Are there any patterns for component versioning and backwards-compatibility using Windsor? - inversion-of-control

I have to support a new input file format in a system which uses Windsor. I also need to support the old version of the input file during a transition phase.
This will probably be repeated in future, and we'll again need to support the new and the next most recent format.
The import processing is handled by a component, and the new version has had significant improvements in the code which makes it lots more efficient compared to the old version. So what I'd like to do is to have the new component and the old component in the system, and dynamically use the new or the old component based upon the file metadata.
Is there a pattern for this type of scenario anyone can suggest?

The fact that you're using Windsor is pretty much irrelevant here. Always strive to find a container-independent solution. Here's one:
interface IImportProcessor {
bool CanHandleVersion(int version);
Stream Import(Stream input);
}
class ImportProcessorVersion1 : IImportProcessor {
public bool CanHandleVersion(int version) {
return version == 1;
}
public Stream Import(Stream input) {
// do stuff
return input;
}
}
class ImportProcessorVersion2 : IImportProcessor {
public bool CanHandleVersion(int version) {
return version == 2;
}
public Stream Import(Stream input) {
// do stuff
return input;
}
}
class MainImportProcessor: IImportProcessor {
private readonly IImportProcessor[] versionSpecificProcessors;
public MainImportProcessor(IImportProcessor[] versionSpecificProcessors) {
this.versionSpecificProcessors = versionSpecificProcessors;
}
public bool CanHandleVersion(int version) {
return versionSpecificProcessors.Any(p => p.CanHandleVersion(version));
}
private int FetchVersion(Stream input) {
// do stuff
return 1;
}
public Stream Import(Stream input) {
int version = FetchVersion(input);
var processor = versionSpecificProcessors.FirstOrDefault(p => p.CanHandleVersion(version));
if (processor == null)
throw new Exception("Unsupported version " + version);
return processor.Import(input);
}
}
Your app would take a dependency on IImportProcessor. The container is wired so that the default implementation of this interface is MainImportProcessor. The container is also wired so that MainImportProcessor gets all other implementations of IImportProcessor.
This way you can add implementations of IImportProcessor and each will be selected when appropriate.
It might be easier to wire things up if MainImportProcessor implements an interface different from IImportProcessor.
Another possibility could be implementing a chain of responsibility.

Related

How do I update a gtk listbox from an async method?

