Incorrect values resulting from to_date on a Time object

[WToa]

We were upgrading our environments to match JRuby 9.2.0.0 and noticed that calling to_date on Time objects lead to different results compared to previous iterations of JRuby and MRI.

Environment

JRuby 9.2.0.0

Expected Behavior

JRuby 9.1.17.0

jruby-9.1.17.0 :006 > Date.today.eql?(Time.now.to_date)
 => true 
jruby-9.1.17.0 :007 > Time.now.to_date.inspect
 => "#<Date: 2018-08-09 ((2458340j,0s,0n),+0s,2299161j)>" 
jruby-9.1.17.0 :008 > Date.today.inspect
 => "#<Date: 2018-08-09 ((2458340j,0s,0n),+0s,2299161j)>" 

Ruby 2.3.3

2.3.3 :005 > Date.today.eql?(Time.now.to_date)
 => true 
2.3.3 :006 > Time.now.to_date.inspect
 => "#<Date: 2018-08-09 ((2458340j,0s,0n),+0s,2299161j)>" 
2.3.3 :007 > Date.today.inspect
 => "#<Date: 2018-08-09 ((2458340j,0s,0n),+0s,2299161j)>" 

Ruby 2.5.1

2.5.1 :004 > Date.today.eql?(Time.now.to_date)
 => true 
2.5.1 :005 > Time.now.to_date.inspect
 => "#<Date: 2018-08-09 ((2458340j,0s,0n),+0s,2299161j)>" 
2.5.1 :006 > Date.today.inspect
 => "#<Date: 2018-08-09 ((2458340j,0s,0n),+0s,2299161j)>" 

Actual Behavior

jruby-9.2.0.0 :002 > Date.today.eql?(Time.now.to_date)
 => false 
jruby-9.2.0.0 :003 > Time.now.to_date.inspect
 => "#<Date: 2018-08-09 ((2458340j,0s,0n),+0s,0j)>" 
jruby-9.2.0.0 :004 > Date.today.inspect
 => "#<Date: 2018-08-09 ((2458340j,0s,0n),+0s,2299161j)>" 

[WToa]

Not sure why this is, we looked around in the commits and found:

f30fdfe#diff-c549944ab214525639bedaeffbaac520R61

It seems that the constant ITALY is being replaced by 0 in this.

[dr-itz]

I run into date related problems on 9.2.0 recently. They were already fixed on master in my case and it looks like this one is ok too:

jruby-head :001 > require 'date'
 => true 
jruby-head :002 > Date.today.eql?(Time.now.to_date)
 => true 
jruby-head :003 > Time.now.to_date.inspect
 => "#<Date: 2018-08-10 ((2458341j,0s,0n),+0s,2299161j)>" 
jruby-head :004 > Date.today.inspect
 => "#<Date: 2018-08-10 ((2458341j,0s,0n),+0s,2299161j)>" 

for reference, it's not exactly the latest, but close enough:

> jruby -v
jruby 9.2.1.0-SNAPSHOT (2.5.0) 2018-07-30 13b2df5 Java HotSpot(TM) 64-Bit Server VM 25.162-b12 on 1.8.0_162-b12 +jit [darwin-x86_64]

So the upcoming 9.2.1 should be ok..

[headius]

@dr-itz Thanks for confirming it's fixed on master! We did do some rework of Date in 9.2. and some subsequent fixes in 9.2.1.

[nightsurge]

I am getting this same exact issue on 9.2.14.0 (among other newer versions I also tried)

Time.parse('0001-01-01').to_date Is returning Jan 3, 0001, instead of Jan 1, 0001.
In our app there are RPG/DB2 legacy tables that store 0001-01-01 as a default date. We are trying to check for that default value and it is now failing because of this bug. It appears when this string is parsed as a Date when creating the Ruby/ActiveRecord object, it must be using Time#to_date at some point, yes?

Thoughts? @dr-itz @headius @kares

[ahorek]

@nightsurge this is expected. If you call Time.parse('0001-01-01') Ruby doesn't know if you mean Gregorian or Julian calendar. There's a leap year's difference. It may be surprising, but this conversion is actually unsafe for years < 1583. I don't think it'll ever be fixed for compatibility reasons, but JRuby behaves the same way as CRuby.

you may consider using DateTime instead

Time.parse("0001-01-01").to_datetime
=> #<DateTime: 0001-01-01T00:00:00+01:00 ((1721423j,82800s,0n),+3600s,2299161j)>
Time.parse("0001-01-01").to_datetime.julian?
=> true

take a look at https://stevemorse.org/jcal/julian.html

[nightsurge]

@ahorek when did this behavior change? We recently updated from Rails 4.1 to Rails 5.2, JRuby 9.1.15 to 9.2.14, and activerecord-jdbc-adapter to the 52-stable version.

Prior to these updates, our database models would correctly translate the '0001-01-01' database value to the correct model attribute of a Date with Jan 1, 0001.

I'm saying that when we do an ActiveRecord lookup policy = Policy.find(1234) and look at policy.some_date we used to get the Date object of Jan 01, 0001, but now doing that same process it equates to Jan 03, 0001. I get that it is related to Gregorian calendars, but I am trying to figure out where this behavior stems from so I can determine if we need to monkey-patch something, or if we have to just update our Rails code to look for Jan 3, 0001 instead of Jan 1, 0001.

I'm not even sure if this is a JRuby issue or something in ActiveModel/ActiveRecord, or JDBC Adapter...

[ahorek]

could you open a new issue? it looks more like a bug somewhere in activerecord-jdbc-adapter.

it would be great if you could track where's the conversion being made, make a test case, or compare it with a different db adapter (on CRuby)

[nightsurge]

I was able to track the change to the adapter when it moved to 50-stable and newer! That's when the gregorian calendar was put in place. I didn't "fix" it by checking for dates and handling pre-post gregorian, but I did make the fix for my scenario seen here:

jruby/activerecord-jdbc-adapter#1084