IRC channel logs


back to list of logs

***cky944 is now known as cky
<wingo>i did a git push --tags and it pushed trash tags
<wingo>serves me right for blindly following
***aburgess_ is now known as aburgess
<lokli>davexunit, hi, you really saved my source to source compiler project. I have total clarity about it now, and actually got to the point of experimenting with type inference. Do you have any insights into the topic?
<davexunit>lokli: glad to hear it! I don't know anything about type inference, though. that's a topic I need to do my homework on still.
<davexunit>I know that, at a high level, I would want to associate each expression with its type.
<davexunit>atoms (numbers, strings, etc.) are self-evident
<davexunit>the type of a nested expression is the result of examining the types of all sub-expressions
<lokli>i am mostly thinking of how to implement contexts and carry them around
<davexunit>I need to read up on how other people actually implement this stuff, but my naive approach would be to tag every expression with a type or maybe a set of types that it could evaluate to.
<davexunit>like, let's say that '+' is a primitive function in our language that may return an integer or a float, depending on arguments.
<davexunit>we can use some simple proofs to narrow that down
<davexunit>if both arguments to '+' are tagged as ints, then the '+' expression can be tagged as an int.
<lokli>davexunit, so, tagging is happening in ast itself, or another, parallel ast is to be created?
<davexunit>lokli: some new structure
<davexunit>I guess
<lokli>for now i am thinking context trees, holding those rules, every added rule is checked to be non-conflicting
<davexunit>this is all very... inexact. it might be totally silly.
<davexunit>AST -> type tagged intermediate structure -> ? -> target language
<lokli>thats a good enough start, prototyping will show if it is workable
<davexunit>having a structure to hold the set of proofs sounds reasonable.
<davexunit>I'm hoping that by reading more of "the little prover" I will be able to transfer what it teaches to this problem.
<lokli>i am having some visions (again) i would like to try. when they fail, will go educate myself on the topic =) also ask people around
<davexunit>good luck!
<wingo>mark_weaver, civodul: should i upload .gz releases too for guile? at some point we should be good with just .xz i guess
<wingo>i'll upload a .gz for this one i suppose
<peterbrett_work1>Is there going to be a 2.0.12?
***peterbrett_work1 is now known as peterbrett_work
<wingo>peterbrett_work: yes, but this is 2.1.2
<peterbrett_work>The latter I had already guessed :) With the precompiled compiler, right?
<peterbrett_work>Any thoughts on whether an architecture-independent .go format is a desirable thing?
<wingo>pretty sure it's not desirable
<wingo>the way that SCM objects are statically embedded in the ELF files depends on the architecture
<wingo>any run-time swizzling would negate benefits of shareable read-only memory
<wingo>and in the future these files will be natively compiled anyway
<wingo>if you wanted some independent representation for other purposes like link-time optimization, you want the CPS, not the bytecode
<peterbrett_work>Fair point
<wingo>so to me architecture-independent .go files is a non-goal
<peterbrett_work>What about shipping the CPS in the tarball rather than the .go files?
<peterbrett_work>Since (AIUI) CPS->bytecode is quite cheap
<wingo>cps->bytecode is not as cheap as you might thing :)
<wingo>it involves register allocation at least
<peterbrett_work>There must be a solution here that doesn't require all non Linux/x86{,-64} users to spend an hour bootstrapping
<davexunit>what's wrong with bootstrapping?
<peterbrett_work>It's bad for the environment
<davexunit>it's the right way to go for trust's sake.
<peterbrett_work>An hour of churning the CPU doing something completely duplicative is a waste of CO2 emissions
<wingo>peterbrett_work: see the you can symlink an architecture to prebuilt iles
<wingo>read the comment
<wingo>feel very free to submit patches adding symlinks
<wingo>and feel free to symlink yourself :)
<davexunit>peterbrett_work: it serves a real purpose: not having to trust pre-built binaries, which is a very desirable thing.
<peterbrett_work>Where did you get your C compiler?
<davexunit>I am thankful that Guile doesn't force me to trust a pre-built compiler.
