x1337807x has quit [Read error: Connection reset by peer]
kgerman has joined #jruby
x1337807x has joined #jruby
x1337807x has quit [Max SendQ exceeded]
x1337807x has joined #jruby
kgerman has quit [Client Quit]
tenderlove has joined #jruby
subbu has joined #jruby
kgerman has joined #jruby
benlovell has joined #jruby
colinsurprenant has quit [Quit: colinsurprenant]
<headius>
nirvdrum_: got it
benlovell has quit [Ping timeout: 272 seconds]
zeroecco has quit [Quit: My MacBook Pro has gone to sleep. ZZZzzz…]
halorgium has quit [Quit: gone]
nirvdrum_ has quit [Ping timeout: 246 seconds]
halorgium has joined #jruby
tylersmith has quit []
<subbu>
headius, toStringInstrs in IRScope will bomb after CFG is build .. we should just rip it out or automatically let it delegate to cfg.toStringInstrs() perhaps.
<subbu>
IRScope.toStringInstrs is old time code from 2009/2010 ... we never took it out .. and it is lingering around.
<subbu>
that said,useful for dumping before cfg is built.
x1337807x has quit [Quit: My MacBook Pro has gone to sleep. ZZZzzz…]
enriclluelles has joined #jruby
tenderlove has quit [Remote host closed the connection]
asarih has joined #jruby
tenderlove has joined #jruby
<subbu>
nirvdrum, so, you are not getting a AIOOBE anymore?
enriclluelles has quit [Ping timeout: 246 seconds]
yfeldblum has quit [Ping timeout: 260 seconds]
nirvdrum_ has joined #jruby
<nirvdrum_>
Sorry, timed out there.
<nirvdrum_>
headius: Let me know if you need anything else.
tenderlove has quit [Ping timeout: 272 seconds]
<subbu>
nirvdrum_, so, you are not getting a AIOOBE anymore?
johnsonch is now known as johnsonch_afk
<nirvdrum_>
subbu: I haven't for a bit now. But I think it went away when adding the logging, replaced by the NPE, then by the CCE.
<subbu>
hmm .. ok.
<subbu>
looks like master has slowed down a lot because of jit-compiler exceptions it seems .. headius fyi.
<headius>
not rubyconf... rubyconf is US conf in San Diego in Nov... RubyKaigi is in Japan now
<subbu>
ok.
tesmar has joined #jruby
kgerman has joined #jruby
toady00 has quit [Ping timeout: 260 seconds]
yfeldblum has joined #jruby
toady00 has joined #jruby
jaffachi_ has joined #jruby
skade has quit [Read error: Connection reset by peer]
nirvdrum has quit [Quit: Leaving]
tenderlove has joined #jruby
nirvdrum_ is now known as nirvdrum
<nirvdrum>
headius: Need anything else from me?
vtunka has quit [Quit: Leaving]
<headius>
nirvdrum: not at the moment
skade has joined #jruby
<nirvdrum>
Great. I just don't want to be a bottleneck.
<Aethenelle>
to make the test_simple_include test pass in the reverse condition (every operation in it updates the obj) I have to invalidate the method cache on all including classes... that seems a bit heavy handed for just creating a method... On the other hand, invalidating only that method in the cache is inefficient and eror prone if I poll and set to null or requires making token non-final in CacheEntry (or changing its use throughout to getter/s
<Aethenelle>
I think the best options are blindly set the cache entry to null or make token non-final...
<Aethenelle>
You have a preference?
tenderlove has quit [Ping timeout: 260 seconds]
<headius>
hmmm
<headius>
Aethenelle: method table modifications in a module already invalidate all hierarchies that module is included into
<Aethenelle>
I'll have to look at that again...
<Aethenelle>
o... right... i knew that... just... testing you...
<JRubyGithub>
[jruby] jrubyci pushed 1 new commit to master: http://git.io/eAPgQA
<JRubyGithub>
jruby/master 41de7c6 Charles Oliver Nutter: Clean up more refactored JIT code and unused logic.
JRubyGithub has left #jruby [#jruby]
tylersmith has joined #jruby
<headius>
subbu: let me know when you have a minute to chat
<subbu>
now is good.
