🦋 Welcome to the MAIN() IRC channel of the Raku Programming Language (raku.org). This channel is logged for the purpose of keeping a history about its development | evalbot usage: 'm: say 3;' or /msg camelia m: ... | Log inspection is getting closer to beta. If you're a beginner, you can also check out the #raku-beginner channel!
Set by lizmat on 25 August 2021.
00:02 reportable6 left 00:40 frost joined 01:03 reportable6 joined 01:19 MasterDuke left 02:19 linkable6 left, evalable6 left 02:20 evalable6 joined 02:38 xinming__ left 02:39 xinming__ joined 03:39 squashable6 left, evalable6 left, tellable6 left, notable6 left, bloatable6 left, nativecallable6 left, quotable6 left, sourceable6 left, coverable6 left, statisfiable6 left, greppable6 left, unicodable6 left, benchable6 left, shareable6 left, bisectable6 left, reportable6 left, committable6 left, releasable6 left 03:40 squashable6 joined, notable6 joined, reportable6 joined 03:41 sourceable6 joined, shareable6 joined, committable6 joined, bloatable6 joined, unicodable6 joined, nativecallable6 joined 03:42 greppable6 joined, releasable6 joined, evalable6 joined 04:20 linkable6 joined 04:41 benchable6 joined 04:42 coverable6 joined, statisfiable6 joined 04:44 frost left, dogbert17 left, Scotteh left, chronon left, goblin left, gfldex left, a3r0 left 04:45 frost joined, dogbert17 joined, Scotteh joined, chronon joined, goblin joined, gfldex joined, a3r0 joined 05:34 tejr left, tejr joined 05:37 casaca left 05:42 tellable6 joined, quotable6 joined 05:54 japhb left 06:02 reportable6 left 06:05 reportable6 joined 06:17 tejr left 06:19 tejr joined 06:23 frost left 06:24 Skarsnik joined 06:49 japhb joined 07:17 seednode left 07:18 seednode joined 07:33 abraxxa joined 07:35 abraxxa left 07:36 abraxxa joined 07:40 abraxxa left 07:41 abraxxa joined 08:22 Sgeo left 08:57 MasterDuke joined 09:10 dakkar joined 09:50 hkdtam joined 09:56 casaca joined
Skarsnik I crashed a vm trying to build rakudo in it, that's new x) 10:29
SmokeMachine is there a way of exporting an EXPORT sub that will export something defined by the unit that imported that EXPORT (without using BEGIN)? 10:41
lizmat If you find a way, let me know :-) 10:42
SmokeMachine I was trying something like: `my %exports; sub red-config($schema) { %exports = $schema.models }; sub EXPORT(--> Map()) { '&EXPORT' => sub { %export.Map } }` and on the importing one: `use Red::Config; use Red; red-config schema Bla, Ble`but it's exporting before populating %exports (and that makes sense...) 10:45
I've also tried: `macro export-red-config(|c) { quasi { sub EXPORT(--> Map()) { red-config({{{ |c }}}) } } }` with no luck (but I'm sure I'm doing the quasi wrong...) 10:47
10:56 linkable6 left, evalable6 left 10:58 linkable6 joined 11:18 eseyman left 11:41 bisectable6 joined 12:02 reportable6 left 12:16 Skarsnik left 12:18 A26F64 joined 12:19 A26F64 left 13:13 thundergnat left 13:27 ufobat joined, Sgeo joined 13:35 A26F64 joined 13:48 Geth left, Geth joined 13:50 thundergnat joined 14:13 jmcgnh left 14:22 jmcgnh joined 14:56 [Coke] left 14:58 evalable6 joined 14:59 [Coke] joined 15:03 reportable6 joined 15:16 squashable6 left 15:24 ramiroencinas joined
ramiroencinas Hi everyone. I have discovered an annoying size limitation in the .write method of IO::Socket::Async class. This .write method doesn't write more than 65535 chars. However, the same method .write from the IO::Socket::Async::SSL class doesn't have this size limitation allowing to write more than 65535 chars. 15:36
I have tested it in Debian and Alpine Linux with the same results. 15:38
I don't know if this size limitation is a bug or not. 15:40
MasterDuke it looks like IO::Socket::Async::SSL uses openssl, but IO::Socket::Async uses libuv 15:51
the docs don't mention a size limit for IO::Socket::Async.write, but it returns the number of bytes written, i guess you're supposed to handle retrying with the rest if not all is written 15:53
ramiroencinas Thanks MasterDuke, I will try to handling the rest of the chars. 15:57
