--- Log opened Fri Feb 22 00:00:08 2013
--- Day changed Fri Feb 22 2013
00:00 <@isaacs> yeah
00:00 <@isaacs> like, changing the message doesn't automatically update the stack trace
00:00 <@isaacs> kinda ruins our AssertionError class
00:02 <@isaacs> ohh... it's like this.message is read-only or something
00:06 <@isaacs> oh, no
00:06 <@isaacs> it's that the stack getter doens't call the toString getter
00:06 <@isaacs> er, the toSTring function
00:16 <@isaacs> trevnorris: https://github.com/isaacs/node/tree/v8-3.16.14
00:16 <@isaacs> or tjfontaine 
00:17 <@isaacs> whoever asked for that.  you guys look the same on irc
00:17 < trevnorris> freaka whoot!
00:17 < trevnorris> tjfontaine: i know you've been around a lot longer than I have and all, but you need to change your nick. ;-)
00:18 < tjfontaine> haha
00:18 < tjfontaine> trevnorris: in irc age I'm much older than you even know :P
00:19 < trevnorris> lol. yeah, I got kind of a late start.
00:20 < bnoordhuis> i think isaacs has the same nick color script i have...
00:21 < tjfontaine> people with good clients should have a problem with us 3, since our lengths are all the same :)
00:21 < TNorris> there, better?
00:21 < tjfontaine> haha
00:21 < tjfontaine> :P
00:22 < trevnorris> was using that for a while, until i figured it would confuse everyone not having the same nick as my github.
00:22 < tjfontaine> thank god, finally got multiarch builds working
00:24 < MI6> joyent/node: bnoordhuis created branch issue4808 - http://git.io/PmtBAw
00:24 < bnoordhuis> ^ someone LGTM that pretty please?
00:25 < tjfontaine> lgtm :)
00:25 < bnoordhuis> merci :)
00:25 < MI6> joyent/node: Ben Noordhuis v0.8 * 22d3eff : doc: add note about child process line buffering  Fixes #4808. - http://git.io/bqfs4w
00:30 <@isaacs> hmmm... this is odd.  apparently there's some issue with V8 3.16 when stdout is a file
00:31 < trevnorris> typeof null === 'object' <- wtf? i knew this, but why oh why?
00:36 < bnoordhuis> trevnorris: what else?
00:39 < trevnorris> bnoordhuis: hm?
00:40 <@isaacs> ok, so, this is weird
00:40 <@isaacs> on 3.16, i'm seeing a timeout (timeout!) from simple/test-stdout-to-file
00:41 <@isaacs> it seems like it's spending many seconds on the data = new Buffer(data, encoding) in fs.SyncWriteStream
00:41 <@isaacs> super strange
00:41 <@isaacs> when i do those same operations in a test file, it takes zero time
00:44 <@isaacs> oh, ok, i can reproduce it in a standalone file
00:44 < trevnorris> isaacs: just fyi, here's a bug I filed about isolates: http://code.google.com/p/v8/issues/detail?id=2487
00:45 < trevnorris> i can't tell if that means the old isolate api is already deprecated in 3.16 or not.
00:45 <@isaacs> hm
00:47 <@isaacs> it happens super fast on the repl, but not in a script
00:48 < tjfontaine> it presents super fast in repl?
00:48 < bnoordhuis> tjfontaine: i can't land that tap pr just yet, i'm 99% it'll break the windows build
00:48 < bnoordhuis> trevnorris: it's deprecated in 3.17
00:49 < bnoordhuis> i was actually working on upgrading to 3.17
00:49 < trevnorris> coolio. i like that idea.
00:49 < tjfontaine> bnoordhuis: at the very least we could drop tap_output = 0; at the top of runner-win.c?
00:50 < bnoordhuis> tjfontaine: i guess that moving `int tap_output` to runner.c would fix it
00:50 <@isaacs> yeah, definitely.  String::Utf8Length() got way way slower.
00:50 <@isaacs> that sucks
00:51 < bnoordhuis> how much slower is way way slower?
00:54 < tjfontaine> bnoordhuis: force-pushed with `int tap_output = 0;` at the top of runner.c
00:54 < tjfontaine> should there be a space between vars there?
