Proposed resolution:

This wording is relative to N4981.

1. Modify [facet.num.put.virtuals] as indicated:

   Table 114 – Floating-point conversions [tab:facet.num.put.fp]

   State	`stdio` equivalent
   `floatfield == ios_base::fixed` && !uppercase	`%f`
   `floatfield == ios_base::fixed`	`%F`
   floatfield == ios_base::scientific && !uppercase	`%e`
   floatfield == ios_base::scientific	`%E`
   floatfield == (ios_base::fixed | ios_base::scientific)` && !uppercase	`%a`
   floatfield == (ios_base::fixed | ios_base::scientific)	`%A`
   `!uppercase`	`%g`
   otherwise	`%G`

[ 2024-05-08; Reflector poll ]

Set priority to 3 after reflector poll. Send to LEWG.

In Table 114 – Floating-point conversions [tab:facet.num.put.fp] we specify that a floating-point value should be printed as if by `%f` when (flags & floatfield) == fixed. This ignores whether `uppercase` is also set in `flags`, meaning there is no way to use the conversion specifier `%F` that was added to `printf` in C99.

That's fine for finite values, because 1.23 in fixed format has no exponent character and no hex digits that would need to use uppercase. But `%f` and `%F` are not equivalent for non-finite values, because `%F` prints `"NAN"` and `"INF"` (or `"INFINITY"`). It seems there is no way to print `"NAN"` or `"INF"` using `std::num_put`.

Libstdc++ and MSVC print `"inf"` for the following code, but libc++ prints `"INF"` which I think is non-conforming:

    std::cout << std::uppercase << std::fixed << std::numeric_limits<double>::infinity();

The libc++ behaviour seems more useful and less surprising.