gh-111968: Use per-thread slice_cache in free-threading (gh-113972)

diff --git a/Include/internal/pycore_freelist.h b/Include/internal/pycore_freelist.h
index 34009435910d99f7d9e044a80fded83627d252b7..faa0c11a49e798ca58cd747e476afe59fd59e5dc 100644 (file)
--- a/Include/internal/pycore_freelist.h
+++ b/Include/internal/pycore_freelist.h
@@ -59,10 +59,19 @@ struct _Py_float_state {
 #endif
 };
 
+struct _Py_slice_state {
+#ifdef WITH_FREELISTS
+    /* Using a cache is very effective since typically only a single slice is
+       created and then deleted again. */
+    PySliceObject *slice_cache;
+#endif
+};
+
 typedef struct _Py_freelist_state {
     struct _Py_float_state float_state;
     struct _Py_tuple_state tuple_state;
     struct _Py_list_state list_state;
+    struct _Py_slice_state slice_state;
 } _PyFreeListState;
 
 #ifdef __cplusplus

diff --git a/Include/internal/pycore_gc.h b/Include/internal/pycore_gc.h
index c029b23930664869049b29ca3236051717da5112..f8e86a22d0fa580a3fb1a48abf9fd3e5f0629c9f 100644 (file)
--- a/Include/internal/pycore_gc.h
+++ b/Include/internal/pycore_gc.h
@@ -249,6 +249,7 @@ extern void _Py_ClearFreeLists(_PyFreeListState *state, int is_finalization);
 extern void _PyTuple_ClearFreeList(_PyFreeListState *state, int is_finalization);
 extern void _PyFloat_ClearFreeList(_PyFreeListState *state, int is_finalization);
 extern void _PyList_ClearFreeList(_PyFreeListState *state, int is_finalization);
+extern void _PySlice_ClearCache(_PyFreeListState *state);
 extern void _PyDict_ClearFreeList(PyInterpreterState *interp);
 extern void _PyAsyncGen_ClearFreeLists(PyInterpreterState *interp);
 extern void _PyContext_ClearFreeList(PyInterpreterState *interp);

diff --git a/Include/internal/pycore_interp.h b/Include/internal/pycore_interp.h
index dadc8e3b91a75dce5fb47d875ec78a31bfa817a6..134a882695bbd715ff2f867b1932ee3557f3d497 100644 (file)
--- a/Include/internal/pycore_interp.h
+++ b/Include/internal/pycore_interp.h
@@ -187,9 +187,6 @@ struct _is {
     struct _Py_long_state long_state;
     struct _dtoa_state dtoa;
     struct _py_func_state func_state;
-    /* Using a cache is very effective since typically only a single slice is
-       created and then deleted again. */
-    PySliceObject *slice_cache;
 
     struct _Py_tuple_state tuple;
     struct _Py_dict_state dict_state;

diff --git a/Include/internal/pycore_sliceobject.h b/Include/internal/pycore_sliceobject.h
index 98665c3859d57461d43ea420150d8a52ec64936f..0c72d3ee6225c51d488bd65687f4c5308861838d 100644 (file)
--- a/Include/internal/pycore_sliceobject.h
+++ b/Include/internal/pycore_sliceobject.h
@@ -11,7 +11,7 @@ extern "C" {
 
 /* runtime lifecycle */
 
-extern void _PySlice_Fini(PyInterpreterState *);
+extern void _PySlice_Fini(_PyFreeListState *);
 
 extern PyObject *
 _PyBuildSlice_ConsumeRefs(PyObject *start, PyObject *stop);

diff --git a/Objects/sliceobject.c b/Objects/sliceobject.c
index a3ed0c096d84ed9db96ae4b03a4d7f92d7a7188e..9ec8ea8e1b307f8adcba3ad0fe9ad668d592bc42 100644 (file)
--- a/Objects/sliceobject.c
+++ b/Objects/sliceobject.c
@@ -103,16 +103,20 @@ PyObject _Py_EllipsisObject = _PyObject_HEAD_INIT(&PyEllipsis_Type);
 
