bpo-35081: Add _PyThreadState_GET() internal macro (GH-10266)

If Py_BUILD_CORE is defined, the PyThreadState_GET() macro access
_PyRuntime which comes from the internal pycore_state.h header.
Public headers must not require internal headers.

Move PyThreadState_GET() and _PyInterpreterState_GET_UNSAFE() from
Include/pystate.h to Include/internal/pycore_state.h, and rename
PyThreadState_GET() to _PyThreadState_GET() there.

The PyThreadState_GET() macro of pystate.h is now redefined when
pycore_state.h is included, to use the fast _PyThreadState_GET().

Changes:

* Add _PyThreadState_GET() macro
* Replace "PyThreadState_GET()->interp" with
  _PyInterpreterState_GET_UNSAFE()
* Replace PyThreadState_GET() with _PyThreadState_GET() in internal C
  files (compiled with Py_BUILD_CORE defined), but keep
  PyThreadState_GET() in the public header files.
* _testcapimodule.c: replace PyThreadState_GET() with
  PyThreadState_Get(); the module is not compiled with Py_BUILD_CORE
  defined.
* pycore_state.h now requires Py_BUILD_CORE to be defined.

20 files changed, 144 insertions, 104 deletions

diff --git a/Include/internal/pycore_state.h b/Include/internal/pycore_state.h
index ff25d2ebf4..9a084f7f4d 100644
--- a/Include/internal/pycore_state.h
+++ b/Include/internal/pycore_state.h
@@ -4,6 +4,10 @@
 extern "C" {
 #endif
 
+#ifndef Py_BUILD_CORE
+#  error "Py_BUILD_CORE must be defined to include this header"
+#endif
+
 #include "pystate.h"
 #include "pythread.h"
 
@@ -214,6 +218,36 @@ PyAPI_FUNC(_PyInitError) _PyRuntime_Initialize(void);
     (_PyRuntime.finalizing == tstate)
 
 
+/* Variable and macro for in-line access to current thread
+   and interpreter state */
+
+/* Get the current Python thread state.
+
+   Efficient macro reading directly the 'gilstate.tstate_current' atomic
+   variable. The macro is unsafe: it does not check for error and it can
+   return NULL.
+
+   The caller must hold the GIL.
+
+   See also PyThreadState_Get() and PyThreadState_GET(). */
+#define _PyThreadState_GET() \
+    ((PyThreadState*)_Py_atomic_load_relaxed(&_PyRuntime.gilstate.tstate_current))
+
+/* Redefine PyThreadState_GET() as an alias to _PyThreadState_GET() */
+#undef PyThreadState_GET
+#define PyThreadState_GET() _PyThreadState_GET()
+
+/* Get the current interpreter state.
+
+   The macro is unsafe: it does not check for error and it can return NULL.
+
+   The caller must hold the GIL.
+
+   See also _PyInterpreterState_Get()
+   and _PyGILState_GetInterpreterStateUnsafe(). */
+#define _PyInterpreterState_GET_UNSAFE() (_PyThreadState_GET()->interp)
+
+
 /* Other */
 
 PyAPI_FUNC(_PyInitError) _PyInterpreterState_Enable(_PyRuntimeState *);
diff --git a/Include/pystate.h b/Include/pystate.h
index 7fc921e9ac..b30c3187b6 100644
--- a/Include/pystate.h
+++ b/Include/pystate.h
@@ -245,15 +245,17 @@ typedef struct _ts {
 PyAPI_FUNC(PyInterpreterState *) PyInterpreterState_New(void);
 PyAPI_FUNC(void) PyInterpreterState_Clear(PyInterpreterState *);
 PyAPI_FUNC(void) PyInterpreterState_Delete(PyInterpreterState *);
+
 #if !defined(Py_LIMITED_API)
+/* Get the current interpreter state.
+
+   Issue a fatal error if there no current Python thread state or no current
+   interpreter. It cannot return NULL.
+
+   The caller must hold the GIL.*/
 PyAPI_FUNC(PyInterpreterState *) _PyInterpreterState_Get(void);
 #endif
-#ifdef Py_BUILD_CORE
-   /* Macro which should only be used for performance critical code.
-      Need "#include "pycore_state.h". See also _PyInterpreterState_Get()
-      and _PyGILState_GetInterpreterStateUnsafe(). */
-#  define _PyInterpreterState_GET_UNSAFE() (PyThreadState_GET()->interp)
-#endif
+
 #if !defined(Py_LIMITED_API) || Py_LIMITED_API+0 >= 0x03070000
 /* New in 3.7 */
 PyAPI_FUNC(int64_t) PyInterpreterState_GetID(PyInterpreterState *);
@@ -286,11 +288,27 @@ PyAPI_FUNC(void) PyThreadState_DeleteCurrent(void);
 PyAPI_FUNC(void) _PyGILState_Reinit(void);
 #endif /* !Py_LIMITED_API */
 
