We Keep Coding

Resolving concrete types when using polymorphism in Autofac

Consider the following code:
public interface IFileBackup
{
Task Backup(byte[] file);
}
public class BackUpMechanismA : IFileBackup
{
//Implementation
public async Task Backup(byte[] file)
{
//Attempts to backup using mechanism A
}
}
public class BackUpMechanismB : IFileBackup
{
//Implementation
public async Task Backup(byte[] file)
{
//Attempts to backup using mechanism B
}
}
Then the calling class looks like this:
public class Caller
{
private readonly IFileBackup _backupA;
private readonly IFileBackup _backupB;
public Caller(IFileBackup backupA, IFileBackup backupB)
{
_backupA = backupA;
_backupB = backupB;
}
public async Task BackupFile(byte[] file)
{
try
{
await _backupA.Backup(file);
}
catch(SomeException)
{
await _backupB.Backup(file);
}
}
}
So what I'm trying to do here is to use polymorphism. So both BackupMechanismA and BackupMechanismB implements the Backup method in their own way. In the caller I want to attempt the first mechanism and if that doesn't work we catch an exception and try the second approach.
I'm having trouble resolving the correct implementations using Autofac. I have tried with:
builder.RegisterType<BackupMechanismA>().As<IFileBackup>().AsSelf();
builder.RegisterType<BackupMechanismB>().As<IFileBackUp>().AsSelf();
But this won't work because I still need to tell the caller which of the types to resolve. How do I do that in the caller?
Also, I'm in doubt whether this design is really the right design to go with. Before this design I just had one class with two different methods, one for mechanism A and one for mechanism B and then the caller would just call the different methods in the try catch. So I wanted to refactor this because the class got quite big and I wanted to separate the two different mechanisms into their own classes.
So, can I resolve this using Autofac? And is it the right design to go with for this scenario?

Agree with Jogge that iterating IFileBackups would be a better option, but creating an interface for each type is a no go. Instead, you could add a class which provides IEnumerable<IFileBackup> (an aggregate). For example:
public class BackupBundle : IEnumerable<IFileBackup>
{
private readonly List<IFileBackup> _backups = new List<IFileBackup>();
// default constructor creates default implementations
public BackupBundle()
: this(new List<IFileBackup> {new BackUpMechanismA(), new BackUpMechanismB()}) {}
// allow users to add custom backups
public BackupBundle(IEnumerable<IFileBackup> backups)
{
foreach (var backup in backups)
Add(backup);
}
public void Add(IFileBackup backup)
{
if (backup == null) throw new ArgumentNullException(nameof(backup));
_backups.Add(backup);
}
public IEnumerator<IFileBackup> GetEnumerator()
{
foreach (var backup in _backups)
yield return backup;
}
IEnumerator IEnumerable.GetEnumerator()
{
return GetEnumerator();
}
}
public class Caller
{
private readonly IEnumerable<IFileBackup> _backups;
public Caller(IEnumerable<IFileBackup> backups)
{
_backups = backups ?? throw new ArgumentNullException(nameof(backups));
}
public async Task BackupFile(byte[] file)
{
foreach (var b in _backups)
{
try
{
await b.Backup(file);
break;
}
catch (Exception e) { }
}
}
}
Registration can be done as follows:
builder.RegisterInstance(new BackupBundle()).As<IEnumerable<IFileBackup>>();
builder.RegisterType<Caller>();
which allows you to resolve by class name:
var caller = scope.Resolve<Caller>();
As you see, the BackupBundle has a dependency of BackUpMechanismA and BackUpMechanismB. You could get rid of it by introducing another layer of abstraction but I'd prefer not to do that. My main concern would be to make Caller more robust. You might want to introduce retry logic, timeouts, etc.

Try registering with a name and then resolve using the name:
builder.RegisterType<BackupMechanismA>().Named<IFileBackup>("BackUpMechanismA");
builder.RegisterType<BackupMechanismB>().Named<IFileBackUp>("BackUpMechanismB");
_backupA = container.ResolveNamed<IFileBackUp>
("BackUpMechanismA");
_backupB = container.ResolveNamed<IFileBackUp>
("BackUpMechanismB");
Resolve the instances during runtime, rather than injecting through the constructor. This will let you resolve to the respective type, as needed.
Let me know if this works.

