Use a stable ordering for renamed symbols

We should generate the exact same bytecode regardless of the order in which
files are loaded. Relying on Symbol.id for ordering reflects the order in which
symbols were accessed, and that depends on what files are compiled. Instead,
we should use source positions wherever possible, since all the renamable symbols
originate from the currently compiled unit.

src/compiler/scala/tools/nsc/transform/LambdaLift.scala

@@ -64,14 +64,21 @@ abstract class LambdaLift extends InfoTransform {
     /** Symbols that are called from an inner class. */
     private val calledFromInner = new LinkedHashSet[Symbol]
 
-    private val ord = Ordering.fromLessThan[Symbol](_ isLess _)
-    private def newSymSet = TreeSet.empty[Symbol](ord)
+    private val posord = Ordering.fromLessThan[Symbol] { (sym1: Symbol, sym2: Symbol) =>
+      if (sym1.pos.isDefined && sym2.pos.isDefined
+        && sym1.pos.source == sym2.pos.source
+        && sym1.pos.point != sym2.pos.point)
+        sym1.pos.point < sym2.pos.point
+      else
+        sym1 isLess sym2
+    }
+    private def newSymSet(ord: Ordering[Symbol]) = TreeSet.empty[Symbol](ord)
 
     private def symSet(f: LinkedHashMap[Symbol, SymSet], sym: Symbol): SymSet =
-      f.getOrElseUpdate(sym, newSymSet)
+      f.getOrElseUpdate(sym, newSymSet(Ordering.fromLessThan(_ isLess _)))
 
     /** The set of symbols that need to be renamed. */
-    private val renamable = newSymSet
+    private val renamable = newSymSet(posord)
 
     /**
      * The new names for free variables proxies. If we simply renamed the