| samcv | jnthn, you should find one of your modules and add a license to meta | 03:27 | |
| we're at exactly 50% things with/without licenses. you could win points if you get us over 50% | 03:28 | ||
| or anybody here! | |||
| lizmat | Files=1192, Tests=56873, 210 wallclock secs (12.46 usr 4.80 sys + 1233.36 cusr 123.08 csys = 1373.70 CPU) | 07:32 | |
| lizmat gave an answer to stackoverflow.com/questions/4370157...5#43704475 | 08:34 | ||
| [Tux] | This is Rakudo version 2017.04.3-112-g1ed76a903 built on MoarVM version 2017.04-44-gf0db8822 | 08:54 | |
| csv-ip5xs 3.230 | |||
| test 13.036 | |||
| test-t 5.347 - 5.379 | |||
| csv-parser 13.109 | |||
| and my computer was not busy doing other stuff | |||
| lizmat | [Tux] could it be that you have a large .precomp dir ? | 09:08 | |
| nine | lizmat: would that make any difference? | 09:09 | |
| lizmat | well, in most OS's dir searches are still lineair | ||
| I should say, filesystems | 09:10 | ||
| if you have 1000's of .precomp entries | |||
| that does make a difference | |||
| [Tux] | 290M .precomp/ | ||
| rm -rf .precomp/ .panda-work/ | 09:11 | ||
| This is Rakudo version 2017.04.3-112-g1ed76a903 built on MoarVM version 2017.04-44-gf0db8822 | 09:14 | ||
| csv-ip5xs 3.034 | |||
| test 13.046 | |||
| test-t 5.107 | |||
| csv-parser 14.313 | |||
| lizmat | so looks like it made a difference :-) | 09:15 | |
| [Tux] | but still not way below 5.000 :/ | 09:27 | |
| lizmat | [Tux]: true :-( | 09:39 | |
| Geth | rakudo/nom: c41a200ade | (Elizabeth Mattijsen)++ | 3 files Make Set.SET-SELF return set() for empty Sets As Sets are immutable, there should only one empty set around. |
09:46 | |
| rakudo/nom: 3d99321f19 | (Elizabeth Mattijsen)++ | src/core/set_operators.pm All other types on (.) and (+) need Bag coercion This fixes an uncaught error on Iterable (+) Iterable if the Iterable contained Pairs. And it improves the speed of Iterable (.) Iterable by a factor of 2.5. |
|||
| rakudo/nom: 6d12cecc95 | (Elizabeth Mattijsen)++ | 9 files Remove trailing whitespace |
09:55 | ||
| rakudo/nom: bdb5391b94 | (Elizabeth Mattijsen)++ | src/core/set_operators.pm Simplify Set creation in several set operators Now that Set.SET-SELF is smart enough to return set() for empty Sets |
10:04 | ||
| pmurias | Unhandled exception: Missing or wrong version of dependency '../MoarVM/install/bin/../share/nqp/lib/MAST/Nodes.nqp' (from 'src/Perl6/Pod.nqp') | 10:08 | |
| what could be causing that while trying to make rakudo? | |||
| nine | Have you tried make clean? | 10:09 | |
| I usually need to do that after updating nqp | 10:10 | ||
| pmurias | I hate Makefiles-- so much :( | 10:11 | |
| timotimo | cross-project makefiles, not much fun | 10:14 | |
| jnthn | samcv: Discovered github.com/jnthn/p6-app-moarvm-heapanalyzer was still without one, so there we go... :-) | 10:59 | |
| samcv | yay! | 11:21 | |
| you win! | |||
| party time | |||
| jnthn | :-) | ||
| samcv | now we can say that *most* modules have license tags | ||
| timotimo | cool | 11:23 | |
| "your chance of a module having a license tag is better than 50/50!" | |||
| samcv | that is unproven. | 11:25 | |
| though | |||
| but if you randomly install all modules then yes | |||
| dogbert17 | does anyone know if slangs have a large overhead? | 12:10 | |
| timotimo | i've never measured, but it shouldn't be terribly big | 12:11 | |
| and it's only compile-time overhead of course | |||
| dogbert17 | I was looking at [Tux] csv code again, i.e. github.com/Tux/CSV/blob/master/test-t.pl | 12:12 | |
| if you put an 'exit 0' at line 8 and then time the script I get 1.3 seconds | |||
| timotimo | and if you rewrite it to be untuxic? | 12:13 | |
