Generic AddHttpClient selected instead of non-generic

[aklefdal]

I try to use Named HttpClient as described in the documentation from a Giraffe application.

    services.AddHttpClient(
        "poc",
        (fun httpClient -> httpClient.BaseAddress <- Uri("https://example.com")))
    |> ignore

Together with:

            let factory = ctx.GetService<IHttpClientFactory>()
            let httpClient = factory.CreateClient("poc")

This does not work since the F#-compiler chooses another overload to be used, the generic one.

Here is a small application that illustrates the problem: https://github.com/aklefdal/FSharpNamedHttpClient

• How am I supposed register and use named HttpClients in F#?
• Do I have to use classes and constructor injection of dependencies to make HttpClientFactory work?
• Can I at least have warning issued from the compiler and possibly a way to force the compiler to use the right overload?

This issue is possibly related to

• Generic arguments of Aspnetcore AddHttpClient method are wrongly inferred (silent bug) #12957
• Get more explicit help in choosing the right overload #11534

There is a workaround, and that is to add HttpClient as a generic argument to the call to AddHttpClient. This is not documented, nor do I know if it works in all cases. I don't even know why it works.

Update:
There is a second workaround, and that is explicitly to cast the second argument to Action<HttpClient>.

[smoothdeveloper]

I believe you can do it with specifying the Action<HttpClient> explicitly:

services.AddHttpClient("should pick the one you want", Action<HttpClient>(fun httpClient -> httpClient.BaseAddress <- Uri("..."))) |> ignore

It would be interesting to make sure every overload can be called and to have non regression tests about the generated IL.

[aklefdal]

Specifying the second parameter as Action<HttpClient> works as well. So this might just be the same issue as #11534

[smoothdeveloper]

@aklefdal it would be interesting to remove the "func is better" thing in the compiler and see how it behaves, and thanks to your issues being tracked, we can, in time, start putting test cases together around overload resolution and keep this in consideration to improve overload resolution.

[gusty]

Might be related to fsharp/fslang-suggestions#905

[0101]

We need to investigate and find out what's happening here exactly and then document the behavior or look if it can be solved as part of an existing suggestion.

We might not be able to fix it without breaking something else related to method resolution.