This prevents loss of precision that occurs with double-precision floats if timestamps become very large. Timestamps are already large if they contain a UNIX time value (requires 60 bits; double has 53 bit resolution).
Both are identical so far, this is just an infrastructure commit.