| dogbert17 | I'd say one second faster | ||
| i.e. I removed the compilation errors I get when I remove the slang | 12:14 | ||
| timotimo | interesting | ||
| wanna --profile-compile that? | 12:15 | ||
| dogbert17 | is there a --profile-compile switch? | ||
| or do you mean perl6 -c --profile test-t ? | 12:16 | ||
| lizmat | m: Bag.new.roll # hmmmm | ||
| camelia | Cannot iterate object with P6opaque representation (Scalar) in block <unit> at <tmp> line 1 |
||
| Zoffix | s: Bag.new, 'roll', \() | ||
| SourceBaby | Zoffix, Sauce is at github.com/rakudo/rakudo/blob/bdb5...gy.pm#L498 | 12:17 | |
| timotimo | there is a --profile-compile switch | ||
| dogbert17 | cool, didn't know that | ||
| lizmat | Zoffix: working on that | ||
| timotimo | :) | ||
| dogbert17 | hmm, wrote perl6 --profile-compile test-t.pl and now it hangs, maybe it's slow | 12:18 | |
| timotimo | it's likely quite slow | ||
| dogbert17 | I'll wait then :) | 12:19 | |
| pmurias | is a raw » allowed in the comments and literals in the MoarVM source | ||
| timotimo | hopefully it won't blow up in memory usage and asplodes :) | ||
| pmurias: i know nothing about the other C compilers we target :) | |||
| but isn't » in latin1? | |||
| m: say "»".ord | 12:20 | ||
| camelia | 187 | ||
| timotimo | that should totally be fine | ||
| pmurias | timotimo: gcc handles that fine, I don't know what else are we using to compile on other despicable platforms | 12:21 | |
| timotimo | hmm | 12:22 | |
| well, you know, we're using {} and other kinds of parenthesis without using trigraphs | |||
| i think a latin1 character should be fine | |||
| Zoffix | kinda weird $*SPEC.curupdir is a `none`, rather than `any` Junction; considering .curdir and .updir are strings ' | 12:25 | |
| '.' and '..' | |||
| timotimo | right, that must be because it was originally used as the test for dir? | 12:28 | |
| which of course could also have used none(any(...)) rather than just the any from curupdir | |||
| or, you know, none(|$*SPEC.curupdir) | |||
| dogbert17 | timotimo: the routines with the largest exclusive times are: mergesubstates (17.99), mergesubrule (14.85) and addedge (14.52) | 12:33 | |
| these are percentages btw | |||
| timotimo | oh, yikes, why isn't the result of that precompiled? | 12:34 | |
| MasterDuke_ | dogbert17: was that for a clean (i.e, hadn't been compiled before) run? | ||
| timotimo | or is Slang::Tuxic exporting a "no precompilation" thing? | 12:35 | |
| dogbert17 | I had run it before but I could try again | ||
| github.com/FROGGS/p6-Slang-Tuxic/b...g/Tuxic.pm | |||
| timotimo | i'unno | 12:36 | |
| dogbert17 runs again with --profile-compile | 12:39 | ||
| 1.3 secs to get the code running is quite much, perhaps it's different on 64 bit systems | 12:40 | ||
| timotimo | well, we have the jit there, so that's something | 12:41 | |
| lizmat | feels to me the 1.3 is mostly Slang::Tuxic causing a redo of the grammar | ||
| timotimo | hm. i wonder if a manual call to precompute_nfas would help | 12:42 | |
| it's a private method on a Grammar | |||
| perhaps that could go into define_slang | 12:43 | ||
| dogbert17 | lizmat: but if it takes the same time on [Tux] machine then that's 20 % of the total execution time | 12:44 | |
| lizmat | dogbert17: have you tried removing Slang::Tuxic from test-t.pl ? | 12:45 | |
| looks like it doesn't need it | 12:46 | ||
| dogbert17 | I do get compilation errors if I remove it | ||
| lizmat | you do ? | ||
| hmmm | |||
| Zoffix | I wanna add a mode argument to nqp::open(). Right now we have no core way to securely create files. | 12:47 | |
| And add a :chmode arg to IO::Handle.open.... :mode would be nice, but it's already taken up :/ | |||
