[Mlir-commits] [mlir] [MLIR] Fix Liveness analysis handling of unreachable code (PR #153973)

[Mehdi Amini]

joker-eph wrote:

> There is likely a problem with:
> 
> ```
>     // the execution of this pass might create new SSA values when erasing some
>     // of the results of an op and we know that these new values are live
>     // (because they weren't erased) and also their liveness is null because
>     // liveness analysis ran before their creation.
> ```
> 
> But we don't have any test coverage for this!! :(

This is fixed in the new approach in this PR. Still sad there is no test for this.

https://github.com/llvm/llvm-project/pull/153973