Numeric segments were parsed into a saturating u64, so any two numbers
above u64::MAX compared equal. Compare the digits instead, by length
and then digit by digit, as systemd does.
When only one version had a digit at the current position, both sides
were parsed as numbers and the letter side read as 0. "0" then compared
equal to "a" and the result depended on what followed. Per the
specification, the side with digits is newer.
The '~' check ran after the end-of-string check, so "1~rc1" compared as
newer than "1". Per the UAPI version format specification, '~' must be
handled first so a pre-release sorts older than the release it
precedes. An empty string is kept as the oldest version, matching
systemd's strverscmp_improved.
alex
Benjamin Leggett