BobFunk has joined #jruby
kgerman has quit [Quit: ChatZilla 0.9.90.1 [Firefox 31.0/20140716183446]]
<headius>
I'm having trouble figuring out how to get AddCallProtocol to run properly when we've already done earlier passes
<headius>
it adds IR that I believe needs to go through AddLocalVariableLoadStore and friends
<subbu>
what is going wrong with the current ordering? ...
<subbu>
let me look at the output of that pass ...
<subbu>
it is only push/pop instrs which shouldn't be using local vars as far as i remember.
<subbu>
headius, what problems are encountering? it is just adding couple of pushes and pops at top and bottom without using any vars.
mkristian has joined #jruby
<headius>
sorry, got disconnected
<headius>
some of these passes aren't repeatable
<headius>
subbu: at least ACP isn't repeatable
<headius>
the changes you made still don't run JIT passes either
zeroecco has joined #jruby
<headius>
because the CFG is already built
<subbu>
prepareForCompilation builds cfg and runs all passes including acp ... do you have to rerun any of them later?
<headius>
prepareForCompilation is happening after prepareForInterpretation, which also builds CFG, so pFC bails out without running passes
<subbu>
oh i see. got it.
<subbu>
now i understand the issue ... because this is now a true jit that runs on interped code.
<headius>
if I force it to run the passes anyway, I can proceed...but I think the IR is getting messed up
<headius>
right
<subbu>
we can get rid of the cfg != null bit ... there is a reset-state method in IRScope i think .. but i may be mistaken ...
<headius>
the change I made the other night was to make "JIT" passes only include ACP and Linearize, which seemed to make better IR, but I was just guessing at how to order these
<subbu>
so, acp is not repeatable, yes.
nanoyak has joined #jruby
<headius>
I think AddLocalVarLoadStore may not be repeatable either
<subbu>
neither is add-load-store-... but, that is not an issue for now .. since we don't plan to run those two passes befor einterping.
<headius>
so ALVLS shouldn't normally run for interp?
<subbu>
let us not run it for interp .. it is expensive for interp use case.
<headius>
ok
<nirvdrum>
What's ACP?
<subbu>
AddCallProtocolInstructions
<nirvdrum>
Thanks.
<Aethenelle>
bugger... I'm at unflatten searchMethod or recurse invalidation....
<subbu>
explicit push/pop of bindings and frames
<headius>
I think we could start sorting out the right passs for each now since the JIT is nearly there
simpleman has joined #jruby
<subbu>
headius, but we can tweak the alvls issue since we can track for some scopes (where binding doesn't escape) and rerunning is a no-op => we can enable it for interp since i can benefit interp as well. but, that is for later.
<Aethenelle>
wait... the cache has always been on the calling object, right?
<subbu>
so for interp ... OptimizeTempVarsPass, LocalOptimizationPass, LinearizeCFG
<Aethenelle>
reciever that is...
<subbu>
and then for JIT, you can run alvls,acp, linearizecfg
<headius>
Aethenelle: the class-level cache, yes
<simpleman>
Hello! I am wondering if anyone has any information on setting up qt4 to be used with jruby? Tried requiring qtjambi.jar and qtjambi-gcc.jar and installing qtjruby-core, but when I do Qt::Application.initialize I get "Application not found in packages" error
<headius>
subbu: should I be running OTVP or LOP after ALVLS and ACP?
<subbu>
no.
<subbu>
not necessary.
<subbu>
otvp is run just once after IR is built.
<subbu>
and that should be it.
<subbu>
LOP can be rerun, but it doesn't do much, so it is wasted effort.
<headius>
and ordering of ACP and ALVLS? ACP adds LocalVariable instrs that I think should pass through ALVLS
<subbu>
ACP doesn't add local-var instrs. as far as i know.
<subbu>
only push/pop frame/binding
<headius>
hmm
<headius>
I didn't see LocalVariable in JIT until I turned ACP on
<headius>
let me check
<subbu>
but, it has been a while ... let mem look at the code.
<headius>
oh you know, here's the other question
<Aethenelle>
that means for an included module to update its including hierarchies, I have to invalidate the cache of all including classes... on every method defn... this is going to get hairy...