-/* Return the current thread state. The global interpreter lock must be held.
- * When the current thread state is NULL, this issues a fatal error (so that
- * the caller needn't check for NULL). */
+/* Get the current thread state.
+
+   When the current thread state is NULL, this issues a fatal error (so that
+   the caller needn't check for NULL).
+
+   The caller must hold the GIL.
+
+   See also PyThreadState_GET() and _PyThreadState_GET(). */
 PyAPI_FUNC(PyThreadState *) PyThreadState_Get(void);
 
+/* Get the current Python thread state.
+
+   Macro using PyThreadState_Get() or _PyThreadState_GET() depending if
+   pycore_state.h is included or not (this header redefines the macro).
+
+   If PyThreadState_Get() is used, issue a fatal error if the current thread
+   state is NULL.
+
+   See also PyThreadState_Get() and _PyThreadState_GET(). */
+#define PyThreadState_GET() PyThreadState_Get()
+
 #ifndef Py_LIMITED_API
 /* Similar to PyThreadState_Get(), but don't issue a fatal error
  * if it is NULL. */
@@ -301,18 +319,6 @@ PyAPI_FUNC(PyThreadState *) PyThreadState_Swap(PyThreadState *);
 PyAPI_FUNC(PyObject *) PyThreadState_GetDict(void);
 PyAPI_FUNC(int) PyThreadState_SetAsyncExc(unsigned long, PyObject *);
 
-
-/* Variable and macro for in-line access to current thread state */
-
-/* Assuming the current thread holds the GIL, this is the
-   PyThreadState for the current thread. */
-#ifdef Py_BUILD_CORE
-#  define PyThreadState_GET() \
-             ((PyThreadState*)_Py_atomic_load_relaxed(&_PyRuntime.gilstate.tstate_current))
-#else
-#  define PyThreadState_GET() PyThreadState_Get()
-#endif
-
 typedef
     enum {PyGILState_LOCKED, PyGILState_UNLOCKED}
         PyGILState_STATE;
@@ -366,11 +372,11 @@ PyAPI_FUNC(PyThreadState *) PyGILState_GetThisThreadState(void);
    The function returns 1 if _PyGILState_check_enabled is non-zero. */
 PyAPI_FUNC(int) PyGILState_Check(void);
 
-/* Unsafe function to get the single PyInterpreterState used by this process'
-   GILState implementation.
+/* Get the single PyInterpreterState used by this process' GILState
+   implementation.
 
-   Return NULL before _PyGILState_Init() is called and after _PyGILState_Fini()
-   is called.
+   This function doesn't check for error. Return NULL before _PyGILState_Init()
+   is called and after _PyGILState_Fini() is called.
 
    See also _PyInterpreterState_Get() and _PyInterpreterState_GET_UNSAFE(). */
 PyAPI_FUNC(PyInterpreterState *) _PyGILState_GetInterpreterStateUnsafe(void);
diff --git a/Modules/_testcapimodule.c b/Modules/_testcapimodule.c
index b2cda51423..018af4a137 100644
--- a/Modules/_testcapimodule.c
+++ b/Modules/_testcapimodule.c
@@ -4160,7 +4160,7 @@ test_PyTime_AsMicroseconds(PyObject *self, PyObject *args)
 static PyObject*
 get_recursion_depth(PyObject *self, PyObject *args)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = PyThreadState_Get();
 
     /* subtract one to ignore the frame of the get_recursion_depth() call */
     return PyLong_FromLong(tstate->recursion_depth - 1);
diff --git a/Objects/call.c b/Objects/call.c
index 9061d0b1ac..48e3aafe5f 100644
--- a/Objects/call.c
+++ b/Objects/call.c
@@ -258,7 +258,7 @@ function_code_fastcall(PyCodeObject *co, PyObject *const *args, Py_ssize_t nargs
                        PyObject *globals)
 {
     PyFrameObject *f;
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     PyObject **fastlocals;
     Py_ssize_t i;
     PyObject *result;
diff --git a/Objects/dictobject.c b/Objects/dictobject.c
index a9ae907d9f..ea564a2b7c 100644
--- a/Objects/dictobject.c
+++ b/Objects/dictobject.c
@@ -1314,9 +1314,9 @@ PyDict_GetItem(PyObject *op, PyObject *key)
     /* We can arrive here with a NULL tstate during initialization: try
        running "python -Wi" for an example related to string interning.
        Let's just hope that no exception occurs then...  This must be
-       PyThreadState_GET() and not PyThreadState_Get() because the latter
+       _PyThreadState_GET() and not PyThreadState_Get() because the latter
        abort Python if tstate is NULL. */
-    tstate = PyThreadState_GET();
+    tstate = _PyThreadState_GET();
     if (tstate != NULL && tstate->curexc_type != NULL) {
         /* preserve the existing exception */
         PyObject *err_type, *err_value, *err_tb;
diff --git a/Objects/genobject.c b/Objects/genobject.c
index 885b3f2eca..7c2948bb52 100644
--- a/Objects/genobject.c
+++ b/Objects/genobject.c
@@ -151,7 +151,7 @@ gen_dealloc(PyGenObject *gen)
 static PyObject *
 gen_send_ex(PyGenObject *gen, PyObject *arg, int exc, int closing)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     PyFrameObject *f = gen->gi_frame;
     PyObject *result;
 
