Bug 115693

Summary:

[1.5][compiler] Unnecessary double checkcast instruction emmited

Product:

[Eclipse Project] JDT

Reporter:

dpr

Component:

Core

Assignee:

Philipe Mulet <philippe_mulet>

Status:

VERIFIED FIXED

QA Contact:

Severity:

minor

Priority:

CC:

frederic_fusier, mlists

Version:

3.2

Target Milestone:

3.2 M4

Hardware:

OS:

Windows XP

Whiteboard:

Attachments:

Description	Flags
Test project	none
Correct bytecode	none
Incorrect bytecode with double checkcast	none

Description dpr

2005-11-09 15:27:10 EST

Just spotted a minor bytecode generation error. I'll try to attach a simple test
project. When this project is built after clean everithing is ok. But when class
D is recompiled incrementally a double checkcast instruction is emmited.

Comment 1 dpr

2005-11-09 15:28:18 EST

Created attachment 29635 [details]
Test project

Comment 2 dpr

2005-11-09 15:29:15 EST

Created attachment 29636 [details]
Correct bytecode

Comment 3 dpr

2005-11-09 15:30:07 EST

Created attachment 29637 [details]
Incorrect bytecode with double checkcast

Comment 4 Philipe Mulet

2005-12-05 06:51:32 EST

Good find, indeed the cast is overkill after the generic cast conversion.
Fixed in latest. Added GenericTypeTest#test874.
No plan to backport to 3.1.2 since really not severe.

Fixed

Comment 5 Frederic Fusier

2005-12-13 09:54:00 EST

Verified for 3.2 M4 using build I20051213-0010

Comment 6 Frederic Fusier

2005-12-13 14:45:14 EST

In fact, GenericTypeTests#test874 already passes with M3. So, it seems that the problem is fixed but we currently need to investigate in order to know precisely which changes has done it
=> reopen until clarified...

Comment 7 Frederic Fusier

2005-12-13 14:46:01 EST

Note also that test passes with 3.1.2 (ie. there's only one checkcast)

Comment 8 Philipe Mulet

2005-12-13 14:52:37 EST

It feels like this regression got introduced after M3; and given it isn't in 3.1.2, it would point at the unification of generic and parameterized bindings.

Need to understand what exactly occurred there; i.e. is fix really needed, or is it working around another issue ?

Comment 9 Philipe Mulet

2005-12-14 15:03:37 EST

The original testcase didn't fail in 3.1.2&3.2M3, since the return type is generic type (instead of parameterized type - pre binding unification); consequently, no generic cast is inserted as generic type interestingly doesn't have its tagbits saying HasTypeVariable true (!?).

In 3.2, it works as expected.

Here is another testcase which would fail in 3.1.2&3.2M3:

class A {}
class D extends A {}
abstract class B<T> {
    public T label(String s) { return null; }
}
final class C extends B<A> {
    public static C instance(String s) { return new C(); }
    @Override public String toString() {
    	return "SUCCESS"; 
    }
}
public class X {
    public static void main(String[] args) {
        D d = (D)C.instance("X").label("Y");
        System.out.println(d);
    }
}

It would issue a double checkcast D inside X#main().
Added GenericTypeTest#test878

Comment 10 Philipe Mulet

2005-12-14 15:25:16 EST

Fixed buildnotes (re-inserted), and tagged for M4.
Frederic - pls verify it.

Comment 11 Olivier Thomann

2005-12-14 19:52:15 EST

Verified in build I20051214-1600.

output is:

  // Method descriptor #15 ([Ljava/lang/String;)V
  // Stack: 2, Locals: 2
  public static void main(java.lang.String[] args);
     0  ldc <String "X"> [16]
     2  invokestatic C.instance(java.lang.String) : C [17]
     5  ldc <String "Y"> [23]
     7  invokevirtual C.label(java.lang.String) : java.lang.Object [25]
    10  checkcast D [29]
    13  astore_1 [d]
    14  getstatic java.lang.System.out : java.io.PrintStream [31]
    17  aload_1 [d]
    18  invokevirtual java.io.PrintStream.println(java.lang.Object) : void [37]
    21  return