AFAICT, there are two potential problems in general:
- If you’re using string identifiers, you potentially run into namespace issues (depending on the language semantics).
- In the case where it’s important to have the body of
mystruct1
available, both struct types would have to appear in the MLIR output. The workaround of which I am aware is – IMO – pretty hack-y.
I think this is possible today, though I would personally like to see a better solution to both of these issues.