| I mean :chmod | 12:48 | ||
| timotimo | opening and at the same time (?) changing the permission flags? | ||
| dogbert17 | lizmat: two terms in a row on e.g. $sum = [+] $csv.getline_all ($*IN)».map(*.elems); | ||
| Zoffix | timotimo: just giving different mode to libuv than DEFAULT_MODE | 12:49 | |
| lizmat | yeah... seeing now too, wonder what happened before | ||
| timotimo | mhm | ||
| Zoffix | timotimo: I'd hope there's no "changing" but rather the file is created in the given permissions from the start | ||
| timotimo | ah | ||
| Zoffix | Otherwise there's still a security issue | ||
| lizmat | dogbert17: rewriting test-t.pl to plain Perl 6 saves me 13% (3877 -> 3406 msecs) | 12:51 | |
| dogbert17 | that's quite a savings | ||
| lizmat | it *would* bring it below 5 seconds again :-) | 12:52 | |
| dogbert17 | but I seem to remember that [Tux] wanted his slang | ||
| [Tux] | correct. no slang = no Text::CSV (at least not by me) | ||
| lizmat | yeah, but one could argue that's not needed in the test file | ||
| [Tux]: I mean, test-t.pl is like a user file, and a user would most likely *not* use Slang::Tuxic | 12:53 | ||
| [Tux] | lizmat, I agree: but if it surfaces additional problems, the canary lives happily everafter | ||
| lizmat | so I think it would be fair to remove Slang::Tuxic from test-t | ||
| [Tux] | lizmat, unless the user is me :) | ||
| lizmat | well, we could make a test-tuxic.pl :-) | 12:54 | |
| timotimo | [Tux]: how many lines is hello.csv supposed to have so i get the same workload as you? | ||
| jnthn | Or try to figure out why Slang::Tuxic requires `no precompilation`, if that's the issue? :) | ||
| Zoffix | huggable: bench csv | ||
| huggable | Zoffix, export PATH=`pwd`/install/bin:$PATH; cd CSV; for i in $(seq 1 10000); do echo 'hello,","," ",world,"!"'; done > /tmp/hello.csv; time perl6 -Ilib -MText::CSV test-t.pl </tmp/hello.csv | ||
| Zoffix | timotimo: ^ | ||
| timotimo | it doesn't have "no precompilation", though | ||
| Zoffix | (ignore the export part) | ||
| timotimo | ah | ||
| dogbert17 | on my system I get (test-t) real 0m8.859s, (test-no-tuxic) real 0m7.871s # this is a crappy i5 660 | 12:56 | |
| jnthn | Ah, OK | 12:58 | |
| lizmat | m: say 8859 / 7871 | ||
| camelia | 1.125524 | ||
| lizmat | yeah, that's about what I saw | ||
| timotimo | hm. i added precompute_nfas to define_slang but it didn't make mergesubstates/optimize/mergesubrule/addedge disappear from the top of the routines tab | ||
| it does have precompute_nfas in it, at the 5th spot, and it takes 73% of total run time | |||
| er, total compile time | 12:59 | ||
| dogbert17 is now grammar expert so github.com/FROGGS/p6-Slang-Tuxic/b...g/Tuxic.pm does not give me any clues | 13:00 | ||
| s/now/no/ :) | |||
| timotimo | dogbert17: It is like a finger pointing away to the moon. Don't concentrate on the finger or you will miss all that heavenly glory. | 13:02 | |
| (the finger is Slang::Tuxic, the moon is rakudo's and nqp's innards) | 13:03 | ||
| dogbert17 | ah | ||
| so IS Slang::Tuxic precompiled? | 13:08 | ||
| lizmat | yes | 13:10 | |
| afaik | |||
| and Text::CSV as well | |||
| afaik | |||
| dogbert17 | so it might be the 'redoing of the grammar' which takes up the time then, like you wrote earlier | 13:12 | |
| lizmat | m: say now - BEGIN now; sub postfix:<!>(\a) { a } # the cost of rewriting the grammar | 13:14 | |
| camelia | 0.28916266 | ||
| lizmat | m: say now - BEGIN now; sub prefix:<!>(\a) { a } | 13:15 | |
| camelia | 0.0146918 | ||
| lizmat | prefix:<!> exists in core, postfix:<!> doesn't | ||
| pmurias | jnthn: I made a pull request to fix the str to num conversion | ||
| jnthn: :16[1,1] is not implemented yet | |||