@@ -1157,7 +1157,7 @@ PyCoro_New(PyFrameObject *f, PyObject *name, PyObject *qualname)
         return NULL;
     }
 
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     int origin_depth = tstate->coroutine_origin_tracking_depth;
 
     if (origin_depth == 0) {
@@ -1267,7 +1267,7 @@ async_gen_init_hooks(PyAsyncGenObject *o)
 
     o->ag_hooks_inited = 1;
 
-    tstate = PyThreadState_GET();
+    tstate = _PyThreadState_GET();
 
     finalizer = tstate->async_gen_finalizer;
     if (finalizer) {
diff --git a/Objects/object.c b/Objects/object.c
index f7395c783a..d3a97f6c5b 100644
--- a/Objects/object.c
+++ b/Objects/object.c
@@ -2136,7 +2136,7 @@ _PyTrash_deposit_object(PyObject *op)
 void
 _PyTrash_thread_deposit_object(PyObject *op)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     _PyObject_ASSERT(op, PyObject_IS_GC(op));
     _PyObject_ASSERT(op, !_PyObject_GC_IS_TRACKED(op));
     _PyObject_ASSERT(op, op->ob_refcnt == 0);
@@ -2174,7 +2174,7 @@ _PyTrash_destroy_chain(void)
 void
 _PyTrash_thread_destroy_chain(void)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     /* We need to increase trash_delete_nesting here, otherwise,
        _PyTrash_thread_destroy_chain will be called recursively
        and then possibly crash.  An example that may crash without
diff --git a/Objects/odictobject.c b/Objects/odictobject.c
index 81c996be56..52ac7e5f88 100644
--- a/Objects/odictobject.c
+++ b/Objects/odictobject.c
@@ -1355,7 +1355,7 @@ static PyGetSetDef odict_getset[] = {
 static void
 odict_dealloc(PyODictObject *self)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
 
     PyObject_GC_UnTrack(self);
     Py_TRASHCAN_SAFE_BEGIN(self)
diff --git a/Objects/typeobject.c b/Objects/typeobject.c
index 9a390b3e50..dedc4f7c87 100644
--- a/Objects/typeobject.c
+++ b/Objects/typeobject.c
@@ -1115,7 +1115,7 @@ subtype_dealloc(PyObject *self)
 {
     PyTypeObject *type, *base;
     destructor basedealloc;
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     int has_finalizer;
 
     /* Extract the type; we expect it to be a heap type */
@@ -7678,7 +7678,7 @@ super_init(PyObject *self, PyObject *args, PyObject *kwds)
         PyFrameObject *f;
         PyCodeObject *co;
         Py_ssize_t i, n;
-        f = PyThreadState_GET()->frame;
+        f = _PyThreadState_GET()->frame;
         if (f == NULL) {
             PyErr_SetString(PyExc_RuntimeError,
                             "super(): no current frame");
diff --git a/Parser/myreadline.c b/Parser/myreadline.c
index d18cf1b61b..f511319814 100644
--- a/Parser/myreadline.c
+++ b/Parser/myreadline.c
@@ -301,7 +301,7 @@ PyOS_Readline(FILE *sys_stdin, FILE *sys_stdout, const char *prompt)
     char *rv, *res;
     size_t len;
 
-    if (_PyOS_ReadlineTState == PyThreadState_GET()) {
+    if (_PyOS_ReadlineTState == _PyThreadState_GET()) {
         PyErr_SetString(PyExc_RuntimeError,
                         "can't re-enter readline");
         return NULL;
@@ -316,7 +316,7 @@ PyOS_Readline(FILE *sys_stdin, FILE *sys_stdout, const char *prompt)
         _PyOS_ReadlineLock = PyThread_allocate_lock();
     }
 
