<headius>
my wheel of attention got pulled off to other bugs and jnr-* PRs for a while
rcvalle has joined #jruby
<headius>
no enebo today, hmm
nicksieger has quit [Read error: Connection reset by peer]
nicksieger has joined #jruby
<etehtsea>
headius there is more to do, but this pr seems pretty complete to solve specific issue (#3229)
<etehtsea>
I hope to review other commits that left in your branch and open different prs with them
nicksieger has quit [Read error: Connection reset by peer]
nicksieger has joined #jruby
<subbu>
headius, i see you worked on #4167 ... looks like one test is still failing .. but, did you notice any perf changes from the change?
<headius>
subbu: the last failure was a timing test...should be green after I kicked it
<headius>
I have not measured for perf yet but it should definitely be faster
<headius>
the old logic loaded nil from context for every scope load
<subbu>
right .. that was my original reasoning that it would inline in more places, plus the null check is gone .. but curious to see what the concrete numbers are .. will test it once it is merged.
pawnbox has quit [Read error: Connection reset by peer]
<headius>
ok, I'll go ahead with it since this was the bulk of what we wanted to do
pawnbox has joined #jruby
<headius>
runtime generating new DynamicScope arities is a longer-term goal
<subbu>
ya .. that is more of a code maintenance / code size reduction thing, right?
<GitHub132>
[jruby] headius closed pull request #4168: Add paths to *DynamicScope that don't check nulls, offsets. (master...unchecked_dynamicscope) https://git.io/vi7Wk
<GitHub93>
jruby/master a7399b7 Charles Oliver Nutter: Add paths to *DynamicScope that don't check nulls, offsets....
<GitHub93>
jruby/master b8e019a Charles Oliver Nutter: Initialize Thread name to nil. It was returning null if unset....
<GitHub93>
jruby/master 7424803 Charles Oliver Nutter: Merge pull request #4168 from headius/unchecked_dynamicscope...
<headius>
subbu: mainly that, but also a suspicion that many captured scopes are larger than 4 variables
<headius>
ideally we'd only have one or two templates and then at runtime we'd generate scope widths as needed
<subbu>
ah, ok.
<subbu>
"[ERROR] Failed to execute goal on project jruby-core: Could not resolve dependencies for project org.jruby:jruby-core:jar:9.1.6.0-SNAPSHOT: Failed to collect dependencies at com.github.jnr:jnr-constants:jar:0.9.3: Failed to read artifact descriptor for com.github.jnr:jnr-constants:jar:0.9.3: Could not transfer artifact com.github.jnr:jnr-constants:pom:0.9.3 from/to central (https://repo.maven.apache.org/maven2): java.lang.RuntimeException: Unexpected
<subbu>
error: java.security.InvalidAlgorithmParameterException: the trustAnchors parameter must be non-empty -> [Help 1]"
<subbu>
know what that is about?
<headius>
running with a weird JDK
<headius>
like an openjdk build or what?
<subbu>
1.7
<subbu>
i guess i need to bump it.
<headius>
I have seen that when I forgot to choose an installed JDK on OS X and it defaulted to some wonky Java 9 build
<subbu>
alright, that is a bigger project for later tonight.
<headius>
also could be that your jdk doesn't have the extended crypto thingy, but I don't know why that would have changed...I've built without installing that
pawnbox_ has joined #jruby
pawnbox has quit [Read error: Connection reset by peer]
<headius>
subbu: dunno if you noticed but my PR doesn't do anything to alter the IR...I just made the JIT use unchecked paths for all scope reads
<subbu>
yes, i saw.
<headius>
if we want full IR to also use unchecked paths, we'll need an unchecked flag in LocalVariable or a new UncheckedLocalVariable or something
<headius>
ok
<headius>
hmm
<headius>
actually maybe this is just an interpreter change? Make Full always use unchecked too?
