Warning FS0073: internal error: Typar not found during XmlDoc generation

[KevinRansom]

[from fsbugs] Ported from Codeplex

Simplified repro of root cause:

let rec foo< > c = bar c
and bar c = 0

changing the order of lines 1 and 2 fixes the problem

Original repro:

 C:\Users\latkin\Desktop\FSharp.DataFrame-typar-error\FSharp.DataFrame-typar-error\src\Frame.fs(251,28): warning FS0073: internal error: Typar not found during XmlDoc generation

/// [category:Series operations]
member frame.SeriesApply<'T>(strict, f:Func<Series<'TRowKey, 'T>, ISeries<_>>) = 
    frame.Columns
    |> Series.mapValues (fun os ->
        match os.TryAs<'T>(strict) with
        | OptionalValue.Present s -> f.Invoke s
        | _ -> os :> ISeries<_>)
    |> Frame<'TRowKey, 'TColumnKey>.FromColumnsNonGeneric

[forki]

I think the code sample is not correctly displayed

[latkin]

Updated with what I believe is proper formatting (though the original repro is probably not useful given the greatly simplified repro), and a link back to original codeplex bug.

[forki]

The foo function still looks strange. Is it really a space between the angle brackets?

[latkin]

The code is not properly formed, but it does repro the bug. Repro also occurs with more fully-formed code:

let rec foo<'t> c = bar c
and bar c = 0

[latkin]

Oddly enough, the workaround compiles successfully, even with the (I would think) invalid code < >:

let rec bar c = 0
and foo< > c = bar c

[vladima]

@forki, foo< > stands for type application with an empty list of type parameters (6.4 Basic application expressions)

[dsyme]

@latkin Hmm... I labelled this as fix ready in May - did the typar check PR fixed this and I just forgot to reference the issue?

[latkin]

With current codebase, repro here #33 (comment) now fails with test.fs(1,9): error FS3188: Type inference caused an inference type variable to escape its scope. Consider adding type annotations to make your code less generic.

That's better than an ICE but it still doesn't compile.

[dsyme]

That's a sufficient resolution. In that case we can close this.