To make your design work, you can try the next approach:
static void Main(string[] args)
{
var builder = new ContainerBuilder();
builder.RegisterType<BackUpMechanismA>().Keyed<IFileBackup>("A");
builder.RegisterType<BackUpMechanismB>().Keyed<IFileBackup>("B");
builder.RegisterType<Caller>()
.WithParameter((p, ctx) => p.Position == 0, (p, ctx) => ctx.ResolveKeyed<IFileBackup>("A"))
.WithParameter((p, ctx) => p.Position == 1, (p, ctx) => ctx.ResolveKeyed<IFileBackup>("B"));
IContainer container = builder.Build();
var caller = container.Resolve<Caller>();
Console.ReadKey();
}
However in my opinion you probably don't need such a polymorphism here. It will be much more obvious and descriptive to implement something like this:
public async Task BackupFile(byte[] file)
{
try
{
await BackUpToAmazonS3(file);
}
catch (AmazonS3LoadingException)
{
await BackUpToLocalDisk(file);
}
}
In this example it is obvious what is going on. And there in BackUpToAmazonS3 you can use some injected AmazonS3FileBackUp and in BackUpToLocalDisk use LocalDiskFileBackUp or whatever. The point is that you don't need a polymorphism, when you don't plan to change the implementation. In your context it should be clear? that you try to put backup to some remote storage, and then, if it fails, put in on local disk. You don't need to hide the meaning here. This is your logic and should be clear, when you read the code, I suppose. Hope it helps.

In my experience, you're better off by creating an interface for each type:
public interface IFileBackup
{
Task Backup(byte[] file);
}
public interface IBackUpMechanismA : IFileBackup
{
}
public class BackUpMechanismA : IBackUpMechanismA
{
//...
}
public interface IBackUpMechanismB : IFileBackup
{
}
public class BackUpMechanismB : IBackUpMechanismB
{
//...
}
If you don't want that, what you could do is getting injected a list of IFileBackup and just iterate them. If you register BackUpMechanismA first it will be the first in the list. I'm not sure if this is guaranteed, you have to look it up.
public class Caller
{
private readonly ICollection<IFileBackup> _fileBackups;
public Caller(ICollection<IFileBackup> fileBackups)
{
_fileBackups = fileBackups;
}
public async Task BackupFile(byte[] file)
{
foreach (var fileBackup in _fileBackups)
{
try
{
await fileBackup.Backup(file);
break;
}
catch { }
}
}
}

Do I need Task.Run to call an async method from a constructor? [duplicate]

I have a project where I'm trying to populate some data in a constructor:
public class ViewModel
{
public ObservableCollection<TData> Data { get; set; }
async public ViewModel()
{
Data = await GetDataTask();
}
public Task<ObservableCollection<TData>> GetDataTask()
{
Task<ObservableCollection<TData>> task;
//Create a task which represents getting the data
return task;
}
}
Unfortunately, I'm getting an error:
The modifier async is not valid for this item
Of course, if I wrap in a standard method and call that from the constructor:
public async void Foo()
{
Data = await GetDataTask();
}
it works fine. Likewise, if I use the old inside-out way
GetData().ContinueWith(t => Data = t.Result);
That works too. I was just wondering why we can't call await from within a constructor directly. There are probably lots of (even obvious) edge cases and reasons against it, I just can't think of any. I've also search around for an explanation, but can't seem to find any.