BobFunk has quit [Ping timeout: 245 seconds]
<headius>
subbu: bak to my email the other night... first I realized ACP wasn't running for JIT, so I turned it on... then I realized GEB was not being included in the list of blocks from IRScope CFG, so I started compiling it manually... and then JIT started seeing LocalVariableInstr
<subbu>
headius, yes, loooking at passes/AddCallProtocolInst... i dont see any l-vars being introduced
<headius>
right, neither do I
<headius>
I think the problem was that I made JIT passes only do ACP and LCFG
<headius>
maybe only those two ended up running?
<headius>
I am going to sort out the transitions for this stuff today
<headius>
now this shows EnsureTempsAssigned running twice
<headius>
but it also shows that ACP finally logic goes into GEB
bbrowning_ has joined #jruby
<simpleman>
anyone got anz pointers?
<subbu>
so, for interp, you dont need EnsureTempsAssigned
<simpleman>
any*
<headius>
if I just use the linearized BB list GEB isn't in it
<Aethenelle>
best option I can think of is recuring up the hierarchy.... wait... we're already doing that just without recursion it won't be any worse than it already is... I wonder if single method invalidation would speed anything up meaningfully...
<nirvdrum>
headius: Is that the %v_2 and %t_a_5 assigns? (Feel free to ignore until you guys are done. I'm just trying to better understand some of this).
<Aethenelle>
... experiment for later i guess....
<headius>
subbu: when I don't explicitly go get GEB and emit code for it, I end up with no finally logic
<headius>
JIT uses linearizedBBList
<subbu>
one sec.
<headius>
nirvdrum: the "v" variables are temp variables used for IR in general, as internal registers essentially
bbrowning has quit [Ping timeout: 246 seconds]
<headius>
the "t" variables are temporary hidden variables used to work with Ruby local variables without loading/storing from/to the heap scope
<subbu>
headius, so, here is how i can see this working .. split up passes between interp and jit ... interp gets otvp, lop, lineraize cfg
<subbu>
jit gets alvls, alp, linearizecfg
bbrowning_ is now known as bbrowning
<headius>
AddLocalVariableLoadStore pass basically turns code that would be constantly reading and writing heap variables (closures, for example) into normal JVM local variable operations until they need to be visible in the scope again
<headius>
subbu: ok, yeah, I'm trying that locally now
<subbu>
but, before running jit passes, you can call resetLinearizationDAta on IRScope .. so that linearizeCFG redoes its thing.
<headius>
ok
colinsurprenant has joined #jruby
<subbu>
or you can all resetState (but that is not necessary .. but if you do, you have to recompute scope flags).
<simpleman>
Hello! I am wondering if anyone has any information on setting up qt4 to be used with jruby? Tried requiring qtjambi.jar and qtjambi-gcc.jar and installing qtjruby-core, but when I do Qt::Application.initialize I get "Application not found in packages" error. Any pointers are welcome I'm starting to go crazy
<nirvdrum>
simpleman: I've never heard of that and don't know anyone using it, unfortunately.
<headius>
subbu: I will go with resetLin for now
<subbu>
k
<subbu>
once enebo is back, we can meet up and fix up some of this dependency business .. it has gone stale.
<simpleman>
nirvdrum: any suggestions about a different gui toolkit similliar to qt4? haven't worked with any other
<subbu>
we still need the flexibility of running passes that are provided on the CLI .. that lets us bypass bugs and also run interp against a full set of passes to keep correctness tested in CI runs.
<headius>
that's looking better
<headius>
subbu: yeah I would expect those passes to be run as interpreter passes... JIT passes are usually going to be pretty strictly controlled
<nirvdrum>
simpleman: I haven't used it, but I've seen discussion in here in the past about jrubyfx.
<subbu>
also, fwiw, when i run specs with all known passes enabled, we have a couple additional spec failures .. i'll work on those one of these days soon.
