gh-112538: Add internal-only _PyThreadStateImpl "wrapper" for PyThreadState (gh-112560)

Every PyThreadState instance is now actually a _PyThreadStateImpl.
It is safe to cast from `PyThreadState*` to `_PyThreadStateImpl*` and back.
The _PyThreadStateImpl will contain fields that we do not want to expose
in the public C API.

Include/internal/pycore_interp.h

@@ -29,6 +29,7 @@ extern "C" {
 #include "pycore_list.h"          // struct _Py_list_state
 #include "pycore_object_state.h"  // struct _py_object_state
 #include "pycore_obmalloc.h"      // struct _obmalloc_state
+#include "pycore_tstate.h"        // _PyThreadStateImpl
 #include "pycore_tuple.h"         // struct _Py_tuple_state
 #include "pycore_typeobject.h"    // struct types_state
 #include "pycore_unicodeobject.h" // struct _Py_unicode_state
@@ -211,7 +212,7 @@ struct _is {
     struct _Py_interp_static_objects static_objects;
 
     /* the initial PyInterpreterState.threads.head */
-    PyThreadState _initial_thread;
+    _PyThreadStateImpl _initial_thread;
     Py_ssize_t _interactive_src_count;
 };
 

Include/internal/pycore_runtime_init.h

@@ -186,7 +186,12 @@ extern PyTypeObject _PyExc_MemoryError;
                 }, \
             }, \
         }, \
-        ._initial_thread = _PyThreadState_INIT, \
+        ._initial_thread = _PyThreadStateImpl_INIT, \
+    }
+
+#define _PyThreadStateImpl_INIT \
+    { \
+        .base = _PyThreadState_INIT, \
     }
 
 #define _PyThreadState_INIT \

Include/internal/pycore_tstate.h

@@ -0,0 +1,26 @@
+#ifndef Py_INTERNAL_TSTATE_H
+#define Py_INTERNAL_TSTATE_H
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifndef Py_BUILD_CORE
+#  error "this header requires Py_BUILD_CORE define"
+#endif
+
+
+// Every PyThreadState is actually allocated as a _PyThreadStateImpl. The
+// PyThreadState fields are exposed as part of the C API, although most fields
+// are intended to be private. The _PyThreadStateImpl fields not exposed.
+typedef struct _PyThreadStateImpl {
+    // semi-public fields are in PyThreadState.
+    PyThreadState base;
+
+    // TODO: add private fields here
+} _PyThreadStateImpl;
+
+
+#ifdef __cplusplus
+}
+#endif
+#endif /* !Py_INTERNAL_TSTATE_H */

Makefile.pre.in

@@ -1874,6 +1874,7 @@ PYTHON_HEADERS= \
 		$(srcdir)/Include/internal/pycore_token.h \
 		$(srcdir)/Include/internal/pycore_traceback.h \
 		$(srcdir)/Include/internal/pycore_tracemalloc.h \
+		$(srcdir)/Include/internal/pycore_tstate.h \
 		$(srcdir)/Include/internal/pycore_tuple.h \
 		$(srcdir)/Include/internal/pycore_typeobject.h \
 		$(srcdir)/Include/internal/pycore_typevarobject.h \

PCbuild/pythoncore.vcxproj

@@ -285,6 +285,7 @@
     <ClInclude Include="..\Include\internal\pycore_token.h" />
     <ClInclude Include="..\Include\internal\pycore_traceback.h" />
     <ClInclude Include="..\Include\internal\pycore_tracemalloc.h" />
+    <ClInclude Include="..\Include\internal\pycore_tstate.h" />
     <ClInclude Include="..\Include\internal\pycore_tuple.h" />
     <ClInclude Include="..\Include\internal\pycore_typeobject.h" />
     <ClInclude Include="..\Include\internal\pycore_typevarobject.h" />

PCbuild/pythoncore.vcxproj.filters

@@ -780,6 +780,9 @@
     <ClInclude Include="..\Include\internal\pycore_tracemalloc.h">
       <Filter>Include\internal</Filter>
     </ClInclude>
+    <ClInclude Include="..\Include\internal\pycore_tstate.h">
+      <Filter>Include\internal</Filter>
+    </ClInclude>
     <ClInclude Include="..\Include\internal\pycore_tuple.h">
       <Filter>Include\internal</Filter>
     </ClInclude>

