»ö« Welcome to Perl 6! | perl6.org/ | evalbot usage: 'perl6: say 3;' or rakudo:, niecza:, std:, or /msg camelia perl6: ... | irclog: irc.perl6.org | UTF-8 is our friend!
Set by sorear on 25 June 2013.
00:00 telex joined, rurban1 joined 00:06 rurban1 left 00:16 Psyche^ left 00:17 Psyche^ joined 00:32 btyler joined
[Coke] lue: github.com/tadzik/Template-Mojo/issues/4 00:38
Anyone talk to supernovus outside of irc? 00:39
01:02 ajr_ left 01:03 rurban1 joined 01:05 bbkr_ joined 01:07 bbkr left, rurban1 left 01:11 genehack left 01:23 leont left 01:39 eternaleye left 01:42 eternaleye joined 01:55 araujo left 01:56 tgt left, araujo joined 02:04 rurban1 joined
TimToady should there be a Promise state of Unwanted that tells the code trying to fill a promise that it no longer needs to try? 02:07
02:08 rurban1 left
TimToady (my brother-in-law Mark asks: if you're racing a bunch of promises, and you only want the first one that finishes, how do you stop the others from wasting energy?) 02:08
btyler SIGGIVEUPYOULOST :) 02:09
TimToady one could, I suppose (with proper documentation) have an agreement that either the promiser or the user can get the vow and keep/break the promise, and if the promiser's promise is broken by the user, stop looping or whatever 02:10
or the promiser gets the vow but puts it somewhere public, if it wants to be cancelable by the user 02:12
on that subject, can the user of channel $c call $c.done? I suspect the method (originally called .close iirc) was envisioned only for use by the sender, but can a channel be shut down by the reader? 02:15
or do we need to have something corresponding to a vow for a channel (like maybe a "quest" object, sort of an extended vow :) 02:17
02:21 genehack joined, lue left 02:22 colomon joined
BenGoldberg What happens if there are two or more users of a channel, and only one of those users stops wanting it's output? It would be bad to call $c.done too early, I think 02:22
TimToady well, at some point civility is required 02:23
but I wonder if we're setting ourselves up for a "core leak" 02:24
how does Go handle this?
obviously if you don't trust the code you're running, you shouldn't be allowing it to use up threads without resource limits of some kind, but even assumning non-malignant intent, you can get long-running promises started accidentally, and you'll like to have some way to deal with that maybe 02:27
02:29 kurahaupo_ joined
TimToady or do we just develop a culture of adding timeout promises to everything? is there any way for an anyof to signal the "promise collector" that a "promise collection run" is desirable? :) 02:29
02:31 [Sno] left, [Sno]_ joined 02:32 jnap left
TimToady is there a need for some kind of fratrcidal anyof that breaks all the eggs that didn't hatch first? 02:32
*tric 02:33
energy usage is a concern to various companies running servers 02:35
02:36 bloonix left
TimToady I suppose you can do a "lazy race" with a bunch of channels, as long as they can exert backpressure and specify how much buffering they desire 02:37
a certainly amount of speculative execution can speed throughput, but too much of it just runs up your electric bill, not to mention your air conditioning bill 02:40
TimToady imagines a Google barge in the Arctic that doesn't use much electricity to cool the servers, and then imagines a pun on "sunk cost" 02:42
TimToady probably needs to douse his head in some cold water...
(or do a bit less speculative execution...) 02:43
02:44 _ilbot left 02:46 _ilbot joined 02:50 lue joined
TimToady ((it would seem post-headache branefuzz is not the best time to do design)) 02:51
japhb_ It is certainly desirable to make M-of-N combinators that when satisfied stop the (M+1)..N work-in-progress. 02:53
TimToady stop and/or suspend 02:54
after the first page of results :) 02:55
TimToady takes pity on wife, who pitiously and/or pitilessly asks him to cook dinner 02:57
02:59 raiph joined
japhb_ Sure. Though I was thinking rather of managing latency long tail. 03:01
(Without generating vast continuing load that is known to not affect results already decided._
03:04 rurban1 joined 03:09 rurban1 left 03:13 kaleem joined 03:29 fridim_ joined 03:48 kaleem left 03:49 guest313 joined 04:05 rurban1 joined 04:06 leont joined 04:10 rurban1 left 04:11 preflex_ joined, ChanServ sets mode: +v preflex_ 04:13 preflex left, preflex_ is now known as preflex 04:25 guest313 left 04:42 araujo left 04:55 xinming left 05:03 xinming joined 05:06 logie left 05:08 kurahaupo_ is now known as kurahaupo 05:12 logie joined 05:23 ponbiki left
kurahaupo tries to remember which bot does store-and-replay messaging (for the benefit of timezone outcasts) 05:27
TimToady preflex: help 05:34
preflex try 'help help' or see 'list' for available commands
TimToady preflex: help list
preflex list - show available plugins
TimToady preflex: list
preflex Botsnack: [botsnack]; Cdecl: [cdecl]; 8ball: [8ball]; excuses: [excuse]; Factoid: [+, -, ., ?, delete, get, store]; Help: [help, list]; Karma: [++, --, karma, karmabot, karmatop]; Nickometer: [nickometer]; Nickr: [nickr]; PlokiRE: [re]; Seen: [seen]; Sixst: [6st, ordinal]; Tell: [ask, clear-messages, messages, tell]; Rot13: [rot13]; Quote: [be, quote, remember]; WCalc: [calc, wcalc]; Version:
[version]; XSeen: [xseen]; ZCode: [zdec, zenc]
05:34 logie left
lue preflex: zenc test 05:37
preflex test
05:37 SamuraiJack__ joined 05:46 jercos left, jercos joined
moritz good morning 05:58
06:06 rurban1 joined 06:11 rurban1 left 06:16 BenGoldberg left 06:19 xinming left, xinming joined 06:36 [Sno]_ left 07:05 kaleem joined 07:07 rurban1 joined 07:10 SamuraiJack joined, SamuraiJack__ left 07:12 rurban1 left 07:15 daveec left, daveec joined 07:23 sivoais left 07:26 sivoais joined 07:29 araujo joined 07:33 fridim_ left 07:45 dmol joined 07:52 btyler left 07:53 darutoko joined
GlitchMr runsvdir -P /etc/service log: /run: line 2: cd: /data/svn/commitbit: No such file or directory ./run: line 2: cd: /data/svn/commitbit: No such file or directory 07:59
hm
08:08 rurban1 joined
moritz GlitchMr: pasting some random command line and error message without any context into the channel is neither interesting for us, nor a good base for discussion 08:09
08:09 zakharyas joined
GlitchMr From ps aux. 08:09
moritz ... on?
GlitchMr I'm not sure if service running on feather should have error messages in arguments. 08:10
moritz well, I'm sure
(that they shouldn't)
08:10 Celelibi left 08:12 thou left 08:13 rurban1 left, nnunley left
moritz GlitchMr: now that you found it, what are going to do with it? 08:14
08:41 FROGGS joined 08:50 salv0 joined 08:55 darutoko left 08:58 hummeleB1 left 09:09 rurban1 joined 09:13 rurban1 left 09:18 FROGGS[mobile] left 09:21 grondilu left 09:29 kivutar joined 09:32 xinming left 09:34 xinming joined 09:36 dakkar joined 10:01 darutoko joined 10:04 tgt joined
jnthn TimToady: Yes, we very much need a cancellation mechanism. 10:06
10:07 ponbiki joined
jnthn TimToady: Though, anyof should probably not use it. A combinator that explicitly races, along with a ..timeout one to factor out that very common idiom, would, though. 10:09
10:10 rurban1 joined
moritz A branch, a tag, and a reflog walk into a bar. The bartender says, "What is this, some sort of rebase?" 10:10
10:10 kurahaupo is now known as kurahaupo_mobile
arnsholt o/ 10:14
10:14 rurban1 left
moritz arnsh\olt 10:14
arnsholt ^_^
jnthn arnsh\o/t 10:15
FROGGS hi :o)
moritz FRO/GGS!
colomon \p 10:17
errr
moritz c\olomo/n 10:18
jnthn r: "\p"
camelia rakudo-parrot 75c45f: OUTPUT«===SORRY!=== Error while compiling /tmp/BsdHz5pnVm␤Unrecognized backslash sequence: '\p'␤at /tmp/BsdHz5pnVm:1␤------> "\⏏p"␤ expecting any of:␤ statement list␤ prefix or term␤ p…»
..rakudo-jvm 75c45f: OUTPUT«===SORRY!=== Error while compiling /tmp/ti0BmLVtsg␤Unrecognized backslash sequence: '\p'␤at /tmp/ti0BmLVtsg:1␤------> "\⏏p"␤ expecting any of:␤ statement list␤ prefix or term␤ pref…»
jnthn Thought so :)
10:18 kivutar left
FROGGS mo/ritz! 10:19
10:24 darutoko left, ponbiki left 10:33 darutoko joined, xenoterracide left 10:35 ponbiki joined
FROGGS jnthn: when storing modules in a sort-of database, do you have any wishes about precomp or extra information that should go along with it? 10:46
like, compiler version or something else?
jnthn FROGGS: compiler version is one good one, 'cus if that changes the pre-comp is invalidated 10:47
FROGGS for all backends?
jnthn Yup 10:48
FROGGS k
jnthn It's 'cus of the BS
You link against the setting.
