format-test failing on ppc64el

[trldp]

When running unit tests on Ubuntu compiled with gcc on the architecture ppc64el, the test format-test fails with the following error:

./test/format-test.cc:1493: Failure
Expected equality of these values:
  buffer
    Which is: "0x1.a7ae147ae147ae147ae147ae148p+1"
  fmt::format("{:a}", 3.31l)
    Which is: "0x0.0000000000001a7ae147ae147bp+54"

The error was detected on Ubuntu 22.04, but seems to occur on all other currently supported versions. Full logs can be found at https://launchpadlibrarian.net/657130777/buildlog_ubuntu-focal-ppc64el.fmtlib_9.1.0-1~git202303230019+5~ubuntu20.04.1_BUILDING.txt.gz.

[phprus]

@trldp

Please, build and run next code:

#include <iostream>
#include <limits>
#include <typeinfo>

int main()
{
    auto v = 3.31l;
    using T = decltype(v);
    std::cout << typeid(T).name() << std::endl;   
    std::cout << std::numeric_limits<T>::is_iec559 << std::endl;
    std::cout << "size:" << sizeof(T) << std::endl;
    std::cout << std::endl;
    std::cout << std::numeric_limits<T>::is_specialized << std::endl;
    std::cout << std::numeric_limits<T>::is_signed << std::endl;
    std::cout << std::numeric_limits<T>::is_integer << std::endl;
    std::cout << std::numeric_limits<T>::is_exact << std::endl;
    std::cout << std::numeric_limits<T>::has_infinity << std::endl;
    std::cout << std::numeric_limits<T>::has_quiet_NaN << std::endl;
    std::cout << std::numeric_limits<T>::has_signaling_NaN << std::endl;
    std::cout << std::numeric_limits<T>::has_denorm << std::endl;
    std::cout << std::numeric_limits<T>::has_denorm_loss << std::endl;
    std::cout << std::numeric_limits<T>::round_style << std::endl;
    std::cout << std::numeric_limits<T>::is_iec559 << std::endl;
    std::cout << std::numeric_limits<T>::is_bounded << std::endl;
    std::cout << std::numeric_limits<T>::is_modulo << std::endl;
    std::cout << std::numeric_limits<T>::digits << std::endl;
    std::cout << std::numeric_limits<T>::digits10 << std::endl;
    std::cout << std::numeric_limits<T>::max_digits10 << std::endl;
    std::cout << std::numeric_limits<T>::radix << std::endl;
    std::cout << std::numeric_limits<T>::min_exponent << std::endl;
    std::cout << std::numeric_limits<T>::min_exponent10 << std::endl;
    std::cout << std::numeric_limits<T>::max_exponent << std::endl;
    std::cout << std::numeric_limits<T>::max_exponent10 << std::endl;
    std::cout << std::numeric_limits<T>::traps << std::endl;
    std::cout << std::numeric_limits<T>::tinyness_before << std::endl;
    return 0;
}

and command:

c++filt -t <first_line_of_output>

[trldp]

This is the output of your code above

g
1
size:16

1
1
0
0
1
1
1
1
0
1
1
1
0
106
31
33
2
-968
-291
1024
308
0
0

and this is de output of the command c++filt -t g

__float128

[phprus]

@trldp
Please compile fmtlib with the flag

CXXFLAGS="-mabi=ieeelongdouble"

and run tests.

An error due to limited support for the double-double format in fmtlib.

I'll make a workaround for tests.

[phprus]

Please test this branch: https://github.com/phprus/fmt/tree/hexfloat-ppc64el

[trldp]

With the flag -mabi=ieeelongdouble, I do get the following error in format-test:

./include/fmt/format.h:3439: assertion failed: value is negativeterminate called without an active exception.

The branch above does works like a charm (without -mabi=ieeelongdouble).

[phprus]

With the flag -mabi=ieeelongdouble, I do get the following error in format-test:

I think this is a different bug not related with this issue.

The branch above does works like a charm

I created PR #3366.
But this is not a very fair fix, since fmt does not support double-double and converts it to double.