MasterDuke fell free to create a rakudo issue if you think it should do something differently (maybe the docs should at least mention the limit for rakudo) 16:00
16:05 Sgeo_ joined, Sgeo left
ramiroencinas Thanks MasterDuke, I have created a Rakudo issue. 16:16
16:16 ramiroencinas left
MasterDuke thanks 16:16
16:25 whatnext joined 16:26 [Coke]_ joined 16:28 [Coke] left, [Coke]_ is now known as [Coke]
whatnext hello all :) question about deleting using Red. I am basically doing the following `Table.^delete` and getting `DB::Pg::Error::FatalError.new(message => "syntax error at end of input" ...` I can't seem to get any visibility on what SQL statement it is trying to execute. Just thought I'd check if anyone had come across this before raising it as an 16:29
issue? :)
SmokeMachine whatnext: could tou try `my $*RED-DEBUG = True` before the delete to show us the SQL, please? 16:30
whatnext: Table.^delete? were you meaning `Table.^all.delete`? 16:31
whatnext I was following this: fco.github.io/Red/tutorials/start.html which says `Person.^delete` at the bottom of the page? 16:32
SmokeMachine whatnext: yes, sorry... I've forgotten about that 16:33
whatnext: but yes... I can reproduce it here... 16:34
whatnext should I be using `Table.^all.delete` then? 16:35
SmokeMachine whatnext: no, both should work... 16:37
whatnext: that's an issue... it seems it's starting a transaction and never finishing it... :(
whatnext: no, sorry, forget about it! it's not working inside a transaction... but it's failing for some reason... 16:38
whatnext I guess this must have broken recently? 16:39
SmokeMachine whatnext: I'll try to fix that after work. Would you mind to create a issue for that, please?
whatnext: yes, that's new...
whatnext Ok will do
andinus is there a way to locate memory leaks? say to narrow it down to a block? 16:42
MasterDuke well, a --profile will give you a count of allocated objects and where they were allocated, which may help 16:45
whatnext SmokeMachine: raised issue here github.com/FCO/Red/issues/525 16:46
SmokeMachine whatnext: thanks!
MasterDuke if you add --profile-kind=heap you'll get a heap snapshot, which can be more useful (but a bit trickier to understand)
or you can try running under something like heaptrack (though you might want to use --full-cleanup with rakudo with that) 16:48
that will give you moarvm-level information, but you might be able to figure out where that corresponds to in your script 16:49
andinus i see, i'll try with heap kind 16:53
MasterDuke i think you'll need p6-app-moarvm-heapanalyzer to open them. maybe comma does to? 16:54
SmokeMachine whatnext: It's going to be an easy fix... sorry for that... 17:02
andinus instrumented profile says, inlining eliminated the need to create .. followed by a very large number
SmokeMachine whatnext: are you liking Red and it's documentation? (there are a lot missing on the documentation...)
andinus js says NaN
# MoarVM oops: MVM_str_hash_entry_size called with a stale hashtable pointer 17:03
^ i get these crashes but i'm not able to reliably reproduce it, should i file an issue for it? 17:04
whatnext SmokeMachine: it's certainly the best ORM for raku right now, and there's a lot I do like about it 17:05
MasterDuke that usually means you're writing to a hash from multiple threads, which isn't allowed 17:06
whatnext documentation is a bit sparse it's true, but to be expected given the early stage of development
MasterDuke you're going to want to look in the allocations tab 17:07
andinus my @p; for @lines -> $iter { if @p.elems == $batch {await @p; @p = [];}} 17:08
lizmat clickbaits rakudoweekly.blog/2021/11/08/2021-...wo-commas/ 17:09
andinus ... push @p, start {} ==> later in the loop it does this
whatnext one thing that I would have preferred is a more DBIx::Class like interface, rather than the SQLAlchemy style. However, currently I use some fairly simple wrapper code which ends up making it pretty similar
andinus is this usage problematic? 17:10
whatnext I guess my main issue is namespacing - because with the SQLAlchemy style you end up `use`ing a module for every table
MasterDuke what are you doing in the start block? or is it really empty?