So when writing UI in GTK it's generally preferrable to handle reading of files, etc. in an Async Method. things such as listboxes, are generally bound to a ListModel, the items in the ListBox updated in accordance with the items_changed signal.
So if I have some class, that implements ListModel, and has an add function, and some FileReader that holds a reference to said ListModel, and call add from an async function, how do i make that in essence triggering the items_changed and having GTK update accordingly?
I've tried list.items_changed.connect(message("Items changed!")); but it never triggers.
I saw this: How can one update GTK+ UI in Vala from a long operation without blocking the UI
but in this example, it's just the button label that is changed, no signal is actually triggered.
EDIT: (Codesample added at the request of #Michael Gratton
//Disclaimer: everything here is still very much a work in progress, and will, as soon as I'm confident that what I have is not total crap, be released under some GPL or other open license.
//Note: for the sake of readability, I adopted the C# naming convention for interfaces, namely, putting a capital 'I' in front of them, a decision i do not feel quite as confident in as I did earlier.
//Note: the calls to message(..) was put in here to help debugging
public class AsyncFileContext : Object{
private int64 offset;
private bool start_read;
private bool read_to_end;
private Factories.IVCardFactory factory;
private File file;
private FileMonitor monitor;
private Gee.Set<IVCard> vcard_buffer;
private IObservableSet<IVCard> _vCards;
public IObservableSet<IVCard> vCards {
owned get{
return this._vCards;
}
}
construct{
//We want to start fileops at the beginning of the file
this.offset = (int64)0;
this.start_read = true;
this.read_to_end = false;
this.vcard_buffer = new Gee.HashSet<IVCard>();
this.factory = new Factories.GenericVCardFactory();
}
public void add_vcard(IVCard card){
//TODO: implement
}
public AsyncFileContext(IObservableSet<IVCard> vcards, string path){
this._vCards = vcards;
this._vCards = IObservableSet.wrap_set<IVCard>(new Gee.HashSet<IVCard>());
this.file = File.new_for_path(path);
this.monitor = file.monitor_file(FileMonitorFlags.NONE, null);
message("1");
//TODO: add connect
this.monitor.changed.connect((file, otherfile, event) => {
if(event != FileMonitorEvent.DELETED){
bool changes_done = event == FileMonitorEvent.CHANGES_DONE_HINT;
Idle.add(() => {
read_file_async.begin(changes_done);
return false;
});
}
});
message("2");
//We don't know that changes are done yet
//TODO: Consider carefully how you want this to work when it is NOT called from an event
Idle.add(() => {
read_file_async.begin(false);
return false;
});
}
//Changes done should only be true if the FileMonitorEvent that triggers the call was CHANGES_DONE_HINT
private async void read_file_async(bool changes_done) throws IOError{
if(!this.start_read){
return;
}
this.start_read = false;
var dis = new DataInputStream(yield file.read_async());
message("3");
//If we've been reading this file, and there's then a change, we assume we need to continue where we let off
//TODO: assert that the offset isn't at the very end of the file, if so reset to 0 so we can reread the file
if(offset > 0){
dis.seek(offset, SeekType.SET);
}
string line;
int vcards_added = 0;
while((line = yield dis.read_line_async()) != null){
message("position: %s".printf(dis.tell().to_string()));
this.offset = dis.tell();
message("4");
message(line);
//if the line is empty, we want to jump to next line, and ignore the input here entirely
if(line.chomp().chug() == ""){
continue;
}
this.factory.add_line(line);
if(factory.vcard_ready){
message("creating...");
this.vcard_buffer.add(factory.create());
vcards_added++;
//If we've read-in and created an entire vcard, it's time to yield
message("Yielding...");
Idle.add(() => {
_vCards.add_all(vcard_buffer);
vcard_buffer.remove_all(_vCards);
return false;
});
Idle.add(read_file_async.callback);
yield;
message("Resuming");
}
}
//IF we expect there will be no more writing, or if we expect that we read ALL the vcards, and did not add any, it's time to go back and read through the whole thing again.
if(changes_done){ //|| vcards_added == 0){
this.offset = 0;
}
this.start_read = true;
}
}
//The main idea in this class is to just bind the IObservableCollection's item_added, item_removed and cleared signals to the items_changed of the ListModel. IObservableCollection is a class I have implemented that merely wraps Gee.Collection, it is unittested, and works as intended
public class VCardListModel : ListModel, Object{
private Gee.List<IVCard> vcard_list;
private IObservableCollection<IVCard> vcard_collection;
public VCardListModel(IObservableCollection<IVCard> vcard_collection){
this.vcard_collection = vcard_collection;
this.vcard_list = new Gee.ArrayList<IVCard>.wrap(vcard_collection.to_array());
this.vcard_collection.item_added.connect((vcard) => {
vcard_list.add(vcard);
int pos = vcard_list.index_of(vcard);
items_changed(pos, 0, 1);
});
this.vcard_collection.item_removed.connect((vcard) => {
int pos = vcard_list.index_of(vcard);
vcard_list.remove(vcard);
items_changed(pos, 1, 0);
});
this.vcard_collection.cleared.connect(() => {
items_changed(0, vcard_list.size, 0);
vcard_list.clear();
});
}
public Object? get_item(uint position){
if((vcard_list.size - 1) < position){
return null;
}
return this.vcard_list.get((int)position);
}
public Type get_item_type(){
return Type.from_name("VikingvCardIVCard");
}
public uint get_n_items(){
return (uint)this.vcard_list.size;
}
public Object? get_object(uint position){
return this.get_item((int)position);
}
}
//The IObservableCollection parsed to this classes constructor, is the one from the AsyncFileContext
public class ContactList : Gtk.ListBox{
private ListModel list_model;
public ContactList(IObservableCollection<IVCard> ivcards){
this.list_model = new VCardListModel(ivcards);
bind_model(this.list_model, create_row_func);
list_model.items_changed.connect(() => {
message("Items Changed!");
base.show_all();
});
}
private Gtk.Widget create_row_func(Object item){
return new ContactRow((IVCard)item);
}
}
Heres the way i 'solved' it.
I'm not particularly proud of this solution, but there are a couple of awful things about the Gtk ListBox, one of them being (and this might really be more of a ListModel issue) if the ListBox is bound to a ListModel, the ListBox will NOT be sortable by using the sort method, and to me at least, that is a dealbreaker. I've solved it by making a class which is basically a List wrapper, which has an 'added' signal and a 'remove' signal. Upon adding an element to the list, the added signal is then wired, so it will create a new Row object and add it to the list box. That way, data is control in a manner Similar to ListModel binding. I can not make it work without calling the ShowAll method though.
private IObservableCollection<IVCard> _ivcards;
public IObservableCollection<IVCard> ivcards {
get{
return _ivcards;
}
set{
this._ivcards = value;
foreach(var card in this._ivcards){
base.prepend(new ContactRow(card));
}
this._ivcards.item_added.connect((item) => {
base.add(new ContactRow(item));
base.show_all();
});
base.show_all();
}
}
Even though this is by no means the best code I've come up with, it works very well.

