Took a different tack on avoiding unnecessarily...

Took a different tack on avoiding unnecessarily final bytecode methods
after discovering the inliner is not a fan of any reduction in finality.
Restored the application of the lateFINAL flag in makeNotPrivate, and
switched to sorting out who is ACC_FINAL and who isn't in genjvm. Review
by dragos.

1 files changed, 7 insertions, 7 deletions

diff --git a/src/compiler/scala/reflect/internal/Symbols.scala b/src/compiler/scala/reflect/internal/Symbols.scala
index 3167bbc5de..30e9f38cbe 100644
--- a/src/compiler/scala/reflect/internal/Symbols.scala
+++ b/src/compiler/scala/reflect/internal/Symbols.scala
@@ -1644,13 +1644,13 @@ trait Symbols extends api.Symbols { self: SymbolTable =>
     final def makeNotPrivate(base: Symbol) {
       if (this.isPrivate) {
         setFlag(notPRIVATE)
-        // Marking these methods final causes problems for proxies
-        // which use subclassing.  If people write their code with no
-        // usage of final, we probably shouldn't introduce it ourselves
-        // unless we know it is safe.
-        //
-        // if (isMethod && !isDeferred)
-        //   setFlag(lateFINAL)
+        // Marking these methods final causes problems for proxies which use subclassing. If people
+        // write their code with no usage of final, we probably shouldn't introduce it ourselves
+        // unless we know it is safe. ... Unfortunately if they aren't marked final the inliner
+        // thinks it can't inline them. So once again marking lateFINAL, and in genjvm we no longer
+        // generate ACC_FINAL on "final" methods which are actually lateFINAL.
+        if (isMethod && !isDeferred)
+          setFlag(lateFINAL)
         if (!isStaticModule && !isClassConstructor) {
           expandName(base)
           if (isModule) moduleClass.makeNotPrivate(base)