JS passes should remove local vars that are written but not read

[niloc132]

On the Java side, ControlFlowAnalyzer looks at local variable refs across the program, finding writes with no reads, and removing them. Nothing seems to do this today on the JS side though, which can result in extra local vars that are assigned but unread or never even assigned.

In the example given by #10399, the inlined local variable directHandlers is only assigned but never read, and should be entirely removed.

[niloc132]

gwt/user/src/com/google/gwt/dom/client/DOMImplMozilla.java

Lines 128 to 142 in 444e4d1

	public native EventTarget eventGetRelatedTarget(NativeEvent evt) /*-{
	// Hack around Mozilla bug 497780 (relatedTarget sometimes returns XUL
	// elements). Trying to access relatedTarget.nodeName will throw an
	// exception if it's a XUL element.
	var relatedTarget = evt.relatedTarget;
	if (!relatedTarget) {
	return null;
	}
	try {
	var nodeName = relatedTarget.nodeName;
	return relatedTarget;
	} catch (e) {
	return null;
	}
	}-*/;

A fix for this optimizes out nodeName in the code above, and then JsStaticEval removes the expression itself, which removes the purpose of the code - the exception can't be thrown if the code is missing.

On the other hand, https://bugzilla.mozilla.org/show_bug.cgi?id=497780 was closed 17 years ago, so the whole thing should be removed.

--

gwt/user/super/com/google/gwt/emul/java/lang/Class.java

Lines 173 to 195 in 444e4d1

	private static native void maybeSetClassLiteral(JavaScriptObject typeId, Class<?> clazz) /*-{
	var proto;
	if (!typeId) {
	// Type is not instantiable, hence not registered in the metadata table.
	return;
	}
	clazz.@Class::typeId = typeId;
	// Guarantees virtual method won't be pruned by using a JSNI ref
	// This is required because deRPC needs to call it.
	var prototype = @Class::getPrototypeForClass(Ljava/lang/Class;)(clazz);
	// A class literal may be referenced prior to an async-loaded vtable setup
	// For example, class literal lives in inital fragment,
	// but type is instantiated in another fragment
	if (!prototype) {
	// Leave a place holder for now to be filled in by __defineClass__ later.
	// TODO(rluble): Do not rely on the fact that if the entry is an array it is a placeholder.
	@com.google.gwt.lang.Runtime::prototypesByTypeId[typeId] = [clazz];
	return;
	}
	// Type already registered in the metadata table, install the class literal in the appropriate
	// prototype field.
	prototype.@java.lang.Object::___clazz = clazz;
	}-*/;

This method declares var proto, but never assigns it. The last time this was assigned was in 2453402, but the compiler hasn't been capable of removing that empty var.

--

These are both easy to fix directly, but improving the compiler so they don't matter to begin with is even better, and improves all apps.