Re: [text] [review] Text formal review result

From: Zach Laine (whatwasthataddress_at_[hidden])
Date: 2020-06-27 01:29:51

On Fri, Jun 26, 2020, 5:22 PM Emil Dotchevski via Boost <
boost_at_[hidden]> wrote:

> On Fri, Jun 26, 2020 at 11:03 AM Vinnie Falco via Boost <
> boost_at_[hidden]> wrote:
> >
> > On Fri, Jun 26, 2020 at 10:43 AM Zach Laine via Boost
> > <boost_at_[hidden]> wrote:
> > > I want to standardize something like the interfaces there for C++23
> >
> > Why are you targeting a particular release date? Is your development
> > strategy going to be different because you want to "get something in
> > to C++23" versus "I want to standardize something after it has
> > achieved sufficient field experience and maturity?"
>
> I was going to ask the same question. Nothing should be standardized that
> isn't already the defacto standard. It isn't likely that between what's
> left of 2020 and 2023 this library or any other will become the defacto
> standard. I know this fails on deaf ears but it needs to be said.
>

Unicode is an actual standard.  The std::ranges library is too.  Not de
facto standards, actual standards.

The Unicode layer of the proposed Boost.Text packages the Unicode
algorithms into a std::ranges-style API.  There is very little invention
going on here.

The text layer is a different matter.  That *is* an invention.  It is also
controversial, as the review demonstrated.  That is precisely why I was not
going to attempt to standardize any of it in the C++23 timeframe, even when
my plan was still to get it into Boost right away.  It needs time for
evaluation and revision -- perhaps several years.

I understand the concerns you have both raised, and I share them.  I'm one
of the few people in LEWG consistently asking "How many users does this
have?" and "Why didn't this go into Boost first?".  If I thought that
either the Unicode standard or the std::ranges work were not fully baked, I
might have reservations about the Unicode layer of Boost.Text.  The only
thing that might need baking is the exact subset of Unicode algorithms that
one needs to do useful work.  The review has pointed out some weak spots,
but if you look at the review, you'll notice that they're all additive.  I
expect all future changes to be additive, too, because there are not
variations of the Unicode algorithms.  There's a fixed set of them.  If we
were to standardize the existing Unicode layer today, we'd be able to use
it for a very, very long time.

You cannot say that with any certainty about most proposals to the standard
library, but then most such proposals are not based on an existing
international standard.  That's why the Unicode layer (and *only* the
Unicode layer) should be standardized in the C++23 timeframe.

Zach