00:54 <@isaacs> this looks mostly nice, though: https://gist.github.com/5009914
00:55 < bnoordhuis> tjfontaine: ../../test/run-tests.c: In function ‘maybe_run_test’:
00:55 < bnoordhuis> ../../test/run-tests.c:158: error: too few arguments to function ‘run_test’
00:55 < bnoordhuis> ../../test/run-tests.c:159: warning: control reaches end of non-void function
00:55 < bnoordhuis> same for run-benchmarks.c
00:55 < tjfontaine> oh oh
00:55 < tjfontaine> lemme fixor this
00:55 < bnoordhuis> oh, also long lines in test/runner.c
00:55 < bnoordhuis> wrap at 80 columns
00:56 < tjfontaine> ah run_test itself is too long, ok
01:01 < tjfontaine> bnoordhuis: a single indent in, or line up with first arg?
01:08 <@isaacs> d'oh.  i broke lint.
01:08 <@isaacs> weird that the tests didn't catch it
01:09 <@isaacs> i swear sometimes that thing is just fickle
01:09 < tjfontaine> I don't get pgrep on smartos
01:12 < tjfontaine> ah, pr_psargs is too small to capture what I'm looking for
01:16 < bnoordhuis> tjfontaine: line up (as in: arguments should line up)
01:17 < tjfontaine> ok
01:55 < bnoordhuis> hm, v8 3.17.2 has some rather spectacular performance regressions
01:55 < bnoordhuis> http/cluster.js type=bytes length=102400 c=50: 5.3300
01:56 < bnoordhuis> that's ~630x slower than 3.15
01:56 < bnoordhuis> with c=500 it doesn't even finish in any reasonable amount of time
02:01 <@TooTallNate> not good...
02:06 < bnoordhuis> mraleph: ^ take heed
02:17 < bnoordhuis> some nice speedups on other benchmarks though
02:17 < bnoordhuis> the http thing seems to be a regression in string consing / flattening?
02:18 < bnoordhuis> something to look into tomorrow
02:18 < bnoordhuis> sleep tight people
02:22 <@isaacs> bnoordhuis: way slower = many thousands of times slower.
02:22 <@isaacs> bnoordhuis: from ms to minutes, on a string 1024*1024 in length
02:34 <@isaacs> Here's what's crazy freaky, though... Buffer.byteLength *ISNT* slower in teh repl.
02:35 <@isaacs> only in non-repl contexts.
02:35 <@isaacs> !
02:35 <@isaacs> 
02:35 <@isaacs> !?
02:37 < Raynos> isaacs: https://github.com/joyent/node/issues/4816#issuecomment-13925530
03:28 <@isaacs> Raynos: it's fine: https://gist.github.com/5010496
03:50 < Raynos> isaacs: let me check for objectMode :p
03:50 < Raynos> oh god I have to build master >_<
03:54 < Raynos> is there an easier way to run an individual test yet?
03:55 <@isaacs> Raynos: ./node path/to/test.js
03:55 <@isaacs> Raynos: or python tools/test.py simple/test-stream*
03:55 <@isaacs> Raynos: keep master and v0.8 checked out in two separate folders.
03:55 <@isaacs> makes swithcing easier
03:55 <@isaacs> ok, i'm out
03:55  * isaacs &
03:55 < LOUDBOT> YOU DO, YOU REALLY REALLY DO
03:56 < Raynos> isaacs: https://gist.github.com/Raynos/b61f1c7d5f0534e33931
04:12 -!- mode/#libuv [+o TooTallNate] by ChanServ
13:43 < MI6> joyent/libuv: Ben Noordhuis master * da0b84d : unix: auto-unref spawn handle on process exit  Consistent, for better or  - http://git.io/xUpXGw
13:44 < bnoordhuis> saghul: ^
13:46 < `3rdEden> bnoordhuis: http://cl.ly/image/3U3q381R0b0z
13:46 < bnoordhuis> what will i see when i click on that link?
13:46 < `3rdEden> that your site is down
13:46 < `3rdEden> or goats
13:46 < bnoordhuis> ah
13:46 < bnoordhuis> wfm
13:47 < `3rdEden> Just thought id let you know :)
13:47 < bnoordhuis> yes, thanks. odd though, curl wfm too
13:49 < `3rdEden> bnoordhuis: ye seems to be up again
13:49 < saghul> bnoordhuis great, thanks a lot!
13:49 <@indutny> morning
13:49 < `3rdEden> might have been a small disruption in cloud flare
13:49 <@indutny> wfm
13:49 <@indutny> too
14:23 < bnoordhuis> v8 is pretty flaky lately
14:24 < bnoordhuis> i'm trying to run the linux-tick-processor but d8 crashes during a gc cycle
14:24 < bnoordhuis> both with 3.16 and 3.17
14:25 < bnoordhuis> and of course it doesn't happen with debug builds...
