josh-k has quit [Remote host closed the connection]
<yopp>
yorickpeterse, hey. How your work on oga is going?
<chrisseaton>
Is this the right place to ask about RubySL? It looks like rubysl-set is out of date compared to MRI and JRuby. Is there an update policy? To be fair I think most of the changes might be documentation or cosmetic.
<yorickpeterse>
yopp: working on replacing the Racc parsers with some custom stuff
<yorickpeterse>
chrisseaton: we generally update them whenever needed, there's no particular policy for it
<yorickpeterse>
chrisseaton: we just copy-paste the changes basically, note the MRI commit in the Git commit and that's about it
max96at is now known as max96at|off
<yorickpeterse>
chrisseaton: the only "policy" is that MRI 2.0 stuff goes on the 2.0 branch, 1.x stuff on the 1.0 branch, etc
justinmcp_ has joined #rubinius
justinmcp has quit [Ping timeout: 272 seconds]
havenwood has joined #rubinius
elia has joined #rubinius
diegoviola has joined #rubinius
nirvdrum has joined #rubinius
<brixen>
cpuguy83: lazy Q, could you please link me to the proposed Dockerfile for rbx?
goyox86 has joined #rubinius
goyox86 has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
JohnBat26 has joined #rubinius
meh` has joined #rubinius
max96at|off is now known as max96at
* yopp
trying to start project under rbx
<brixen>
yopp: what kind of project?
<yopp>
a HUGE pice of enterprise shit :)
<yopp>
I'd like to see, if JIT can improve some stuff
<brixen>
oh, you mean a migration?
<yopp>
Maybe, if it will fly without major changes
<brixen>
ok, keep me posted
<brixen>
I just now added some logging to the JIT on what's not inlining
<brixen>
I'm trying to push out little pieces of JIT improvement
<yopp>
btw, is there any way to tell rbx to JIR earlier, without pre-heating codepaths?
<yopp>
•JIT
<brixen>
the present JIT is probably about 10% effective given its potential
<brixen>
what would be the reason to JIT earlier?
<brixen>
you can reduce the call threshold
<yopp>
To get an idea how it might help in my case. I don't have much performance tests for this thing, and I'd like to take a quick look, to get an idea
<brixen>
well, here's the issue with that
<brixen>
you want your execution to settle down some
<brixen>
if the type profiles are changing a lot, you're just going to thrash
<brixen>
do you have the metrics going anywhere now?
<yopp>
nope
<brixen>
you should set that up
<brixen>
I'm making sure the JIT metrics are working right now
<yopp>
It's running in very strict environment, we can't do that atm
<brixen>
you can't use the influxdb-grafana container?
<yopp>
I can try to agree on getting it, but mainly nope.
elia has quit [Quit: Computer has gone to sleep.]
<brixen>
well, if you want to actually do perf testing, you need to find a way to set that up
<yopp>
Okay, good news, seems like it installed all the deps
<yopp>
And server is started!
<yopp>
wow
<brixen>
my whole reason to put that stuff in is so I can help you fix stuff without seeing your code
<brixen>
and you'll save a lot of time by just running your app and using the metrics to guide you
<yopp>
I can try doing this locally, it's not quite the same, but might be helpful
<brixen>
rather than wasting time with synthetic bullshit benchmarks
<yopp>
wow
<yopp>
it's actually running
<brixen>
sweet
<brixen>
note that load times for rbx is still shit
<brixen>
working on that as soon as I get psych sorted
<yopp>
~300ms vs ~100ms on mri
<brixen>
it's not a lot of work to fix the load times but it keeps getting pushed behind other stuff
<brixen>
ok
<yopp>
how many calls required, so calls will be JITed?
<yopp>
I can run ab agains api, to see how it reacts
<yopp>
also, I have 2.2.10
<brixen>
you should try master
<brixen>
the default threshold is 32k
<brixen>
I can't stress this enough though, you need to get the metrics set up
<brixen>
then you can see shit like total heap sizes, concurrent GC time, stop-the-world time, etc
<brixen>
what's a request doing? rendering json, html, other, etc?