Develop Reference

Clean Architecture: How to reflect the data layer's changes in the UI

I'm trying to make a design based on the Uncle Bob's Clean Architecture in Android.
The problem:
I'd like to solve is how to make the changes generated in one repository to be reflected in other parts of the app, like other repositories or Views.
The example
I've designed a VERY simplified example for this example. Please notice that boundary interfaces has been removed to keep the diagrams small.
Imagine an app that shows a list of videos (with title, thumnail and like count), clicking a video you can see the detail (there you can like/dislike the video).
Additionally the app has an statistics system that counts the number of videos the user liked or disliked.
The main classes for this app could be:
For the Videos part/module:
For the Stats part/module:
The target
Now imagine you check your stats, then navigate the list of videos, open the detail of one, and click the like button.
After the like is sent to the server, there are several elements of the apps that should be aware of the change:
Of course the detail view, should be updated with the changes (this can be made through callbacks so no problem)
The list of videos should update the "likes" count for the given video
The StatsRepository may want to update/invalidate the caches after voting a new video
If the list of stats is visible (imagine a split screen) it should also show the updated stats (or at least receive the event for re-query the data)
The Question
What are the common patterns to solve this kind of communication?
Please make your answer as complete as you can, specifying where the events are generated, how they get propagated though the app, etc.
Note: Bounties will be given to complete answers

Publish / Subscribe
Typically, for n:m communication (n senders may send a message to m receivers, while all senders and receivers do not know each other) you'll use a publish/subscribe pattern.
There are lots of libraries implementing such a communication style, for Java there is for example an EventBus implementation in the Guava library.
For in-app communication these libraries are typically called EventBus or EventManager and send/receive events.
Domain Events
Suppose you now created an event VideoRatedEvent, which signals that a user has either liked or disliked a video.
These type of events are referred to as Domain Events. The event class is a simple POJO and might look like this:
class VideoRatedEvent {
/** The video that was rated */
public Video video;
/** The user that triggered this event */
public User user;
/** True if the user liked the video, false if the user disliked the video */
public boolean liked;
}
Dispatch events
Now each time your users like or dislike a video, you'll need to dispatch a VideoRatedEvent.
With Guava, you'll simply pass an instantiated event object to object to EventBus.post(myVideoRatedEvent).
Ideally the events are generated in your domain objects and are dispatched within the persisting transaction (see this blog post for details).
That means that as your domain model state is persisted, the events are dispatched.
Event Listeners
In your application, all components affected by an event can now listen to the domain events.
In your particular example, the VideoDetailView or StatsRepository might be event listeners for the VideoRatedEvent.
Of course, you will need to register those to the Guava EventBus with EventBus.register(Object).

