🦋 Welcome to the MAIN() IRC channel of the Raku Programming Language (raku.org). Log available at irclogs.raku.org/raku/live.html . If you're a beginner, you can also check out the #raku-beginner channel!
Set by lizmat on 6 September 2022.
00:00 reportable6 left 00:02 reportable6 joined 00:21 cacci left
leont I'm observing that «raku -e 'unit sub MAIN(:$foo);' --foo» gives warnings from rakudo's own argument handling (as in of the binary, not the Raku level argument handling). I'm pretty sure it didn't use to do that (but I don't have an old rakudo laying around) 01:01
Oh, it's more fascinating than that. It works as I expect when there's a space between -e and its argument, but breaks if there isn't. That has got to be a bug. 01:05
lucs <docs.raku.org/language/modules.html> mentions: 01:12
"「require」 loads a compunit and imports definite symbols at runtime." 01:13
What are "definite symbols"?
guifa_ probably should read "defined symbols", as in the symbols defined as exported by the module
actually, "exported symbols" is probably even better
lucs I guess that would make sense. 01:14
I wonder how sentences like that get produced...
guifa_ probably by people whose native language isn't English
definite and definite are fairly close in meaning
err
defined and definite
lucs Right. 01:15
01:15 zebzzw joined
lucs But even "defined" doesn't seem to convey what is meant. 01:15
guifa_ defined in the sense of "defined by the module" (which is different from being defined in the assigned context)
hooray for English ambiguity lol 01:16
01:17 clsn_ left 01:25 initrd joined
lucs Welp, after a bunch of tests, 'require Foo;' appears to be unable to import anything, no matter how they are declared in Foo.rakumod. 01:43
I hope I'm just missing something. 01:44
guifa_ So if my memory serves me correctly, `require Foo` might only give you access to `our` scoped variables
rather than exported 01:45
via Foo::whatever
in other words, the docs are really poorly worded
lucs Was about to say... :) 01:46
Which is sad, because hey, they're the docs.
guifa_ Tbh, require isn't used very often 01:50
most people just use, well, use
lucs I understand, but still, LTA. 01:51
01:51 razetime joined
lucs If I understood it more, I'd offer a pull request, but right now, the best I can do is raise and issue that would basically say "That doc doesn't work.", and that's not very helpful. 01:52
*an
guifa_ just put in an issue instead 01:53
lucs (I'm not even able to work out a correct "our" usage to make an export work.) 01:54
guifa_ seriously, that's hugely helpful, anytime a doc isn't clear, that's something that needs to be fixed
so in theory with our
it should be
module Foo { our $foo = 42);
then later
require Foo; say $Foo::foo;
or
say Foo::<$foo> 01:55
guifa_ is saying this from memory so don't quote him
lucs I was trying with class Foo {our method foo {say "FOO"} , and later sa
Dammit, pressed return before finishing editing, please disregard. 01:56
Your example works. 01:58
guifa_ -
if I can ask, is there any reason you're using require instead of use? 01:59
lucs For help with development: given a foo.raku, I want to either 'require FooVersionOne' or '...VersionTwo' for example, based on some condition. 02:00
It's not intended to be in released code. 02:01
There are othe ways to do this of course, but I wanted to try with 'require'. 02:02
02:04 cacci joined
guifa_ yeah I think the run-time loading of symbols is a little tricky, because at compile time it can't be sure what's available 02:05
so you have to indirect references
in the docs that's mentioned as
::("Foo::EXPORT::exportpackage::symbol_name") 02:06
lucs Ah, I see that, right.
guifa_ (indirects are always resolved at runtime, which is needed because `requries` is a runtime thing) 02:07
lucs I seem to remember it was easier in Perl, where 'use' was kind of like 'require', then 'import'.
