cpp_int fails to construct correctly from minimum negative

[Jan Bouwer <JBouwer@…>]

A cpp_int fails to construct correctly with the minimum negative value of a (platform dependent?) signed integral type.

Example:

template<typename T, class I>
void test_boost_multiprecision()
{
    T negMin = std::numeric_limits<T>::min();
    I test1(negMin), test2( negMin + 1 );
    --test2;
    
    assert( test1 == test2 ); // This fails on my platform for T = long
    assert( test1 != 0 );
    assert( test1 == negMin );
    assert( test2 != 0 );
    assert( test2 == negMin ); // This fails on my platform for T = long
}

test_boost_multiprecision<int64_t, boost::multiprecision::checked_cpp_int>(); // Fails on my platform

// The following hold for my platform: x86_64-apple-darwin12.4.0,
// using: Apple LLVM version 4.2 (clang-425.0.28) (based on LLVM 3.2svn)
// with: -std=c++11 -stdlib=libc++
// static_assert(std::numeric_limits<long long>::min() == std::numeric_limits<int64_t>::min(), "Ok");
// static_assert(std::numeric_limits<long>::min() == std::numeric_limits<int64_t>::min(), "Ok");

[Jan Bouwer <JBouwer@…>]

A demonstration of the error.

[Jan Bouwer <JBouwer@…>]

A naive patch suggestion.

[John Maddock]

Confirmed, investigating.

[John Maddock]

(In [84897]) Add some typecasts to ensure conversion from negative values proceeds correctly. Fix masking of values during construction so overflow wraps around in unchecked fixed precision integers. Add tests for above changes. Refs #8711.

[Jan Bouwer <JBouwer@…>]

This indeed fixes the issue I experienced (and others along the same line).

Thanks - your work (here and elsewhere) are much appreciated.