IRC channel logs

2017-05-10.log

back to list of logs

<davexunit>paroneayea: this tutorial is really nice. good job!
<davexunit>er not tutorial, spec
<paroneayea>davexunit: thank you :)
<paroneayea>davexunit: well it has a tutorial at the top!
<rlb>guile-2.2 (2.2.2+1-1) has been uploaded to debian unstable.
<rlb>ACTION wonders whether they're going to have to increase the build timeouts for some of the architectures...
<daviid>rlb: congratulation! can't answer your quiz, but is this based on the tarball or a git master clone?
<rlb>based on the git tag, otherwise it wouldn't be so bad.
<rlb>i.e. no prebuilts
<rlb>takes ~5-6 hours on my machine, though arguably not the fastest.
<daviid>oh, then it really can take hours to compile, some reported up to a day on old and slow machines
<rlb>mips is going to be *ugly*
<rlb>I suspect
<daviid>rlb: you should ask the guix folks, they have a good experience about that
<daviid>rlb: on #guix of course
<rlb>Well, in the end, doesn't matter - it is what it is unless I decide to try to use the archive .go files, and I'm currently inclined against, unless we have to. We'll see what the buildds say.
<rlb>Eventually we'll want guile to be a reproducible build, but one thing at a time.
<daviid>rlb: I agree with you, and guile build itself is now reproducible, if not mistaken, so it shouldn't be too hard to make guile debian packages reproducible as well
<daviid>rlb: did you succeed in having 2.0.14 in stable?
<rlb>daviid: not yet, and may not happen, but we'll plan for all of the important bits from 14 to be there even if we can't get everything, and/or change the version.
<rlb>daviid: I don't think there's a big difference atm, but I need to double-check.
<rlb>Might also be able to switch to 14 in the first stable update, but we'll see.
<daviid>rlb: ok, i beleive there has been impotant changes in between 2.0.11 and 2.0.14, but most importantly, i beleive for users, have the upstream stable in a stable distro is better (maybe only on the psy side :)
<rlb>daviid: oh, you meant jessie, I was thinking about stretch.
<rlb>stretch has 13 plus some patches.
<daviid>ah ok
<rlb>jessie is further back, certainly, and harder to change.
<daviid>ok, it would be cool to have 14 in stretch, imo (i think 14 is reproducible, but that would have to be confirmed with civodul, or looking at the patches)
<rlb>If nothing else, I think I'll at least plan to upload 14 for the first stretch update after the release.
<daviid>rlb: really cool! tx for all this work
<rlb>Then I need to get started on trying to remove 2.0 from buster.
<rlb>Hopefully that won't be nearly as bad as 1.8 was...
<daviid>hopefully
<rlb>daviid: you're certainly welcome.
<rlb>I'm also not sure how soon 2.2 might make it in to sid from new - don't know if stretch is consuming the same reviewers.
<rlb>bbl
<paroneayea>rlb: wow nice! :D
<nalaginrut>after this release, I can't wait for the further optimizing of Guile compiler
<nalaginrut>btw, guile-tjit has excellent optimizing for things like event-loop, I'd love to have it. I'll spend time on hacking guile-tjit
<wingo>nice
<nalaginrut>wingo: do you have plan for AOT now? ;-)
<davexunit>you mean native compilation? we have AOT
<nalaginrut>davexunit: AOT is a term to imply the native compilation
<nalaginrut>not what it looks like ;-)
<wingo>nalaginrut: i was thinking about it this morning :) probably some form of "instruction explosion" first
<wingo>will make things slower before they become faster tho. anyway, have to branch off stable-2.2 first
<manumanumanu>I admit defeat. I can't make our wordcount faster in any meaningful way
<ArneBab_>wingo: can I make string-split faster somehow?
<ArneBab_>wingo: that’s the core limitation manumanumanu and I face
<wingo>probably string-split isn't the right thing to use for counting words, if that's the question
<ArneBab_>that’s the question. read-line is nice and fast, but then splitting the lines gets slow
<wingo>don't split then. count the words in the line directly via a loop and string-index, or just a loop and string-ref
<wingo>dunno :)
<manumanumanu>Oh, but we need the word as well to get the frequency per word
<ArneBab_>(and we need to output it in the end)
<manumanumanu>I have to go make some baby food.
<manumanumanu>bbl
<ArneBab_>cu
<paroneayea>moin moin
<ArneBab_>moin paroneayea
<nalaginrut>wingo: is it possible to just add a backend for native code to replace bytecode part? I guess it's probably flexible enough to do so, or maybe not yet
<nalaginrut>fortunately we have Chez for inspiration