(Perl was easier, but a lot less powerful, there is that :-) ) 02:08
guifa_ I think the general idea is that MOST of the time you just want to use, well, `use`
lucs That is certainly my experience. 02:09
guifa_ Raku really pushes hard the idea of torturing the developer to benefit the user 02:10
optional importing is definitely a module developer level thing -- an end user should know what modules they have and just install accordingly
lucs Hehe! I like that, even though I'm not much of a masochist.
guifa_ I definitely take it to the extreme in my modules. I like to make it so they ultimately export a single sub or two, and make the end user completely unaware of all the stuff going on in the background 02:11
lucs I totally agree. The interface to a module should be as simple as possible. 02:17
And if there are many features, good defaults should cover the most frequent usages. 02:18
(reported the doc issue) 02:27
guifa_ Right now I'm working on an international number formatter 02:29
99% of users will just use `format-number $number`
but I do have an advanced interface for power users who need extra control, by way the more traditional `number-formatter(%options)`, that produces a callable 02:30
lucs I know there are comma vs period delimiter differences, but I don't know of others. 02:31
Will be interesting to see.
guifa_ ha there's a lot
what symbols are used for digits (01234, versus eastern arabic, versus a bunch of other systems)
lucs I'm sure there are. I've been confronted to my ignorance often enough to expect unexpected stuff a lot :) 02:32
guifa_ how large numbers are separated (e.g. English in most countries uses 123,456,789; but in India, it's 1,234,567,89)
02:33 initrd left
guifa_ and I think some places do 1,2345,6789 because their systems are based on myriads 02:33
guifa_ is afk 02:38
lucs (crazier than I'd ever have thought) 02:41
03:10 cacci left 03:17 rf left 03:25 razetime left 04:00 razetime joined 04:34 razetime1 joined 04:36 razetime left, razetime1 is now known as razetime 05:21 razetime1 joined 05:22 razetime left, razetime1 is now known as razetime 05:51 razetime1 joined, razetime left, razetime1 is now known as razetime 05:52 razetime left 05:54 nort joined 05:55 razetime joined 06:00 reportable6 left 06:01 razetime left 06:02 reportable6 joined 06:35 abraxxa joined 06:49 tea3po left, tea3po joined 06:50 japhb left 06:55 Sgeo left 06:57 japhb joined 07:07 tea3po left, tea3po joined 07:33 jpn joined 07:48 jpn left 08:05 jpn joined 08:15 jpn left 08:57 dakkar joined 09:22 jpn joined 09:33 jpn left 09:35 jpn joined 10:17 cacci joined
Nemokosch Yesterday, I messed around with RakuConfig and string serialization overall, because of the syntax highlighter 11:06
Today, it paid off: there is a Template6 issue I can finally take up on.
11:12 peder left 11:17 sena_kun joined
Geth Raku-Steering-Council/main: 89250c6fa4 | (Elizabeth Mattijsen)++ | minutes/20230318.md
Add minutes of 18 March 2023 meeting
11:18
11:26 peder joined 11:42 peder left
lizmat and yet another Rakudo Weekly News hits the Net: rakudoweekly.blog/2023/03/20/2023-12-priced-2/ 11:43
11:45 peder joined 12:00 reportable6 left 12:02 reportable6 joined 12:23 mtj left
Nemokosch the docs say: "Heredocs include the newline from before the terminator." 12:34
lizmat and they do? 12:35
Nemokosch what if I don't want that newline to be added? do I just chomp it or is there a more official way?
lizmat chomp is what I always use 12:37
q:to/FOO/.chomp
or .chop 12:38
Nemokosch hm okay, .chomp will do I suppose. It's not the worst thing to happen that something can be solved from code, without entering yet another tricky escape situation 12:44
lizmat if you're worried about speed, .chop is quicker I think, but .chomp more conveys meaning 12:45
13:12 kolibrie joined
kolibrie I have a program that I intend to run for days or weeks at a time, but it is being killed by the OS after a day or so, I think because it is using up too much memory (growing over time). I would like to be able to debug the memory leak and find out where it is happening, so I can fix it. What tools, modules, or procedures do you recommend for 13:23
tracking down a memory leak? The program itself is short, but uses a multitude of modules, some of which are custom built for this program, and some that are open source. What seems ideal to me would be to have a list of which objects exist, with snapshots of that over time, so I can see which objects are growing in number, and which are staying
relatively stable. Or maybe there is something better. What do you recommend?
