Thanks for the comments. I now get more sense of it, and have a few "thoughts" inline.

Kind regards,

---
Yavuz Selim Yilmaz
SUNY at Buffalo
Computer Science and Engineering
PhD Candidate

On Jul 26, 2013, at 10:42 AM, Summers Pittman <supittma@redhat.com> wrote:

Be warned I'm only on my first cup of coffee :)

On 07/25/2013 12:20 PM, Yavuz Selim YILMAZ wrote:
Hi all,

When I was using AeroGear iOS, I needed to read from pipe and execute the success and failure blocks in background (for background data syncronization, a basic one which provides eventual data consistency). What I did first was this:

dispatch a thread
    read from pipe
        --- READ ---
        success
            --- SUCCESS BLOCK ---
        failure
            --- FAILURE BLOCK ---
            
However, when I checked, I realized that the success and failure blocks were being executed on the main thread, not on the thread that I initiated the read. Then I needed to modify the code and do like this (I know all "--- READ ---"s are already in background):

read from pipe
    --- READ ---
    success
        dispatch a thread
            --- SUCCESS BLOCK ---
    failure
        dispatch a thread
            --- FAILURE BLOCK ---
            
At that time, I was going to ask the reason behind dispatching back to the main queue for success and failure blocks, but then I thought the idea is coming from Android's AsyncTask. So my wrong assumption made sense as I thought having similar way of handling the end-developer (hope you get what I mean by end-developer) operations (in this case read from pipe) would help to use same/similar software architectures for different platforms (iOS and Android in this case).

Then I started to taste AeroGear Android from the last week. Keeping my hard times with multithreading on iOS in mind, I was expecting to see the read from pipe code something like this:

create AsyncTask
    doInBackground
        --- READ ---
    onPostExecute
        success
            --- SUCCESS BLOCK ---
        failure
            --- FAILURE BLOCK ---
            
However, what I've seen on RestAdapter was quite different:

create new thread
    --- READ ---
    success
        --- SUCCESS BLOCK ---
    failure
        --- FAILURE BLOCK ---

And here, success and failure blocks are being executed on the background thread instead of on the UI thread. I checked why not using AsyncTask for this operation (as I was using it for my network operations if I need to touch the UI thread at onProgress and at onPostExecute), and I realized that there is a reason behind using ThreadPoolExecutor (no deep research though, I only know it is recommended considering the operations might take longer).
The reason we use the thread pool executor is because the onPostExcute method runs on the UI thread.  If your Pipe is running form a service that means that things happening on onSuccess would block the UI thread.  This wouldn't be obvious to the user.


Makes perfect sense. :)


So, in Android case, making sure the availability of the UI thread when executing success and failure blocks is left to the end-developer as far as I could see (please correct me if I am wrong),
So the threading bits aren't explicitly defined in the Pipes 101 doc.  What the docs say to do will get you around this issue though.  If you use the Pipeline get methods which take in some form of context (Activity, Fragment, fragment activity, etc) then you will have onSuccess run on the UI thread.  Otherwise you want.

The way pipeline handles the lifecycle of the request and threads makes perfect sense. My concern now is, as an outside developer, I start with getting started guides, and going through step by step, pipe (RestAdapter - "first example") provides me a way to read from endpoint, but it doesn't fit to the "most common" scenario, which I think is (as Christios mentioned) "networking on the background, update UI on the success block" (though I can still update the UI, the activity might or might not be available at that moment). So I feel like, even if I don't know what pipeline is (at that moment), the "first example" should fit to the basic scenario so it should use pipeline (in current aerogear-android implementation). But mine is just an idea. :)


Sounds like a trip to the JIRA mobile for better docs.

In general you want to use these methods because it tells lets AG do some smart things behind the scenes with respect to managing the network, data caching, etc.

