Camelia, the Perl 6 bug

IRC log for #darcs, 2013-02-13

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

All times shown according to UTC.

Time Nick Message
01:00 mizu_no_oto joined #darcs
01:57 favonia joined #darcs
01:59 mizu_no_oto joined #darcs
03:38 mizu_no_oto joined #darcs
03:45 mizu_no_oto joined #darcs
04:53 preflex_ joined #darcs
05:44 favonia joined #darcs
06:52 favonia joined #darcs
06:57 favonia joined #darcs
08:38 gh_ joined #darcs
08:51 nomeata joined #darcs
08:58 raichoo joined #darcs
08:59 jeltsch joined #darcs
08:59 jeltsch Hi, I’m trying to install darcs-2.8.4 from source under Fedora 17.
08:59 jeltsch I get the error message, “setup: Unable to link against the iconv library.”
09:00 jeltsch Any ideas how to fix this?
09:04 gh_ jeltsch, looks like I'm experiencing the same issue with ubuntu 13.04 and HEAD
09:12 kmels joined #darcs
09:15 jeltsch gh_: On #fedora, I was told that iconv is part of glibc.
09:16 jeltsch So if you have the development packages for glibc installed, it might already work.
09:16 jeltsch But it doesn’t work for me.
09:16 jeltsch In darcs.cabal, I have found the option libiconv.
09:17 jeltsch Surprisingly, this option should be disabled by default, so -liconv shouldn’t be used by defaultj.
09:24 jeltsch gh_: Ah, I didn’t read the comment in the cabal file correctly. Disabled libiconv flag means that it is autodetected whether -liconv is necessary (and this autodetection seems to fail). Enabled libiconv means, “Just use -liconv”.
09:24 jeltsch I will try with an enabled libiconv flag now.
09:26 jeltsch Hmm, this doesn’t work either.
09:26 gh_ I installed the package libglib2.0-dev and compilation starts normally now
09:27 jeltsch Why can’t I just *disable* the use of -liconv. After all, this shouldn‘t be necessary if iconv is in glibc, right?
09:27 jeltsch gh_: You are lucky.
09:27 jeltsch gh_: But what has GLib to do with glibc?
09:27 gh_ at the configure stage, I had a message "checking whether to use -liconv... not needed."
09:28 gh_ jeltsch, no idea
09:29 jeltsch gh_: So probably, iconv is somewhere in the dependencies of GLib.
09:29 gh_ jeltsch, I'm compiling darcs HEAD, but the patches that introduce the iconv dependency should be the same in both branches
09:33 jeltsch SOLVED. :-)
09:34 jeltsch gh_: I also had to install curl development files. After I had done this, it was detected that iconv is not needed.
09:34 gh_ oh
09:35 gh_ I did that too :)
09:35 gh_ so lack of libcurl makes cabal output an iconv error
09:37 jeltsch gh_: Yes, this is annoying. :-(
09:37 gh_ jeltsch, one long-term wish we have with darcs is to get rid of curl, using a 100% haskell replacement
09:38 gh_ jeltsch, but there's not really anyone working on that :(
09:47 owst joined #darcs
10:28 jeltsch left #darcs
10:34 schlaftier joined #darcs
11:07 lelit joined #darcs
11:08 gal_bolle_ joined #darcs
11:08 gal_bolle hi all
11:09 owst Hey gal_bolle
11:09 florent_ joined #darcs
11:10 gal_bolle regarding the sprint, I could not get rid of my teaching on friday, I'll join you saturday and sunday
11:12 owst gal_bolle: what are you teaching?
11:13 gal_bolle I don't really know ;-)
11:13 gal_bolle It's a module called applications of regalar languages
11:13 gal_bolle for which I have 32 hours of content to find
11:13 owst ...urk!
11:14 gal_bolle so it's becoming "stuff about text algorithm gal_bolle likes"
11:14 owst *applications* of regular languages, that's tough
11:14 gal_bolle s/hm/hms/
11:14 gal_bolle tough indeed
11:14 gh_ hi gal_bolle
11:14 gal_bolle so I got rid of both the "applications" part
11:14 gal_bolle and the "regular languages" part
11:14 owst ;-)
11:15 gal_bolle it's turning out ok
11:15 gal_bolle these kids just need some algorithmic exercise
11:16 gal_bolle and text algorithms are nice as the theory to concrete application threshold is thin
11:16 gal_bolle s/threshold/gap/
11:21 iago joined #darcs
12:18 owst joined #darcs
13:21 mizu_no_oto joined #darcs
13:44 plat0 joined #darcs
13:45 plat0 Hi all.  What are the arguments for using git instead of darcs these days?  I love darcs (and dislike git).  Is darcs performance still unpredictable enough that developers want to avoid it?