-    _PyOS_ReadlineTState = PyThreadState_GET();
+    _PyOS_ReadlineTState = _PyThreadState_GET();
     Py_BEGIN_ALLOW_THREADS
     PyThread_acquire_lock(_PyOS_ReadlineLock, 1);
 
diff --git a/Python/_warnings.c b/Python/_warnings.c
index 619ec6f26b..9b50289bb6 100644
--- a/Python/_warnings.c
+++ b/Python/_warnings.c
@@ -671,7 +671,7 @@ setup_context(Py_ssize_t stack_level, PyObject **filename, int *lineno,
     PyObject *globals;
 
     /* Setup globals, filename and lineno. */
-    PyFrameObject *f = PyThreadState_GET()->frame;
+    PyFrameObject *f = _PyThreadState_GET()->frame;
     // Stack level comparisons to Python code is off by one as there is no
     // warnings-related stack level to avoid.
     if (stack_level <= 0 || is_internal_frame(f)) {
diff --git a/Python/ceval.c b/Python/ceval.c
index 5599b6efa3..ac9db1533b 100644
--- a/Python/ceval.c
+++ b/Python/ceval.c
@@ -157,7 +157,7 @@ PyEval_InitThreads(void)
     if (gil_created())
         return;
     create_gil();
-    take_gil(PyThreadState_GET());
+    take_gil(_PyThreadState_GET());
     _PyRuntime.ceval.pending.main_thread = PyThread_get_thread_ident();
     if (!_PyRuntime.ceval.pending.lock)
         _PyRuntime.ceval.pending.lock = PyThread_allocate_lock();
@@ -175,7 +175,7 @@ _PyEval_FiniThreads(void)
 void
 PyEval_AcquireLock(void)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     if (tstate == NULL)
         Py_FatalError("PyEval_AcquireLock: current thread state is NULL");
     take_gil(tstate);
@@ -185,10 +185,10 @@ void
 PyEval_ReleaseLock(void)
 {
     /* This function must succeed when the current thread state is NULL.
-       We therefore avoid PyThreadState_GET() which dumps a fatal error
+       We therefore avoid PyThreadState_Get() which dumps a fatal error
        in debug mode.
     */
-    drop_gil(PyThreadState_GET());
+    drop_gil(_PyThreadState_GET());
 }
 
 void
@@ -222,7 +222,7 @@ PyEval_ReleaseThread(PyThreadState *tstate)
 void
 PyEval_ReInitThreads(void)
 {
-    PyThreadState *current_tstate = PyThreadState_GET();
+    PyThreadState *current_tstate = _PyThreadState_GET();
 
     if (!gil_created())
         return;
@@ -462,7 +462,7 @@ Py_SetRecursionLimit(int new_limit)
 int
 _Py_CheckRecursiveCall(const char *where)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     int recursion_limit = _PyRuntime.ceval.recursion_limit;
 
 #ifdef USE_STACKCHECK
@@ -543,7 +543,7 @@ _PyEval_EvalFrameDefault(PyFrameObject *f, int throwflag)
     int oparg;         /* Current opcode argument, if any */
     PyObject **fastlocals, **freevars;
     PyObject *retval = NULL;            /* Return value */
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     PyCodeObject *co;
 
     /* when tracing we set things up so that
@@ -3702,7 +3702,7 @@ _PyEval_EvalCodeWithName(PyObject *_co, PyObject *globals, PyObject *locals,
     }
 
     /* Create the frame */
-    tstate = PyThreadState_GET();
+    tstate = _PyThreadState_GET();
     assert(tstate != NULL);
     f = _PyFrame_New_NoTrack(tstate, co, globals, locals);
     if (f == NULL) {
@@ -4003,7 +4003,7 @@ do_raise(PyObject *exc, PyObject *cause)
 
     if (exc == NULL) {
         /* Reraise */
-        PyThreadState *tstate = PyThreadState_GET();
+        PyThreadState *tstate = _PyThreadState_GET();
         _PyErr_StackItem *exc_info = _PyErr_GetTopmostException(tstate);
         PyObject *tb;
         type = exc_info->exc_type;
@@ -4275,7 +4275,7 @@ call_trace(Py_tracefunc func, PyObject *obj,
 PyObject *
 _PyEval_CallTracing(PyObject *func, PyObject *args)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     int save_tracing = tstate->tracing;
     int save_use_tracing = tstate->use_tracing;
     PyObject *result;
@@ -4329,7 +4329,7 @@ maybe_call_line_trace(Py_tracefunc func, PyObject *obj,
 void
 PyEval_SetProfile(Py_tracefunc func, PyObject *arg)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     PyObject *temp = tstate->c_profileobj;
     Py_XINCREF(arg);
     tstate->c_profilefunc = NULL;
@@ -4346,7 +4346,7 @@ PyEval_SetProfile(Py_tracefunc func, PyObject *arg)
 void
 PyEval_SetTrace(Py_tracefunc func, PyObject *arg)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     PyObject *temp = tstate->c_traceobj;
     _Py_TracingPossible += (func != NULL) - (tstate->c_tracefunc != NULL);
     Py_XINCREF(arg);
@@ -4366,21 +4366,21 @@ void
 _PyEval_SetCoroutineOriginTrackingDepth(int new_depth)
 {
     assert(new_depth >= 0);
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     tstate->coroutine_origin_tracking_depth = new_depth;
 }
 