13:33 jgaz joined 13:38 rf joined
lizmat kolibrie perhaps commaide.com/features 's features wrt to running and debugging could be of help ? 13:39
you would need the paid version for the profiler though
rf Morning folks 13:41
lizmat rf o/
13:42 peder left
Anton Antonov @rf "Humming-Bird is a very unopinionated web framework." -- Yeah, right. 13:44
rf :^) it's just functions
13:44 peder joined
Anton Antonov @rf Sure. Basically that is a popular excuse to be lazy and not make the necessary functionalities that facilitate rapid development. 13:45
jast can't have everything :) 13:46
Anton Antonov In other words -- opinionated frameworks are more useful.
jast (I have no familiarity with Humming-Bird but more generally you can't check all the boxes in *any* project)
depends on what you want
things designed for rapid development can have downsides for more complex / more mature projects
Anton Antonov @jast Not everything, just the most common, popular, or useful.
rf I disagree. I made Humming-Bird so that people could structure, design, and create systems that make sense to them
I don't want to impose a strict guideline on how to build systems, rather, I'd prefer to give them the tools to do so 13:47
jast I do like unopinionated projects, and a subset of opinionated ones
Anton Antonov @rf I get it. (Still sounds like excuse for laziness.)
jast of course some things do suffer from being unopinionated, too 13:48
Anton Antonov @rf Generally speaking, documenting other approaches might be useful.
@rf It looks like that blog post refers to two, right? (Or is it just one.) 13:49
rf The only opinions Humming-Bird has is that your routes type Request, Response -> Response and I prefer it that way. If I wanted an opinionated framework I would have used Cro
jast the straightforward way to make something unopionated more friendly for rapid development is to layer an optional, more opinionated thing on top of it
and whether that's up to the original maintainer or other people... well, matter of opinion :) 13:50
kolibrie lizmat Thanks for your recommendation. Looking at a summary of what the comma profiler reports on, I'm not sure how that would help me. Have you used it for tracking down memory leaks?
Anton Antonov @rf I have only heard (or remembered) leont making statements like that. (Including Cro.)
jast looking at other languages there's a huge market for very unopinionated web frameworks that are mostly "just functions" 13:51
lizmat kolibrie: no, I'm stockholm syndromed to vim :-)
jast come to the dark side (emacs + evil-mode) :-)
Anton Antonov Amsterdam syndromed too. 13:52
leont Anton Antonov: what did I say?
lizmat well, I was born in Amsterdam, so that's hard to forget :-)
Anton Antonov @leont 🙂 Hi!
@leont "The distance to Cro matters." (For Raku projects.)
kolibrie @lizmat I'm also a vim user. 13:53
leont was also born in Amsterdam and a vim user 😛
Anton Antonov @lizmat You might have seen this : "They say I have been in Amsterdam, but I don't remember..."
jast I've been to Amsterdam and I used to use vim pretty much exclusively
so you're basically past me 13:54
lizmat Anton Antonov then you must have had a good time :-)
Anton Antonov So, my speculation is that lizmat tries other non-vi(m) editors, but remembers/recalls only vi(m) the next day.
lizmat :-) 13:55
Anton Antonov @lizmat Just in passing to a conference in Egmond aan Zee. 13:56
rf Anton another thing is, I think web-development should be fun too. I find it's a lot more enjoyable to work on less-opinionated frameworks. 13:58
14:00 kolibrie left
Nemokosch The problem with opinionated framework is that you learn something that you don't even know if you need, or why you need it 14:02
Anton Antonov @rf I get it. I am saying that you should/could demonstrate that the framework is non-opinionated by building different, opinionated solutions with it. Ideally, for the same problem (but that might be boring.)
