Behaviour of Date

SAURAV_KOTHARI · 5 April 2016 14:21

Hello,

I am facing this situation where the how the date calculations are made
seem to not make sense.

Say the date is set to *31st March 2016*

d = Date.parse('31-03-2016')
// d = Thu, 31 Mar 2016

d + 1.month = Sat, 30 Apr 2016

d - 1.month = Mon, 29 Feb 2016

Is this how it is supposed to be evaluating? Shouldn't adding *1.month* to
date and subtracting it return the original date?

Regards,
Saurav Kothari

Bryce_Kerley · 5 April 2016 14:23

That’s what happens, but not what you’re doing. `Date#+` doesn’t mutate the date, it returns a new one.

···

On 5 Apr 2016, at 10:21, Saurav Kothari wrote:

    d = Date.parse('31-03-2016')
    // d = Thu, 31 Mar 2016

    d + 1.month = Sat, 30 Apr 2016

    d - 1.month = Mon, 29 Feb 2016

Is this how it is supposed to be evaluating? Shouldn't adding *1.month* to
date and subtracting it return the original date?

SAURAV_KOTHARI · 5 April 2016 14:27

Yes. Modifying the date is not my intention in this example.

But say it were. Say I were to to modify the current date to some date in
the future. A simple reversal of steps will not give me the original value
for certain dates.

Can something else be done to make such a calculation reversible?

Saurav Kothari

stomar · 5 April 2016 19:57

Considering the way #month behaves (which obviously is intentional),
that is indeed not possible for some dates.

For reference here a (working) code example:

  require "date"
  require "active_support"
  require "active_support/core_ext"

Date.parse("2016-03-31") + 1.month - 1.month # => Wed, 30 Mar 2016

  Date.parse("2016-03-31") - 1.month # => Mon, 29 Feb 2016
  Date.parse("2016-03-30") - 1.month # => Mon, 29 Feb 2016
  Date.parse("2016-03-29") - 1.month # => Mon, 29 Feb 2016
  Date.parse("2016-03-28") - 1.month # => Sun, 28 Feb 2016

Date.parse("2016-03-31") + 1.month # => Sat, 30 Apr 2016
Date.parse("2016-03-30") + 1.month # => Sat, 30 Apr 2016

Date.parse(2016-03-31) + 12.month # => Fri, 31 Mar 2017 (ok)

  d = Date.parse("2016-03-31")
  12.times { d = d + 1.month }
  d # => Tue, 28 Mar 2017 (!!!)

Regards,
Marcus

···

Am 05.04.2016 um 16:21 schrieb Saurav Kothari:

Is this how it is supposed to be evaluating? Shouldn't adding /1.month/
to date and subtracting it return the original date?

--
GitHub: stomar (Marcus Stollsteimer) · GitHub
PGP: 0x6B3A101A

Andrew_Vit · 7 April 2016 03:34

Ruby also has `<<` and `>>` methods on Date that do the same thing as
ActiveSupport's `date + 1.month`:

     mar31 = Date.new(2016, 3, 31)
     feb29 = mar31 << 1
     mar29 = feb29 >> 1

A date object only holds a particular day, there is no special notion
of anchoring it to the end of the month:

feb29 = Date.new(2016, 2, -1)

You can create a date like that, but it doesn't have memory of the
negative offset that was passed in, it just becomes the 29th day of the
month.

If you want to, you could create your own wrapper for Date which holds
this original value, something like this I just wrote for fun:

gist.github.com

https://gist.github.com/avit/96f07db8a2ed83c436aff0ed61466e3a

anchored_date.rb

require 'delegate'
require 'date'

class AnchoredDate < Delegator
  def initialize year, month=1, day=1
    case year
    when Fixnum
      @year = year
      @month = month
      @day = day

This file has been truncated. show original

Andrew Vit

···

On 2016-04-05 07:21, Saurav Kothari wrote:

Is this how it is supposed to be evaluating? Shouldn't adding /1.month/
to date and subtracting it return the original date?

Reid_Thompson1 · 5 April 2016 15:11

what certain dates?

irb(main):006:0> d = Date.parse('31-03-2016')
=> #<Date: 2016-03-31 ((2457479j,0s,0n),+0s,2299161j)>

irb(main):008:0> d = d + 30
=> #<Date: 2016-04-30 ((2457509j,0s,0n),+0s,2299161j)>

irb(main):009:0> d - 30
=> #<Date: 2016-03-31 ((2457479j,0s,0n),+0s,2299161j)>

···

On Tue, 2016-04-05 at 19:57 +0530, Saurav Kothari wrote:

Yes. Modifying the date is not my intention in this example.

But say it were. Say I were to to modify the current date to some date in the
future. A simple reversal of steps will not give me the original value for
certain dates.

Can something else be done to make such a calculation reversible?

Saurav Kothari

SAURAV_KOTHARI · 5 April 2016 15:34

Reid, the examples you have demonstrated are calculations where you add and
*substract* *days *which will certainly be reversible.
But in the event of adding/subtractin *month *(as in the first thread), you
can see that the operation is not reversible. Can that be made reversible?

We already use a system where we set a future date (+ 1.month) for further
actions and do not store the original date. We use -1.month to get the
original date whenever we need it but that fails for a few cases as shown.