13:47 owst plat0: conflict handling in darcs is broken in some (complicated) cases.
13:47 owst plat0: General performance and usability are fine for darcs.
13:48 owst the git model also lets you identify unqiue revisions much more easily than darcs "what happened, when?" is an easier question to answer.
13:49 plat0 I guess I have to learn more about git.
13:49 plat0 I was under the impression that darcs merges could hang indefinitely, though maybe that problem was solved years ago.
13:49 owst Some of that problem is solved.
13:49 owst I wouldn't say it's solved
13:49 plat0 OK
13:50 owst The solution breaks other things (some operations fail, that shouldn't, for instance)
13:51 plat0 git and mercurial seem a lot more complicated than darcs
13:51 plat0 which is why I am reluctant to really start using them
13:53 owst It seems that way to me too
13:53 plat0 Maybe I don't understand them properly, but darcs seems to stay out of my way and they seem to get in my way
13:54 plat0 I guess there's one benefit to having multiple branches in the same directory, and that is if you are compiling a large project you avoid a lot of recompilation if you switch to a different branch.
13:55 plat0 Other than that, I can't see the point of them.
13:58 gh_ that's already a good reason
13:59 plat0 It is indeed.
13:59 plat0 But is there another?
14:05 dixie probably other reason is better encapsulation of the project history. Sparse copies of repository for each branch is easier to be lost than in-repository branches...
14:18 mizu_no_oto joined #darcs
14:45 iago owst, I would say that the Darcs worflow may require tagging more frequently
14:46 owst iago: sure, for performance reasons, which sucks.
15:01 sm_ joined #darcs
15:04 gbeshers joined #darcs
15:06 iago plat0, in what sense mercurial feels *a lot more* complicated than Darcs?
15:07 iago git UI is terrible, but IMO Hg is quite sane
15:07 MasseR plat0: Featurebranches. You have a master branch which contains your stable version. You got a dev branch which gots completed features that haven't been vetted yet. Then you have feature-branches which start from the development branch where all bets are off. And of course you could have something like production-bugfixes branch which is basically as stable (or by definition more stable than) as the master branch
15:17 mizu_no_oto joined #darcs
15:28 plat0 iago: Well I haven't used Hg at all so perhaps I am wrong!  I got the impression it is quite like Git from the tutorials I read.
15:28 plat0 MasseR: Sure but you can do that with Darcs's model.
15:30 MasseR plat0: Yes and no
15:30 MasseR Gits branches are more powerful than darcs's implicit branches
15:30 MasseR Yet they are (immensely) lighter than real darcs branches (clones)
15:31 owst Depends what you mean by "more powerful"
15:32 MasseR owst: Complete control which branch a commit belongs to. As far as I understand if I create a commit in darcs that depends on some other commit thats in 'other' branch, the new commit is automatically in that branch. Whether I want it or not
15:33 owst So you want a patch, without its dependency?
15:33 MasseR And of course the tree model is more intuitive
15:33 owst What does it mean for a commit to belong to a branch?
15:33 owst Git doesn't have anything of the sort
15:34 owst It "belongs" to a branch if you can reach it by traversing the DAG starting at the branch leaf pointer
15:34 MasseR Fine, the DAG model is more intuitive :)
15:35 owst If you say so ;-)
15:35 MasseR Branch A has its history wherein branch B has its own history. Whatever I do in A doesn't affect B (unless I explicitly make it happen)
15:35 owst The same is true in darcs
15:35 owst Repo A and Repo B are distinct, unless you push/pull between them
15:36 MasseR What happens in darcs in this situation:
15:36 MasseR I'm working on feature A, it consists of multiple files
15:36 MasseR Suddenly I need to work on B
15:36 MasseR But in B I need to make modifications where I have modifications in A
15:37 MasseR At least I don't have the same baseline branch, the modifications to A are visible in B
15:38 MasseR But the real question was, that B isn't independent of A anymore isn't it?
15:38 MasseR With git and featurebranches they would be separate (as they would start on their own baseline)
15:38 gh_ yeah, in that case you clone your darcs repo and work on B
15:38 iago plat0, then maybe you should give Hg a try first, Hg + a couple of extensions is what I'm using since a year ago and I'm very happy with that, I rarely miss Darcs
15:39 plat0 iago: Yes I will do, thanks.  What extensions do you use?
15:40 MasseR gh_: Then we get back to the "but git branches are immensely lighter than darcs" :)
15:40 MasseR (creating a branch in git is in milliseconds, in darcs it could be minutes)
15:40 gh_ MasseR, I agree.
