Extremely long literals are not handled well

[akrzemi1]

The following literal does not yield the correct decimal value

1000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000_DD

Not that I would ever need it. Just wanted to break the library. It looks like the literal operator taking unsigned long long shadows the one taking const char *. If I remove the former, the above literal starts to work.

[mborland]

Interesting, this seems to be a GCC problem. Clang gives the following:

/Users/mattborland/Documents/boost/libs/decimal/test/test_literals.cpp:138:22: error: integer literal is too large to be represented in any integer type
138 | const auto val = 1000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000_DD;
|

[ckormanyos]

Is that a bug or does the standard itself limit literals? I know that the standards for C and C++ have very liberal limits. But I have never wondered if literal numeric values are limited by the standard?

[ckormanyos]

For instance, this works on MSVC and it prints 1e+100, via the standard long double suffix .0L.

#include <boost/decimal/decimal64_t.hpp>
#include <boost/decimal/iostream.hpp>

#include <iostream>


auto main() -> int
{
  boost::decimal::decimal64_t googol { 1'0000000000'0000000000'0000000000'0000000000'0000000000'0000000000'0000000000'0000000000'0000000000'0000000000.0L };

  std::cout << googol << std::endl;
}

[akrzemi1]

From the C++ Standard: https://eel.is/c++draft/lex.ext#3.
the unsigned long long overload should be preferred to const char* one. But if the value doesn't fit into unsigned long long, then one cannot expect good results. Clang's error is better than GCC's silent change in meaning.

Still, the best course of action seems to be to remove the unsigned long long overload. The const char* one is a proper superset of the former.

[mborland]

Still, the best course of action seems to be to remove the unsigned long long overload. The const char* one is a proper superset of the former.

That is what I've done in the linked PR. Seems to be running fine across platforms and toolchains.