 /* Slice object implementation */
 
-
-void _PySlice_Fini(PyInterpreterState *interp)
+void _PySlice_ClearCache(_PyFreeListState *state)
 {
-    PySliceObject *obj = interp->slice_cache;
+    PySliceObject *obj = state->slice_state.slice_cache;
     if (obj != NULL) {
-        interp->slice_cache = NULL;
+        state->slice_state.slice_cache = NULL;
         PyObject_GC_Del(obj);
     }
 }
 
+void _PySlice_Fini(_PyFreeListState *state)
+{
+    _PySlice_ClearCache(state);
+}
+
 /* start, stop, and step are python objects with None indicating no
    index is present.
 */
@@ -122,11 +126,11 @@ _PyBuildSlice_Consume2(PyObject *start, PyObject *stop, PyObject *step)
 {
     assert(start != NULL && stop != NULL && step != NULL);
 
-    PyInterpreterState *interp = _PyInterpreterState_GET();
+    _PyFreeListState *state = _PyFreeListState_GET();
     PySliceObject *obj;
-    if (interp->slice_cache != NULL) {
-        obj = interp->slice_cache;
-        interp->slice_cache = NULL;
+    if (state->slice_state.slice_cache != NULL) {
+        obj = state->slice_state.slice_cache;
+        state->slice_state.slice_cache = NULL;
         _Py_NewReference((PyObject *)obj);
     }
     else {
@@ -354,13 +358,13 @@ Create a slice object.  This is used for extended slicing (e.g. a[0:10:2]).");
 static void
 slice_dealloc(PySliceObject *r)
 {
-    PyInterpreterState *interp = _PyInterpreterState_GET();
+    _PyFreeListState *state = _PyFreeListState_GET();
     _PyObject_GC_UNTRACK(r);
     Py_DECREF(r->step);
     Py_DECREF(r->start);
     Py_DECREF(r->stop);
-    if (interp->slice_cache == NULL) {
-        interp->slice_cache = r;
+    if (state->slice_state.slice_cache == NULL) {
+        state->slice_state.slice_cache = r;
     }
     else {
         PyObject_GC_Del(r);

diff --git a/Python/pylifecycle.c b/Python/pylifecycle.c
index 4198f6a38f0e56cfdbf65ae7b86e24f818f6cb5b..c33892af1c91da6f5c0af5d7e161c3347a0c4edf 100644 (file)
--- a/Python/pylifecycle.c
+++ b/Python/pylifecycle.c
@@ -1752,15 +1752,13 @@ finalize_interp_types(PyInterpreterState *interp)
     _PyUnicode_ClearInterned(interp);
 
     _PyDict_Fini(interp);
-
-    _PySlice_Fini(interp);
-
     _PyUnicode_Fini(interp);
 
     _PyFreeListState *state = _PyFreeListState_GET();
     _PyTuple_Fini(state);
     _PyList_Fini(state);
     _PyFloat_Fini(state);
+    _PySlice_Fini(state);
 
 #ifdef Py_DEBUG
     _PyStaticObjects_CheckRefcnt(interp);

diff --git a/Python/pystate.c b/Python/pystate.c
index eaf77b0da62a843e2aca6b8b91cd22ea13482863..01dc86feabfb2f8dcfa1ebc8ce25e8be2a871c5a 100644 (file)
--- a/Python/pystate.c
+++ b/Python/pystate.c
@@ -1549,6 +1549,7 @@ PyThreadState_Clear(PyThreadState *tstate)
     // Each thread should clear own freelists in free-threading builds.
     _PyFreeListState *freelist_state = &((_PyThreadStateImpl*)tstate)->freelist_state;
     _Py_ClearFreeLists(freelist_state, 0);
+    _PySlice_ClearCache(freelist_state);
 #endif
 
     _PyThreadState_ClearMimallocHeaps(tstate);