Bug in javac ?

[Andrew Haley]

On 10/31/2011 09:49 PM, Dr Andrew John Hughes wrote:
> On 16:48 Mon 31 Oct     , Andrew Haley wrote:
>>
>>> Am 31.10.2011 17:00, schrieb Deepak Bhole:
>>>
>>>> It looks like a known bug in the 6 compiler related to interface 
>>>> inheritance and covariant return types. I think this is the commit 
>>>> that fixed it in 7:
>>>>
>>>> http://hg.openjdk.java.net/jdk7/jdk7/langtools/rev/4a3b9801f7a0
>>>>
>>>> If you have code that resembles the above and would like to see the
>>>> fix backported to 6, please feel free to open a bug.
>>>
>>> Thanks. Bug is open (#750301)
>>
>> This one should be submitted to the tools people @ Oracle rather than
>> just patched into Icedtea, I think.  Their input might avoid us breaking
>> anything.
> 
> It's worth a try, but they've been pretty unhelpful in the past.  The
> OpenJDK6 toolset is a mess (being some fairly random point in the
> development of the 7 compiler) and their main contribution has being
> to break the TCK run.

Perhaps, but we really need their input for anything non-trivial on
the compiler.  The type inference logic is very convoluted, and it is
extremely difficult for anyone to look at a patch that fixes one thing
and know that it doesn't break something else.  It's really easy for a
fix to "work" for, say, building JBoss but not Eclipse.

It's arguable whether a patch for something that has been wrong for
the whole life of OpenJDK 6 should go in now.

> On that note, a good initial test for this would be to see if the patched
> version passes the TCK.  Then post to compiler-dev and jdk6-dev.

Right.  That's the right way to start.

Andrew.