Shared backend

[DarkDimius]

Can compile hello world, can compile patterns and tailrec methods. Can compile array operations. Can compile both classes and objects.

[DarkDimius]

Just for fun:

dark@lamppc17 ~/workspace/dotty/tests/pos (shared-backend) $ time scalac HelloWorld.scala

real    0m1.546s
user    0m4.674s
sys 0m0.317s
dark@lamppc17 ~/workspace/dotty/tests/pos (shared-backend) $ time dotc HelloWorld.scala
warning: encountered F-bounded higher-kinded type parameters for type scala$collection$generic$GenMapFactory$$CC; assuming they are invariant
warning: encountered F-bounded higher-kinded type parameters for type scala$collection$generic$MapFactory$$CC; assuming they are invariant
warning: encountered F-bounded higher-kinded type parameters for type scala$collection$generic$GenSetFactory$$CC; assuming they are invariant
warning: encountered F-bounded higher-kinded type parameters for type scala$collection$generic$SetFactory$$CC; assuming they are invariant
warning: encountered F-bounded higher-kinded type parameters for type scala$collection$generic$ImmutableMapFactory$$CC; assuming they are invariant
warning: encountered F-bounded higher-kinded type parameters for type scala$collection$generic$ImmutableSetFactory$$CC; assuming they are invariant
6 warnings found

real    0m1.422s
user    0m3.931s
sys 0m0.346s

and bytecode is the same

diff --git a/HelloWorld$.class b/HelloWorld$.class
index d2b517a..bf3b04f 100644
--- a/HelloWorld$.class
+++ b/HelloWorld$.class
@@ -11,6 +11,13 @@ public final class HelloWorld$ {
 ##: invokespecial ###                 // Method "<init>":()V
 ##: return

+  public HelloWorld$();
+    descriptor: ()V
+    Code:
+##: invokespecial ###                 // Method java/lang/Object."<init>":()V
+##: putstatic     ###                 // Field MODULE$:LHelloWorld$;
+##: return
+
   public void main(java.lang.String[]);
     descriptor: ([Ljava/lang/String;)V
     Code:
@@ -18,11 +25,4 @@ public final class HelloWorld$ {
 ##: ldc           ###                 // String hello world
 ##: invokevirtual ###                 // Method scala/Predef$.println:(Ljava/lang/Object;)V
 ##: return
-
-  private HelloWorld$();
-    descriptor: ()V
-    Code:
-##: invokespecial ###                 // Method java/lang/Object."<init>":()V
-##: putstatic     ###                 // Field MODULE$:LHelloWorld$;
-##: return
 }

[DarkDimius]

Status update: Arrays and <label> DefDefs are supported(can compile while cycles, patterns, and tailrec methods).
SeqLiterals and Closures aren't handled.

And there are some errors that I see why applying Identity tranformer that just collects information:

./src/dotty/tools/dotc/core/Uniques.scala:25: error: wrong number of parameters for ($this: dotty.tools.dotc.core.Uniques$(dotty.tools.dotc.core.Uniques), tp: dotty.tools.dotc.core.Types.dotty$tools$dotc$core$Types$$Type)Unit(
  dotty.tools.dotc.core.Uniques
.recordCaching); expected: 2
    if (monitored) recordCaching(tp)

It seems that we have a phase that did FullParametrization but didn't transform actual calls. I guess this is PrivateToStatic that has a transformApply that assumes that fun is a Select while it can also be an Ident. Will get to root of it soon.

[odersky]

Did we try to fuse with previous group ending in RestoreScopes?

[odersky]

I only checked the patches to previously existing files in detail, and did a quick browse over the backend code. It would be good if somebody who knows the backend could comment. But I agree we should not wait for a long time for it to happen.

Except for the remarks I added, LGTM as far as I am concerned.

[DarkDimius]

I've created several issues to track what stops backend from working on other phases: #289 and #288
There's also seems to be a bug in the backend itself that manifests itself late in asm by IndexOutOfBounds exception. It seems that I break the stack somehow. I'll need to track this down.

In the meanwhile I've enabled quite a few tests that compile to bytecode without errors.

[DarkDimius]

I haven't actually changed any backend code here for a week already, all changes are in phases. So I propose to merge.

[samuelgruetter]

Since this PR, it's not possible any more to compile dotty with Java 7; Java 8 is needed. I guess this is intended and not considered a problem, right?

[odersky]

On Thu, Dec 18, 2014 at 5:23 PM, Samuel Gruetter [email protected]
wrote:

Since this PR, it's not possible any more to compile dotty with Java 7;
Java 8 is needed. I guess this is intended and not considered a problem,
right?

Yes, it's inevitable. We said from the start we'd be Java 8 only.

• Martin

—
Reply to this email directly or view it on GitHub
#251 (comment).

Martin Odersky
EPFL