perryprog lizmat what's that picture?
lizmat it's from a number of chocolates presented by the late Jeff Goff at the 2015 YAPC::NA in Salt Lake City 17:11
SmokeMachine whatnext: I'm "fixing" that with this: github.com/FCO/Red/pull/524
andinus MasterDuke: its this script github.com/andinus/fornax/blob/mai...LI.rakumod
perryprog oh wow, nice
andinus it's using Cairo togenerate lots of pngs
i suspect it's Cairo that's causing the memory leak 17:12
SmokeMachine whatnext: with that, you can create a module like this: github.com/FCO/Red/pull/524/files#...aedfeR1-R9 and where you use it, you'll import all your models, the schema and the connection... 17:13
MasterDuke andinus: any particular reason you're doing manual batching and starts instead of using `race for @lines.skip.race.kv <...>`? 17:17
oh, you're also using run. i think that's known to (at least appear to) leak 17:19
andinus yes, race didn't do anything last i checked, i'll check it again (2 mins)
run is used at the end
andinus.unfla.me/writings/2021/for...rames.html
MasterDuke ah, right
andinus here i summarize the leak, it has screenshots as the script progressed
you can see the memory usage rising as the iterations pass
whatnext SmokeMachine: can I ask when you will merge into master? :) 17:20
andinus also seems like --full-cleanup always terminates with `SIGSEGV (Address boundary error)` on OpenBSD
i dont have comma, i'll get the heapanalyzer 17:21
MasterDuke turns out comma doesn't analyze heap snapshots yet, but they're in the process of adding that 17:22
andinus: i can just try some of the .fornax files in the repo to test with? 17:24
SmokeMachine whatnext: I'm still trying to find out a way to the user not need to manually write the EXPORT sub... (I don't think there Is a way... but when I'm sure I'll probably merge that)
andinus MasterDuke: yes, 50 is around 1200 iterations, i run, raku -Ilib bin/fornax --skip-video resources/solutions/DFS-50.fornax 17:27
just tested with .race, doesn't help, with manual batching, i see noticeable speedup 17:28
17:28 evalable6 left, linkable6 left
SmokeMachine whatnext: github.com/FCO/Red/commit/5dcfbc13...5cf818df74 17:29
17:31 linkable6 joined
whatnext @SmokeMachine: ok I see - that does look like a simple fix :) 17:32
thanks
I guess currently this is only an issue if no filter terms are specified - would that be correct? 17:33
SmokeMachine whatnext: yes, exactly
whatnext: and that wasn't being tested... :( 17:34
whatnext: if you test it and that's working, would you mind to close the issue, please? 17:35
17:37 dakkar left
lizmat ok 17:37
whatnext SmokeMachine: ok will do :) 17:38
SmokeMachine whatnext: thanks! 17:39
whatnext: and if you'd like to help even more, I'm needing all possible help to close all this issues (github.com/FCO/Red/projects/3) to finally launch our first stable version 17:41
MasterDuke andinus: turns out race doesn't support multi-arg blocks. but if you switch to .pairs instead of .kv and then pull out the $idx and $iter manually it works. a little bit simpler overall 17:46
still does have the occasional MVM_oops 17:47
andinus ah i see, i'll try with .pairs
what could be causing the MVM_oops after this?, nothing is shared (written) between threads 17:48
