<chrisseaton>
tenderlove: yes and I read his issue in the Ruby bug tracker
<tenderlove>
what do you think about it?
<chrisseaton>
It looks like a sensible approach to making a real JIT for MRI
<tenderlove>
(I'm still catching up on the thread, so if you've commented there I'll get to it :) )
<chrisseaton>
I disagree with how he's measuring performance and assessing TruffleRuby :)
<chrisseaton>
Our basic benchmarks are actually a bit broken at the moment as we've been focusing on C extensions, but we might post a friendly rebuttal to some of the measurements at some point
<chrisseaton>
It sounds like he's focusing on startup and warmup as well as peak, while we're trying to get the best peak that we can
<chrisseaton>
I think objectively, measuring from cold isn't the right way to assess the code quality from TruffleRuby, which is what he says he's trying to do in his issue
<chrisseaton>
I wasn't going to wade into his issue and criticise anything though - it looks like good work
<chrisseaton>
For the benchmarks he has, we should be able to optimise most of them to nothing, but I think they may have been broken in the version he tried
<chrisseaton>
I think he should use benchmarks-ips basically
<tenderlove>
I guess startup speed is important to MRI
<tenderlove>
but maybe not other implementations
<chrisseaton>
Yes if that's what he's aiming at then that's a completely legitimate goal to have and maybe TruffleRuby isn't the tool for that job (but SVM could be)
<chrisseaton>
But I think he says concludes from his benchmarks that his code is better quality which isn't a conclusion that makes sense from that experiment
<chrisseaton>
And I mean why guess? Look at the assembly from both
<chrisseaton>
"I think Graal performance based on some articles I read is somewhere in the middle between JVM client and server JIT compilers" I think even someone outside the TruffleRuby project, like Charles, would say that's not remotely the case
<lopex>
lol, like the perf could be determined by some articles
<lopex>
our vm is not fast enough, lets write an article
hobodave has quit [Quit: Computer has gone to sleep.]