Re: Decimal: Formal review begins.

From: Ruben Perez (rubenperez038_at_[hidden])
Date: 2025-01-19 13:04:36

Hi,

Some problems I've found:

The decimal32 doc page states that the minimum normalized value is
1e-95. The std::numeric_limits<>::min() function agrees with that.

However, fpclassify and similar functions don't seem to agree:

// Both of the above print "1"
std::cout << issubnormal(std::numeric_limits<boost::decimal::decimal32>::min())
<< std::endl;
std::cout << issubnormal(1.0e-95_df) << std::endl;

Printing very small values seem to do nothing:

// This prints an empty line
auto v = 1.0e-94_df;
std::cout << v << std::endl;

Calling to_chars with subnormal values fails with
std::errc::not_supported, regardless of the chars_format used:

// This prints: Error: generic:95
void f(decimal32 v)
{
    char buff[64]{};
    auto r = boost::decimal::to_chars(buff, buff + sizeof(buff), v);
    std::cout << "Error: " << std::make_error_code(r.ec) << std::endl;
}

int main()
{
    auto v = 1e-95_df;
    f(v);
}

Am I doing something stupid?

Regards,
Ruben.