| jnthn: paste.scsys.co.uk/559078 | 13:16 | ||
| dogbert17 | so the thing that didn't exist tool ~0.3 secs to compile then | ||
| pmurias | jnthn: that the test file for that PR | ||
| Geth | roast: d543e756ad | (Zoffix Znet)++ | S32-io/io-handle.t [io grant] Test IO::Handle.DESTROY closes the handle |
13:20 | |
| timotimo | i don't konw how grammar::tracer fails to invoke the original $meth | 13:21 | |
| oh, it could very well be it's deeper inside than i thought | 13:22 | ||
| we really are calling invoke_o on VMNull | 13:25 | ||
| lizmat | Zoffix: s/It's worth nothing/It's worth noting/ ? | 13:27 | |
| Geth | rakudo/nom: 9e7d0b36cb | (Elizabeth Mattijsen)++ | src/core/Baggy.pm Make Baggy.roll about 15% faster Also fix issue for Bag.new.roll, which used to fail. |
13:30 | |
| Zoffix | lizmat++ thanks fixed. | 13:31 | |
| timotimo | jnthn: could the fact that match objects and cursors are now the same thing (and derived from the grammar if i understand correctly?) have some bearing on why grammar::tracer's wrapcache explodes? | 13:34 | |
| the last thing it prints from my debugspam before exploding is that it's just made an unwrapped version of !alt | 13:37 | ||
| huh, *weird* | 13:39 | ||
| say "something" gives an "unable to invoke this object (REPR: Null, VMNull)" | |||
| evalable6 | timotimo, Full output: gist.github.com/9a2cf767c4a6bcb4a8...1f30c40286 | ||
| (exit code 1) 04===SORRY!04=== Error while compiling /tmp/LApikm2HtN Two ter… |
|||
| timotimo | but nqp::say("something") works | ||
| thank you, evalable6 | |||
| dogbert17 | timotimo: so we have a couple of broken modules as well | 13:44 | |
| timotimo | what the flying fuck. | 13:45 | |
| even $indent++ or $indent = $indent + 1 breaks | |||
| jnthn: could it be that the created wrapper function somehow isn't correctly closing over the perl6 core setting? | 13:47 | ||
| timotimo pretty stumped | 13:49 | ||
| [Tux] re-runs after lizmat's commit | 14:00 | ||
| note that this did neither change Text::CSV not perl6 (rakude/nqp), but *just* the invoking script | 14:01 | ||
| s/not/nor/ | 14:02 | ||
| 4.436 | |||
| lizmat | whee! :-) | ||
| [Tux] | I still think this time change isn't fair | 14:03 | |
| lizmat | [Tux]: I think it would only be unfair if you were the only user of Text::CSV, which I sincerely hope is *not* the case | 14:04 | |
| afaik, test-t is about testing Text::CSV performance, not Slang::Tuxic performance | 14:05 | ||
| Geth | rakudo/nom: aeecfc445d | (Elizabeth Mattijsen)++ | 5 files Move QuantHash helper methods to Rakudo::QuantHash Since I will probably add a few more in the near future, and I don't want to make Rakudo::Internals even more behemothy than it already is. |
14:07 | |
| dogbert17 | lizmat++, [Tux]++ | 14:08 | |
| at least we now have an idea as to the current compile/execution overhead wrt Slang::Tuxic | 14:09 | ||
| lizmat | $ 6 '' | 14:10 | |
| real0m0.120s | |||
| $ 6 'use Slang::Tuxic' | |||
| real0m0.557s | |||
| [Tux] | lizmat, I mean that it now *looks* like the speed improved, but nothing changed | ||
| and there still remains the undelying performance loss of using a slang in two layers | 14:11 | ||
| lizmat | [Tux]: two layers ? | ||
| there is only a compile time performance issue | |||
| since we don't pre-compile the script, we see that in every run | 14:12 | ||
| [Tux] | Text::CSV already uses Slang::Tuxic, so I see no reason why having the calling script also using Slang::Tuxic causes that much overhead | ||
| ah, oké, that explains | |||
| lizmat | because the precomped version of Text::CSV doesn't need It anymore | ||
| [Tux] | right | ||
| timotimo | right, removing Slang::Tuxic from the rest of Text::CSV will not have a performance impact on test-t performance | 14:17 | |
