Camelia, the Perl 6 bug

IRC log for #darcs, 2010-08-12

| Channels | #darcs index | Today | | Search | Google Search | Plain-Text | summary

All times shown according to UTC.

Time Nick Message
00:43 sshc joined #darcs
01:03 darcswikibot joined #darcs
01:12 zooko joined #darcs
02:06 abuiles joined #darcs
02:25 abuiles left #darcs
02:28 Gracenotes joined #darcs
03:26 lispy mornfall: one thing we might need to do is figure out what RULES are baked into GHC for lists and port them to FL/RL
03:26 lispy mornfall: it's entirely possible, our custom lists are a huge pain performance wise
03:28 lispy ?tell exlevan if you're having trouble with the type witnesses, I can help.
03:28 lambdabot Consider it noted.
03:37 zooko The Tahoe-LAFS project is about to start using bzr in addition to darcs because GHC isn't ported to OpenBSD/sparc64 but Python is.
03:39 lispy zooko: it's been ported to OpenSolaris
03:39 lispy zooko: are you sure it's not on OpenBSD?
03:39 lispy and why bzr instead of mercurial?
03:40 twb joined #darcs
03:40 zooko Because someone will volunteer to support the darcs->bzr updater and because then Tahoe-LAFS can be automatically built nightly for Ubuntu packaging and integrated into Launchpad.
03:40 zooko I'm sure GHC doesn't support the platform.
03:41 zooko Hm, I would show you the ticket, but http://hackage.haskell.org/trac/ghc isn't responding.
03:43 lispy hackage is down
03:43 lispy http://www.mail-archive.com/glasgow-h​askell-bugs@haskell.org/msg23780.html
03:43 lispy zooko: but, the code gen has now been ported
03:44 lispy That ticket is old, before the port I think
03:44 lispy Igloo: can you confirm?
03:44 zooko Yes, that's the ticket.
03:46 lispy Right at 5pm our internet went out at work, hackage has been down since then
03:46 lispy I speculate that the construction workers in the building broke the phones/internet, but I can't say for sure
03:48 sshc joined #darcs
03:51 zooko It would be great if this buildslave could start participating in building Tahoe-LAFS: http://tahoe-lafs.org/buildbot/​builders/Ben%20OpenBSD-sparc64
03:55 twb lispy: solar flares!
04:22 FunctorSalad_ joined #darcs
04:38 twb joined #darcs
04:59 sshc_ joined #darcs
05:04 sshc joined #darcs
05:15 sshc joined #darcs
06:58 balor joined #darcs
07:00 burp bbbbbbbbbbbbbb
07:00 burp oops
07:02 darcscommitbot joined #darcs
07:03 darcswikibot joined #darcs
07:24 sxs joined #darcs
07:37 mornfall *yawn*
07:37 mornfall Finally some substantial sleep.
07:51 lispy mornfall: I tried switching the parser monad to a CPS style with church encoded Maybe, but it slowed it down.
07:52 lispy mornfall: I tried adding strategic bang patterns, but it slowed it down.
07:53 lispy mornfall: Using data a :*: b = !a :*: !b, instead of (,) makes it about 10 ms faster on massive input (meaning, it's negligible)
07:54 mornfall Interesting.
07:54 mornfall Do you have the benchmarks? I'd be interested to have a look.
07:54 lispy I haven't really been saving them just because they're not faster
07:55 lispy Or do you mean the means by which I experiment?
07:55 mornfall Yea.
07:55 mornfall (The code.)
07:56 * mornfall disappears in a shower for a bit, while the rice's cooking...
08:00 lispy mornfall: darcs get http://code.haskell.org/~dagit/darcs-parser
08:00 lispy mornfall: Step 1) it reads from the file "patch" in the top level of the repo
08:00 Gracenotes joined #darcs
08:01 lispy mornfall: Step 2) it depends on the version 2.4.98.2 of the darcs library.  I achieve this by bumping the version in the .cabal file of the version of darcs that Im testing with
08:01 lispy so, once you edit the cabal file, you should be able to build/install darcs
08:01 lispy Then, cabal clean && cabal configure && cabal build && ./dist/build/darcs-parser/darcs-parser
08:02 lispy Note:  You need to reclean after every modification to the darcs source, otherwise cabal won't properly relink the darcs-parser binary
08:06 mornfall OK, thanks, I'll have a look.
08:13 JaffaCake joined #darcs
08:37 lispy I think I just found the world's slowest compilation options for the darcs source
08:37 lispy add -funfolding-use-threshold=16 -optc-O3 makes it really slow!
08:41 twb WPO would be slower
08:41 lispy WPO?
08:41 twb whole-program optimization
08:41 twb Normally optimization is only done within a translation unit
08:41 mornfall twb: That's not entirely true. Inlining is cross-module.
08:42 twb To WPO the (naïve) compiler needs to store the entire program in memory at once.
08:42 twb mornfall: yeah, ok, hand-waving away details
08:42 mornfall :D
08:42 lispy yeah, GHC's optimizer isn't WPO or supercompilation (yet) but cross-module inlining is activated with -O and higher
08:42 twb And I was mainly talking about C, not hs
08:43 twb I remember there was a Scheme compiler that did whopr
08:43 twb Er, WPO
08:43 mornfall twb: :))) I do WPO in C++. It's called templates.
