[clang] [Clang] enhance loop analysis to handle variable changes inside lambdas (PR #135573)

[Aaron Ballman via cfe-commits]

AaronBallman wrote:

> > I question if -Wloop-analysis should be completely rewritten using dataflow analysis, rather than AST based matching.
> 
> should changes inside functions be handled in the following cases?
> 
> https://github.com/llvm/llvm-project/blob/db0f754c5af8e6c96770533520bf8b17fc0dc977/clang/test/SemaCXX/warn-loop-analysis.cpp#L23
> 
> https://github.com/llvm/llvm-project/blob/db0f754c5af8e6c96770533520bf8b17fc0dc977/clang/test/SemaCXX/warn-loop-analysis.cpp#L8
> 
> > rewritten using dataflow analysis
> 
> [`ExprMutationAnalyzer`](https://github.com/llvm/llvm-project/blob/main/clang/lib/Analysis/ExprMutationAnalyzer.cpp)?
> 
> /cc @AaronBallman @zyn0217 @cor3ntin

Oh that's a fun case. I think the behavior of that test makes sense if there was no definition for `by_ref` because we'd want to assume the non-`const` reference parameter is modified. But because we can see the definition for `by_ref`, we know the body is empty and so the variable is not modified, and so issuing the diagnostic there would be reasonable. WDYT?

https://github.com/llvm/llvm-project/pull/135573