Use C++20 concepts to inform code completion in templates

[HighCommander4]

In the following code example:

template <typename T>
concept Fooable = requires (T t) { t.foo(); };

class Waldo {};

template <typename T> requires Fooable<T>
void bar(T t) {
  t.^
}

it would make sense to offer foo as a completion proposal.

[sam-mccall]

Indeed. It looks like core concepts stuff (constraints and such) is still landing in clang, e.g. recent https://reviews.llvm.org/D50360. The implementation of this is going to depend on what the AST looks like.

If you're able to follow that closely enough, have a sense of how the AST could be structured to help enable such code completion, it might be worth prototyping and talking to the authors (looks like Saar Raz is implementing most with Richard Smith reviewing) while things are still in flux.

Otherwise it might be worth holding off on this for a bit until it's more stable, and working out how to produce sensible suggestions on top of it. My read is an AST walk and a few heuristics will give us method names with argument counts (and maybe types), nested types, and that might be about as much as we get.

[sam-mccall]

What the heck, here's a rough prototype (no tests, and some cases not covered: completion after :: doesn't work, and completion of nested templates doesn't work)

https://reviews.llvm.org/D73596

Have been trying it out with this snippet:

template <typename T> concept X = requires(T t) { t.xxx(); typename T::x_t; };
template <typename T> concept Y = requires(T t) { t.yyy(); };
template <typename A, typename B> concept W = requires(A a, B b) { { b.www() } noexcept; };
template <typename T> concept Z = requires(T t) { t.zzz(); requires W<int, T>; };
template <X T> requires Y<T> &&requires(T *t) { t->aaa(); }
void foo(T t) requires Z<T> || requires(T &t) { t.bbb(); t->bb(); } {
  t.
}

[HighCommander4]

What the heck, here's a rough prototype (no tests, and some cases not covered: completion after :: doesn't work, and completion of nested templates doesn't work)

https://reviews.llvm.org/D73596

I remember reviewing this code, but I guess it was in a re-upload as a new review request. Did it end up landing?

[HighCommander4]

Here's the newer review request: https://reviews.llvm.org/D73649, and it did in fact land.

Perhaps we should close this issue, and file follow-ups for additional cases we'd like to support?

[sam-mccall]

Agree, we have some basic support. I'm sure there's things to be improved but let's close this until we have a better idea what's high priority.