With pipeline, yes, aerogear does all the magic. :)


while in iOS case the execution of the success or failure blocks indicates that the main thread is up and running at that time. The difference is maybe because of the way iOS and Android handles the execution of the background threads. But, when first tasting the libs for both platforms, I did not feel comfortable with (it was not so obvious) on which thread the success and failure blocks are executed.

So, my first question is, can somebody point me where these threading designs were discussed?
Probably not.  The best thread I can think of would be back from February but it really doesn't define anything.
I am curious about the idea behind these designs (platform pushes that way? easy to implement? common way of doing network operations? or any other specific reason?), and hoping to learn new things from there (actually I am sure I will).
The goals behind the design were to leverage as much of than Android API to handle the activity lifecycle as possible.  It did mean that there are somethings which just have to be "known" and the docs should highlight them.  IE if you are calling things from an Activity or Fragment then you should pass the context to the pipeline.get method.  If the docs aren't clear (or if we need to explain the full implications) then that is a JIRA. 

This I think goes to my "the 'first example' should fit to the most common scenario" statement, if the first example does so, things are getting clear while going further. But I won't expect everybody to go over "all" the guides and docs before using AeroGear (e.g. maybe my bad, but I'm using many libs which I didn't read "all" of their docs and guides yet), and even if the developers don't know how AeroGear does the things, they should be able to use its magic from the start. :) (just an opinion though)

For iOS, I checked why AFNetworking defaults to the main thread instead of whichever thread initiates the call, and I've seen an odd explanation. Here is the brief scenario:

The first idea comes from this guy and he provides (to me the solution which makes perfect sense) this PR:
https://github.com/AFNetworking/AFNetworking/pull/107

This simply works as follows: Whichever thread initiates the request is default execution thread for success/failure blocks, and developer has option to override (to some other thread or to main thread). The AFNetworking maintainers find this solution confusing. After 5 months, they provide this PR:
https://github.com/AFNetworking/AFNetworking/pull/202

This uses the main thread as default, and developer has option to override it (I think AeroGear developer doesn't even have such an option for now). Here is the explanation why not the first but the second solution is merged (from one of the second option supporting developer): "This is discussable, but probably lead to may less bugs due to people not understanding what they are doing :)" However, this solution and explanation doesn't make sense to me at all. I think if I'm initiating the request from the main thread, I might or might not know what I am doing, but if I am dispatching another thread and requesting from there, I at least know what multithreading is. :) And I did expect my success/failure to run on the initiator thread when I was trying it out (maybe it's only me, but I felt it's the natural way - and AeroGear Android pipeline works this way as far as I could see).

At the least, as Christos said, nothing prevents me to work on the threading, and to deal with life cycles on my own, but I'm just expecting all the magic (at least for the common scenario) is done by AeroGear (it's in the end much more powerful than what I can come up with in a limited time :)). And to me, what AFNetworking provides does not seem to be a full magic. :) Maybe it's  only me who thinks the common scenario is different.


If people have better ideas then that is a ML thread and then maybe a JIRA :)

Second and the last thing, I think somewhere in the guides and docs, the threading (best practices when using AeroGear) should be emphasized/mentioned/explained. Because, although things are smooth and easy when the operations take little amount of time, things are becoming more complicated when the operations take longer as the device users' app usage pattern is unknown. (e.g. for Android, I needed to check thread name/id to actually realize the success/failure blocks were being executed on the background thread (e.g. Pipes 101 mentions to be aware of Android Activity Lifecycle, but not so clear what happens during the execution - is it going to stop? is it going to execute on UI thread or background? will I need to check if UI thread is there or not? etc.).

And for iOS, I needed to have a really blocking success/failure code to realize it was running on the main thread, since then I was looking for some other parts to see why scrolling was not smooth).

Kind regards,

---
Yavuz Selim Yilmaz
SUNY at Buffalo
Computer Science and Engineering
PhD Candidate



_______________________________________________
aerogear-dev mailing list
aerogear-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/aerogear-dev

_______________________________________________
aerogear-dev mailing list
aerogear-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/aerogear-dev