<headius>
riight now LocalVariable.retrieve uses getValue(offset, depth) and does its own null check
<subbu>
the interpreters are shared iirc .. i don't remember the changes that tom made there.
<subbu>
so, the paths will need some splitting ... unless the full interp uses a different instance .. in which case, yes, the changes there would be similar to the jit changes.
<headius>
ok
<headius>
enebo may know exactly what to change, so I'll let him weigh in
nicksieger has quit [Read error: Connection reset by peer]
nicksieger has joined #jruby
nicksieger has quit [Read error: Connection reset by peer]
nicksieger has quit [Read error: Connection reset by peer]
nicksieger has joined #jruby
nicksieg_ has joined #jruby
nicksieger has quit [Read error: Connection reset by peer]
nicksieg_ has quit [Read error: Connection reset by peer]
subbu|lunch is now known as subbu
nicksieger has joined #jruby
akp has quit [Remote host closed the connection]
akp has joined #jruby
akp has quit [Remote host closed the connection]
akp_ has joined #jruby
akp_ has quit [Remote host closed the connection]
akp has joined #jruby
<subbu>
chrisseaton, good overview paper of ruby dynamic patterns + techniques to address them.
<chrisseaton>
subbu: yeah it's just a minor survey paper to wrap up all the ideas into one package
<chrisseaton>
the main take-away is that if you make caching hard, people will not do it in all places. If you make it easy you can cache everywhere, which is what Ruby needs.
<subbu>
i think survey papers are always useful .. i appreciated them when i was in grad school since i didn't have to go digging in lots of places for an overview.
<chrisseaton>
Rubinius and JRuby (to a lesser extent) are missing lots of caches because it requires work to add them
<subbu>
even if in this case, the overview is ruby-specific.
<subbu>
i think deopt is the biggest barrier.
jensnockert has joined #jruby
<chrisseaton>
Yeah, deopt is a massive enabler. If you can deopt it's a whole different world.
<subbu>
alright, i am going to crawl back into my wmf world again.
temporalfox has joined #jruby
hobodave has joined #jruby
nicksieg_ has joined #jruby
nicksieger has quit [Read error: Connection reset by peer]
nicksieg_ has quit [Read error: Connection reset by peer]
nicksieger has joined #jruby
hobodave has quit [Quit: Computer has gone to sleep.]
nicksieger has quit [Read error: Connection reset by peer]
nicksieg_ has joined #jruby
nicksieg_ has quit [Read error: Connection reset by peer]
nicksieger has joined #jruby
hobodave has joined #jruby
tcrawley is now known as tcrawley-away
prasunanand has joined #jruby
pawnbox_ has quit [Remote host closed the connection]
claudiuinberlin has quit []
nicksieg_ has joined #jruby
nicksieger has quit [Read error: Connection reset by peer]
pawnbox has joined #jruby
nicksieg_ has quit [Read error: Connection reset by peer]
nicksieger has joined #jruby
bbrowning has quit [Quit: Leaving]
pawnbox_ has joined #jruby
pawnbox has quit [Read error: Connection reset by peer]
nn0 has joined #jruby
<nn0>
Hi Everyone .
<chrisseaton>
nn0: hi!
<GitHub163>
[jruby] BanzaiMan opened issue #4171: JRuby HEAD gets NPE when running `bundle install` https://git.io/vidPr
nnzero has joined #jruby
nnzero has quit [Client Quit]
akp_ has joined #jruby
akp has quit [Ping timeout: 244 seconds]
nicksieger has quit [Read error: Connection reset by peer]
nicksieger has joined #jruby
Aethenelle has quit [Quit: Aethenelle]
akp_ has quit [Ping timeout: 244 seconds]
nn0 has left #jruby ["WeeChat 1.4"]
nicksieger has quit [Read error: Connection reset by peer]
nicksieger has joined #jruby
akp has joined #jruby
nicksieger has quit [Read error: Connection reset by peer]
nicksieger has joined #jruby
akp has quit [Ping timeout: 240 seconds]
tenderlove has joined #jruby
tenderlove has joined #jruby
<lopex>
nirvdrum: the pathological cases for ropes made me think of something, like having a "mutable node types in ther tree" but you need to track that
nicksieger has quit [Read error: Connection reset by peer]
<lopex>
of course a node like "ASetByteAtIndex" list node wouldnt be very optimal too
<lopex>
er, rope node
<lopex>
I guess I'm reinventing something so disregard that
nicksieger has joined #jruby
<lopex>
nirvdrum: the other thing is the rack for example, does it use String as the output so the rope has to be flattened at least once ?
