15:36 <enebo[m]> kares: rdubya I put out new point releases of arjdbc 50-60. These just end up dr-itz fixes. Someone asked for new 50.x about a week or so ago.

16:06 <headius[m]> enebo: just pushed a change to make evalType final and at least specs passed locally

16:07 <headius[m]> getting down there on a simple times benchmark now

16:10 <headius[m]> about half as long to run 100_000_000.times {} on graalvm EE and 10x on CE (which ends up faster than EE somehow)

16:10 <headius[m]> compared to 9.2.9 with indy yields turned on (they were off by default)

16:11 <headius[m]> Java 8 comparison is better than 2x faster on my branch

16:12 <headius[m]> if I don't enable indy yields on 9.2.9 then it's more like 4.5x

16:25 <enebo[m]> headius: sounds really good

16:50 <enebo[m]> huiming: https://github.com/jruby/jruby/wiki/OperationDiscourse

16:50 <enebo[m]> It is a start. Once I restart that work I will update those getting started bits.

16:58 <headius[m]> oh nice

17:26 <headius[m]> seems to have passed with final evalType

17:27 <headius[m]> so that eliminates a thread-local set on every block entry

17:27 <headius[m]> eliminates the thread-local and the set

17:48 <headius[m]> so this looks decent so far

17:49 <headius[m]> I'm looking at the ideap graph for this on graal CE and considering next steps

17:49 <headius[m]> ideal graph

17:49 <headius[m]> the null self seems like the next thing to deal with

17:49 <headius[m]> we pass a null self into most yield paths, basically to indicate "get self from binding"

17:49 <headius[m]> that also triggers it to set self onto the frame

17:51 <headius[m]> so next question is why binding has a self when it already has a frame with self

18:11 <enebo[m]> we talked about this a couple of weeks ago...i think one thought was instance_eval

19:29 <headius[m]> yeah I suppose there are cases where you want the frame to still be the same but update only self

19:29 <headius[m]> need to make those the rare case rather than doing this dance all the time

19:35 <enebo[m]> yeah agreed. and documented since the number of things which set frame self is basically only two cases