Nemokosch which breaks down especially badly if there are more meaningful approaches
14:02 kolibrie joined
rf I write a lot of Spring-Boot for my job, it's super duper opinionated, I hate it. Mainly because the ideas don't cross-over to other languages or frameworks. That's one thing I wanted to avoid as well. 14:04
14:10 sena_kun left 14:15 kolibrie left
jast somewhat unrelated: I'm working on designing (not implementing yet :-)) a somewhat opinionated (non-Raku) web framework where data is virtually always operated on in RAM without fetches from disk/network, but still allowing Web Scale™. It's... complicated. :-) 14:20
currently researching ways to achieve fast reads on distributed consensus based top-level metadata without breaking the current design. fun stuff. 14:21
El_Che jast: everything in RAM, mm. When I think about springboot I think of our containerized apps that use 2-4 gig RAM to do less that out Go apps do with 20-40 mb :) 14:23
jast the goal was to do it in a language that doesn't treat RAM like candy :-)
El_Che jast: so the "adding everything in RAM" sound weird because I thought maybe they were already doing that
:)
*sound 14:24
jast the thing they're doing is probably slightly different, it's a complicated technology called "bloat"
El_Che they are good at it
:)
jast I was planning to give the application code the illusion that all the data it needs is always available in RAM, except in certain cases where it needs to work on data dependencies involving billions of objects 14:25
El_Che talking about bload Java-based Keycloak moved from wildfly/jboss to quarkus and now it uses 1/4 of the ram as before in my usedcase (a busy oidc provider)
jast and in well-defined cases all of the data needed for an operation *is* actually in RAM so you get free linearization of operations and ultra fast write throughput 14:26
14:26 Sgeo joined
jast I like thinking about weird stuff like this :)\ 14:26
El_Che well, if ram behaviour is known before hand, there is no problem. Lost of people wouldn't mind adding ram for lots of speed
however, web scalable == expenisve 14:27
jast as keeping everything in RAM (even across many machines) gets too expensive, the application can transition towards having some of its data use more traditional access patterns where async fetches will be necessary
El_Che you'll get blog post on how X moved from Ruby to Go/Rust/Whatever and now uses 50x less RAM :) 14:28
jast which gives you the best of both worlds
assuming I can make it work............
small "if" :)
El_Che so caching, essentially?
jast that's more of an afterthought
the cornerpoint of distributing the dataset is being confident that you know which node has primary authority over each bit of data 14:29
getting that and having it be both fast and 100% robust is... non-trivial
El_Che that's a hard problem
jast so I think there'll be a way to specify that certain data can be eventually consistent using simple conflict resolution rules
for instance, submitting a log entry to the wrong node by accident is healable, log entries (ideally) don't depend on each other so it can be inserted in the right place later on without violating the semantic integrity of the log 14:30
(assuming that it's fine for log entries to show up slightly delayed) 14:31
El_Che isn't that what mongodb is trying?
jast have I mentioned that distributed computing is super fun, and impossibly difficult?
but really the key of what I'm doing is to make it possible to control which guarantees you need for which of your application data, and have simple code initially which you can extend to deal with more difficult distribution requirements over time 14:32
mongodb is more of a "this is what you get for all your data, take it or leave it" and personally I wouldn't trust it with anything important :)
14:33 cacci left 14:37 rypervenche joined, cacci joined
El_Che haha 14:42
[Coke] We're using it for a large internal project, so far so good 14:58
exp for what purpose [Coke]? 15:03
Anton Antonov @Coke What? A you referring to: (1) a Raku package (e.g. "Humming-Bird"), (2) Raku in general, or (3) MongoDB?
exp jast: i'm very sceptical about the proposition that "log entries ideally don't depend on each other" 15:05
jast I mean obviously it depends on the exact thing you're using a log for 15:10
but time series data often has a lot of data independence as long as there's an eventual clear ordering of events 15:11
for instance, recording a series of measurements, if you send in a measurement out of order the other measurements don't suffer as a result, and if the measurement arrives in the correct spot before you actually do anything with the measurements, it doesn't matter 15:14
going back to MongoDB, I suspect that in the presence of network congestion or partitions there can be stale reads. stale reads are really hard to prevent without significantly affecting read throughput 15:15
[Coke] 1/2/3 - 3. MongoDB. I use Raku for lots of automation, tooling, but it's not part of our main app 15:20
tonyo .
Anton Antonov @Coke -- good to know. 15:21
There are at least a dozen of Raku packages for doing Artificial Intelligence (AI) and Machine Learning (ML). I consider making some likely and unlikely combinations of them. What would be a good namespace for those? Here are few names I consider: "AI::Fusion", "ACME::AI::Blending" , "ACME::AI::MushUp" . 15:24
perryprog ... there's ML stuff for Raku? That is a surprise. 15:25
Anton Antonov I think "AI::Fusion" might be misleading -- seen as referring to a concrete AI contraption.
