[Mlir-commits] [mlir] [mlir][Transforms] Dialect Conversion: Fix folder implementation (PR #150775)

[Matthias Springer]

matthias-springer wrote:

> It could be the case, for instance, that the new location being written has a pointer to the same context. In which case there is a true race on location, but there's no observable difference.

The context should always be the same. So indeed no observable difference.

> However, as I mentioned above, we can't do this for verification of TensorShardingAttr as a discardable attr: see [SdyDialect::verifyOperationAttribute](https://github.com/openxla/shardy/blob/main/shardy/dialect/sdy/ir/verifiers.cc#L1552). Maybe that's fine as that is invoked differently to OpTy::verify?

I'm not sure how `verifyOperationAttribute` is hooked up and still don't fully understand why there is a race in the first place. Given that switching to `SymbolUserOpInterface` fixes the issue for now, I'd go with that. (Whether this is worth investigating further is up to you. I don't see any immediate problem in the code.)


https://github.com/llvm/llvm-project/pull/150775