<lopex>
if ruby had streams it would have to right ?
<lopex>
er, at most once I meant (wrt flattening)
<lopex>
doh, and it wouldnt if it had streams
drbobbeaty has quit [Quit: My MacBook Pro has gone to sleep. ZZZzzz…]
nicksieger has quit [Remote host closed the connection]
tcrawley-away is now known as tcrawley
temporalfox has quit [*.net *.split]
Liothen has quit [*.net *.split]
justinmcp has quit [*.net *.split]
m4rCsi has quit [*.net *.split]
kegster has quit [*.net *.split]
kylo has quit [*.net *.split]
subbu has quit [*.net *.split]
lopex has quit [*.net *.split]
fidothe has quit [*.net *.split]
mjc_ has quit [*.net *.split]
mccraig_ has quit [*.net *.split]
snkcld has quit [*.net *.split]
joevandyk has quit [*.net *.split]
cremes has quit [*.net *.split]
ale has quit [*.net *.split]
bruceadams has quit [*.net *.split]
guilleiguaran__ has quit [*.net *.split]
atambo has quit [*.net *.split]
deepak_ has quit [*.net *.split]
phlebas has quit [*.net *.split]
quadz has quit [*.net *.split]
ruurd has quit [*.net *.split]
haze has quit [*.net *.split]
tcrawley has quit [*.net *.split]
byteit101 has quit [*.net *.split]
zph has quit [*.net *.split]
jhass has quit [*.net *.split]
beawesomeinstead has quit [*.net *.split]
cschneid has quit [*.net *.split]
amdprophet has quit [*.net *.split]
Scorchin has quit [*.net *.split]
andrewvc has quit [*.net *.split]
qmx has quit [*.net *.split]
samuelkadolph has quit [*.net *.split]
mberg has quit [*.net *.split]
lanceball is now known as lance|afk
subbu has joined #jruby
mjc_ has joined #jruby
lopex has joined #jruby
mccraig_ has joined #jruby
joevandyk has joined #jruby
fidothe has joined #jruby
cremes has joined #jruby
snkcld has joined #jruby
ale has joined #jruby
bruceadams has joined #jruby
deepak_ has joined #jruby
phlebas has joined #jruby
atambo has joined #jruby
guilleiguaran__ has joined #jruby
johnsonch is now known as johnsonch_afk
snkcld has quit [Max SendQ exceeded]
andrewvc has joined #jruby
qmx has joined #jruby
samuelkadolph has joined #jruby
mberg has joined #jruby
amdprophet has joined #jruby
cschneid has joined #jruby
beawesomeinstead has joined #jruby
jhass has joined #jruby
byteit101 has joined #jruby
mjc_ has quit [Ping timeout: 252 seconds]
mccraig_ has quit [Ping timeout: 252 seconds]
fidothe has quit [Ping timeout: 252 seconds]
guilleiguaran__ has quit [Ping timeout: 252 seconds]
m4rCsi has joined #jruby
Liothen has joined #jruby
justinmcp has joined #jruby
temporalfox has joined #jruby
electrical has quit [Ping timeout: 265 seconds]
kegster has joined #jruby
kylo has joined #jruby
ruurd has joined #jruby
haze has joined #jruby
quadz has joined #jruby
tcrawley-away has joined #jruby
lopex has quit [Ping timeout: 265 seconds]
deepak_ has quit [Ping timeout: 265 seconds]
zph has joined #jruby
fidothe has joined #jruby
mccraig_ has joined #jruby
electrical has joined #jruby
guilleiguaran__ has joined #jruby
hobodave has quit [Ping timeout: 244 seconds]
deepak_ has joined #jruby
mjc_ has joined #jruby
Scorchin has joined #jruby
lopex has joined #jruby
enebo has quit [Quit: enebo]
snkcld has joined #jruby
lopexx has joined #jruby
akp has joined #jruby
drbobbeaty has joined #jruby
akp has quit [Ping timeout: 244 seconds]
lopexx has quit []
akp has joined #jruby
akp_ has joined #jruby
akp has quit [Ping timeout: 276 seconds]
akp_ has quit [Remote host closed the connection]
akp has joined #jruby
<lopex>
chrisseaton, nirvdrum: if you speculate on something I guess it's like a new dimension of "cost the the guard" / "payoff" now ?
