[Mlir-commits] [mlir] [MLIR][LLVM] Add bytecode support for several attributes (PR #162577)

[Tobias Gysi]

gysit wrote:

> Not in the way we'd like (a nice tablegen choke perhaps) but because these tablegen generate call to the builders, we'd get an error for the builder call mismatch, which is not ideal but works (for most part?). Note that the [first extra commit](https://github.com/llvm/llvm-project/pull/162577/commits/aa91a1abc7283626694e3168297295b740ac6a3b) in this PR was due to exactly this: when bringing the work upstream, DICompositeTypeAttr had a different order and that was caught at build time.

That is indeed not optimal. E.g. the builder may still match if the position of two integers is swapped. I guess we have to rely on roundtrip tests then. 

https://github.com/llvm/llvm-project/pull/162577