@perryprog Are you joking, provocative, or you really do not know? 🙂 15:26
dakkar (very tangential: it's usually "mash up", not much "must up"; also traditionally it's spelled `Acme` not `ACME`; then again, it's just names)
perryprog I really didn't know!
dakkar (and of course I typo that… ☹)
Anton Antonov @perryprog Here is reference: raku.land/?q=ML .
perryprog But I'm also still in the mindset of assuming no-one uses Raku in production for some reason. 15:27
... hopefully that makes sense and isn't as cynical as I think it is 15:28
Nemokosch yeah, go figure lol
Anton Antonov @perryprog Yeah, I use DSLs like raku.land/zef:antononcube/DSL::Eng...yWorkflows to help production. But that is not directly in production...
perryprog That counts!
I would count "in production" as "used on a preferably non-solo team for non-personal reasons". I think that definition works. 15:29
El_Che perryprog: you underestimate the non-solo team in big places 15:30
Anton Antonov I am currently trying to convince a certain company to use Raku in production. (Again, for natural DSL-related projects.)
perryprog I said "non-solo" team because it'd be easy to have a personal workflow that consists of raku scripts being part of a runbook for example (which is basically personal), where Raku was chosen due to you being the sole member, and not the product of a group decision on how to architect some system. 15:31
Anton Antonov @El_Che ... and non-multi cohorts in small places... 15:32
perryprog Since the idea behind "in production" is really that "people agree that this thing is good to use in a professional environment, with dynamic groups of people working on the project the thing is being used for"
Anton Antonov I am solo Raku programmer in that company, but I there are at least a few consumers of the Raku-mande functionalities (via Cro and Docker.) 15:33
15:33 buffet left
Nemokosch this definition also rules out a lot of consulting, pretty much 15:34
Anton Antonov My motivation for implementing Raku ML packages initially was mostly to support my DSLs. 15:35
15:35 buffet joined
rf In production means its accessible to stake-holders and customers imo 15:36
Anton Antonov * ... "Raku-made" above, not "Raku-mande"...
perryprog As an example, I'm reminded of the classic Hacker Scripts. I wouldn't count whatever language these were written in to be used "in production", in most contexts. github.com/NARKOZ/hacker-scripts#readme
Nemokosch, disagree speaking from personal experience. 15:37
(albeit limited)
rf If when you use my product, Raku code is even touched, it's in production.
perryprog also we should get raku added to that repo
rf, but what about Raku used in build automation? (CI/CD) 15:38
Anton Antonov @rf Then this is Raku-product then : antononcube.shinyapps.io/DSL-evaluations/ .
rf perryprog: Sure, CI/CD is part of a production pipeline therfor it is part of production 15:39
perryprog But that's not always accessible to stake-holders and customers, right?
rf The stake-holders are the developers
perryprog Oh
rf the customers are the developers, anaylysts, etc
perryprog neeeeeeeevermind
Anton Antonov @rf Is this a Raku-production: pypi.org/project/DSLTranslation/ ? (For access of a Raku Cro service within Python sessions.) 15:40
rf Probably?
Anton Antonov @rf For me to claim "production level" effort means extensive documentation. (Also, often related demo-movies.) 15:42
15:42 mtj joined
Nemokosch I don't know. It seems a bit bizarre to say it's not production if, for example, others don't have to know that the tool you created for a particular problem, is written in Raku 15:43
It would make more sense for the outcome to matter.
rf Anton: Documentation has nothing to do with something being used in production. Look at most internal libraries, they're typically not documented at all but still used extensively 15:45
If you want people to adopt your library documentation is important, but it is not determanistic of how the software is used 15:46
Anton Antonov To a large extend, my efforts to make Raku "work" with/within Jupyter, Markdown, and Mathematica are for production purposes. My target users are data analysts and data scientists. Hence, transparency of the algorithms and easy to understand usage documentation are essential.
rf Yeah that makes sense, you're targeting your audience 15:47
Anton Antonov @rf "If you want people to adopt your library documentation is important, but it is not determanistic of how the software is used" -- unfortunately that is not true for code generating DSL packages. People are really not paying attention to the results, they just get stuck "why is this working" / "why are showing me this" . 15:48
[Coke] another ping: any raku folks attending TPRC in Toronto? 15:49
Voldenet I've worked with tons of production-ready software that was massively underdocumented 15:50
apparently business needs features, not docs. eh
rf Typically haha
Typically haha 15:51
Nemokosch Anyway, the social aspect is very important, and one part of it is to set the right expectations, rather than falling into the nirvana trap
Anton Antonov @Coke FWIW I considered the idea to go TPRC this year, and decided not to go. For logistics / time related reasons...