08:43 lispy mornfall: oh hrm.  Changing the compilation flags gives a big boost
08:43 mornfall If your program is a template instantiation...
08:43 lispy Well, the biggest boost I've seen
08:43 twb It took nine days compiling itself, at which point my 3GHz P4 hit a thermal cutout and I gave up.
08:43 mornfall :)
08:44 twb stalin, that was its name
08:44 mornfall twb: Dear lord.
08:44 mornfall Oh!
08:44 mornfall Yeah, stalin is cool.
08:44 mornfall If by nothing else, the name.
08:44 mornfall (And how it's fitting.)
08:44 twb It was intended for schemers doing numerical analysis on big iron, I think
08:44 twb Or something batshit insane along those lines
08:44 mornfall That's even more fitting.
08:45 mornfall Stalin goes great with big iron.
08:45 twb It's still in Debian, if you care
08:46 mornfall I know
08:46 mornfall .
08:46 lispy mean: 766.2472 ms, lb 757.0107 ms, ub 774.9432 ms, ci 0.950  down to mean: 705.9286 ms, lb 701.0311 ms, ub 710.6611 ms, ci 0.950
08:46 mornfall lispy: I'm not sure we want to go via-C though.
08:47 mornfall We could hope for early LLVM merge, though.
08:48 lispy I doubt the -optc-O3 one does anything.  I think it's really the inlining thresshold at 16 that is making a big difference
08:48 twb Wow, the .debian.diff.gz for stalin is 11MB
08:48 lispy granted, I get to this mean: 736.6959 ms, lb 731.5933 ms, ub 741.7336 ms, ci 0.950
08:48 lispy just by using the :*: tuple
08:48 mornfall twb: It's probably including last phase output. :D
08:48 mornfall Post-opt, pre-codegen.
08:49 mornfall Otherwise, buildd's would choke.
08:49 twb Yep
08:49 twb Fuck em
08:50 mornfall From: debian@debian.org Subject: [VAC] Stalin has been uploaded, archive suspended for a month.
08:51 mornfall Another rollbaker.
08:51 mornfall Amaran rocks.
08:52 mornfall (Sorry I am still a bit incoherent post that filepath carnage.)
08:52 lispy filepath carnage?
08:54 mornfall lispy: See darcs-users@. I ripped out most of the path handling.
08:54 mornfall I mean, ripped through. The paths are still handled.
08:55 mornfall But they are bytestring-ish and have a safer interface. And there are no longer half-a-dozen incompatible path representations either.
08:56 twb So there are half-a-dozen+1 now? ;-P
08:56 lispy mornfall: [darcs-users] [patch343] First stab at a hashed-storage 0.6 port.(and 14 more)   Petr Ro?kai ?
08:57 mornfall That one yes.
08:57 mornfall twb: They reduced to type Foo = Bar.
08:59 lispy mornfall: One of the best things we can do to fix the pathing stuff is throw away our whitespace encoding/decoding
09:00 lispy mornfall: it would be much better if paths were stored in some other form, such as <length of path><path>
09:00 lispy So that you, do len <- readInt; fp <- take len;
09:00 lispy and then fp is just done.  No whitespace non-sense to deal with
09:01 lispy (stored, I mean serialized)
09:01 * lispy should be asleep
09:03 mornfall lispy: That's true, but we can't change it without breaking formats.
09:03 mornfall lispy: At which point we can throw out all of the parser and start over with a much simpler format.
09:04 lispy mornfall: well, if I can get this parser stuff to a point where you won't be grumpy if I stop working on it the first item on my list is to write a Cereal instance for Named (FL Patch)
09:05 mornfall cereal? :)
09:05 lispy I can't spell
09:05 mornfall No that's ok.
09:05 mornfall I'm just wondering what it was.
09:05 lispy http://hackage.haskell.org/package/cereal
09:05 lispy It's a better version of binary
09:06 lispy faster, stricture, more better
09:06 mornfall Hackage's back.
09:06 mornfall Lennart. :D
09:06 lispy oh yeah, it is.  I just used it without even thinking about it
09:07 mornfall Storable is still faster, presumably. But you have to do nasty stuff that way.
09:07 dons joined #darcs
09:07 lispy I was thinking we could pretty easily modify the patch parser to be backwards compatible without things are now, but to also check for a 'magic number' sort of thing.  IF the magic number is there it uses cereal to parse it instead of text based parsing
09:07 mornfall (That's how _darcs/index is done -- pre-computed size, Storable poking into mmap...)
09:08 lispy mmap is the devil
09:08 mornfall Well, it's *fast*.
09:08 lispy yeah, until it breaks
09:08 lispy or you need to mmap something large
09:08 mornfall True. But it doesn't seem to be breaking.
09:08 mornfall It even seems to work on windows, most of the time.
09:08 lispy the index might be fine.  I'm worried about mmap patches though.
09:09 mornfall Ok, that's a different story yes. :)
09:09 lispy anyway, I need sleep
09:09 mornfall Goodnight. :)
09:57 kowey joined #darcs
09:59 kowey good morning
10:08 kowey latex2html -long_titles 2 will help our user doc stability (getting_started.html instead of node2.html)
10:08 kowey now to fix the section names...