FROGGS ahh, yes
jnthn Which links against the bootstrap and MOP, which in turn is against a particular NQP...
10:49 dmol left 10:52 fhelmberger joined 10:53 berekuk left
FROGGS lizmat: atm it tries site first: github.com/tadzik/panda/blob/maste...ler.pm#L20 10:56
and for CompUnitRepo.install I think panda just calls its .install with: %dist-info, $from-path, $to-path-without-prefix 10:58
or maybe it just passes the %dist-info along will to-be-installed files along the "subdirectory" they should appear in 10:59
dunno how we're going to support Makefiles btw
maybe we need some sort of standalone installer that can be invoked like ExtUtils 11:00
still, you need to supply (trustworthy) information about the distribution that is going to be installed 11:01
11:08 darutoko left 11:10 cibs left, rurban1 joined 11:15 ssutch left, rurban1 left 11:20 [Sno] joined 11:21 raiph left 11:22 cibs joined 11:40 dmol joined
timotimo ohai 11:44
11:55 hummeleB1 joined, pernatiy left 12:11 rurban1 joined 12:16 rurban1 left
timotimo i had a dream where some big news site showed a story about an open source developer from finland having the nickname "timotimo1" because obviously some horrible commoner has taken "timotimo" 12:27
moritz literally lols
jnthn lol 12:28
tadzik :D 12:31
timotimo huh, it occurs to me, that that may very well be an expression of my imposter syndrome 12:33
tadzik great. I got it and now everyone has it 12:34
imposters!
timotimo wow, vi hart's newest video about logarithms is pretty stunning
jnthn bbl & 12:38
lizmat FROGGS: wrt to standalone installer: that is what CompUnitRepo.install should do 12:43
note that the first parameter to .install is the *source* to be installed 12:44
*not* a filename
.install should check the pod for module name / auth / version info and fail if it can't find the data it needs to install the module 12:45
and create a pre-compiled version if it thinks that's necessary 12:46
FROGGS lizmat: what if a distribution ships other data than source code too? how should a command line tool know which dist/auth/ver it is? 12:52
lizmat hmmm... do you have examples ? 12:53
12:54 dmol left
lizmat just to be able to think of use cases 12:54
FROGGS example: perl6-debug ships a batch file, so you can actually type `perl6-debug`
there is no pod :/
so, an "ExtUtils::CompUnitRepo --install" would need a hash of information directly from e.g. panda 12:55
lizmat hmmm... maybe the source parameter could be a hash of some sort 12:56
FROGGS like when the CompUnitRepo wants to store the source location of the distribution, only panda knows where it got the dist from
feels a bit icky
lizmat I see panda as a thin client of CompUnitRepo 13:00
so we need to put any needed functionality somehow in an API in CompUnitRepo
maybe .install is a bit too simplistic 13:01
;-)
13:02 dmol joined
FROGGS depends on what it should install... it can either handle single files (of whatever type) or distros 13:04
lizmat is thinking 13:09
.oO( what are you sinking about? )
13:12 rurban1 joined
FROGGS *g* 13:13
german cost guard :o)
13:16 rurban1 left 13:17 ajr joined, ajr is now known as Guest74822, Guest74822 is now known as ajr_ 13:20 tgt left
lizmat ok, thought about it 13:32
.install is simple enough
it takes *one* file and installs that
because that is what the system needs to know when it wants to -use- it
the "database" in the top dir should only keep information that is needed to be able to *execute* the code 13:33
that database needs to be lean and mean 13:34
any additional introspection should be optional and probably live in a different database
so the oneliner should only be used to install a single file 13:35
if a distro needs to install multiple files, it should call the oneliner multiple times
FROGGS sure, but the installer needs to know all about the distro it belongs to 13:38
even if it is a one-file-distro
and especially if not
lizmat hmmm.. would we allow installation of pre-compiled bytecode ? 13:40
FROGGS why not?
lizmat I guess we would
FROGGS like perl6-debug compiles itself using a Makefile
lizmat indeed... so .install can't in that case find out what it is it's installting
FROGGS you would install the executable and the source then
maybe we would even install the needed things to rebuild it on demand 13:41
13:41 tgt joined
lizmat but we need to keep remembering this: CompUnitRepo is what is needed to be able to *find* code for execution 13:42
nothing more
and more specifically: find using the "use" command (and friends)
FROGGS it is not that easy
lizmat nothing else
finding code to be executed ? 13:43
FROGGS who is in charge to install other files that belong to a dist?
lizmat not the CompUnitRepo, is what I just decided and will put into the spec :-)
FROGGS I am thinking about File::ShareDir now btw
you do: use Foo::Bar(v1.0.2..*), and somehow this must match a File::SharDir lookup 13:44
so an installed module can ask: Where is my stuff?
lizmat indeed: a compunit should be able to tell where it came from 13:45
FROGGS I don't wanna have two installers, one for source code, and another for bytecode and other shared files
lizmat hmmm... 13:47
FROGGS as I may have mentioned: IMO the CompUnitRepo is in charge to install a distribution (including (compiled) stuff from installation process)
13:48 rindolf joined
lizmat so your saying, it's the CompUnitRepo's responsability to make sure that the database for looking up -use- files is as lean as possible 13:48
while granting it the possibility to have another database with much more information 13:49
Hmmm.. I would think that the information the CompUnitRepo's database would live in memory anyway
13:49 kivutar joined
lizmat perhaps even preload all @*INC CompUnitRepo's at startup, as to provide a "frozen" image of the outside during execution 13:50
13:50 kaleem left
lizmat so maybe we should free up whatever parameters we thought about with .install 13:51
because it's always going to be an "external" process doing the install
so maybe we should just spec what is going to be needed for ::Installation 13:53
aka, the "old" CPAN-like installation
hmmm.. but would we allow modules to be installed by something like dpkg in the same directory as being serviced by ::Installation? 13:54
I guess not, right ?
FROGGS lizmat: we don't need to freeze anything during execution for -use-, since that is compile time... 13:56
13:56 kaare_ joined
FROGGS question is: if you require something, should that take the newest database there is? I'd say so 13:57
lizmat indeed, a require should reload if it sees that the database in memory is out of date with the one on disk
FROGGS lizmat: I'd give dpkg its own tree with its own database (maybe even with a debiadn specific database format)
lizmat I guess even a -use- should do that 13:58
FROGGS so you newer install to the debian tree (aka vendor) using panda
never
never*
lizmat ah, ok
*phew*
FROGGS :o)
sorry, I am a bit distracted atm 13:59
lizmat so am I, actually
FROGGS the next thing is that we should spec that if a module installs a binary, it is not guaranteed that it will be the only version there is, and may get the version-number appended by default 14:00
our auth, for that matter 14:01
or just get a hash
14:01 tgt left 14:02 tgt joined
lizmat well, -use- will just ask the CompUnitRepo for candidates matching the given longname/ver/auth/from 14:02
so it's the CompUnitRepo's responsability to favour binary versions over source version or not
FROGGS I am not talking about use, more about executing a program
like, perl6-debug, panda, ...
p6doc 14:03
if I install three versions of perl6-debug, clearly the best one should be in PATH (if that bin dir happens to be in path)
but what is the best one if they all come from different auths?
