Fixed bug #77243 (Weekdays are calculated incorrectly for negative years)

[derickr]

Ideally this should be done by reimporting timelib, but I haven't made a release of that yet, so we'll do it with a "hot fix".

[cmb69]

Do you have a reference on the weekdays of the dates in the tests? Respective tests with ext/calendar (since there is no year 0 there, I decreased the years by one):

var_dump(
    jddayofweek(gregoriantojd(1, 1, -1501), CAL_DOW_SHORT),
    jddayofweek(gregoriantojd(1, 1, -2147483649), CAL_DOW_SHORT),
    jddayofweek(gregoriantojd(1, 27, -292277022658), CAL_DOW_SHORT),
);

gives:

string(3) "Fri"
string(3) "Tue"
string(3) "Mon"

[derickr]

@cmb69, you were right about jddayofweek(gregoriantojd(1, 1, -1501), CAL_DOW_SHORT), having to return Friday. There was indeed another bug in my code with multiples of 100 (negative).

The other two can't be checked with ext/calendar, as its algorithm states:

 * VALID RANGE
 *
 *     4714 B.C. to at least 10000 A.D.

[derickr]

And FWIW, the results now match with what https://www.fourmilab.ch/documents/calendar/ does.

[cmb69]

I think we're good to go. :)