Properly raise div-by-zero errors in more cases under LSL

Author: HaroldCindy

VM/src/lvmexecute.cpp

@@ -208,7 +208,7 @@ inline bool luau_skipstep(uint8_t op)
 }
 
 
-static const char * const MATH_ERROR_STR = "Math error: division or modulus (possibly by 0) resulted in NaN";
+static const char * const MATH_ERROR_STR = "Math error: division or modulus by 0";
 
 template<bool SingleStep>
 static void luau_execute(lua_State* L)
@@ -1774,9 +1774,8 @@ static void luau_execute(lua_State* L)
                     // ServerLua: ints not represented in here, they only support idiv.
                     const double res = nvalue(rb) / nvalue(rc);
 
-                    // ServerLua: Any division that results in NaN is an error under Mono
-                    // Generally this is div by zero, but there might be others.
-                    if (LUAU_UNLIKELY(LUAU_IS_LSL_VM(L) && std::isnan(res)))
+                    // ServerLua: In LSL (Mono), division by zero or a NaN result is an error.
+                    if (LUAU_UNLIKELY(LUAU_IS_LSL_VM(L) && (nvalue(rc) == 0.0 || std::isnan(res))))
                     {
                         // VM protection is necessary here because PC may be invalid if we've
                         // not called `VM_PROTECT()` in this function yet.
@@ -2092,12 +2091,26 @@ static void luau_execute(lua_State* L)
                 // fast-path
                 if (LUAU_LIKELY(ttisnumber(rb)))
                 {
-                    setnvalue(ra, nvalue(rb) / nvalue(kv));
+                    const double res = nvalue(rb) / nvalue(kv);
+
+                    // ServerLua: In LSL (Mono), division by zero or a NaN result is an error.
+                    if (LUAU_UNLIKELY(LUAU_IS_LSL_VM(L) && (nvalue(kv) == 0.0 || std::isnan(res))))
+                    {
+                        VM_PROTECT(luaG_runerrorL(L, MATH_ERROR_STR));
+                    }
+
+                    setnvalue(ra, res);
                     VM_NEXT();
                 }
                 // ServerLua: int intentionally not handled, it uses idivk
                 else if (ttisvector(rb))
                 {
+                    // ServerLua: In LSL (Mono), vector division by zero is a runtime error.
+                    if (LUAU_UNLIKELY(LUAU_IS_LSL_VM(L) && nvalue(kv) == 0.0))
+                    {
+                        VM_PROTECT(luaG_runerrorL(L, MATH_ERROR_STR));
+                    }
+
                     const float* vb = vvalue(rb);
                     float nc = cast_to(float, nvalue(kv));
                     setvvalue(ra, vb[0] / nc, vb[1] / nc, vb[2] / nc, vb[3] / nc);
@@ -3113,7 +3126,15 @@ static void luau_execute(lua_State* L)
                 // fast-path
                 if (LUAU_LIKELY(ttisnumber(rc)))
                 {
-                    setnvalue(ra, nvalue(kv) / nvalue(rc));
+                    const double res = nvalue(kv) / nvalue(rc);
+
+                    // ServerLua: In LSL (Mono), division by zero or a NaN result is an error.
+                    if (LUAU_UNLIKELY(LUAU_IS_LSL_VM(L) && (nvalue(rc) == 0.0 || std::isnan(res))))
+                    {
+                        VM_PROTECT(luaG_runerrorL(L, MATH_ERROR_STR));
+                    }
+
+                    setnvalue(ra, res);
                     VM_NEXT();
                 }
                 else if (ttisvector(rc))

tests/LSL.test.cpp

@@ -1082,9 +1082,13 @@ TEST_CASE("Integer MulAssign Float Valid Contexts")
     runConformance("mul_assign_valid.lsl");
 }
 
-TEST_CASE("Integer Division By Zero Error")
+TEST_CASE("Division By Zero Error")
 {
     runConformance("div_by_zero.lsl", nullptr, nullptr, nullptr, "Math error");
+    runConformance("float_div_by_zero.lsl", nullptr, nullptr, nullptr, "Math error");
+    runConformance("vector_div_by_zero.lsl", nullptr, nullptr, nullptr, "Math error");
+    runConformance("zero_vector_div_by_zero.lsl", nullptr, nullptr, nullptr, "Math error");
+    runConformance("inf_div_inf.lsl", nullptr, nullptr, nullptr, "Math error");
 }
 
 TEST_CASE("Mono Strings")

tests/conformance/float_div_by_zero.lsl

@@ -0,0 +1,9 @@
+// Test that float division by zero throws a runtime error
+default {
+    state_entry() {
+        float a = 1.0;
+        float b = 0.0;
+        float c = a / b;  // This should error
+        print((string)c);
+    }
+}

tests/conformance/inf_div_inf.lsl

@@ -0,0 +1,8 @@
+// Test that Infinity / Infinity throws a runtime error (NaN result)
+default {
+    state_entry() {
+        float inf = (float)"inf";
+        float result = inf / inf;  // This should error
+        print((string)result);
+    }
+}

tests/conformance/vector_div_by_zero.lsl

@@ -0,0 +1,7 @@
+// Test that vector division by zero throws a runtime error
+default {
+    state_entry() {
+        vector v = <1, 1, 1> / 0;  // This should error
+        print((string)v);
+    }
+}

tests/conformance/zero_vector_div_by_zero.lsl

@@ -0,0 +1,7 @@
+// Test that zero vector division by zero throws a runtime error
+default {
+    state_entry() {
+        vector v = <0, 0, 0> / 0;  // This should error
+        print((string)v);
+    }
+}