10:14 Igloo__ joined #darcs
10:17 mornfall Hmm! :)
10:34 kowey "I am increasingly thinking that Darcs' set-of-patches model would work better for this than Git's DAG-of-commits model."
10:34 kowey hmm, wonder what that's about http://lackingrhoticity.blogspot.com/2010​/08/my-workflow-with-git-cl-rietveld.html [have not read, just scanned]
10:48 * mornfall implements more neat features into cmdlib...
10:48 mornfall (default command, possibility to disable the "help" command, option to not permute options and non-options)
10:50 kowey why would disabling "help" be a good idea?
10:51 mornfall if your commands start all with @ and non-@ things fall through to a default command which runs shell...
10:51 mornfall (my current use-case... there may be others)
11:00 mornfall kowey: Have you seen my arguably-cool patch for h-s 0.6 (and more importantly, path handling)?
11:03 kowey not yet (aside from seeing the title), but hopefully we can chat more about it after my last meeting with abuiles
11:04 mornfall When is that? (Not weekend this time, presumably?)
11:16 kowey in about 45 min
11:16 * kowey continues to unpile...
11:18 mornfall Ah.
11:19 kowey hmm, we may need to nail down our terminology for caches and sources a bit
11:19 mornfall I also opened the path to help --manpage, since "help" is now a regular command (which gets a list of all commands, including itself :D).
11:19 kowey ah, it used to be baked in?
11:19 mornfall Yes.
11:20 mornfall It does get added by default by the dispatcher, but you can say dispatch [NoHelp] to suppress that. And provide your own help command if you like.
11:20 mornfall I guess not many people are going to use that, though.
11:29 kowey (as an aside, I think I've mentioned hsgtd as a potential user, the small twist being its shell-like ui)
11:30 mornfall Fortunately cmdlib works with [String] input and does not call getArgs itself. :)
11:31 mornfall So you can just dispatch on whatever words and readline give you.
11:32 kowey should try that, was blocked on cmdargs not accepting non-String for argpos args
11:32 kowey although Neil's fixed that in his big rewrite
11:34 mornfall Oh.
11:34 mornfall I guess I can't do that either. :(
11:35 mornfall It seemed tricky to do.
11:35 kowey uh-oh! http://code.google.com/p/ndm​itchell/issues/detail?id=221
11:36 kowey ah, so much adhoc command line handling in the haskell world to set alight (flamethrower)
11:36 kowey whether the flamethrower be cmdargs or cmdlib, there's quite a bit of code to burn: twidge, cabal-install, ghc-pkg, hsgtd, criterion...
11:37 mornfall cabal-install isn't going to fly
11:37 kowey minimal deps required?
11:37 mornfall it's using what Cabal provides which has to be very portable
11:37 mornfall which rules out Data.Typeable, presumably
11:37 kowey that would make sense
11:38 mornfall (for Setup.hs)
11:38 mornfall I mean, what Setup.hs uses is also what cabal-install uses, and the former needs to be portable. :)
11:39 mornfall I'll have to look in parsing argpos-thingies.
11:39 mornfall (I guess I call them Positional though.)
11:40 mornfall (Maybe it's not so tricky afterall.)
11:49 mornfall kowey: Fixed.
11:50 mornfall (Will be in 0.2.)
11:50 kowey not so tricky after all?
11:52 mornfall Nope. Just needed to flip a set to a parse.
11:52 mornfall (The only tricky bit was that parse had not been exported from the flag wrappers, which now is and all is good...)
11:53 mornfall 3-line patch overall.
11:53 mornfall (I still consider positional arguments to be slightly gay. :P)
11:55 abuiles joined #darcs
11:55 abuiles hi
11:55 abuiles kowey, hi
11:57 kowey hey, abuiles
11:57 kowey well! congratulations on a successful GSoC!
11:57 kowey (and thanks for your kind words on the final blog post)
11:58 abuiles thanks to you, It has been a nice experience :)
11:59 kowey so any feedback for us on how we can improve the Darcs GSoC experience for 2011?
12:00 * mornfall wonders how much truth's in that an adult man can live off a bowl of rice a day... already had two today and I'm fairly sure my balance is still negative
12:00 mornfall (and it's just 2pm, too)
12:01 kowey hmm, I suppose the size of the bowl could factor in
12:01 mornfall Obviously. :D
12:02 abuiles kowey, well, certainly mention about downloading the wiki ;), so is easier not to lose the changes...
12:02 mornfall But mine isn't exactly tiny. If I fill it to the rim, it's probably a reasonable western meal (but we still get 3 a day or so).
12:02 * kowey starts http://wiki.darcs.net/GoogleS​ummerOfCode/GuideForStudents
12:03 mornfall kowey: It'd look much nicer if it was wiki.darcs.net/GSOC/Guide instead.
12:03 mornfall GSoC if you like
12:04 kowey hmm, yeah
12:04 kowey I do wish gitit had a renaming mechanism for stability
12:05 mornfall But thumbs up. Hopefully we have enough documentation by the start of next GSoC.
12:05 abuiles kowey, your administrative thing should be a must follow for all the project... maybe linking to my project or mornfall's would be a good idea.
12:05 kowey abuiles: interesting, what do you mean exactly?