14:26 < bnoordhuis> 41670   99.9%   99.9%  syscall <- umm...
15:15 < bnoordhuis> 95.73%  perf  [unknown]          [.] 0x00000000007a219b <- that's not really helping...
15:22 < bnoordhuis> for some reason, the perf regressions from last night has disappeared...
15:23 < bnoordhuis> without anything changing
15:25 < bnoordhuis> http/cluster.js type=bytes length=102400 c=50: 2438.2 <- that was 5.3 last night
15:28 < bnoordhuis> hm, maybe something did change - but in node, not v8. i see a couple of streams2 patches landed
16:15 < MI6> joyent/libuv: Ben Noordhuis master * c98083e : unix: short-circuit on no-op io watcher changes  Don't add the io watcher - http://git.io/kiWqBw
16:18 < mmalecki> bnoordhuis: hey
16:20 <@isaacs> Raynos: that's not what i'm seeing on master.
16:21 < mmalecki> so Bert is sleeping by a pool, any idea what joke can I pull on him?
16:24 <@isaacs> Raynos: but, it's not actually calling read(0) unless it needs readable, so that only happens once.  hrm.
16:26 <@indutny> mmalecki: sell him as slave
16:31 <@isaacs> bnoordhuis: what's your opinion about v8 versions to ship 0.10 with?
16:31 < mmalecki> indutny: it's illegal
16:31 <@isaacs> bnoordhuis: 16 or 17
16:31 <@indutny> mmalecki: ok, just sell his return ticket and buy one to russia
16:31 <@indutny> i'll sell him there
16:31 <@indutny> its illegal too
16:32 <@indutny> but... you know
16:32 <@indutny> isaacs: are we currently using 16?
16:32 <@isaacs> indutny: we're currently using 15
16:32 <@isaacs> but 16 would be rad
16:32 <@indutny> ok
16:32 <@indutny> yep
16:33 <@indutny> what about tests?
16:33 <@isaacs> i've got 16 passing
16:33 <@indutny> are they passing for both 16 and 17?
16:33 <@isaacs> but there's this weird Buffer.byteLength() taking forever issue.
16:33 <@indutny> oh
16:33 <@isaacs> going to try to create a reduced test case for V8 today
16:33 <@isaacs> it's not slow in the repl, only in a script
16:33 <@isaacs> weird, right?
16:33 <@indutny> yes
16:34 <@indutny> may be it happens because of closure?
16:34 <@indutny> models are rning in clousre
16:43 <@isaacs> hm.
16:44 <@isaacs> maybe, i don't know
16:44 <@isaacs> indutny: fwiw, it's the same with NODE_MODULE_CONTEXTS=1
16:44 <@indutny> oh
16:44 <@indutny> even that
16:46 <@isaacs> with NMC, it's not run in a closure, it's run in a separate context, like the repl
17:15 <@isaacs> hm.  ok, the program is spending 92s in ConsStringIteratorOp
17:15 < tjfontaine> 92 seconds, jeepers
17:18 <@isaacs> yeah, fer realsies
17:18 <@isaacs> it'd take less time to just walk the string and do it manually in js
17:21 <@isaacs> hm.  ConsStringIteratorOp::Search does have a while(true) in it, so this is probably got a bug somewhere.
17:49 <@isaacs> indutny: http://code.google.com/p/v8/issues/detail?id=2551&thanks=2551&ts=1361555104
17:49 <@indutny> great
17:49 <@isaacs> indutny: (if you're curious)
17:49 <@indutny> starred
17:50 <@isaacs> we definitely cannot release v0.10 with that bug, so if it's fixed soon, we can upgrade to v8 3.16.14, and if not, i guess just go out with 3.15 like w ehave now, and probably update to 3.17 in 0.12
17:51 <@isaacs> bnoordhuis: ^
17:51 < tjfontaine> we could float a patch if we fix it first though, right?
17:52 <@isaacs> sure
17:52 <@isaacs> or even if they fix it, and don't backport, or something
17:52 <@isaacs> but 3.16 is their current stable branch, so it'll probably get landed relatively quickly
17:53 < tjfontaine> nod
18:22 < creationix> hmm, can't create an ArrayBuffer in node 0.8.20
18:22 < creationix> new ArrayBuffer(1024*1024)
18:22 < creationix> using official node binary for OSX
18:22 < trevnorris> wtf? how'd that get dropped.
18:22 < creationix> smaller values work
18:23 < creationix> I can create a node Buffer that size no problem
18:24 < trevnorris> hm, just tried in linux. worked.
