I'm working with an application which was originally designed to make heavy use of static-variables and functions to impose singleton-style access to objects. I've been utilizing Parsley to break apart some of the coupling in this project, and I'd like to start chiseling away at the use of static functions.
I will explain the basic architecture first, and then I'll ask my questions.
Within this application exists a static utility which sends/receives HTTP requests. When a component wishes to request data via HTTP, it invokes a static function in this class:
Utility.fetchUrl(url, parameters, successHandler, errorHandler);
This function calls another static function in a tracking component which monitors how long requests take, how many are sent, etc. The invocation looks very similar in the utility class:
public static function fetchUrl( ... ):void {
Tracker.startTracking(url, new Date());
...
Tracker.stopTracking(url, new Date());
}
Any components in this application wishing to dispatch an HTTP request must do so via the web utility class. This creates quite a bit of coupling between this class and other components, and is just one example of several where such reliance on static functions exists. This causes problems when we're extending and refactoring the system: I would like to decouple this using events.
Ultimately, I'd like each component to instantiate a custom event which is dispatched to a high-level framework. From there, the framework itself would relay the event to the correct location. In other words, those components which need to perform an HTTP request would dispatch an event like this:
dispatchEvent(new WebRequestEvent(url, parameters, successHandler, errorHandler));
From there, Parsley (or another framework) would make sure the event is sent to the correct location which could handle the functionality and perform whatever is necessary. This would also allow me a stepping-stone to introducing a more compartmentalized MVC architecture, where web request results are handled by models, injected by the framework into their own respective views.
I'd like to do the same with the tracking functionality.
Are there drawbacks from using an event-based mechanism, coupled with a framework like Parsley? Is it better to stick with static functions/variables and use the singleton-style access? If so, why? Will this end up causing more trouble in the future than it's worth?
So, short answer on Events drawbacks:
Slightly more weight on the system to use the events. Listeners, bubbling, capture, etc.. all have to be managed by the system. Much less of an issue when you're outside the display hierarchy, but still more expensive than straight functions. (then again, avoid pre-optimization, so get it right, then get it fast).
"Soft" circular dependencies can occur in complicated asynchronous systems. This means you end up with a case where A triggers an event. B notices A has changed, so updates C. C triggers an event. D notices C has changed and updates A. Doesn't usually max the CPU, but is still a non-terminating loop.
Sometimes you need to have forced buffering / locking of functions. Say component A and B both trigger the same event. You might not want C to be triggered twice (e.g., fetching new data from server) so you have to make sure C is marked as "busy" or something.
From personal experience, I haven't seen any other issues with event systems. I'm using the PureMVC framework in a relatively complicated system without issue.
Good luck.
Related
I'm about to start a new project and I'd like to use Single Page Application technique. Since I'll be using ASP.NET I think the easiest way will be using Angular, which I'm new with.
Anyway, what scares me the most about Angular (or any other JS/TS tech), is since I don't have much time, I can't afford to rewrite all the models/entities to another language. The code and maintenance cost of this is too high for me.
tl;dr
So my question is, is there a way to have Angular use the original model/entity names so I can use them in the page without the need to rewrite any unnecessary code?
Will the .NET attributes take action?
I guess your concern is, that your business object world (entity model) needs to be reflected in your client/angular app as models (javascript objects) ?? The need for them comes also from typing errors you get in angular 2.
Creating and maintaining a transparent model world spanning server and client part is way too much effort for real world applications, although it would be nice.
I decided to use and receive the model as a result from a remote call via AJAX/WebAPI and work with these "models" in my client applications. The result then reflects your business model (entities) you have probably defined already.
this.dataService.getRecords('MT_MyEntity')
.subscribe((data: any[]) => {
var response: any = data; // Do this to avoid typing errors
var resprecords: any = response.items;
// Here you get entities;
// Deal with your business objects fetched from remote system; Use it to show in forms, ....
},
error => {
// your error handling
});
In your application you can use the entity and attributes names you have defined in your server side model (take care of upper/lower case modifications)
For me this is a pragmatic way to deal with that and it works very well.
For any decent size application the benefits of creating client side model far outweigh the effort required to create and maintain them.
