Camelia, the Perl 6 bug

IRC log for #darcs, 2010-02-11

← Previous day | Index | Channel Index | Today | Next day → | Search | Google Search | Plain-Text | plain, newest first

All times shown according to UTC.

Time Nick Message
01:44 vsync is darcs supposed to eat all my memory just from a patch with some photoshop files or whatever?
01:45 i know that large patches are "non-optimal" but this is insane
01:45 and if i can't fix it i may have to move the project from darcs to probably Hg
02:19 sm vsync: how big are the files in a patch ?
02:35 vsync individual files, or the patch itself?
02:35 10-15 20MiB files
02:35 sm: darcs ended up eating 1.5GiB of RAM with no sign of stopping
02:40 sm so a 200-300Mb patch ?
02:42 others will speak on this in a while. I don't know how darcs reacts to that
03:07 twb joined #darcs
04:30 gwern joined #darcs
04:30 gwern joined #darcs
05:36 twb joined #darcs
07:29 sm joined #darcs
07:43 tux_rocker_ joined #darcs
07:53 tux_rocker_ joined #darcs
07:59 balor joined #darcs
08:02 darcscommitbot joined #darcs
08:02 vsync joined #darcs
08:03 darcswikibot joined #darcs
08:04 lispy joined #darcs
08:05 lelit joined #darcs
08:25 gh_ joined #darcs
08:26 gh_ vsync, the explanation to your problem is here: http://bugs.darcs.net/issue80
08:27 vsync, the solution would be for you to create one patch by 20MB file, not just one patch for all of them
08:28 there have been recent discussions on the mailing list about directions to make darcs use exactly 1x memory in size of the patch
08:29 http://lists.osuosl.org/piperm[…]ember/022517.html
08:53 twb joined #darcs
09:06 mornfall joined #darcs
09:06 mornfall joined #darcs
09:15 twb joined #darcs
09:35 gal_bolle joined #darcs
09:40 JaffaCake joined #darcs
09:51 gal_bolle morning all
10:01 Heffalump hiya
10:04 gh_ joined #darcs
10:07 gh__ joined #darcs
10:13 twb joined #darcs
10:22 kowey joined #darcs
10:22 kowey morning
10:23 Heffalump: do you have any spare cycles to look over the latest hashed-storage work?
10:23 I think mornfall is itching for a release :-)
10:26 (oh yeah, I ought to try and catch you when you're at home)
10:27 mornfall kowey: Well, he did look at them and the comment was that functionally they are OK but the performance impact is not quite clear on him.
10:27 But I have to go now. I'll be around in the evening if I manage to not fall asleep.
10:29 kowey cool
10:42 * Heffalump is on holiday today, actually
10:42 Heffalump though with stuff to do for the rest of the day in an hour or so
10:42 and yes, what mornfall said.
10:42 I didn't spot a formal patch submission to review, otherwise I'd have put the comments there.
10:43 what do I have to do in an email to get it sent as a comment to a specific issue?
10:46 kowey it has to go to bugs@darcs.net and have [issueXXXX] in the subject line
10:47 Heffalump anywhere in the subject line?
10:50 kowey I'm not 100% sure, but I think it's anywhere
10:50 I'm having a look at the config.ini file to see if I can glean any clues
10:51 Heffalump don't worry, I've just tried it
10:51 we'll find out :-)
10:51 kowey :-)
10:52 it sounds like issue1742 may be also related to the use case of people wanting to put hunks in different bins in one go
10:52 (like you suggested in FOSDEM'05)
10:53 Heffalump I think here he wants the same bin. Just two different hunks.
10:53 Or perhaps one hunk would be good enough
10:53 kowey or rather, I didn't fully understand what John was trying to do, but the idea of being able to merge two hunks reminded me of the bins thing
10:53 Heffalump the whole revert during record thing is a good example of bins
10:53 there's been some discussion on a bug, right?
10:54 kowey there is this http://bugs.darcs.net/issue1734 which talks about revert-on-record
10:55 twb joined #darcs
10:55 Heffalump ta
10:56 I think there's a fairly simple UI for this: add !0123456789 to the list of keys you can press during record (perhaps called something different like select), and have those select the bins with ! being revert.
10:58 kowey that does sound fairly simple
10:58 should that become a catch-all bug for bins?
10:58 I'm still a bit nervous about this long-term in the "just because we can, doesn't mean we should" sense
10:59 Heffalump true. FWIW I find myself wanting it quite often.