 int
 _PyEval_GetCoroutineOriginTrackingDepth(void)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     return tstate->coroutine_origin_tracking_depth;
 }
 
 void
 _PyEval_SetCoroutineWrapper(PyObject *wrapper)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
 
     Py_XINCREF(wrapper);
     Py_XSETREF(tstate->coroutine_wrapper, wrapper);
@@ -4389,14 +4389,14 @@ _PyEval_SetCoroutineWrapper(PyObject *wrapper)
 PyObject *
 _PyEval_GetCoroutineWrapper(void)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     return tstate->coroutine_wrapper;
 }
 
 void
 _PyEval_SetAsyncGenFirstiter(PyObject *firstiter)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
 
     Py_XINCREF(firstiter);
     Py_XSETREF(tstate->async_gen_firstiter, firstiter);
@@ -4405,14 +4405,14 @@ _PyEval_SetAsyncGenFirstiter(PyObject *firstiter)
 PyObject *
 _PyEval_GetAsyncGenFirstiter(void)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     return tstate->async_gen_firstiter;
 }
 
 void
 _PyEval_SetAsyncGenFinalizer(PyObject *finalizer)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
 
     Py_XINCREF(finalizer);
     Py_XSETREF(tstate->async_gen_finalizer, finalizer);
@@ -4421,7 +4421,7 @@ _PyEval_SetAsyncGenFinalizer(PyObject *finalizer)
 PyObject *
 _PyEval_GetAsyncGenFinalizer(void)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     return tstate->async_gen_finalizer;
 }
 
@@ -4465,7 +4465,7 @@ PyEval_GetGlobals(void)
 PyFrameObject *
 PyEval_GetFrame(void)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     return _PyThreadState_GetFrame(tstate);
 }
 
@@ -4566,11 +4566,11 @@ call_function(PyObject ***pp_stack, Py_ssize_t oparg, PyObject *kwnames)
        presumed to be the most frequent callable object.
     */
     if (PyCFunction_Check(func)) {
-        PyThreadState *tstate = PyThreadState_GET();
+        PyThreadState *tstate = _PyThreadState_GET();
         C_TRACE(x, _PyCFunction_FastCallKeywords(func, stack, nargs, kwnames));
     }
     else if (Py_TYPE(func) == &PyMethodDescr_Type) {
-        PyThreadState *tstate = PyThreadState_GET();
+        PyThreadState *tstate = _PyThreadState_GET();
         if (nargs > 0 && tstate->use_tracing) {
             /* We need to create a temporary bound method as argument
                for profiling.
@@ -4640,12 +4640,12 @@ do_call_core(PyObject *func, PyObject *callargs, PyObject *kwdict)
     PyObject *result;
 
     if (PyCFunction_Check(func)) {
-        PyThreadState *tstate = PyThreadState_GET();
+        PyThreadState *tstate = _PyThreadState_GET();
         C_TRACE(result, PyCFunction_Call(func, callargs, kwdict));
         return result;
     }
     else if (Py_TYPE(func) == &PyMethodDescr_Type) {
-        PyThreadState *tstate = PyThreadState_GET();
+        PyThreadState *tstate = _PyThreadState_GET();
         Py_ssize_t nargs = PyTuple_GET_SIZE(callargs);
         if (nargs > 0 && tstate->use_tracing) {
             /* We need to create a temporary bound method as argument
diff --git a/Python/context.c b/Python/context.c
index b1f67b5aa7..1fb2a5de2b 100644
--- a/Python/context.c
+++ b/Python/context.c
@@ -112,7 +112,7 @@ PyContext_Enter(PyObject *octx)
         return -1;
     }
 
-    PyThreadState *ts = PyThreadState_GET();
+    PyThreadState *ts = _PyThreadState_GET();
     assert(ts != NULL);
 
     ctx->ctx_prev = (PyContext *)ts->context;  /* borrow */
@@ -138,7 +138,7 @@ PyContext_Exit(PyObject *octx)
         return -1;
     }
 
-    PyThreadState *ts = PyThreadState_GET();
+    PyThreadState *ts = _PyThreadState_GET();
     assert(ts != NULL);
 
     if (ts->context != (PyObject *)ctx) {
@@ -178,7 +178,7 @@ PyContextVar_Get(PyObject *ovar, PyObject *def, PyObject **val)
     ENSURE_ContextVar(ovar, -1)
     PyContextVar *var = (PyContextVar *)ovar;
 
