[PATCH] D72041: [clangd] Handle go-to-definition in macro invocations where the target appears in the expansion multiple times

[Nathan Ridge via Phabricator via cfe-commits]

nridge marked 4 inline comments as done.
nridge added inline comments.


================
Comment at: clang-tools-extra/clangd/Selection.cpp:261
+      // consider it selected.
+      if (!SeenMacroCalls.insert(ArgStart).second) {
+        return NoTokens;
----------------
sammccall wrote:
> sammccall wrote:
> > Given the following program:
> > ```
> > #define SQUARE(x) x * x;
> > int four = [[SQUARE(2)]];
> > ```
> > We're going to now report the binary operator and one of the operands as selected and not the other, which doesn't seem desirable.
> > 
> > I think we want to accept macro-selected || arg-selected, so probably doing the current "non-argument macro expansion" first unconditionally or factoring it out into a function.
> > 
> > This will change the behavior of `int four = [[SQUARE]](2)` to consider the literal children selected too, I think this is fine.
> I don't think it's a good idea to add hidden state and side-effects to testChunk() - it breaks a lot of assumptions that help reason about the code, and using `mutable` hides the violation of them.
> (And a possible source of bugs - this is first in traversal order rather than first in source order - these are mostly but IIRC not always the same).
> 
> Instead I think you can do this statelessly: from the top-level spelling location, walk down with `SM.getMacroArgExpandedLocation` until you hit the target FileID (this is the first-expansion of first-expansion of first-expansion...) or the FileID stops changing (you've reached the innermost macro invocation, and your target location was on a different branch).
I agree that adding state is not great. I thought it was icky as I was writing it, I just couldn't think of an alternative. Thank you for suggesting one!

I implemented what you suggested, and it seems to work. I did want to ask a clarifying question to make sure I understand correctly: when an argument occurs multiple times in a macro exapnsion, the occurrences will have distinct `FileID`s (as opposed just different offsets in the same macro `FileID`)?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D72041/new/

https://reviews.llvm.org/D72041