<headius>
ok... "def foo; p block_given?; end; foo" seems to compile ok with these passes
<headius>
(it needs frame)
<headius>
foo() {} and foo() both produce expected results
tenderlove has joined #jruby
<headius>
simpleman: I don't know how Qt or qtjambi work...most folks that do GUI stuff on JRuby use SWT or Swing
<simpleman>
nirvdrum: i'm pretty new to this whole jruby thing, but I'm pretty sure the jruby-core gem is out of date and thats why qt4 bindings are not working..time for a replacement. Going to try jrubyfx then, thanks for the suggestion :)
yfeldblu_ has joined #jruby
<simpleman>
headius: Yeah I guess I'll need to go down that road..its just that swt and swing styles are not that good looking in my opinion, thats why I wanted to roll with qt4, but the bindings are outdated so no luck
<headius>
mmm yeah :-(
<headius>
I believe SWT uses gtk widgets
yfeldblum has quit [Ping timeout: 246 seconds]
tenderlove has quit [Ping timeout: 258 seconds]
<simpleman>
I'm going to try jrubyfx,swt and swing and just make a choice since these should work as expected. Thank you headius and nirvdrum for your input!
yfeldblu_ has quit [Ping timeout: 255 seconds]
<headius>
ok
<headius>
have fun!
<simpleman>
Who knows, maybe when I get the grasp of all the jruby stuff I'll start working on some qt4 bindings ;)
<headius>
that would be excellent!
DrShoggoth has quit [Quit: Leaving]
<tarcieri>
goddamn my phone's battery is so fucked :(
<tarcieri>
it lasts like 4 hours now
tesmar has quit [Quit: tesmar]
<headius>
turn off LTE?
tesmar has joined #jruby
robbyoconnor has quit [Ping timeout: 272 seconds]
iamjarvo has joined #jruby
DrShoggoth has joined #jruby
mister_solo has joined #jruby
JRubyGithub has joined #jruby
<JRubyGithub>
jruby/master 5ff9567 Charles Oliver Nutter: Tweak IR passes and when they run for JIT. Fix spec.
<JRubyGithub>
[jruby] jrubyci pushed 2 new commits to master: http://git.io/dNOe_A
<JRubyGithub>
jruby/master 4475d60 Charles Oliver Nutter: Fix AliasInstr in JIT to do toString on operand results.
JRubyGithub has left #jruby [#jruby]
skade has quit [Quit: Computer has gone to sleep.]
yfeldblum has quit [Remote host closed the connection]
Hobogrammer has joined #jruby
tenderlove has joined #jruby
jaffachi_ has quit [Quit: Computer has gone to sleep.]
diegoviola has joined #jruby
elia has joined #jruby
tenderlove has quit [Ping timeout: 255 seconds]
kgerman has quit [Quit: kgerman]
iamjarvo has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
mkristian has quit [Quit: bye]
iamjarvo has joined #jruby
<dfr|work>
whee! I've wrestled some more code away from jruby-rack! :D Now neither RackServlet nor AbstractRackDispatcher are in my processing path. :D
tesmar has quit [Quit: tesmar]
<headius>
dfr|work: excellent!
<headius>
subbu: oops, added just to see the resulting IR, but forgot to remove
<headius>
subbu: I'm looking at the logic for defining a method...IRRuntimeHelpers.findInstanceMethodContainer
<dfr|work>
headius, are you also a jruby-rack hater? :)
<headius>
it requires a dynamic scope always, because that's where it gets "evalType" from
<headius>
but I'm thinking eval type should be a property of the static scope, and when we want to use a block in an instance_eval we clone the static scope too
<headius>
this is for blocks used in instance_eval, module_eval, etc, where the "def" target changes from when the block is run normally
<headius>
I'm trying it now...patch may be more coherent than me
postmodern has joined #jruby
bbrowning is now known as bbrowning_away
<headius>
oh darn...block bodies aren't cloneable
etehtsea has joined #jruby
tesmar has joined #jruby
toshetm has joined #jruby
subbu is now known as subbu|away
robbyoconnor has joined #jruby
robbyoconnor has quit [Client Quit]
<headius>
hmmm
<headius>
is it ever the case that dynScope.parent.staticScope != dynScope.staticScope.parent?