MasterDuke so locally i have `race for @lines.skip.pairs.race(:degree($batch)) {` and in the block `my ($idx, $iter) = .key, .value;`, and i removed @p entirely
andinus i see, makes sense
whatnext SmokeMachine: well I won't rule out pitching in on that completely - I am pretty stretched right now though, so it might be some time before I get to it
possibly I could pick up some of the documentation if it's still not done when I have more free time 17:49
17:49 abraxxa left
SmokeMachine whatnext: that would help a lot! Thanks! 17:50
MasterDuke the thread that oopsed was in Cairo's write_png 17:51
timo: ping 17:52
andinus is .race helping when you run locally? here it's not making much of a difference, manual batching seems to do better 17:54
MasterDuke it was roughly the same, maybe 1s faster with .race
andinus ah i see
yes, and with manual batching the speed is ~4x, as expected 17:55
MasterDuke e.g., 22s vs 21s with default batch size
andinus ah wait, that might be it, default batch is 64 iirc 17:56
MasterDuke i mean .race (with a batch of 4) was the same speed as manual batching (with a batch of 4) 17:57
timo sorry, i'm just heading out the door
MasterDuke --batch=8 is almost twice as fast 17:59
andinus MasterDuke: ah, not able to reproduce that behaviour, (openbsd), manual batching (4) is faster than .race(:4batch)
MasterDuke ah, maybe you're confusing race's batch and degree 18:00
andinus ah got it, :degree($batch), i should rename the var
indeed, .race seems faster, testing on DFS-10 (30 iterations) seems like manual batching beats it, maybe it performs well with more iterations 18:02
If there's an I/O operation inside the loop, there might be some contention so please avoid it.
^ .race docs do say this 18:03
18:03 reportable6 left
ugexe a lot of it is going to depend on your cpu 18:04
MasterDuke what it you do .race(:1batch, :degree($batch))
and the specifics of the workload, yeah 18:05
gfldex andinus: you could try to use a Channel to steam to-be-written data into a separat thread that does the IO.
MasterDuke i'm not saying you should definitely switch to .race, it just looked like the manual batching was pretty straightforward and a good candidate for converting 18:06
andinus :1batch doesnt change much.
without .race/batch: 24s, with .race (:degree($batch)): 12s, with manual batching: 6s
:1batch :degree($batch) should reproduce manual batching behaviour though 18:07
gfldex: i see, i'll read about Channels 18:08
MasterDuke on my system .race and manual is pretty much the same, even with :1batch or not and different values of :degrees 18:09
gfldex You can `use Telemetry;` to get an idea how well the worker threads are satured. see: gfldex.wordpress.com/2017/11/05/racing-rakudo/
How long does take one iteration on a single core? 18:10
lizmat -Msnapper also nowadays :-)
MasterDuke i'm actually seeing the same scaling factor for both ways, but manual consistently being a couple seconds slower 18:11
andinus i just tried it on a ubuntu machine, manual: 2.88, .race: 3.40, single core: 4.78 18:16
maybe the iterations are too low (40) for .race to benefit 18:17
MasterDuke: do you have .pairs.race or .race.pairs?
MasterDuke pairs.race 18:18
andinus ah, both same times
race.pairs would be better right?
