[llvm] [IPO] Avoid crash when DISubprogram has no type (PR #193420)

[Jiang Ning via llvm-commits]

JiangNingHX wrote:

> If we're not aware of any frontend that produces this IR, then I would prefer to reject it in the verifier, so we don't have to deal with it everywhere.

Thanks, I followed up on #186557 and confirmed that the reproducer is crafted/mutated IR, not frontend-emitted IR.

So I agree the current PR is framed too strongly. Given the current LangRef requirement for `DISubprogram::type`, verifier rejection seems like the better direction.

Unless there are objections, I'll stop pursuing this PR in its current form and follow up with a verifier-side fix instead.


https://github.com/llvm/llvm-project/pull/193420