iamjarvo has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
subbu|away has quit [Ping timeout: 260 seconds]
iamjarvo has joined #jruby
mister_solo has quit [Ping timeout: 260 seconds]
subbu|away has joined #jruby
<headius>
yeah this is going to have to change...there's a race too
yfeldblum has joined #jruby
skade has joined #jruby
yfeldblum has quit [Ping timeout: 260 seconds]
yfeldblum has joined #jruby
tenderlove has joined #jruby
subbu|away has quit [Ping timeout: 272 seconds]
mister_solo has joined #jruby
simpleman has quit [Ping timeout: 246 seconds]
tenderlove has quit [Ping timeout: 244 seconds]
mister_solo has quit [Ping timeout: 240 seconds]
toady00 has quit [Remote host closed the connection]
e_dub has quit [Quit: ZZZzzz…]
toady00 has joined #jruby
dabradley has quit [Ping timeout: 255 seconds]
toady00 has quit [Ping timeout: 246 seconds]
subbu has joined #jruby
nanoyak has quit [Quit: Computer has gone to sleep.]
Hobogrammer has quit [Ping timeout: 255 seconds]
dabradley has joined #jruby
BobFunk has quit [Ping timeout: 246 seconds]
guilleiguaran___ has quit [Ping timeout: 258 seconds]
elia has quit [Quit: Computer has gone to sleep.]
BobFunk has joined #jruby
skade has quit [Quit: Computer has gone to sleep.]
guilleiguaran___ has joined #jruby
skade has joined #jruby
tlarevo has joined #jruby
<headius>
subbu: I think I'm going to perform some surgery on how instance_eval and friends are set up
<subbu>
i see .. ok.
<headius>
what we have now is spread all over the place and harms the genericity of the jit
<headius>
e.g. jit wants to optimize a block to have no dynscope, but instance_eval stores evalType on dynscope
<headius>
bbiab
<subbu>
right, ideally I would have set it on static-scope .. but i couldn't find a realiable way of setting it there and getting it to where it was needed .. so, evalType on dynscope ws a hack that worked.
<headius>
right, no problem...static scope is the right place, but even getting it there is complicated because we can't clone the block body
<headius>
what I'm trying to figure out is how to localize all block pre/post logic into the block body impls
<headius>
including instance_eval pre/post
<subbu>
k
JRubyGithub has joined #jruby
<JRubyGithub>
[jruby] jrubyci pushed 3 new commits to master: http://git.io/jzcp7g
<JRubyGithub>
jruby/master 35141a4 Charles Oliver Nutter: Clean up block cloning for instance_eval and friends.
<JRubyGithub>
jruby/master 2d71826 Charles Oliver Nutter: Clean up and thread-localize evalType from IR block body.
<JRubyGithub>
jruby/master 8c49c0d Charles Oliver Nutter: Doco fix.
JRubyGithub has left #jruby [#jruby]
tcrawley is now known as tcrawley-away
rimenes has quit [Quit: My MacBook Pro has gone to sleep. ZZZzzz…]
drbobbeaty has quit [Quit: My MacBook Pro has gone to sleep. ZZZzzz…]
mister_solo has joined #jruby
rimenes has joined #jruby
skade has quit [Quit: Computer has gone to sleep.]
fridim__ has quit [Ping timeout: 272 seconds]
mister_solo has quit [Ping timeout: 245 seconds]
Hobogrammer has joined #jruby
jamo_ has quit [Ping timeout: 245 seconds]
donV has joined #jruby
mpapis has quit [Ping timeout: 240 seconds]
jamo_ has joined #jruby
mpapis has joined #jruby
johnsonch is now known as johnsonch_afk
<headius>
subbu: heh, I think I know where some of my confusion was coming from... ACP doesn't do anything to blocks
<headius>
blocks still expect their scope to be already pushed on ThreadContext stack
<subbu>
ah, ok ... i forget why i don't run it on blocks.
<headius>
I was trying to figure why I couldn't remove the pre/post logic in CompiledIRBlockBody with ACP active, and this is why :-)
<headius>
I will get things working now with full scope/frame logic around blocks
yfeldblu_ has joined #jruby
<subbu>
i can take a look later why i dont add those instrs. for blocks.
<headius>
in order for blocks to run their call protocol, I think we'd need to pass more stuff into the interpreter
<subbu>
but, i think i had a good reason for it :)
<headius>
it needs to get the parent dynamic scope, current evalType, block type, and maybe more
<subbu>
ok.
yfeldblum has quit [Ping timeout: 272 seconds]
BobFunk has quit [Ping timeout: 255 seconds]
<subbu>
compared to couple months back, the interpreter seems to have slowed down by 20-35% ... we can revisit this once the JIT is done.
