Cleanup init of interp->pystats_struct.

Add the _PyStats_InterpInit() function which will set pystats_enabled
flag and allocate pystats_struct as required.  We shouldn't be putting
logic in _PyObject_InitState().  This also fixes a bug where
interp->pystats_struct was allocated repeated rather than once per
interpreter.

Author: nascheme

Include/internal/pycore_stats.h

@@ -110,6 +110,7 @@ PyAPI_FUNC(PyObject*) _Py_GetSpecializationStats(void);
         RARE_EVENT_INTERP_INC(interp, name); \
     } while (0); \
 
+PyStatus _PyStats_InterpInit(PyInterpreterState *);
 bool _PyStats_ThreadInit(PyInterpreterState *, _PyThreadStateImpl *);
 void _PyStats_ThreadFini(_PyThreadStateImpl *);
 void _PyStats_Attach(_PyThreadStateImpl *);

Objects/object.c

@@ -2418,13 +2418,6 @@ _PyObject_InitState(PyInterpreterState *interp)
     if (refchain_init(interp) < 0) {
         return _PyStatus_NO_MEMORY();
     }
-#endif
-#ifdef Py_STATS
-    if (interp->config._pystats) {
-        // start with pystats enabled, can be disabled via sys._stats_off()
-        // this needs to be set before the first tstate is created
-        interp->pystats_enabled = 1;
-    }
 #endif
     return _PyStatus_OK();
 }

Python/pylifecycle.c

@@ -26,6 +26,7 @@
 #include "pycore_runtime.h"       // _Py_ID()
 #include "pycore_runtime_init.h"  // _PyRuntimeState_INIT
 #include "pycore_setobject.h"     // _PySet_NextEntry()
+#include "pycore_stats.h"         // _PyStats_InterpInit()
 #include "pycore_sysmodule.h"     // _PySys_ClearAttrString()
 #include "pycore_traceback.h"     // _Py_DumpTracebackThreads()
 #include "pycore_typeobject.h"    // _PyTypes_InitTypes()
@@ -652,6 +653,14 @@ pycore_create_interpreter(_PyRuntimeState *runtime,
         return status;
     }
 
+#ifdef Py_STATS
+    // initialize pystats.  This must be done after the settings are loaded.
+    status = _PyStats_InterpInit(interp);
+    if (_PyStatus_EXCEPTION(status)) {
+        return status;
+    }
+#endif
+
     // initialize the interp->obmalloc state.  This must be done after
     // the settings are loaded (so that feature_flags are set) but before
     // any calls are made to obmalloc functions.
@@ -2349,6 +2358,14 @@ new_interpreter(PyThreadState **tstate_p,
         return status;
     }
 
+#ifdef Py_STATS
+    // initialize pystats.  This must be done after the settings are loaded.
+    status = _PyStats_InterpInit(interp);
+    if (_PyStatus_EXCEPTION(status)) {
+        return status;
+    }
+#endif
+
     // initialize the interp->obmalloc state.  This must be done after
     // the settings are loaded (so that feature_flags are set) but before
     // any calls are made to obmalloc functions.

Python/pystats.c

@@ -4,6 +4,7 @@
 #include "pycore_pyatomic_ft_wrappers.h"
 #include "pycore_pylifecycle.h"     // _PyOS_URandomNonblock()
 #include "pycore_tstate.h"
+#include "pycore_initconfig.h"      // _PyStatus_OK()
 #include "pycore_uop_metadata.h"    // _PyOpcode_uop_name
 #include "pycore_uop_ids.h"         // MAX_UOP_ID
 #include "pycore_pystate.h"         // _PyThreadState_GET()
@@ -725,23 +726,31 @@ _Py_PrintSpecializationStats(int to_file)
     return 1;
 }
 
+PyStatus
+_PyStats_InterpInit(PyInterpreterState *interp)
+{
+    if (interp->config._pystats) {
+        // start with pystats enabled, can be disabled via sys._stats_off()
+        // this needs to be set before the first tstate is created
+        interp->pystats_enabled = 1;
+        interp->pystats_struct = PyMem_RawCalloc(1, sizeof(PyStats));
+        if (interp->pystats_struct == NULL) {
+            return _PyStatus_ERR("out-of-memory while initializing interpreter");
+        }
+    }
+    return _PyStatus_OK();
+}
+
 bool
 _PyStats_ThreadInit(PyInterpreterState *interp, _PyThreadStateImpl *tstate)
 {
-    STATS_LOCK(interp);
-    if (interp->pystats_enabled) {
-        PyStats *s = PyMem_RawCalloc(1, sizeof(PyStats));
-        if (s == NULL) {
-            STATS_UNLOCK(interp);
+#ifdef Py_GIL_DISABLED
+    if (FT_ATOMIC_LOAD_INT_RELAXED(interp->pystats_enabled)) {
+        assert(interp->pystats_struct != NULL);
+        tstate->pystats_struct = PyMem_RawCalloc(1, sizeof(PyStats));
+        if (tstate->pystats_struct == NULL) {
             return false;
         }
-        FT_ATOMIC_STORE_PTR_RELAXED(interp->pystats_struct, s);
-    }
-    STATS_UNLOCK(interp);
-#ifdef Py_GIL_DISABLED
-    tstate->pystats_struct = PyMem_RawCalloc(1, sizeof(PyStats));
-    if (tstate->pystats_struct == NULL) {
-        return false;
     }
 #endif
     return true;