lizmat in my view, that lies outside the responsability of CompUnitRepo, but in yours not, so I'm trying to adapt
FROGGS yeah, I am thinking about the whole installer thing, maybe CompUnitRepo is just one part of it :o) 14:04
lizmat I'm going to see sno this evening, at the Niederrhein.pm meeting
FROGGS (an important one of course :o)
lizmat well, as I said, I wanted to keep CompUnitRepo as simple as possible, to reduce RAM/CPU requirements 14:05
FROGGS yes, we don't need to blow that up
we can have other parts that care about the rest 14:06
lizmat otoh, loading a compilation unit may be such an intensive task, that we could actually be a lot fatter in CompUnitRepo before anybody starts to notice
still, I think leaner from the start is nearly always better
FROGGS yes
btw, I think we should split rip the install part out of the code that is active while locating modules to use them 14:08
s/split//
14:08 tgt left
FROGGS gah, my english is terrible :( 14:08
lizmat well, that's another reason I wanted to keep that simple :-) 14:09
so it wouldn't be a lot of code
.install could just be a stub that requires the necessary code and does the right thing 14:10
away for half an hour&
FROGGS jnthn: if we had a huge database of installed modules with a lot of information... I guess it would make sense to compile something for use/need/require? 14:13
14:13 rurban1 joined
FROGGS jnthn: so that we can just deserialize it and have a hash, only with the information we would need? 14:13
so we don't have to parse json or whatever at that point for example 14:14
14:15 leont left 14:17 rurban1 left 14:21 colomon left 14:26 iSlug joined 14:29 smash_ is now known as smash 14:32 PacoAir joined
jnthn FROGGS: Perhaps, yeah. I guess that we get to keep this format under our control, rather than it being spec? 14:37
FROGGS jnthn: that is an implementation detail, yes 14:38
it is even specific to a given module loader (aka CompUnitRepo)
jnthn ok 14:39
FROGGS but it would be nice if this format survives a rakudo upgrade/rebuild
hmmm 14:40
or we rebuild that cache when rebuilding racudo
rakudo
jnthn Well, or it detects it's out-dated on first "use" it has to process and re-builds then or something. 14:41
FROGGS that would work too
btw, I imagine that we put that CompUnitRepo in src/core along with the libraries.cfg in root dir, and include CompUnitRepo as the last thing in the setting 14:43
14:44 kaleem joined 14:47 btyler joined, bluescreen10 joined
lizmat FROGGS: agree with that last statement 14:48
FROGGS k 14:49
14:49 berekuk joined 14:55 Gwyxx joined
lizmat FROGGS: hmm... what's libraries.cfg ? 14:57
15:02 jnap joined 15:04 logie joined 15:06 kaleem left
dalek ecs: 0dd084d | (Elizabeth Mattijsen)++ | S03-operators.pod:
Small change in hash name to avoid confusion
 15:07
lizmat TimToady: been looking in the spec for the place where map { } is specified
I can't find it :-(
method map, no problem 15:08
hmmm. I guess S32/Containers:326 mentions it 15:09
15:13 darutoko joined, salv0 left, rurban1 joined 15:17 colomon joined 15:18 rurban1 left, nnunley joined 15:20 ajr_ left 15:21 ajr joined 15:22 ajr is now known as Guest52379, Guest52379 is now known as ajr_
FROGGS lizmat: libraries.cfg is the resident version of -I/--install 15:27
15:29 FROGGS left
dalek ecs: dabb21e | (Elizabeth Mattijsen)++ | S32-setting-library/Containers.pod:
Spec the listless map { }
 15:30
lizmat FROGGS: but why can't I find that file anywhere on my system ? 15:31
15:35 salv0 joined 15:39 ajr_ left, darutoko left 15:42 woolfy left, brrt joined 15:44 dmol left 15:45 scottp_ joined 15:47 Ulti_ joined 15:48 Juerd_ joined, pmichaud_ joined, masak_ joined, zakalwe_ joined, BooK joined, zakalwe_ left, zakalwe_ joined, Hor|zon_ joined 15:49 tgt joined 15:51 darutoko joined, sorear_ joined, [Sno] left 15:52 ivan``_ joined, scottp left, zakalwe left, tadzik left, pmichaud left, masak left, Juerd left, ivan`` left, BooK_ left, Ulti left, sjohnson left, Yappo__________ left, Hor|zon left 15:53 Juerd_ is now known as Juerd
lizmat Niederrhein.pm& 15:53
15:53 lizmat left, brrt left, salv0 left 15:56 thou joined 15:57 kaare_ left 15:59 prammer joined, jeffreykegler joined, sjohnson joined 16:00 Yappo__________ joined, tadzik joined 16:01 salv0 joined 16:03 fridim_ joined 16:06 kaare_ joined 16:09 sjohnson left, sjohnson joined 16:12 FROGGS[mobile] joined 16:14 rurban1 joined
FROGGS[mobile] lizmat: that file would be new when my changes get merged 16:15
16:19 rurban1 left
timotimo FROGGS[mobile]: are there long-term plans for rakudo-p5? 16:19
FROGGS[mobile] timotimo: yes, there are 16:20
timotimo you're doing S11 for now, and then back to p5?
FROGGS[mobile] that is my plan, yes
jnthn
.oO( When somebody isn't trying to distract him with even moar to do... :) )
 16:21
FROGGS[mobile] I hope that the test cicle gets shorter when we have moar
because after every change I often have to spectest, which takes 40min on four cores 16:22
timotimo ouch :(
FROGGS[mobile] and then at some point diakopter++ wants to have that parser 16:24
diakopter o_O
timotimo you mean the integration of perl5 into moarvm? 16:25
FROGGS[mobile] :o)
timotimo is any of you going to the chaos communication congress in hamburg this year per chance? 16:26
FROGGS[mobile] timotimo: yes, when the first step (not inlined code) wollen probably not need it
timotimo: no
jnthn The first step is gonna be knitted? :D 16:27
FROGGS[mobile] Bah!
mobile phone typing sucks :p
jnthn timotimo: Not me, my only two remaining events this year are Nordic Perl Workshop and Build Stuff
tadzik not LPW? 16:28
FROGGS[mobile] (Build Stuff)++
jnthn tadzik: No, 'fraid not.
tadzik: My work schedule is nuts. I'm only making NPW 'cus it's just a short train journey away. 16:29
tadzik I see
jnthn Woulda been fun to do LPW, though. It's always a great time.
Ah well, it's yearly. :)
tadzik I was considering going
FROGGS[mobile] what is LPW again?
jnthn FROGGS[mobile]: London ...
timotimo london perl workshop
tadzik that would've been like 12th flight this month though, I'm starting to hate those :P
FROGGS[mobile] ahh yes
jnthn tadzik: Yes, while I technically could squeeze it in, I'm feeling kinda the same about travel :) 16:30
FROGGS[mobile] I always hate flighs
diakopter EMOARFLIGHTSPLEASE
tadzik I miss trains. Trains are so fun 16:31
jnthn Yeah, I do train when I can.
tadzik hah
I see what you did there :P
jnthn But not always practical.
timotimo so what you're trying to say is ... "i like trains"?
jnthn :D
tadzik :D
zzzzzzumm
jnthn hah, I was thiking about "I like trains" :)
tadzik now diakopter gets that joke too :P 16:32
diakopter no, llama, no
16:40 dmol joined 16:42 kaleem joined, FROGGS joined, FROGGS[mobile] left
timotimo now can't get the thought of obama singing "i like change" out of his head 16:42
tadzik imagines obama with his head opening like that of canadians in South Park, so he can sing out of it 16:43
16:44 zakharyas left
diakopter PIE FLAVOR 16:45
16:45 raiph joined
tadzik isn't that Flavour? :P 16:46
diakopter FEED ME PAPER 16:47
THERE'S NO DOG THERE
I AM YOUR SANDWICH
TimToady and you pronounced PIE wrong
16:48 fhelmberger_ joined
diakopter oh wait, this isn't a camera 16:48
WHY WAIT 16:50
16:51 berekuk left
diakopter hello mine turtle. hello. 16:51
16:51 fhelmberger left
diakopter no because I work at teh muffin factory 16:51
tadzik funny, it always sounded like "ze muffin factory" to me 16:53
16:53 fhelmberger_ left
TimToady shakes his head for various reasons 16:54
FROGGS thinks that music is one of the best 16:55
moritz or drying your hair!
diakopter I'm gonna do an internet ... MUSIC
moritz ok, my hair isn't long enough for that to make any sense
diakopter 3p6gXgiVEzY 16:56
FROGGS hehe 16:59
moritz: mine almost is by now :o)
16:59 ajr joined 17:00 ajr is now known as Guest8089
TimToady I believe the DSM-5 has a category for this channel...probably several categories... 17:00
17:00 Guest8089 is now known as ajr_ 17:04 GlitchMr left
timotimo lamecia pls 17:07
jnthn Did we start on advent calendar signups yet? 17:09
tadzik yes
hurry up, while there are still some slots :) 17:10
jnthn tries to find it
Oh...there it is
uh, yeah, hot competition for slots here :P 17:11
jnthn wonders what he's done in the last year and could talk about... :)
FROGGS humm 17:12
I should probably write about S11
and v5?
17:12 kaleem left
jnthn Good ideas. 17:12
I guess I could do one about JVM.
As in, Rakudo on JVM 17:13
FROGGS and parallism
jnthn Yes, that too
FROGGS and one about moar perhaps
dalek : 948d9d4 | jonathan++ | misc/perl6advent-2013/schedule:
Take three slots.
 17:15
17:16 fhelmberger joined
jnthn FROGGS: There's moar moar hackers than me, so I think maybe somebody else might pick that up... 17:16
FROGGS yeah, maybe somebody does :o)
17:16 GlitchMr joined, [Sno] joined
jnthn If we were to put a Moar one towards the end of December we stand a good chance of having Rakudo on MoarVM running stuff by it. 17:16
17:16 xinming left 17:17 xinming joined
jnthn oh, I shoulda maybe put my name on those entries, so folks know I'm on the hook for 'em... 17:17
dalek : b64fb21 | jonathan++ | misc/perl6advent-2013/schedule:
Mark who's doing my days.
 17:18
kudo-star-daily: e25a847 | coke++ | log/ (5 files):
today (automated commit)
 17:19
rl6-roast-data: 483c822 | coke++ | / (5 files):
today (automated commit)
17:20 Celelaptop joined
FROGGS jnthn: I think that "we" are on a good trak about perl6-m :o) 17:20
17:20 rurban1 joined
jnthn I may well get an hour or two for it tonight. We'll see. 17:21
Got ahead on the $dayjob stuff yesterday, but then lost a chunk of today to anesthetic and drilling... 17:22
:/
[Coke] Ow.
FROGGS uhh :o/
youtube: "Larry Wall - That Goes Without Saying (or Does It?)"++ 17:24
dalek ecs: c5b7d73 | larry++ | S17-concurrency.pod:
winner ~~ s/nobody/wait 0/
 17:25
17:31 jnap left
jnthn Hmm. 17:32
Guess that works.
TimToady the problem with using anyof is that the degenerate case of timeout isn't guaranteed to poll
jnthn Poll? 17:33
TimToady anyof($p1, $timer) can just give you the timer promise if it's 0 without checking $p1 17:34
jnthn Ah, true.
TimToady whereas winner's 'wait 0' guarantees to be checked last 17:35
jnthn Well, it doesn't give you anything per se.
anyof is just kept with True and it's up to you to go look at what that means.