Since it is not possible to make an async constructor, I use a static async method that returns a class instance created by a private constructor. This is not elegant but it works ok.
public class ViewModel
{
public ObservableCollection<TData> Data { get; set; }
//static async method that behave like a constructor
async public static Task<ViewModel> BuildViewModelAsync()
{
ObservableCollection<TData> tmpData = await GetDataTask();
return new ViewModel(tmpData);
}
// private constructor called by the async method
private ViewModel(ObservableCollection<TData> Data)
{
this.Data = Data;
}
}

Constructor acts very similarly to a method returning the constructed type. And async method can't return just any type, it has to be either “fire and forget” void, or Task.
If the constructor of type T actually returned Task<T>, that would be very confusing, I think.
If the async constructor behaved the same way as an async void method, that kind of breaks what constructor is meant to be. After constructor returns, you should get a fully initialized object. Not an object that will be actually properly initialized at some undefined point in the future. That is, if you're lucky and the async initialization doesn't fail.
All this is just a guess. But it seems to me that having the possibility of an async constructor brings more trouble than it's worth.
If you actually want the “fire and forget” semantics of async void methods (which should be avoided, if possible), you can easily encapsulate all the code in an async void method and call that from your constructor, as you mentioned in the question.

Your problem is comparable to the creation of a file object and opening the file. In fact there are a lot of classes where you have to perform two steps before you can actually use the object: create + Initialize (often called something similar to Open).
The advantage of this is that the constructor can be lightweight. If desired, you can change some properties before actually initializing the object. When all properties are set, the Initialize/Open function is called to prepare the object to be used. This Initialize function can be async.
The disadvantage is that you have to trust the user of your class that he will call Initialize() before he uses any other function of your class. In fact if you want to make your class full proof (fool proof?) you have to check in every function that the Initialize() has been called.
The pattern to make this easier is to declare the constructor private and make a public static function that will construct the object and call Initialize() before returning the constructed object. This way you'll know that everyone who has access to the object has used the Initialize function.
The example shows a class that mimics your desired async constructor
public MyClass
{
public static async Task<MyClass> CreateAsync(...)
{
MyClass x = new MyClass();
await x.InitializeAsync(...)
return x;
}
// make sure no one but the Create function can call the constructor:
private MyClass(){}
private async Task InitializeAsync(...)
{
// do the async things you wanted to do in your async constructor
}
public async Task<int> OtherFunctionAsync(int a, int b)
{
return await ... // return something useful
}
Usage will be as follows:
public async Task<int> SomethingAsync()
{
// Create and initialize a MyClass object
MyClass myObject = await MyClass.CreateAsync(...);
// use the created object:
return await myObject.OtherFunctionAsync(4, 7);
}

if you make constructor asynchronous, after creating an object, you may fall into problems like null values instead of instance objects. For instance;
MyClass instance = new MyClass();
instance.Foo(); // null exception here
That's why they don't allow this i guess.

In this particular case, a viewModel is required to launch the task and notify the view upon its completion. An "async property", not an "async constructor", is in order.
I just released AsyncMVVM, which solves exactly this problem (among others). Should you use it, your ViewModel would become:
public class ViewModel : AsyncBindableBase
{
public ObservableCollection<TData> Data
{
get { return Property.Get(GetDataAsync); }
}
private Task<ObservableCollection<TData>> GetDataAsync()
{
//Get the data asynchronously
}
}
Strangely enough, Silverlight is supported. :)

I was just wondering why we can't call await from within a constructor directly.
I believe the short answer is simply: Because the .Net team has not programmed this feature.
I believe with the right syntax this could be implemented and shouldn't be too confusing or error prone. I think Stephen Cleary's blog post and several other answers here have implicitly pointed out that there is no fundamental reason against it, and more than that - solved that lack with workarounds. The existence of these relatively simple workarounds is probably one of the reasons why this feature has not (yet) been implemented.

calling async in constructor maybe cause deadlock, please refer to
http://social.msdn.microsoft.com/Forums/en/winappswithcsharp/thread/0d24419e-36ad-4157-abb5-3d9e6c5dacf1
http://blogs.msdn.com/b/pfxteam/archive/2011/01/13/10115163.aspx

Some of the answers involve creating a new public method. Without doing this, use the Lazy<T> class:
public class ViewModel
{
private Lazy<ObservableCollection<TData>> Data;
async public ViewModel()
{
Data = new Lazy<ObservableCollection<TData>>(GetDataTask);
}
public ObservableCollection<TData> GetDataTask()
{
Task<ObservableCollection<TData>> task;
//Create a task which represents getting the data
return task.GetAwaiter().GetResult();
}
}
To use Data, use Data.Value.

