[Interpreter] Marks that 'throw' has returned a value.

src/interpreter/bytecode-generator.cc

Index: src/interpreter/bytecode-generator.cc
diff --git a/src/interpreter/bytecode-generator.cc b/src/interpreter/bytecode-generator.cc
index 96b513c8b065e14b13ab504e113d8fd6ab9387b0..ada82a7c0043f053b3fc4fa08e9a4bb4cd7a115e 100644
--- a/src/interpreter/bytecode-generator.cc
+++ b/src/interpreter/bytecode-generator.cc
@@ -1539,6 +1539,14 @@ void BytecodeGenerator::VisitYield(Yield* expr) { UNIMPLEMENTED(); }
 void BytecodeGenerator::VisitThrow(Throw* expr) {
   VisitForAccumulatorValue(expr->exception());
   builder()->Throw();
+  // For script and eval scopes expression statements are converted to
+  // assignment statements in Rewriter::ReWrite pass. An assignment statement
+  // expects a value in the accumulator. This is a hack to avoid DCHECK fails
+  // assert accumulator has been set.
+  if (info()->literal()->scope()->is_script_scope() ||

[Michael Starzinger, 2016/01/18 21:23:03]

I wouldn't even narrow this to script or eval scopes, but in general just mark
all throw expressions as delivering the completion value in the accumulator
unconditionally.

As discussed offline today: This is a short-coming of our AST, where "throw"is
modeled as an expression instead of a statement (as the spec defines it),
exactly because of the internal replacement we perform. With proper
do-expressions such a rewriting could be done without this hack AFAICT. But that
is cleanup for another day.

[mythria, 2016/01/19 10:48:04]

Done.

+      info()->literal()->scope()->is_eval_scope()) {
+    execution_result()->SetResultInAccumulator();
+  }
 }