TimToady right, but it's a funny little race condition there
might be harmless, might not 17:36
jnthn The fact it doesn't give you back a promise and makes you decide what to check is part of what helps make it most likely harmless. :)
17:36 fhelmberger left
TimToady I thought it does give back a promise, just a different one 17:36
jnthn Yeah, sorry, misspoke 17:37
It gives you one, but when it's kept it's just kept with True.
17:37 fhelmberger joined
TimToady anyway, we need something like winner if we're going to set up gladiatorial winner semantics 17:37
jnthn As opposed to giving you the promise/value that made it be considered as kept. 17:38
TimToady which winner can do, but is not required to
btw, I'm assuming we're demoting winner from statement_control to a mere (but fancy) term 17:39
jnthn In general, though, I think anyof/allof are useful in the cases where you want to do something afterwards which is not based on the results of the input promises.
As in, "once once of them has done it's thing then go on and do..."
Typically you're probably doing something else with the rsult. 17:40
So they're primarily useful for synchronization. rather than racing to a result like winner is good for.
17:40 ssutch joined
dalek d: d075830 | larry++ | STD.pm6:
demote winner/combine to term
 17:40
17:41 kaare_ left, fhelmberger left
dalek : 5f9d992 | (Tobias Leich)++ | misc/perl6advent-2013/schedule:
took two slots
 17:43
TimToady I think we don't need special syntax to implement gladiation; the winning entity has sufficient information to optionally kill the losers if the losers provide a passable (pun intended) api.
17:46 jnap joined
TimToady ESL lesson: "Elves are immortal, but passable." 17:46
17:48 darutoko left
huf TimToady: you'd need a very large intestine... 17:48
jnthn We can write a timeout combinator for the simple cases too :) 17:49
TimToady revokes huf's pass.
geekosaur well, we do use double colons for objects :p
TimToady jnthn: treu
huf TimToady: so i ... i cannot pass?
TimToady I have a bridge to sell you. It's not all it's cracked up to be... 17:50
jnthn *lol*
17:50 stevan__ joined
huf heh :) 17:51
TimToady you fell for it, did you?
17:53 stevan_ left 17:58 jeffreykegler left 18:01 colomon left
dalek ast: 259ed8d | coke++ | test_summary:
FIX jvm fudgespec skip
 18:01
18:03 dakkar left 18:04 jeff_s2 joined 18:05 jeff_s1 left 18:09 baest left, baest_ joined 18:10 colomon joined 18:15 mattp_ left
FROGGS .oO( say fast "fudgespec skip, fudgespec skip, fudgespec skip, fudgespec skip, fudgespec skip" ) 18:18
18:18 aindilis` joined, aindilis left
TimToady
.oO( can I just think it fast? )
 18:18
FROGGS if you imagine saying it :o) 18:19
TimToady I imagined saying it fast. :P
timotimo goalpost movement again, he? 18:20
eh?
TimToady just generalized the goalpost to a finish line :)
diakopter wheh 18:21
TimToady heh, pick two. 18:22
timotimo is not sure any more what exactly he has to change
if i change winner to a term, i can keep more, done and later (albeit renamed) as statement_modifier?
TimToady only winner and combine need to be terms 18:23
so we don't have to keep putting 'do' in front of them
timotimo i don't know about combine yet 18:24
what does that do?
is that for combining multiple channels into one?
TimToady combinates taps rather than promises/channels
timotimo ah, ok 18:25
TimToady and a timeout probably doesn't make sense for it, unless it does.
jnthn combine basically sets up an actor that is also tappable.
TimToady but there's no polling, so 'wait 0' makes no sense 18:26
jnthn And that taps many other supplies
18:26 mattp_ joined
jnthn The actor-ness needed 'cus you don't get to control when the supplies you tap throw stuff at you. 18:26
But you do want to write code without having to lock everywhere.
shop & 18:28
TimToady channels are locking in the sense that if you exert any queing backpressure, the throwers have to stop throwing until you catch up
timotimo jnthn: do you expect me to implement combine as well until your talk? :)
TimToady it's probably easier than winner :)
you don't worry about the scheduling, only composition of callbacks 18:29
18:30 jeff_s2 left
jnthn timotimo: Well, it's only syntaxing the "on" that I already implemented... 18:30
I don't think it's undergone any big semantic changes.
TimToady not really, though thinking about changing tap to take one positional and two named args rather than three positionals 18:31
timotimo i don't know what the on thing is you're talking about :P
18:31 jeff_s1 joined
timotimo but i'll dive into my code again 18:31
TimToady so the named args can look like they do in combine
timotimo: it's just like .then on a promise, only more so :) 18:32
timotimo ah
that doesn't seem like a syntaxing thing to me. more like a method call
jnthn TimToady: I think I'm good with the named args approach 18:33
TimToady still not sure I want to call failure 'fail' though, since that conflicts with the builtin function
jnthn Well, I named it "fail" on purpose to say "you're shoving an exception somewhere else" 18:34
It felt strangely consistent.
18:34 lizmat joined
jnthn really shop & 18:35
TimToady it is, but it prevents us from ever defining winner/combine like we do class bodies: run the non-declarative statements, then take the declarations and compose something
in fact, winner/combine should probably be called declarators 18:36
ajr_ I just noticed the update to S32, about listless map {} - is that an especially lazy form of the instruction? 18:38
timotimo no, it's just as lazy as map 0..* {} 18:39
lizmat map {}, 0..* actually
timotimo er, yes
18:40 pecastro_ left
lizmat but hopefully without the overhead of an actual counter going from 0 to Inf 18:40
18:40 pecastro joined
timotimo yup 18:41
that wouldn't be much, but still.
18:44 [Sno] left
timotimo chased the renames and turning winner into a term 18:44
18:44 ggoebel joined, [Sno] joined
TimToady is turning them into declarators now :) 18:44
timotimo oh, but wait needs to take an xblock now, too
TimToady yep
timotimo do declarators already exist?
TimToady hang on
diakopter all the things to declarators 18:45
timotimo there are many proto tokens something_declarator 18:46
dalek d: 87ac055 | larry++ | STD.pm6:
add combinator_declarator
timotimo ah, indeed, there's a tokne declarator
18:46 masak_ is now known as masak
masak oh hai, #perl6 18:46
FROGGS masak: \o/
diakopter TimToady: but the marcos 18:47
timotimo now we have catch and CATCH?
diakopter ermasak, macros
masak :P
FROGGS masak: welcome back :o)
masak thank you :)
moritz masakro!
TimToady catch is only for use inside 'combine'
18:48 PZt left
timotimo hi masak :) 18:48
TimToady it's better than 'fail', and since the failure comes in as $_, it might end up looking a lot like a CATCH on the inside anyway
masak "I love it when a unification comes together"
timotimo how's life? how's t4? how's that super cool trick with feeds to implement your blog in only 5 lines of code?
masak timotimo: thank you for asking. 18:49
18:49 PZt joined
masak timotimo: (a) life, of the kind that doesn't happen on #perl6 but happens out in Cartesian space, with other people, is catching up with me. it's great, but leads to little time for #perl6. 18:50
timotimo: (b) t4 is standing still right now. it's at the top of my list, though.
timotimo: (c) that super cool trick is still alive and well. stand by. 18:51
man, I nees to backlog.
need*
timotimo sounds good, glad to hear :) 18:52
masak how does 'done' as a combinator declarator rhyme with 'done;' as a Test.pm sub? 18:53
18:55 prevost joined
dalek kudo/WINNER: 5ba95a7 | (Timo Paulssen)++ | src/vm/jvm/core/asyncops.pm:
a single named positional is fine for done/more, too.
 18:56
kudo/WINNER: dcc8b05 | (Timo Paulssen)++ | src/ (5 files):
chase lots of changes to winner:

  - all the statements are now in the new class "combinator_declarator"
  - "later" is now called "wait" and always takes a parameter.
  - "combine parses, but throws NYI for now.
FROGGS tadzik: there is a panda PR 18:59
timotimo aaaand it doesn't compile 19:02
moritz ♪ the winner took it all ♫
lizmat masak: I guess it will confuse the fudger a lot
masak not sure the fudger cares about 'done'
19:02 rindolf left
masak but I was thinking, how does the Perl 6 parser distinguish? 19:03
jnthn r: sub when() { }; when;
FROGGS you'd have to use done(), right?
camelia ( no output )
jnthn Apparently it can there...
dalek ecs: d0fb631 | larry++ | S17-concurrency.pod:
s/fail/catch/ on receiving end, tap has named args
timotimo i'm not sure why it's wrong ... 19:04
diakopter masak: I'm afraid you'd never finish backlogging... it's grown tremendously lately
TimToady is tired of collisions with Test.pm
masak "ypu'd have to use done()" sounds like a rather big change to the existing spectest suite.
TimToady just a perl -i :) 19:05
masak and to the ecosystem in general.
timotimo why would winner $self { more * { $_ }\ndone * { last } } give TTIAR?
FROGGS masak: I didn't say that I would like or recommend it :o)
masak big breaking change.