switch UIThreadDispatcher in platform (Unity) that does not have a built-in dispatcher

I am using mvvmcross with Unity, but it doesn't have a built-in UIThreadDispatcher, there are many people had written it owns dispatcher which doesn't have a common interface. The following is my own implementation. Instead of create different MvxXXXUIThreadDispatcher, is there any other approach ?
public abstract class MvxUnityUIThreadDispatcher
: MvxMainThreadDispatcher
{
protected MvxUnityUIThreadDispatcher()
{
}
public bool RequestMainThreadAction(Action action)
{
UIThreadDispatcher.Instance.InvokeOrEnqueueOnMainThread(() => ExceptionMaskedAction(action));
return true;
}
}
The thread dispatcher is used by Mvx to make sure that UI controls are not called on non-UI threads.
If Unity doesn't need this, then you can just call action() in your implementation.
In the https://github.com/et-nowis/mvx-unity-ngui MvvmCross implementation it looks like they used SynchronizationContext:
public abstract class MvxUnityUIThreadDispatcher
: MvxMainThreadDispatcher
{
private readonly SynchronizationContext _uiSynchronizationContext;
protected MvxUnityUIThreadDispatcher()
{
if (SynchronizationContext.Current == null)
SynchronizationContext.SetSynchronizationContext(new UnitySynchronizationContext());
_uiSynchronizationContext = SynchronizationContext.Current;
}
public bool RequestMainThreadAction(Action action)
{
if (_uiSynchronizationContext == SynchronizationContext.Current)
action();
else
_uiSynchronizationContext.Post(state => ExceptionMaskedAction(action), null);
return true;
}
}
from https://github.com/et-nowis/mvx-unity-ngui/blob/master/Assets/External/MvvmCross/Cirrious/Cirrious.MvvmCross.Unity/Views/MvxUnityUIThreadDispatcher.cs - see source file there for their license terms on this code

RxJava (or Rx.NET) equivalent of ReactiveCocoa's RACObserve

Given an arbitrary field on a Java object, I want to create an Observable that will watch that field and push a new result to an Observer every time the value of the field changes. ReactiveCocoa has a macro called RACObserve, which appears to do exactly this.
I want to know how to implement similar functionality using RxJava.
For example, say I had the following simple class:
public class Foo {
enum State {
Idle,
Ready,
Error
}
private State currentState = State.Idle;
//methods that can change currentState
}
I want to create an Observable<State> that will push the new state to an Observer every time something changes the value of currentState.
In ReactiveCocoa, it looks like I would write something sort of like the following (please excuse my pseudo Objective-C):
[RACObserve(self, currentState) subscribeNext:^(NSString *newState) {
NSLog(#"%#", newState);
}];
How would I achieve similar functionality in RxJava? I'm thinking that I may need to wrap all changes to currentState in a setter, but it's not clear to me where I should then call Observable.create and how to feed the changes of currentState to an Observer.
ReactiveCocoa is actually more similar to ReactiveUI (http://www.reactiveui.net) than just plain Rx. And in ReactiveUI, you can use this.WhenAnyValue(x => x.PropName) to do exactly what you want.
I stumbled across this same problem recently, I ended up using PropertyChangeListener, which will emit an object when a property is changed, see the following:
Update Listener:
public class GameUpdateListener {
public static Observable<Object> changed(Game game) {
final BehaviorSubject<Object> subject = BehaviorSubject.create((Object)game);
game.addPropertyChangeListener(new PropertyChangeListener() {
#Override
public void propertyChange(PropertyChangeEvent propertyChangeEvent) {
subject.onNext( (Object)propertyChangeEvent.getNewValue());
}
});
return subject;
}
}
Some custom object:
public class Game {
private PropertyChangeSupport pcs = new PropertyChangeSupport(this);
...
public setSomeField(String field){
this.field = field;
pcs.firePropertyChange("field", this.field, field);
}
public void addPropertyChangeListener(PropertyChangeListener propertyChangeListener) {
pcs.addPropertyChangeListener(propertyChangeListener);
}
...
}
Observe:
Game game = new Game();
GameUpdateListener listener = new GameUpdateListener();
final Observable<Object> gameObserver = listener.changed(game);
gameObserver.subscribe(new Action1<Object>() {
#Override
public void call(Object o) {
Log.e(TAG, "Object Changed");
}
});
game.setSomeField("New value");
This will work fine as long as you don't need to instantiate your object again. Perhaps a solution to this is to create a local setter method and emit a change there.
Since your question title contains "or Rx.NET", here is my suggestion (I dunno bout RxJava, you may find something similar).
You probably will have to leverage some sort of mechanism in the setter. The standard way in .NET is by using the INotifyPropertyChanged interface.
Then by firing the events, you can create an IObservable<T> from this stream by using
Observable.FromEvent<TEvent, TArgs>()
You can find a really good example of what you want to do (.NET) here.
(credits to Rob Foncesa-Ensor)
I think what you are after is a Subject<T>. It implements IObserver<T>, so you can call OnNext(T) to fire a new value, as well as IObservable<T>, which you can expose it as publicly so it can be subscribed to.
If you need it to fire the latest value to new subscribers, you can use a ReplaySubject<T> with a buffer size of 1.
Here's a basic implementation:
public class SomeService
{
private Subject<int> values = new Subject<int>();
public IObservable<T> Values
{
get
{
// AsObservable prevents it from being cast back to Subject
return values.AsObservable();
}
}
// Private; called by some internal mechanism
private void SetValue(int newValue)
{
newValue.OnNext(newValue);
}
}

