Exclude additive classifications from "/highlight" requests

[DasCleverle]

Currently the behavior of the "/highlight" request is not deterministic for static symbols. This is because the Classifier class returns two ClassifiedSpans for static symbols (one with the actual kind and one with kind "static symbol"). And depending on which of the two spans is returned first in the result we get a different highlight span.

I propose a change where we remove these "additive" kinds from the result, only returning the actual kind of the span.

Fixes #1576

[nickspoons]

I can confirm that this PR fixes #1576, thanks @DasCleverle!

I'm not sure if it is related but I notice that WriteLine has kind method name when it is contains anything valid but has kind identifier when it contains e.g. an undeclared variable:

Console.WriteLine("a string");
Console.WriteLine(5);
object o = "another string";
Console.WriteLine(o.ToString());
Console.WriteLine(undeclared_variable.ToString());

Is this another situation where there are multiple spans?

[DasCleverle]

No, it seems that Roslyn can't determine the actual kind when there is a semantic error. This is what I got for the sample you provided:

[filipw]

thank you, this is a very good PR.
it's the same that Roslyn does internally https://github.com/dotnet/roslyn/blob/9ee3ac69f99cbcd645d494e0b0ee44616c1725e0/src/Workspaces/Core/Portable/Classification/ClassifierHelper.cs#L49-L55

(we might want to switch to their classifier in the future completely too, but that's a separate story)

[nickspoons]

I don't think the CI errors have anything to do with the PR - can someone trigger them for a rebuild?

[dnfclas]

Thank you for your submission, we really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

❌ DasCleverle sign now
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[DasOhmoff]

Thank you all very much for fixing this issue. It is very appreciated <3