[Lldb-commits] [lldb] [LLDB] Add AST node classes, functions, etc. for Data Inspection Lang… (PR #95738)

[via lldb-commits]

cmtice wrote:

> Even though you've removed the smart pointer checking function (the most obvious example of language-specific behavior), I'm still worried about whether the overall design of this code will allow you to avoid it.
> 
I believe I have fixed all the code to avoid any need to to smart pointer checking -- I went back and looked very closely at what the current frame variable implementation does when handling smart pointers, and tried to do the same thing, which appears to have worked.

> Looking at the `IdentifierInfo` class, I see that it can be constructed with only a type (the value is optional). I think this is going to be very hard to make work, since a type does not know if its a smart pointer (only a value does). Have you already figured out a way to make that work?
> 
The case where it is constructed only with a type is meant to be when handling LLDB convenience variables (the dynamic variables names "$..."); I'm not sure if those can be smart pointers?

> IndentifierInfo isn't the only suspicious place though. All of the places that are doing something with bitfield offsets are very suspicious to me, as that's something that the Values should already know.
> 
> I haven't seen the parser, but I wouldn't be surprised if this has something to do with the vexing parse issues of C languages, where one needs to know if something is a type just to parse the expression correctly. Now, if that's the case, and if really need to know if something is a smart pointer (i.e. whether it is dereferencable) without having access to a value of that type, then we may need to have a bigger discussion about what to do about that.

I would be happy to share the parser code with you and walk/talk you through it, if that would help?


https://github.com/llvm/llvm-project/pull/95738