[kaffe] generic collections

[Dalibor Topic]

Hallo Matthias,

[Hi Tom, Eric & Martin, I've CC:ed you to hear your plans for adding 
support for Generics in your compilers]

Matthias Pfisterer wrote:

> Yea, I'm already wondering: are you working 24 hours a day for kaffe?

No, not really. I work hard on keeping the image, though ;)

> Well, in general you're right. There is a early access version of the 
> JDK 1.5.0 (from java.sun.com) without docs. In the case of generics, the 
>  list of affected classes in the collection and reflection APIs has long 
> been published as part of the public review draft of JSR-14 
> (http://jcp.org/aboutJava/communityprocess/review/jsr014/index.html). 
> The license allows "internal review", but one could arguement that part 
> of an internal review is to check how easy the specification can be 
> implemented by doing some actual coding, as long as the result is not 
> officially released. BTW, the first official beta of the JDK 1.5.0 is 
> expected for end of february.

The trouble is, anyone wanting to follow what a developer interested in 
generics does would have to agree to the same 'internal review' terms, 
which would make discussing it in public (i.e. externally, on mailing 
lists for example) quite hard for everyone involved, I guess. Something 
we can not talk about, is kind of hard to implement and evaluate 
collaboratively, spirit of GPL and all that. ;)

>>> I noticed that kjc can already handle generics. However, the propably 
>>> bigger problem might be the collection framework.
>>
>>
>>
>> I don't know how well kjc supports generics in the form in which they 
>> are supposed to be used in JDK 1.5. I also don't know if there is any 
>> development on jikes to support new java language constructs from 1.5. 
>> As far as I know, there is some work going on in the gcj camp to 
>> support those features in the future.
> 
> 
> Yes, I just found a bug in kjc's generics support! On the other hand, it 
> is basically working, while I had no success with gcj and jikes.

I know that the kjc developers would be glad to hear about bugs in their 
compiler implementation, so send those reports over to the kjc mailing 
lists ;)

>> So the first problem, once 1.5 is out, and it becomes clear what the 
>> specs demand, will be to make sure the build tools can cope with it. 
>> Then we can move on further to actually implement the spec ;)
> 
> 
> I agree that we have to make sure the tool chain supports generics. But 
> we won't find out and improve the tools without playing with generics. 
> The kjc bug mentioned above is an example of this: I wrote a simple list 
> class to test the generics support and zong! encountered a bug. So I 
> think we have to start implementing generics to get the tools right.

I think the JACKS compiler testing suite would be a great place to 
contribute such tests. I believe it's being used by all major open 
source java compiler developers to check their implemenations against 
the language specs.

I hope Tom Tromey from gcj can chip in with some details on what his 
plans for generics in gcj are, as well as other open source java 
compiler authors (I've CC:ed Eric Blake from Jikes, as well as Martin 
Lackner from kjc).

>> Check out the Classpath web pages for requirements, and get your 
>> paperwork done with the FSF to contribute the code to GNU Classpath. 
>> Then all free runtimes (not just kaffe) can profit from your work on 
>> reimplementing the collections framework.
> 
> 
> Ok, I'll do that.

Welcome to the cabal :)

cheers,
dalibor topic