On Tue, Apr 30, 2024 at 12:41:56PM +0200, Corentin Jabot via SG16 wrote:
> On Tue, Apr 30, 2024 at 12:45 AM Tom Honermann <tom_at_[hidden]> wrote:
>
> > On 4/29/24 4:11 PM, Peter Dimov via SG16 wrote:
> > > Tom Honermann wrote:
> > >> I'm not entirely sure that cout << std::format("{}", u8"...") is
> > We'll have to see what Corentin proposes :)
> >
>
> Very rough draft https://isocpp.org/files/papers/D3258R0.pdf

Thanks for working on this!

I know it is an early rushed draft, but I have some concerns regarding
the wording changes of [format.arg]

You did not add char8_t, char16_t, and char32_t to the variant, so these
are in stored in the handle. However you add wording

  otherwise, if TD is a Unicode character type, initializes value with v

This requires storing the Unicode character types in the variant instead
of in the handle.

You add 6 (or 9) types to the variant. However all major implementations
use a tagged union instead of a variant. Based on

[format.args]/1

  An instance of basic_format_args provides access to formatting
  arguments. Implementations should optimize the representation of
  basic_format_args for a small number of formatting arguments.

  [Note 1: For example, by storing indices of type alternatives
  separately from values and packing the former. — end note]

all implementations store the tag in a bit field
libc++ 5 bits [1]
libstdc++ 5 bits [2]
MSVC STL 4 bits [3]

So these new elements don't fit for MSVC STL. Looking at the number of
extension types in libstdc++'s enum type greatly reduces the number of
future extensions possible. I'm not sure whether the change in the
number of variant alternates will be an ABI-issue for MSVC STL or
libstdc++. (libc++ has sufficient space at the moment.)

An alternative for implementations is storing these values in the handle
object. However this would be non-conforming. The difference is
observable when using basic_format_arg::visit(); elements stored in the
handle can't be visited.

Another small nit

  otherwise, if TD is a specialization of basic_string_view or
  basic_string and TD::value_type is char_type <ins>or a Unicode
  character type</ins>, initializes value with
  basic_string_view<<del>char_type</del><ins>TD<ins>>(v.data(), v.size());

This should be TD::value_type instead of TD.

[1] https://github.com/mordante/llvm-project/blob/main/libcxx/include/__format/format_arg.h#L46
[2] https://github.com/gcc-mirror/gcc/blob/master/libstdc%2B%2B-v3/include/std/format
[3] https://github.com/microsoft/STL/blob/main/stl/inc/format#L2001

Cheers,
Mark