1) I agree that a Java wrapper is preferred, and I'd contemplate a JNI 
approach only the wrapper approach hit some unsurmountable roadblock
2) A separate repository requires overhead and introduces opportunities for 
breakage. As I said before, a single repo approach helps avoid breaking 
behavior from slipping into the C implementation. It should also make it 
less likely that a gem update is released for the C derived platforms 
without a corresponding update for the JRuby variant.
3-5) Per Mike's feedback about the Google Cloud gem's requirements, I'd 
want to get all the existing tests to pass that are needed to make a viable 
client, and I'd be willing to settle for raising an exception that the 
operation is not supported on the rest of the API as a starting point. The 
eventual goal would be to get complete API compatibility, but a client-only 
approach would unblock the use of the Google Cloud by JRuby applications 
the fastest, and that is my core motivation.

I'm willing to contribute a proof of concept PR into the existing GRPC 
repository. I haven't started such a branch yet, as I didn't want to start 
such an effort before I got consensus that such a contribution would be 
welcome in principle, so I have yet to look at the API surfaces in general.

Lastly, I'm at a conference this week, so please accept my apologies if I 
am not responsive until next week.

 - Jason

On Thursday, February 16, 2017 at 3:17:42 PM UTC-8, [email protected] 
wrote:
>
> Hi Jason,
>
> Thanks for sending this proposal, I took a look and had some comments:
>
> 1) As per the proposal, I think that the recommended approach is indeed go 
> down the path of a java library wrapper. JNI wrapping is another 
> possibility, but we should first investigate whether the Java wrapper is 
> not possible before considering it.
> 2) IMO the jruby implementation can live in a separate repository and I 
> would recommend this to start with. In the future, when the wrapped 
> implementation is stable and passes interop tests we can consider merging 
> it into the main repo. I would recommend starting this work outside of the 
> main gRPC repo. The grpc-ecosystem org <http://github.com/grpc-ecosystem> 
> is a good candidate. This will allow for maximum flexibility.
> 3) It would be useful to define some shallow tests that cover the API. 
> Such tests should be added to the existing Ruby implementation first, and 
> can then be used to ensure that the jruby implementation conforms. See 
> https://github.com/grpc/grpc/blob/master/src/python/grpcio_tests/tests/unit/_api_test.py
>  
> for a starting point. I must reiterate that this type of testing is very 
> shallow and is not sufficient in itself, but a good addition to the interop 
> tests.
> 4) For any gRPC implementation, we need to ensure that the interop tests 
> pass. See https://github.com/grpc/grpc-java/tree/master/interop-testing, 
> https://github.com/grpc/grpc/tree/master/src/ruby/pb/test and 
> https://github.com/grpc/grpc/blob/master/doc/interop-test-descriptions.md. 
> The jruby implementation needs to pass these in order to be considered 
> conforming. We recommend re-using the ruby interop test suite with the 
> jruby wrapper to the extent possible, independent of the java suite (of 
> course, these can be leveraged as needed). The nice thing is that if the 
> APIs fully conform, the ruby interop tests should just pass.
> 5) Given the difference between Java and C stacks, there may be API 
> incompatibilities that might not be surmountable, especially in situations 
> where the C core implementation specifics is directly exposed upward. Such 
> situations need to be quantified and further down the line, we can decide 
> how to tackle them and weigh the cost of doing just that vs a full JNI 
> implementation.
>
> Some minor comments on the proposal doc itself: 
>
> 1) Please add details on implementation ownership, so that it can be 
> discussed. Its unclear if this is something that the proposer is signing up 
> to implement, is that the plan?
>
> 2) Could you please update the proposal to cover the implementation 
> approach in more detail based on some of these points. Ie: Location of the 
> prototype repo, some discussion on the API surfaces that need to be 
> conformed to and potential areas you have identified as problematic. This 
> can go hand in hand with a prototype of the java wrapped approach so that 
> we can tackle issues as they happen.
>
>
> Thanks!
> Kailash
>
>
>
>
>
>
> On Thursday, February 16, 2017 at 10:36:25 AM UTC-8, Mike Moore wrote:
>>
>> On Thu, Feb 16, 2017 at 9:08 AM, Jason Lunn <[email protected]> wrote:
>>
>>> It has probably been a decade since I had a reason to try to use JNI. 
>>> Does anyone have any experience building Gems using that approach? Would it 
>>> be the case that there would be one JAR per supported platform (
>>> x64-mingw32, x86_64-linux, universal-darwin, x86-mingw32), or would it 
>>> be a fat jar that include multiple JNI wrappers?
>>>
>>> I am not clear on what capabilities of the existing gem are needed to 
>>> satisfy the dependency of the Google Cloud gem (which is my personal 
>>> motivation to see this move forward). Maybe Mike Moore knows better if the 
>>> client side would be enough?
>>>
>>
>> The Google Cloud gems are clients only, and so supporting the client-half 
>> of GRPC would be sufficient for our use of GRPC.
>>  
>>
>>> I think I prefer to see the JRuby support sit alongside the existing 
>>> Ruby code in the same repository, and tested in a consistent way. This will 
>>> help avoid breaking behavior from slipping into the C implementation that 
>>> isn't observed until someone thinks to update the other repo. It should 
>>> also make it less likely that a gem update is released for the C derived 
>>> platforms without a corresponding update for the JRuby variant.
>>>
>>> On Wednesday, February 15, 2017 at 1:30:59 PM UTC-5, [email protected] 
>>> wrote:
>>>>
>>>> As has been noted before, I think this would be possible by doing a 
>>>> pure implementation, wrapping the java client, or with a JNI wrapper over 
>>>> the C-core (these implementations probably end up relatively different 
>>>> from 
>>>> C-wrapping grpc-ruby).
>>>>
>>>> Actually one thing hasn't been noted yet: If only a client-side library 
>>>> is needed, then perhaps the jruby-platform target could only need to 
>>>> emulate the client-half of the c-wrapping grpc-ruby (would probably 
>>>> simplify things).
>>>>
>>>> As for where the project could live though, one option is it could go 
>>>> into the same grpc repo as current grpc-ruby. Another is it could be done 
>>>> entirely as a third party project in its own repo, for which we could give 
>>>> some guidance.
>>>>
>>>> On Wednesday, February 15, 2017 at 8:59:31 AM UTC-8, Jason Lunn wrote:
>>>>>
>>>>> There is a GRFC <https://github.com/grpc/proposal/pull/13> to add 
>>>>> support for the JRuby <http://jruby.org/> interpreter, which runs on 
>>>>> the JVM.
>>>>>
>>>>> Please provide all feedback on this thread.
>>>>>
>>>>
>>

-- 
You received this message because you are subscribed to the Google Groups 
"grpc.io" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To post to this group, send email to [email protected].
Visit this group at https://groups.google.com/group/grpc-io.
To view this discussion on the web visit 
https://groups.google.com/d/msgid/grpc-io/dbae8b24-dcc6-453f-a6c5-92219429b1c0%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Reply via email to