Test failures of `to_static_string` with floating point types when using libstdc++-15 / C++26

[Flamefire]

On Ubuntu 25.04 the to_static_string tests fail for floating point types, likely because the underlying libstdc++ is that of GCC 15, as I see the same failures for the used Clang versions.

See https://github.com/boostorg/static_string/actions/runs/18156408010/job/51686048189

libs/static_string/test/static_string.cpp(3955): test 'str == std_res' ('0.100000' == '0.1') failed in function 'void boost::static_strings::testToStaticString()'
libs/static_string/test/static_string.cpp(3956): test 'str == std_res' ('0.000000' == '1e-07') failed in function 'void boost::static_strings::testToStaticString()'
libs/static_string/test/static_string.cpp(3957): test 'str == std_res' ('-0.000000' == '-1e-07') failed in function 'void boost::static_strings::testToStaticString()'
libs/static_string/test/static_string.cpp(3958): test 'str == std_res' ('-0.100000' == '-0.1') failed in function 'void boost::static_strings::testToStaticString()'
libs/static_string/test/static_string.cpp(3959): test 'str == std_res' ('1234567890.000100' == '1234567890.0001') failed in function 'void boost::static_strings::testToStaticString()'
libs/static_string/test/static_string.cpp(3960): test 'str == std_res' ('1.123457' == '1.123456789012345') failed in function 'void boost::static_strings::testToStaticString()'
libs/static_string/test/static_string.cpp(3961): test 'str == std_res' ('-1234567890.123400' == '-1234567890.1234') failed in function 'void boost::static_strings::testToStaticString()'
libs/static_string/test/static_string.cpp(3962): test 'str == std_res' ('-1.123457' == '-1.123456789012345') failed in function 'void boost::static_strings::testToStaticString()'
libs/static_string/test/static_string.cpp(3968): test 'str == std_res' ('0.000000' == '3.3621031431120935063e-4932') failed in function 'void boost::static_strings::testToStaticString()'
libs/static_string/test/static_string.cpp(3969): test 'str == std_res' ('0.000000' == '1.1754944e-38') failed in function 'void boost::static_strings::testToStaticString()'

Note that I changed the tests to use BOOST_TEST_EQ to see the actual values on failure.

'-1.123457' == '-1.123456789012345' seems most worrying as the rounding is wrong: The value is cut off

[gennaroprota]

@vinniefalco: Is it a requirement that the result be the same as std::to_string()?

[Flamefire]

Shouldn't it be at least exact according to rounding rules? -1.123456789012345 roundtrips with the stdlib but causes a truncation with static_string. I'd expect it at least to round to `-1.123458'

I can't find documentation of the behavior of to_static_string. Given the resemblance to std::to_string I'd intuitively expect the same behavior

[vinniefalco]

to_static_string should produce results identical as std::to_string although if there is a problem with that design choice then it needs discussion (maybe on the list but we aren't there yet)

[gennaroprota]

@Flamefire: I don't see how the standard result can be something like "0.1", given that std::to_string( double ) must behave as if it used std::sprintf() with a "%f" specifier, and that has a default precision of 6. In fact, this gives "0.100000".

[Flamefire]

Good question, seems for c++2c this changes: https://godbolt.org/z/7K84j5sEW

Double checking https://en.cppreference.com/w/cpp/string/basic_string/to_string for that I found

Converts a numeric value to a string as if by std::format("{}", value). (since C++26)

And

__cpp_lib_to_string 202306L (C++26) Redefining std::to_string in terms of std::format

Following that leads to https://en.cppreference.com/w/cpp/utility/format/spec.html

Decimal format. Produces the output as if by calling std::to_chars(first, last, value).

https://en.cppreference.com/w/cpp/utility/to_chars.html

value is converted [...] in the default ("C") locale. [...] shortest representation: the string representation consists of the smallest number of characters such that there is at least one digit before the radix point (if present) and parsing the representation using the corresponding std::from_chars function recovers value exactly. If there are several such representations, one with the smallest difference to value is chosen, resolving any remaining ties using rounding according to std::round_to_nearest

So 0.1 is correct for C++26

Not sure how hard this would be, you could of course just adjust the tests to check against snprintf("%f" and state that clearly in the documentation.

But 0.000000 for to_static_string(1e-7) seems odd as 1e-7 is possible. Not sure how hard that would be to implement.
Maybe in C++26 std::to_chars could simply be used? Looks trivial to do and likely faster

[vinniefalco]

C++26 introduced a change that causes different output for the same code?

[Flamefire]

C++26 introduced a change that causes different output for the same code?

Yes, it redefined std::to_string to be more consistent with the "new" std::format to output the shortest sequence.
I don't see a big issue of outputting 0.1 instead of 0.100000000000 in general although of course it is possible that some might be affected.

[gennaroprota]

The documentation of the to_static_[w]string() overloads just says "Converts value to a static_[w]string", so we have the option to either make it return the same result as to_[w]string(), which means we need two different implementations, one for pre-C++26 and one for C++26 or later, or stick with one implementation and possibly document how we do the conversion. @vinniefalco: Which option do you prefer?

[Flamefire]

C++17 has std::to_chars and C++11 Boost.Charconv can be used prior to that as a drop-in. Maybe just always use (either of) that?

[vinniefalco]

I don't know what the best solution is, and I think we should strive to satisfy users first and foremost. What do they want?

[gennaroprota]

Fixed with 2cc22bf.