10:59 kowey I guess that the scary situation I want to avoid is us having a lot of features that only a very small number of us actually use (but those that do, care passionately about)
10:59 Heffalump I don't know if bins apply sanely to things like push/obliterate too.
10:59 kowey same here, although maybe less frequently
11:00 the way the darcs UI would continue feeling good is if we had as few ideas/conventions/etc as possible, and if they applied as broadly as possible (/me says fuzzily)
11:01 for example, interactive selection of patches - works everywhere... learn the one idea and you're set
11:01 Heffalump so there's one thing for interactive changes, another for managing patches
11:03 kowey sorry?
11:03 do you mean that the current SelectChanges UI behaves in two separate ways, one for primitives and one for named patches?
11:03 or something else?
11:03 Heffalump I mean one set of things you can do with primitives, and one set of things you can do with recorded patches
11:04 kowey eg. hunk editing for primitives
11:04 Heffalump I meant more in terms of commands: you can revert, record and amend-record primitives
11:04 kowey ah ok
11:04 Heffalump and you can push, obliterate, rollback and unrecord recorded patches
11:04 kowey and some commands working on both levels: amend and rollback
11:05 Heffalump hmm, perhaps rollback is more complicated because it lets you select hunks inside the patch. But amend is really about unrecorded primitives only.
11:05 kowey or rather there are two rollback actions, named-rollback which invokes prim-rollback
11:06 but broadly speaking the kinds of things you can do break down into primitives vs recorded patches
11:06 Heffalump a few other random ideas for UI features include removing hunks (and deps) from already recorded patches and merging two recorded patches
11:07 kowey my difficulty is that I have a vague idea of what I want ("unity"), with no idea how to achieve it!
11:08 oh boy, this is starting to be a big list... but it's giving me hope... maybe we can tie it all together somehow
11:09 http://bugs.darcs.net/issue1700 <-- amend-record on steroids
11:09 http://bugs.darcs.net/issue1470 <-- amend by removing hunks
11:10 Heffalump 1700 will be covered by rebase
11:10 kowey oops, it even says so
11:11 http://bugs.darcs.net/issue1559 <-- maybe this is what I was thinking of
11:12 oh and then (I don't know if we're getting really far off track here), the thing about being able to break a dependency (I guess by introducing a conflict)
11:12 Heffalump I think that one is far off track.
11:13 and I think the title is wrong
11:13 I think there are two ways out of the situation msg7230 mentions
11:13 firstly, forcible commutation maintaining patch identity, which is what I talk about in my comment on issue1559
11:14 and secondly, amending the initial patch to remove changes/split them into a separate patch
11:14 which I guess might be what the current title of issue1559 is about
11:15 kowey OK, so I'll change the title of issue1559, and create a new ticket for the second one?
11:15 and then hopefully try to come back to John's problem and bins, etc
11:16 Heffalump sounds good
11:22 * kowey abuses the wiki a bit - http://wiki.darcs.net/Ideas/FancyRecording
11:23 kowey this contains the current list of "fancy recording" tricks that we may like to have
11:23 merging hunks, recording in bins, revert-on-record, removing hunks via amend
11:24 it's quite nice (imho) that hunk editing/splitting are actually the same feature (unity); and likewise bins and revert
11:25 but I still wonder if we can do better.  It doesn't have to be the ubershell; I'm skeptical of that idea too
11:27 err, I should probably go do work work
15:30 kpreid_ joined #darcs
15:41 lispy Good morning
16:03 kpreid_ joined #darcs
17:43 tux_rocker_ joined #darcs
17:48 kpreid_ joined #darcs
17:51 * tux_rocker_ solved issue1739 in my local dev repo
17:54 gh_ yay
17:57 tux_rocker the ColorPrinter code is fed strings with char's > 255, which makes its poor head explode
17:57 someone even left a friendly comment there stating this explicitly
18:52 sm nice
18:52 always good to see an honest-to-god bugfix
19:32 nomeata joined #darcs
21:09 exlevan joined #darcs
21:47 nomeata joined #darcs
23:25 JaffaCake joined #darcs

← Previous day | Index | Channel Index | Today | Next day → | Search | Google Search | Plain-Text | plain, newest first