TimToady timotimo: shouldn't, if that's parsed as a statement list
jnthn masak: I think my example of when; above working shows that this may also not cause a problem. 19:06
timotimo i implemented winner as <sym> <xblock> and more and friends as <sym> <xblock(1)>
and combinator_declarator is inside declarator
masak jnthn: indeed.
r: sub when() { }; when 19:07
camelia ( no output )
FROGGS rp: sub say () { say 42 }; say
TimToady it could be forced with a dynvar
camelia rakudo-parrot 75c45f: OUTPUT«===SORRY!===␤CHECK FAILED:␤Calling 'say' will never work with argument types (int) (lines 1, 1)␤ Expected: :()␤»
TimToady most keywords require a space after, so done; is fine 19:08
fail "foo"; # not so much
timotimo sadly, rxtrace is immensely unhelpful 19:09
19:09 diakopter is now known as TimToTimTo
FROGGS hmmm, I feel like panda needs to adjust its shebang line when it installs itself (or other executables for that matter) 19:10
TimToady std: my $self; winner $self { more * { $_ }␤done * { last } } 19:11
camelia std d075830: OUTPUT«ok 00:01 125m␤»
FROGGS maybe it should even create panda[-p|-j] variants
japhb FROGGS: +!
FROGGS: +1
FROGGS japhb: to which statement? :o)
timotimo could it be declarator in rakudo is significantly different from the one in std? 19:12
japhb There are some times when I want to install a module to All The Backends, but other times that I have a module that only works on a limited number of backends (and will fail tests on the others)
TimToady did you add <combinator_declarator> as an alternate in declarator?
timotimo yes
FROGGS hmmm 19:13
japhb Also, I am also +1 to adjusting shebang lines to match the backend a script is being "compiled" for.
TimToady and added the syntax category?
and didn't misspell anything?
timotimo the ... syntax category?
FROGGS japhb: well, that would introduce a problem actually 19:14
timotimo i have a proto token if you mean that
japhb_ timotimo: See the beginning of TimToady's STD patch
japhb FROGGS, oh?
FROGGS japhb: maybe I can only solve it by S11 itself
19:14 SamuraiJack left
TimToady timotimo: yeah, the first green bit of github.com/perl6/std/commit/87ac055582 19:14
19:14 TimToTimTo is now known as diakopter
timotimo no, i did not do such a thing, i thought that was an std thing 19:14
TimToady it's a Perl 6 thing :) 19:15
FROGGS japhb: ahh, not sure, need to think more careful
japhb Thinking: It's dangerous.
timotimo huh, i don't see an equivalent in rakudo's grammar
TimToady wonders why the lexer generator didn't complain about a missing <combinator_declarator> method...
FROGGS japhb: panda should get at least a --backends option 19:16
japhb YES
19:16 sqirrel joined
TimToady timotimo: it's usually called a proto rule or some such in rakudo 19:16
FROGGS otherwise you could only install modules for your default backend
timotimo i have a proto token combinator_declarator { <...> } 19:17
lizmat japhb: like p5, I think you should only be able to install for that version of perl you're currently running with 19:18
ggoebel PO 19:19
japhb lizmat: Sure, but in the current world, you can have perl6-p, perl6-j, perl6-m all in your path, and perl6 symlinked to one of them. Panda should get the same treatment.
FROGGS good point, so I#d have to: `which panda` and then `perl6-p /path/to/panda` 19:20
19:20 fhelmberger joined
ggoebel sorry... cat typing 19:20
diakopter meow
FROGGS cat++
19:20 prevost left
TimToady duck typing, monkey typing--what's cat typing? ;) 19:21
19:21 xinming left
TimToady category theory, I suppose... 19:21
19:21 xinming joined
geekosaur cats do what they damned well please, therefore... 19:22
huf "cat" and "ego" go well together 19:24
and if you dont pay attention to that ego, things can get "gory"
so this seems to fit
masak * lizmat just realized that (+=) for atomic updates is bogus 19:25
TimToady timotimo: do you need to insert <.end_keyword> to require the ' ' after the keyword?
timotimo oh, i may
masak lizmat: so, how about we revert 71fd6d3abb1c5b360eae4d424a342bc6a0cb03da?
TimToady otherwise you tie with <ident>
lizmat masak: no, not completely 19:27
diakopter lizmat: you didn't reply to my question about that 19:28
timotimo i'm still getting TTIAR
diakopter how is it bogus
TimToady wonders why we don't just have a cas($var,$newvalue) function or macro
diakopter lizmat: er, I didn't see a reply
timotimo oh, i may have to remove the whitespace between <sym> and <.end_keyword>
TimToady or make it a token, yes
masak lizmat: well, it was worth a shot.
lizmat why was it worth a shot? 19:29
masak asking for it to be reverted.
timotimo that seems to have fixed it, yay
i spoke too soon
masak lizmat: oh, "why", not "what"...
lizmat but why would *you* want to have it reverted, and are ok with it staying ?
why are you Dutch so suspicious? 19:30
masak I don't believe that there is a sane connection between set operations and (+=)
PerlJam The capitalization in S17:807 makes me think it should be ACS instead of CAS ... unless this is another one of those french things like UTC
synopsebot Link: perlcabal.org/syn/S17.html#line_807
lizmat
.oO( why do you want to know ? )
TimToady No true Scotsman is Dutch.
masak I don't believe we should have operators for CAS things. at least not in core.
diakopter but why wouldn't the operator just do the right thing for cas variables 19:31
TimToady so what is += supposed to return?
dalek ecs: 73f60f7 | (Elizabeth Mattijsen)++ | S17-concurrency.pod:
Remove (+=) and friends cas operators
lizmat yes, "is cas" makes much more sense 19:32
TimToady CAS is all about known absolute values, not relative values
diakopter why was the += bogus?
TimToady ^^
lizmat because then it would handle things hopefully correctly
because CAS should be hooked to a variable, not to an operation
TimToady dislikes 'is cas' changing the return semantics of $var = $newvalue
lizmat you want *all* accesses to a var to be CAS, not just some 19:33
diakopter that's not true
but you do want the defaults to be cas usually
jnthn I think cas has to be about operations, not about variables...
TimToady +1
it's a very low-level operation 19:34
there's very little point in trying to abstract it
jnthn Well, also, it's kinda transaction-y
Meaning you need to know the bounds of yoru transaction.
masak I totally don't see why CAS can't just be, you know, a sub.
lizmat " is cas" bounds the transaction to updates to a single variable ? 19:35
masak trying to *design* it to be something else, without trying it practically in Perl 6, is premature to say the least.
TimToady lizmat: what is the lifetime of that variable? what if you return it?
masak the only reason we should spec a thing like that is if we like spec that turns out wrong over time.
lizmat TimToady: I see "is cas" hanging off of the containerdescriptor, like "is default" 19:36
diakopter masak: it can't be a sub because the compiler needs to know how to get at the variables' memory directly
TimToady sentences everyone who is confused to reread en.wikipedia.org/wiki/Compare-and-swap 19:37
masak diakopter: that sounds like 'is rw' semantics to me.
TimToady that's why I said it might be a macro even
or a built-in, I suppose I should say
19:37 kurahaupo_mobile left
jnthn masak: I did it as a sub in the spec, fwiw 19:37
TimToady it can certainly parse like a sub :)
masak I suppose I meant "look like a sub". it's fine if it's a built-in.
diakopter jnthn: but that sub in the spec makes no sense 19:38
jnthn diakopter: How so? :)
diakopter: I didn't put anything into S17 that I didn't already know how to implement, or already had implemented :P
TimToady diakopter: he said "something like", so it's fine :P
diakopter jnthn: I'm thinking of the one that had the + operation in it 19:39
jnthn diakopter: cas($a, { $_ + 1) # like this?
uh, but with the }
diakopter yes 19:40
how does the compiler make that cas
TimToady I don't see anything like that at S17:816
synopsebot Link: perlcabal.org/syn/S17.html#line_816
masak "using internals"
jnthn diakopter: It doesn't need to, given $a is a scalar container, so we operate on updating its contents.
timotimo i think it's parsing winner, then starts parsing an expression and then gets surprised by the {
jnthn diakopter: But it's just a generalization of the usual CAS pattern.
timotimo how do i make that work?
diakopter jnthn: but how does the read of the prior value work with the + 19:41
19:41 kivutar_ joined, kivutar_ left
diakopter (where does it inject the atomic cpu instruction) 19:41
TimToady timotimo: s/block/xblock/ everywhere?
lizmat I always interpreted S17:816 as pseudocode, to get the point across 19:42
synopsebot Link: perlcabal.org/syn/S17.html#line_816
19:42 colomon left
FROGGS why does `time perl6-p panda` take 6.3s? 19:42
jnthn loop { my \orig = nqp::decont($a); my \new = nqp::decont(the_block(orig)); last if cas_scalar($attr, new, orig) == orig; }
timotimo TimToady: yes, i did that
TimToady FROGGS: maybe it's calling out to perl6-j :)
FROGGS TimToady: no, that'd only need 4.7s :o( 19:43
diakopter jnthn: okay, but that eliminates the possibility of using the increment atomic op
or the add atomic op
TimToady that's a matter for the optimizer 19:44
jnthn diakopter: Oh...then the real thing at issue is I wrote a crappy example that's better done with something else :)
FROGGS the timomizer?
diakopter jnthn: well, I'm sure it's possible to optimize to incr or add, but I can't see how yet
TimToady maybe we should switch to the CAS style that returns success or failure 19:45
19:45 woolfy joined
TimToady rather than doing our own comparison 19:45
jnthn The thing about all of this is that I see cas as a *low level* thing. 19:46
TimToady metoo
if an architecture needs a different primitive, we may be talking about conditional compilation
jnthn The sugar form I suggested is just abstracting a common "read an immutable data structure, make a new one incorporating the change, and try to "commit" the reference update" pattern
TimToady but most modern architectures support CAS now 19:47
in some form or other
albeit with a rather limited number of bits in some cases 19:48
CAS on an int32 is a much different matter than CAS on an Int
the latter is likely close to impossible without going toward STM 19:49
japhb_ ... and double-wide CAS
TimToady still limited
diakopter we have to do our own comparison on the CAS on the Int, and are updating pointers
(and definitely can't do the atomic incr or add)
TimToady the transaction becomes much larger
timotimo TimToady: could there be a reason why combinator_declarator isn't the right category for something that takes an xblock?