This effect is more pronounce with TypeScript as it allows compile time checking of the contracts. As we move more and more code to client side and use frameworks like Angular, having a clearly defined model helps us understand what is happening. We derive the same benefits that we derive when type checking is available on server.
Having a separate client side model also allows us to adapt such model to client side UI needs (albeit sometime we create viewmodel to satisfy such requirements)
The approach of generating these client side contracts, as highlighted by #Ivan can help us to reduce the overall effort.
I am currently working on several flex projects, that have gone in a relative short amount of time from prototype to almost quite large applications.
Time has come for some refactoring to be done, so obviously the mvc principle came into mind.
For reasons out my control a framework(i.e. robotlegs etc.) is not an option.
Here comes the question...what general guidelines should I take into consideration when designing the architecture?
Also...say for example that I have the following: View, Ctrl, Model.
From View:
var ctrlInstance:Ctrl= new Ctrl();
ctrl.performControllerMethod();
In Controller
public function performControllerMethod():void{
//dome some sort of processing and store the result in the model.
Model.instance.result = method_scope_result;
}
and based on the stored values update the view.
As far as storing values in the model, that will be later used dynamically in the application, via time filtering or other operation, everything is clear, but in cases where data just needs to go in(say a tree that gets populated once at loading time), is this really necessary to use the view->controller->model->view update scheme, or can I just make the controller implement IEventDispatcher and dispatch some custom events, that hold necessary data, after the controller operation finished.
Ex:
View:
var ctrlInstance:Ctrl= new Ctrl();
ctrl.addEventListener(CustomEv.HAPPY_END, onHappyEnd);
ctrl.addEventListener(CustomEv.SAD_END, onSadEnd);
ctrl.performControllerMethod();
Controller
public function performControllerMethod():void{
(processOk) ? dispatchEvent(new CustomEv(CustomEv.HAPPY_END, theData)) : dispatchEvent(new CustomEv(CustomEv.SAD_END));
}
When one of the event handlers kicks into action do a cleanup of the event listeners(via event.currentTarget).
As I realize that this might not be a question, but rather a discussion, I would love to get your opinions.
Thanks.
IMO, this whole question is framed in a way that misses the point of MVC, which is to avoid coupling between the model, view, and controller tiers. The View should know nothing of any other tier, because as soon as it starts having references to other parts of the architecture, you can't reuse it.
Static variables that are not constant are basically just asking for trouble http://misko.hevery.com/2009/07/31/how-to-think-about-oo/. Some people believe you can mitigate this by making sure that you only access these globals by using a Controller, but as soon as all Views have a Controller, you have a situation where the static Model can be changed literally from anywhere.
If you want to use the principles of a framework without using a particular framework, check out http://www.developria.com/2010/04/combining-the-timeline-with-oo.html and http://www.developria.com/2010/05/pass-the-eventdispatcher-pleas.html . But keep in mind that established frameworks have solved most of the issues you will encounter, you're probably better off just using one.
I am looking into utilizing Google Maps API to do some geocoding. I want to implement client side geocoding, to remove the possibility of request limitation.
I need to do some fairly complex logic on the result set, and I would prefer to do that in C# as it is a ASP.NET MVC application. However part of that logic is possibly makeing subsequent follow up requests and that again would require JavaScript.
So, my first thought is to make a service in my application to pass JSON results to and certain return types to trigger the subsequent request. That seems a little convoluted and want to know from the community if this seems like the best approach and if there are any libraries/third party tools that can help handle this situation.
I've an app that does something similar, with the complexity somewhat decoupled by using standardized events (within this app, not a W3 standard or anything)
Client uses native geolocation, SimpleGeo and Google Loader to guess where the user is and AJAX's that to the server.
Server uses client data, MaxMind, and user preferences to decide where to treat the user as being.
Server response is generic event data (as JSON response) that is converted by a generic AJAX response handler into one or more events triggered against the body element.
Depending on the page, various listeners are bound to the events and or namespaces (see jQuery namespaced events) and they handle the updated location events, e.g., getting different weather data, changing local search results
Some of those listeners in turn trigger other AJAX requests, the responses to those may also carry generic events to triggered...
