Merge pull request #2748 from jamescowens/fix_fraction_class_addition…

…_overflow

util: Adjust Fraction class addition overload overflow tests

src/test/util_tests.cpp

@@ -1648,8 +1648,8 @@ BOOST_AUTO_TEST_CASE(util_Fraction_addition_overflow_2)
 
 BOOST_AUTO_TEST_CASE(util_Fraction_addition_overflow_3)
 {
-    Fraction lhs(-(std::numeric_limits<int64_t>::max() / 2 + 1), 1);
-    Fraction rhs(-(std::numeric_limits<int64_t>::max() / 2 + 1), 1);
+    Fraction lhs(std::numeric_limits<int64_t>::min() / 2 + 1, 1);
+    Fraction rhs(std::numeric_limits<int64_t>::min() / 2, 1);
 
     std::string err;
 
@@ -1664,8 +1664,8 @@ BOOST_AUTO_TEST_CASE(util_Fraction_addition_overflow_3)
 
 BOOST_AUTO_TEST_CASE(util_Fraction_addition_overflow_4)
 {
-    Fraction lhs(-(std::numeric_limits<int64_t>::max() / 2 + 1), 1);
-    Fraction rhs(-(std::numeric_limits<int64_t>::max() / 2 + 2), 1);
+    Fraction lhs(std::numeric_limits<int64_t>::min() / 2, 1);
+    Fraction rhs(std::numeric_limits<int64_t>::min() / 2, 1);
 
     std::string err;
 

src/util.h

@@ -684,7 +684,14 @@ class Fraction {
 
         if (a < 0 && b < 0) {
             // Remember b is negative here, so the difference below is GREATER than std::numeric_limits<int64_t>::min().
-            if (a >= std::numeric_limits<int64_t>::min() - b) {
+            //
+            // The reason for the + 1 below is that the first case handled in the addition overflow operator method above,
+            // the fraction addition with a common denominator, also specifies simplification of the resultant fraction.
+            // This will call std::gcd on the numerator and denominator, which then calls std::abs. If std::abs is called on
+            // std::numeric_limits<int64_t>::min(), and -D_GLIBCXX_ASSERTIONS is set, then the program will abort on a glibc
+            // assertion, because the abs will overflow. To prevent this, we increase (make less negative) by 1, which will
+            // ensure the call to std::abs will succeed at the extreme case.
+            if (a >= std::numeric_limits<int64_t>::min() + 1 - b) {
                 return a + b;
             } else {
                 throw std::overflow_error("fraction addition of a + b where a < 0 and b < 0 results in an overflow");