C# doesn't allow async constructors. Constructors are meant to return fast after some brief initialization. You don't expect and you don't want to wait for an instance i.e. the constructor to return. Therefore, even if async constructors were possible, a constructor is not a place for long-running operations or starting background threads. The only purpose of a constructor is initialization of instance or class members to a default value or the captured constructor parameters. You always create the instance and then call DoSomething() on this instance. Async operations are no exception. You always defer expensive initialization of members.
There are a few solutions to avoid the requirement of async constructors.
A simple alternative solution using Lazy<T> or AsyncLazy<T> (requires to install the Microsoft.VisualStudio.Threading package via the NuGet Package Manager). Lazy<T> allows to defer the instantiation or allocation of expensive resources.
public class OrderService
{
public List<object> Orders => this.OrdersInitializer.GetValue();
private AsyncLazy<List<object>> OrdersInitializer { get; }
public OrderService()
=> this.OrdersInitializer = new AsyncLazy<List<object>>(InitializeOrdersAsync, new JoinableTaskFactory(new JoinableTaskContext()));
private async Task<List<object>> InitializeOrdersAsync()
{
await Task.Delay(TimeSpan.FromSeconds(5));
return new List<object> { 1, 2, 3 };
}
}
public static void Main()
{
var orderService = new OrderService();
// Trigger async initialization
orderService.Orders.Add(4);
}
You can expose the data using a method instead of a property
public class OrderService
{
private List<object> Orders { get; set; }
public async Task<List<object>> GetOrdersAsync()
{
if (this.Orders == null)
{
await Task.Delay(TimeSpan.FromSeconds(5));
this.Orders = new List<object> { 1, 2, 3 };
}
return this.Orders;
}
}
public static async Task Main()
{
var orderService = new OrderService();
// Trigger async initialization
List<object> orders = await orderService.GetOrdersAsync();
}
Use an InitializeAsync method that must be called before using the instance
public class OrderService
{
private List<object> orders;
public List<object> Orders
{
get
{
if (!this.IsInitialized)
{
throw new InvalidOperationException();
}
return this.orders;
}
private set
{
this.orders = value;
}
}
public bool IsInitialized { get; private set; }
public async Task<List<object>> InitializeAsync()
{
if (this.IsInitialized)
{
return;
}
await Task.Delay(TimeSpan.FromSeconds(5));
this.Orders = new List<object> { 1, 2, 3 };
this.IsInitialized = true;
}
}
public static async Task Main()
{
var orderService = new OrderService();
// Trigger async initialization
await orderService.InitializeAsync();
}
Instantiate the instance by passing the expensive arguments to the constructor
public class OrderService
{
public List<object> Orders { get; }
public async Task<List<object>> OrderService(List<object> orders)
=> this.Orders = orders;
}
public static async Task Main()
{
List<object> orders = await GetOrdersAsync();
// Instantiate with the result of the async operation
var orderService = new OrderService(orders);
}
private static async Task<List<object>> GetOrdersAsync()
{
await Task.Delay(TimeSpan.FromSeconds(5));
return new List<object> { 1, 2, 3 };
}
Use a factory method and a private constructor
public class OrderService
{
public List<object> Orders { get; set; }
private OrderServiceBase()
=> this.Orders = new List<object>();
public static async Task<OrderService> CreateInstanceAsync()
{
var instance = new OrderService();
await Task.Delay(TimeSpan.FromSeconds(5));
instance.Orders = new List<object> { 1, 2, 3 };
return instance;
}
}
public static async Task Main()
{
// Trigger async initialization
OrderService orderService = await OrderService.CreateInstanceAsync();
}

you can use Action inside Constructor
public class ViewModel
{
public ObservableCollection<TData> Data { get; set; }
public ViewModel()
{
new Action(async () =>
{
Data = await GetDataTask();
}).Invoke();
}
public Task<ObservableCollection<TData>> GetDataTask()
{
Task<ObservableCollection<TData>> task;
//Create a task which represents getting the data
return task;
}
}

you can create a wrapper and inject a functor representing the constructor:
class AsyncConstruct<T>
where T: class
{
private readonly Task<T> m_construction;
private T m_constructed;
public AsyncConstruct(Func<T> createFunc)
{
m_constructed = null;
m_construction = Task.Run(()=>createFunc());
}
public T Get()
{
if(m_constructed == null)
{
m_constructed = m_construction.Result;
}
return m_constructed;
}
}