-    PyThreadState *ts = PyThreadState_GET();
+    PyThreadState *ts = _PyThreadState_GET();
     assert(ts != NULL);
     if (ts->context == NULL) {
         goto not_found;
@@ -382,7 +382,7 @@ context_new_from_vars(PyHamtObject *vars)
 static inline PyContext *
 context_get(void)
 {
-    PyThreadState *ts = PyThreadState_GET();
+    PyThreadState *ts = _PyThreadState_GET();
     assert(ts != NULL);
     PyContext *current_ctx = (PyContext *)ts->context;
     if (current_ctx == NULL) {
diff --git a/Python/errors.c b/Python/errors.c
index 14a70d9ce8..4c6c34c74f 100644
--- a/Python/errors.c
+++ b/Python/errors.c
@@ -28,7 +28,7 @@ _Py_IDENTIFIER(stderr);
 void
 PyErr_Restore(PyObject *type, PyObject *value, PyObject *traceback)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     PyObject *oldtype, *oldvalue, *oldtraceback;
 
     if (traceback != NULL && !PyTraceBack_Check(traceback)) {
@@ -82,7 +82,7 @@ _PyErr_CreateException(PyObject *exception, PyObject *value)
 void
 PyErr_SetObject(PyObject *exception, PyObject *value)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     PyObject *exc_value;
     PyObject *tb = NULL;
 
@@ -175,7 +175,7 @@ PyErr_SetString(PyObject *exception, const char *string)
 PyObject* _Py_HOT_FUNCTION
 PyErr_Occurred(void)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     return tstate == NULL ? NULL : tstate->curexc_type;
 }
 
@@ -334,7 +334,7 @@ PyErr_NormalizeException(PyObject **exc, PyObject **val, PyObject **tb)
 void
 PyErr_Fetch(PyObject **p_type, PyObject **p_value, PyObject **p_traceback)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
 
     *p_type = tstate->curexc_type;
     *p_value = tstate->curexc_value;
@@ -354,7 +354,7 @@ PyErr_Clear(void)
 void
 PyErr_GetExcInfo(PyObject **p_type, PyObject **p_value, PyObject **p_traceback)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
 
     _PyErr_StackItem *exc_info = _PyErr_GetTopmostException(tstate);
     *p_type = exc_info->exc_type;
@@ -371,7 +371,7 @@ void
 PyErr_SetExcInfo(PyObject *p_type, PyObject *p_value, PyObject *p_traceback)
 {
     PyObject *oldtype, *oldvalue, *oldtraceback;
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
 
     oldtype = tstate->exc_info->exc_type;
     oldvalue = tstate->exc_info->exc_value;
diff --git a/Python/pylifecycle.c b/Python/pylifecycle.c
index d349aaf6de..160f30ce6d 100644
--- a/Python/pylifecycle.c
+++ b/Python/pylifecycle.c
@@ -530,7 +530,7 @@ _Py_InitializeCore_impl(PyInterpreterState **interp_p,
     /* bpo-34008: For backward compatibility reasons, calling Py_Main() after
        Py_Initialize() ignores the new configuration. */
     if (_PyRuntime.core_initialized) {
-        PyThreadState *tstate = PyThreadState_GET();
+        PyThreadState *tstate = _PyThreadState_GET();
         if (!tstate) {
             return _Py_INIT_ERR("failed to read thread state");
         }
@@ -1009,7 +1009,7 @@ Py_FinalizeEx(void)
     wait_for_thread_shutdown();
 
     /* Get current thread state and interpreter pointer */
-    tstate = PyThreadState_GET();
+    tstate = _PyThreadState_GET();
     interp = tstate->interp;
 
     /* The interpreter is still entirely intact at this point, and the
@@ -1406,7 +1406,7 @@ Py_EndInterpreter(PyThreadState *tstate)
 {
     PyInterpreterState *interp = tstate->interp;
 