This is my personal 5cents and maybe not closely enough related to your example of "The Clean Architecure".
I usually try to force a kind of MVC upon androids activities and fragments and use publish/subscribe for communication. As components I have model classes that handle business logic and the data state. They data changing methods are only to be called by the controller classes which usually is the activity class and also handles session state. I use fragments to manage different view parts of the application and views under those fragments (obviously). All fragments subscribe to one or more topics. I use my own simple DataDistributionService which handles different topics, takes messages from registered publishers and relays them to all subscribers. (partly influenced by the OMGs DDS but MUCH MUCH more primitive) A simple application would only have a single topic e.g. "Main".
Every part of view interaction (touches etc) is handled by its fragment first. The fragment can potentially change a few things without sending notifications. E.g. switching the subrange of rendered data elements if the rest of the app does not need to know/react. Otherwise the fragment publishes a ViewRequest(...) containing the necessary parameters to the DDS.
The DDS broadcasts that message and at some point reaches a controller. This can simply be the main activity or a specific controller instance. There should be only ONE controller so that the request is only handled once. The controller basically has a long list of request handling code. When a request arrives the controller calls to the business logic in the model. The controller also handles other view related things like arranging the view (tabs) or starting dialogs for user input (overwrite file?) and other things that the model is not supposed to know about but influences (Throw new NoOverWritePermissionException())
Once the model changes are done the controller decides if an update notification has to be send. (usually it does). That way the model classes do not need to listen or send messages and only take care of busines logic and consistent state. The update notification ist broadcasted and received by the fragments which then run "updateFromModel()".
Effects:
Commands are global. Any ViewRequest or other kind of request can be send from anywhere the DDS can be accessed. Fragments do not have to provide a listener class and no higher instance has to implement listeners for their instanced fragments. If a new fragment does not require new Requests it can be added without any change to controller classes.
Model classes do not need to know about the communication at all. It can be hard enough to keep consistent state and handle all the data management. No message handling or session state handling is necessary. However the model might not be proteced against malicous calls from the view. But that is a general problem and cannot really be prevented if the model has to give out references at some point. If your app is fine with a model that only passes copies/flat data its possible. But at some point the ArrayAdapter simply needs access to the bitmaps he is supposed to draw in the gridview. If you cannot afford copies, you always have the risk of "view makes a changing call to the model". Different battlefield...
Update calls might be too simple. If the update of a fragment is expensive (OpenGL fragment reloading textures...) you want to have more detailed update information. The controler COULD send a more detailed notification however it actually should not have to/be able to know what parts of the model exactly changed. Sending update notes from the model is ugly. Not only would the model have to implement messaging but it also gets very chaotic with mixed notifications. The controler can divide update notifications and others a bit by using topics. E.g. a specific topic for changes to your video resources. That way fragments can decide which topics they subscribe to. Other than that you want to have a model that can be queried for changed values. Timestamp etc. I have an app where the user draws shapes on canvas. They get rendered to bitmaps and are used as textures in an OpenGL view. I certainly don't want to reload textures everytime "updateFromModel()" is called in the GLViewFragment.
Dependency Rule:
