Bug #79580 [Com]: date_create_from_format misses leap year

From: Date: Sun, 10 May 2020 18:14:42 +0000
Subject: Bug #79580 [Com]: date_create_from_format misses leap year
References: 1  Groups: php.bugs 
Request: Send a blank email to php-bugs+get-226980@lists.php.net to get a copy of this message
Edit report at https://bugs.php.net/bug.php?id=79580&edit=1

 ID:                 79580
 Comment by:         rowan dot collins at gmail dot com
 Reported by:        mcab at acm dot org
 Summary:            date_create_from_format misses leap year
 Status:             Verified
 Type:               Bug
 Package:            Date/time related
 Operating System:   Linux
 PHP Version:        7.2.30
 Block user comment: N
 Private report:     N

 New Comment:

Branch updated with an alternative implementation that makes mixing 'z' with any month or
day specifiers an error.


Previous Comments:
------------------------------------------------------------------------
[2020-05-10 17:21:29] rowan dot collins at gmail dot com

Test and tentative fix here: https://github.com/php/php-src/compare/PHP-7.3...IMSoP:bug-79580
(will actually need a change to timelib, but I'm not sure how to test it there).

This implements the approach in my last comment, but results in a change of behaviour for the
slightly odd case of specifying a day-of-year and then a month within the same time string:

echo DateTimeImmutable::createFromFormat("z m Y", '58 3
2019')->format('z: Y-m-d'), "\n";
# Previously "86: 2019-03-28"
# Now "117: 2019-04-28"

I'm not sure what to do about that, or even what the correct behaviour for that combination
should be.

------------------------------------------------------------------------
[2020-05-10 13:45:17] rowan dot collins at gmail dot com

It looks like the problem is that the 'z' specifier works by setting "month=1,
day=X" (in this case, "the 60th of January"), then normalising the date immediately.
But at that point, it doesn't know what year to use for the normalisation, so picks 1970 (which
wasn't a leap year).

https://heap.space/xref/php-src/ext/date/lib/parse_date.c?r=9c608bd1#25213

If you reverse the format, you get the expected result, because the correct year has already been
selected when the "normalise" call happens:

$TZ=timezone_open( "UTC" );
$refdate=date_create_from_format ( "Y z" ,"2020 60" ,$TZ );
echo date_format($refdate,"d/m/Y");

# 01/03/2020


A possible fix would be to delay the call to timelib_do_normalize until after the whole input has
been parsed, including the year specifier.

------------------------------------------------------------------------
[2020-05-10 13:41:20] cmb@php.net

Confirmed: <https://3v4l.org/Drfk6>.

------------------------------------------------------------------------
[2020-05-10 13:22:00] mcab at acm dot org

Description:
------------
date_create_from_format when format is 'z Y' is a day out when day number is after feb28.


Test script:
---------------
$TZ= timezone_open( "UTC" );
$refdate=date_create_from_format ( "z Y" ,"31 2020" ,$TZ );
echo date_format($refdate,"d/m/Y")."<br>";

$refdate=date_create_from_format ( "z Y" ,"60 2020" ,$TZ );
echo date_format($refdate,"d/m/Y")."<br>";

$refdate=date_create_from_format ( "z Y" ,"91 2020" ,$TZ );
echo date_format($refdate,"d/m/Y")."<br>";

$refdate=date_create_from_format ( "z Y" ,"121 2020" ,$TZ );
echo date_format($refdate,"d/m/Y")."<br>";

$refdate=date_create_from_format ( "z Y" ,"130 2020" ,$TZ );
echo date_format($refdate,"d/m/Y")."<br>";

/*  Output is 
01/02/2020
02/03/2020   <- This has missed the fact that 2020 is a leap year
02/04/2020   <- All future date are wrong
02/05/2020
11/05/2020
*/



------------------------------------------------------------------------



--
Edit this bug report at https://bugs.php.net/bug.php?id=79580&edit=1


Thread (8 messages)

« previous php.bugs (#226980) next »