See also: IRC log
paul: we talked with TAG
... on Issue 72 refactoring
... one thing came up with the GENIVI call
... architectural pattern we agreeable
... logical architecture for that
... plugin-based and service-based
... LBS uses the same pattern
... any thoughts?
<ted> Kevin and I are the only ones on so far it seems
<ted> dial in worked
dave: in favor of service-based approach
paul: AdamC?
hashimoto: no specific opinion
kaz: don't have strong preference
myself
... just wanted to mention the WoT group's work
... they're working on both the socket-based interface (WoT API)
... and the device-internal script interface (Script API)
peter: think service-based approach is promising
powell: websocketty approach
mentioned in the issue 72
... service-based approach is the right way to go
urata: basically with the
service-based approach
... but concerned with the possible schedule delay
... could someone clarify the good/bad points?
paul: would leave it the group
(lost audio...)
kevin: @@@
dave: makes sense
... the partial answer to Urata-san is that the current
approach lacks flexibility
... we need to move to high level API
ted: in favor of service-based
approach
... to separate implementations from plugins
... for testing, etc.
urata: I understand that choosing
the websocket approach doesn't mean we don't need JavaScript
APIs
... JavaScript APIs are useful for developers
... because easy to use
paul: layered approach would be
good
... service/websocket approach exposes the functionality using
services
... how JS approach could work with it?
... the current API definition's expectation is working within
the runtime
dave: as a third party developer,
would prefer simplicity
... easy enhancement
peter: agree
... if you want to go more code completion function
... it's quite easy
powell: @@@
peter: also service-based
approach makes the testing easy
... quite a few advantages
urata: understand the good point
that it would be easy
... not opposing the websocket approach
... just would like to clarify the situation
... we're using promise style APIs
... if we use the websocket approach, we can't use the promise
style
dave: can create promise style API by wrapping the low-level API
kevin: can have high-level
library
... high level representation
dave: kind of like Node.js
... still have underlying HTTP server
kevin: make sense but would loose
advantage of high-level APIs
... OEMs would benefit with high-level spec
dave: it's spec vs library
paul: you want to have standardized vendor neutral APIs
kevin: yes
paul: the question is that in
this specific group
... would it make sense to define the spec?
... focus on the value
kevin: probably the low-level
idea has attraction
... but as an implementer, it's complicated
... several signals responding to the real world
... signals treated separately
... we can correct the data together
... avility there
dave: regarding the low-level
approach
... depending on how low will you go
... APIs will be very simple if it's low-level enough
... the data spec itself could be tweaked to meet with the
websocket approach
paul: Kevin, make sense to you?
kevin: would see pros/cons of the websocket approach
dave: have created gist
... short brainstorm
... can open a new issue
powell: happy to help out
... examples for better understanding
paul: one thing useful is what
discussed during the GENIVI call
... clarifying architecture pattern
... what goes to where
... as far as the timeline scope, this might impact the
Charter
ted: hopefully could fresh out
quickly
... need to revise the Charter
... sounds like lean on the approach
... high-level API and low-level API
... competition anyhow
... developers may contribute on GitHub
paul: to me, would like to do is
getting a new issue
... what we need to do is getting consensus
... target architecture, strawman on what to decide
... anyone to volunteer?
dave: can help the strawman
peter: also would look into
... see the concrete work
ted: would contribute
kaz: can also help
... also
think we might want to work for both the high-level API and the
low-level API
... and see if the high-level API is implementable using the low-level API
... would be useful to define standard high-level interface (=standard implementation of the high-level API using the low-level API) for developers
dave: conversation on pros/cons
of the websocket approach
... strawman discussion
... high-level interface
paul: architecture and high-level interface discussion
dave: also data definition from WebIDL to JSON
kevin: not objecting the vote on
high-level vs low-level
... realistic implementers' security concern should be
considered
... don't mind having both the high-level API and the low-level
API
dave: first action is strawman
discussion on how the websocket approach works
... and the second architecture picture
... third is WebIDL to JSON
... fourth is existing high-level API can be implemented using
the newly proposed low-level API
paul: Urata-san, if you have specific opinions, please speak out
urata: not sure how much time for
this topic
... so need to talk within the company
... but would like to participate
paul: good with this
urata: about the service-based
approach, the Generic Sensor API itself is not this type
... anybody aware of any spec which is really web-socketty?
kaz: WoT IG is working on WoT API
using socket level RESTful connection
... but they're still an IG, and would transition to a WG
shortly
... Web&TV IG guys are also working to extend the web
platform testing mechanism
ted: TV Control API CG is working on API for TV tuners
kaz: the v1 spec draft for TV
Control API is device internal plugin style
... maybe they might want to think about websocket approach for
the v2 version
[ adjourned ]