bin/111077: /bin/date -j -f "%b %Y" "Feb 2007" +%m returns 03 for Feb!!

[Giorgos Keramidas]

The following reply was made to PR bin/111077; it has been noted by GNATS.

From: Giorgos Keramidas <keramida at freebsd.org>
To: Ryan Pavely <paradox at nac.net>
Cc: bug-followup at freebsd.org
Subject: Re: bin/111077: /bin/date -j -f "%b %Y" "Feb 2007" +%m returns 03
	for Feb!!
Date: Mon, 2 Apr 2007 05:07:05 +0300

 On 2007-03-31 21:56, Ryan Pavely <paradox at nac.net> wrote:
 > >Number:         111077
 > >Category:       bin
 > >Synopsis:       /bin/date -j -f "%b %Y" "Feb 2007" +%m returns 03 for Feb!!
 [...]
 > >Environment:
 > FreeBSD #####.nac.net 4.6-STABLE FreeBSD 4.6-STABLE #0
 > FreeBSD #####.nac.net 4.8-RELEASE FreeBSD 4.8-RELEASE
 > FreeBSD #####.nac.net 4.9-RELEASE FreeBSD 4.9-RELEASE
 > FreeBSD #####.nac.net 5.2.1-RELEASE FreeBSD 5.2.1-RELEASE
 > FreeBSD #####.nac.net 5.3-RELEASE FreeBSD 5.3-RELEASE
 > FreeBSD #####.nac.net 5.4-RELEASE FreeBSD 5.4-RELEASE
 > FreeBSD #####.nac.net 5.5-RELEASE FreeBSD 5.5-RELEASE
 > FreeBSD #####.nac.net 6.0-RELEASE FreeBSD 6.0-RELEASE
 > FreeBSD #####.nac.net 6.1-RELEASE FreeBSD 6.1-RELEASE
 > FreeBSD #####.nac.net 6.2-RELEASE FreeBSD 6.2-RELEASE
 >
 > >Description:
 > Date input of "mmm yyyy" for Feb always returns 03.
 >
 > Flaw exists across all know bsd versions, intel, amd, 64bit, not, etc.
 
 Hmmm, not that we are in the 2nd of April, I cannot seem to reproduce
 this by running /bin/date on the command line, on a 6.2-RELEASE or a
 7.0-CURRENT system:
 
   $ date -j -f '%b %Y' 'Feb 2007' '+%m'
   02
   $
 
 As I found out, this happens when the current value of %d when /bin/date
 runs is larger than the number of days February has.  Then date(1) gets
 the value of %d from the current time, and this overflows from February
 into March (i.e. if you ran the tests on the same day that you submitted
 this bug report, the value of %d was 31, which is clearly not a valid %d
 value for any February).
 
 The date utility initializes a `struct tm' structure in setthetime()
 with the current value of date/time using localtime().  Then strptime()
 is called with the format specified and it parses *only* the parts which
 are explicitly mentioned in the format string "%b %Y".  The value of the
 current day-of-the-month should be left untouched by strptime(), and it
 is.  But I think that strptime() tries to rationalize an invalid value,
 when it finds one.  It's easy to reproduce this by setting a breakpoint
 in setthetime() while /bin/date runs, and tweaking the value of "day of
 the month" which is returned by localtime() in place:
 
 % keramida at kobe:/home/keramida/tmp/date$ gdb date
 % GNU gdb 6.1.1 [FreeBSD]
 % Copyright 2004 Free Software Foundation, Inc.
 % GDB is free software, covered by the GNU General Public License, and you are
 % welcome to change it and/or distribute copies of it under certain conditions.
 % Type "show copying" to see the conditions.
 % There is absolutely no warranty for GDB.  Type "show warranty" for details.
 % This GDB was configured as "i386-marcel-freebsd"...No symbol table is loaded.  Use the "file" command.
 %
 % (gdb) b setthetime
 % Breakpoint 1 at 0x8049213: file date.c, line 189.
 % (gdb) run -j -f "%b %Y" "Feb 2007" +%m
 % Starting program: /home/keramida/tmp/date/date -j -f "%b %Y" "Feb 2007" +%m
 %
 % Breakpoint 1, setthetime (fmt=0xbfbfe9a7 "%b %Y", \
 %     p=0xbfbfe9ad "Feb 2007", jflag=1, nflag=0) at date.c:189
 % 189             if (fmt != NULL) {
 % (gdb) n
 % 190                     lt = localtime(&tval);
 % (gdb)
 % 191                     t = strptime(p, fmt, lt);
 % (gdb) print *lt
 % $1 = {tm_sec = 12, tm_min = 54, tm_hour = 4, tm_mday = 2, tm_mon = 3,
 %       tm_year = 107, tm_wday = 1, tm_yday = 91, tm_isdst = 1,
 %       tm_gmtoff = 10800, tm_zone = 0x28184270 "EEST"}
 % (gdb) print lt->tm_mday =31
 % $2 = 31
 % (gdb) print *lt
 % $3 = {tm_sec = 12, tm_min = 54, tm_hour = 4, tm_mday = 31, tm_mon = 3,
 %       tm_year = 107, tm_wday = 1, tm_yday = 91, tm_isdst = 1,
 %       tm_gmtoff = 10800, tm_zone = 0x28184270 "EEST"}
 % (gdb) n
 % 192                     if (t == NULL) {
 % (gdb) c
 % Continuing.
 % 03
 %
 % Program exited normally.
 % (gdb)
 
 By asking strptime() to parse a struct tm which contains tm_mday set to
 31 with a format specifier of "%b" and an input string which says "Feb"
 we get "03" in the output (i.e. "March").
 
 I don't know if strptime() should return an error in this case, instead
 of trying to "overflow" into the next calendar month.  Both cases have,
 arguably, a logical explanation, but the overflow case is surprising.
 
 My own personal preference would be that strptime() returns an error in
 this case.  This will certainly cause a mild disturbance when users run
 /bin/date in the case you tried and get an error if the current day of
 the month is more than 28 (or 29 on leap years), but when /bin/date is
 used without an explicit %d value in both the format string of -j and
 the input data, it runs with an underspecified input value.  Giving
 input which is underspecified and getting an error is, IMHO, slightly
 better than getting surprising results.
 
 - Giorgos