Fix DateTimeValueCleaner for years in the far future

Since PHP 8.3.25, cal_days_in_month() refuses to operate outside of what
we previously called the “sane range” here, throwing a ValueError
instead of silently giving a slightly wrong result (28 instead of 29
days for February 92000000200). To avoid this error, bring the year into
a PHP-safe range using string operations (so we don’t introduce any
numerical imprecision / overflow issues).

Note that this *changes* the result for affected dates (year 2^31 or
higher, precision day or higher) in all PHP versions, but in a way that
I think is more correct. (And, really, I very much doubt anybody will
notice in practice. As of this writing, there are only 14 distinct time
values in Wikidata in excess of 2^31 years, and none of them are in
February or have a precision higher than year: https://w.wiki/Fwqd.)

Bug: T409269
Change-Id: Ibde22c1e8832b0c46301529eb613fe45dc6f4703

Author: lucaswerkmeister

repo/includes/Rdf/DateTimeValueCleaner.php

@@ -76,16 +76,24 @@ protected function cleanupGregorianValue( $dateValue, $precision ) {
 
 		// check if the date "looks safe". If not, we do deeper check
 		if ( !( $d <= 28 || ( $m != 2 && $d <= 30 ) ) ) {
+			// clamp the day to the last day in the month according to PHP
+			// (note: $safeYear is only used for cal_days_in_month())
+
 			// PHP source docs say PHP gregorian calendar can work down to 4714 BC
 			// Use float conversion here since we don't care about precision but don't want overflows.
 			if ( $minus && (float)$y >= 4714 ) {
 				$safeYear = -4713;
 			} else {
-				$safeYear = (int)$y * ( $minus ? -1 : 1 );
+				// since PHP 8.3.25, PHP also requires a year less than 2147483646 (2^31), see T409269;
+				// as the Gregorian calendar has a 400-year period and 10000 is a clean multiple of 400,
+				// shift overlong years into the safe range [100000, 199999] without losing integer precision
+				$safeYear = (int)(
+					strlen( $y ) >= 10 ? '1' . substr( $y, -5 ) : $y
+				) * (
+					$minus ? -1 : 1
+				);
 			}
 
-			// This will convert $y to int. If it's not within sane range,
-			// Feb 29 may be mangled, but this will be rare.
 			$max = cal_days_in_month( CAL_GREGORIAN, $m, $safeYear );
 			// We just put it as the last day in month, won't bother further
 			if ( $d > $max ) {

repo/tests/phpunit/includes/Rdf/DateTimeValueCleanerTest.php

@@ -58,6 +58,11 @@ public static function provideDates() {
 			[ '+02000000200-00-00T00:00:00Z', $greg, '2000000200-01-01T00:00:00Z' ],
 			[ '+92000000200-05-31T00:00:00Z', $greg, '92000000200-01-01T00:00:00Z', $year1m ],
 			[ '+92000000200-05-31T00:00:00Z', $greg, '92000000200-05-31T00:00:00Z' ],
+			[ '+92000000200-02-31T00:00:00Z', $greg, '92000000200-02-28T00:00:00Z' ],
+			[ '+92000000201-02-31T00:00:00Z', $greg, '92000000201-02-28T00:00:00Z' ],
+			[ '+92000000204-02-31T00:00:00Z', $greg, '92000000204-02-29T00:00:00Z' ],
+			[ '+92000000400-02-31T00:00:00Z', $greg, '92000000400-02-29T00:00:00Z' ],
+			[ '+92000000000-02-31T00:00:00Z', $greg, '92000000000-02-29T00:00:00Z' ],
 			[ '-02000000200-05-22T00:00:00Z', $greg, '-2000000200-05-22T00:00:00Z' ],
 			[ '-02000000200-02-31T00:00:00Z', $greg, '-2000000200-02-29T00:00:00Z' ],
 			[ '+00000000200-02-31T00:00:00Z', $greg, '0200-02-28T00:00:00Z' ],