)
 
     rev = reversed(s)
+    if not env.GRAALPY:
         assert cstats.values() == ["of size", "5"]
 
     rev2 = s[::-1]
+    if not env.GRAALPY:
         assert cstats.values() == ["of size", "5"]
 
     it = iter(m.Sequence(0))
     for _ in range(3):  # __next__ must continue to raise StopIteration
         with pytest.raises(StopIteration):
             next(it)
+    if not env.GRAALPY:
         assert cstats.values() == ["of size", "0"]
 
     expected = [0, 56.78, 0, 0, 12.34]
@@ -140,10 +145,14 @@ def test_sequence():
     assert rev == rev2
 
     rev[0::2] = m.Sequence([2.0, 2.0, 2.0])
+    if not env.GRAALPY:
         assert cstats.values() == ["of size", "3", "from std::vector"]
 
     assert rev == approx([2, 56.78, 2, 0, 2], rel=1e-05)
 
+    if env.GRAALPY:
+        pytest.skip("ConstructorStats is incompatible with GraalPy.")
+
     assert cstats.alive() == 4
     del it
     assert cstats.alive() == 3

tests/test_smart_ptr.py

@@ -2,10 +2,13 @@ from __future__ import annotations
 
 import pytest
 
+import env  # noqa: F401
+
 m = pytest.importorskip("pybind11_tests.smart_ptr")
 from pybind11_tests import ConstructorStats  # noqa: E402
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_smart_ptr(capture):
     # Object1
     for i, o in enumerate(
@@ -118,6 +121,7 @@ def test_smart_ptr_refcounting():
     assert m.test_object1_refcounting()
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_unique_nodelete():
     o = m.MyObject4(23)
     assert o.value == 23
@@ -129,6 +133,7 @@ def test_unique_nodelete():
     assert cstats.alive() == 0
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_unique_nodelete4a():
     o = m.MyObject4a(23)
     assert o.value == 23
@@ -140,6 +145,7 @@ def test_unique_nodelete4a():
     assert cstats.alive() == 0
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_unique_deleter():
     m.MyObject4a(0)
     o = m.MyObject4b(23)
@@ -156,6 +162,7 @@ def test_unique_deleter():
     assert cstats4b.alive() == 0
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_large_holder():
     o = m.MyObject5(5)
     assert o.value == 5
@@ -165,6 +172,7 @@ def test_large_holder():
     assert cstats.alive() == 0
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_shared_ptr_and_references():
     s = m.SharedPtrRef()
     stats = ConstructorStats.get(m.A)
@@ -196,6 +204,7 @@ def test_shared_ptr_and_references():
     assert stats.alive() == 0
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_shared_ptr_from_this_and_references():
     s = m.SharedFromThisRef()
     stats = ConstructorStats.get(m.B)
@@ -242,6 +251,7 @@ def test_shared_ptr_from_this_and_references():
     assert y is z
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_move_only_holder():
     a = m.TypeWithMoveOnlyHolder.make()
     b = m.TypeWithMoveOnlyHolder.make_as_object()
@@ -253,6 +263,7 @@ def test_move_only_holder():
     assert stats.alive() == 0
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_holder_with_addressof_operator():
     # this test must not throw exception from c++
     a = m.TypeForHolderWithAddressOf.make()
@@ -283,6 +294,7 @@ def test_holder_with_addressof_operator():
     assert stats.alive() == 0
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_move_only_holder_with_addressof_operator():
     a = m.TypeForMoveOnlyHolderWithAddressOf.make()
     a.print_object()

tests/test_stl.py

@@ -2,6 +2,7 @@ from __future__ import annotations
 
 import pytest
 
+import env  # noqa: F401
 from pybind11_tests import ConstructorStats, UserType
 from pybind11_tests import stl as m
 
@@ -362,6 +363,7 @@ def test_function_with_string_and_vector_string_arg():
     assert m.func_with_string_or_vector_string_arg_overload("A") == 3
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_stl_ownership():
     cstats = ConstructorStats.get(m.Placeholder)
     assert cstats.alive() == 0

tests/test_type_caster_pyobject_ptr.cpp

@@ -37,7 +37,8 @@ struct WithPyObjectPtrReturnTrampoline : WithPyObjectPtrReturn {
 
 std::string call_return_pyobject_ptr(const WithPyObjectPtrReturn *base_class_ptr) {
     PyObject *returned_obj = base_class_ptr->return_pyobject_ptr();
-#if !defined(PYPY_VERSION) // It is not worth the trouble doing something special for PyPy.
+// It is not worth the trouble doing something special for PyPy/GraalPy
+#if !defined(PYPY_VERSION) && !defined(GRAALVM_PYTHON)
     if (Py_REFCNT(returned_obj) != 1) {
         py::pybind11_fail(__FILE__ ":" PYBIND11_TOSTRING(__LINE__));
     }

tests/test_virtual_functions.py

@@ -4,7 +4,7 @@ import sys
 
 import pytest
 
-import env  # noqa: F401
+import env
 
 m = pytest.importorskip("pybind11_tests.virtual_functions")
 from pybind11_tests import ConstructorStats  # noqa: E402
@@ -82,6 +82,9 @@ def test_override(capture, msg):
     """
     )
 
+    if env.GRAALPY:
+        pytest.skip("ConstructorStats is incompatible with GraalPy.")
+
     cstats = ConstructorStats.get(m.ExampleVirt)
     assert cstats.alive() == 3
     del ex12, ex12p, ex12p2
@@ -91,6 +94,7 @@ def test_override(capture, msg):
     assert cstats.move_constructions >= 0
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_alias_delay_initialization1(capture):
     """`A` only initializes its trampoline class when we inherit from it
 
@@ -130,6 +134,7 @@ def test_alias_delay_initialization1(capture):
     )
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_alias_delay_initialization2(capture):
     """`A2`, unlike the above, is configured to always initialize the alias
 
@@ -188,7 +193,7 @@ def test_alias_delay_initialization2(capture):
 
 # PyPy: Reference count > 1 causes call with noncopyable instance
 # to fail in ncv1.print_nc()
-@pytest.mark.xfail("env.PYPY")
+@pytest.mark.xfail("env.PYPY or env.GRAALPY")
 @pytest.mark.skipif(
     not hasattr(m, "NCVirt"), reason="NCVirt does not work on Intel/PGI/NVCC compilers"
 )