It's just a long-standing quirk in the parser. It parses the numeric part as an unsigned 64-bit number, then applies the sign afterwards. And the result can be

It's just a long-standing quirk in the parser. It parses the numeric part as an unsigned 64-bit number, then applies the sign afterwards. And the result can be approximately stuffed into a floating point value. The behavior for integer fields is different; compilation will fail with an out of range error.