Probably not respected all the time. If the controller handles a tab switch it can simply call "seletTab()" on a TabHost and therefore have a dependency to outer circles. You can turn it into a message but then it is still a logical dependency. If the controller part has to organize some elements of the view (show the image-editor-fragment-tab automatically after loading an image via the image-gallery-fragmen-tab) you cannot avoid dependencies completely. Maybe you can get it done by modelling viewstate and have your view parts organize themselves from viewstate.currentUseCase or smth like that. But if you need global control over the view of your app you will get problems with this dependency rule I'd say. What if you try to save some data and your model asks for overwrite permission? You need to create some kind of UI for that. Dependency again. You can send a message to the view and hope that a DialogFragment picks it up. If it exists in the extremely modular world described at your link.
Entities:
are the model classes in my approach. That is pretty close to the link you provided.
Use Cases:
I do not have those explicitly modelled for now. Atm I am working on editors for videogame assets. Drawing shapes in one fragment, applying shading values in another fragment, saving/loading in a galleryfragment, exporting to a texture atlas in another one ... stuff like that. I would add Use Cases as some kind of Request subset. Basically a Use Case as a set of rules which request in which order are allowed/required/expected/forbidden etc. I would build them like transactions so that a Use Case can keep progressing, can be finished, can be cancelled and maybe even rolled back. E.g. a Use Case would define the order of saving a fresh drawn image. Including posting a Dialog to ask for overwrite permission and roll back if permission is not give or time out is reached. But Use Cases are defined in many different ways. Some apps have a single Use Case for an hour of active user interaction, some apps have 50 Use Cases just to get money from an atm. ;)
Interface Adapters:
Here it gets a bit complicated. To me this seems to be extremely high level for android apps. It states "The Ring of Interface Adapters contains the whole MVC architecture of a GUI". I cannot really wrap my head around that. Maybe you are building far more complicated apps than I do.
Frameworks and Drivers:
Not sure what to think of this one. "The web is a detail, the database is a detail..." and the graphic contains "UI" in this Ring as well. Too much for my little head
Lets check the other "asserts"
Independent of Frameworks. The architecture does not depend on the existence of some library of feature laden software. This allows you to use such frameworks as tools, rather than having to cram your system into their limited constraints.
Hm yeah well, if you run your own architecture that is what you get.
Testable. The business rules can be tested without the UI, Database, Web Server, or any other external element.
As in my approach model classes neither know about controllers or views nor about the message passing. One can test state consistency with just those classes alone.
Independent of UI. The UI can change easily, without changing the rest of the system. A Web UI could be replaced with a console UI, for example, without changing the business rules.
Again a bit overkill for android is it not? Independence yes. In my approach you can add or remove fragments as long as they do not require explicit handling somewhere higher up. But replacing a Web UI with a console UI and have the system run like before is a wet dream of architecture freaks. Some UI elements are integral part of the provided service. Of course i can easily swap the canvas drawing fragment for a console drawing fragment, or the classic photo fragment for a 'take picture with console' fragment but that does not mean the application still works. Technically its fine in my approach. If you implement an ascii console video player you can render the videos there and no other part of the app will necessarily care. However it COULD be that the set of requests that the controller supports does not align well with the new console UI or that a Use Case is not designed for the order in which a video needs to be accessed via a console interface. The view is not always the unimportant presenting slave that many architecture gurus like to see it as.
Independent of Database. You can swap out Oracle or SQL Server, for Mongo, BigTable, CouchDB, or something else. Your business rules are not bound to the database.
Yeah, so? How is that directly related to your architecture? Use the right adapters and abstraction and you can have that in a hello world app.
Independent of any external agency. In fact your business rules simply don’t know anything at all about the outside world.
Same here. If you want modularized independent code then write it. Hard to say anything specific about that.

