The following example compiles with gcc and fails with clang. I believe clang is correct but can't prove it.

======= begin example =============
#include <ranges>

void foo() {
    std::ranges::iota_view iota(2, 10);
    iota.begin();
}
======== end example ==============

Clang complains that iota_view doesn't have a begin() member (which it does). I think the reason is that the constraint is evaluated before iota_view is fully defined. gcc evaluates lazily so it doesn't stumble on the problem.

The circular chain is:

view_interface<iota_view> is instantiated as a base class of iota_view. Clearly iota_view isn't defined at this stage, it's just a forward-declared name.

view_interface instantiates iterator_t, which is an alias to std::__detail::__range_iter_t.

__range_iter_t instantiates __ranges_begin.


__ranges_begin requires that its template parameter is a __member_begin<> (among other options, but this is the valid one here).

__member_begin requires that the type (iota_view) have a begin() function. But the type isn't defined yet.

I believe clang is correct, mostly because I believe concept evaluation should be eager and not lazy, not because I know it for a fact.

Apologies for posting a clang issue here, I wouldn't if gcc could get asan+coroutines working. The problem will hit gcc if it implements eager evaluation too.