perryprog Voldenet but it's self documenting!! :P
Voldenet perryprog: well yes, I can read that disassembly just fine, it's _relatively_ self-documenting :> 15:52
Nemokosch we don't know if Raku will ever be rationally (whatever we mean by that) chosen over XYZ super duper popular general purpose language 15:53
but I'd say that goes for most languages, and there are many popular languages that don't have a "killer feature", or perhaps never even had one
15:53 derpydoo joined
Voldenet Raku was chosen over XYZ many times by now 15:54
by me at least 15:55
You can't exclude people from your statement, otherwise you'd end up with a tautology :)
Anton Antonov @perryprog "but it's self documenting" -- I know where these kind statements make discussion go. Here is A relevant summary with ML-context flavor: cdn.discordapp.com/attachments/768...nknown.png
Nemokosch same - whether that was a "rational" choice, is another story 🙂
perryprog ... lol 15:56
not totally fair, but still funny
rf Do I need performance? C++. Do I need to formally verify? OCaml. Else, Raku.
Anton Antonov @rf "Do I need to formally verify? OCaml." -- Another non-Mathematica user. (Thinking of non-opinionated-ness.) 15:58
rf Coq and OCaml > Mathematica
Nemokosch from what I can recall, Mathematica was very much "pay to win" software 😆 15:59
rf Yeah, you pay someone else to do it for you
Because its not a very good language :^)
Anton Antonov @rf "Coq and OCaml > Mathematica" -- Maybe true in some specific, idiosyncratic universe (populated mostly with hummingbirds.) 16:00
rf Hahahah
Voldenet I like TLA+ more than coq
maybe because I use it for distributed concurrency 16:02
Anton Antonov @perryprog -- thanks! 16:03
[Coke] .ask tbrowder - why is github.com/tbrowder/Windows-Test/b...6.json.pdf ?? 16:12
tellable6 [Coke], I'll pass your message to tbrowder__
[Coke] (it looks like it's a META6.json for PDF, not a pdf of the json file) 16:13
perryprog rf, what about Julia? :> 16:23
rf It's a fine language, just not my cup of tea 16:24
Anton Antonov @perryprog Julia is little bit like Obama -- it got an award for just showing up. 16:27
perryprog I don't follow.
But personally I'm quite a fan of Julia, I've used it pretty often for a few years now.
Cliche but I suspect it will be a fairly Big Deal™ in a few years in a lot of environments 16:28
Anton Antonov Since Julia is a Lisp-based -- and so are R and Mathematica -- I do not see Julia worth learning. (If I am not payed to use it.)
rf I think one benefit of Julia over R is it's very fast but idk if that's important in data land 16:30
Nemokosch isn't the "killer feature" of Julia being fast?
perryprog It's one of them. 16:31
rf, it very much is
Anton Antonov @perryprog "I don't follow." -- see en.wikipedia.org/wiki/J._H._Wilkin...l_Software . The way I see it, Julia got an award for releasing version 1.0. Nothing about Julia is that new or outperforming existing systems. (Mathematica, R, Python, etc...)
perryprog (Though I'm an outsider to data land, so take that with a grain of salt)
Oh, that's not at all true IMO Anton
Anton Antonov Of course speed is important in the data land. 16:32
16:32 derpydoo left
It has to be balanced -- again, of course -- with easy-ness of computation specifications. 16:32
16:32 cacci left
R is way too arcane in that regard. (Because it was made statisticians.) But the ecosystem of packages by RStudio makes R much more useful. 16:33
Voldenet Julia is nice, because you can even replace core types 16:34
perryprog There's a lot of excellent stuff Julia does in a way that makes it uniquely well suited for a lot of tasks. It has a quite good dynamic type system, it has multiple dispatch (which is super, super, super awesome), great C and python interop (among other things), amazing support for distributed computing, as well as threading and coroutines, extremely good handling of typecasting between different number types... 16:35
Voldenet the worst part is that "you can replcae core types"
perryprog lol
Voldenet now, imagine amount of combinations you can produce with it 16:36
and debugging all of it
rf I wonder how outlandish it would be to run Raku on Beam
perryprog multiple dispatch also lends itself extremely well to interop between different packages. There's a Measurements package which provides a generic Measurement which wraps a numerical value, letting you say 1.0 ± 0.1.