Android Entry Points

Does anybody know the process of entry point generation for program with multiple entry points, specifically Android APK. I already tried to look at Flowdroid and Scandroid, but I don't understand the terms "Source" and "Sinks" and their relation to entry point generation.
If anybody could point me to learning resource for newbies, maybe some good books, I would really appreciate it. (I already wasted a few days on this)
Thank you.
Edit, after further reading on FlowDroid:
Sources and Sinks have no relation to entry point generation but they are used for data flow analysis. That's why when running FlowDroid with --nocallbacks option, it will not generate dummyMainClass.class. Amirite? Or am I still wrong?
Source:
Furthermore, FlowDroid needs a complete modeling of Android’s lifecycles and callbacks. Because sources and sinks for Android are provided by SuSi, we only have to look for entry points. Along with necessary meta information they are extracted from Android’s manifest file, dex files and layout xml files. The latter allow us to consider user interaction callbacks defined in XML (for example button clicks) and discover additional sources in terms of password fields. Because user interaction cannot be predicted statically, FlowDroid generates a special main method which considers all possible combinations to make sure no taint is lost.

--nocallbacks
Disables the emulation of Android callbacks (button clicks, GPS location changes, etc.; this option reduces the runtime, but may miss some leaks.
What are callbacks?
From FlowDroid paper
Callbacks The Android operating system allows applications to
register callbacks for various types of information, e.g., location
updates or UI interactions. FLOWDROID models these callbacks in
its dummy main method, for instance to recognize cases where an
application stores the location data that the framework passes to the
callback as a parameter, and later sends this data to the Internet when
the activity is stopped. The order in which callbacks are invoked
cannot generally be predicted, which is why FLOWDROID assumes
that all callbacks can be invoked in any possible order. However,
callbacks can only happen while the parent component (e.g. activity)
is running. For precision, FLOWDROID thus associates components
(activities, services, etc.) with the callbacks they register. An activity
may, for instance, register callbacks that get invoked when a button
is pressed. The respective callback handler would then have to be
analyzed between the onResume() and onPause() events of this
activity only.
There are two different ways to register callback handlers on the
Android platform. Firstly, callbacks can be defined declaratively in
the XML files of an activity. Alternatively, they can also be registered
imperatively using well-known calls to specific system methods.
FLOWDROID supports both ways. Additionally, for malware there
is the risk that an attacker registers undocumented callbacks by
overwriting methods of the Android infrastructure, some of which
could even be called by native code. FLOWDROID recognizes such
overwritten methods, handling them similar to normal callback
handlers such as button clicks.
For finding callbacks registered in the application code, FLOWDROID
first computes one call graph per component, starting at the
lifecycle methods (onCreate(), onStop(), etc.) implemented in the
respective component class. This call graph is then used to scan for
calls to Android system methods that use one of the well-known
callback interfaces as a formal parameter type. Afterwards, the call
graph is incrementally extended to include these newly discovered
callbacks, and the scan is run again since callback handlers are free
to register new callbacks on their own, potentially requiring FLOWDROID
to re-extend the call graph and re-analyze until a fixed point
it reached. While this method is more expensive than just scanning
for classes implementing the callback interfaces, it delivers a more precise mapping between components and callbacks. This does not
only reduce false positives, but we also found it to considerably decrease
the runtime of the following taint analysis. Once the dummy
main method has been constructed, FLOWDROID computes a final
call graph using this method as the app’s entry point.
For callbacks defined in the layout XML files, the respective
XML file is mapped to one or more application components using
the respective layout controls. A button-click handler, for instance,
is only valid for the activity that hosts the respective button. FLOWDROID
analyzes each activity to see which identifiers from the XML
file it registers. This information is then used to create the mapping
I'm sorry, we have the same problem: there isn't a good documentation about Soot/FlowDroid. But I often find the solution to my problems by searching the mailing list.

GWT: Places, History and MVP by Example

Say I want to create a GWT app that consists of two simple "screens":
A Main Menu that users see when the app first starts up (they go to specific URL) - forget login screens, etc., for this simple example; and
An Edit Preferences "page"/screen/view that takes the place of the Main Menu when the user clicks a button on the Main Menu; this Edit Preferences screen also has a button that takes the user back to the Main Menu; thus
The user can simply go back and forth between the Main Menu and Edit Preference "pages" (this is GWT so its actually a single-page app) all day long
So the pseudo-code for this might look like:
public class MainMenuPresenter implements Presenter {
// mainMenuScreen may be a Composite that contains a Button for
// switching/navigating to the EditPreferencesView
private MainMenuView mainMenuScreen;
}
public class EditPreferencesPresenter implements Presenter {
// editPrefsScreen may be a Composite that contains a Button for
// switching/navigating back to the MainMenuView
private EditPreferencesView editPrefsScreen;
}
Questions:
How does a Place and its associated Activities fit into this paradigm? If it's a choice between MVP and Places/Activities, then how would this example look using the Activities/Places paradigm?
If MVP and Activities/Places can play nicely together (and are not mutually exclusive), then how could we implement GWT's History API here so that we can set MainMenuView as one "place" in history, EditPreferencesView as a second "place", and then allow the user to hit the Back/Forward browser buttons and keep toggling between them?
If MVP and Activities/Places are mutually exclusive, and I have to choose between the two, is there anything different about the code in Question #2 above that needs to change?
I'm also interested in how a Place ties into an ActivityManager, and how/who fires/handles PlaceChangeEvents, but I'll save that for later; I need to understand these concepts before I can move on to those. Thanks in advance!

