Camelia, the Perl 6 bug

IRC log for #darcs, 2010-02-19

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

All times shown according to UTC.

Time Nick Message
00:04 twb What's the darcs-beta repo's URL?
00:04 twb nm, it's in topic
00:18 * twb grumbles about "darcs dist" and "cabal sdist" being sucky
00:57 markstos joined #darcs
03:15 lispy joined #darcs
05:54 mornfall twb: You might want to print "use runghc Setup sdist" unless someone fixed cabal-install to actually call Setup for sdist. Otherwise, the tarball may be bogus, if it is created at all...
05:55 mornfall (Ack, still broken with my 0.8.0...)
07:02 twb Yeah, it was explained to me just now that "cabal sdist" and "runghc Setup sdist" actually do COMPLETLY DIFFERENT THINGS
07:07 mornfall Yes. Bummer.
07:08 twb I grumbled at #hackage about that
07:31 tux_rocker joined #darcs
08:02 darcscommitbot joined #darcs
08:03 darcswikibot joined #darcs
08:23 lelit joined #darcs
08:45 gh_ joined #darcs
09:21 gal_bolle joined #darcs
09:57 kowey joined #darcs
09:57 kowey morning
10:02 gal_bolle morning
10:02 gal_bolle do you know if there are tools for exploiting .prof files? searching through them is a bit of a pain
10:14 kowey I wish there were.  Anyone know?
10:15 kowey I think one of the lessons in my fantasy Haskell performance book is "how to look at a profile", giving lots of examples and showing the kinds of things your eyes should be landing on, and why
10:16 kowey (although that's not very related to your question)
10:17 gal_bolle well it would be the following question
10:19 kowey http://book.realworldhaskell.org/re​ad/profiling-and-optimization.html sorta does it, but I'm still a bit too tl;dr-ish for it (which is unfortunate)
10:19 kowey what I would want to see is a profile, with segments of it highlighted (circled in a red pen)
10:20 kowey and little arrows pointing from one circled bit to another - sort of a "how your eyes *ought* to be scanning this page"
10:20 gal_bolle rwh is a bit light on time profiles
10:20 kowey (the other thing in my fantasy book is lots and lots and lots of case studies)
10:21 kowey basically each chapter being a case study focusing on a particular kind of pathalogical behaviour and maybe also lots of mistaken attempts at making things go faster
10:23 gal_bolle tickets for the sprint bought. I resisted the temptation to use the night train for extra time sunday afternoon
10:25 kowey you may want to add your plans on http://wiki.darcs.net/Sprints/2010-03 in case people want to do stuff
10:26 gal_bolle hmm looks like i'll be missing guillaume by 1/2hr each way
10:33 nwf joined #darcs
10:50 kowey our bugtracker has now passed the 10000 message mark
11:35 gypsymauro joined #darcs
11:36 gypsymauro hi
11:36 gypsymauro there is a way to add all new files to a repository?
11:37 gypsymauro I can find them with darcs what -ls
11:42 kowey hello! I think you're looking for darcs add -r
11:42 kowey but if you're repository is large, watch out for http://bugs.darcs.net/issue80 :-(
11:45 gypsymauro kowey:  I think I've got a mess, if I try to add the folder it says:
11:45 gypsymauro The following files and directories are already in the repository:
11:46 gypsymauro and a list of file
11:46 gypsymauro and at the end
11:46 gypsymauro darcs failed:  No files were added
11:46 kowey that sounds fairly normal; darcs just reminds you if you've already added the files
11:46 gypsymauro can I remove them in a step?
11:46 gypsymauro it's a css folder
11:46 kowey the new darcs 2.4 has darcs remove -r
11:47 gypsymauro darcs remove css fails
11:47 gypsymauro oh
11:47 kowey but it's in beta - cabal update; cabal install darcs-beta if you have the Haskell Platform and are willing to help use test
11:47 gypsymauro 2.3.0 (release)
11:47 kowey I believe http://wiki.darcs.net/FrequentlyAskedQuestions may have an answer for you, but I don't remember
11:48 kowey (i.e. one trick that doesn't involve upgrading Darcs just yet)
11:48 gypsymauro how can I know what I neet to delete befor trying the add again?