Anton Antonov @perryprog In the Rraku DSL Data Wrangling package I mentioned above I show code generation for Julia. In those experiments -- 1-2 years ago -- Julia was always slower than R and Python. (But I was comparing only CSV file ingestion and a particular paradigm of data transformations. See Julia's "DataFrames".)
Voldenet basically this: yuri.is/not-julia/ 16:37
apparently julialang is too flexible at some points
tbrowder__ [Coke]: thats correct, i should mv it to a dev subdir 16:38
perryprog It then automatically propagates that error according to normal rules, so ((1.0 ± 0.1) * 2)^2 = 4.0 ± 0.8. The way it's implemented also takes advantage of automatic differentiation of functions which basically means any package instantly supports this propagation. (Except for some things that only work with strict Float64 :P)
tellable6 2023-03-20T16:12:28Z #raku <[Coke]> tbrowder - why is github.com/tbrowder/Windows-Test/b...6.json.pdf ??
perryprog Anton, are you including the JIT compilation cost? While it's fair to sometimes include it in some benchmarks, it's worth noting there's now ways to minimize it. 16:39
Voldenet in fact for large things non-flexible languages are better
Anton Antonov @perryprog Below WL == Mathematica 1. "It has a quite good dynamic type system" -- so does WL. 2. "it has multiple dispatch (which is super, super, super awesome)," -- so does WL 3. "great C and python interop (among other things)" - so does WL 4. "amazing support for distributed computing" -- so does WL 5. "as well as threading and coroutines" -- cannot comment now... 6. "extremely good handling of typecasting between different
number types..." -- so does WL
perryprog Voldenet, while Yuri has some good points in that now-infamous article, there have since been like eight hundred responses to that essay on why some stuff they say isn't totally fair. 16:40
Nemokosch Voldenet: enter Go xD
perryprog Anton, WL is icky!!
checkmate freedom hater
Anton Antonov Now, these "responses" of mine are just to support my claim that Julia does not bring something particularly new.
perryprog Totally fair
Anton Antonov @perryprog Your attitude is something Stephen Wolfram does not understand. 16:41
perryprog lol
Nemokosch Stephen 'Sigma' Wolfram
Anton Antonov Julia is not different -- it really feels "same family" language if you know LISP, Mathematica, (or R). 16:42
perryprog Don't get me wrong, WL is genuinely incredible, but I don't think it's ethical to rely on closed source algorithms or implementations in academic research. You're locking your methodology behind a paywall.
Voldenet Nemokosch: well yes, golang is awfully simple and non-abstract, it's nice for doing things
tellable6 Voldenet, I'll pass your message to Nemokosch
[Coke] tbrowder__: my recommendation would be to rename it to meta6-pdf.json - the last extension should be the type.
Voldenet but at some point you start hating how non-generic and inelegant it is, it feels too simple 16:43
Anton Antonov @perryprog Simply not true. Wolfram Engine is free for developers -- you get it using an email.
Nemokosch it's almost like you are biased towards the Wolfram system, Anton
:DD
perryprog The algorithms are still closed source and proprietary!
Anton Antonov WL has first class documentation -- I do not know any other system for mathematical computations with better documentation.
perryprog Oh that's 100% true 16:44
It's absurdly good
Like, it's unfair how good it is
rf octave.org/
Nemokosch by the way, I only read like the first two paragraphs of that "why I can't recommend Julia anymore" and like... do y'all not realize that it could be literally written about Raku?
rf It could also be written about 95% of languages 16:45
Nemokosch too many tricky interactions, impossible to prove anything on the most superficial level
perryprog Anton, perhaps I should've been clearer: I view the closed source implementation bit as a much bigger detail than WL being paywalled (in some cases, in some cases like you said it's free).