I'll try to outline an example setup that could work in your situation. There are lots of variations, but I'd like to concentrate on a simple one (and encourage you to think up where this can be adjusted).
1. Define URL tokens
Choose some URL tokens like "#mainmenu" and "#editprefs" which will be appended to the host page's URL. They will be used for the browser history, so the back and forward buttons work etc.
The URL handling will be done automatically for you by DefaultHistorian. The PlaceChangeEvent will be fired by PlaceController.
2. Map tokens to Place objects
A Place object is simply the object oriented abstraction of a token - this is useful, because more advanced tokens can also take parameters which need to be parsed. You'll need a way to map the tokens to Place objects. This is the responsibility of PlaceHistoryMapper.
In my example, we'd simply implement PlaceHistoryMapper manually to map "#mainmenu" to a MainMenuPlace and "#editprefs" to an EditPreferencesPlace.
[Alternatively, it's also possible to use the #WithTokenizers annotation and implement an (empty) PlaceTokenizer for every type of place. You could then use the #Prefix annotation to specify "mainmenu" and "editprefs" as the tokens.]
3. Map the Places to Activities
A Place object itself doesn't do anything - as explained above, it's basically just an abstract token. The actual code will be run in an Activity. So you'll have to map Places to Activities. This is the responsibility of ActivityMapper.
In my example you'd implement it to map MainMenuPlace to a MainMenuActivity and EditPreferencePlace to an EditPreferenceActivity.
4. Activities and Presenters
For simplicity, in my example the Activities would also implement the Presenter. So MainMenuActivity would implement MainMenuPresenter. This is not necessary at all, but maybe a nice starting point. And this is where Places+Activities can connect with MVP. The two concepts don't require each other, but they work nicely together:
Activities+Places is basically about the connection between the history token and an Activity.
MVP is basically about the connection between a Presenter and a View.
If you let an Activity (or one of its delegates) implement a Presenter, you have connected both.
5. Quick overview
"#mainMenu"
---(PlaceHistoryMapper)---> MainMenuPlace
---(ActivityMapper)---> MainMenuActivity implements MainMenuPresenter

Android: Using observer pattern with Views (possibly MVC, MVVM related)

I've spent some hours reading various questions and answers regarding implementing the various MVC-type patterns in Android. I've seen a few code examples posted in various blogs. I would, however, still appreciate some ideas and opinions on what I am trying to achieve. Specifically, I'd like to work out the best code mechanism to inform a single View, or a group of Views, that a particular item of data has been changed.
My application is quite simply one which obtains measurement data from a hardware device via Bluetooth, and displays and logs that data. At present, I have a Service which takes care of Bluetooth communications and background logging. I have a 'global' data store class that is an extension of Application.
As measurement data is polled from the external device, the measurement data (which is in reality about thirty bytes of data) is updated in the data store object (which, in MVC terms, I'm guessing is the 'model').
At any time, only a small subset of that data will be displayed by UI Views. Typically, a given View will only be interested in representing one particular byte of measurement data. As the user moves to different Activity classes, other Views will be displayed which would display a different subset of that data.
So to get to the point, I'm trying to choose the best way to cause invalidate() to be invoked on interested Views when a given data item is changed.
The options seem to be:
Make use of the existing Observer class, and related classes.
Kind of 'roll my own' observer pattern, by creating my own register() and unregister() functions in the data model. Observer Views would be held in an ArrayList (or perhaps a more complex arrangement of one observer List per data item). I'd loop through this ArrayList each time data are updated and call invalidate() (or postInvalidate() of course, depending on my threading arrangement).
Are there any reasons why I should use one of the above over the other? And is there any other 'observer' mechanism I should consider?

Many views in Android are backed by some subclass of BaseAdapter which has a method notifyDataSetChanged() which instructs the view to refresh itself. If you are using a view (such as ListView or GridView or any descendent of AdapterView) then it is backed by a BaseAdapter and you can simply update that Adapter and the view will refresh itself.
I guess this means, I vote that you use the built-in observer pattern. If you are using a custom view then obviously this won't work and you would have to use a custom method of refreshing anyway.

Another Option would be to use the Android Intent framework. When new data is received in the service set the data to the universal model and broadcast an intent that the data has been updated using the Context.broadcastIntent(Intent) method. Any view that is interested in that data would register and unregister receivers using the Context.RegisterReceiver(Receiver) and Context.unregisterReceiver(Receiver) methods. From there the view would retrieve the data from the universal model and update the view accordingly.
I think this might be what the observer pattern in Android.Lifecycle.Observer package is doing behind the scenes.

