[clangd] Include cleaner doesn't use forward declarations

[torshepherd]

Include cleaner is great, but often in a header file I don't need to include full definitions. Adding a config setting "UseForwardDeclarations" that uses forward declarations instead of includes in header files, for instance, would help cut down transitive includes.

Currently there isn't really a way to do this without scrutinizing the header file to see if the symbol is used as a full definition or just a declaration

[llvmbot]

@llvm/issue-subscribers-clang-include-cleaner

Author: Tor Shepherd (torshepherd)

Include cleaner is great, but often in a header file I don't need to include full definitions. Adding a config setting "UseForwardDeclarations" that uses forward declarations instead of includes in header files, for instance, would help cut down transitive includes.

Currently there isn't really a way to do this without scrutinizing the header file to see if the symbol is used as a full definition or just a declaration

[jpothier-figma]

Just adding a +1 for this. My company is currently looking at how we might implement this ourselves in include-cleaner. There is a lot of great prior art in IWYU that could be helpful in implementing this, though that tool makes some opinionated design choices about what "should be" forward-declared (e.g., function return types are not consider forward-declarable), and include-cleaner would have to choose a stance on that question as well.

[EugeneZelenko]

I noticed this too while using Clang Include Cleaner (via Clang-Tidy) on my work project.

[sethrj]

This has been a longstanding issue.

Perhaps the most egregious example is that it suggests you remove #include <typeinfo> from code that uses the typeid operator:

#include <typeinfo>

class BaseClass;

char const* typeinfo(BaseClass* bc) { return typeid(bc).name(); }