Nathan Ridge

> * Could I ask someone more familiar with the codebase than I check to make sure the source-file-based inference is being properly used here, parallel to https://reviews.llvm.org/D116167? I tried...

> [As an aside, returning an empty compile command seems like an unnecessarily destructive way for `QueryDriverDatabase` to fail. It could at least just return the incoming compile command unmodified,...

> I will try to write a patch that fixes the main issue here (and #1089 as well) by changing the order of `QueryDriverDatabase` and `CommandMangler`. I'm close to a...

Proposed patches: https://reviews.llvm.org/D133756 https://reviews.llvm.org/D133757 They also fix #1089 and #1263

Sorry, I've been out of commission for a bit due to travel... I'm back now though and getting the review comments on those patches addressed is very near the top...

@cpsauer Patches just merged! Once you or your users get a chance to test this, I would be curious to know if it in fact resolved this issue (and #1263).

> Best way is to wait for the next weekly prerelease and then have folks test, right? On Linux you can get nightly packages from https://apt.llvm.org/, but otherwise yeah.

Is this referring to the "Extract variable" refactoring, or a new refactoring? An example with code before/after would be helpful to illustrate.

The exact behaviour likely depends on the macro as well. Here's the first thing I tried: ```c++ struct Point { int x; int y; }; #define ADD(x, y) x +...

With #1142 fixed, the diagnostics about unrecognized arguments could be suppressed using https://clangd.llvm.org/config.html#suppress. Not sure what would need to happen to clangd to actually recognize these arguments.