gfldex: thanks for the hint! 18:19
18:19 A26F64 left
andinus i'll document these and update the writing later tomorrow 18:20
MasterDuke i think it's really the same, adding the .race is only needed/used to set the degree and batch. it's the race prefix that's doing the work
andinus ah, i was thinking .pairs is applied first so it would be single core and then .race multi threads the loop 18:21
MasterDuke andinus: btw, you can return a value from a given block. so something like `my IterStatus $status = do given $iter.substr(0, 1) { when '|' { Completed } ... };` should work 18:22
andinus ah,i do remember trying something similar 18:23
i was missing "do"
18:31 evalable6 joined
codesections How do I tell zef that I want it to run tests that live somewhere other than ./t ? zef test takes a path, but pointing it to a different directory (or test file) doesn't seem to do the trick 18:43
ugexe there is no way 18:44
codesections Oh, interesting. Thanks 18:45
ugexe a test has no way of communicating where the distribution its testing lives 18:47
and with `zef test .` the `.` is referring to a distribution 18:48
if it was like `zef test a/b/c/d` it would be impossible to tell if that is supposed to be a distribution or a test directory
codesections yeah, that makes sense. I wasn't thinking that it would let me put the tests _outside_ of the distribution. But I was thinking I could specify a different directory so long as a META6.json was somewhere in its ancestors 18:50
18:50 ufobat left
ugexe there is no "go up a level till you find a META6.json", which would mean you might not be able t have e.g. test files called META6.json 18:51
18:51 bdju left
ugexe plus you can run tests against installed versions of a distribution instead of e.g. -Ilib 18:51
there is probably a sane way to expand that `zef test $dist` to account for specific test files but it might be a bit verbose like having --test-file=... --test-file=... 18:52
codesections Makes sense. I'm not complaining – I just figured I was missing something obvious :)
Could it be a field in the META6? 18:53
ugexe Probably, although it is also weird that it would be referencing paths that are nowhere else in the META6.json 18:55
codesections wouldn't is just be a path starting from the same relative root as the paths in "provides" ? 18:56
ugexe yeah but it doesnt refer to anything after the distribution is installed 18:57
18:58 bdju joined
ugexe that doesn't discount the idea, but to me it is weird that an installed META6.json would refer to paths/files that are no longer part of it as a whole 18:58
we could of course installed tests >:)
which would let people do stuff like $dist = CURI.candidates("MyFoo"); CURI2.install($dist); run-tests-on-curi2($dist) 19:00
MasterDuke ugexe: i just did a `zef search Cairo` and get
0 |Zef::Repository::LocalCache |Cairo:ver<0.2.4>
8 |Zef::Repository::Ecosystems<p6c> |Cairo:ver<0.2.7>
but `zef upgrade Cairo` says `All requested distributions are already at their latest versions`
ugexe what version is installed? 19:01
MasterDuke 0.2.4 (which oddly enough was from a `zef install Cairo` just earlier today) 19:02
gfldex codesections: I got a script that likes to test things. If you run it with `raku-test-all test ./some-dir/ it does what you need. raw.githubusercontent.com/gfldex/b...u-test-all 19:03
codesections Hmm, I think that's getting deeper into zef's implementation than where I can have an informed opinion. I'd kind of thought that zef *did* install the tests somewhere in the sources/ folder of hashed paths
19:03 reportable6 joined
MasterDuke ugexe: oh, that was happening when i was using zef from its checkout. if i use the zef in <prefix>/share/perl6/site/bin/ an upgrade works fine 19:06
ugexe if those are both the same version they should work the same 19:07
might be worth doing --version to be sure the zef in checkout isnt loading some old zef from the e.g. home repo or some such 19:08
MasterDuke ah, checked out one is 0.13.0, installed one is 0.13.1
i need to remember to not use the checked out one
ugexe fwiw you should update zef again 19:09
to at least 0.13.3
MasterDuke just upgraded to 0.13.4, thanks 19:10
19:11 melezhik joined
ugexe codesections: it probably could similar to what we do for bin/ files, although really we need a spec for that too instead of just having zef grep for all files in that directory (some of which might not even be raku and thus shouldnt have the bin shim applied to them by rakudo) 19:12
at a minimum users would need to list files in bin/ they want installed, and also a way to identify that e.g. bin/zef is a raku script and not like a bash script 19:15
