Run the GC only on eval breaker

pablogsal · Oct 8, 2022 · Oct 4, 2022 · Oct 5, 2022 · Oct 7, 2022 · Oct 7, 2022
commit e6b7b2cc1799bb0ea4fd170dae51daace94f0124
diff --git a/Include/internal/pycore_gc.h b/Include/internal/pycore_gc.h
@@ -202,6 +202,8 @@ extern void _PyList_ClearFreeList(PyInterpreterState *interp);
 extern void _PyDict_ClearFreeList(PyInterpreterState *interp);
 extern void _PyAsyncGen_ClearFreeLists(PyInterpreterState *interp);
 extern void _PyContext_ClearFreeList(PyInterpreterState *interp);
+extern void _Py_ScheduleGC(PyInterpreterState *interp);
+extern void _Py_RunGC(PyThreadState *tstate);
 
 #ifdef __cplusplus
 }

@@ -49,6 +49,8 @@ struct _ceval_state {
     _Py_atomic_int eval_breaker;
     /* Request for dropping the GIL */
     _Py_atomic_int gil_drop_request;
+    /* The GC is ready to be executed */
+    _Py_atomic_int gc_scheduled;
     struct _pending_calls pending;
 };
 

@@ -2252,6 +2252,14 @@ PyObject_IS_GC(PyObject *obj)
     return _PyObject_IS_GC(obj);
 }
 
+void
+_Py_ScheduleGC(PyInterpreterState *interp)
+{
+    struct _ceval_state *ceval = &interp->ceval;
+    _Py_atomic_store_relaxed(&ceval->gc_scheduled, 1);
+    _Py_atomic_store_relaxed(&ceval->eval_breaker, 1);
+}
+
 void
 _PyObject_GC_Link(PyObject *op)
 {
@@ -2269,12 +2277,19 @@ _PyObject_GC_Link(PyObject *op)
         !gcstate->collecting &&
         !_PyErr_Occurred(tstate))
     {
-        gcstate->collecting = 1;
-        gc_collect_generations(tstate);
-        gcstate->collecting = 0;
+        _Py_ScheduleGC(tstate->interp);
     }
 }
 
+void
+_Py_RunGC(PyThreadState *tstate)
+{
+    GCState *gcstate = &tstate->interp->gc;
+    gcstate->collecting = 1;
+    gc_collect_generations(tstate);
+    gcstate->collecting = 0;
+}
+
 static PyObject *
 gc_alloc(size_t basicsize, size_t presize)
 {

diff --git a/Modules/signalmodule.c b/Modules/signalmodule.c
@@ -1798,6 +1798,19 @@ int
 PyErr_CheckSignals(void)
 {
     PyThreadState *tstate = _PyThreadState_GET();
+
+    /* Opportunistically check if the GC is scheduled to run and run it
+       if we have a request. This is done here because native code needs
+       to call this API if is going to run for some time without executing
+       Python code to ensure signals are handled. Checking for the GC here
+       allows long running native code to clean cycles created using the C-API
+       even if it doesn't run the evaluation loop */
+    struct _ceval_state *interp_ceval_state = &tstate->interp->ceval;
+    if (_Py_atomic_load_relaxed(&interp_ceval_state->gc_scheduled)) {
+        _Py_atomic_store_relaxed(&interp_ceval_state->gc_scheduled, 0);
+        _Py_RunGC(tstate);
+    }
+
     if (!_Py_ThreadCanHandleSignals(tstate->interp)) {
         return 0;
     }

@@ -5,6 +5,7 @@
 #include "pycore_pyerrors.h"      // _PyErr_Fetch()
 #include "pycore_pylifecycle.h"   // _PyErr_Print()
 #include "pycore_initconfig.h"    // _PyStatus_OK()
+#include "pycore_interp.h"        // _Py_RunGC()
 #include "pycore_pymem.h"         // _PyMem_IsPtrFreed()
 
 /*
@@ -938,6 +939,13 @@ _Py_HandlePending(PyThreadState *tstate)
 {
     _PyRuntimeState * const runtime = &_PyRuntime;
     struct _ceval_runtime_state *ceval = &runtime->ceval;
+    struct _ceval_state *interp_ceval_state = &tstate->interp->ceval;
+
+    /* GC scheduled to run */
+    if (_Py_atomic_load_relaxed_int32(&interp_ceval_state->gc_scheduled)) {
+        _Py_atomic_store_relaxed(&interp_ceval_state->gc_scheduled, 0);
+        _Py_RunGC(tstate);
+    }
 
     /* Pending signals */
     if (_Py_atomic_load_relaxed_int32(&ceval->signals_pending)) {
@@ -947,20 +955,19 @@ _Py_HandlePending(PyThreadState *tstate)
     }
 
     /* Pending calls */
-    struct _ceval_state *ceval2 = &tstate->interp->ceval;
-    if (_Py_atomic_load_relaxed_int32(&ceval2->pending.calls_to_do)) {
+    if (_Py_atomic_load_relaxed_int32(&interp_ceval_state->pending.calls_to_do)) {
         if (make_pending_calls(tstate->interp) != 0) {
             return -1;
         }
     }
 
     /* GIL drop request */
-    if (_Py_atomic_load_relaxed_int32(&ceval2->gil_drop_request)) {
+    if (_Py_atomic_load_relaxed_int32(&interp_ceval_state->gil_drop_request)) {
         /* Give another thread a chance */
         if (_PyThreadState_Swap(&runtime->gilstate, NULL) != tstate) {
             Py_FatalError("tstate mix-up");
         }
-        drop_gil(ceval, ceval2, tstate);
+        drop_gil(ceval, interp_ceval_state, tstate);
 
         /* Other threads may run now */
 
@@ -989,7 +996,7 @@ _Py_HandlePending(PyThreadState *tstate)
     // value. It prevents to interrupt the eval loop at every instruction if
     // the current Python thread cannot handle signals (if
     // _Py_ThreadCanHandleSignals() is false).
-    COMPUTE_EVAL_BREAKER(tstate->interp, ceval, ceval2);
+    COMPUTE_EVAL_BREAKER(tstate->interp, ceval, interp_ceval_state);
 #endif
 
     return 0;