11:49 kowey ah, here we are: http://wiki.darcs.net/HintsAndTips​#tips-for-older-versions-of-darcs
11:49 kowey actually: allow me to probe that question a little bit
11:50 kowey why bother deleting and adding them again?
11:50 kowey is it because you think you've added too much and want to step back a little?
11:50 kowey you can do darcs whatsnew -s to see what you've added
11:51 kowey OR if you just want to record your changes in smaller bits and pieces, you could just do darcs record and then interactively select 'n' on the bits you don't want to record yet
11:52 gypsymauro damn it fails
11:52 gypsymauro I tried the xarg line
11:52 gypsymauro but if dails
11:52 gypsymauro kowey: just cause it fails for case names
11:53 gypsymauro and now I can't delete nor add them again :(
11:53 gypsymauro :)
11:53 gypsymauro I'm lost
11:53 gypsymauro and I'm new to darcs :)
11:53 gypsymauro if I try to add again them it fails
11:53 gypsymauro files that differ only in case. Use --case-ok to override this:
11:54 gypsymauro with darcs add -r css
11:54 gypsymauro and then
11:54 gypsymauro darcs add -r  --case-ok css
11:54 gypsymauro The following files and directories are already in the repository:
11:54 gypsymauro a lot of file and then
11:54 gypsymauro darcs failed:  No files were added
11:58 kowey sorry, I was away for a bit
11:59 kowey OK hang on... what operating system are you on, please?
12:00 kowey I'm just curious... I'll bet you're on some variant on Linux just because you're using a case-sensitive filesystem
12:00 kowey (or appear to be, in that you have different filenames that differ by case only; or maybe I'm wrong)
12:01 kowey here's a pastebin you can use if you'd like to plug your commands and error messages into them - http://pastebin.ca/
12:03 gypsymauro sorry I've to leave maybe we will check later, tank you a lot for your help :)
12:04 kowey hmm: I wonder how we could have helped with that user experience
12:05 kowey interesting observation here is that we've got a case of user inferring that something is "wrong" when really from our point of view, things are perfectly "OK"
12:05 kowey in the sense that darcs add "fails" to re-add the user's file
12:06 kowey ... which we think of as a good thing because it's good to know you've already added them, but which the user here perceives as "argh, scary!"
12:14 kolmodin joined #darcs
12:43 lelit joined #darcs
13:02 lelit` joined #darcs
13:18 nomeata joined #darcs
13:19 kpreid_ joined #darcs
14:17 kpreid_ joined #darcs
14:56 kpreid_ joined #darcs
14:58 markstos joined #darcs
14:59 markstos I just tried a basic push with the 2.4beta3, and got this: "Thread 860c400 has exited with leftover thread-specific data after 4 destructor iterations"  I compiled with GHC 6.10.4 on FreeBSD.
14:59 markstos It's not clear that darcs crashed, but the message isn't encouraging.
15:35 lispy joined #darcs
15:39 gal_bolle left #darcs
15:44 mornfall Wtf roundup. First it loses my comment and then it appears out of nowhere? BLah.
15:50 kowey wait, what?
15:51 kowey you got to keep me honest about this... what do you mean by it losing your comment?
15:51 kowey (meanwhile I'm worrying a bit about the too-many-open-files pb, likely unrelated)
15:51 gbeshers joined #darcs
15:56 lispy joined #darcs
16:00 mornfall kowey: I enter comment, hit submit, comment nowhere.
16:00 mornfall kowey: So I type it again, hit submit and the *old* one is there.
16:00 mornfall Dunno.
