[Lldb-commits] [lldb] [LLDB][NFC] Added the interface DWARFExpression::Delegate to break dependencies and reduce lldb-server size (PR #131645)

[Pavel Labath via lldb-commits]

labath wrote:

> This type of patch also slightly worries me in that it could make it harder to unify the LLDB dwarf data structures with the LLVM ones

That is something I considered, but I think this might actually help with that. At this point the LLDB and llvm dwarf classes are so different, that the only way I see to unify them is to do it piecemeal, and this kind of decoupling helps with that. So, if you wanted to replace LLDB DWARFUnit with llvm's, you could do that without touching DWARFExpression. It's true lldb's DWARFUnit now inherits from some DWARFExpression thingy, but that's easy to replace with composition. I alluded to that in the earlier comments, but in the end I didn't insist on it. Maybe I should have. If you think it's better to do that now, I'd be certainly fine with that.

This doesn't help with replacing LLDB's DWARFExpression with the llvm version, but I don't think it makes it worse either. I see a couple of possible ways forwards there:
- introduce a similar abstraction to llvm DWARFExpression. I think that would make sense since dwarf expressions are used in contexts (eh_frame) where there is no DWARFUnit to speak of.
- replace (non-unwinding) uses of DWARF Expression with a higher level abstraction. It's use in Function and Variable classes is breaking abstractions, and the PDB plugin has to jump through hoops to create DWARF expressions for these objects. That would leave DWARFExpression only used for unwinding in the core code, where it would be easier to replace this with llvm::DWARFExpression (and at that point, it may not matter that it depends on llvm::DWARFUnit, since this class -- unlike the lldb counterpart -- does not depend on the whole world)

https://github.com/llvm/llvm-project/pull/131645