12:06 abuiles kowey, well, what we used to do with the weekly meeting and evaluate how much I was advancing, and giving percentages to each of the goals.
12:07 abuiles I always wanted to keep all the goals in done, and I would feel mad at myself seeing 50%, 75%
12:08 abuiles I think is more realistic to talk in terms of small goals, rather than a single goal ( which could have lots of small tasks).
12:09 kowey so the weekly meeting and percentages thing is something we should continue doing, but with more well-defined goals?
12:10 kowey or perhaps we should try something else?
12:11 kowey I'm not sure students getting mad at themselves for not reaching targets is a good thing
12:12 mornfall I also think we need a dual-mentoring scheme.
12:12 mornfall I.e. every student gets Kowey and one other mentor. :)
12:12 abuiles I think the weekly meeting is fine..
12:12 abuiles I don't mean mad as whipping myself.. maybe I choose the wrong word..
12:13 kowey :-) well, if you think it helped you to feel motivated, I'm reassured!
12:13 abuiles thanks, that's the word, motivation.
12:13 kowey there's different schools of thoughts on the whole goals thing, btw
12:13 kowey and there's a romantic side of me that's a bit reluctant to apply it
12:14 mornfall There seem to be interesting commonalities in English vocabularies of South American folks.
12:14 kowey but I hope the overall model of weekly meetings when the mentor helps the student to define his own goals and track them is correct
12:14 mornfall kowey: But you didn't apply it to me, did you?
12:14 abuiles agree
12:15 kowey I *might* have done a sort of primitive version of it in http://wiki.darcs.net/Goog​leSummerOfCode/2009-Hashed
12:15 kowey although I think I was a lot more prescriptive then in the "please work on X" sense :-)
12:15 abuiles I must admit that I needed more discipline, I failed with 3 blog entries :S
12:16 abuiles kowey, but I'm sure you know how to handle that ;)
12:16 kowey well, I appreciated the entries you did write :-)
12:16 * mornfall replaces otb.sh with otb.hs.
12:17 mornfall Seems to be working pretty well.
12:18 mornfall Now I wanted to plug in a --monitor option right... inotify to the rescue.
12:19 abuiles1 joined #darcs
12:19 abuiles1 sorry, ISP problems.
12:21 kowey so, any more tips for us?
12:22 * abuiles1 thinks
12:27 abuiles1 kowey, I can't think of any other at the moment,  the administrative side works fine for me as we did. I thought sometime as adding a 10 minutes meet in the middle of the week, maybe to resolve any doubts, but then I thought that's not necessary since you can always ask here and someone will help you
12:28 kowey OK, well thanks for the advice so far and do let us know if something comes to mind :-)
12:28 kowey so what's the future looking like for you?
12:28 kowey lots of work at University?
12:28 abuiles1 yeah, lots of readings
12:29 abuiles1 and midterms coming next week.
12:29 kowey already!
12:29 abuiles1 well, next week will be one month since it started...
12:30 abuiles1 also I must admit I can't understand still why Colombia's universities work so different to the rest of the world.
12:30 abuiles1 kowey, I forgot to mention http://wiki.darcs.net/FAQ#does-darcs-have-a-timeo​ut-feature-so-that-it-will-give-up-trying-to-pull​-from-a-repo-if-it-cant-find-it-after-x-seconds
12:31 kowey ah yes, thanks!
12:31 kowey hmm, I think we need a shorter URL for that
12:31 abuiles1 in fact, someone was asking about that, last week.
12:33 kowey nice!
12:33 kowey how's http://wiki.darcs.net/FAQ#ca​n-i-change-the-darcs-timeout ?
12:33 gbeshers joined #darcs
12:33 kowey as for universities, I think I've seen quarter, trimester and semester based systems
12:34 kowey and at least in the UK, people sometimes complaining about how crazy the scheduling of the breaks is
12:34 kowey funny to see Real World complexities, something like timing University holidays being contingent on a bunch of random factors
12:35 kowey like wanting kids to be available to work on the farm (in the past), or wanting to align with school holidays, etc
12:43 abuiles joined #darcs
12:43 * abuiles back
12:43 abuiles kowey, seems better the new name.
12:44 kowey thanks
12:44 abuiles kowey, also here we have normally like 3 midterms of 25%
12:44 kowey per year
12:44 abuiles not semester
12:45 abuiles the first semester is from january to end of may-beginning of jun, then from July to November.
12:45 abuiles and undergraduate is 4 and half year :S
12:47 abuiles kowey, I need to get ready for Uni, we'll talk next week..
12:48 kowey ok! see you then! (would be curious to hear about what darcs work you may fancy work on in your spare time)
12:49 abuiles I'd like to work in the interface with curl.
12:49 kowey :-)
12:50 kowey ok, well, it was really nice to work with you this summer... gotta go do work myself. catch you later!
12:51 abuiles left #darcs
14:52 sshc joined #darcs
14:54 burp joined #darcs
14:55 kowey joined #darcs
14:55 kowey issue1898 (the set-defaults notification) is surprisingly tricky
14:56 kowey have discovered that the --remote-repo flag is not being fixed
15:01 mornfall joined #darcs
15:01 mornfall joined #darcs
15:09 balor joined #darcs
15:34 lispy joined #darcs
15:42 mornfall Oh dear. Installing windows is such a chore.....