<lopex>
I wonder if it could be classified like rope operations but on a whole new level
<chrisseaton>
Yes but the cost of the guard is almost always a single word comparison, and often nothing at all
<lopex>
chrisseaton: you think ruby's arrays might be made persistent vecors ?
<lopex>
but then you'd also hit the append case
<lopex>
*vectors
<chrisseaton>
Could do, but would be quite a bit of work to even try it
<lopex>
the pathological aset is something do deal with of course
<lopex>
but can be treated differently
<lopex>
*to
<lopex>
same for hashes
<lopex>
well, it all boils down to clojure core structs
<lopex>
since it's dynamic
<lopex>
scala's vector got it's persitence very late on
<lopex>
chrisseaton: but since you have spacialization why not proliferate on rope node types ?
<lopex>
you can du anything
rcvalle has quit [Quit: rcvalle]
<lopex>
nirvdrum: the talk was awesome btw
<lopex>
if one had a single "RopeGrowableNode" for those cases
<lopex>
it's superfast to chek whether tree contains one right ?
<lopex>
you can/or not flatten one accordingly
<lopex>
I guess I'm trying to avoid COW now
<lopex>
but I'm not
<lopex>
er, avoid the word
<lopex>
of course I'm not talking about COW
pawnbox_ has quit [Remote host closed the connection]
<lopex>
so the question is the tracking
<lopex>
chrisseaton: can you speculate on whether an object was seen by one thread ?
<lopex>
ont like na escape analysis
<lopex>
*not
<chrisseaton>
Yeah eregon has written a paper about doing that
jensnockert has quit [Remote host closed the connection]
<lopex>
chrisseaton: you think having a mutable node in rope (with all it;s consequences and additional tracking) - whatever that idea might be just to help the pathological case is worth pursuing ?
<chrisseaton>
We do have a mutable native rope node for C extensions
<lopex>
and for the append and aset case in ruby ?
<lopex>
wow i;ts super interesting
<lopex>
I guess fallbacking to array one is a dead end
<lopex>
or at least short sited
<lopex>
sighted?
<lopex>
anyways Kevin made my eyes open wrt ropes
<lopex>
wrt that familaiarity argument
<lopex>
damn ruby
<lopex>
in fp it's all so simple
<lopex>
you use lenses and zippers
<lopex>
but will all end up same datastructures undeneath
<lopex>
chrisseaton: I very recently watched SPJ video on GHC aproaches to inlining and optimization
<lopex>
but they can use algebra
<lopex>
the case statement inlining and processing was qiote interesting
<lopex>
*their
<chrisseaton>
So can we in many cases - not all of Ruby has side effects
<lopex>
sure
<chrisseaton>
1 + 2 is pure with deopt
<chrisseaton>
So is [1,2,3].sort
<lopex>
yeah
<lopex>
1 + 2 is definitely easier
<lopex>
literal construction is pure
<lopex>
+ is speculative
<lopex>
the sort is like you said "you make a single memalloc you loose"