add qualified_name_of

[KHicketts]

Adds qualified_name_of plus unit tests per p2996
(I have only tested the unit tests as a main, not by building them from the unit test folder)

[katzdm]

A few stylistic nits, hope you don't mind:

• I've been trying to keep line lengths capped at 80. We can break up lines like this by writing e.g.,:

    constexpr auto expectedQNofType = "outer_namespace::a_namespace::"
                                      "an_inner_namespace::ClassInsideSomeNamespaces";

• super nit: please include spaces between identifiers and == operators in static assertions
• Since std::meta:: is an associated namespace of all reflection values, we can usually omit the leading std::meta:: from metafunctions. So let's rewrite the static assertions like:

static_assert(expectedQNofDeclaration == qualified_name_of(^aString));

[katzdm]

Let's add a case to verify that qualified_name_of(^SomeTemplateClass<int>) doesn't include the trailing <int>, since P2996 specifically calls out:

For template instances, the name does not include the template argument list.

[KHicketts]

waaaaah! this is broken. I'll take a look...

[katzdm]

Rather than template-ize this function, we may be able to make DT& declType a NamedDecl *, as NamedDecl is an ancestor of ValueDecl, TemplateDecl, NamespaceDecl, and TypeDecl. Passing by reference as you've done here (i.e., NamedDecl &) would be just as good, but let's stick with passing by pointer since that's what's done throughout the rest of this file.

[KHicketts]

I could but then I have to cast everything which is a bit ugly IMHO. But then I guess it would match the rest of the clang codebase so...

[katzdm]

nits for consistency: space after if, no padding around parens, { on the same line as if.

if (makeCString(Result, Name, S.Context, Evaluator)) {

[katzdm]

No need to return true after llvm_unreachable - Depending on the build configuration, this will either halt execution, or probably get optimized away entirely by the compiler (that is...the compiler building the compiler 😵 ).

[katzdm]

Although this is what we do for name_of, it's not quite correct - P2996 specifies:

For all other reflections, an empty string_view is produced.

So that should include this case as well. Let's just stack this case on top of RK_const_value.

case ReflectionValue::RK_const_value:
case ReflectionValue::RK_data_member_spec: {
  if (makeCString(Result, "", S.Context, Evaluator))
    llvm_unreachable("failed to create empty string");
  return false;
}

(bonus points if you want to fix that in name_of, while you're in here 😄 )

[katzdm]

Is this cast necessary? I think getAsTemplateDecl() already returns a TemplateDecl *.

[katzdm]

This, surprisingly, actually isn't quite right - in addition to a NamespaceDecl, getReflectedNamespace() might be a NamespaceAliasDecl * or a TranslationUnitDecl * (i.e., for reflections like ^:: of the global scope). Neither of these has NamespaceDecl as an ancestor base class, so this cast<NamespaceDecl> will crash with an assert-failure.

I think what you'll want to do here is:

1. Get R.getReflectedNamespaced() as a Decl *.
2. Check the special case of whether it's a TranslationUnitDecl. If it is, return the empty string (the global namespace isn't a "declared entity", so P2996 says this should return empty).
3. Otherwise cast it to a NamedDecl * (which is a base class of both NamespaceDecl and NamespaceAliasDecl), and pass that to getQualifiedName.

We should probably add test cases for ^:: and for namespace aliases as well.

[katzdm]

Sorry, @KHicketts ! We ended up ditching this function from the paper :(