RFC: Refactoring SystemABI

[Maurizio Cimadamore]

On 04/12/2018 15:51, Jorn Vernee wrote:
> Maybe this concern is being raised prematurely. I think we can see how 
> the current plans pans out and then look at this again if needed.
>
Yes, let's revisit this later. I have a couple of different ideas on how 
to get there, but it doesn't make a lot of sense to discuss them here, 
as it's probably too early - and I don't want to add to the confusion.

To recap (for Henry): let's have SystemABI::upcallStub return a Symbol 
and take it from there.

Thanks for your comments, as usual.

Maurizio

> Jorn
>
> Jorn Vernee schreef op 2018-12-04 16:34:
>> Maurizio Cimadamore schreef op 2018-12-04 16:07:
>>> On 04/12/2018 14:23, Jorn Vernee wrote:
>>>> We have Scope::allocateCallback to replace step 2. and 3., but that 
>>>> requires an annotated functional interface object. What if a user 
>>>> just has a MethodHandle?
>>>
>>> Not sure I follow - a Callback is based on the idea that you _do have_
>>> a functional interface.
>>
>> The scenario I'm talking about still uses a functional interface
>> _type_. But the Scope::allocateCallback api expects _an instance_ of
>> such a functional interface type; So if you just have a MethodHandle
>> you would first have to turn it into an instance of this functional
>> interface type and then call Scope::allocateCallback. This adds a
>> level of indirection, and AFAIK there's no straight forward API for
>> wrapping a MethodHandle in a functional interface in the JDK
>> (MethodHandleProxies comes close but it relies on reflection, so it's
>> slow).
>>
>> So what I'm talking about is that a user can use System::upcallStub to
>> make the MethodHandle callable from native, but if they have some
>> other method that accepts a Callback there still needs to be a way to
>> convert the value returned from upcallStub into a Callback.
>>
>>> I don't think it makes sense to mix the abstraction levels - the
>>> binder will represent all callbacks using Callback, so I don't expect
>>> a user of the high-level binder to mess with SystemABI interface to
>>> allocate stubs.
>>
>> I was thinking about the inverse; we have a user who is using the
>> low-level SystemABI, and wants to have the high-level addition of
>> Callback. This also makes sense where low-level code has to interact
>> with high-level code.
>>
>> Jorn