Hi,
Yes, I've come to the same conclusion (i.e. the numeric value in "+3 days"
or "-2 days" being treated as time zone). The 8.5 coreutils I tested on
is in a different time zone so it is possible it doesn't exhibit the same
behavior because of that, and as such I can't actually tell whether the
behavior is actually resolved in coreutils 8.5.
I confirm that if I type,
mrjb@THE-D-MRJB:~$ date --utc -d "2014-03-11 12:34:56 0 -1 day" +"%Y-%m-%d
%H:%M:%S"
(i.e. using "0" as time zone), this gives the expected output,
2014-03-10 12:34:56
Best,
Marc
On Wed, 2 Apr 2014, Eric Blake wrote:
reopen 17161
thanks
On 04/02/2014 02:17 AM, Marc R.J. Brevoort wrote:
Hello,
As a heads-up, the explanation that this is due to time zone+locale
doesn't make sense:
mrjb@THE-D-MRJB:~$ date --utc -d "2014-03-11 12:34:56 -1 day" +"%Y-%m-%d
%H:%M:%S"
2014-03-12 13:34:56
mrjb@THE-D-MRJB:~$ date --utc -d "2014-03-11 12:34:56 -2 days"
+"%Y-%m-%d %H:%M:%S"
2014-03-12 14:34:56
mrjb@THE-D-MRJB:~$ date --utc -d "2014-03-11 12:34:56 -3 days"
+"%Y-%m-%d %H:%M:%S"
2014-03-12 15:34:56
The more days I subtract, the more hours are added. If this were a time
zone issue, the amount of hours added would be constant but the
resulting date would not.
Thanks for persisting. I was able to reproduce using coreutils.git, and
I can explain what's happening. Date is parsing this as:
"(2014-03-11 12:34:56 -3) (days)"
that is, -3 is being treated as the timezone, since it occurs next to
the hour specification; then treating "days" as "+1 day" or "+24 hours".
Watch what happens when you reorder the line to force a different parse:
$ src/date --utc -d "-3 days 2014-03-11 12:34:56" +"%Y-%m-%d %H:%M:%S"
2014-03-08 12:34:56
It does indeed seem awkward that date is not parsing this as:
"(2014-03-11 12:34:56) (-3 days)"
but fixing that will require tweaks to the getdate.y parser in gnulib.
I've reopened the bug to track the issue with awkward precedence in the
parser.
--
Eric Blake eblake redhat com +1-919-301-3266
Libvirt virtualization library http://libvirt.org
