fix [#28114] Render Crash

existing check for driver to use GIL was not thread safe and could cause, details in the report.

This bug was caused by a check to avoid hanging, a fix for [#27683] that worked in 2.4x because the UI didn't use python to draw while rendering.

Apply a different fix for [#27683], when calling an operator, call PyEval_SaveThread(), then PyEval_RestoreThread() so the GIL can be aquired by threads started by the operator - in this case bake starting a thread that evaluates drivers.

1 files changed, 20 insertions, 1 deletions

diff --git a/source/blender/python/intern/bpy_operator.c b/source/blender/python/intern/bpy_operator.c
index b8883e655f2..4a17c45ae38 100644
--- a/source/blender/python/intern/bpy_operator.c
+++ b/source/blender/python/intern/bpy_operator.c
@@ -55,6 +55,10 @@
 #include "BKE_report.h"
 #include "BKE_context.h"
 
+/* so operators called can spawn threads which aquire the GIL */
+#define BPY_RELEASE_GIL
+
+
 static PyObject *pyop_poll(PyObject *UNUSED(self), PyObject *args)
 {
 	wmOperatorType *ot;
@@ -219,7 +223,22 @@ static PyObject *pyop_call(PyObject *UNUSED(self), PyObject *args)
 			reports= MEM_mallocN(sizeof(ReportList), "wmOperatorReportList");
 			BKE_reports_init(reports, RPT_STORE | RPT_OP_HOLD); /* own so these dont move into global reports */
 
-			operator_ret= WM_operator_call_py(C, ot, context, &ptr, reports);
+#ifdef BPY_RELEASE_GIL
+			/* release GIL, since a thread could be started from an operator
+			 * that updates a driver */
+			/* note: I havve not seen any examples of code that does this
+			 * so it may not be officially supported but seems to work ok. */
+			{
+				PyThreadState *ts= PyEval_SaveThread();
+#endif
+
+				operator_ret= WM_operator_call_py(C, ot, context, &ptr, reports);
+
+#ifdef BPY_RELEASE_GIL
+				/* regain GIL */
+				PyEval_RestoreThread(ts);
+			}
+#endif
 
 			error_val= BPy_reports_to_error(reports, PyExc_RuntimeError, FALSE);