18:24 < creationix> strange
18:24 < creationix> I'm away from home, I can't test on my other machines
18:24 < creationix> does V8 not have it's own typed array implementation? or is that part of webkit
18:25 < creationix> ok, the large values does eventually work in osx node, it's just really slow (~10 seconds)
18:26 < creationix> hmm, or maybe that's inspect that's killing me
18:26 < creationix> Buffer has a custom inspect
18:27 < creationix> yep, write a dummy ArrayBuffer.prototype.inspect and it's fast
18:27 < creationix> sorry for the noise
18:27 < creationix> I wonder if we could add custom .inspect methods to the the typed array family
18:28 <@isaacs> creationix: sure, ++ for master/unstable, sure
18:28 <@isaacs> creationix: not high priority, though.  patch welcome etc.
18:28 < creationix> or at least, make the offsets not enumerable to match the browser
18:28 <@isaacs> or that
18:28 < creationix> isaacs: yeah, who uses typed arrays in node anyway
18:28 < trevnorris> isaacs: you gotta get a faster machine man. righty - done in 0.00001067s. ;-)
18:28 <@isaacs> hahh
18:28 < creationix> I do but because I'm writing browser code I want to test using node
18:28 <@isaacs> trevnorris: you've got one of those new joyent macbooks
18:29 <@isaacs> trevnorris: when my 3.0ghz comes in, i'll be flyin
18:29 <@isaacs> oh, wait, you're trevnorris not tjfontaine 
18:29 <@isaacs> hahah
18:29 < tjfontaine> :)
18:29  * isaacs actual honest mistake ^
18:29 <@isaacs> not making jokes
18:30 < creationix> isaacs: btw, what did you think of asm.js.  I couldn't tell how much you were being sarcastic on twitter
18:34 <@isaacs> creationix: no, i really think it'd be nice to have a very explicit way to say "Write your code like this and you can have things go very fast, and not have toworry about gc pauses, etc."
18:34 < creationix> I'm just not sure how this would work for JS code
18:34 < trevnorris> isaacs: oh, freak. on master righty -  0.00001067s; 3.16.14 - 0.040659s
18:34 <@isaacs> creationix: i didn't realize at first that there's no way to deal with strings or arbitrary struct datas
18:35 < creationix> all data structures have to be re-implemented in the heap
18:35 < creationix> right
18:35 <@isaacs> creationix: i mean, i'm sure that the browserid folks LOVE this :)
18:35 < creationix> which for me is acceptable since I'm writing a new VM anyway
18:35 < creationix> but for someone wanting to write faster JS, we need something else I think
18:35 <@isaacs> yeah
18:35 < trevnorris> isaacs: that's 3810x's slower...
18:35 < creationix> though maybe parts of node could be implemented in asm.js.  I wonder how fast asm.js md5 would be
18:36 <@isaacs> trevnorris: what about lefty?
18:36 < creationix> my pure js version is already faster than node's bindings because of the C++ tax
18:36 <@isaacs> creationix: right, but it's only faster whne you run it a bazillion times
18:36 <@isaacs> creationix: if you run it once, then the JS tax kills you
18:36 < creationix> true, JITs are faster when hot
18:36 < creationix> but asm.js could AOT
18:37 < creationix> not that MD5 is a bottleneck for anyone
18:37 <@isaacs> creationix: we could also rejigger it so that we create a buffer, call the openssl stuff with the buffer, then read the data out of it when it's done, without ever actually taking the result from C++ to JS
18:37 < creationix> though RSA may be
18:37 < trevnorris> isaacs: still running...
18:37 < creationix> isaacs: heh, that's what my fast md5 does actually, just in all js
18:37 < creationix> creating the buffer is super slow
18:38 < trevnorris> though, what's the likelihood of v8 supporting asm.js?
18:39 < trevnorris> and still running.....
18:40 < trevnorris> this thing's going to burn through my battery before it finishes.
18:41 -!- mode/#libuv [+o TooTallNate] by ChanServ
18:41 < creationix> trevnorris: well, even if they ignore the pragma and don't write a special AOT compiler, the style is code that can be optimized.  I imagine V8 optimizes it some.
18:41 < creationix> except we're using our own implementation of typed arrays that V8 doesn't know much about
18:41 < creationix> so that could be a problem
18:43 < trevnorris> creationix: but there are better optimizations for v8. (e.g. ((n % 8) / 8) is way way faster than ~~(n / 8))
18:43 < trevnorris> or (n/8)|0 or whatever their spec uses.
