<GitHub1>
[jruby] pitr-ch pushed 3 new commits to master: http://git.io/vuZYC
<GitHub1>
jruby/master fbb40d1 Petr Chalupa: [Truffle] code style
<GitHub1>
jruby/master 1eb3fc0 Petr Chalupa: [Truffle] Print warning when running without Graal
<GitHub1>
jruby/master dc46771 Petr Chalupa: [Truffle] tag slow spec
<GitHub5>
[jruby] pitr-ch commented on commit 1eb3fc0: cc @jruby/truffle http://git.io/vuZY0
<GitHub64>
[jruby] chrisseaton commented on commit 1eb3fc0: I think this message could perhaps be more explicit, with a 'warning:' prefix, and mention that it will run slower without it. http://git.io/vuZOz
dfr has quit [Ping timeout: 260 seconds]
tcrawley-away is now known as tcrawley
dfr has joined #jruby
vtunka has quit [Quit: Leaving]
vtunka has joined #jruby
nirvdrum has joined #jruby
<GitHub51>
[jruby] pitr-ch pushed 1 new commit to master: http://git.io/vuZuM
<GitHub51>
jruby/master 94ad479 Petr Chalupa: [Truffle] Update warning message when not running on Graal
<GitHub176>
[jruby] pitr-ch commented on commit 1eb3fc0: updated in 94ad479 http://git.io/vuZub
<GitHub87>
[jruby] headius pushed 1 new commit to ruby-2.3: http://git.io/vuZ2p
<GitHub87>
jruby/ruby-2.3 f1d645c Charles Oliver Nutter: Merge remote-tracking branch 'origin/master' into ruby-2.3
<GitHub92>
[jruby] headius pushed 1 new commit to ruby-2.3+socket: http://git.io/vuZav
<GitHub92>
jruby/ruby-2.3+socket 1b0ce40 Charles Oliver Nutter: Merge branch 'ruby-2.3' into ruby-2.3+socket
<headius>
one of the complaints the travis folks had that has forced them to move stuff to MRI was that their many little services all took up too much memory running independently on JRuby
<headius>
our base memory size to even start up is 20x MRI's
<headius>
of course JVM folks would ask why they're running many independent JVMs but that's ignoring the fact that people want to be able to deploy microservices independently
<headius>
chrisseaton: I don't know when it will become a priority, but memory use and startup time will need to be significantly for anyone to adopt the truffle stuff, since they're both even worse than JRuby proper
<headius>
er, significantly better
<headius>
and without a closed-source option like SVM
<headius>
don't get me wrong...I wish we lived in a world of infinite memory :-)
<nirvdrum>
codefinger: Surely you guys set a max heap for each of the dynos though, no?
<codefinger>
yea, 350m by default
<codefinger>
but, as detailed in that mailing list thread, there is a lot of native memory being allocated
<nirvdrum>
All metaspace? Or stuff from jnr?
<codefinger>
both
<codefinger>
metaspace is about 60m for a basic app
<nirvdrum>
Wow.
lanceball is now known as lance|afk
<headius>
the biggest single barrier to adoption for either JRuby or JRuby+Truffle is startup time, followed by memory size
<headius>
we, I guess that's the biggest two barriers
<headius>
it has never been about performance
<headius>
to that end I really hope SVM becomes OSS at some point :-)
<codefinger>
jruby has the biggest problem with startup time because it's going head to head with MRI. but many of the jvm langs struggle with it. I know of Clojurescript apps that have chosen to run on Node.js for this reason.
praveen has joined #jruby
<headius>
right, it's not really our fault in the end
<headius>
it's a peril of running on a late-optimizing runtime
<headius>
unfortunately truffle is an even later-optimizing runtime :-\
yfeldblum has joined #jruby
praveen has quit [Ping timeout: 245 seconds]
bbrowning is now known as bbrowning_away
<chrisseaton>
we're tracking this stuff in our CI and will work on it at some point
<chrisseaton>
Truffle doesn't inherently have to be very late optimising - I think we have fairly late counters for compiling at the moment but that's just configuration
<chrisseaton>
and we do things like compiling blocks and even just while loop bodies, which is something JRuby and Rubinius don't do I think
<headius>
chrisseaton: sure, that makes sense...but then there's the cost of compiling too early to balance as well
<headius>
we pay that too of course
<headius>
and we do compile blocks independently now
<headius>
rbx has for a long time
<headius>
neither of us do anything for loops, though...kinda falls into the domain of OSR to me
<chrisseaton>
it doesn't do while loops does it? i don't think so anyway
<headius>
while loops aren't blocks :-)
<headius>
so no
<headius>
it's certainly something we could do, but like OSR in Hotspot, it's almost never useful for real apps
<chrisseaton>
yeah it's OSR - we have an OSR node - Truffle does that - the JRuby part doesn't know about it
<chrisseaton>
we're focusing on getting rails running for the next few months - we know there are other things that are important as well, but at the 3 year point we need to be running Rails or it's starting to look like there's a reason why we can't (there isn't)
<chrisseaton>
especially with OMR being able to run Rails it's a point I want to make clear that we can (will) as well
pawnbox has joined #jruby
<headius>
I have no doubts you'll get rails working in some capacity this year
<headius>
AR will be the big question in that
<headius>
OMR?
<chrisseaton>
We have AS, and I think Arel was working as well
<chrisseaton>
IBM's OMR J9
<chrisseaton>
The OMR doesn't stand for anything apparently
<headius>
oh, the MRI thing
<headius>
yeah, run rails slightly slower than MRI proper
<headius>
they obviously have some work to do too
<headius>
chrisseaton: I don't mean to sound down on the truffle work...I just know the #1 complaint that's kept people from using jruby proper
gaustin has quit [Quit: gaustin]
pawnbox has quit [Ping timeout: 245 seconds]
gaustin has joined #jruby
pitr-ch has quit [Quit: My MacBook Pro has gone to sleep. ZZZzzz…]
gaustin has quit [Client Quit]
pitr-ch has joined #jruby
oblutak has left #jruby [#jruby]
pitr-ch has quit [Quit: My MacBook Pro has gone to sleep. ZZZzzz…]
<chrisseaton>
we've already overcome a lot of potential deal breakers - JVMCI is already in JDK 9 EA, startup time with the SVM (I know that's not public yet), I'm confident we can overcome this one as well
<headius>
yeah great progress on all those fronts
<headius>
has there been any talk of making SVM public?
<chrisseaton>
not that I know of
hobodave has quit [Quit: Computer has gone to sleep.]