16:00 lispy kowey: hello
16:00 lispy kowey: that chunky ticket, is that one about changing in memory or on disk formats?
16:05 kpreid_ joined #darcs
16:08 gbeshers joined #darcs
16:13 kowey mornfall: OK, so web interface problem... hmm, I'll have to think about this
16:13 kowey proxy server? browser cache?
16:16 kowey lispy: complex answer.  I think it just means "disk format".  Traditionally "chunky hunks" just refers to the fact of coalescing the hunks into a single region
16:16 kowey lispy: droundy tried to implement both changing the disk format and the internal representation, but ran into bugs
16:16 kowey so (on the phone), he suggested backing off into just changing the disk format
16:16 lispy kowey: ah yeah.  I'd like to be more incremental in that regard.
16:17 lispy My HunkHandles still use the same representation for Hunks
16:17 kowey later on, mornfall in his blog post introduced a 3rd idea of storing the chunk on disk and referring to it as a hash
16:17 kowey so I was wondering if HunkHandle was related to this third idea
16:18 lispy kowey: did my ticket update clarify what a HunkHandle is?  I think I sent in my patches and a description at one point
16:18 kowey I'm trying to digest it
16:18 lispy kowey: http://lists.osuosl.org/pipermail/d​arcs-users/2010-January/022760.html
16:18 lispy that email is way more concrete
16:18 kowey I regret that I never really got a chance to understand what this work was about :-(
16:19 lispy It's simple
16:19 lispy Dead simple
16:19 lispy Do a pass over the patch bundle
16:19 kowey sounds like a temporary cache of the hunk for processing
16:19 lispy If you find a Hunk, record where it was in the stream
16:20 kowey well, not so much a cache, but just a way of dumping to disk so you're not keeping in memory, anti-cache
16:20 lispy (but don't load the hunk)
16:20 lispy Later, when you need to hunk use readHunkHandle :: HunkHandle -> IO Hunk, to get the hunk
16:20 kowey where it was in the stream, meaning its offset in the file?
16:20 lispy yup
16:20 lispy dead simple
16:21 kowey ok, so if I understand correctly, what you're trying to achieve is to make it so that Darcs only keeps offsets in memory
16:21 kowey because that's all it needs for hunk-hunk commutation
16:21 lispy Bingo
16:21 lispy Well
16:21 lispy No
16:21 kowey and in the rare case where you do need the hunk contents (eg. hunk-replace) you read them from disk
16:21 lispy commute will not change (at least initially)
16:21 kowey oh...
16:22 lispy Later, it should be possible to apply Mornfall's ideas
16:22 lispy So that we can do what you just suggested
16:22 lispy But, first things first
16:22 kowey wait, what did I suggest?
16:22 lispy You suggested the "3rd" idea again
16:22 lispy using hunk meta data to do commutes
16:23 kowey OK, I think I can see structurally that there is a "4th" idea
16:23 kowey but it's going to take me some more walking through before I can understand the distinction between the 3rd and 4th idea
16:24 lispy Kind of.  But, I see it as a prefactor to the 3rd idea :)
16:24 kowey I mean, in my way of talking about it, I didn't really see it as "commutation changing"
16:24 kowey because I thought it was just a fundamental reality that commutation does not affect the hunk contents during hunk-hunk commutation
16:25 kowey so I guess I'm confused by why doing the 4th thing doesn't mean you get the 3rd thing for free
16:25 lispy I'm talking about code that has to be changed
16:26 kowey oh, so you're saying that to implement the 3rd idea, the commutation code necessarily has to change
16:26 lispy commute won't show up in any of my diffs, I don't think
16:26 lispy Yeah, I think so
16:26 kowey (ah because it involves changing the internal representation of hunks to talk about hashes)
16:26 lispy Yeah
16:26 kowey OK, that's *starting* to make sense
16:27 kowey and your hunkhandle work is completely orthogonal to the 1st (in-memory representation) and 2nd ideas (on-disk representation)
16:27 kowey #2 I can see quite clearly
16:27 lispy I do change the internal rep slightly
16:27 lispy Now a PatchSet doesn't have Hunks in it, it has HunkHandles
16:27 lispy and things that actually look at the Hunks will need to request them from disk
16:28 lispy But otherwise, it's the same
16:28 lispy (same as what we have now)
16:28 kowey oh wait, is the reason why you think of #4 as being separate from #1
16:28 kowey that you still use hcOld :: ![B.ByteString]
16:28 lispy Yeah
16:28 kowey whereas #1 would mean using hcOld :: !B.ByteString?