FROGGS ahh, the slow thing is probably that it checks for the first writeable diretory and does other setup tasks
lizmat couldn't CAS just not work on an container descriptor attribute that gets incremented on every update ?
japhb_ ISTR that PowerPC has a different primitive than CAS that it prefers, but I don't recall what it is offhand.
timotimo because it doesn't seem like the xblock will terminate the <EXPR> at the right moment for a block to be acceptable 19:50
TimToady it shouldn't be any different from an ordinary xblock in a statement list in that regard
japhb_ wonders how many canon_path calls get done during panda startup 19:51
TimToady but maybe rakudo adds more semantics to xblock that are interfering?
lue I'm assuming compiling r-m, I should expect a bunch of "WARNING: nominal type check NYI" compiling CORE.setting, right? :)
TimToady STD is just a parser, after all
timotimo lue: yes
lizmat jnthn, diakopter: couldn't CAS just not work on an container descriptor attribute that gets incremented on every update ?
diakopter I don't know about such attributes
19:52 fhelmberger left
japhb_ Actually FROGGS: I'd lay some odds it's either A) you're running an not-yet-installed panda, so no modules are compiled, or B) the cost of parsing the module list JSON 19:52
timotimo TimToady: would it be terrible if i turned it into a term for the time being? because ISTR that worked
19:52 kurahaupo joined
FROGGS japhb: it is an installed panda 19:52
19:52 fhelmberger joined
TimToady shrugs 19:52
lizmat two threads update the same variable: each one checks the value of the container attribute before and after the change is completed:
timotimo i'll try that again then. 19:53
diakopter lizmat: well the low-level cas operation returns the orig value
TimToady lizmat: you're starting to talk STM, not CAS
diakopter so it doesn't/can't check it afterwards
lizmat TimToady: perhaps, I still like STM :-)
timotimo though i think i can only change winner and combine into term and leave the others statement_control again
because done, more, catch and wait wouldn't be wanted in expressions inside winner/combine blocks 19:54
TimToady is fine for now
19:57 fhelmberger left
diakopter lizmat: the low-level cas operation (probably repr "method^Wsomething less inflammatory") returns the original, so that's the value that's checked to see whether it was successful 19:59
(whether it's checked within the op or not - either way) 20:00
dalek kudo/nom: a7ae9a9 | (Elizabeth Mattijsen)++ | docs/release_guide.pod:
A friendly reminder that we have a release soon
 20:01
diakopter lizmat: oh now I see what you were suggesting with the incremented version number; that's a form of a lock 20:02
lizmat indeed 20:03
timotimo i can't seem to get it to work at all right now :o
FROGGS japhb: 0.6s slurping a 35kb file, 4.5s spent on from-json for that file 20:05
20:05 dwarring joined
diakopter jnthn: OHHH you said "reference update" 20:05
I thought we were talking about integer instructions.. didn't realize we were talking bout boxed thingsies 20:06
jnthn diakopter: oh!
:)
diakopter jnthn: so yes, clearly no possibility for xincr or xadd there
jnthn diakopter: Yeah, I shoulda spotted the mis-understanding when you said that. Sorry. Not at my sharpest today... 20:07
20:07 sqirrel left
japhb FROGGS, yeah, both of those numbers are horrid. 20:07
That's why github.com/japhb/perl6-bench/blob/...parse-json
diakopter jnthn: however, my question then is about primitive vars.. how to use the special syntax to do atomic incr/add 20:08
jnthn diakopter: Yeah, I didn't try to attach either of those. :)
diakopter attach? 20:09
attack?
jnthn uh, attack 20:10
wtf :)
diakopter i like trainses
lizmat
.oO( until your car detaches )
FROGGS japhb: I see
diakopter lizmat: it's an allusion to the asdfmovie 1-6 where the kid who likes trains get smushed by trains often 20:11
lizmat fortunately we only get smashed
by trains of thought here
masak is on a train
20:12 dave1c_ joined
timotimo wtf is this i don't even 20:12
lizmat get cut short ? 20:13
20:13 iSlug left
timotimo tries putting the rules someplace else 20:13
dalek d: 8370f3f | larry++ | STD.pm6:
undo 87ac0555827223eec4da40c72dfeab90b996f525
20:14 dave1c left
timotimo oh 20:14
did my plight inspire pity in you?
masak TimToady: when you undo commits, do you perchance use `git revert`? 20:15
20:16 xenoterracide joined 20:17 kurahaupo left
timotimo i made it work >_< 20:18
it was about a missing <.ws>
i'm such a derp sometimes.
lizmat r: (1).map: start {...} # strange failure in the bowels of reification
camelia rakudo-parrot 75c45f: OUTPUT«===SORRY!=== Error while compiling /tmp/lvCNQE7osl␤Undeclared routine:␤ start used at line 1. Did you mean '&spurt', '&sqrt', '&sort'?␤␤»
..rakudo-jvm 75c45f: OUTPUT«Unhandled exception: Method 'count' not found for invocant of class 'Promise'␤ in (gen/jvm/CORE.setting:7210)␤ in reify (gen/jvm/CORE.setting)␤ in (gen/jvm/CORE.setting:7122)␤ in reify (gen/jvm/CORE.setting)␤ in gimme (gen/jvm/CORE.setting:…»
timotimo r: (1).map: do start {...} 20:19
camelia rakudo-jvm 75c45f: OUTPUT«Unhandled exception: Method 'count' not found for invocant of class 'Promise'␤ in (gen/jvm/CORE.setting:7210)␤ in reify (gen/jvm/CORE.setting)␤ in (gen/jvm/CORE.setting:7122)␤ in reify (gen/jvm/CORE.setting)␤ in gimme (gen/jvm/CORE.setting:…»
..rakudo-parrot 75c45f: OUTPUT«===SORRY!=== Error while compiling /tmp/tvuzJQIX9y␤Undeclared routine:␤ start used at line 1. Did you mean '&spurt', '&sqrt', '&sort'?␤␤»
lizmat it's a runtime error, it compiles ok
timotimo ah 20:20
i see that now
20:20 spider-mario joined
jnthn lizmat: It's the same as if you pass map anything that's not a block. 20:21
I'm surprised it lets you pass a non-callable at all..
lizmat argh
of course
timotimo oh duh :D 20:22
i'm not the only derp around today
jnthn We can all derp together!
lizmat j: my $seen; await( (1..3).map: { start { $seen++ for ^1000 }}); say $seen 20:23
camelia rakudo-jvm 75c45f: OUTPUT«1379␤»
timotimo depr yourself together, man!
lizmat j: my $seen; await( (1..3).map: { start { $seen++ for ^1000 }}); say $seen
camelia rakudo-jvm 75c45f: OUTPUT«3000␤»
lizmat j: my $seen; await( (1..3).map: { start { $seen++ for ^1000 }}); say $seen
camelia rakudo-jvm 75c45f: OUTPUT«2663␤»
lizmat j: my $seen; await( (1..3).map: { start { $seen++ for ^1000 }}); say $seen
camelia rakudo-jvm 75c45f: OUTPUT«3000␤»
lizmat this worries me
timotimo your point being? :)
lizmat j: my $seen; await( (1..3).map: { start { $seen++ for ^1000 }}); say $seen
camelia rakudo-jvm 75c45f: OUTPUT«1792␤»
jnthn lizmat: *sigh*
Yes, I know I owe you a blog post on why. 20:24
lizmat I want to be able to say something about $seen to have it always be 3000
timotimo by putting winner and combine as combinator_declarators, i get Unexpected block in infix position again!
whyyyyyyy~
it worked so flawlessly as statement_control :( 20:25
Util #ps in 5m
timotimo parsings are hard :( 20:26
lizmat jnthn: I think I know why, but I'd like to see that blog post anyway
timotimo it would be cool if target=parse could dump the innermost $/ before an error is thrown 20:27
lizmat it's just that I'm *very* worried we wind up in Perl 5.005 threads land :-(
20:28 colomon joined
timotimo lizmat: parallelism in perl6 relies on composable thingies, which i don't think shared state counts as 20:33
lizmat $seen++ seems pretty composable to me 20:34
but really, this is about perception and previous experiences
people will come to Perl6 and see promises, and start to use them like above 20:35
and things will not always work
and won't generate any errors either
timotimo hrm 20:36
TimToady why shouldn't reference to an unprotected external lexical cause an error?
timotimo i'd say you could use a KeyReducer for this case :P
lizmat it should!
timotimo because that's totally composable :)
lizmat but it doesn't
TimToady I've said many times that lexicals outside of threads should have extra constraints 20:37
timotimo that's acceptable.
lizmat outside of threads?
