Burn down list of issues raised by module owners

[mhdawson]

Community reported issues:

• Microtask queue interaction with N-API (Problems with N-API that prevent us from using N-API exclusively, instead of V8 and Node internal APIs #279)
• Callback to js from native threads (Problems with N-API that prevent us from using N-API exclusively, instead of V8 and Node internal APIs #279 (comment))
• 3x perf overhead in napi version of native-hdr-histogram (3x overhead on Node 6 node-addon-api#102) (@digitalinfinity)
• Convert.js not updating dependencies in binding.gyp (convert.js does not add node-addon-api to the dependencies in binding.gyp node-addon-api#98) (@sampsongao)
• Make Buffer extend Uint8Array (Make Buffer extend Uint8Array? node-addon-api#109) (@sampsongao)
• Enforce HandleScope semantics more strictly (N-API: Enforce handle scope semantics more strictly node#16175)
• Resolve promises asynchronously without napi_create_async_work (N-API feature request: expose API to resolve promises asynchronously without napi_create_async_work node#15604)

[digitalinfinity]

I'll edit the issue to include the links to the issues that are being tracked by this meta-issue

[mhdawson]

#279

[digitalinfinity]

@mhdawson I've added a few issues here- feel free to update the issue to add any I might have missed (and I'll add more to this list as they come in)

[mhdawson]

@digitalinfinity thanks !

[YafimK]

• create custom env - when creating an async worker, you may want to manipulate env to be with another isolate. A bit closer to Node ENV.

[digitalinfinity]

Thanks @YafimK - can you create a separate issue in nodejs/node with more details on what change you'd like to see, perhaps with the use case- I can then update the above list and link to that issue.

[digitalinfinity]

For node-native-histogram, I've determined that the benchmark is not doing a significant amount of work, so its effectively just measuring the difference between napi_call_function and nan's function call API. Given that, I don't think we have to necessarily address this immediately since in most benchmarks where the benchmark does non-trivial work, the overhead difference would be less significant and so I'll check this off our list. However, I'll continue to look at whether there are more clever things we can do to reduce the absolute difference in overheads but napi_call_function will always have greater overhead since it's a more general API than nan.

[gabrielschulhof]

https://github.com/MayhemYDG/iltorb/tree/napi documents a slight memory leak wrt. V8 bindings.

[gabrielschulhof]

nstepien/iltorb#70 (comment)

[NickNaso]

Hi everyone I want report this issue kelektiv/node.bcrypt.js#709 developers are starting use ative addons in worker_threads so they need to refactor to implement a context aware addon.

[gabrielschulhof]

We discussed a possible solution to the accumulation of native finalizers during a tight loop that creates wrapped objects here:

nodejs/node-addon-api#1140