[Lldb-commits] [lldb] [lldb][test] Combine libstdc++ and libc++ std::variant tests into generic test (PR #147253)

[Pavel Labath via lldb-commits]

https://github.com/labath approved this pull request.

This looks good. The problem with caching is not related to this patch.

> For now i think I'll probably i'll reset the formatters in the teardown hook, but @jimingham comment on #110767 might be a good follow-up:
> 
> > Since this is just an optimization, we should first prove to ourselves that failing to put one of these type names in the exact matches really makes enough difference to bother with a complex solution.  If the gains are some but not huge, some boneheaded solution like never caching anything where the from type has a '<' and a '>' is probably sufficient.

However, I don't see how templates are related to this. I think the same thing can happen if I have `using MyType = TypeA;` in one binary and `using MyType = TypeB;` in another. If I cache the formatter for TypeA as the formatter for `MyType` (because that's correct for one binary) and then use that in the other binary, things will blow up. Technically, even `TypeA` in one binary need not have anything to do with `TypeA` in the second one, but we currently don't have a way to say which version of TypeA we are trying to format.

I think my ideal solution would be to have the formatter cache be per-target. I like that because it opens the door to having data formatters be registered per-target, which then means you don't have to have uber-formatters which know how to format every version of some struct.

https://github.com/llvm/llvm-project/pull/147253