[PATCH] D95644: Ensure that we traverse non-op() method bodys of lambdas

[Richard Smith - zygoloid via Phabricator via cfe-commits]

rsmith added inline comments.


================
Comment at: clang/include/clang/AST/RecursiveASTVisitor.h:2065
     if (const CXXRecordDecl *RD = MD->getParent()) {
-      if (RD->isLambda()) {
+      if (RD->isLambda() && RD->getLambdaCallOperator() == MD) {
         VisitBody = VisitBody && getDerived().shouldVisitLambdaBody();
----------------
In principle there can be multiple declarations of the lambda call operator if we merge multiple lambdas from different modules; we should check for any declaration of the `operator()` here rather than the exact one that `getLambdaCallOperator` returns.

I'd previously thought we could use `isLambdaCallOperator` here, but I think that's wrong: for a generic lambda, that would skip instantiations of the templated `operator()`, whereas I think you only want to skip the body of the primary template in that case, and still visit the instantiations, right? Eg, given:

```
auto x = [](auto n) { return n; };
int n = x(0);
```
... you'd want to skip the body of the template `<lambda>::operator()(T)`, but you'd still want to visit the body of `<lambda>::operator()<int>(int)`, right?


================
Comment at: clang/unittests/ASTMatchers/ASTMatchersTraversalTest.cpp:485
+  A a;
+  auto l = [a] { };
+  auto lCopy = l;
----------------
steveire wrote:
> I don't know how to create a lambda with a default ctor body.
I think that's probably not actually possible, sorry for misleading you on that! You can introduce a copy constructor with a body (by giving `A` a non-trivial copy constructor), though, if you want to test that.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D95644/new/

https://reviews.llvm.org/D95644