18:43 < creationix> I see
18:44 < creationix> I know mraleph didn't seem to excited about the asm.js spec
18:44 < trevnorris> creationix: the asm.js spec was created w/ emscripten specifically in mind.
18:44 <@isaacs> creationix: he just wasn't so sure that it'd be super useful for node
18:44 < creationix> I do like the idea of implementing my own data structures on top of a "heap"
18:44 < creationix> my new language semantics don't match lua or js semantics
18:45 < creationix> so I'll essentially skip the native stuff and write my own GC
18:45 < creationix> hopefully it won't be too slow
18:45 < trevnorris> isaacs: left master - 0.000010956; 3.16.14 - 348.98;  that's 31,852,866x's slower.
18:46 <@isaacs> trevnorris: wow, that's even a bigger diff than i'm seeing
18:46 < trevnorris> yeah...
18:46 <@isaacs> so, for the time being, i guess we're staying with 3.15
18:47 < trevnorris> until that's fixed, um yeah.
18:47 < trevnorris> isaacs: also, it was using ~100MB ram while doing that operation.
18:47 < trevnorris> but the string was only 1MB is size.
18:48 <@isaacs> trevnorris: yeah, something is busted there.
18:48 <@isaacs> indutny: feel like digging into some c++? ;)
19:09 < Raynos> Isaac's: its definitely a bug. Ill pr a broken test. What branch should I pr into?
19:17 <@isaacs> Raynos: master
19:17 <@isaacs> Raynos: i have a patch that might be relevant
19:18 <@isaacs> https://gist.github.com/5015824
19:18 <@isaacs> Raynos: ^
19:20 < tjfontaine> well this test-cluster-disconnect test seems to have been racey on smartos for a while (back into .8 days) but in .9 it seems the ECONNRESET stuff makes it die in a more predictable manner
19:20 <@isaacs> tjfontaine: interesting
19:21 < tjfontaine> isaacs: in .8 world when things go wrong it would hang, assuming it's waiting for the listening event, in .9 the handle just isnt there or we get the reset error (which I presume means destory already removed the handle)
19:22 < tjfontaine> there's not a particualrly good reason for this test to be racey as far as I can tell though
19:25 <@isaacs> Raynos: btw, there is no more lowWaterMark
19:25 <@isaacs> lowWaterMark no longer exists.
19:30 <@isaacs> Raynos: https://github.com/joyent/node/pull/4833
19:30 <@isaacs> TooTallNate: ^
19:31 <@isaacs> Raynos: push() triggers a read(0) *on nexttick)
19:31 <@isaacs> Raynos: so, if you push(), push(), push(), that's fine.
19:31 <@isaacs> Raynos: if you push() nexttick(push()), not so fine
19:38 <@isaacs> TooTallNate: ok, the leak test still is fine with this chnage.
19:38 <@isaacs> i'm gonna land on v0.8
19:38 <@isaacs> since we have to do a new one for the EINPROGRESS bs anyway
19:38 <@TooTallNate> isaacs: cool, sgtm
19:39 < MI6> joyent/node: isaacs v0.8 * f9a0140 : http: Handle hangup writes more gently - http://git.io/j6GDUg
19:41 <@isaacs> TooTallNate: how do you feel about the "preemptively read up to hwm" change?
19:43 <@TooTallNate> isaacs: seems fine to me, as long as Raynos is happy
19:49 <@isaacs> Raynos: ?
19:49 <@isaacs> ^
19:49 <@isaacs> bbiab
19:49  * isaacs &
19:49 < LOUDBOT> INTERESTS FROM VIRGINIA NORTHWARD TO NEW ENGLAND SHOULD MONITOR THE PROGRESS OF EARL.
19:49 < Raynos> Good to me
19:50 < Raynos> Preemptive upto hwm is ideal.
19:52 < Raynos> Isaacs: maybe also change defualt high water mark when objectMode is true
19:52 <@indutny> isaacs: sure
19:52 <@indutny> isaacs: what's up?
19:52 <@indutny> that v8 stuff?
19:54 < tjfontaine> more than likely
20:43 < trevnorris> indutny: yeah, the perf regression shown in http://code.google.com/p/v8/issues/detail?id=2551
20:44 < trevnorris> ok, in dns.js there's a variable called "resolver" in exports.resolve that gets assigned a function, but there's also an external function called "resolver".
20:44 < trevnorris> that doesn't seem right...
