[clang-tools-extra] clangd: Make callHierarchy follow inheritance (PR #163024)

[via cfe-commits]

================
@@ -94,6 +106,9 @@ class MemIndex : public SymbolIndex {
   static_assert(sizeof(RelationKind) == sizeof(uint8_t),
                 "RelationKind should be of same size as a uint8_t");
   llvm::DenseMap<std::pair<SymbolID, uint8_t>, std::vector<SymbolID>> Relations;
+  // Reverse relations, currently only for OverridenBy
+  llvm::DenseMap<std::pair<SymbolID, uint8_t>, std::vector<SymbolID>>
----------------
timon-ul wrote:

I am a bit torn, I absolutely get that saving some memory here is a good idea, but also it makes me question a bit that all the changes have been kept generic. What I mean is:

- `reverseRelations` is a generic name, even though it will be very specific for `OverriddenBy`
- `reverseRelations` gets a RelationsRequest, but again, we only accept `OverriddenBy` anyway
- easily extendible

Now of course there is a difference, these things are exposed to the outside while the map itself is an implementation detail, but I think the idea of removing the RelationKind also kinda puts in question why we keep `reverseRelations` generic. We could instead also just create a map called something like `OverriddenByMap`, call the function something related too and be very specific what we are doing here. It kinda comes down to if we expect this to be expanded in the future or not.

Now let me finish this by saying I am not against this change, I just have some mixed feelings and am myself unsure, these are just some thoughts that popped into my mind.

https://github.com/llvm/llvm-project/pull/163024