Bug 199527 – Fix DFGArgumentEliminationPhase's NewArrayBuffer check

• Component: JavaScriptCore
• Status: NEW

[Michael Catanzaro]

Coverity has flagged this code from r226033 "[FTL] NewArrayBuffer should be sinked if it is only used for spreading" in ArgumentsEliminationPhase::eliminateCandidatesThatEscape as a likely copy/paste error:

                case LoadVarargs:
                    if (node->loadVarargsData()->offset && (node->child1()->op() == NewArrayWithSpread || node->child1()->op() == Spread || node->child1()->op() == NewArrayBuffer))
                        escape(node->child1(), node);
                    break;
                    
                case CallVarargs:
                case ConstructVarargs:
                case TailCallVarargs:
                case TailCallVarargsInlinedCaller:
                    escape(node->child1(), node);
                    escape(node->child2(), node);
                    if (node->callVarargsData()->firstVarArgOffset && (node->child3()->op() == NewArrayWithSpread || node->child3()->op() == Spread || node->child1()->op() == NewArrayBuffer)) // <-- likely error here
                        escape(node->child3(), node);
                    break;

It looks like this was probably supposed to be node->child3()->op() == NewArrayBuffer, but I don't pretend to understand anything under Source/JavaScriptCore. Yusuke should look at it.

[Saam Barati]

Yeah I think coverity is right!

[Radar WebKit Bug Importer]

<rdar://problem/55507104>