08:24 <aolko> hey there, do you guys have a metaprogramming crutch for `packages` construct?

11:31 <headius[m]> Like for importing?

17:18 <headius[m]> aolko What do you mean?

17:57 <headius[m]> kares: I might step away from the lazyfication of JI binding...the branch I have there already reduces overhead and cleaned up a lot of logic that was spread all over the place

17:57 <headius[m]> I'll merge in what I have but would appreciate a quick review...might just be easier to look at the new MethodGatherer class that I pulled out from the "State" class that was there before...all the logic is in there, should be easier to start making more of it lazy or caching better

18:30 <headius[m]> I documented how .jruby.java_opts files would work, if we include this support: https://twitter.com/headius/status/1163518349848039429

19:30 <kares[m]> sounds good, altough I might have missed the point for optimizing JI binding this far ...

19:31 <kares[m]> actually hoped to do pretty much the opposite -> bind all impls of virtual methods (that blocked us from delivering toString like inspect for Java types)

19:45 <headius[m]> kares: can you elaborate on that?

19:46 <headius[m]> my main goal here was to try to defer as much of the additional method processing until the methods are actually called

19:46 <headius[m]> so basically set up the proxy class as fast as possible without chewing on so many methods that we won't actually call

19:46 <headius[m]> JI boot time has been an issue, which I think is why you moved a bunch of JI enhancements out of Ruby, right?

19:47 <headius[m]> I also wanted to clean this logic up and I think it's much nicer in the PR, even if it doesn't fully make the heavy processing lazy

19:47 <headius[m]> in any case further lazy work can build on this, now that it's all sorted out into its own class

19:55 <kares[m]> JRuby does little JI on most we do

19:56 <kares[m]> and doing JI on JRuby's own types has other issues

19:57 <kares[m]> so its best to not do any JI there ...

19:58 <kares[m]> JI provided exts are lazy to load

19:59 <kares[m]> I think we only load a few, namely Map due org.jruby.util.SystemPropertiesMap

19:59 <kares[m]> that one is also unnecessary and could be avoided

20:00 <kares[m]> not 100% sure but we could reach zero JI on jruby boot

20:01 <headius[m]> I'm getting some numbers on whether this helps startup of JI stuff

20:01 <headius[m]> the overarching goal though is reducing JI binding costs to what you actually use

20:02 <headius[m]> we process thousands upon thousands of methods evenif you only call one

20:05 <headius[m]> yeah it does help

20:06 <headius[m]> for --disable-gems --dev -e 1 it's probably in the neighborhood of noise, but I got startup as low as 1.232 versus 1.268

20:06 <headius[m]> for -e creating a JFrame total run goes from 3.3 at best to 3.15

20:06 <headius[m]> binding more classes will add to that

20:07 <kares[m]> okay so its for reducing JI cost specifically for libraries

20:08 <headius[m]> yeah that was original goal...I didn't get as far as I wanted but there's a lot of code cleanup and allocation reduction plus a small gain

20:08 <headius[m]> I would like to do more to reduce this cost but need to get back to 9.2.9 critical path stuff

20:09 <headius[m]> I'll put together a quick report on a few startup cases

20:09 <headius[m]> I thikn this also moves us closer to being able to cache on disk a list of all methods for each class, so we just bind them super fast

20:10 <headius[m]> like on a per-JDK basis

20:10 <headius[m]> general idea is binding all names we would bind, but pushing most processing of Method[] lists to later

20:12 <headius[m]> kares: been seeing this in builds lately btw

20:16 <kares[m]> bleh, guess Ruby conventions do no translate to Java ... will clean those out

20:18 <kares[m]> had a clash of not using the proper block variable when there was nesting - so I started naming the unused ones, well Ruby style

20:26 <headius[m]> you can use $

20:26 <headius[m]> "unused" might be better though

20:26 <headius[m]> at least I think you can use $

22:14 <headius[m]> kares: oh btw I think I'm also seeing a lot of deprecation warnings from your builtin changes

22:14 <headius[m]> [WARNING] /Users/headius/projects/jruby/core/src/main/java/org/jruby/RubyClass.java:[630,37] [deprecation] retrieveCache(RubyClass) in CachingCallSite has been deprecated

22:14 <headius[m]> etc

22:16 <headius[m]> I made a couple more tweaks to the JI binding PR and will merge if there's no other issues