Consider the following spreadsheet formulas:

=MONTHS("1599-12-22";"1599-12-23";1)
=MONTHS("1599-12-31";"1600-01-01";1)
=YEARS("1599-12-22";"1599-12-23";1)
=YEARS("1599-12-31";"1600-01-01";1)

They return 0, 1, 0, and 1, respectively, as one would expect.

Now consider these:

=MONTHS("1499-12-22";"1499-12-23";1)
=MONTHS("1499-12-31";"1500-01-01";1)
=YEARS("1499-12-22";"1499-12-23";1)
=YEARS("1499-12-31";"1500-01-01";1)

They return 1, 0, 1, 0, respectively - which is completely unexpected.
The problem is, that the former set of formulas is after 1582-10-15, which is when Gregorian calendar was introduced, and the latter set is before that date; the dates sent to the functions in the latter set get converted to serial dates as Julian dates, but MONTHS and YEARS use proleptic Gregorian to determine the result when calculating with calendar months/years.

In OOo 3.2 and LO 3.3.0.4, the latter set resulted in #VALUE!; while LO 3.4.0.1 already produces the values described above. Not a regression though, rather implementation error?