Python/pystate.c

@@ -1353,20 +1353,19 @@ allocate_chunk(int size_in_bytes, _PyStackChunk* previous)
     return res;
 }
 
-static PyThreadState *
+static _PyThreadStateImpl *
 alloc_threadstate(void)
 {
-    return PyMem_RawCalloc(1, sizeof(PyThreadState));
+    return PyMem_RawCalloc(1, sizeof(_PyThreadStateImpl));
 }
 
 static void
-free_threadstate(PyThreadState *tstate)
+free_threadstate(_PyThreadStateImpl *tstate)
 {
     // The initial thread state of the interpreter is allocated
     // as part of the interpreter state so should not be freed.
-    if (tstate == &tstate->interp->_initial_thread) {
+    if (tstate == &tstate->base.interp->_initial_thread) {
         // Restore to _PyThreadState_INIT.
-        tstate = &tstate->interp->_initial_thread;
         memcpy(tstate,
                &initial._main_interpreter._initial_thread,
                sizeof(*tstate));
@@ -1385,9 +1384,10 @@ free_threadstate(PyThreadState *tstate)
   */
 
 static void
-init_threadstate(PyThreadState *tstate,
+init_threadstate(_PyThreadStateImpl *_tstate,
                  PyInterpreterState *interp, uint64_t id, int whence)
 {
+    PyThreadState *tstate = (PyThreadState *)_tstate;
     if (tstate->_status.initialized) {
         Py_FatalError("thread state already initialized");
     }
@@ -1444,13 +1444,13 @@ add_threadstate(PyInterpreterState *interp, PyThreadState *tstate,
 static PyThreadState *
 new_threadstate(PyInterpreterState *interp, int whence)
 {
-    PyThreadState *tstate;
+    _PyThreadStateImpl *tstate;
     _PyRuntimeState *runtime = interp->runtime;
     // We don't need to allocate a thread state for the main interpreter
     // (the common case), but doing it later for the other case revealed a
     // reentrancy problem (deadlock).  So for now we always allocate before
     // taking the interpreters lock.  See GH-96071.
-    PyThreadState *new_tstate = alloc_threadstate();
+    _PyThreadStateImpl *new_tstate = alloc_threadstate();
     int used_newtstate;
     if (new_tstate == NULL) {
         return NULL;
@@ -1482,14 +1482,14 @@ new_threadstate(PyInterpreterState *interp, int whence)
     }
 
     init_threadstate(tstate, interp, id, whence);
-    add_threadstate(interp, tstate, old_head);
+    add_threadstate(interp, (PyThreadState *)tstate, old_head);
 
     HEAD_UNLOCK(runtime);
     if (!used_newtstate) {
         // Must be called with lock unlocked to avoid re-entrancy deadlock.
         PyMem_RawFree(new_tstate);
     }
-    return tstate;
+    return (PyThreadState *)tstate;
 }
 
 PyThreadState *
@@ -1678,7 +1678,7 @@ zapthreads(PyInterpreterState *interp)
     while ((tstate = interp->threads.head) != NULL) {
         tstate_verify_not_active(tstate);
         tstate_delete_common(tstate);
-        free_threadstate(tstate);
+        free_threadstate((_PyThreadStateImpl *)tstate);
     }
 }
 
@@ -1689,7 +1689,7 @@ PyThreadState_Delete(PyThreadState *tstate)
     _Py_EnsureTstateNotNULL(tstate);
     tstate_verify_not_active(tstate);
     tstate_delete_common(tstate);
-    free_threadstate(tstate);
+    free_threadstate((_PyThreadStateImpl *)tstate);
 }
 
 
@@ -1701,7 +1701,7 @@ _PyThreadState_DeleteCurrent(PyThreadState *tstate)
     tstate_delete_common(tstate);
     current_fast_clear(tstate->interp->runtime);
     _PyEval_ReleaseLock(tstate->interp, NULL);
-    free_threadstate(tstate);
+    free_threadstate((_PyThreadStateImpl *)tstate);
 }
 
 void
@@ -1751,7 +1751,7 @@ _PyThreadState_DeleteExcept(PyThreadState *tstate)
     for (p = list; p; p = next) {
         next = p->next;
         PyThreadState_Clear(p);
-        free_threadstate(p);
+        free_threadstate((_PyThreadStateImpl *)p);
     }
 }