<wingo>how to trust the c compiler is a different question :)
<davexunit>peterbrett_work: sure, I know where you're going, but do you think that adding an *additional* compiler to the trust chain improves the situation?
<peterbrett_work>If I'm happy to trust Fedora's C compiler, then I'm happy to trust the FSF's Guile Scheme compiler
<peterbrett_work>(Actually, more so)
<peterbrett_work>And the same applies for the vast majority of devs
<davexunit>the reproducible-builds project is actually looking for ways to bootstrap a C compiler without trusting another pre-built C compiler.
<davexunit>but there's also an approach that can be taken by finding a fixed point.
<davexunit>but this is a yak for another day.
<civodul>wingo: i prefer to upload both .tar.gz and .tar.xz, it doesn't cost much
<peterbrett_work>I'm not sure that there are only 4 variants of the bytecode
<civodul>congrats, wingo :-)
<peterbrett_work>I think it's big/little vs LP32/LP64/LLP64
<peterbrett_work>(Unless I've missed something)
<peterbrett_work>That reminds me that I need to test mark_weaver's LLP64 support patch
<davexunit>civodul: how do we disable the use of pre-built binaries in our guix package for guile 2.1?
<wingo>ACTION tries to remember how to use CVS
<civodul>davexunit: yeah, we could remove stuff in a snippet
<wingo>davexunit: make -C prebuild clean before building
<wingo>see NEWS
<wingo>you could of course nuke any .go in the tarball
<wingo>that would work too
<davexunit>okay, cool.
<peterbrett_work>wingo: You need to update the copyright date on NEWS
***karswell` is now known as karswell
<wingo>man, friggin cvs
<davexunit>the worst
<wingo>i am thinking it is useless to list md5 or sha1 sums, we should just do sha256 sums
<wingo>savannah has allows you to log in over http, weird
<bavier>yay for the 2.1.2 release!
<roelj>I'm going to check out the 2.1.2 release :D
<roelj>Hmm, I saw the the 2.1.2 tag when pulling. Now built and installed it, but 'guile --version' says:
<roelj>Wow, it is fast
<roelj>Thanks wingo
<roelj>for all your work
<wingo>tx for kind words :)
<mark_weaver>roelj: you need to run ./configure again to fix the version number
<mark_weaver>(or maybe even autoreconf -vfi, not sure)
<wingo>ok guil release done
<wingo>web site and all
<davexunit>thanks wingo
<davexunit>guile rocks
<wingo>looking forward to seeing it in guix :)
<mark_weaver>sweet! :)
<wingo>how should guix work with guile prereleases? just -next or should we have a guile-2.2 package or what?
<wingo>s/prereleases/eventual 2.2 release/
<mark_weaver>wingo: we have a 'guile-next' package. I guess it should be updated :)
<civodul>wingo: yeah it has a different name so that --upgrade don't switch to 2.1 inadvertently
<mark_weaver>and also so --install doesn't choose it over guile-2.0
<nalaginrut>ACTION glad to see 2.1+ new release 
<peterbrett>:) I'm looking forward to trying it out too
<peterbrett>Even though I've barely got gschem users even using 2.0 yet ;)
<amz3>congrats for the new release :)
<paroneayea>ooh, there's a new release?
<paroneayea>I'll promote it when I reboot back into debian and have access to my package manager again :)
<wilo_>Hi Community
<wilo_>Good Afternoon
<wilo_>I'm a python programmer who wants to learn Guile, a good book that can recommend me apart from the official documentation
<daviid>wilo_: here:
<wilo_>daviid: thanks :D
<paroneayea>wilo_: I'm also working on a tutorial, but it isn't really ready!
<paroneayea>hello buharin
<paroneayea>welcome to #guile!
<davexunit>here are you complimentary parentheses.
<wilo_>paroneayea: cool!, could see your tutorial
<paroneayea>warning: it's very early!
<wilo_>jeej no problem
<paroneayea>wilo_: happy to hear feedback though (with the caveat that I am mostly just "writing ahead" at the moment)
<wilo_>I'll track this page :D