Please bump this language request:
https://github.com/dotnet/csharplang/discussions/419
The amount of boilerplate code everyone needs to write to have a fully initialized async object is crazy and completely opposite of the trend in C# (less boilerplate).

I would use something like this.
public class MyViewModel
{
public MyDataTable Data { get; set; }
public MyViewModel()
{
loadData(() => GetData());
}
private async void loadData(Func<DataTable> load)
{
try
{
MyDataTable = await Task.Run(load);
}
catch (Exception ex)
{
//log
}
}
private DataTable GetData()
{
DataTable data;
// get data and return
return data;
}
}
This is as close to I can get for constructors.

I use this easy trick.
public sealed partial class NamePage
{
private readonly Task _initializingTask;
public NamePage()
{
_initializingTask = Init();
}
private async Task Init()
{
/*
Initialization that you need with await/async stuff allowed
*/
}
}

I'm not familiar with the async keyword (is this specific to Silverlight or a new feature in the beta version of Visual Studio?), but I think I can give you an idea of why you can't do this.
If I do:
var o = new MyObject();
MessageBox(o.SomeProperty.ToString());
o may not be done initializing before the next line of code runs. An instantiation of your object cannot be assigned until your constructor is completed, and making the constructor asynchronous wouldn't change that so what would be the point? However, you could call an asynchronous method from your constructor and then your constructor could complete and you would get your instantiation while the async method is still doing whatever it needs to do to setup your object.

How can I call async method from constructor?

I need to call a async method from my Form1 constructor. Since a constructor can't have a return type, I can't add a async void. I read that static constructor can be async but I need to call methods from constructor that aren't static, such as InitializeComponent() (since it's the Form's constructor).
The class is:
public partial class Form1 : Form
{
InitializeComponent();
//some stuff
await myMethod();
}
I read this one too but I still don't know how to implement this (in my case) since the method still requires to use async.

Don't do this in the constructor but in the loaded event of the window instead.
You can mark the loaded eventhandler as async.

You can use a static method that returns an instance of your form
public class TestForm : Form
{
private TestForm()
{
}
public static async Task<TestForm> Create()
{
await myMethod();
return new TestForm();
}
}

My sample is to call student details from page constructor
1- the calling of navigation page
void Handle_ItemTapped(object sender, Xamarin.Forms.ItemTappedEventArgs e)
{
Student _student = (Student)e.Item;
Navigation.PushAsync(new Student_Details(_student.ID));
}
2 - the details page
public partial class Student_Details : ContentPage
{
public Student_Details(int id)
{
InitializeComponent();
Task.Run(async () => await getStudent(id));
}
public async Task<int> getStudent(int id)
{
Student _student;
SQLiteDatabase db = new SQLiteDatabase();
_student = await db.getStudent(id);
return 0;
}
}

While common advice dictates you generally shouldn't do it in the constructor, you can do the following, which I have used in apps, such as console apps, where I need to call some existing async code:
DetailsModel details = null; // holds the eventual result
var apiTask = new Task(() => details = MyService.GetDetailsAsync(id).Result); // creates the task with the call on another thread
apiTask.Start(); // starts the task - important, or you'll spin forever
Task.WaitAll(apiTask); // waits for it to complete
Philip is correct that, if you can avoid doing this in a constructor, you should.

