[clang] [Clang][Sema] Retain the expanding index for unevaluated type constraints (PR #109518)

[Younan Zhang via cfe-commits]

zyn0217 wrote:

> @zyn0217 
> 
> 
> 
> I am having second thoughts on both this PR and the follow up fix https://github.com/llvm/llvm-project/commit/adb0d8ddceb143749c519d14b8b31b481071da77
> 
> 
> 
> We shouldn't be adding information that is required for correct substitution into `SubstTemplateTypeParmType `, it's sugar after all and doesn't survive canonicalization. That's the reason we have created `SubstTemplateTypeParmPackType`, that one is canonical and is able to hold out information so we can finish substitution correctly later.
> 
> 
> 
> I think neither of these patches should have touched `SubstTemplateTypeParmType`, this is probably an issue with not holding out into `SubstTemplateTypeParmPackType`.
> 
> 
> 
> I know I reviewed the follow up patch, but I must have missed it was not touching the 'Pack' version of the node instead.

Thanks for revisiting the patch. I think it's worth exploring the idea. I'll take a look if I can move that to the Pack node. (Or if you have a patch, let me know so we don't step on each others' work)

I added that to the non-Pack node in the first place because of the fact that the node already has an optional Index to track exactly what I want, while the Pack node doesn't. And if I recall it correctly, the Pack node needs an expanded template arguments, and it needs some effort to find out that at the point we rebuild the constraint. So I took the shortcut.


https://github.com/llvm/llvm-project/pull/109518