pausing multiple times with n_run can crash scsynth

[Sciss]

Environment

• SuperCollider version: 3.10.2, 3.10.4
• Operating system: Linux

Steps to reproduce

s.waitForBoot { fork {
  s.sendBundle(nil,
		["/g_new", 1009, 0, 0],
		["/n_run", 1009, 0],
	);
  SynthDef("foo", {
    Out.ar(0, DC.ar(0.0))
	}).send(s);
	s.sync;
  s.sendBundle(nil,
		["/s_new", "foo", 1011, 1, 1009]
	);
	1.0.wait;
	"tada!".postln;
	s.sendBundle(nil,
		["/n_run", 1011, 0],
		["/n_run", 1011, 0], // this is needed. SO WEIRD
		["/n_run", 1011, 1],
		["/n_run", 1009, 1],
	);
}};

Expected vs. actual behavior

The above example crashes the server (when sending the last bundle after "tada"). It makes a difference whether the synth is "paused twice" or not. I.e. if the line "this is needed" is removed, scsynth no longer crashes. Seems quite a fundamental/severe bug to me. We tried to different versions and computer so far.

[mossheim]

@Sciss i think this patch will fix it:

diff --git a/server/scsynth/SC_Node.cpp b/server/scsynth/SC_Node.cpp
index 92a6236a5..520f52af4 100644
--- a/server/scsynth/SC_Node.cpp
+++ b/server/scsynth/SC_Node.cpp
@@ -260,7 +260,9 @@ void Graph_NullFirstCalc(Graph* inGraph);
 void Node_SetRun(Node* inNode, int inRun) {
     if (inRun) {
-        if (inNode->mCalcFunc == &Node_NullCalc) {
+        if (inNode->mCalcFunc == &Node_NullCalc || inNode->mCalcFunc == (NodeCalcFunc)&Graph_NullFirstCalc) {
             if (inNode->mIsGroup) {
                 inNode->mCalcFunc = (NodeCalcFunc)&Group_Calc;
+            } else if (inNode->mCalcFunc == (NodeCalcFunc)&Graph_NullFirstCalc) {
+                inNode->mCalcFunc = (NodeCalcFunc)&Graph_FirstCalc;
             } else {
                 inNode->mCalcFunc = (NodeCalcFunc)&Graph_Calc;
@@ -270,5 +272,5 @@ void Node_SetRun(Node* inNode, int inRun) {
     } else {
         if (inNode->mCalcFunc != &Node_NullCalc) {
-            if (!inNode->mIsGroup && inNode->mCalcFunc == (NodeCalcFunc)&Graph_FirstCalc) {
+            if (!inNode->mIsGroup && (inNode->mCalcFunc == (NodeCalcFunc)&Graph_FirstCalc || inNode->mCalcFunc == (NodeCalcFunc)&Graph_NullFirstCalc)) {
                 inNode->mCalcFunc = (NodeCalcFunc)&Graph_NullFirstCalc;
             } else {

basically we need to keep better track of the 'first-calc'-ness of the node. there could be more bugs in this code, but it passes your test case. if i have more time to test this week i will submit a PR. thanks for the report!