[PATCH] D132142: [analyzer] Prefer wrapping SymbolicRegions by ElementRegions

[Balázs Benics via Phabricator via cfe-commits]

steakhal added a comment.

In D132142#3733629 <https://reviews.llvm.org/D132142#3733629>, @NoQ wrote:

> Like I vaguely mentioned in the thread, I'm really curious whether it's possible to canonicalize the *absence* of element regions instead, as it allows modeling pointer casts as no-op and avoids problems like D38797 <https://reviews.llvm.org/D38797> where results of even very basic operations are impossible to represent.

Yes, that would make more sense if we were getting rid of the current memory model. However, I'm not feeling confident working on that before finalizing a detailed plan.
For that, we would need to discuss the potential pros and cons thoroughly and then approach how it should look & we could incrementally land it. So, it definitely does not feels like a cheap thing to do.

For now, I believe this change makes the modeling of trivial copies a bit better from the user's perspective. In the future, we could still get rid of the memory model and implement a new one if we want.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D132142/new/

https://reviews.llvm.org/D132142