<nalaginrut>artanis-0.2 released, now I can have my supper casually
<manumanumanu>ArneBab_: This was fun. Are there any other challenges like that?
<manumanumanu>I have done the first 50 project euler challenges, but they have started becoming too hard for me
<ArneBab_>manumanumanu: there’s always the debian benchmarks game, but that needs some reading to start
<ArneBab_> http://benchmarksgame.alioth.debian.org/
<ArneBab_>there are existing racket operations which could be ported
<manumanumanu>ArneBab_: they are always doing so many weird things those racketeers.
<amz3>io
<ArneBab_>yes, I also stumbled over that
<nikita1>amz3: for plotting I use ploticus project
<manumanumanu>but (for (in-lines (current-input-port)) seems pretty convenient :)
<ArneBab_>manumanumanu: I just got it another 10% faster by first splitting by tab, then by newline and finally by space
<manumanumanu>ArneBab_: Great!
<ArneBab_>read-delimited \\n isn’t actually slower than read-line
<amz3>nikita1: tx
<ArneBab_>(that’s then using domain knowledge, though)
<ArneBab_>manumanumanu: this is the fastest I got now, execute with time taskset -ac 0 cat data/huwikisource-latest-pages-meta-current.xml | taskset -ac 0,1 guile -L guile guile/wordcount.scm >/dev/null
<ArneBab_>(taskset reduces the CPU switching: it doesn’t actually make the script faster, but reduces the amount of work done on other processors)
<ArneBab_>(I’m pretty sure that we’re in low-level stuff with that)
<ArneBab_> https://paste.pound-python.org/show/DbpBlsG3xYa2LfPRVg0d/
<ArneBab_>manumanumanu: I also created the much slower version which uses skip-lists — but in turn almost halves the memory requirements
<ArneBab_>manumanumanu: correction: the full dataset is 40x as large as the reduced one. The 700MiB memory our code needs right now will blow up to far too much.
<manumanumanu>ArneBab_:I can speed that up a bit :)
<manumanumanu>read-word is needlessly complicated
<manumanumanu>ArneBab_: got it down another 3 seconds.
<manumanumanu>!!
<manumanumanu>ArneBab_: https://paste.pound-python.org/show/mbGa1rOYFpyGv5Y4pVsf/
<manumanumanu>We are doing a hell of a lot of appends though. Append is O(n), which is not optimal
<manumanumanu>oh, you can delete the (define buf [...]-line. We don't use that anymore
<manumanumanu>we can remove that. it is better to just iterate over the sublists
<paroneayea>amz3: so you were playing with biwascheme right?
<paroneayea>amz3: do you happen to know if sxml runs in it?
<paroneayea>hey rekado
<paroneayea>didn't you have a multipart form-data library?
<paroneayea>rekado: I'm thinking of putting together a guile-webutils library under LGPLv3+. would you be interested in including something based off your code for that?
<paroneayea>maybe GPLv3+
<paroneayea>not sure
<amz3`>paroneayea: no vanilla biwascheme doesn't support sxml
<paroneayea>amz3`: kk :)
<paroneayea>eventually I'd love an sxml option in the browser
<paroneayea>where I can render templates the same both client and server side
<paroneayea>that would be super nice
<amz3`>I have that in my todo list, not far away
<paroneayea>amz3`: planning on porting the sxml stuff from the sourceforge release
<paroneayea>?
<amz3`>I don't know that sourceforge release
<amz3`>got it
<paroneayea>amz3`: http://ssax.sourceforge.net/
<paroneayea>iirc it's where guile's sxml comes from
<amz3`>hmmm, I don't manipulate the dom directly in my work
<amz3`>tho, I do declare the content of the page in sxml
<amz3`>I am not sure how ssax.sf.net can be helpful
<paroneayea>maybe it's not what you want, which is ok!
<paroneayea>but I'd like to render the same kind of templates I'm writing server side also client-side
<paroneayea>thatt way if I have people using stuff without js, but I have some js components
<paroneayea>I can use the same templates for both.
<amz3`>ok got ti
<manumanumanu>ArneBab_: down another second
<manumanumanu>ArneBab_: https://paste.pound-python.org/show/PN196Kso9K1UPIVwg8hD/
<rekado>paroneayea: my code is here: https://github.com/BIMSBbioinfo/rcas-web/
<rekado>paroneayea: I only wrote enough to be able to use the “fine-uploader” javascript thingie
<rekado>paroneayea: here’s something to handle multipart stuff: https://github.com/BIMSBbioinfo/rcas-web/tree/master/rcas/web
<rekado>I don’t think it’s very pretty, but I was under a deadline and haven’t worked on it since
<rekado>paroneayea: the rcas-web thing is under AGPLv3+, but you’re free to snarf whatever you want under GPLv3+ as well.
<paroneayea>rekado: \\o/
<civodul>rekado: is there an instance running somewhere?
***ertesx is now known as ertes