15:41 gh_ MasseR, an alternative would be to "stash" your A changes using "darcs obliterate -O" and then work on B in the same repository
15:41 iago at least record, fetch, color and extdiff; but also others like crecord (this is particularly cool, although darcs hunk editor is more powerful)
15:42 MasseR gh_: Might be me, but I'm bad with stashes. After a month I have multiple levels of stashes and I don't have a clue what they consist :D (with git)
15:43 gh_ MasseR, the "stashing" you do with "obliterate -O" has the form of a patch bundle in your repo directory, that you can later apply with "darcs apply"
15:43 iago plat0, transplant is useful too, although you won't use it unless you use branches often
15:43 gh_ MasseR, so maybe that's more explicit to you like this
15:44 MasseR gh_: Could be. I'll try to remember that
15:44 plat0 iago: thanks
15:46 MasseR (but then again, branching and faster clone/pull/push/etc are the only things I miss from git. For now anyway)
15:46 MasseR Does darcs have something like bisect?
15:47 gh_ MasseR, yes "darcs test --bisect"
15:47 MasseR Yup, then my previous statement is true
15:47 gh_ MasseR, althought "darcs test --backoff" may be faster in many cases
15:48 donri joined #darcs
15:48 MasseR gh_: How come?
15:49 gh_ MasseR, because --backoff starts close to the end of the history and then does some exponential jumping in the past
15:50 plat0 MasseR: Can you briefly summarise you thoughts on Git vs Hg?
15:50 MasseR plat0: Nope, never touched hg
15:50 plat0 Oh OK
15:51 MasseR But as I understand it has a worse branching model and if you read my lines above, I appreciate highly good branching
15:51 MasseR gh_: That's nice. Especially if it's more efficient than bisect
15:51 gal_bolle MasseR: regarding branch creation, with darcs get --lazy, you get branch creation within seconds unless your repo is really huge
15:52 owst gal_bolle: wouldn't --lazy just prevent the hard link creation? (Isn't that fast, anyway?)
15:53 MasseR gal_bolle: A while back when trying to undestand darcs, I tested cloning some repos with and without lazy
15:53 lelit it does not copy the whole history
15:53 MasseR It's still magnitudes slower than git
15:53 gal_bolle well it avoids the parsing of inventories before hard-linking
15:53 MasseR Especially with cold cache
15:53 owst gal_bolle: right
15:53 gh_ owst, hard link creation of a big repo (like darcs' one) takes time
15:54 * gh_ wonders whether --lazy should be default
15:54 owst gal_bolle: so it just copies the inventories, and the pristine cache?
15:54 gal_bolle cache shouldn't matter for --lazy, but you're right that it's still slower than in git
15:54 gal_bolle owst: i think so, and working
15:54 gh_ owst, latest inventory I think
15:54 owst gh_: no, not until we are able to tell users why darcs suddenly "hangs"
15:54 owst right
15:55 owst That's sort of what I thought happened
15:55 gal_bolle gh_: it could be for local get
15:55 MasseR gal_bolle: Don't know about if it's cache but first run takes usually magnitudes longer than subsequent runs. And even the subsequent runs are magnitudes slower than git branches :)
15:55 owst Ah yes, local could be fine
15:55 gal_bolle but then you have to decide how smart we are about the "local" part
15:55 owst gal_bolle: yeah, NFS anyone?
15:55 owst :-)
15:55 gh_ yeah, darcs already proposes to ctrl+c when doing get
15:56 gal_bolle we could use darcs branch as an alias for darcs get --lazy
15:56 gal_bolle if we feel gittish
15:57 owst I'd rather not associate the term "branch" with darcs, until people can recognise a feature that is somewhat like branches elsewhere
15:57 owst (I don't feel gittish)
15:57 owst :-)
15:57 gal_bolle but I think that the conclusion of last sprint was that in-repo branching was something we'd eventually get
15:57 owst Yeah, I think so
15:57 gal_bolle keyword being eventually, of course
15:57 MasseR gal_bolle: Rather not. IMO either no branches or good/better branches :)
15:58 MasseR :s/good/equal/
15:58 owst gal_bolle: I think branches will exasperate the broken conflictors stuff, so they'll need to be fixed at the same time
15:59 owst (certainly it happens when importing git branches into darcs via the (broken) bridge)
15:59 gal_bolle owst: why would that be?
15:59 owst People will start using branches ;-)
15:59 owst By that I mean, creating more complicated conflict situations, and expecting darcs to handle it ok
16:00 owst Whereas now, I'd guess people stay away from branches
16:00 owst Maybe not, it's just a feeling
16:00 gal_bolle MasseR: what do you mean by good branches? What are your requirements?