Task.Run(async () => await YourAsyncMethod());

Generic constraint based on non-implementation of interface

I have an application with a factory service to allow construction of instances while resolving the necessary dependency injection. For instance, I use this to construct dialog view models. I have a service interface that looks like this:
public interface IAsyncFactory
{
Task<T> Build<T>() where T: class, IAsyncInitialize;
}
Ideally, what I'd like to have is something like this (pseudo-syntax, as this isn't directly achievable)
public interface IFactory
{
Task<T> Build<T>() where T: class, IAsyncInitialize;
T Build<T>() where T: class, !IAsyncInitialize;
}
The idea here is that if a class supports IAsyncInitialize, I'd like the compiler to resolve to the method that returns Task<T> so that it's obvious from the consuming code that it needs to wait for initialization. If the class does not support IAsyncInitialize, I'd like to return the class directly. The C# syntax doesn't allow this, but is there a different way to achieve what I'm after? The main goal here is to help remind the consumer of the class of the correct way to instantiate it, so that for classes with an asynchronous initialization component, I don't try to use it before it has been initialized.
The closest I can think of is to create separate Build and BuildAsync methods, with a runtime error if you call Build for an IAsyncInitialize type, but this doesn't have the benefit of catching errors at compile time.

In general Microsoft suggests to add async suffix when naming asynchronous methods. Thus, your assumption of creating two methods named as Build and BuildAsync makes sense.
I think there is no way to enforce something like "all types that do not implement IAsyncInitialize shall use Build method instead of BuildAsync" unless you force the developers to mark synchronous methods with another interface like ISynchronousInitialize.
You may try the following;
instead of having to separate methods, just implement one BuildAsync method which has the following signature:
Task<T> BuildAsync<T>() where T: class
In the BuildAsync method check whether T implements IAsyncInitialize. If this is the case, just call related initialization code after creating the object of type T. Otherwise, just create a TaskCompletionSource object and run the synchronous initialization code as if it is asynchronous.

The following approach might not be the best, but I find it very convenient. When both asynchronous and synchronous initializers are available (or possibly can be available), I wrap the synchronous one as asynchronous with Task.FromResult, and only expose the asynchronous method to the client:
public interface IAsyncInitialize
{
Task InitAsync();
int Data { get; }
}
// sync version
class SyncClass : IAsyncInitialize
{
readonly int _data = 1;
public Task InitAsync()
{
return Task.FromResult(true);
}
public int Data { get { return _data; } }
}
// async version
class AsyncClass: IAsyncInitialize
{
int? _data;
public async Task InitAsync()
{
await Task.Delay(1000);
_data = 1;
}
public int Data
{
get
{
if (!_data.HasValue)
throw new ApplicationException("Data uninitalized.");
return _data.Value;
}
}
}
This leaves only the asynchronous version of the factory:
public interface IAsyncFactory
{
// Build can create either SyncClass or AsyncClass
Task<T> Build<T>() where T: class, IAsyncInitialize;
}
Furthermore, I prefer to avoid dedicated initializer methods like InitAsync, and rather expose asynchronous properties directly as tasks:
public interface IAsyncData
{
Task<int> AsyncData { get; }
}
// sync version
class SyncClass : IAsyncData
{
readonly Task<int> _data = Task.FromResult(1);
public Task<int> AsyncData
{
get { return _data; }
}
}
// async versions
class AsyncClass : IAsyncData
{
readonly Task<int> _data = GetDataAsync();
public Task<int> AsyncData
{
get { return _data; }
}
private static async Task<int> GetDataAsync()
{
await Task.Delay(1000);
return 1;
}
}
In either case, it always imposes asynchrony on the client code, i.e.:
var sum = await provider1.AsyncData + await provider2.AsyncData;
However, I don't think it's an issue as the overhead of Task.FromResult and await Task.FromResult for the synchronous version is quite low. I'm going to post some benchmarks.
The approach using asynchronous properties can be further improved with Lazy<T>, e.g. like this:
public class AsyncData<T>
{
readonly Lazy<Task<T>> _data;
// expose async initializer
public AsyncData(Func<Task<T>> asyncInit, bool makeThreadSafe = true)
{
_data = new Lazy<Task<T>>(asyncInit, makeThreadSafe);
}
// expose sync initializer as async
public AsyncData(Func<T> syncInit, bool makeThreadSafe = true)
{
_data = new Lazy<Task<T>>(() =>
Task.FromResult(syncInit()), makeThreadSafe);
}
public Task<T> AsyncValue
{
get { return _data.Value; }
}
}