15:42 mornfall Will do later.
16:04 jadrian joined #darcs
16:05 jadrian I've never used a  distributed revision control system... for solo projects (single author) would there be any advantages of using darcs over svn?
16:07 kowey jadrian: have you ever used RCS?
16:07 jadrian kowey, only cvs and svn
16:07 kolmodin are there any severe bugs found in 2.4.98.2?
16:07 lambdabot kolmodin: You have 6 new messages. '/msg lambdabot @messages' to read them.
16:07 burp nah there is not a single advantage ;-)
16:07 kowey jadrian: OK, RCS was a version control system that pre-dated CVS, right? [sorry, I'm being long-winded]
16:07 jadrian kowey, I think I've heard of it
16:08 kowey jadrian: ... but there was one thing about it that would keep me using it even in situations where I knew I "should" have been CVS
16:08 kowey jadrian: and that was pure and simple laziness
16:08 kowey I hated having to go through the hoops of setting up the central repository
16:08 kowey hoops like thinking about where the right place for it would be, etc
16:09 kowey so sometimes I'd be really really really lazy and just use RCS and feel kinda guilty about it the whole time
16:09 kowey ...
16:09 jadrian I'm listening...
16:09 kowey sorry, this is a really roundabout way to answer your question
16:10 kowey one of the reasons I like distributed revision control systems is that on the one hand, they're a lot more powerful than the centralised ones (may be debatable...)
16:10 kowey but on the other hand, they're a lot *easier* to use in some ways
16:11 kowey with Darcs (same with Git, etc), I can have the convenience of just knocking out an instant-repo with darcs init (RCS-style convenience)
16:11 kolmodin kowey: do I dare let my users test to use darcs-beta 2.4.98.2? :) I thought of adding it, but masked so that you have to explicitly require this version
16:11 burp and it can be faster
16:11 kowey but without the guilt, jadrian
16:12 kowey dunno if this makes any sense? power and yet more convenience
16:12 jadrian kowey, yes I was just looking for specifics... for instance I moved from cvs to svn mainly for one motive, renaming and moving files
16:12 kowey *nod*
16:13 kowey kolmodin: well, it is a beta...
16:13 kowey but more testers would indeed be very welcome
16:13 jadrian when I think about a distributed system for solo work it confuses me a bit
16:13 kowey right, so what I'm saying is that distributed version control is even better for solo work than svn
16:13 burp don't let the "distributed" disturb you
16:13 jadrian right now I always commit and I know that the central repository contains my latest version
16:13 kolmodin kowey: sure, but you have at least not found any dangerous data corrupting stuff so far? :D
16:13 jadrian and all the tags
16:14 kowey because the "distributed" part means you can just do "darcs init" and go, which eliminates the laziness barrier to getting started
16:14 jadrian I see
16:14 kowey the tradeoff to using distributed version control is that (A) being able to locally commit [we call this record] patches and push them at leisure is great
16:14 kowey but (B)... sometimes you forget to push
16:15 kolmodin I use a huge amount of short lived darcs repos
16:15 kowey (for [B], the bzr people have a special command that you can tell users to use, which would force all commits to be centralised-style]
16:15 jadrian ah sounds good
16:15 kolmodin and it's so easy to "darcs get foo foo-branch-feature-x" that might or might not get pushed back to the master repo, without having to bother people with my branch if I decide to remove it
16:15 jadrian what's bzr though?
16:16 kowey but jadrian: for me, the usefulness of (A) plus the laziness factor far outweigh the inconveniences caused by (B)
16:16 kowey bzr is a distributed revision control system, a competitor :-)
16:16 jadrian kowey, got it
16:16 jadrian ah, hadn't heard of that one
16:16 kowey normally, most people have heard of git, hg and bzr but not darcs
16:16 burp hm, haven't tried bzr, but the only usable vcs are git and darcs for me :>
16:17 jadrian I've also heard of mercurial actually
16:17 kowey as for why you may want to consider darcs... well, that's a separate discussion :-)
16:17 burp jadrian: hg is mercurial
16:17 jadrian darcs has been of my radar for a long time because I use haskell a lot
16:17 jadrian burp, oh
16:18 burp jadrian: chemical symbol for mercurial ;)
16:18 jadrian burp, yeap I got it now :)
16:18 kowey in my opinion, deciding to use darcs over git/hg/bzr requires a similar leap as deciding to use distributed revision control over centralised
16:18 burp I don't really like hg, it's missing so much usefull stuff that darcs and git have
16:18 kowey not saying it's necessarily better, but there's a certain ah-hah! in seeing what makes it compelling
16:19 jadrian right, so much choice so little time
16:19 kowey but helping people get to that "ah-hah!" is still a bit tricky because on the surface, they all look the same
16:20 kowey but don't let the choice overwhelm you; the sooner you go distributed, the better, imho
16:20 mornfall kowey: About those witnessed path types, I am not sure. Also the package you mentioned has a *huge* API.
16:20 burp hg has the strange concept of "closing branches"� once you opened one you can't just trash them :/
16:20 burp so the nice throwaway branch technique available to git and darcs is not possible
16:20 mornfall kowey: I'll keep to "just" keeping the paths sane by only exposing sane operations on them.