16:28 lispy Right
16:29 kowey OK! thank-you!
16:29 lispy you're welcome
17:14 kowey I wonder what happened to kolibrie's darcs record in http://wiki.darcs.net/Benchmarks
17:16 lispy kowey: are those old benchmarks?  I can't tell what version 2.3.99 is
17:16 kowey those are the darcs 2.4 beta
17:16 kowey it's just using our hackage-friendly versioning scheme
17:17 lispy :(
17:17 lispy Do you know which beta though?
17:18 * lispy really dislikes the duplicity
17:18 kowey perhaps darcs-benchmark should be a little evil and systematically replace N.99.C with N+1 rc C
17:18 lispy Can we stop calling them darcs 2.4 betas and start calling them darcs 2.3.99?
17:18 kowey I'm not entirely sure about which beta it is - I think folks just followed my requests
17:18 kowey I still think the best thing to do is to flip to the odd/even scheme next time
17:18 lispy My poor little brain just can't be trouble to store so many assosciative data structures
17:19 kowey so that those versions would just be darcs 2.4 and darcs 2.5 respectively
17:19 kowey and when it becomes stable, darcs 2.6
17:19 kowey easy to understand, no tricks
17:52 kpreid_ joined #darcs
18:14 kowey lispy: I think my point is that this will indirectly eliminate the need for duplication
18:15 kowey I think we tend to talk about darcs 2.4 beta X instead of darcs 2.3.98 X because the latter sounds too much like darcs 2.3
18:15 lispy Well, if beta/alpha/RC is our natural language then we should use it and fix cabal :)
18:15 kowey by giving us a distinct version number that does not sound like the previous one (darcs 2.5 vs darcs 2.4), we will be less likely to need to use two names
18:16 kowey so what I'm saying is that having darcs 2.5.x means that we'll be less likely to want to call it darcs 2.6 beta
18:16 kowey see what I mean?
18:16 lispy If we abolish the beta/alpha/RC terminology and stick to numbers I'm okay too.  I just want:  One source version, one version number/name.
18:17 lispy My brain is too small to do the duplicity thing
18:17 kowey right... so what I'm proposing is a way of making it comfortable to stick to numbers
18:17 kowey because right now, if you say darcs 2.4.1 and darcs 2.4.99.1 in the same sentence, *my* brain gets confused and thinks you're still talking about darcs 2.4
18:18 kowey so if we adopt the hybrid scheme, hopefully we're both happy - no duplication for you, and no accidentally-thinking-theyre-the-same for me
18:18 markstos left #darcs
18:19 lispy In your email you still called 2.5.x the 2.6 beta
18:20 lispy So I would want to stop doing that
18:20 lispy Either stick with 2.6 beta or with 2.5.x
18:20 kowey fine!