Nemokosch from the OCaml nihilist's perspective, maybe
Anton Antonov @perryprog "The algorithms are still closed source and proprietary!" -- 1) So? 2) Not entirely true, many of the complicated algorithms have published WL-written prototypes. 3) Especially, for ML algorithms -- they are verifiable (in a variety of ways.) 16:46
Nemokosch but this is something Raku does particularly bad at, thanks to the Perl heritage
rf I can use dependent typing
perryprog "To reproduce our results just run this black box that we created with these inputs"
rf It is semi-provable
Nemokosch the control flow is crazy though 16:47
(and the type system is stuck in the "weird flex" position - that I really wouldn't call "production ready" at all)
perryprog I just don't think it's in the spirit of academic research to have the algorithms (among other details) that some research was done using to be just as open as the results of the research. 16:48
Anton Antonov @perryprog I am somewhat biased toward WL (as @Nemokosch was readily kind enough to mention above.)
perryprog That's totally fine, no worries
It'd be like omitting the "Procedure" section saying "To reproduce the results, hire one of of our lab employees who will perform the test in our lab with no witnesses and will send you the results" 16:49
(And these analogies aren't totally fair, I'm just using them to illustrate where I'm coming from—they aren't meant to be the basis of my points.)
Anton Antonov @perryprog I am / was a Mathematica power user. So, the WL-community is somewhat surprised I am spending so much time with Raku. (BTW, that includes Stephen.) 16:50
perryprog Lol
Stephen is a very interesting guy.
His personality is... something. 16:51
Anton Antonov Absolutely! Some of the most insightful comments about computing I have heard from him.
perryprog Heh. I meant that in the slightly negative connotation. 16:52
I don't agree with a lot of his views.
Anton Antonov It is just academia for reason decided to "hate" Stephen. (And understand why, but I also understand why he finds that surprising.)
perryprog I mean, the reason he finds that surprising is part of the reason for that hate, I think. 16:53
rf The only person I know who uses WL besides anton is Xah Lee, and he is pretty crazy
tonyo stephen diehl ?
perryprog Stephen Wolfram 16:54
Anton Antonov @perryprog I know. 🙂 That is why I produce and compare same computational workflows in different systems. Mathematica often wins, and in many cases is just decisively better.
perryprog But he's very, very, very self-centered on his work. This isn't necessarily a bad thing but it makes a lot of his discussions end up going from a normal topic (like PL theory) to "oh here's how WL does that" and then a giant tangent about WL and how great it is.
I really just can't get by the fact that it's closed source. To me that's a fundamental reason to never use it for anything that's not intended only for yourself. 16:55
Anton Antonov @rf You realize, I might mention idiosyncratic, hummingbird universes, again, right!
perryprog oh goodness 16:56
Anton Antonov @perryprog "But he's very, very, very self-centered on his work." -- One of my friends summarized Mathematica as "made for brilliant physicists." 16:57
perryprog s/brilliant physicists/SW
Anton Antonov Fine -- I am just saying there are other people thinking in those lines.
Nemokosch so Mr Wolfram is a physicist I guess? xD
perryprog You aren't good at Mathematica. You're good at temporary emulating a fraction of Stephen Wolfram's mindset. —Joseph-Louis Lagrange 16:58
Oh I'm just being facetious, sorry
Nemokosch Lagrange xdd
perryprog Also, correct me if I'm wrong, but isn't Mathematica only fast for... /some/ things, but way slower for any task that isn't just normal numerical analysis? Like, isn't it slow in the areas outside of the parts that have fine-tuned optimized compiled implementations? 16:59
Anton Antonov I need to go to a meeting -- I will tell more about Julia's mandelbrot-ness later.
Nemokosch The French are historically outstanding at claiming people who would have never been French in a less imperialistic world
perryprog Anton hasn't responded, therefore I wasn't wrong, and therefore WL is slow. 17:01
I win.
Also very relevant: blog.wolfram.com/2019/04/02/why-wo...n-reasons/ 17:03
Anton Antonov @perryprog "[...] normal numerical analysis" -- symbolic mathematics of WL is fast.
perryprog Oh, that too yeah
I mean stuff outside of like. Math stuff.
Anton Antonov Like what? 🙂 17:04