TimToady and that's the main reason almost everything is lexical in the first place, to automatically get thread-local storage for any internal lexicals
lizmat so, every container remembers in which thread they're made 20:38
and if they're changed from another thread, they should bomb?
TimToady various semantic approaches are possible, once we notice the problem 20:39
timotimo how do we notice if the user protected the lexicals with some locking scheme that we can't immediately see?
TimToady we can force them to declare it so
jnthn The challenge is distinguishing the cases we do want to allow it, like in combine, because we know that is safe. 20:40
timotimo my $foo is protected_by($lock)?
jnthn otoh, that's syntactic now :) 20:41
TimToady is inclined to make outside lexicals simply not visible to the inside of a start
diakopter TimToady: but what about all the other lvalues, like array and hash slots
TimToady s/lexicals/variables/
TimToady still likes single-owner semantics too 20:42
but I recognize it's fraught
diakopter -_-
lizmat hmmm.... so a thread/promise starts with a fresh lexical stack
diakopter but closures are essential in many cases.. 20:43
TimToady well, or just skips to SETTING
lizmat indeed
skip to SETTING
TimToady for variables
it's just an idea
lizmat I guess that would stop all of the accidental accesses
I like the idea
diakopter okay, but you can still pass a closure value to one of these blocks and it'll still have access to its lexical scope
lizmat it forces people to think about cocurrency
yes, you still can 20:44
TimToady anyway, I'm certainly not against imposing some kind of discipline on outer shared state
dalek kudo/WINNER: 7213c53 | (Timo Paulssen)++ | src/Perl6/ (2 files):
the previous attempt was intent on TTIAR'ing :(
kudo/nom: b774722 | moritz++ | docs/ChangeLog:
[doc] extend ChangeLog a bit
lizmat I wonder whether also keeping the creation thread ID with each container 20:45
timotimo yay, a thing i made got mentioned in the changelog \o/
lizmat and checking that on update, wouldn't fix the closure issue
jnthn Maybe it's more about scope ownership.
"Who created the invocation record"
lizmat but scopes can leak, so we need to have something to catch that
TimToady doubts there is a perfect solution (outside of making everything immutable) 20:46
but we can certainly decrease the likelihood of erroneous code
moritz and making everything immutable isn't a perfect solution either :-)
lizmat indeed, that is my main concern: code that doesn't complain, works ok most of the time, until we get a load on the system 20:47
jnthn lizmat: That's kinda what I was trying to address. If a lookup in the thread running something crosses into a invocation record created by another thread to do a lookup...
TimToady we can at least complain bitterly if we notice such a thing happening 20:48
lizmat but wouldn't you need to be able to do that at container level, because of closures ?
masak invocation records feel like exactly the right level for closures. 20:49
jnthn lizmat: Closures are what I'm (badly :)) trying to address too. When you take a closure, it has an outer frame pointer. If we access soemthing that was closed over, then the lookup is crossing from an invocation record created by the thread (when the closure was invoked) to the outer one created by the original thread. 20:50
TimToady the important thing about these shared things is not that they're shared, but that they're not transactional
jnthn In generaly, my worry about making things like the "$a++" example just work is that as soon as it does, people will then start to write The Next Idiom 20:51
for 1..4 { start while @foo { @foo.pop } } # race condition!
lizmat indeed
and we need to fail that before it starts 20:52
diakopter why!??!!?
jnthn And the problem here is that it's not obvious where the transaction starts/ends.
diakopter how is that a race condition?
(why does the user always need to care about the order of popping)
lizmat because you would be getting the same value from the array twice or not
jnthn diakopter: It's not about order
diakopter what?!?! 20:53
jnthn No, it's not about that either. Let's suppose pop is atomic.
lizmat (it's not currently, afaik)
jnthn It's still a race because @pop boolifies @foo to tell you if there's anything more
lizmat but let's assume :-)
jnthn Then next you try to pop
But somebody else might have beat you to pop the last thing.
TimToady iow, it doesn't compose
jnthn That's why things like the .Net concurrent collections don't have a Pop method
They have a TryPop
diakopter that's simply solved by adding a try
yeah, that 20:54
jnthn I guess this is all a "forgiveness > permission" thing :)
In cncurrent situations you can't ask if you can, then do it, 'cus the answer may change underneath you.
TimToady well, and making it likelier that they'll pick the composable primitives over the non-composable
diakopter so why can't the while's block have a try
TimToady we're starting to talk about STM again, here, basically--keep trying to do "this" until we succeed 20:55
STM just has a funny way of defining "this"
diakopter it's not talking about STM 20:56
jnthn diakopter: My point wasn't that you can't find *some* way to fix these, though "just add a try" works in this case provided you only catch the appropriate kind of exception.
diakopter: My point was that there's not a *general* approach.
diakopter why would you need a general approach 20:57
TimToady generality is generally...er...more general
jnthn Because solving these problems case by case is *far too hard* for the typical programmer.
Which is why .Net folks are pushed away from threads/locks and towards Tasks/TPL/Rx etc, for example. 20:58
Which all revolve around breaking problems into manageable pieces and making synchronization happen in well-defined places. 20:59
lizmat and threads are currently wide open on Perl6
indeed
jnthn lizmat: They are in Java and C# too, fwiw.
lizmat but Perl6 should be better
TimToady threads are sort of the Ur-DIHWIDT
jnthn lizmat: Not arguing that we should follow that. Just saying that it's a situation in languages where people actually do use threads day to day.
lizmat experienced people 21:00
jnthn No, most of them aren't, and most of them screw it up.
lizmat well, there you go
I don't want them to screw up, I want them to be steered into a direction that will do what they want 21:01
just maybe not in the way they originally thought
jnthn Me too, which is why I've been pushing the whole promise/channel/supply thing.
lizmat and compile/runtime errors should lead them that way
diakopter adding checks for thread parity everywhere can't be the right solution
jnthn It's just that I've been focused on the carrot (hey look, nice primitive)
And now we're discussing the stick. :)
diakopter: Can't be as in, we can't afford to slow everything down with those? 21:03
diakopter yes, but it also feels wrong just because it's checks the user has to imagine
lizmat so each scope gets a thread ID ?
jnthn diakopter: If so, I agree, but iiuc what TimToady is saying, the point of having things lexical is that we can know a lot of places we don't need to do any of this.
TimToady it's just tilting the playing field, not necessarily blowing up the other guy's goal 21:04
jnthn lizmat: fwiw, that already implicitly happens on both Moar and JVM, in that the scope gets a ThreadContext attached :)
21:05 kurahaupo joined
TimToady if we can detect hazardous sharing 50% of the time and slap the user's wrist, they'll tend to migrate over to the safer constructs, even if it's not perfect 21:05
since perfect usually implies 'unusably slow'
diakopter jnthn: I don't know what was meant by "having things lexical"
TimToady this is in contrast to P5 where way too many vars are package/global 21:06
it's the main motivation for making $*FOO work via search rather than "local $foo" as it is in P5 21:07
jnthn diakopter: I mean that a lot of variable accesses, you can look between where the variable is mentioned and where it's declared and determine statically that it's impossible that access could be on a different thread, even accounting for closure semantics.
21:07 hummeleB1 left
diakopter alright, that optimizes for the single-threaded case 21:08
lizmat my $seen; await( (1..3).map: { start { $seen++ for ^1000 }}); say $seen # could be caught at compile time
diakopter but what about when the user wants to parallelize things with shared data strutures
TimToady most single threads are single threaded :)
lizmat they're very singular about that :-) 21:09
TimToady we slap their wrists when we notice them using non-sharable data structures for that
diakopter lizmat: what would be caught there
what's the problem with that
21:10 sqirrel joined
TimToady what's the problem with what? 21:10
jnthn diakopter: Then the onus is on them to (a) say they're doing so, and (b) be extremely careful, just like you have to be in every other languages where you mutate data structures from multiple threads.
diakopter what's the problem with liz's example
jnthn diakopter: That's a "hard thing possible" rather than an "easy thing easy".
diakopter I don't understand why making all the changes atomic doesn't solve it 21:11
er, thread-safe.
b/c if it's just "when you're popping from multiple threads, you need to handle for when pop fails" how is that a problem 21:13
a problem for the user to remember/handle, I mean
TimToady "all the changes" is transaction scoping, and users tend to think in transactions that are not atomic unless we warn them off of it where we can
pragmatically, we admit the possibility of DIHWIDT, but we also vaccinate people and treat them when their sick 21:14
dealing with users is sometimes more like being a doctor than like being a starship engineer :) 21:15
so we need to say "Well then, don't do that." a little more often than we like 21:16
diakopter I don't see how you can ever possibly detect that
TimToady you can detect some of it
diakopter with any reliability
does anyone know what's the problem with liz' example? 21:17
jnthn On "what can be detected", it may serve us well to look at things like TSAN, Go's race detector, etc.
lizmat some is already good
jnthn It's not like we're the first people looking at this problem.
lizmat in any case, I think we need to make sure that multi-thread access to hashes, need to be protected
diakopter I feel like we're not looking at the problem; we're launching nuclear missles at it from 3 feet away 21:18
lizmat we don't want that to be able to corrupt a hash
hhe
diakopter that was never at issue
lizmat ok, I'll go stand a little further away
TimToady "I see you've used a single-threaded implemented of Scalar for variable $seen there, and referred to it from inside a thread. Howsabout you declare it 'my $seen is threadsafe' to get a Scalar with atomic increment at line 42"
lizmat which was "is cas" was all about
TimToady cas is the wrong name for that 21:19
lizmat granted
it defines the implementation
"is safe" maybe better ?
