[Lldb-commits] [clang] [clang-tools-extra] [lldb] [clang] Reland: Instantiate alias templates with sugar (PR #101858)

[Matheus Izvekov via lldb-commits]

mizvekov wrote:

> I agree that we don't need `SubstTemplateTypeParmType` nodes if all resuraging that we ever do is related to types that the Clang frontend itself knows. However that is not universally true.
> 
> For example, we (Google) have a tool for inferring and checking nullability contracts (https://github.com/google/crubit/tree/main/nullability), and a tool for lifetime contracts (https://github.com/google/crubit/tree/main/lifetime_annotations). These tool rely on `SubstTemplateTypeParmType` nodes to propagate nullability and lifetime annotations through template signatures. We can't rely on the Clang frontend doing this propagation because this logic is in an external tool, and the whole notion of nullability and lifetimes is absent in Clang's upstream repository.
> 

Can we extend Clang in other ways where this propagation can happen within our transforms?

> To be clear: our tool right now is partially broken (when alias templates are involved), and I am not sure how to fix it when `SubstTemplateTypeParmType` nodes are not going to be available.

To be clear, it has been broken as of a couple of years already for some cases, we already perform this 'Final' substitution when instantiating default arguments for templates, for example.

This patch extends that to alias templates.

As for my first question, can you think of an alternative way we can go about this? What would you need to be upstreamed in order for our own transforms to handle this for you?

It sounds like it will be a lot of waste if we have our own resugarer within clang, but then you would be forced to implement your own external resugarer anyway.

https://github.com/llvm/llvm-project/pull/101858