This way there's no sequential code I have to write, i.e., I can add or remove behaviors (simple or complex) without changing anything else. jQuery Events are all I use, really nothing much to it after you decide how you'll pattern things.
Let me know if that's interesting to you and you want me to expand or clarify a part of it.
You may want to try this API:
http://code.google.com/apis/maps/documentation/geocoding/
It's far more REST like - no Javascript required. May work better with C#
In the end I found the best solution was to do as I stated in my question. Pass the JSON object to controller, do work, then return. Worked pretty well.
This is quite a lengthy post, so bear with me. I'm not sure whether it is primarily about ASP.NET Session State behaviour, NInject, application design, or refactoring. Read on and then you can decide... :-)
Background
First, a bit of background. We are working on trying to refactor a large webshop into a more maintainable , structured design. The webshop is currently running on .NET 3.5, but the design is more of a hangover from the classic ASP days. Obviously we cannot tackle everything in one go, so many of the features / technologies / approaches have to be taken as a given. With that in mind...
The app maintains everything to do with the current session (user profile, cart, session choices, etc.) in a context object which is simply a large XML document that gets serialized to and deserialized from the Session as a string. The XML format is also important because the rendering is done via XSLT.
This has led to a number of problems :
It's a kind of God object with far
too many concerns.
It's loosely typed and relies too much on XML manipulation / XPath.
There is no standard way / pattern for retrieving the session xml document or for writing it back. We have a horrible mixture of methods that take the document in as a parameter, modify it and return it, methods that retrieve it themselves, modify it and save it back to session, etc, etc. This has lead to a lot of hard to trace bugs, over-use of serializing /deserializing from the Session, etc.
Our Solution
What we have done is try to introduce a strongly -typed wrapper around the xml document, which breaks it up into different concerns and to manage the lifecycle transparently to the rest of the app.
What we are aiming for is the following workflow:
Beginning of the request, we populate
the session document from the xml
string stored in the session.
The rest of the app interacts with it
only through the strongly typed
wrapper. The whole app uses the same
instance and does not have to worry
about when to retrieve or save the
state back to session.
At the end of the request, the underlying xml document is serialized back to the Session.
Since we are using NInject(v1) as the IOC of choice, we decided to use this to manage the lifecycle of our context object. The context object was wrapped with the OnePerRequest attribute and the dispose method was hooked up to a method that would save the xml document back to Session as a string.
It doesn't work...
We soon encountered a problem that the NInject OnePerRequest module didn't appear to have access to SessionState. The first thing we tried was a hack that we would keep the Session object in a variable to make sure we could still write to it. This appeared to work on a development machine but it became obvious it didn't when moving to out of process state.
It still doesn't work...
We tried inheriting from the OnePerRequest behaviour / module, and adding the IRequiresSessionState marker interface (OnePerRequestRequiresSessionState). However, this was not enough as the method which NInject uses to release references and clean up gets hooked up to the EndRequest method. Session is available in EndRequest but it has already been serialized to the out of process state server so changing something now is not reflected when the session string is retrieved at the beginning of the next request.
We then decided to change the even t to hook up to. We ditched EndRequest and hooked up our OnePerRequestRequiresSessionState "release all" method to the PostRequestHandlerExecute event, which is BEFORE the session data gets serialized out of process.
It works... then it doesn't...
This seemed to work. On a single server and on a web farm. Then we noticed weird behaviour. There seemed to be two different versions of the context and you would randomly switch between them. Add something to the cart, it's not there. Go to browse to another product and the previous product would show up in the cart.
After some tracing, we discovered the culprit: Response.Redirect. Sprinkled throughout the site in literally hundreds of places is Response.Redirect(url);. With this version of the redirect, the execution of the page is stopped immediately. This means that PostRequestHandlerExecute is not fired and the current version of the Context object is not thrown away by NInject... and everything falls apart. New versions are not created properly, etc. EndRequest is fired which is why the normal NInject OnePerRequest module works fine with it, just not our bastardized version that tries to use session state.
Of course, there is an override to Response.Redirect where you can pass a boolean value in to tell it whether to terminate the existing page or continue to execute - Response.Redirect(url,false). Continuing obviously fires our event and everything works but... it continues to execute the rest of the page! This means executing everything that comes after the call to Redirect and we have absolutely no idea what that means (since the existing site expects it to stop).
