[PATCH] D117846: [ASTMatchers] Add `isConstinit` matcher

[Evgeny Shulgin via Phabricator via cfe-commits]

Izaron marked an inline comment as done.
Izaron added inline comments.


================
Comment at: clang/include/clang/ASTMatchers/ASTMatchers.h:5223
+///   matches the declaration of foo and bar.
+AST_MATCHER(VarDecl, isConstinit) { return Node.hasAttr<ConstInitAttr>(); }
+
----------------
aaron.ballman wrote:
> This isn't quite correct -- there are two forms of `ConstInitAttr`:
> ```
> constinit int i = 0; // We want to match this one
> [[clang::require_constant_initialization]] int j = 0; // We don't want to match this one
> ```
> We typically want AST matchers to model the AST and this one does not... but, I think it is okay. `constinit` was based on the implementation experience we got with `[[clang::require_constant_initialization]]` and so we continue to model it as an attribute rather than a bit on the AST node. Having this matcher helps to distinguish the cases (otherwise we could just use the `hasAttr()` matcher instead).
> 
> I think the logic here should be:
> ```
> if (const auto *CIA = Node.getAttr<ConstInitAttr>())
>   return CIA->isConstinit();
> return false;
> ```
> and you should add a test case + documentation to show that we don't match the attribute form.
Thank you! Very valuable information. It opened my eyes why is `constinit` unexpectedly parsed like an attribute. Another plus of the new matcher - users won't search for `constinit` incorrectly as I did before the fix.

I added an explicit example to docs showing that the clang attribute won't work.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D117846/new/

https://reviews.llvm.org/D117846