[llvm] [RFC][TableGen] Require DAG argument for complex operands in InstAlias (PR #136411)

[Rahul Joshi via llvm-commits]

jurahul wrote:

> > is this backward in-compatible change?
> 
> It isn't. That's why I'd like to gather opinions, whether the new slightly more verbose syntax is helpful and whether it justifies rewriting `InstAlias` definitions. The latter may be non-trivial, as AArch64 changes show.

Thanks. Should we be concerned with downstream backends breaking due to this (if its approved)? Especially if adopting this is non-trivial. I guess we can discuss once we reach some consensus on the upstream change first.

https://github.com/llvm/llvm-project/pull/136411