| anyway, ideally we'd be able to just re-use the precompiled result of Perl6::Grammar + Slang::Tuxic::Grammar and then the performance hit on test-t would be gone regardless of using Slang::Tuxic or not | 14:18 | ||
| lizmat | timotimo: I'm not seeing how you could merge the precomped Grammars | 14:23 | |
| timotimo | hm. because the EXPORT is run every time the user script is run? | 14:24 | |
| you're right, there's nothing we could really do about that :\ | 14:25 | ||
| lizmat would have loved to be proved wrong | |||
| timotimo | a little spark of hope was to check if what we're being passed is the pristine Perl6::Grammar and just return a pre-made one from a module we've precomped | 14:26 | |
| but i don't think we can just change the type from underneath the instance? | |||
| i mean, we do have a type change thing that exists. not sure if that'd work here | |||
| lizmat neither :-) | 14:27 | ||
| timotimo | of course, test-t could become a module that just gets used and it'd do its thing in the mainline :) | 14:29 | |
| lizmat | yes, but *that* feels like cheating, as we lose the compile overhead of test-t altogether then | 14:32 | |
| timotimo | yes | ||
| lizmat | Text::CSV is typically used in one/few-off situations | ||
| timotimo | it'll give an unexplained change in all timings | ||
| lizmat | if anything, perhaps we should deduct .3 from all measurements so far | 14:33 | |
| timotimo | mayhaps | ||
| anyway, the whole mergesubrules and stuff business can become faster from a thing TT has been pondering | |||
| so maybe we'll get to chop a few second-parts from compile time there | 14:34 | ||
| lizmat | that would be brill :-) | ||
| dogbert17 | as far as the rest of the code I remember [Tux] pointing to the slowness of the NEXT phasers | 14:37 | |
| Geth | rakudo/nom: 94390ae9ca | (Elizabeth Mattijsen)++ | src/core/Baggy.pm Use the more ideomatic .raw_hash instead of nqp |
14:40 | |
| rakudo/nom: 2bd1d9ec77 | (Elizabeth Mattijsen)++ | 3 files Move R:I:WeightedRole to R:QuantHash To make Rakudo::Internals less behemothy still |
|||
| lizmat | afk& | 14:54 | |
| [Tux] | dogbert17, that is because return/next/last is handled as exception (as jnthn explained), so only an optimizer can be used to boost the use of next | 15:38 | |
| where a relatively simple looking loop using next and last to jump out of it can be re-written to be a huge if/else tree | 15:39 | ||
| Geth | rakudo/nom: d672424efc | (Elizabeth Mattijsen)++ | src/core/Rakudo/QuantHash.pm Reorder methods into logical sections Purely to make maint easier |
18:40 | |
| rakudo/nom: 07feca6747 | (Elizabeth Mattijsen)++ | 3 files Add R:Q.BAG-TOTAL, make BagHash.roll 1.7x as fast Optimize the adding up of values in a Bag/BagHash. Is not as important for Bag's as is it cached there upon first usage. But since BagHashes are mutable, the total is currently calculated each call. |
|||
| MasterDuke_ | s: &infix:<%%>, \(4, 2) | 19:22 | |
| SourceBaby | MasterDuke_, Sauce is at github.com/rakudo/rakudo/blob/07fe...ic.pm#L221 | ||
| Geth | rakudo/nom: 5e459bce12 | (Elizabeth Mattijsen)++ | 4 files Add R:Q.MIX-TOTAL, make MixHash.total 1.4x faster Optimize the adding up of values in a Mix/MixHash. Is not as important for Mixes as is it cached there upon first usage. But since MixHashes are mutable, the total is currently calculated each call. This does not affect Mixy.roll, as that uses another logic. |
19:31 | |
| rakudo/nom: b1fbd1331d | (Elizabeth Mattijsen)++ | 3 files Some more uses of .raw_hash Preparing some more for %!elems -> $!elems shift. Also some logic folding to handle empty (Bag|Mix)Hashes faster. |