Correct way to call async methods from within a data-bound property setter?

Now I know properties do not support async/await for good reasons. But sometimes you need to kick off some additional background processing from a property setter - a good example is data binding in a MVVM scenario.
In my case, I have a property that is bound to the SelectedItem of a ListView. Of course I immediately set the new value to the backing field and the main work of the property is done. But the change of the selected item in the UI needs also to trigger a REST service call to get some new data based on the now selected item.
So I need to call an async method. I can't await it, obviously, but I also do not want to fire and forget the call as I could miss exceptions during the async processing.
Now my take is the following:
private Feed selectedFeed;
public Feed SelectedFeed
{
get
{
return this.selectedFeed;
}
set
{
if (this.selectedFeed != value)
{
this.selectedFeed = value;
RaisePropertyChanged();
Task task = GetFeedArticles(value.Id);
task.ContinueWith(t =>
{
if (t.Status != TaskStatus.RanToCompletion)
{
MessengerInstance.Send<string>("Error description", "DisplayErrorNotification");
}
});
}
}
}
Ok so besides the fact I could move out the handling from the setter to a synchronous method, is this the correct way to handle such a scenario? Is there a better, less cluttered solution I do not see?
Would be very interested to see some other takes on this problem. I'm a bit curious that I was not able to find any other discussions on this concrete topic as it seems very common to me in MVVM apps that make heavy use of databinding.
I have a NotifyTaskCompletion type in my AsyncEx library that is essentially an INotifyPropertyChanged wrapper for Task/Task<T>. AFAIK there is very little information currently available on async combined with MVVM, so let me know if you find any other approaches.
Anyway, the NotifyTaskCompletion approach works best if your tasks return their results. I.e., from your current code sample it looks like GetFeedArticles is setting data-bound properties as a side effect instead of returning the articles. If you make this return Task<T> instead, you can end up with code like this:
private Feed selectedFeed;
public Feed SelectedFeed
{
get
{
return this.selectedFeed;
}
set
{
if (this.selectedFeed == value)
return;
this.selectedFeed = value;
RaisePropertyChanged();
Articles = NotifyTaskCompletion.Create(GetFeedArticlesAsync(value.Id));
}
}
private INotifyTaskCompletion<List<Article>> articles;
public INotifyTaskCompletion<List<Article>> Articles
{
get { return this.articles; }
set
{
if (this.articles == value)
return;
this.articles = value;
RaisePropertyChanged();
}
}
private async Task<List<Article>> GetFeedArticlesAsync(int id)
{
...
}
Then your databinding can use Articles.Result to get to the resulting collection (which is null until GetFeedArticlesAsync completes). You can use NotifyTaskCompletion "out of the box" to data-bind to errors as well (e.g., Articles.ErrorMessage) and it has a few boolean convenience properties (IsSuccessfullyCompleted, IsFaulted) to handle visibility toggles.
Note that this will correctly handle operations completing out of order. Since Articles actually represents the asynchronous operation itself (instead of the results directly), it is updated immediately when a new operation is started. So you'll never see out-of-date results.
You don't have to use data binding for your error handling. You can make whatever semantics you want by modifying the GetFeedArticlesAsync; for example, to handle exceptions by passing them to your MessengerInstance:
private async Task<List<Article>> GetFeedArticlesAsync(int id)
{
try
{
...
}
catch (Exception ex)
{
MessengerInstance.Send<string>("Error description", "DisplayErrorNotification");
return null;
}
}
Similarly, there's no notion of automatic cancellation built-in, but again it's easy to add to GetFeedArticlesAsync:
private CancellationTokenSource getFeedArticlesCts;
private async Task<List<Article>> GetFeedArticlesAsync(int id)
{
if (getFeedArticlesCts != null)
getFeedArticlesCts.Cancel();
using (getFeedArticlesCts = new CancellationTokenSource())
{
...
}
}
This is an area of current development, so please do make improvements or API suggestions!
public class AsyncRunner
{
public static void Run(Task task, Action<Task> onError = null)
{
if (onError == null)
{
task.ContinueWith((task1, o) => { }, TaskContinuationOptions.OnlyOnFaulted);
}
else
{
task.ContinueWith(onError, TaskContinuationOptions.OnlyOnFaulted);
}
}
}
Usage within the property
private NavigationMenuItem _selectedMenuItem;
public NavigationMenuItem SelectedMenuItem
{
get { return _selectedMenuItem; }
set
{
_selectedMenuItem = val;
AsyncRunner.Run(NavigateToMenuAsync(_selectedMenuItem));
}
}
private async Task NavigateToMenuAsync(NavigationMenuItem newNavigationMenu)
{
//call async tasks...
}