20:57 <@indutny> ok, looks like I'm just too tired
20:58 <@indutny> with translation of all that papers for tax service
20:58 <@indutny> ttyl guys
20:58  * indutny &
20:58 < LOUDBOT> THEN WHY DID YOU WANT ME TO PRODUCE SOMETHING CONTAINING HALF OF YOUR GENETIC CODE
21:09  * CoverSlide remembers to never left-concatenate anything ever
21:30 -!- mode/#libuv [+o TooTallNate] by ChanServ
21:40 < tjfontaine> bnoordhuis: btw I did push those style changes for tap this morning
21:40 < tjfontaine> well and the build fix for benchmarks
21:52 < bnoordhuis> tjfontaine: i'll look at it in a few
21:52 < bnoordhuis> working through the github notifications as we speak
21:52 < tjfontaine> k take your time, I just wanted to update you
21:56 < tjfontaine> I'm seeing this weird race in test-cluster-disconnect, where getServers gets a tcpSelf with a null ._handle, but if I change the worker count to 1 I can't get it to reproduce, and seems slightly related to the econnreset changes (but it was definitely racey on v0.8 as well)
21:58 < joyent-jenkins> Project nodejs-v0.8 » x64,godfrey build #7: SUCCESS in 8 min 44 sec: https://jenkins.joyent.us/job/nodejs-v0.8/./DESTCPU=x64,label=godfrey/7/
22:10 < joyent-jenkins> Project nodejs-v0.8 » ia32,godfrey build #7:SUCCESS in 20 min: https://jenkins.joyent.us/job/nodejs-v0.8/./DESTCPU=ia32,label=godfrey/7/
22:10 < tjfontaine> you're weird jenkins, turning on colors makes you lose spaces
22:11 < MI6> joyent/libuv: Timothy J Fontaine master * bfe269b : test: add tap output  Given UV_TAP_OUTPUT being set, test result output s - http://git.io/huqqCw
22:13 < tjfontaine> bnoordhuis: thanks
22:14 < bnoordhuis> np
22:14 < tjfontaine> oh you wanted run_tests even more different
22:14  * tjfontaine makes note of style intention
22:15 < bnoordhuis> tjfontaine: if you look around the files in src/ you'll get a feel of the libuv style
22:15 < tjfontaine> I will do that more yes
22:15 < bnoordhuis> though admittedly uv-unix and uv-win differ sometimes
22:16 < bnoordhuis> but when in doubt, go with the uv-unix style - it's provably superior
22:16 < tjfontaine> indeed
22:18 < bnoordhuis> the 3.17 perf regression is back...
22:18 < bnoordhuis> this is spooky, someone hold me
22:20 < tjfontaine> hm a control char must be still leaking through the LOG stuff
22:21 < bnoordhuis> 99.04%  node  node                [.] v8::internal::ConsStringIteratorOp::Search(unsigned int*, int*, unsigned int*)
22:21 < tjfontaine> oh
22:21 < bnoordhuis> what. the. hell?
22:21 < tjfontaine> ya that's what isaacs was seeing as well
22:21 < tjfontaine> he says there's a while(true) in that
22:22 < bnoordhuis> not sure if it's actually an infinite loop
22:22 < bnoordhuis> but i'll find out
22:23 < tjfontaine> I did miss a piece in this that I didn't see in the console but see in the jenkins output, rewind_cursor
22:42 <@isaacs> tjfontaine: joyent-jenkins!!
22:42 <@isaacs> tjfontaine: nice!
22:43 <@isaacs> bnoordhuis: it's definitely not an *infinite* loop
22:43 <@isaacs> bnoordhuis: but it's a "way way way longer than it should be" loop
22:44 <@isaacs> bnoordhuis: it finishes, just ~36M times slower than 3.15.11.15
22:44 < bnoordhuis> yes, i can confirm that
22:44 <@TooTallNate> what's joyent-jenkins 
22:44 <@TooTallNate> ?
22:44 < bnoordhuis> it's our own skynet
22:44 < tjfontaine> isaacs: ya, right now it's tuned for "only when shit breaks" but the tap plugin atm isn't properly failing the build when tests fail
22:44 <@isaacs> tjfontaine: well, it's a while true, but with return statements in a few places.
22:47 <@isaacs> bnoordhuis: yes.  the proof is that it runs on unix.
22:47 <@isaacs> oh, was scrolled back :)
22:48 <@isaacs> (re: uv-unix being provably superior)
22:49 <@isaacs> TooTallNate: you're +1 on preemptive buffering up to hwm?