Why people use message/event buses in their code? [closed]

Closed. This question is opinion-based. It is not currently accepting answers.
Closed 4 years ago.
Locked. This question and its answers are locked because the question is off-topic but has historical significance. It is not currently accepting new answers or interactions.
I think that you have heard of message/event buses, it's the single place when all events in the system flow. Similar architectures are found in computer's motherboards and LAN networks. It's a good approach for motherboards and networks as it reduces the number of wires, but is it good for software development? We don't have such restrictions as electronics does.
The simplest implementation of message bus/event bus can be like:
class EventBus {
void addListener(EventBusListener l}{...}
void fireEvent(Event e) {...}
}
Posting events is done with bus.fireEvent(event), receiving messages is enabled by bus.addListener(listener). Such architectures are sometimes used for software development, for example MVP4G implements similar message bus for GWT.
Active projects:
Google Guava EventBus
MBassador by Benjamin Diedrichsen
Mycila PubSub by Mathieu Carbou
mvp4g Event Bus
Simple Java Event Bus
Dormant/Dead projects:
Sun/Oracle JavaBeans InfoBus
https://eventbus.dev.java.net/ [Broken link]
It's just the popular Observer (Listener) pattern made 'globally' - each object in the system can listen to each message, and I think it's bad, it breaks the Encapsulation principle (each object knows about everything) and Single Responsibility principle (eg when some object needs to a new type of message, event bus often needs to be changed for example to add a new Listener class or a new method in the Listener class).
For these reasons I think, that for most software, Observer pattern is better than event bus. What do you think about event bus, does it make any good sense for typical applications?
EDIT: I'm not talking about 'big' enterprise solutions like ESB - they can be useful (what's more ESB offers much, much more than just an event bus). I'm asking about usefulness of using message bus in 'regular' Java code for object-to-object connection - some people do it, check the links above. Event bus is probably best solution for telephone-to-telephone communication or computer-to-computer communication because each telefone (or computer) in a network can typically talk to each other, and bus reduces the number of wires. But objects rarely talk to each other - how many collaborators one object can have - 3, 5?

I am considering using a In memory Event Bus for my regular java code and my rationale is as follows
Each object in the system can listen to each message, and I think it's
bad, it breaks the Encapsulation principle (each object knows about
everything)
I am not sure if this is really true, I class needs to register with the event bus to start with, similar to observer pattern, Once a class has registered with the Event Bus, only the methods which have the appropriate signature and annotation are notified.
and Single Responsibility principle (eg when some object needs to a
new type of message, event bus often needs to be changed for example
to add a new Listener class or a new method in the Listener class).
I totally disagree with
event bus often needs to be changed
The event bus is never changed
I agree with
add a new Listener class or a new method in the Listener class
How does this break SRP ?, I can have a BookEventListener which subscribes to all events pertaining to my Book Entity, and yes I can add methods to this class but still this class is cohesive ...
Why I plan to use it ? It helps me model the "when" of my domain ....
Usually we hear some thing like send a mail "when" book is purchased
we go write down
book.purchase();
sendEmail()
Then we are told add a audit log when a book is purchased , we go to the above snippet
book.purchase();
sendEmail();
**auditBook();**
Right there OCP violated
I Prefer
book.purchase();
EventBus.raiseEvent(bookPurchasedEvent);
Then keep adding handlers as needed Open for Extension Closed for Modification
Thanks

Some people like it because it is the embodiment of the Facade pattern or Mediator pattern. It centralizes cross-cutting activities like logging, alerting, monitoring, security, etc.
Some people don't like it because it is often a Singleton point of failure. Everyone has to know about it.

I use it heavily in JavaScript. There can be so many various widgets that all need to do some sort of action whenever something else happens -- there is no real hierarchy of ownership of objects. Instead of passing references of every object to every object, or just making every object global, when something significant happens inside a particular widget, I can just publish "/thisWidget/somethingHappened" -- instead of filling that widget with all kinds of code specific to the API of other widgets. The I have a single class that contains all the "wiring", or "plubming" as they like to call it in the Java Spring framework. This class contains references to all of my widgets, and has all of the code for what happens after each various event fires.
It is centralized, easy to access and maintain, and if one thing changes or I want a new process to occur on a specific event, I don't have to search through every single class/object/widget to try to find out where something is being handled. I can just go to my "operator" class -- the one that handles all the "wiring" when a particular event happens, and see every repercussion of that event. In this system, every individual widget is completely API agnostic of the other widgets. It simply publishes what has happened to it or what it is doing.