Using IoC container as a service locator for HttpHandler

This question relates to my other post.
Ok so after a bit more messing around I decided to do it this way. Which seems to work fine when I run it, although I'm getting the following error in NUnit: Could not load file or assembly 'Castle.Core, Version=1.0.3.0, Culture=neutral, PublicKeyToken=407dd0808d44fbdc' or one of its dependencies. The located assembly's manifest definition does not match the assembly reference. (Exception from HRESULT: 0x80131040) So not sure what is happening there???
Just wanted to know what others thought about the design and if there are any obvious 'no no's' or improvements. I.e. Is the constructor of the base handler a good place to instantiate the windsor component or is there a better place to do this? As I said in the original post the idea behind doing things this way was to keep the components nicely decoupled and to make unit testing easy. I should also add I'm new to unit testing, mocking. Thanks!
public abstract class BaseHttpHandler : IHttpHandler
{
private HttpContext _httpContext;
private ILogger _logger;
private IDataRepository _dataRepository;
protected HttpRequest Request { get { return _httpContext.Request; } }
protected HttpResponse Response { get { return _httpContext.Response; } }
protected bool IsRequestFromUAD { get { return Request.UserAgent == null ? false : Request.UserAgent.Equals("UAD"); } }
protected ILogger Logger { get { return _logger; } }
protected IDataRepository DataRepository { get { return _dataRepository; } }
public virtual bool IsReusable { get { return false; } }
public BaseHttpHandler()
{
var container = new WindsorContainer(new XmlInterpreter(new ConfigResource("castle")));
_logger = container.Resolve<ILogger>();
_dataRepository = container.Resolve<IDataRepository>();
}
public void ProcessRequest(HttpContext context)
{
_httpContext = context;
ProcessRequest(new HttpContextWrapper(context));
}
public abstract void ProcessRequest(HttpContextBase context);
}
public class UADRecordHttpHandler : BaseHttpHandler
{
public override void ProcessRequest(HttpContextBase context)
{
if (IsRequestFromUAD)
{
using (var reader = new StreamReader(context.Request.InputStream))
{
string data = reader.ReadToEnd();
if (Logger != null)
Logger.Log(data);
if(DataRepository != null)
DataRepository.Write(data);
context.Response.Write(data);
}
}
else
ReturnResponse(HttpStatusCode.BadRequest);
}
}
That's a very bad thing to do, what you're doing here. You should have one instance of the container per application, while with this code you will have one per each request.
About the error in NUnit: make sure you don't have other versions of Castle assemblies in the GAC. If so, uninstall them.
About your BaseHttpHandler: the problem with this implementation is that you're creating a new container. Instead, use a single container per application, like Krzysztof said. Use a static service locator, e.g. CommonServiceLocator. (I never recommend this but it's one of the few places where it does make sense).