[libcxx-commits] [PATCH] D117449: [libc++] Fix common_iterator for output_iterators

[Arthur O'Dwyer via Phabricator via libcxx-commits]

Quuxplusone added a comment.

I assume this is still OK, and still necessary, even after D117400 <https://reviews.llvm.org/D117400> which I assume must have merge-conflicted with it a lot?



================
Comment at: libcxx/test/std/iterators/predef.iterators/iterators.common/plus_plus.pass.cpp:154-155
+
+  // Increment a common_iterator<output_iterator>: iter_value_t is not valid for output iterators,
+  // so this gets tricky when we define operator++(int).
+  {
----------------
ldionne wrote:
> CaseyCarter wrote:
> > Quuxplusone wrote:
> > > IIUC this comment is misleading. `iter_value_t` for an ordinary C++11 output iterator is just `ThatType::value_type`. It's only if the iterator doesn't define a `value_type`, or defines it as a non-object type, that `iter_value_t` might be ill-formed.
> > > https://godbolt.org/z/1Ybjv7o9G
> > > Our test iterator's `output_iterator<int*>::value_type` is `void`, so its `iter_value_t` is indeed ill-formed, but that's a special property of //our// `output_iterator`, not all output iterators in general.
> > > 
> > > I'd like to see test coverage for "output iterator with valid value_type" added to `iterator_traits.compile.pass.cpp` above, even if this means defining a little helper type a la the `OutIt` in my godbolt.
> > "need not be valid for output iterators" would be fine here.
> Agreed, the comment was wrong, I'll fix it. However:
> 
> > I'd like to see test coverage for "output iterator with valid `value_type`" added to `iterator_traits.compile.pass.cpp` above, even if this means defining a little helper type a la the `OutIt` in my godbolt.
> 
> The problem is that even though `iter_value_t<OutputIter>` *can* be non-`void`, `iter_value_t<common_iterator<OutputIter>>` will never be, because we are now properly constraining the `iterator_traits<common_iterator<...>>` specialization to only work on `input_iterator`s. So I can't add a test that checks that `iter_value_t<common_iterator<OutputIter>>` is going to be non-`void`.
> 
> Please LMK if you think I'm mistaken here.
I haven't investigated so I assume you're not mistaken. Then it just sounds like my request shifts from test coverage for `common_iterator<OutputIterWithNonVoidValueType>::iterator_category`, to test coverage that `common_iterator<OutputIterWithNonVoidValueType>` is SFINAE-friendly ill-formed. I believe we have some prior art, shaped roughly like
```
template<class T>
concept HasCommonIterator = requires { typename std::ranges::common_iterator<T>; };
static_assert(!HasCommonIterator<OutputIterWithNonVoidValueType>);
```


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D117449/new/

https://reviews.llvm.org/D117449