Implement Java Streams

[gzm0]

Following up from
#4925 (comment)_

IMHO Java streams are sufficiently important that if we cannot support them, we should at the very least formally document that (and why).

[gzm0]

I've had a look at the java.util.stream package and I really do not see any major reason why we cannot support it.

Of course, we will not be able to offer parallel streams. This means, we should probably not even define all the parallel methods.

There is the (unfortunate) nuisance that the methods on StreamSupport take parallel as an argument. So we'd have to resort to throwing UnsupportedOperationException in this case (which given how low-level the class is, seems acceptable to me).

@sjrd, what am I missing? :)

[gzm0]

Note on Collection#parallelStream.

Returns a possibly parallel Stream with this collection as its source. It is allowable for this method to return a sequential stream.

So just ignoring request for parallelism might be just fine.

[sjrd]

I had another look and I cannot find anything that is a total deal-breaker. The StreamSupport issue is annoying, but perhaps we can simply not provide StreamSupport at all. It is a static-only helper class, which is not interleaved in the rest of the API.

The fact that parallel() is abstract in BaseStream and its public subinterfaces means that third-party code will have trouble implementing their own stream subclasses: if they don't declare an implementation of parallel(), the compiler will complain that it is left abstract; but if they do, they inevitably break its contract. Some IR post-processing may fix that issue, but that's harsh to ask of third-party implementers.

Note on Collection#parallelStream.

Returns a possibly parallel Stream with this collection as its source. It is allowable for this method to return a sequential stream.

So just ignoring request for parallelism might be just fine.

It does have an Implementation Requirements section that says that the default implementation returns a parallel stream, so I think it is still off-limits.

[gzm0]

The StreamSupport issue is annoying, but perhaps we can simply not provide StreamSupport at all. It is a static-only helper class, which is not interleaved in the rest of the API.

Fair enough. Feels to me like we definitely want something like this to implement streams, but at least for starters, it can be private to the javalib without the parallel parameter.

Some IR post-processing may fix that issue, but that's harsh to ask of third-party implementers.

True. But I guess we agree that providing streams simply for usage is worth it in its own right at least as a first step?

It does have an Implementation Requirements section that says that the default implementation returns a parallel stream, so I think it is still off-limits.

True :-/

In any case, do you agree that based on this, we should re-open #4925?

[sjrd]

Some IR post-processing may fix that issue, but that's harsh to ask of third-party implementers.

True. But I guess we agree that providing streams simply for usage is worth it in its own right at least as a first step?

Yes, it's probably viable for usage-only at first.

In any case, do you agree that based on this, we should re-open #4925?

Reopened.