16:21 mornfall kowey: (And distinguishing relative/absolute, since that's a path property.)
16:21 kowey mornfall: ok
16:21 mornfall kowey: Whether something is a file or directory, well that tricky.
16:21 mornfall that's
16:22 mornfall Witnesses have the downside that anything that comes from outside needs to be annotated.
16:22 sm jadrian: yes, there are advantages to darcs over svn for a single user, such as: simplicity of creating and storing the repo; patch cherry-picking; hunk cherry-picking; ease of sharing repos/patches with others
16:22 mornfall I would like this path stuff to be simple and easy. Maybe not super-safe...
16:23 jadrian sm, alright sounds good
16:23 sm the ability to add commits wiki-style using gitit...
16:24 jadrian pretty conviced it's worth the try then
16:24 * sm can't think of anything else significant right now.. anyone ?
16:25 kowey cherry picking may sound like some fancy thing, but it actual has very simple/practical implications
16:25 zooko joined #darcs
16:25 kowey like being able to safely undo something even though you've commit lots of stuff on top of it
16:28 lispy1 joined #darcs
16:30 mornfall The downside is that SVN doesn't mind a repository that has a gigabyte of openoffice crap in it.
16:30 mornfall Whereas darcs *does* mind.
16:33 sm yes, there's a couple of disadvantages too
16:34 kowey another nice feature of CVS and SVN is subtree checkouts
16:38 burp hm, git can do that too right?
16:39 burp aka git submodule
16:40 gh_ joined #darcs
16:40 gh_ hi
16:40 kowey I'm afraid I don't know my way around git enough to comment.  But http://lackingrhoticity.blogspot.com/201​0/08/cvss-problems-resurface-in-git.html seems to complain
16:40 gh_ jadrian, did you know about this page http://wiki.darcs.net/DifferencesFromSubversionCVS ?
16:41 jadrian gh_, ah nice didn't know about it
16:42 kowey yay! yes, Conspicuous Use of Archives http://producingoss.com/en/​growth.html#using-archives (which I keep forgetting about)
16:42 zooko lispy told me last night that maybe GHC has been ported to OpenBSD/sparc64. Does anyone know about that?
16:42 zooko The Tahoe-LAFS project is about to set up a darcs->bzr gateway so that we can get our OpenBSD/Sparc64 contributor able to do buildbot builds.
16:43 kowey perhaps the folks at #ghc would know something?
16:43 burp wish there was ghc on openbsd/macppc for my router :>
16:44 kowey oh, I see zooko has filed http://hackage.haskell.org/trac/ghc/ticket/3511
16:45 zooko Yes, that ticket hasn't been updated, but lispy's comments made me think maybe GHC hackers made progress but they didn't know about GHC ticket 3511.
17:12 arjanb joined #darcs
17:37 zooko joined #darcs
17:46 exlevan joined #darcs
19:30 tux_rocker joined #darcs
19:31 tux_rocker hi
19:32 tux_rocker i see "src/Darcs/Commands/Convert.lhs:190:51:     y  not in scope because it has a wobbly type (solution: add a type annotation)"
19:32 tux_rocker when pushing stuff to the HEAD darcs darcs repo
19:32 Heffalump that's with 6.8, which we've dropped support for
19:32 tux_rocker hmmm
19:32 Heffalump but darcs.net hasn't been updated yet
19:33 tux_rocker what distro is it on?
19:33 Heffalump not certain. An old Debian I think.
19:34 mornfall Debian stable.
19:34 mornfall Eric said he'll do something about that.
19:35 Heffalump ok, current Debian stable
19:35 Heffalump I guess we may be optimistically hoping for the next Debian release before our next release
19:39 mornfall It could happen. Although it could also not happen.
19:40 burp testing is already in freeze afaik
19:40 Heffalump hmm, so on our GHC support wiki page we promise copy and pastable instructions for installing a recent GHC on Debian stable, but on our Debian wiki page we just provide textual instructions (which I think I wrote)
19:40 Heffalump we should resolve this conflict
19:41 mornfall burp: It is. But that doesn't need to mean anything (speaking with my DD hat on). : - P
19:41 mornfall Regarding the actual time of release, anyway.
19:41 mornfall But I am hopeful.
19:43 burp ah I see :D
20:05 Igloo joined #darcs
20:15 lispy1 mornfall: I was just talking to dons about the parser.  he recommended that my strict tuple look like this data State a = !a :*: !B.ByteString
20:15 lispy1 mornfall: then ghc will be able to specialize it more better on the ByteString side
20:15 lispy1 mornfall: were you able to do any benchmarking with the repo I shared?
20:16 mornfall lispy1: Not yet. I am stuck with other stuff.
20:17 lispy1 He said I can get even better if I know "a", but in this case I can't
20:17 mornfall lispy1: Btw. what was that about PReach and galois? You know anything about that?
20:17 lispy1 mornfall: sorry, I missed that talk
20:17 lispy1 mornfall: it's here if you want to watch it http://www.galois.com/blog/2010​/08/09/tech-talk-video-preach-–-a-distributed-murphi-based-model-checker/
20:18 lispy1 oh wait, no I went to that talk
20:18 lispy1 what am I saying :)
20:18 mornfall I know I have seen the vid (or, most of it).
