202404 – [1.5][compiler] Incorrect handling of type variable bounds issued by capture conversion

Bug 202404 - [1.5][compiler] Incorrect handling of type variable bounds issued by capture conversion

Summary: [1.5][compiler] Incorrect handling of type variable bounds issued by capture ...

Status:	VERIFIED FIXED

Alias:	None

Product:	JDT
Classification:	Eclipse Project
Component:	Core (show other bugs)
Version:	3.4
Hardware:	PC Windows XP

Importance:	P3 normal (vote)
Target Milestone:	3.3.2
Assignee:	Philipe Mulet
QA Contact:

URL:
Whiteboard:
Keywords:

Duplicates (1):	179902 (view as bug list)
Depends on:
Blocks:	203318
	Show dependency tree

Reported:	2007-09-06 05:03 EDT by Philipe Mulet
Modified:	2008-01-24 07:09 EST (History)
CC List:	3 users (show)

See Also:

Flags:	philippe_mulet: pmc_approved+

Attachments
work in progress (15.39 KB, patch) 2007-09-12 12:15 EDT, Philipe Mulet	no flags	Details \| Diff
Proposed patch (13.60 KB, patch) 2007-09-13 16:13 EDT, Philipe Mulet	no flags	Details \| Diff
View All Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.

Description Philipe Mulet

2007-09-06 05:03:28 EDT

Build 3.3.0

The following program shouldn't raise any compiler error:

[X.java]
class A {}
class D<U extends A, V extends U> {}
public class X {
	void foo() {
		D<?, ? super A> d1 = null;
		D<?, ? extends A> d2 = null;
	}
}

The compiler currently rejects d2 declaration:
----------
1. ERROR in X.java (at line 6)
	D<?, ? extends A> d2 = null;
	     ^^^^^^^^^^^
Bound mismatch: The type ? extends A is not a valid substitute for the bounded parameter <V extends U> of the type D<U,V>
----------

Comment 1 Philipe Mulet

2007-09-06 05:04:39 EDT

(comment from spec lead)

Consider the generic type
   class D<X extends A, Y extends X>...

And the parameterized type
   D<?, ? super A>

which capture-converts to
   D<#1 extends A super null, #2 extends #1 super A>

This is legal because there is a possible instantiation for #1 and #2: A.

Now consider the parameterized type
   D<?, ? extends A>

which capture-converts to
   D<#1 extends A super null, #2 extends glb(A, #1) super null>

Whatever #1 is, it extends A by its own definition, so glb(A, #1) is 
trivially well-formed and #2 can be inhabited. So I think D<?, ? extends 
A> is legal.

Comment 2 Philipe Mulet

2007-09-12 12:14:50 EDT

The bound check is not truly performing a capture conversion, and rather tries to emulate it (to avoid creating captures on the fly). This results in slight differences in behavior.

Investigating.

Comment 3 Philipe Mulet

2007-09-12 12:15:31 EDT

Created attachment 78199 [details]
work in progress

This is not final. It breaks existing working scenarii.

Comment 4 Philipe Mulet

2007-09-12 12:20:04 EDT

Added GenericTypeTest#test1158-1162

Comment 5 Philipe Mulet

2007-09-13 15:22:25 EDT

Also see bug 159214

Comment 6 Philipe Mulet

2007-09-13 16:13:16 EDT

Created attachment 78381 [details]
Proposed patch

Alternate solution, better aligned with existing story. Basically, it doesn't induce more captures to be created, but simply tune the bound check test to better emulate the true capture.

Note that this patch is combined with the changes for bug 159214 and bug 203318 since all these issues are intricated.

Comment 7 Philipe Mulet

2007-09-13 18:29:30 EDT

Released for 3.4M2.
Fixed

Comment 8 Philipe Mulet

2007-09-13 18:42:17 EDT

*** Bug 179902 has been marked as a duplicate of this bug. ***

Comment 9 David Audel

2007-09-18 06:28:40 EDT

Verified for 3.4M2 using build I20070918-0010

Comment 10 Philipe Mulet

2007-10-10 04:23:33 EDT

Released for 3.3.2
Fixed

Comment 11 Philipe Mulet

2007-10-10 04:24:03 EDT

reopening to remove verify state

Comment 12 Philipe Mulet

2007-10-10 04:24:30 EDT

Released for 3.3.2
Fixed

Comment 13 Maxime Daniel

2007-10-29 12:41:17 EDT

Only adding Verified for 3.4M2 to the status whiteboard, since I believe this got lost.

Comment 14 Jerome Lanneluc

2007-11-22 10:16:58 EST

The compiler semantics near capture conversion need to be tuned to match the spec. Without these changes, we see quite many corner cases failing to compile properly or being rejected incorrectly, and numerous customers starting to notice it.

Comment 15 Philipe Mulet

2007-11-23 09:18:50 EST

+1 for 3.3.2

Comment 16 Frederic Fusier

2008-01-24 07:09:02 EST

Verified for 3.3.2 using build M20080123-0800.