-    if (tstate != PyThreadState_GET())
+    if (tstate != _PyThreadState_GET())
         Py_FatalError("Py_EndInterpreter: thread is not current");
     if (tstate->frame != NULL)
         Py_FatalError("Py_EndInterpreter: thread still has a frame");
@@ -1928,7 +1928,7 @@ fatal_error(const char *prefix, const char *msg, int status)
        and holds the GIL */
     PyThreadState *tss_tstate = PyGILState_GetThisThreadState();
     if (tss_tstate != NULL) {
-        PyThreadState *tstate = PyThreadState_GET();
+        PyThreadState *tstate = _PyThreadState_GET();
         if (tss_tstate != tstate) {
             /* The Python thread does not hold the GIL */
             tss_tstate = NULL;
diff --git a/Python/pystate.c b/Python/pystate.c
index c77902a7f9..c193a10c8c 100644
--- a/Python/pystate.c
+++ b/Python/pystate.c
@@ -307,7 +307,7 @@ _PyInterpreterState_DeleteExceptMain()
 PyInterpreterState *
 _PyInterpreterState_Get(void)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     if (tstate == NULL) {
         Py_FatalError("_PyInterpreterState_Get(): no current thread state");
     }
@@ -689,7 +689,7 @@ tstate_delete_common(PyThreadState *tstate)
 void
 PyThreadState_Delete(PyThreadState *tstate)
 {
-    if (tstate == PyThreadState_GET())
+    if (tstate == _PyThreadState_GET())
         Py_FatalError("PyThreadState_Delete: tstate is still current");
     if (_PyRuntime.gilstate.autoInterpreterState &&
         PyThread_tss_get(&_PyRuntime.gilstate.autoTSSkey) == tstate)
@@ -703,7 +703,7 @@ PyThreadState_Delete(PyThreadState *tstate)
 void
 PyThreadState_DeleteCurrent()
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     if (tstate == NULL)
         Py_FatalError(
             "PyThreadState_DeleteCurrent: no current tstate");
@@ -758,14 +758,14 @@ _PyThreadState_DeleteExcept(PyThreadState *tstate)
 PyThreadState *
 _PyThreadState_UncheckedGet(void)
 {
-    return PyThreadState_GET();
+    return _PyThreadState_GET();
 }
 
 
 PyThreadState *
 PyThreadState_Get(void)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     if (tstate == NULL)
         Py_FatalError("PyThreadState_Get: no current thread");
 
@@ -776,7 +776,7 @@ PyThreadState_Get(void)
 PyThreadState *
 PyThreadState_Swap(PyThreadState *newts)
 {
-    PyThreadState *oldts = PyThreadState_GET();
+    PyThreadState *oldts = _PyThreadState_GET();
 
     _PyThreadState_SET(newts);
     /* It should not be possible for more than one thread state
@@ -807,7 +807,7 @@ PyThreadState_Swap(PyThreadState *newts)
 PyObject *
 PyThreadState_GetDict(void)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     if (tstate == NULL)
         return NULL;
 
@@ -958,7 +958,7 @@ PyThreadState_IsCurrent(PyThreadState *tstate)
 {
     /* Must be the tstate for this thread */
     assert(PyGILState_GetThisThreadState()==tstate);
-    return tstate == PyThreadState_GET();
+    return tstate == _PyThreadState_GET();
 }
 
 /* Internal initialization/finalization functions called by
@@ -1085,7 +1085,7 @@ PyGILState_Check(void)
         return 1;
     }
 
-    tstate = PyThreadState_GET();
+    tstate = _PyThreadState_GET();
     if (tstate == NULL)
         return 0;
 
diff --git a/Python/symtable.c b/Python/symtable.c
index dae6fda0e4..48e1515fa3 100644
--- a/Python/symtable.c
+++ b/Python/symtable.c
@@ -265,7 +265,7 @@ PySymtable_BuildObject(mod_ty mod, PyObject *filename, PyFutureFeatures *future)
     st->st_future = future;
 
     /* Setup recursion depth check counters */
-    tstate = PyThreadState_GET();
+    tstate = _PyThreadState_GET();
     if (!tstate) {
         PySymtable_Free(st);
         return NULL;
diff --git a/Python/sysmodule.c b/Python/sysmodule.c
index cb13e21c94..830f0a88e4 100644
--- a/Python/sysmodule.c
+++ b/Python/sysmodule.c
@@ -336,7 +336,7 @@ PyDoc_STRVAR(excepthook_doc,
 static PyObject *
 sys_exc_info(PyObject *self, PyObject *noargs)
 {
-    _PyErr_StackItem *err_info = _PyErr_GetTopmostException(PyThreadState_GET());
+    _PyErr_StackItem *err_info = _PyErr_GetTopmostException(_PyThreadState_GET());
     return Py_BuildValue(
         "(OOO)",
         err_info->exc_type != NULL ? err_info->exc_type : Py_None,
@@ -565,7 +565,7 @@ function call.  See the debugger chapter in the library manual."
 static PyObject *
 sys_gettrace(PyObject *self, PyObject *args)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     PyObject *temp = tstate->c_traceobj;
 
     if (temp == NULL)
@@ -603,7 +603,7 @@ and return.  See the profiler chapter in the library manual."
 static PyObject *
 sys_getprofile(PyObject *self, PyObject *args)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     PyObject *temp = tstate->c_profileobj;
 
