Re: Should boost::core::string_view be made public?

From: Andrey Semashev (andrey.semashev_at_[hidden])
Date: 2024-02-08 16:11:17

On 2/8/24 18:54, Peter Dimov via Boost wrote:
> Andrey Semashev wrote:
>> On 2/8/24 18:14, Peter Dimov via Boost wrote:
>>> Andrey Semashev wrote:
>>>> Which they are not, as boost::core::string_view doesn't support
>> char_traits.
>>>> So it looks like a non-starter to me.
>>>
>>> Do you really need or support traits other than std::char_traits<Ch>?
>>> Why and where?
>>
>> It's not that I need it. It's that std::basic_string(_view) are defined this way (for
>> better or worse), and for our string_view to properly emulate and
>> inperoperate with std::string_view, it should have it as well.
> 
> That's not true at all.
> 
>> You're not winning anything by not supporting it either as you're using
>> std::char_traits internally anyway.
> 
> core::string_view is intended to be used in this manner:
> 
>   void my_api_function( core::string_view sv );
> 
> which then allows the user to pass std::string, std::string_view,
> boost::string_view.
> 
> It's not intended to be used in this manner
> 
>   template<class Ch>
>   void my_api_function( core::basic_string_view<Ch> sv );
> 
> because then conversions don't work.
> 
> There is no benefit in also supporting
> 
>   template<class Ch, class Tr>
>   void my_api_function( core::basic_string_view<Ch, Tr> sv );
> 
> because it doesn't work either.

This implies that the API only works with one character type, which is
not always the case, as shown in Boost.Log.

Template interface parity is useful in cases like this:

https://github.com/boostorg/container/blob/6e697d796897b32b471b4f0740dcaa03d8ee57cc/include/boost/container/string.hpp#L685-L691

https://github.com/boostorg/container/blob/6e697d796897b32b471b4f0740dcaa03d8ee57cc/include/boost/container/string.hpp#L2314-L2316

It also allows to interoperate with strings with custom char traits just
in case if someone actually uses them.