<headius>
ok
<headius>
driving me nuts because the actual JIT is almost done...it's just this call protocol stuff now
DrShoggoth has quit [Quit: Leaving]
BobFunk has joined #jruby
etehtsea has quit [Quit: Computer has gone to sleep.]
tenderlove has joined #jruby
<chrisseaton>
we're doing quite a bit of CI benchmarking of difference JRuby branches, so now including the IR and the IR JIT - so I can let you know how that looks over time if you want
<chrisseaton>
currently most of them don't compile, but that might be something in the harness
enriclluelles has joined #jruby
yfeldblu_ has quit [Remote host closed the connection]
yfeldblum has joined #jruby
tenderlove has quit [Ping timeout: 260 seconds]
yfeldblum has quit [Read error: Connection reset by peer]
yfeldblum has joined #jruby
yfeldblum has quit [Remote host closed the connection]
<headius>
chrisseaton: yeah hopefully soonish
yfeldblum has joined #jruby
jaffachi_ has joined #jruby
yfeldblu_ has joined #jruby
mister_solo has joined #jruby
yfeldblum has quit [Ping timeout: 272 seconds]
yfeldblu_ has quit [Ping timeout: 244 seconds]
<headius>
mmm 9 failures
<subbu>
headius, know that there are 4 additional failures when i enable ACP in interp mode .. so maybe 5 actually.
<subbu>
heading out shortly .. so, catch you tomorrow.
<headius>
ahh, interesting... ok
mister_solo has quit [Ping timeout: 255 seconds]
johnsonch_afk is now known as johnsonch
etehtsea has joined #jruby
iamjarvo has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
elia has joined #jruby
brettporter has joined #jruby
brettporter has joined #jruby
anaeem1_ has quit [Remote host closed the connection]
elia has quit [Quit: Computer has gone to sleep.]
subbu has quit [Ping timeout: 245 seconds]
<headius>
ahhh, crap
<headius>
er...no...hmm
diegoviola has quit [Remote host closed the connection]
etehtsea has quit [Quit: Computer has gone to sleep.]
towski has quit [Quit: goodbye friends I am gone]
<headius>
ok, yes crap
tenderlove has joined #jruby
towski has joined #jruby
tenderlove has quit [Ping timeout: 260 seconds]
rimenes has quit [Quit: My MacBook Pro has gone to sleep. ZZZzzz…]
mje113__ has quit [Quit: Connection closed for inactivity]
donV has quit [Quit: donV]
towski has quit [Remote host closed the connection]
towski has joined #jruby
rimenes has joined #jruby
tenderlove has joined #jruby
iamjarvo has joined #jruby
zeroecco has quit [Read error: Connection reset by peer]
justinmcp has joined #jruby
jaffachi_ has quit [Quit: Computer has gone to sleep.]
havenwood has quit [Remote host closed the connection]
diegoviola has joined #jruby
havenwood has joined #jruby
jaffachi_ has joined #jruby
havenwood has quit [Remote host closed the connection]
yfeldblum has joined #jruby
tesmar has joined #jruby
havenwood has joined #jruby
tenderlo_ has joined #jruby
marr has quit [Ping timeout: 272 seconds]
tenderlove has quit [Ping timeout: 245 seconds]
JRubyGithub has joined #jruby
<JRubyGithub>
[jruby] jrubyci pushed 3 new commits to master: http://git.io/HdaEVw
<JRubyGithub>
jruby/master 219902c Charles Oliver Nutter: Break will exit method, so insert a return to appease JVM.
<JRubyGithub>
jruby/master 5ed09db Charles Oliver Nutter: Add evalType logic to compiled blocks.
<JRubyGithub>
jruby/master c795b8f Charles Oliver Nutter: Only methods that weren't delegating.
JRubyGithub has left #jruby [#jruby]
enriclluelles has quit [Remote host closed the connection]
tenderlo_ has quit [Remote host closed the connection]
tenderlove has joined #jruby
tesmar has quit [Quit: tesmar]
jaffachi_ has quit [Quit: Computer has gone to sleep.]