I'm having trouble understanding what you're really asking in your question. You give an example of a simple event bus which is actually just Observable with a different name, then you say;
For these reasons I think, that for
most software, Observer pattern is
better than event bus. What do you
think about event bus, does it make
any good sense for typical
applications?
..but given your example, they are the same. This makes me wonder if you have ever used something like a Enterprise Service Bus. At a base level an ESB logically does the same thing as the observer pattern, but commercial products add much, much more. Its like an event bus on steroids. They are complicated software products and offer;
Message pickup
Generate events by listening to various endpoints. The endpoint can be a listener (such as a HTTP server), a messaging system (such as JMS), a database or pretty much anything else you want.
Message routing
Take your event and send it to one/many endpoint. Routing can be pretty smart, the bus might route the message depending on the message type, the message contents or any other criteria. Routing can be intelligent and dynamic.
Message Transformation
Transforms your message into another format, this can be as simnple as from XML to JSON or from a row on a database table to a HTTP request. Transformation can occur within the data itself, for example swapping date formats.
Data Enrichment
Adds or modifies data in your message by calling services along the way. For example if a message has a postcode in it the bus might use a postcode lookup service to add in address data.
..and lots, lots more. When you start looking into the details you can really start to see why people use these things.

Because it can be an important step in the way to decouple the application modules to a service based architecture.
So in your case if you have not the intention to decouple the modules of your application into isolated services then the native implementation of the Observer pattern will make it a simpler solution.
But If you want to build lets say a micro-services architecture the event-bus will allow to get the benefits of this architecture style so you could for instance update and deploy just some part of your application without affect others, because they are just connected through the event-bus.
So the main question here is the desired level of application components decoupling.
Some references about it:
http://microservices.io/patterns/data/event-driven-architecture.html
http://tech.grammarly.com/blog/posts/Decoupling-A-Monolithic-Server-Application.html

A good analogy is that of a telephone exchange, where every handset can dial to every other handset. A compromised handset can tune into other conversations. Program control flows like wires(cyclomatic complexity anyone!) This is similar to the requirement of having a connection/physical medium between two end points. This is So for N handsets instead of having NC2 (Combinatorial logic) flows for every new handset we tend to get N flows.
A reduction in complexity implies easy to understand code. Lets start with the prominent points you have highlighted: 1. Global knowledge 2. Intrusive modifications.
Global Knowledge: Consider message event to be an envelop. From event handler/sender perspective there is no data being exposed, it is seeing an envelop (unless an derived class tries to do some inspection using 'instanceof' checks). In a good OOP design, this would never occur.
Intrusive modifications: Instead of having a event specific listener approach, one can use a global event handling approach. As such we have a global event type (on which data is piggy backed and down-casted). This is much like the PropertyBeanSupport model in Java. With a single event type we are required to have a single sender and listener types. This implies you need not modify the bus/listeners every time you see something new. The ugly down-casting can be soothened using Adapter pattern (Please don't start that another level of redirection quote!). Programmers can write assembly in any language. So need for commonsense and smartness can not be substituted. All I intend to state is it can be a effective tool.
The actual event receivers can use the listeners (composition/proxy) easily. In such a Java code base, listeners would look like stand alone inner class declarations (with unused warning being flagged in certain IDEs). This is akeen to two players on the beach playing a ball game, the players don't react until they see the ball.
'#duffymo' points out another interesting aspect: 'Single point of failure'. This would/can in theory effect any object residing in memory(RAM) and not specific to MessageHandlers.

As a practical example, our app sync's with a web service every x number of minutes, and if any new data is received, we need to update the GUI. Now, because the SyncAdapter runs on a background thread, you can't simply point to a textview and modify its properties, you have to bubble up an event. And the only way to make sure you catch that event is if you have a shared (static,singleton) object passing that event around for subscribers to handle.