[PATCH] D133029: [Sema] Allow to diagnose the references to std::vector<T> with incomplete T

[Shafik Yaghmour via Phabricator via cfe-commits]

shafik added a comment.

In D133029#3763688 <https://reviews.llvm.org/D133029#3763688>, @ilya-biryukov wrote:

> In D133029#3763120 <https://reviews.llvm.org/D133029#3763120>, @shafik wrote:
>
>> Do you have an idea of how common this might be?
>
> I don't have the numbers yet, but this broke the builds in C++20 mode inside the core libraries we are using after libc++ change <https://github.com/llvm/llvm-project/commit/98d3d5b5da66e3cf7807c23a0294280bb796466b> that makes vector `constexpr`.
> I suspect every medium-to-large codebase will have a few instances.
> One common pattern that I have seen goes something like:
>
>   // foo.h
>   struct Inner;
>   class Outer {
>     std::vector<Inner> elements;
>   private:
>     size_t size() { return elements.size(); }
>     void some_other_method();
>   };
>   
>   struct Inner {
>     Outer* outer;
>   };
>   
>   // foo.cpp
>   Outer::some_other_method() { /* uses members that would actually fail with incomplete type */ }

This old cfe-dev thread might be relevant: https://lists.llvm.org/pipermail/cfe-dev/2018-June/058338.html

It appears that folks take advantage of this working a lot. It sounds like your saying w/ constexpr std::vector this would be ill-formed in a constant expression context now. Which won't effect most of the existing code I think?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D133029/new/

https://reviews.llvm.org/D133029