FYI gcc warns of 'const qualifier ignored on return type'

[willwray]

https://wandbox.org/permlink/EjDDKW72cTOKB3et
^^^ cut n paste the span snippet to wandbox with recent gcc gives vvv

4x warning: type qualifiers ignored on function return type on cbegin() and cend()
(and a couple of signed/unsigned comparison warning in the test code)

I managed to quieten the warnings by

• adding a const_pointer typedef with const added to the type
• updating const_iterator typedef from const pointer to const_pointer
  (or, just update const_iterator but I prefer to also have a const_pointer typedef).

    using const_pointer  = _Tp const*;     // added
    using const_iterator = const_pointer;  // updated

[mclow]

Thanks for the report - we're going to be looking at bugs in <span> in Rapperswil next month. I'll make sure that this gets discussed (and probably updated in the standard).

[mclow]

And I'll make sure that those warnings are squashed before I land this in libc++

[willwray]

Thanks for posting these snippets - I'd studied the earlier array_view snippet too.

The same bug was lying latent in my code - without warning - so it's useful that gcc flushed it out here. Clang doesn't warn (with the same flags). Coincidentally, on cpplang slack #general today arthur-odwyer found this same const pointer bug in some other multidimensional array code. Maybe there's a common ancestor. Can't remember if I'd copied or concocted-it-up.

It does sound like a storm is brewing over span's cbegin return:
Again on slack today, in #future_standard, a >100 thread was sparked by tvaneerd's "irregular span" paper, which itself references 'a giant reflector discussion', with caseycarter noting: I think it's an abomination that cbegin(span) and span.cbegin() have different meaning