9.2.0.0 regression in Date#to_time for BC dates resulting in Time off by 1 year

[jeremyevans]

Environment

Tested on:

jruby 9.2.0.0 (2.5.0) 2018-05-24 81156a8 OpenJDK 64-Bit Server VM 25.144-b01 on 1.8.0_144-b01 +jit [OpenBSD-x86_64]
jruby 9.2.0.0 (2.5.0) 2018-05-24 81156a8 Java HotSpot(TM) 64-Bit Server VM 25.131-b11 on 1.8.0_131-b11 +jit [mswin32-x86_64]

Expected Behavior

CRuby 2.3, CRuby 2.5, and JRuby 9.1 all return correct results:

$ ruby -r date -ve "d = Date.new(-1); p d; p d.to_time"
ruby 2.3.0p0 (2015-12-25 revision 53290) [x64-mingw32]
#<Date: -0001-01-01 ((1720693j,0s,0n),+0s,2299161j)>
-0001-01-01 00:00:00 -0800

$ ruby -r date -ve "d = Date.new(-1); p d; p d.to_time"
ruby 2.5.0p0 (2017-12-25 revision 61468) [x64-mingw32]
#<Date: -0001-01-01 ((1720693j,0s,0n),+0s,2299161j)>
-0001-01-01 00:00:00 -0800

$ jruby -r date -ve "d = Date.new(-1); p d; p d.to_time"
jruby 9.1.12.0 (2.3.3) 2017-06-15 33c6439 Java HotSpot(TM) 64-Bit Server VM 25.131-b11 on 1.8.0_131-b11 +jit [mswin32-x86_64]
#<Date: -0001-01-01 ((1720693j,0s,0n),+0s,2299161j)>
-0001-01-01 00:00:00 -0752

Actual Behavior

JRuby 9.2 has the Time value wrong:

$ jruby -r date -ve "d = Date.new(-1); p d; p d.to_time"
jruby 9.2.0.0 (2.5.0) 2018-05-24 81156a8 Java HotSpot(TM) 64-Bit Server VM 25.131-b11 on 1.8.0_131-b11 +jit [mswin32-x86_64]
#<Date: -0001-01-01 ((1720693j,0s,0n),+0s,2299161j)>
-0002-01-01 00:00:00 -0752

While not shown here, this also affects DateTime#to_time, which I'm guessing uses the same implementation.

Also note that both JRuby versions have a different timezone offset compared to MRI, but I'm not sure if that is a bug or just the result of using different timezone settings between Java and C.

I thought this may be related to #5191, but looking at the fix for that, I think it may be a different issue. However, I did not check the master branch to see if this has already been fixed.

[kares]

thanks - as you guessed ... should be the same cause as in #5191 (but seems it isn't)