We could migrate to adding/subtracting 30.days in the future but how do we
handle existing date?

Saurav Kothari

SAURAV_KOTHARI · 6 April 2016 05:16

Thank you @Martin for putting it like I was intending to. I put up a wrong
example in the original post.

@Marcus: yes this is indeed the quirk that has me puzzled. So that is just
how it works internally. Will have to explore other options (like +/-
30.days).

Thank you all.

Saurav Kothari

Martin_DeMello · 5 April 2016 18:47

I believe he is referring to the fact that Date("2016-03-31") + 1.month -
1.month = Date("2016-03-30"). I don't see what can be done about that,
since two days in March (30, 31), map to a single day in April (30), so the
mapping is clearly not reversible.

martin

···

On Tue, Apr 5, 2016 at 10:32 AM, <sto.mar@web.de> wrote:

Am 05.04.2016 um 17:34 schrieb Saurav Kothari:
> Reid, the examples you have demonstrated are calculations where you add
> and *substract* /days /which will certainly be reversible.
> But in the event of adding/subtractin /month /(as in the first thread),
> you can see that the operation is not reversible. Can that be made
> reversible?

Saurav, you did _not_ show in your first post that the operation is
not reversible, you added / subtracted 1.month from the same date
(2016-03-31), because the original date saved in d was not changed,
as Bryce already pointed out.

Note also that Integer#month is not part of Ruby core or stdlib.
So without knowing what gems you required it's not easy to check
your example:

$ irb
2.3.0 :001 > require "date"
=> true
2.3.0 :002 > d = Date.parse('31-03-2016')
=> #<Date: 2016-03-31 ((2457479j,0s,0n),+0s,2299161j)>
2.3.0 :003 > d = d + 1.month # NoMethodError !!!

(I assume that's what you meant to do.)

You are possibly using Rails?

Regards,
Marcus

--
GitHub: https://github.com/stomar/
PGP: 0x6B3A101A

Unsubscribe: <mailto:ruby-talk-request@ruby-lang.org?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-talk>

stomar · 5 April 2016 17:32

Saurav, you did _not_ show in your first post that the operation is
not reversible, you added / subtracted 1.month from the same date
(2016-03-31), because the original date saved in d was not changed,
as Bryce already pointed out.

Note also that Integer#month is not part of Ruby core or stdlib.
So without knowing what gems you required it's not easy to check
your example:

$ irb
2.3.0 :001 > require "date"
=> true
2.3.0 :002 > d = Date.parse('31-03-2016')
=> #<Date: 2016-03-31 ((2457479j,0s,0n),+0s,2299161j)>
2.3.0 :003 > d = d + 1.month # NoMethodError !!!

(I assume that's what you meant to do.)

You are possibly using Rails?

Regards,
Marcus

···

Am 05.04.2016 um 17:34 schrieb Saurav Kothari:

Reid, the examples you have demonstrated are calculations where you add
and *substract* /days /which will certainly be reversible.
But in the event of adding/subtractin /month /(as in the first thread),
you can see that the operation is not reversible. Can that be made
reversible?

--
GitHub: https://github.com/stomar/
PGP: 0x6B3A101A

Xavier_Noria · 6 April 2016 10:07

The problem of working with durations is that what a month is is
ill-defined in that sense. So the designer of the API needs to define
something that makes sense in edge cases.

For example, the intention of adding a month is generally speaking to
increment the month number in the date, or reset to 1 if 12 incrementing
the year.

But that leaves you with situations like adding a month to 31 March,
because 31 April does not exist. You need to define what to do there, and
the choice is to keep the "increment the month by one", rather than jumping
to May. For that to be possible, the natural choice is "same day or end of
the month".

And that implies that for some dates, adding a month gives the same date.
Therefore you cannot reverse that unless you store the original value.

You need to get the business rule right taking this ambiguity into account.
For example, if the app switches to 30.days then it has to be OK that 1
July yields 31 July, and that 31 January may yield 2 March.

The Active Support Core Extensions guide documents edge cases of these APIs.

···

--
Sent from Gmail Mobile

Murad_Yusufov · 6 April 2016 07:55

You can use "end_of_month" method for this particular case.
http://apidock.com/rails/DateAndTime/Calculations/end_of_month

···

On Wed, 6 Apr 2016 at 08:17, Saurav Kothari <sauravkothari2@gmail.com> wrote:

Thank you @Martin for putting it like I was intending to. I put up a wrong
example in the original post.

@Marcus: yes this is indeed the quirk that has me puzzled. So that is just
how it works internally. Will have to explore other options (like +/-
30.days).

Thank you all.

Saurav Kothari

Unsubscribe: <mailto:ruby-talk-request@ruby-lang.org?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-talk>

Topic		Replies	Views
How do I add x months to a date ruby-talk	9	169	7 December 2007
Calculate last day of month ruby-talk	19	151	16 July 2008
How to get 1 month in the future ruby-talk	7	123	12 January 2007
Calculating a future date ruby-talk	7	86	8 February 2007
Problem with date.succ ruby-talk	3	100	16 August 2006

Behaviour of Date

Related topics