Using IObservable<T> to keep track of current state

Suppose I have an object that observes an IObservable so that it's always aware of the current state of some external source. Internally my object has a method that uses that external value as part of the operation:
public class MyObject
{
public MyObject(IObservable<T> externalSource) { ... }
public void DoSomething()
{
DoSomethingWith(CurrentT);
}
}
What's the idomatic 'reactive' way of using IObservable for 'tracking current state' instead of 'responding to stream of events'.
Idea #1 is to just monitor the observable and write down values as they come in.
public class MyObject
{
private T CurrentT;
public MyObject(IObservable<T> externalSource)
{
externalSource.Subscribe((t) => { CurrentT = t; });
}
public void DoSomething()
{
DoSomethingWith(CurrentT);
}
}
And that's fine, but keeping track of the state in a class member seems very un-reactive-y.
Idea #2 is to use a BehaviorSubject
public class MyObject
{
private readonly BehaviorSubject<T> bs;
public MyObject(BehvaiorSubject<T> externalSource)
{
this.bs = externalSource
}
public void DoSomething()
{
DoSomethingWith(bs.Value);
}
}
But using subjects directly seems to be frowned upon. But at least in this case I have the ability to use a readonly field to store the behaviorsubject.
The BehaviorSubject (or ReplaySubject) does seem like it was made for this purpose, but is there some other better way here? And if I should use the subject, would it make more sense to take the subject as an injected parameter, or take the original observable and build the subject locally in the constructor?
(by the way I'm aware about the need to deal with the 1st value if the source observable hasn't fired yet. Don't get hung up on that, that's not what I'm asking about)

I'd go with a generic solution utilizing the ReactiveUI library. RUI has a standard way of mapping IObservable<T> to an INotifyPropertyChanged stateful property.
public class ObservableToINPCObject<T> : ReactiveObject, IDisposable
{
ObservableAsPropertyHelper<T> _ValueHelper;
public T Value {
get { return _ValueHelper.Value; }
}
public ObservableToINPCObject(IObservable<T> source, T initial = default(T))
{
_ValueHelper = source.ToProperty(this, p=>p.Value, initial);
}
public Dispose(){
_ValueHelper.Dispose();
}
}
ValueHelper is contains both the current state of the observable and automatically triggers the correct INPC notification when the state changes. That's quite a bit of boiler plate handled for you.
and an extension method
public static class ObservableToINPCObject {
public static ObservableToINPCObject<T> ToINPC<T>
( this IObservable<T> source, T init = default(T) )
{
return new ObservableToINPCObject(source, init);
}
}
now given an
IObservable<int> observable;
you can do
var obj = observable.ToINPC(10);
and to get the latest value
Console.WriteLine(obj.Value);
also given that Value is an INPC supporting property you can use it in databinding. I use ToProperty all the time for exposing my observables as properties for WPF databinding.

To be Rx-ish I'd suggest avoiding the second option and go with your first, but modified in one of two ways.
Either (1) make your class disposable so that you can cleanly close off the subscription to the observables or (2) make a method that lets you clean up individual observables.
(1)
public class MyObject : IDisposable
{
private T CurrentT;
private IDisposable Subscription;
public MyObject(IObservable<T> externalSource)
{
Subscription = externalSource
.Subscribe((t) => { CurrentT = t; });
}
public void Dispose()
{
Subscription.Dispose();
}
public void DoSomething()
{
DoSomethingWith(CurrentT);
}
}
(2)
public class MyObject
{
private T CurrentT;
public IDisposable Observe(IObservable<T> externalSource)
{
return externalSource
.Subscribe((t) => { CurrentT = t; });
}
public void DoSomething()
{
DoSomethingWith(CurrentT);
}
}
Both allow proper clean-up and both don't use a subject.