[Mlir-commits] [mlir] [MLIR][MemRef] Fix DimOfReifyRankedShapedTypeOpInterface IR-change on failure (PR #188973)

[Mehdi Amini]

joker-eph wrote:

> with a different approach that can explicitly track all operations created and erase them on failure instead of having to walk the program back.

On which metric would that really be better?

I would agree for compile time (and probably engineering robustness) that **not creating** useless operations in the first place instead of having to clean them up would be a better approach, but that's not what you're proposing here...

https://github.com/llvm/llvm-project/pull/188973