redo date.rb (most) parts in native #5076

kares · 2018-03-02T16:54:27Z

still using JODA but all of the JI stuff in date.rb is gone behind ext classes, motivation :

for a while its 'hard' to create Date instances in AR-JDBC
some near era conversion had a nasty bug in all 9K which constantly pops up in AR-JDBC
some isolated AR benchmarks showed its slower than MRI
there's going to be some nice JI bonuses - toJava working as with RubyTime etc.

some parts were already in native (formatting) so the base setup was already there.

NOTE: in MRI pretty much all of the date.rb is in native - so there's still space for improvement.
e.g. _parse would make sense to end up in .java maybe someone will attempt it afterwards.

p.s. left behind one minor suite failure (argument validation hour: 24+1.to_r/2)
while its simple the bigger picture (of MRI argument handling) feels a bit weird.
some parts handle the overflow some not, smt Float do not work while Rational is ok - clunky.
would eventually circle back as I feel like that might end up as a MRI report/feature request.

needs rebase (waiting for tests) but would like to get this in so I can spent time testing + adding JI stuff.

mkristian · 2018-03-05T07:32:28Z

❤️
might ping you guys on irc on day to talk about repackage an/or third-party jars and its risks for jruby-9.2

kares · 2018-03-08T16:54:06Z

already tested with AR-JDBC - all seems well (and working backwards compatible even on master).
likely going to merge this soon, so I can finish up some JI stuff that will require a set of second eyes.

…ight

…on new

…impl

… this

…oring

... this is simply distated by MRI's test suite we do not handle "big-integer" but that shouldn't matter for real-world

…pec-ed

esp. since numeric handling through code-base feels a bit inconsistent

…form

NOTE: MRI does not care for `Time.local` re-definition with `date.to_time`

... driven by one of the last failures -> strftime-ing pico-secs precision

esp. early years has been a consitent surprise around AR-JDBC testing

... somehow naively assumed its re-calculating with the DateTime ctor

resolves GH-5110 ... (another follow-up on date.rb refactoring #5076)

kares added this to the JRuby 9.2.0.0 milestone Mar 2, 2018

kares requested review from headius and enebo March 2, 2018 16:54

kares added 25 commits March 12, 2018 14:35

first stab at moving more (internal) date.rb pieces into native Java

fa4f0dd

backport Date::Infinity from MRI + get .rb + native loading (order) r…

89c555d

…ight

fix regression of not doing year shift properly + some more date tests

8803406

move more of date.rb (internals) into native + cleanup un-used .rb code

32a425d

refactor to bootstrap Date/DateTime classes from native part, first

2632530

dry-out (internal) instantiation - despite initialize not being used …

34e53ad

…on new

missed removal of old (internal) new! alias + some commented .rb piece

f787378

use Numeric's (call-site-ised) <=> as fallback comparison (after coerce)

97a64ad

re-implement Date#inspect in native (Java)

e5f58f0

move/port internal Date helpers - similar to how MRI has them in C

5edf1ca

[test] a few (edge) cases - for the date.rb -> native .java rewrite

3e66fb6

re-implement Date.civil (new) as native code - similar to custom .rb …

1953706

…impl

don't take a completely different path if day < 0 ... joda can handle…

dd0ff50

… this

make sure proper chronology used from civil constructor + some logging

b430eb1

[test] more 'fast' (and edge) checks to go with date.rb native refact…

4366ff7

…oring

refactor and public-ize getEnvTimeZone helper - to be re-used in Date

5f5bda2

implement Date.today in native + privatize less for DateTime sub-class

7333289

implement native Date#to_datetime

340a546

re-implement DateTime#to_date + preliminary DateTime#now

a960d12

make sure DateTime.now respects ENV[TZ] (like it did with date.rb)

f3f1b3c

fix class-index -> we only ever get Integer in 2.4

22cb217

re-implement DateTime.civil (new alias) in native Java

437a99a

cleanup - remove date.rb pieces implemented natively (civil methods)

addc170

redo _valid_time? internal helper as native

4e15ce0

already moved DateTime.now to native (see RubyDateTime.java)

4de04df

kares added 23 commits March 12, 2018 14:36

get Date/DateTime dup-ing and clone (back) working

98602fd

one more special case to handle on DateTime.civil - with hour == 24

dc5780b

[refactor] keep sg for RubyDate as long - not worth the complication

c2355d3

[test] more tests for a few cases (+ partialy moved from MRI excludes)

b36ff8f

[test] include proper description for test:mri exclude

acee58f

formatter tweaks - can now accept a RubyNumeric sub-millis argument

db36ca2

avoid array copy-ing on format getters + set enc based on pattern string

a136a44

should use long/long precision for Date's sub-millis + dry-out jd_to_ajd

0e0afcd

... this is simply distated by MRI's test suite we do not handle "big-integer" but that shouldn't matter for real-world

[test] more Date tests; move partial (passing) bits from (excluded) MRI

4cce264

Date.valid_jd? needs to return false for nil arg - this is actually s…

b96d727

…pec-ed

fix to_date conversion from DateTime - properly switch to 00:00 UTC

48665e0

simplify (and correct) Time#to_datetime conversion using JODA

d50da5b

[test] untag passing; tag fractional DateTime specs - quite low priority

0f59222

esp. since numeric handling through code-base feels a bit inconsistent

properly inverse DateTime#to_time and Time#to_datetime conversion

471826a

need to account for JODA's julian year rulez (there's no year 0)

7e8759d

account for send not routing properly to over-loads (+ 3 args case)

bdcf332

[refactor] no need to reduce rational - can be sure its in canonical …

8967e16

…form

at last, also moving Date#to_time to native (for better MRI compat)

a87a2bc

NOTE: MRI does not care for `Time.local` re-definition with `date.to_time`

use long precision every-where - not that much users would need this

50e3f4c

... driven by one of the last failures -> strftime-ing pico-secs precision

[test] some more (JRuby only) Date/DateTime tests

d817251

esp. early years has been a consitent surprise around AR-JDBC testing

[refactor] there's already a Java implementation of (MRI's) zone_to_diff

051694d

[refactor] (and simplify) strptime parsing - BigInt only when necessary

3615ebb

work-around JODA considering 'missing' (reform) dates as invalid

2d16dd2

... somehow naively assumed its re-calculating with the DateTime ctor

kares force-pushed the date_native2 branch from c5fa846 to 2d16dd2 Compare March 12, 2018 13:36

kares merged commit cf89614 into master Mar 12, 2018

kares deleted the date_native2 branch March 12, 2018 14:12

kares mentioned this pull request Mar 19, 2018

DateTime.jd returns incorrect times with 9.2.0.0-SNAPSHOT #5092

Closed

philr mentioned this pull request Mar 25, 2018

DateTime.jd doesn't set sec_fraction with 9.2.0.0-SNAPSHOT #5110

Closed

kares added a commit that referenced this pull request Mar 26, 2018

[fix] make sure we account for fractional part on jd/ordinal/civil etc.

25cc7a0

resolves GH-5110 ... (another follow-up on date.rb refactoring #5076)

headius mentioned this pull request Mar 28, 2018

JRuby 9.2 Projects #5119

Closed

15 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

redo date.rb (most) parts in native #5076

redo date.rb (most) parts in native #5076

kares commented Mar 2, 2018

mkristian commented Mar 5, 2018

kares commented Mar 8, 2018

redo date.rb (most) parts in native #5076

redo date.rb (most) parts in native #5076

Conversation

kares commented Mar 2, 2018

mkristian commented Mar 5, 2018

kares commented Mar 8, 2018