TimToady well, that's a bit too general :) 21:20
diakopter I don't understand what happened to the auto-upgrading threadsafe containers we've talked about in moar for ages (and that can be similarly done on jvm) 21:21
TimToady but I could see requiring a user to declare a lock-free hash implementation is to be used for a shared hash, or an array that doesn't support pop
jnthn "is cafe" :P
TimToady diakopter: if we can do that, great
diakopter when was there ever a question we could do that
I thought that was the plan
TimToady but it doesn't solve the boolean vs pop issue necessarily, from a transactional point of view, if you still support pop 21:22
the whole problem with transactions is that they are not reductionistic
just because all the operations are atomic doesn't make the composed operation atomic 21:23
jnthn diakopter: That aspect solves the VM safety level problems, for certain, but the issues is...what TimToady just put better than I was going to.
TimToady so lock-free structures are a good start
jnthn diakopter: If your problem involves anything more than that single data structure, you're still left with a problem.
diakopter jnthn: I think I was trying to express earlier that since you can't solve the general case (or come anywhere close, from what I can see), it seems trying to warn in particular other cases will just be confusing 21:24
because you can't warn on the passing closures as parameters
or lexicals
or container slot values 21:25
TimToady all-or-nothing is not PerlThink
jnthn Lock free data structures are really useful *if* your problem reduces to things that they can do atomically, which sometimes hapens. A Channel is basically a lock free queue.
diakopter or-anywhere-close
TimToady well, you can block on backpressure, and have a cyle of that, which is a kind of deadlock
jnthn True 21:26
21:26 beastd joined
timotimo considers commuting home again 21:27
yup, i'll do that.
TimToady sounds like a complicated transaction :)
diakopter jnthn: "you're still left with a problem" - no one has explained the problem except "I don't want to have to teach people to handle a failed pop, or I don't want pop to ever fail" 21:28
jnthn diakopter: At this point I fear if I showed you 10 more cases where the race isn't immediately obvious, then we found a way to fix it, you'd still be telling me the same thing. :/ 21:29
Even though the fixes would likely be different each time. 21:30
lizmat Niederrhein about to wrap up, decommuting soon
diakopter I don't think it's a valid worry; I haven't understood the way to fix the first one yet
I'm confused by various peoples' descriptions of what they're imagining/proposing because they all seem conflicting 21:31
it's like, let's all throw word soup on the wall and see what sticks 21:32
also, Wall
TimToady the basic problem is threefold: 1) nobody understands how to scale up transactions, and 2) an all-or-nothing approach is doomed to fail, and 3) nobody is happy with a partial solution, but that's what we're *guaranteed* to end up with, so how can we best push the user toward enlightement? 21:33
lizmat well, I think it's better than everybody being silent and ignoring the concurrent elephant in the room 21:34
diakopter no one has answered why "+=
"+= is bogus" 21:35
TimToady it's fine, but I wouldn't call it CAS, is all
diakopter since jnthn seemed to reply that it should be on the operations, not variables
TimToady he was talking specifically about CAS I believe
not about how to make += safe 21:36
s/safe/atomic/
jnthn Yes, I was talking about CAS on the operations not variables thing.
TimToady the S stands for Swap, darn it, not increment :)
diakopter okay, I was asking about the primitive case again 21:37
TimToady asking what?
diakopter (afaict there can be an atomic += for primitives)
lizmat for natives you mean? 21:39
TimToady was just questioning how you signal failure on that
diakopter yes
lizmat int instead of Int
diakopter atomic add or incr can't possibly fail
it always succeeds
that's the benefit 21:40
21:40 sqirrel left
TimToady but that's not how CAS works, hence the confusion 21:40
jnthn Depends what hardware you have, mind...if it doesn't support doing that, then it'll have to be implemented out of something like CAS that can fail.
And then emulate the "not fail" by retrying.
TimToady which goes back to what I said about conditional compilation
different arches have different primitives
alas 21:41
at some point it becomes less meaningful to talk about "primitives" and just talk about atomicity and/or transaction safety
21:42 kurahaupo is now known as kurahaupo2
TimToady and there's a bumpy transition from atomic to transactional, if atomic always succeeds, but a transaction has to be rollbackable 21:43
atomic doesn't imply commit/rollback semantics
21:44 lizmat left, lizmat joined
TimToady but all the the approaches I've seen to composing transactions involve either the possibility of deadlock/livelock, or the ability to just keep hammering on it till you get a commit to go through, à la STM 21:45
but even STM doesn't scale to disk or database transactions
21:47 stevan__ left, eternaleye left
TimToady and if you compose transactions inside transactions, you find yourself in the odd position of having to rollback inner transactions that have been "committed" 21:48
21:48 [Sno] left 21:49 eternaleye joined
lizmat decommiute& 21:49
21:49 lizmat left 21:50 woolfy left
masak clearly the inner transactions need to have only been "speculatively committed" until the outer transaction is truly committed. 21:50
TimToady
.oO(The all-or-nothing never works out well because, while we all understand 'nothing' pretty well, nobody understands 'all'...)
*approach
masak: which means there can never be final commitment without someone say "That's all, folks!" 21:51
*saying 21:52
masak well, there can be an outermost transaction.
diakopter masak: I can only hope you're tongue-in-cheek there
TimToady gah, can't type to keep up with my brane today, gotta start thinking slower...
masak: the question is, who's to be master, that's all. 21:53
diakopter the outerost transaction is the big bang
masak diakopter: no, it would be wise to assume ignorance rather than irony in my case right now.
21:53 spider-mario left
masak I seldom include the "big bang" abstraction in my programs. 21:53
TimToady diakopter: how do you know that's the outer one?
diakopter in that time dimension anyway.. 21:54
masak and yet they commit to things.
&
TimToady Maybe our entire universe is answering a what-if question, and once we get a result of 42, we're good to go. :) 21:56
21:57 rurban1 left 21:59 colomon left
TimToady Maybe our universe is being evaluated lazily, and we'll never actually get to the Big Rip, or Bound, or Whimper, because nobody will ask for the result... 21:59
*Bounce
21:59 rjbs joined
rjbs Whoops, I didn't rejoin after my screen-to-tmux conversion! 21:59
diakopter rjbs: HOW CUTE ARE YOU
nwc10 rjbs: you took the plunge? 22:00
rjbs nwc10: Yes, a bit over a week now.
nwc10 not that I have the font for it locally, but I think my irssi is now not being stupid so I can send 🐫
rjbs Yes, tmux let me do that again, too. 🐫 22:01
22:02 stevan_ joined
TimToady shouldn't that be 🐪 22:03
22:07 dmol left 22:09 dmol joined 22:27 [Sno] joined
dalek nda: 8375930 | (Tobias Leich)++ | / (2 files):
use $*EXECUTABLE_NAME instead of hard-coded 'perl6'

Because we now have perl6-p and perl6-j binaries.		 22:30
nda: 41a73f3 | tadzik++ | / (2 files):
Merge pull request #60 from FROGGS/master

use $*EXECUTABLE_NAME instead of hard-coded 'perl6'
tadzik FROGGS: thanks!
FROGGS tadzik: pleasure :o) 22:31
lue preflex list 22:35
preflex Botsnack: [botsnack]; Cdecl: [cdecl]; 8ball: [8ball]; excuses: [excuse]; Factoid: [+, -, ., ?, delete, get, store]; Help: [help, list]; Karma: [++, --, karma, karmabot, karmatop]; Nickometer: [nickometer]; Nickr: [nickr]; PlokiRE: [re]; Seen: [seen]; Sixst: [6st, ordinal]; Tell: [ask, clear-messages, messages, tell]; Rot13: [rot13]; Quote: [be, quote, remember]; WCalc: [calc, wcalc]; Version:
[version]; XSeen: [xseen]; ZCode: [zdec, zenc]
22:36 dmol left 22:38 dmol joined 22:39 PacoAir left 22:45 stevan_ left 22:54 dmol left 22:59 BenGoldberg joined 23:01 colomon joined 23:02 arnsholt_ joined 23:03 arnsholt left 23:08 stevan_ joined 23:13 beastd left 23:22 btyler left 23:23 jnap left, bluescreen10 left
japhb_ Is there a Perl 6-standard (as opposed to Rakudo-specific) way to refer to the argument capture of a routine that does not specifically specify a capture in the signature? My immediate use case is being able to refer to the arguments of MAIN as a unit, instead of a bunch of named args (which the auto-USAGE code wants to see) 23:24
23:36 lizmat joined
jnthn japhb_: Is there a reason not to specify a capture in the siggy? 23:39
Noting that you can do it *and* then write a normal signature too...
I'm not aware of a way besides that, though... 23:40
And I'd somewhat rather we keep it in the signature
As it's easy to detect if we have to build it. 23:41
And a raft of the opts I've got coming up to better use invokedynamic probably rely on knowing this...
japhb_ Ah, interesting, that's a really good data point.
23:42 woolfy joined
japhb_ In that case, I guess the best option is to teach the USAGE generator to ignore the capture. 23:42
23:46 ajr_ left 23:55 PZt left

Powered by Cro and the Raku® Programming Language.

Please report any issues / comments / feature requests as an issue on App::Raku::Log.

Thank you!