mirror of
https://github.com/nlohmann/json
synced 2024-11-23 12:30:06 +00:00
parent
266399d8e5
commit
b8d05f72ea
@ -2284,11 +2284,7 @@ class basic_json
|
||||
{
|
||||
// 15 digits of precision allows round-trip IEEE 754
|
||||
// string->double->string
|
||||
using std::snprintf;
|
||||
const auto sz = static_cast<unsigned int>(snprintf(nullptr, 0, "%.15g", m_value.number_float));
|
||||
std::vector<typename string_t::value_type> buf(sz + 1);
|
||||
snprintf(&buf[0], buf.size(), "%.15g", m_value.number_float);
|
||||
o << buf.data();
|
||||
o << std::setprecision(15) << m_value.number_float;
|
||||
return;
|
||||
}
|
||||
|
||||
|
@ -2284,11 +2284,7 @@ class basic_json
|
||||
{
|
||||
// 15 digits of precision allows round-trip IEEE 754
|
||||
// string->double->string
|
||||
using std::snprintf;
|
||||
const auto sz = static_cast<unsigned int>(snprintf(nullptr, 0, "%.15g", m_value.number_float));
|
||||
std::vector<typename string_t::value_type> buf(sz + 1);
|
||||
snprintf(&buf[0], buf.size(), "%.15g", m_value.number_float);
|
||||
o << buf.data();
|
||||
o << std::setprecision(15) << m_value.number_float;
|
||||
return;
|
||||
}
|
||||
|
||||
|
Loading…
Reference in New Issue
Block a user