     if (temp == NULL)
@@ -722,7 +722,7 @@ sys_setrecursionlimit(PyObject *self, PyObject *args)
        the new low-water mark. Otherwise it may not be possible anymore to
        reset the overflowed flag to 0. */
     mark = _Py_RecursionLimitLowerWaterMark(new_limit);
-    tstate = PyThreadState_GET();
+    tstate = _PyThreadState_GET();
     if (tstate->recursion_depth >= mark) {
         PyErr_Format(PyExc_RecursionError,
                      "cannot set the recursion limit to %i at "
@@ -1362,7 +1362,7 @@ purposes only."
 static PyObject *
 sys_getframe(PyObject *self, PyObject *args)
 {
-    PyFrameObject *f = PyThreadState_GET()->frame;
+    PyFrameObject *f = _PyThreadState_GET()->frame;
     int depth = -1;
 
     if (!PyArg_ParseTuple(args, "|i:_getframe", &depth))
@@ -1745,7 +1745,7 @@ static int
 _PySys_ReadPreInitOptions(void)
 {
     /* Rerun the add commands with the actual sys module available */
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     if (tstate == NULL) {
         /* Still don't have a thread state, so something is wrong! */
         return -1;
@@ -1796,7 +1796,7 @@ get_warnoptions(void)
 void
 PySys_ResetWarnOptions(void)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     if (tstate == NULL) {
         _clear_preinit_entries(&_preinit_warnoptions);
         return;
@@ -1835,7 +1835,7 @@ PySys_AddWarnOptionUnicode(PyObject *option)
 void
 PySys_AddWarnOption(const wchar_t *s)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     if (tstate == NULL) {
         _append_preinit_entry(&_preinit_warnoptions, s);
         return;
@@ -1922,7 +1922,7 @@ error:
 void
 PySys_AddXOption(const wchar_t *s)
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     if (tstate == NULL) {
         _append_preinit_entry(&_preinit_xoptions, s);
         return;
diff --git a/Python/thread_nt.h b/Python/thread_nt.h
index 61fa8619bc..21ef5556a6 100644
--- a/Python/thread_nt.h
+++ b/Python/thread_nt.h
@@ -189,7 +189,7 @@ PyThread_start_new_thread(void (*func)(void *), void *arg)
         return PYTHREAD_INVALID_THREAD_ID;
     obj->func = func;
     obj->arg = arg;
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     size_t stacksize = tstate ? tstate->interp->pythread_stacksize : 0;
     hThread = (HANDLE)_beginthreadex(0,
                       Py_SAFE_DOWNCAST(stacksize, Py_ssize_t, unsigned int),
@@ -334,13 +334,13 @@ _pythread_nt_set_stacksize(size_t size)
 {
     /* set to default */
     if (size == 0) {
-        PyThreadState_GET()->interp->pythread_stacksize = 0;
+        _PyInterpreterState_GET_UNSAFE()->pythread_stacksize = 0;
         return 0;
     }
 
     /* valid range? */
     if (size >= THREAD_MIN_STACKSIZE && size < THREAD_MAX_STACKSIZE) {
-        PyThreadState_GET()->interp->pythread_stacksize = size;
+        _PyInterpreterState_GET_UNSAFE()->pythread_stacksize = size;
         return 0;
     }
 
diff --git a/Python/thread_pthread.h b/Python/thread_pthread.h
index 697140558f..6da8b3a447 100644
--- a/Python/thread_pthread.h
+++ b/Python/thread_pthread.h
@@ -174,7 +174,7 @@ PyThread_start_new_thread(void (*func)(void *), void *arg)
         return PYTHREAD_INVALID_THREAD_ID;
 #endif
 #if defined(THREAD_STACK_SIZE)
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     size_t stacksize = tstate ? tstate->interp->pythread_stacksize : 0;
     tss = (stacksize != 0) ? stacksize : THREAD_STACK_SIZE;
     if (tss != 0) {
@@ -591,7 +591,7 @@ _pythread_pthread_set_stacksize(size_t size)
 
     /* set to default */
     if (size == 0) {
-        PyThreadState_GET()->interp->pythread_stacksize = 0;
+        _PyInterpreterState_GET_UNSAFE()->pythread_stacksize = 0;
         return 0;
     }
 
@@ -608,7 +608,7 @@ _pythread_pthread_set_stacksize(size_t size)
             rc = pthread_attr_setstacksize(&attrs, size);
             pthread_attr_destroy(&attrs);
             if (rc == 0) {
-                PyThreadState_GET()->interp->pythread_stacksize = size;
+                _PyInterpreterState_GET_UNSAFE()->pythread_stacksize = size;
                 return 0;
             }
         }