18:20 kowey I'm using the old terminology to bridge to the new terminology
18:21 lispy ah
18:21 kowey I don't think you'll be able to stop people from saying 2.6 beta informally, but I think that will just wither away on its own, because it's just easier to say darcs 2.5.x to mean darcs 2.5.x
18:21 lispy That was not clear
18:22 lispy Cabal should probably have support for release candidates and/or hackage may need a sense of 'unstable' like the more mature distribution systems (see Debian/Ubuntu)
18:22 lispy But...That's a lot to change
18:22 lispy And there seems to be resistance
18:23 kowey so this is an easy way forward, and it's a common one in the FOSS world
18:23 lispy I've never noticed the .99 stuff before, but odd/even is common yeah
18:23 kowey no, the .99 stuff is ours
18:24 lispy I've never cared much about version numbers.  I just want it to be easy to know when you have same or different versions :)
18:24 kowey because we're one of the rarer projects that are (a) big enough to need a multi-stage release procedure and (b) use hackage
18:24 lispy True
18:53 kpreid_ joined #darcs
19:37 arossato joined #darcs
19:38 arossato left #darcs
20:15 Heffalump aargh, Easyjet have cancelled my flight to the hackathon
20:16 Heffalump mornfall: you around?
20:17 Igloo How come?
20:17 Heffalump no idea why
20:17 Heffalump they just sent me an email saying "we've cancelled it, here are your options"
20:17 Igloo :-(
20:18 Heffalump well, the travel options to/from the airport were pretty nasty and I was beginning to wonder if I shouldn't have done something else anyway, so it might not be too bad
20:18 Heffalump but now I need to investigate the options
20:20 Heffalump gah, bastards:
20:20 Heffalump >> What would you like to do with your flight? (Please
20:20 Heffalump >> note that you can only select one option and your
20:20 Heffalump >> decision is final)
20:20 Heffalump options being "Get a refund" and "A free transfer", where there's no indication of what the options might be for the free transfer.
20:29 arjanb joined #darcs
20:34 lispy joined #darcs
21:28 lispy joined #darcs
22:13 gwern joined #darcs
22:13 gwern joined #darcs
22:30 kowey joined #darcs
22:35 lambdabot joined #darcs
23:02 kpreid_ joined #darcs
23:10 lispy joined #darcs
23:22 arossato joined #darcs
23:33 kowey oof, I think I finally have a version of darcs-benchmark that takes repo variants into acct
23:33 kowey although there is a significant risk I'm displaying the tables completely wrong, thereby making the results totally wrong...
23:45 Heffalump cool
23:45 Heffalump Easyjet cancelled my outward flight, btw.
23:46 kowey :-(
23:47 kowey hmm... yeah I noticed in the log
23:47 kowey hopefully they'll have more info soon.  that's weird
23:48 Heffalump was mainly wanting to warn you about yours, but since it seems to still be offered (whereas mine isn't) I guess it's ok
23:49 Heffalump I think I can pay about £60 more and go with Swiss at somewhat nicer times, anyway.
23:49 kowey probably worth it in fatigue-avoidance terms
23:50 Heffalump yeah, I'd already been vaguely wondering to myself if I'd made a good choice with Easyjet, so I'm not too fussed
23:53 Heffalump given the continuing noise about performance regressions, I'm starting to think skipping 2.4 might make sense
23:54 kowey Petr does seem to think that it's an unlikely corner case, the last bit of news
23:54 Heffalump the Haskell-Cafe report?
23:55 kowey uh-oh, I must have missed something recent then
23:55 kowey I really need to get back into reading that list
23:55 Heffalump Ben Franksen on Haskell-Cafe
23:55 Heffalump very recent thread
23:56 * kowey reads
23:56 Heffalump (what corner case are you referring to?)
23:57 kowey argh... I didn't really get a chance to understand the thread, but it's worth holding off on this
23:58 kowey this corner case: http://bugs.darcs.net/issue1746
23:59 Heffalump oh, and while reviewing Petr's latest stuff I found a really egregious bug in darcs, which I'm a bit surprised hasn't been found before (at least, I couldn't spot a test case for it)
23:59 kowey oh good
23:59 Heffalump basically darcs add dir ; darcs add dir/file ; darcs rec ; darcs remove dir ; darcs rec ; darcs oblit <last patch> ; darcs check fails
23:59 kowey oh wow

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