[libcxx-commits] [PATCH] D65721: Make rotate a constexpr

[Louis Dionne via Phabricator via libcxx-commits]

ldionne added inline comments.


================
Comment at: libcxx/include/algorithm:2308
 template <class _ForwardIterator>
-_ForwardIterator
+_LIBCPP_CONSTEXPR_AFTER_CXX17 _ForwardIterator
 __rotate_left(_ForwardIterator __first, _ForwardIterator __last)
----------------
mclow.lists wrote:
> zoecarver wrote:
> > ldionne wrote:
> > > This function can be made constexpr in all standard modes that support generalized constexpr (so C++14 and above).
> > > 
> > > That holds for all helper functions below.
> > Was `std::rotate` updated as a paper or an issue? Or have we decided to extend constexpr support back to C++14 independent of the standard? 
> We don't have the freedom to add `constexpr` to things in the standard.
> We do have the freedom to add `noexcept`.
> 
> Odd, but that's the world we've got.
My comment was only about helper functions. Make the helper functions constexpr whenever possible, but only mark the actual algorithm `std::rotate` constexpr when the Standard says so.


================
Comment at: libcxx/test/std/algorithms/alg.modifying.operations/alg.rotate/rotate.pass.cpp:458
+#if TEST_STD_VER > 17
+    static_assert(test_constexr<forward_iterator<int*> >());
+    static_assert(test_constexr<bidirectional_iterator<int*> >());
----------------
zoecarver wrote:
> ldionne wrote:
> > It would be nice if we could run all the same tests for constexpr and non-constexpr. It's not clear to me why we can't just run `test<....>` in a constexpr context. As for `test1`, it allocates memory but it might be possible to change the test to avoid that?
> > 
> > Generally speaking, I'd like for us to develop a fairly mechanical way of testing things in a constexpr context, since we're going to have to do that for almost all of `<algorithm>`. I think it's important to have the same test coverage for constexpr and non-constexpr (whenever it makes sense, of course).
> I agree. What if we decided that we test constexpr by adding `return true` to the end of the test function and modifying the function to be a constexpr that returns a bool. This should work for almost all `<algorithm>` tests. 
Yes, something along those lines is what I was thinking about. Can you see if that works?


Repository:
  rCXX libc++

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D65721/new/

https://reviews.llvm.org/D65721