[PATCH] D67901: [clangd] Improve semantic highlighting in dependent contexts (fixes #154)
Haojian Wu via Phabricator via cfe-commits
cfe-commits at lists.llvm.org
Fri Sep 27 05:20:12 PDT 2019
hokein added inline comments.
================
Comment at: clang-tools-extra/clangd/SemanticHighlighting.cpp:181
+ addToken(E->getMemberLoc(), E->getQualifier()
+ ? HighlightingKind::StaticField
+ : HighlightingKind::Field);
----------------
nridge wrote:
> hokein wrote:
> > ilya-biryukov wrote:
> > > nridge wrote:
> > > > hokein wrote:
> > > > > This could be member functions, a case is like
> > > > >
> > > > > ```
> > > > > template<class T>
> > > > > class Foo {
> > > > > public:
> > > > > void foo() {
> > > > > this->foo();
> > > > > }
> > > > > };
> > > > > ```
> > > > Thanks for the example.
> > > >
> > > > Do you have a suggestion for how to discriminate this case? To me, it would seem logical to do it based on syntax (highlight as a member function if the expression forms the function name of a function call expression). That would require navigating from the expression to its parent node. Is there a way to do that?
> > > There is no way to do this in C++.
> > > Even if the name is followed by a pair of parenthese, this could either be a field with overloaded `operator()` (e.g. a `std::function<void()> field`) or a function with the same name.
> > >
> > > It's much better to pick a separate highlighting kind for dependent names, this follows the actual semantics of C++.
> > +1, I think we should just highlight them as a dependent type.
> Of course, any attempt to disambiguate between a member function and a field would be heuristic only. I figured that would be better than nothing. But if you prefer using a separate highlighting for dependent names that resolve to a function or a variable, we could do that.
>
> (Hokein, I assume you don't actually mean using the dependent *type* highlighting. Using a type highlighting for something we know is not a type, but rather a function or variable, would be rather confusing.)
> But if you prefer using a separate highlighting for dependent names that resolve to a function or a variable, we could do that.
I'd suggest we'd better follow this. The heuristic could fail in some cases. For these cases, the wrong highlighting'd confuse users too.
> (Hokein, I assume you don't actually mean using the dependent *type* highlighting. Using a type highlighting for something we know is not a type, but rather a function or variable, would be rather confusing.)
sorry for the confusion, I meant for anything that could not resolve to a specific declaration (e.g. `CXXDependentScopeMemberExpr`, `UnresolvedLookupExpr`), we use the new `dependent` highlighting kind.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D67901/new/
https://reviews.llvm.org/D67901
More information about the cfe-commits
mailing list