Sam McCall

@ArcsinX > Seems preamble on disk is not enough. We also need other fields of PreambleData structure (Diags, Includes, etc.) + index update (which happens during preamble build). This is...

This sounds plausible, but without an example that clearly shows the problem, it's not actionable. There are many possible causes, I'd guess the most likely are: - libraries, particularly the...

Right, the server has a choice of how much text to the right of the cursor to eat when accepting a completion. My experience is that (non-clangd) editors are inconsistent...

I do understand what you're asking for, what's missing is a rationale for this being better in general. What is a user hoping to achieve when they trigger completion at...

We have more reports of leaking PCH files on crash, where signal handlers aren't doing the trick. (Possibly in an environment where our signal handlers are not being used) I'd...

Yep. We have the "define-outline" code action which is very closely related, so a workaround is: - replace `;` with `{}` - trigger "define outline" But we should support this...

Oh, there is a catch: it's kind of bad to offer this if a definition exists. This is mostly relevant in a header file, where the existing definition (if it...

Less clear examples: ``` struct Point { float x,y; }; array line = {1,2,3,4}; // {[0].x=1, [0].y=2, [1].x=3, [1].y=4} // {.x=1, .y=2, .x=3, .y=4} // {[0].x=1, .y=2, [0].x=3, .y=4} //...

So, what can we do? - tell people to use offUnlessPressed - doesn't work for other editors, and we're fighting defaults - make this configurable, and tell people to change...

This is a symptom of the AST being deficient here: there's no TypeLoc for Foo, and in the AST for `using enum a::Foo` there's no record of `a` at all....