Return-Path: MIME-Version: 1.0 In-Reply-To: References: <1299058614-8904-1-git-send-email-lkslawek@gmail.com> Date: Thu, 3 Mar 2011 16:54:25 -0300 Message-ID: Subject: Re: [PATCH 0/4] Message Access Profile plugin From: Luiz Augusto von Dentz To: Slawomir Bochenski Cc: linux-bluetooth@vger.kernel.org Content-Type: text/plain; charset=ISO-8859-1 Sender: linux-bluetooth-owner@vger.kernel.org List-ID: Hi, On Thu, Mar 3, 2011 at 12:20 PM, Slawomir Bochenski wrote: > On Thu, Mar 3, 2011 at 4:08 PM, Luiz Augusto von Dentz > wrote: >> Hi, >> >> On Thu, Mar 3, 2011 at 11:09 AM, Slawomir Bochenski wrote: >>> On 3/3/11, Luiz Augusto von Dentz wrote: >>>> Hi, >>>> >>>> On Thu, Mar 3, 2011 at 6:58 AM, Slawomir Bochenski >>>> wrote: >>>>> Hi, >>>>> >>>>> On Wed, Mar 2, 2011 at 10:12 PM, Luiz Augusto von Dentz >>>>> wrote: >>>>>> I guess naming the file and plugin 'map' would make more sense here, >>>>>> we don't want to confuse people what this file is about. >>>>> >>>>> There are two things which makes MAP - MAS and MNS. It is not like any >>>>> other plugin we have here. The communication goes both ways, i.e. in >>>>> full implementation there are OBEX server and OBEX client present on >>>>> _both_ sides. As generally architecture of obexd and obex-client >>>>> currently allows OBEX servers to be implemented only in obexd and OBEX >>>>> clients in obex-client, it is easy to imagine that in order to add >>>>> full MAP client role, there would be need for another plugin in obexd >>>>> for MNS. >>>> >>>> But that doesn't solve any problem since the profile as a whole need >>>> both client and server side it make no sense to have one without the >>>> other, we can still have different drivers for each sides but the >>>> plugin should be the same, it is a single profile not two. >>>> >>> >>> Let me rephrase this: >>> +-------+-------------+---------------+ >>> |MAP ? ?| obex-client | client/mns.c ?| >>> |Server | ? ? ? ? ? ? | ? ? / \ ? ? ? | >>> |role ? | ? ? ? ? ? ? | ? ? ?| ? ? ? ?| >>> | ? ? ? +-------------| ? ?D-Bus ? ? ?| >>> | ? ? ? | obexd ? ? ? | ? ? ?| ? ? ? ?| >>> | ? ? ? | ? ? ? ? ? ? | ? ? \ / ? ? ? | >>> | ? ? ? | ? ? ? ? ? ? | plugins/mas.c | >>> +-------+-------------+---------------+ >>> >>> +-------+-------------+---------------+ >>> |MAP ? ?| obex-client | client/mas.c ?| >>> |Client | ? ? ? ? ? ? | ? ? / \ ? ? ? | >>> |role ? | ? ? ? ? ? ? | ? ? ?| ? ? ? ?| >>> | ? ? ? +-------------| ? ?D-Bus ? ? ?| >>> | ? ? ? | obexd ? ? ? | ? ? ?| ? ? ? ?| >>> | ? ? ? | ? ? ? ? ? ? | ? ? \ / ? ? ? | >>> | ? ? ? | ? ? ? ? ? ? | plugins/mns.c | >>> +-------+-------------+---------------+ >>> >>>>> So there can be mns.c and mas.c, both in obexd, both completely >>>>> independent - each one of them having nothing in common, i.e. user can >>>>> run mns when he wants to be MAP client, mas when he wants to be MAP >>>>> server or mas and mns when he likes to be both. >>>>> >>>>> Therefore naming it "map" would confuse more those who know what is >>>>> MAP and what they really want. >>>> >>>> Again if they cannot be qualified separately then it make no sense to >>>> separate them in two plugin, the logical separation can happen on >>>> driver level. >>> >>> So yes, they most definitely can be qualified separately. >> >> Sorry but I doubt you can, MAS and MSN are not profiles they just >> represent services, besides it is mandatory to support both MAS and >> MNS in MSE see Table 4-1:MAP features, so at least for obexd it >> doesn't make much sense to have them separated in two plugins, they >> can be separated in different files which are used by the same plugin, >> this make it easier to enable/disable MAP as o whole. > Please read me again. Especially take a longer look at the graphics > with roles presented. The plugins/mns.c presented here _does not_ have > _anything_ to do with plugins/mas.c - they are used in completely > different roles. Once again: they are NOT part of the same thing. See > chapter 2.1 in MAP specification. What will be qualified _together_ is > plugin/mas.c and client/mns.c. Yep, another design decision you took without consulting us, I believe MSE should be completely implemented on obexd and MCE on obex-client, why do you thing involving two processes here would make sense? There is nothing prevent us to implement client code on obexd nor server code on obex-client, this was a design choice you took which doesn't mean it can't work using MSE/MCE separation. >>>>>> Also Ive been thinking on removing this internal APIs for backend, >>>>>> each would implemented as plugin/mimetype driver directly and we can >>>>>> create basic drivers for pbap and map and export its callbacks on >>>>>> pbap.h and map.h respectively as we do for pcsuite which uses ftp >>>>>> driver callbacks. >>>>> >>>>> I really hope that you will keep this in "thinking stage" for now. I >>>>> see problems coming. I'd rather prefer to use what we have now and >>>>> what works. There might be some parts that won't fit well in this new >>>>> philosophy. It would be better to postpone considering such changes in >>>>> MAP to the point when it will be in repository in a more complete >>>>> form. >>>> >>>> It should not cause any problem, because core only knows about the >>>> mimetype drivers, the backend interface is a plugin specific API. In >>>> theory one could re implement pbap plugin which would have another >>>> backend interface. If we have the backend implementing the mimetype >>>> driver the only thing that changes is that no API is needed between >>>> e.g. pbap plugin and the backend. >>>> >>>> Note that one of the worst problems with current pbap implementation >>>> is the backend API, because it has to handle things like asynchronous >>>> requests and cancel requests which comes from mimetype driver it had >>>> to change several times during the development, I don't want this to >>>> happen with map. >>> >>> This already happened with MAP and it works. It is done differently than >>> in PBAP, for example handling of application parameters and cases when >>> body is sent and when it is not is done more cleanly. And I've already >>> seen the problems closer to the end of the MAP implementation road. What >>> you're proposing will make this road more bumpy. >> >> Sorry, but until this reach upstream you cannot assumed it has >> happened. That why we suggest sending us patches earlier so we can >> review and discuss architect and design aspects not only code, now >> that we had more experience with the backend interface we could >> actually experiment the direct approach without creating more APIs, >> usually it is easier to add API but hard to remove them once a lot of >> code depend on them. What you may suggest is to integrate MAP >> implementation as it is and latter change it, but please communicate >> before about design decisions, there is no need to develop this >> completely before sending upstream. > PBAP is much simpler than MAP. Thus it may be tempting to do > assumptions about MAP using previous experiences. This may not lead to > good conclusions. If mimetype driver API was not able to implement the drivers for MAP you would have to change it, I don't see any modification to mimetype.h, so it looks exactly like pbap, mimetype driver is used to extract the request information to pass to backend via some API. Please if you have a use case for MAP that with mimetype driver it cannot be implement please provide it, otherwise I would consider that you meant the backend is much more complex, but the requests are quite similar. -- Luiz Augusto von Dentz Computer Engineer