On Thu, May 22, 2008 at 3:07 PM, Craig A. Berry <[EMAIL PROTECTED]> wrote:
> At 1:50 PM -0500 5/22/08, Steve Peters wrote:
>
>>
>> > Here's VMS:
>>>
>>> $ perl -MTestInit -"MTime::Piece" -"E" "say join '|',
>>> Time::Piece::_strptime('2001-2-29 12:34:56', '%Y-%m-%d %H:%M:%S');"
>>> Error parsing time at -e line 1.
>>>
>>> Here's Mac OS X:
>>>
>>> % ./perl -MTestInit -"MTime::Piece" -E "say join '|',
>>> Time::Piece::_strptime('2001-2-29 12:34:56', '%Y-%m-%d %H:%M:%S');"
>>> 56|34|12|1|2|101|4|59|0|0|0
>>>
>>> If I'm reading this right, Mac OS X (and probably other BSDs) is
>>> interpreting this as the first of March, which doesn't make much
>>> sense to me, but may have a rationale behind it. In any case,
>>> Time::Piece seems to depend on this non-portable behavior, and even,
>>> as John noted, to be specifically testing for it.
>>>
>>
>>Actually, there isn't any non-portable stuff happening regarding
>>dates. The _strptime routine is supposed to see 2001-2-29 and realize
>>that you really mean 2001-3-1.
>
> Why on earth would it think that? Is 2001-2-30 interpreted as 2001-3-2?
>
Actually, yes, it seems to.
[EMAIL PROTECTED]:~/perl-current$ ./perl -Ilib -"MTime::Piece" -"E" "say
join '|', Time::Piece::_strptime('2001-2-30 12:34:56', '%Y-%m-%d
%H:%M:%S');"
56|34|12|2|2|101|5|60|0|0|0
I get identical answers on Win32, Solaris, Cygwin and Linux.
>>For some reason, the code works pretty
>>much everywhere else, but HP-UX and VMS.
>
> Because it's not the same code. As far as I can tell, it's just
> calling the C library's strptime() and however that interprets
> 2001-2-29 determines what happens.* Since 2001-2-29 is an invalid
> date, returning an error seems to me like the right thing to do, and
> that's what VMS and apparently HP-UX do. But different implementors
> have had different ideas, and as I mentioned before, no standard I
> can find says anything about what conditions should be considered
> error conditions. For systems that guess that you meant something
> other than what you said, is there any documentation that describes
> the intention? Do we even know that interpreting 2001-2-29 as
> 2001-3-1 is intentional and not just an untrapped error?
>
> * The exception to that is that on Win32 the PowerDog strptime()
> implementation is used by Time::Piece, and this is the same
> implementation used by the BSDs.
>
It would seem to me that an error is correct, but, obviously, that's
not how they were implemented anywhere except on HP-UX, VMS, and Tru64
(see
http://www.nntp.perl.org/group/perl.daily-build.reports/2008/05/msg55675.html).
So, saying its an untrapped error seems unlikely. It just seems that
HP-UX, VMS, and Tru64 are the exception rather than the rule.
Steve Peters
[EMAIL PROTECTED]
