Sam McCall

So this "fixes" ifdef guards (makes them considered controlling if unclosed at end of preamble): ``` if (PP->isRecordingPreamble() && PP->isInPrimaryFile()) { PP->setRecordedPreambleConditionalStack(ConditionalStack); + if (!ConditionalStack.empty()) + MIOpt.ExitTopLevelConditional(); ConditionalStack.clear(); } ```...

> Which I guess isn't totally implausible Actually I think it's fine. PCH files are only used for `-include` on the command-line, and the `-cc1 -include-pch` replaces the `-include` if...

Haha, of course replacing the preamble buffer with nulls/zeros doesn't work - the idea is to match the size of the buffer we're using the preamble with, but of course...

OK, I have a fairly ugly fix for this: for the mainfile we store the HeaderFileInfo twice, once with a size of zero. When loading a HeaderFileInfo, if a) we're...

Patch in https://reviews.llvm.org/D78038, not sure whether we should try to land it.

This is ugly and inconsistent for sure. An attempt to explain the behaviors (though not defend them): - documentHighlight: our implementation has no macro support. (There's a FIXME, patches welcome!)...

I think filtering out edits where the replaced text != the original selected identifier would go a long way - we do that for index results IIRC. I know it...

> I do wonder why references can't report the same set of results that rename would change They're certainly similar operations but i don't see how we can make them...

Yeah, that seems a bit unfortunate. It falls out of other decisions. In this case we're not indexing reserved names at all, so renaming them wouldn't be safe. The goals...

Yeah, new file latency is the worst. And we don't have big plans to fundamentally reduce it. Can you explain a bit more about the workflow where you think this...