20:05 | ||
| nqp: MasterDuke17++ created pull request #356: Some cleanup to the SQL profile output |
20:14 | ||
| lizmat | MasterDuke_: it looks like your PR breaks the JVM build ? | 20:38 | |
| MasterDuke_ | t/qast/01-qast.t failed, but i don't see how i could have effected that at all | 20:39 | |
| Geth | rakudo/nom: 7404c706ff | (Elizabeth Mattijsen)++ | src/core/Rakudo/QuantHash.pm Add R:Q.A.MIX-TOTAL-POSITIVE Needed for streamlining Mix.roll |
20:42 | |
| rakudo/nom: 762fd23906 | (Elizabeth Mattijsen)++ | 5 files Make Bag|Mix.new return bag()|mix() directly Without needing to create an intermediate Bag|Mix object. Since new() lived in the role, this shouldn't actually add any new entries to the MMD tables. |
|||
| japhb | .tell nine With a Rakudo built from scratch a few minutes ago, I am unable to install Inline::Perl5 using zef, despite tests passing. Error: ===> Install [FAIL] for Inline::Perl5:ver('0.26'):auth('github:niner'): ===SORRY!=== Could not find Inline::Language::ObjectKeeper at line 4 in: ... | 21:04 | |
| yoleaux | japhb: I'll pass your message to nine. | ||
| Geth | rakudo/nom: fec547a1ed | (Elizabeth Mattijsen)++ | 2 files Abstract Baggy.roll() into R:Q.BAG-ROLL So we can streamline Baggy.roll(x) and Mixy.roll(x). No noticeable performance difference on Baggy.roll. |
21:07 | |
| lizmat | hmmm.... Geth and/or GitHub a little slow | 21:50 | |
| anyways, I've just committed github.com/rakudo/rakudo/commit/a2...60f746b601 | 21:51 | ||
| MasterDuke_ | gfldex was mentioning in #perl6 that he was testing github's api today and it was very slow and causing retries | ||
| timotimo | status.github.com/ | ||
| you see that? | |||
| lizmat | yeah, looks all green no? | 21:52 | |
| timotimo | "time to first byte" went up to almost 20 seconds (or is that just 2s?) | ||
| lizmat | looks like it peaked just below 4 secs | ||
| timotimo | right, not 20 %) | 21:53 | |
| lizmat | but mean hook delivery time is still < 1 second | ||
| so maybe Geth has an issue ? | |||
| anyways, looking at Baggy.grab(*) | 22:00 | ||
| currently, that's eager | 22:01 | ||
| I'm thinking of making it lazy, and having it return a Seq | |||
| timotimo | hmm | 22:05 | |
| so as you're iterating over the list, the bag slowly gets mutated? | |||
| lizmat | yeah, so I'm not sure whether that would be a bug or a feature | 22:10 | |
| I mean, if you want immediate removal, then make it eager :-) | |||
| docs don't state anything about laziness | |||
| specs also don't say anything about laziness | 22:11 | ||
| timotimo | right | 22:13 | |
| interesting little tidbit | 22:14 | ||
| lizmat | I could see uses for a lazy grab | ||
| timotimo | yeah | ||
| kind of | |||
| lizmat | anyways, I will sleep on it... | ||
| enough hacking today... | |||
| good night, #perl6-dev! | |||
| timotimo | gnite lizmat! | 22:15 | |
| BenGoldberg | How can you be going to bed, it's only 6 pm! ;) | 22:32 | |
| Zoffix | "◀▬▬ │ Geth (zofbot-get@perl6.party) has quit (Remote host closed the connection)" | 23:42 | |
| It wasn't even in the channel when you were discussing it.. | |||
| And I don't know why it died because I still didn't add any logging to it. Its screen was just not found :/ | |||
| timotimo | oof | 23:43 | |
| MasterDuke_ | fyi, timotimo merged github.com/perl6/nqp/pull/356 while Geth was still away | 23:55 | |
| Zoffix | hm | 23:59 | |
| m: my @COMMIT-FILTERS where .all ~~ Callable && .all.name = sub foo ($) {} | |||
| camelia | Sub object coerced to string (please use .gist or .perl to do that) in block <unit> at <tmp> line 1 |
||
| Zoffix | m: my @COMMIT-FILTERS where {.all ~~ Callable && .all.name} = sub foo ($) {} | ||
| camelia | ( no output ) |