What next?
So, any suggestions on what to do? So far we've discussed :
Abstracting our redirect behaviour
and going through a central method
that controls the redirect (perhaps
hacking out a way to call the
PostRequestHandlerExecute even t or
maybe a custom Redirect event that
our NInject module can also
subscribe to and clean up).
Seeing if there is a way we can
force the Session object to save in
EndRequest if it hasn't been saved
previously in
PostRequestHandlerExecute, and do
the ninject clean up in EndRequest
Remove our dependency on Session
completely and use another storage
mechanism: DB, document DB,
distributed HashTable, etc. Any
advice? Suggestions we haven't
thought of? Things you've tried
that have / haven't worked?
I think you're on the right track. Here's some thoughts I had:
in addition to the strongly typed wrapper you have, I'd suggest a facade for accessing the context object that returns your wrapper, something like an IContextProvider. that way you can introduce it piece-meal, and then when it's fully integrated, you can refactor the provider without breaking the things that use it. I can't tell, but you might have already done this. it'll also be easier to change your persistence mechanism if you choose to. if you can do this, I would suggest once you get all the dependencies isolated from the context object, change it to not persist as XML. the SessionState will store a binary object much faster, and you can always serialize to XML if you need to do transforms.
I don't think that Ninject is the correct mechanism for what you're trying to do. it's difficult to signal end of the request in Ninject, since garbage collection can't be depended on. have you considered using an IHttpModule instead? you can use the AcquireRequestState and ReleaseRequestState or EndRequest to handle getting/setting the context in Session. only allow the app to get to the context object through the facade.
if you're on a webfarm, you're probably using a database for your Session storage anyway, so putting your context into a DB won't be much different.
Firstly, while it's good to demonstrate you've put in the work, (and I and others may not have replied if it wasn't clear how much you're interested in a resolution)... that's a massive wall of text! Here's a +1 on your way to investing in a bonus for a complete response that talks about the Ninject ASP.NET extensions and how they apply to each individual element of your issue. Having said that, hopefully someone will come along with a real resolution for you.
Even though it's [very] 2.0 specific, Nate's Cache and Collect Post is required reading. While it seems you're pretty au fait with the tradeoffs involved and have debugged deep in, the article is well worth a few reads.
I'd also consider moving to V2 of Ninject - a lot of this stuff has been revised significantly. It's not magically going to work, but represents a mature rewrite based on a lot of learning from V1. Have you read the (V1 or) V2 unit tests for Ninject? They'll show you the low level tools at your disposal in order to realise your goals.
Bottom line for me is that you need to work out a strategy for your state management independent of DI, and then by all means use the container/DI system as a part of the implementation.
I am currently trying to use the "Invocation tags" of Mate to call my web services and delegate the WS-responses to my fault/result handlers.
I want to use the generated proxies, provided by the Flex Builder, and not the plain <WebService> or <WebServiceInvoker> tags.
I actually failed using several techniques:
<WebServiceInvoker> does not work with the generated proxies.
<AsyncMethodInvoker> needs some complicated successType that I could not get to work with the WS-calls. And defining the events seems redundant to me. I want it simple and easy to read, the code will be touched by other people than me!
<MethodInvoker> can't use instances, and it also can't handle the proxies' AsyncToken
<DelegateInvoker> Looked fine at first. It calls the service but doesn't fire valid result events (infinite busy cursor). Even though i can successfully bind to the XYZ_lastResult of the WS-proxies, and a WS-call results in getting valid data from the WS-backend, the <faultHandlers> and <resulthandlers> are not executed. There is some solution for the DelegateInvoker that changes code in the generated proxies, which i definately do not want to do!
So here is my question: Is there a simple(!) way of using default Flexbuilder generated proxies with the Mate Invocation tags?
It appears that your request is not that uncommon to Mate. Check out this couple of threads in their forum:
http://mate.asfusion.com/forums/topic.php?id=424
http://mate.asfusion.com/forums/topic.php?id=421
The solution is to modify some bits of the auto-generated code... which in a way ruins the whole point of using code generation.