22:49 <@TooTallNate> isaacs: yes
22:50 <@isaacs> k, landing
22:50 < MI6> joyent/node: isaacs master * 27d1bab : streams: Pre-emptively buffer readables up to the highWaterMark  Also, th - http://git.io/YEzV1A
22:51 <@isaacs> Raynos: send a PR for objectMode default hwm setting
22:51 < Raynos> is 100 a sensible thing?
22:52 <@isaacs> Raynos: also, i think that objectMode: false should mean "never ever object mode; just throw on non-stringish data"
22:52 < Raynos> blargh it depends how big objects are
22:52 < Raynos> isaacs: I think auto conversion into objectMode can be binned
22:52 <@isaacs> k, i'm cool with that
22:53 < Raynos> but you may want to talk to dominictarr, substack about that
22:54 < Raynos> i pinged them :P
22:54 < Raynos> its upto them to get involved if they care now
23:02 <@isaacs> Raynos: i'm kind of thinking that if you opt into objectMode, then that's an advanced use case, and you should set your highWaterMark intelligently
23:02 <@isaacs> and it should not happen automatically.  tha'ts too clever.
23:02 <@isaacs> and these debugger tests are already frustrating me, so i'm going to write that up and PR it.
23:03 < Raynos> isaacs: this.highWaterMark = (hwm || hwm === 0) ? hwm : (objectMode ? 100 : 16 * 1024);
23:03 <@isaacs> meh.
23:04 <@isaacs> what if you really wnat 16K objects?
23:04 <@isaacs> what if they're {x:1}
23:04 <@isaacs> that's not very big
23:05 <@isaacs> or if they're all references to the SAME object, even
23:05 < Raynos> isaacs: I dont know what a good default is
23:05 < Raynos> But bh
23:05 < dominictarr> isaacs: you should never do that
23:05 < dominictarr> that is abuse
23:05 < Raynos> isaacs: actually one thing that would be useful
23:05 < Raynos> is to have a transform have the same hwm as what ever source is piped into it
23:06 < dominictarr> streams are about serialization, but it's okay to send something that isn't serialized, if it COULD be
23:06 <@isaacs> i think node streams should assume buffer/string usage, but not enforce that.
23:06 <@isaacs> in order to not enforce it, we need to have this objectMode flag, but i think YOU should set that
23:06 <@isaacs> node shouldn't.
23:06 < Raynos> I agree
23:06 <@isaacs> if i do fsStream.write({not: "a stringish"}) then it should be a bug
23:06 <@isaacs> and should emit an error
23:06 < Raynos> so if I have an objectMode source (like a mongodb cursor stream)
23:06 < Raynos> it knows what a sensible hwm is
23:06 <@isaacs> the hwm, as well, i mean.. if you're setting non-default stuff, then just set it to a different default.
23:07 <@isaacs> right
23:07 <@isaacs> be sensible
23:07 <@isaacs> it might be 5
23:07 < Raynos> but what about the transform
23:07 <@isaacs> it migth be 5M
23:07 <@isaacs> who knows?
23:07 < Raynos> source.pipe(someTransform())
23:07 < Raynos> how does the transform know what a good hwm is
23:07 <@isaacs> it also takes a highWaterMark arg
23:07 <@isaacs> all duplexes do
23:07 < Raynos> true
23:07 <@isaacs> so, the transform should know what's right for the transform
23:07 < Raynos> but then its source.pipe(someTransform({ highWaterMark: source._readableState.highWaterMark })
23:07 <@isaacs> back to libertarian unix style principles.
23:08 < Raynos> if its an arbitrary transform
23:08 <@isaacs> from each according to his stdout, to each according to his stdin
23:08 < Raynos> then it can work on small and big objects
23:08 <@isaacs> Raynos: well, look, you're creating these things.  so give it an argument.
23:08 < Raynos> also note that with pre-emptive reads a chain of pipe calls effectively pipelines data transformations
23:08 <@isaacs> if you want it to be different, say so.
23:08 <@isaacs> if you want the default, then trust that it knows a good default for itslef.
23:09 <@isaacs> node is going to have sensible defaults for piping buffers and strings around.
23:09 <@isaacs> but all this is changeable.
23:09 < Raynos> isaacs: true.
23:09 < Raynos> i think im worrying too much
23:10 < dominictarr> isaacs: Raynos I think the problem here is that you are trying to write one baseclass that fits too many cases
23:10 < dominictarr> most transforms don't even need water marks
23:10 < dominictarr> from what people actually use…
23:11 < dominictarr> implementing compression streams is a special case.
