[Lldb-commits] [PATCH] D63240: [Core] Generalize ValueObject::IsRuntimeSupportValue

Jim Ingham via Phabricator via lldb-commits lldb-commits at lists.llvm.org
Wed Jun 26 16:01:31 PDT 2019

jingham added a comment.

In D63240#1559931 <https://reviews.llvm.org/D63240#1559931>, @xiaobai wrote:

> In D63240#1559913 <https://reviews.llvm.org/D63240#1559913>, @jingham wrote:
> > Shouldn't ValueObjectVariables figure out their runtime language from their defining frame, not their CompilerType?  For a ValueObject you get from an expression, you probably can't do that.  But we're always talking about hiding locals or args here - i.e. they are all ValueObjectVariables.  And it seems to me that in that case getting the RuntimeLanguage from the containing frame is much more useful than from the CompilerType.
> Does every variable have an associated frame? I imagine things like global variables wouldn't. I'm not sure if any language or implementation of a language has global variables as compiler/runtime helper variables, but I'm not comfortable making that assumption since one of my goals is more generalized language support.

To be more precise, "frame" is the wrong word to use, Variables have "scopes"...  All Variables have a scope, though not every scope is contained in a Function.

But just to back up a little bit.

We're getting into problems, it seems to me, because we're asking ValueObjects questions that should be asked of their containing scope.  For instance, if you are in a Function, and want to list the locals and arguments, you want to know what the rules for the Function are for which artificial variables should and should not be printed, not any of its value objects.

For instance, ObjC could decide that it wants to reserve the word "this" in methods of ObjC classes in ObjC++ files, and use it as a should-be-hidden artificial variable that will always store a pointer to C++ object for some evil purposes.  Asking the runtime language of this artificial "this" variable whether the word "this" is runtime whitelisted seems like the wrong thing to do.  After all, the Variable has a CompilerType that's clearly a C++ object, so you might reasonably expect its runtime language to be C++.  But then you would get the wrong answer.  You really need to ask the Function "what are your rules for hiding artificial variables".

Getting to the right actor to ask about "which artificial variables to show" was the reason for suggesting that the runtime language of a Variable should be that of its containing scope.  But then you get into the weird situation where a C++ Variable is claiming a runtime language of ObjC++, which seems a little odd.  Instead, it makes more sense to get the RuntimeLanguage of the current frame, and then iterate over the variables and ask the frame's runtime whether they should be shown or not.

I'm not sure what to do about global variables.  Reasoning by analogy, that's something that the language of the CompileUnit which contains the global variables should answer.



More information about the lldb-commits mailing list