On Fri, Apr 12, 2019 at 10:02 AM JeanHeyd Meneide <phdofthehouse@gmail.com> wrote:
On Fri, Apr 12, 2019 at 6:45 AM Lyberta <lyberta@lyberta.net> wrote:
I guess at least teachability and clean structure. The guidance would
be: "stuff in std is old and unusable for text, stuff in std::text is
new and usable".

Having `std::text::text` is a bit of a weird class type (unless we give it a new name), and it's impossible to have `std::text` as a type and `std::text` as a namespace at the same time.

The sub-namespace isn't really necessary here because we are not in competition for certain names or algorithms, save for the 3 names I want to specifically name `std::text_decode`, `std::text_encode`, `std::text_transcode`, and similar because it internally implies other semantics and I do not want to steal the names `decode` and `encode` when those are much more broad terms.

Other names such as `std::utf8`, `std::utf32`, `std::utf16`, `std::wide_execution`, and `std::narrow_execution` are fairly specific to the text domain and I don't see them clashing.

Names such as `std::rope`, `std::text`, and `std::text_view` will speak for themselves. There are a few traits types that might be introduced to the standard, but as far as I can tell none of these will clash either.

`std::uni` is an OK namespace for the unicode properties.

Regarding earlier points on what the standard does provide: the standard needs to provide encodings for all the encoding types that are (currently) pushed out by the standard, and nothing more. This includes: std::utf8, std::utf16, std::utf32, std::wide_execution, and std::narrow_execution. The standard should not vend any other encodings, but the Encoding and Decoding interfaces should be standard -- much like Allocator -- that allows a user to swap in their own class type and object that replaces the use of an encoding in any interface / function standard templates provide. (Similar to char_traits, except not as useless.) This means users can employ whatever encoding or power they have under the hood and enjoy fast and correct text processing so long as they follow the required semantics.

Note that we cannot only ship utf8 as an encoding, because the standard already ships and acknowledges more than utf8 as one of the encoding for string literals. It would be highly dysfunctional to have utf16 string literals that the standard library itself cannot process in a reasonable manner.

It would be fine to do just that,  We'd would be signalling that utf16 is on the chopping block, to be deprecated some day, or at least ignored going forward.  Same way we ignore, or give lower priority to, other things in the language and library that we feel were a mistake and are not worth our time.

For example, no one ever stops a proposal by saying "but that doesn't support valarray", because no one cares about valarray.  Even though it is right there in std::.

Now, you could say some people care about uft16, and that would be a reason to continue to support it, but I think eventually (5-10 years) we will find that no one cares about utf16, and it is just cruft.

Of course, supporting it now might be better for consensus, but "because it exists" isn't reason enough.


--
Be seeing you,
Tony