23:12 < Raynos> dominictarr: anything that needs buffering needs watermark
23:12 < Raynos> through is a transform that buffers 
23:12 < Raynos> its just that through has a hwm of 0
23:13 < dominictarr> I disagree.
23:13 < dominictarr> make the ends smart
23:13 < dominictarr> FS.ReadStream is smart about buffering
23:14 < dominictarr> and let the through streams in the middle just propagate their pause state rapidly
23:14 < dominictarr> (currently it takes several chunks to get the pause state all the way back up the chain)
23:14 < Raynos> thats a different approach
23:15 < tjfontaine> bnoordhuis: what are your thoughts for what might cause tcpSelf._handle in cluster._getServer (specifically in tcpSelf.once('listening')) to end up being null? what should I be looking for here, a worker that is closing the handle? 
23:15 < Raynos> if each transform had a hwm you could effectively build a pipeline
23:15 < Raynos> but maybe your right
23:15 < Raynos> through should just set hwm to 0
23:16 < bnoordhuis> tjfontaine: um, is 'no idea' an acceptable answer?
23:16 < tjfontaine> certainly, because that's where I am
23:20 < tjfontaine> with NODE_DEBUG=net this is what I get https://gist.github.com/tjfontaine/5017342
23:28 < trevnorris> bnoordhuis: mentioned you're checking out the v8 regression is from?
23:30 < trevnorris> bnoordhuis: also noticed that while it's 3e6x's slower, it also was using 90+ MB ram while doing it.
23:36 < bnoordhuis> trevnorris: https://code.google.com/p/v8/issues/detail?id=2551#c2
23:37 < bnoordhuis> i'm kind of disappointed with the v8 people that they let something like this slip through
23:37 < tjfontaine> that's crazy
23:37 < tjfontaine> it has to be hitting them somewhere else
23:38 < bnoordhuis> yeah. i can't imagine where the only ones that use Utf8Length() :)
23:39 < bnoordhuis> *we're
23:40 < bnoordhuis> tjfontaine: re test/simple/test-cluster-disconnect.js, is that a sporadic thing or ?
23:41 < tjfontaine> bnoordhuis: it's non-existent on linux and osx, but I can hit some form of it every 3 or 5 runs, sometimes quite a few in a row fail
23:41 < tjfontaine> bnoordhuis: sometimes I'll get a ECONNRESET or an ECONNREFUSED
23:43 < bnoordhuis> oh, only on smartos you mean?
23:43 < tjfontaine> yes
23:43 < bnoordhuis> i so not want to log onto a smartos machine right now :/
23:43 < tjfontaine> I put assert(tcpSelf._handle) in once('listening')
23:43 < bnoordhuis> truth be told, i'm not sure what machine i'm supposed to be using nowadays
23:44 < tjfontaine> I'm not sure either really, I'm on one in some lab somewhere
23:45 < tjfontaine> https://gist.github.com/tjfontaine/5017342/raw/9f876059c6aa04bf7ff29cf2ab59e2ad3bac2c2c/few-runs.txt
23:45 < tjfontaine> with the assert
23:46 < tjfontaine> if I change workers to 1 in the test I never get a failure, but in other tests increasing the amount of workers doesn't seem to break them
23:46 < bnoordhuis> logging onto umcats, one sec
23:46 < tjfontaine> k
23:47 <@isaacs> bnoordhuis: btw, umcats is kind of a wimpy machine
23:47 <@isaacs> bnoordhuis: you should log into the smartos machine
23:47 < tjfontaine> oh correction, I am getting it with `workers = 1` I must just have been lucky the last time I tried it
23:47 <@isaacs> bnoordhuis: ssh smartos.nodejs.org
23:47 < trevnorris> bnoordhuis: looks like you can also do a v8::Utf8Length(String, Isolate), or is that just for internal use?
23:49 < bnoordhuis> isaacs: what machine / address is that?
23:49 <@isaacs> bnoordhuis: it's 67.188.180.100
23:49 <@isaacs> bnoordhuis: you have root on it
23:50  * bnoordhuis updates .ssh/config
23:51 < bnoordhuis> trevnorris: i think that's an internal one (just logged out of the machine i was testing on)
23:51 < trevnorris> bnoordhuis: yeah, you're right. not in v8.h
23:52 <@isaacs> bnoordhuis: does dns not find that machine for you at smartos.nodejs.org?
23:52 < trevnorris> nice job btw. that issue is insane
23:56 <@ryah> 0.10 is going to be great
23:58 < bnoordhuis> isaacs: it does now, it seems
--- Log closed Sat Feb 23 00:00:20 2013