SpanArguments with an exception handler

[parsonsmatt]

SpanArguments should have a field for determining what happens with a thrown exception. The existing default behavior is to do this:

        forM_ e $ \(SomeException inner) -> do
          setStatus s $ Error $ T.pack $ displayException inner
          recordException s [] Nothing inner

I think that recordException s [] Nothing inner is probably good to leave in, and the "customizable' stuff will be the setStatus s $ Error $ T.pack $ displayException inner.

Related: #90

Comment: #91 (comment)

I think I'd like to see something like:

newtype SpanExceptionBehavior { unSpanExceptionBehavior :: SomeException -> Maybe (Span -> IO ()) }
  deriving newtype (Semigroup, Monoid)

spanExceptionSetStatusError :: SpanExceptionBehavior
spanExceptionSetStatusError = 
  SpanExceptionBehavior \(SomeException inner) ->  
    Just \s -> setStatus s $ Error $ T.pack displayException inner

addSpanExceptionBehavior :: SpanExceptionBehavior -> SpanArguments -> SpanArguments
addSpanExceptionBehavior new sa = sa { exceptionBehavior = new <> exceptionBehavior sa }

That allows you to compose and combine them. However, it does not allow you to override them. So I think we need something other than Maybe there -

data OverrideOrCombine a 
    = Override a
    | Combine a
    | Nothin

instance Semigroup a => Semigroup (OverrideOrCombine a) where
    Override a <> _ = Override a
    _ <> Override a = Override a
    Combine a <> Combine b = Combine (a <> b)
    Nothin <> r = r
    l <> Nothin = l

Then, for the ExitSuccess, we can do an override:

ignoreExitSuccess :: ExceptionBehavior
ignoreExitSuccess = ExceptionBehavior \exn ->
    case fromException exn of
        exitCode 
            | ExitSuccess <- exitCode -> 
                Override mempty
        _ -> 
            Nothin

[parsonsmatt]

I've run into this issue twice now: once for ExitSuccess and again for SIGINT parsonsmatt/hotel-california#8

So I am interested in providing some good work on this 😄

[michaelpj]

Another case: ki kills threads with exception when you leave a scope, but this can be totally fine and not bad at all.