most of this seems like it could apply to tests as well
although tests have the benefit of file extensions 19:16
codesections those all sounds like good ideas (and things that it'd be good to get spec'd before the ecosystem gets too huge)
19:16 whatnext left
ugexe yeah the npm thing the other day got me thinking of how we can allow users to disallow installing bin scripts 19:16
codesections Yeah. Though if we do have a supply chain vulnerability issue at some point, Raku's flexibility will really work against us. 19:18
ugexe hopefully having a thoroughly strict ecosystem and disabling non strict ecosystems can mitigate a lot of issues 19:20
in the future non-strict ecosystems would ideally be limited to darkpans 19:21
codesections yeah. And having a culture that doesn't encourage piles of untrusted transitive dependencies 19:22
I was just looking at a report the other day that tracked average dependencies per project by language, and there's a huge swing
from Swift (4) and Go (13) to JS (377) 19:23
i.blackhat.com/USA-20/Wednesday/us...alysis.pdf
19:33 eseyman joined
ugexe yeah thats just dumb. a developer should have a good understanding of all of their dependencies and that isn't happening with 377 19:34
offloading that understanding to the first transitive dependency's author is a bad solution 19:35
but hey then you dont have to care
19:38 melezhik left 19:43 Geth left, TempIRCLogger left 19:45 RakuIRCLogger left, lizmat left 19:47 lizmat joined 19:48 Geth joined 19:51 Kaipi joined 19:53 childlikempress joined, bingos_ joined, gordonfish- joined 19:54 simcop2387_ joined 19:55 broquain1 joined, moritz_ joined, bdju_ joined 19:56 avarab joined, BinGOs left, bingos_ is now known as BinGOs 19:57 BinGOs left, BinGOs joined 19:58 gordonfish left, gordonfish- is now known as gordonfish
MasterDuke andinus: btw, i have a patch for rakudo that fixes the MVM_oops you're seeing with fornax 20:00
20:00 bdju left, moritz left, broquaint left, Kaiepi left, discord-raku-bot left, avar left, moon-child left, simcop2387 left, simcop2387_ is now known as simcop2387 20:02 Skarsnik joined 20:16 melezhik joined 20:23 melezhik left 20:26 childlikempress is now known as moon-child 20:31 melezhik joined 20:39 melezhik left 21:16 squashable6 joined 21:30 discord-raku-bot joined 22:13 daxim left
Nemokosch but you know it's funny because reading on, it turns out that most JS libraries are stable in return 22:24
unlike PHP where I have no idea what they managed to mess up so much... 22:27
22:34 Skarsnik left 22:47 A26F64 joined
tonyo hasn't been my experience with JS libs..some of the ultra popular ones are very stable _now_ but still have dependency hell 22:49
Nemokosch dependency is not "hell" per se. Javascript doesn't come with a huge runtime like Java or .NET so it's obvious the common things people are using will be more visible 22:53
22:57 melezhik joined
melezhik . 22:57
I am not sure of how many people have heard of #raku-news channel , but here one can get some new - comments from mybfio related to Raku 22:58
logs.liz.nl/raku-news/live.html
for old entries 22:59
logs.liz.nl/raku-news/index.html
tonyo 377 dependencies is significantly buigger than go's 13. it isn't like go has a huge runtime built into it 23:00
and building dependencies for something like raku makes 377 a bear
though, eventually, i suppose the dependency builder will get merged in with zef and it can async build most of them. iirc 2018 PTS is when that was built but the async portion wasn't stable enough to precomp everything 23:02
Nemokosch I only know Go as the language that explicitly advocated copypasting tbh
23:08 melezhik left
tonyo it does advocate that for the people who don't want to screw around with generators. there isn't a huge runtime for it, similarly to java or .net. it's not like js doesn't also promote copy and paste, either 23:12
it'd be easy to generate those numbers for raku
23:15 A26F64 left 23:16 melezhik joined
tonyo not sure what method they're using for the percentiles. raku came out for 10% = 0 depends, mean = 1, 90% = 4.. this could be tweaked to get the right numbers if they're doing something different curl -s 360.zef.pm|jq '.[].depends|length' 2>&1|sort -n|perl -e 'my @x; while (<>) { chomp; push @x, $_; }; my @N = @x[map $_*@x, .1, .5, .9]; printf "10%: %d\n50%: %d\n90%: %d\n\n", @N;' 23:39
23:42 melezhik left 23:44 dogbert17 left
Nemokosch Well JS nowadays is a lot less minimalistic than Go was designed to be, I don't think it's a bad thing that many things are offered as libraries on top of the language 23:58
I genuinely don't get why someone would use is-array for one but it would be a bit harsh to call this phenomenon "hell" lol