20:18 lispy1 Did you have a more specific question?
20:18 mornfall I have been in the distributed model checking business for a couple years now. :)
20:18 lispy1 Are you already familiar with model checking?
20:18 lispy1 ah
20:19 mornfall I was just wondering what was galois doing with modelchecking.
20:19 mornfall I have a sketchy idea what the company is doing overall.
20:19 lispy1 we have a lot of formal methods people
20:19 lispy1 have you seen our work with NASA?
20:20 mornfall Unlikely. From NASA, all I know is SPIN.
20:20 lispy1 http://www.galois.com/blog/2008/11/11/​galois-awarded-nasa-research-contract/
20:21 lispy1 I wish I had more bg in model checking.  We had an internal course in Isabelle and that gave me some FM background, but not enough :)
20:21 lispy1 http://www.galois.com/blog​/category/formal-methods/
20:22 mornfall Model checking is kind of boring. It's mostly about cramming as a big model into as little RAM as possible. :)
20:22 mornfall And exploring it as fast as possible.
20:24 mornfall (Well, if galois wasn't so damn far away, I'd be tempted.)
20:34 Igloo joined #darcs
20:38 Heffalump have you considered applying for an internship there?
20:39 britneypire joined #darcs
20:40 mornfall It's half the world away.
20:41 lispy1 mornfall: we have an intern this year.  I bet we'll have more next year.
20:42 mornfall lispy1: But galois only exists in the US right?
20:42 lispy1 mornfall: Yes.  So far :)
20:43 mornfall Nevertheless, I already have a job and I am not intending to move out of Europe anytime soon. When the Muhammad doesn't come to the mountain...
20:47 Heffalump surely as a PhD student you can stop PhDing temporarily for internships?
20:47 mornfall I mean a real job, aside from PhD. :P
20:48 Heffalump ah :-)
20:48 Heffalump is your PhD part-time then?
20:48 mornfall My everything is part-time. Although officially my PhD is probably fulltime.
20:48 mornfall (Even my existence is parttime. :)
20:50 Heffalump that sounds complicated :-)
20:50 mornfall It sure is.
20:51 mornfall And my model checker keeps segfaulting.
20:51 mornfall I hate MPI.
21:02 mornfall Oh dear lords.
21:05 Heffalump ?
21:07 mornfall Nothing really. I am too tired it seems. :|
21:07 mornfall Simple things take long time.
21:08 kowey joined #darcs
21:08 Heffalump newset makes rebase much faster, which is nice :-)
21:09 Heffalump hey kowey
21:09 mornfall Cool. :)
21:09 * mornfall upgrades valgrind.
21:09 mornfall This old junk is missing exceptions.
21:10 kowey hey Heffalump, all
21:10 mornfall Evening kowey.
21:11 kowey hmm, should really announce the sprint soon so people can plan travel
21:11 mornfall Yes, should.
21:13 lispy1 There is a sprint?
21:14 kowey it's that time of year: http://wiki.darcs.net/Sprints/2010-10
21:15 mornfall Heffalump, kowey, btw. do you think the huge path business restructuring is feasible? (for darcs 2.6)
21:15 kowey we've just been harvesting availability/dates so far
21:15 kowey but it's time to fix the date, I think
21:15 mornfall (the h-s 0.6 patchset)
21:15 Heffalump mornfall: I only skimmed so far. I'll take a proper look now.
21:15 exlevan joined #darcs
21:15 mornfall It is a bit risky I guess, but I speculate that it's a good thing.
21:16 kowey mornfall: I'm afraid I haven't really been following along yet, sorry!
21:16 mornfall (To rip out all those incompatible path representations.)
21:16 kowey but it seems like if we agree that something is good in principle
21:16 kowey and if it may be risky
21:16 mornfall kowey: Yes, that's what I was asking, primarily.
21:16 kowey then earlier may be a lot better than later
21:16 lispy1 I think 1) it NEEDS to happen.  2) vetting it for corner cases is extremely important.
21:16 mornfall It doesn't make sense to do if people are happy about what we have now.
21:16 lispy1 Can you use filepath to sort of verify it against?
21:16 kowey lispy1: so if you fancy a trip to France...
21:16 Heffalump I'm not happy with the code. I'm not sure whether we've had any actual reliability problems with it, though.
21:17 kowey 15-17 October, pending final confirmation from Florent
21:17 lispy1 kowey: oh, I don't think Octember will work for me
21:17 Heffalump in the abstract, I like the idea of Ben Moseley's "safe" filepaths, but we already have lots of type witnesses in darcs
21:17 kowey one day! one day we'll bring darcs-USA and darcs-Europe together
21:18 mornfall Heffalump: Well, it seems to work. But I am worried every time I need to touch some path since we have a few incompatible representations.
21:18 Heffalump (but that's a separate issue arising from kowey's suggestion, not a specific point about your patch)
21:18 mornfall Heffalump: And darcs keeps re-norming paths at random places which I am not very happy about either.
21:18 kowey is that the path fixing code?
21:18 lispy1 kowey: but maybe I can run another sprint here at Galois in Octember
21:18 mornfall Yes, I was thinking about witnessing paths, but I am a bit worried it's too complex for general-purpose coding.