16:00 MasseR gal_bolle: In this situation I mean speed :)
16:00 MasseR Effortless branching
16:01 MasseR I don't enjoy waiting for cli commands :P
16:01 gal_bolle I think the bridge part would be the cause for conflict rather than the branches part (or at least, we should check that it is by real world experiments)
16:02 gal_bolle MasseR: I think you'd get good performance easily if we implemented in-repo branching
16:02 owst Maybe so, but I'm importing a real-world repo, that uses Git branches, so people clearly use branching in the way that triggers it
16:02 MasseR gal_bolle: No objections to that. Only to the clone --lazy
16:02 gal_bolle MasseR: makes sense
16:03 MasseR owst: I thought darcs was supposed to handle conflicts better than competitors. What do you mean with problematic conflicts?
16:03 gal_bolle owst: and also because they are using brain-dead git+bridge rather than excellent darcs to record with
16:03 owst gal_bolle: ;-)
16:03 gal_bolle (I'm not saying what's the precedence of operators concat and + in "brain-dead git+bridge"
16:04 owst MasseR: the answer is: "maybe".
16:04 owst MasseR: I don't know what you mean by "handle conflicts"
16:04 gal_bolle MasseR: it's nice in more cases, but when it's ugly, it's uglier
16:05 MasseR gal_bolle: "uglier"? As in total fubar or a bit more effort for the user to fix conflicts
16:05 gal_bolle and some of it's niceness (is/used to be) ugly too :-(
16:06 gal_bolle total fubar can happen but is rare (I think), more effort from the user to understand the situation (ugly conflict marking) does happen
16:07 MasseR Sorry stupid questions, but I'm curious. By ugly conflict marking do you mean invalid markers or for example too wide markers
16:07 gal_bolle didn't we have a page on the wiki with good examples? let me look
16:08 owst Markers don't have any sensible order (and don't tell you which patch they came from)
16:08 MasseR owst: Thank you, that would be important
16:09 MasseR owst: But wouldn't that be an UI issue? Nothing too fundamental
16:09 gal_bolle and you darcs tends to miss the fact that 3 of your 7 conflicting versions are in fact identical
16:10 gal_bolle it's an UI issue that's hard to correct because of a fundamental deficience of niceness
16:10 gal_bolle it's not purely UI, but it can be corrected without rewriting the whole of darcs
16:11 gal_bolle heffalump did make the situation appreciably better, but not nice enough yet
16:12 owst gal_bolle: as we know Duplicates are strange, and wrong. :-)
16:13 owst s/know Duplicates/know, duplicates/
16:13 gal_bolle yes, they should be reported to the authorities as plagiarism and the user shot on the spot by a DMCA-drone. Simpler for everyone involved
16:14 owst heh
16:18 gbeshers joined #darcs
16:28 gal_bolle see you later guys
16:32 raichoo joined #darcs
16:49 whaletechno joined #darcs
16:59 nomeata joined #darcs
17:50 gbeshers joined #darcs
18:47 owst joined #darcs
18:52 dcoutts joined #darcs
18:52 dcoutts joined #darcs
19:14 mizu_no_oto joined #darcs
20:00 gpiero joined #darcs
20:15 amgarchIn9 joined #darcs
21:13 * dixie study the Paris public transport :)
21:29 Heffalump :-)
21:29 Heffalump it's pretty easy
21:36 dixie looks good, even street view is available on various places. I'm not very often traveler :]
21:55 Heffalump there are these little green card tickets that are one metro journey each
21:55 Heffalump and you can get them in books of 10 at a slight discount, or individually
22:21 * owst is looking forward to stopping thinking about this paper, and starting thinking about darcs :-)
22:22 dixie what is "this paper" ? :)
22:23 owst I'm writing a paper with my supervisor for ICALP'13, I need to get it done for tomorrow, before I leave for Paris!
22:24 owst If you want to know what it's about, well, I'd have to kill you ;-)
22:26 Heffalump :-)
22:26 Heffalump owst: are you bringing a 4-way block?
22:26 owst Yep
22:31 owst I just need to find it :-)
22:31 dixie owst: no, I don't want to know :)
22:32 owst Ok found it
22:32 owst dixie: You're safe then!
22:51 gpiero joined #darcs
22:53 * Heffalump prints out his tickets and considers editing the ticket PDF first so I don't waste toner on the advert
22:55 owst heh
22:55 * owst remembers to do that tomorrow
22:57 dixie I'm going to sleep, see you tomorrow.
22:59 owst see you dixie!
22:59 Heffalump I appear to have accidentally printed two copies of my tickets and one of owst's. Oh well.
23:01 owst heh, well done Heffalump.
23:02 owst You may as well bring it along, then :-)
23:02 Heffalump will do
23:07 owst Heffalump: coming to the next LHUG meetup on teh 27th?
23:16 Heffalump oh, gah, that clashes with something else I want to do...
23:35 owst Well you're out of luck if you want to read a talk abstract to decide if it's worth going to
23:38 Heffalump yeah, I just spent a while searching for one

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