https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118016
--- Comment #2 from geza.herman at gmail dot com --- I disagree how the standard is interpreted. If I write "1.1", it is a double literal. Its value should be the closest double to "1.1". It is fine, if later, the compiler treats this value as long double, but it should still use the numerical value of "1.1", not "1.1L". If I write "1.1/3", then this **expression** (the division) can be evaluated using long double. But the input values should be "1.1" as double and "3" as double. Then the values can be converted to long double, and then the division operation can be evaluated using long doubles. Just like how it would be done runtime, if the CPU evaluates "a/b", where both "a" and "b" are doubles with the values of "1.1" and "3" respectively. (I understand that "1.1" is also an expression. But its value shouldn't be "1.1L", but "1.1".) As far as I see, this interpretation also obeys the standard. I'm not a native english speaker, but this sentence "evaluate all operations and constants to the range and precision of the long double type." could easily mean my interpretation: - "1.1" is a constant, with the type of double, and with the value of the closest double value to "1.1" - then, this value is "evaluated" using the range and precision of the long double type. This "evaluation" doesn't change the value, just for further calculations, this value will use the long double type. What is the benefit of GCC's current interpretation? I only see drawbacks of the current approach. The fact that "9000000000000001.499999" will be used as "9000000000000002" instead of "9000000000000001" should be a sign that the current approach has problems, in my opinion. I tried to find some discussion regarding this, but didn't find anything. Was there any?