21:18 Heffalump in principle re-architecting it should be fairly mechanical, right? Fix on the library and then change existing code to use it.
21:18 kowey one potentially stupid question: do we need to care about remote paths too?
21:19 kowey oh! now I understand what Octember means, gotcha
21:19 Heffalump I mean that the change of existing code should be fairly mechanical once the library is selected.
21:19 mornfall Heffalump: In darcs?
21:19 Heffalump yes
21:19 lispy1 kowey: ah yeah, should actually be December since it's 10th month
21:19 Heffalump I guess you've actually already done that bit? I didn't read your patch fully.
21:20 mornfall Heffalump: Well, there are lots of subtle cases and a lot of code that does its own path munging that depends on things like leading dot, or no leading dot... it's pretty scary.
21:20 Heffalump I guess another question is how this interacts with the whole withCurrentDirectory issue.
21:21 mornfall I tried to get everything fixed, but there are places where I had to resort to feeding pre-refactor style paths to blocks of code.
21:21 mornfall I think the worst is Population and related code.
21:22 mornfall In other words, I couldn't get annotate to work without stuffing it with string-based paths, which it then repacks and unpacks and repacks a few times for a good measure.
21:22 Heffalump ouch
21:23 mornfall (It has a completely custom tree representation with own set of path functions which are embedded in the tree traversal code. It's a major pita.)
21:23 mornfall | BC.unpack (nameI i) == "." =
21:23 mornfall | BC.unpack (nameI i) == takeWhile (/='/') d =
21:24 mornfall where dropDS ('.':'/':f) = dropDS f
21:24 mornfall etc.
21:24 mornfall I'll probably have a go at it later.
21:26 mornfall "Windows has detected that the computer is connected to network." Congrats.
21:31 sshc_ joined #darcs
21:36 lispy1 mornfall: ah yeah.  We almost need to put a temporary ban on String in the darcs source
21:36 lispy1 Just till we sort all the packing unpacking
21:39 sshc_ joined #darcs
21:39 Igloo joined #darcs
21:39 arjanb joined #darcs
21:39 mornfall joined #darcs
21:39 burp joined #darcs
21:39 dons joined #darcs
21:39 FunctorSalad_ joined #darcs
21:39 pierreee joined #darcs
21:39 kolmodin joined #darcs
21:39 sm joined #darcs
21:39 dcoutts joined #darcs
21:39 Jaak joined #darcs
21:39 Korusef joined #darcs
21:39 alexsuraci joined #darcs
21:39 tmug joined #darcs
21:39 thorkilnaur joined #darcs
21:39 dino- joined #darcs
21:39 alpounet joined #darcs
21:39 felipe joined #darcs
21:39 dleverton_ joined #darcs
21:39 kosmikus joined #darcs
21:39 idnar joined #darcs
21:41 tmug joined #darcs
21:41 Korusef_ joined #darcs
21:41 mornfall (You may notice that I got quite hooked on view patterns. :)
21:41 sshc joined #darcs
21:41 burp joined #darcs
21:41 alpounet joined #darcs
21:42 * Heffalump likes view patterns
21:42 Heffalump though I like F#'s active patterns more
21:42 Heffalump well, F#'s active patterns look nicer at the use site. The semantics at definition time are rather opaque.
21:42 * mornfall has never seen F#.
21:44 tmug joined #darcs
21:45 mornfall Are you sure?
21:45 mornfall Looking at F# wikibook, it doesn't seem very nice to me.
21:46 Heffalump well, at the use site they look just like a discriminated union
21:46 Heffalump which IMO is nice (if the pattern was well-designed to begin with)
21:51 mornfall F# looks a bit like funny ML.
21:52 lispy1 mornfall: F# is Ocaml on .NET
21:52 lispy1 mornfall: so it should look like ML :)
21:53 Heffalump well, it originated as OCaml. It's changed quite a bit.
21:53 Heffalump it's quite messy syntactically as a result
21:54 mornfall How Microsoft. : - P
21:56 Heffalump changing it made perfect sense for the niche it was designed for
21:56 Heffalump I reckon abandonning the OCaml compatibility would have been better though
22:00 mornfall What was it designed for, actually?
22:02 Heffalump FP cleanly integrated with .NET
22:06 mornfall I'm wondering how it compares to Scala.
22:08 Heffalump more functional
22:09 Heffalump better type inference, and nicer syntax for working with functions. [this is all hearsay, I know very little about Scala]
22:09 mornfall I guess I'll stick to Haskell. :)
22:10 mornfall (And, well, C++.)
22:11 burp hm, can I amend-record in an easy way just to change the commit message?
22:11 mornfall --edit
22:11 mornfall and then 'd'
22:11 burp or is unrecord + record the best?
22:11 burp ah thanks
22:11 mornfall Amend is safer (you won't accidentally mix up the patch with unrecorded changes)
22:25 mornfall clang: error: compiler command failed due to signal 11 (use -v to see invocation)
22:25 mornfall I guess the C++ support isn't quite ready yet... :(
22:51 mornfall If single item on your backtrace needs more than 20 lines, you wish C++ would use more compact names for template instances.

| Channels | #darcs index | Today | | Search | Google Search | Plain-Text | summary