Perl 6 - the future is here, just unevenly distributed

IRC log for #darcs, 2014-06-25

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

All times shown according to UTC.

Time Nick Message
00:12 mizu_no_oto joined #darcs
00:27 favonia joined #darcs
00:58 mizu_no_oto joined #darcs
01:13 siel joined #darcs
01:37 mizu_no_oto joined #darcs
05:43 favonia joined #darcs
07:37 lelit joined #darcs
08:05 iago joined #darcs
08:10 c74d joined #darcs
08:37 kowey joined #darcs
08:42 gal_bolle joined #darcs
09:32 kowey we have a patch in reviewed that's not in screened
10:45 mizu_no_oto joined #darcs
12:46 lelf joined #darcs
13:25 mdiaz joined #darcs
14:12 mizu_no_oto joined #darcs
14:39 lelf joined #darcs
15:15 sm darcs hub seems to have been hung the last 6 hours
15:16 sm restarted
16:12 lelit joined #darcs
16:16 amgarchIn9 joined #darcs
16:48 capicue joined #darcs
17:04 dolio joined #darcs
17:37 lelf joined #darcs
17:40 amgarchIn9 joined #darcs
17:54 favonia joined #darcs
18:08 amgarchIn9 joined #darcs
18:14 amgarchIn9 joined #darcs
18:42 mdiaz_ joined #darcs
20:09 capicue joined #darcs
20:57 c74d joined #darcs
21:12 stepkut joined #darcs
21:13 * Heffalump appears for a bit
21:13 Heffalump mdiaz_: if you're around we could chat for a bit now
21:50 mdiaz_ Hi Heffalump
21:51 Heffalump hi
21:54 mdiaz_ I have doubts about the garbage collection of the global cache
21:54 Heffalump oh?
21:56 mdiaz__ joined #darcs
21:57 mdiaz__ I'm not sure if that's the best way to do it...
21:57 mdiaz__ What do you think?
21:57 Heffalump do you mean the hardlink based approach, or the collecting repositories approach?
21:58 Heffalump or whether it should be done at all
21:58 mdiaz__ All of that :)
21:59 Heffalump my main worry with hardlinks is what happens on funny filesystems (FAT, NFS, ...)
21:59 mdiaz__ What can happen?
21:59 Heffalump well, for example FAT doesn't have hardlinks :-)
22:00 Heffalump oh, also if the cache isn't on the same filesystem as the repositories, you can't hardlink
22:00 Heffalump so there are lots of cases where you have to make sure that you don't just wipe out the entire cache because it seems to be unused
22:01 mdiaz__ Ok, then we can rule out that approach...
22:01 Heffalump well, in situations where the hardlink count is accurate, it's ideal as you don't have to find all the repos
22:02 Heffalump but I tend to agree that it's best to rule it out given it's not general
22:03 mdiaz__ But if you have many folders, pass them as arguments would not be very nice
22:03 Heffalump we could search for them recursively
22:04 Heffalump your home directory would be the obvious starting point
22:05 mdiaz__ And the user would select then which repository he want to consider?
22:05 Heffalump I think darcs would just search all of them
22:08 mdiaz__ But then darcs keep all the files of those repos found?
22:08 Heffalump it would delete things from the cache that weren't in any of those repos
22:08 Heffalump I think that's the idea of garbage collection, isn't it?
22:09 mdiaz__ I'm not sure.
22:10 mdiaz__ Which are the cache files that we no longer need...
22:11 Heffalump yeah, it's hard to be certain whether we'd ever need one or not
22:11 Heffalump but that's a problem with the hardlink approach too
22:12 mdiaz__ I thought to give the user a way to decrease the size of the cache, as flexible as possible.
22:12 mdiaz__ Then the user decides which are the unnecessary files.
22:13 Heffalump wouldn't that be a lot of information for them to process?
22:14 mdiaz__ Perhaps...
22:15 mdiaz__ But I don't know if the alternatives will be useful.
22:15 Heffalump I think it's a reasonably plausible for a user to decide that anything that isn't in a current repo is garbage
22:15 Heffalump it's fairly similar to garbage collecting an individual repo
22:19 mdiaz__ But the files that we have in a repo are not useful in the cache...
22:20 Heffalump they are if we are likely to get a sibling of that repo
22:20 mdiaz__ And that is likely?
22:20 Heffalump if it's not, why do we bother having a cache at all?
22:22 mdiaz__ I guess that's a yes ... :)
22:22 mdiaz__ Ok.
22:23 Heffalump :-)
22:23 Heffalump well, I get clones of darcs.net quite a bit
22:23 Heffalump I do slightly wonder if a cache will be any use if/when we do in-repo branching
22:24 mdiaz__ And in the case of lazy repositories. I have to keep all files, even those that are not in the repo?
22:25 Heffalump oh, hmm
22:25 Heffalump that implies you have to traverse inventories too to figure out which patches to keep?
22:25 mdiaz__ I think so..
22:26 Heffalump hmm. I wonder what Guillaume thinks :-)
22:27 mdiaz__ :)
22:27 Heffalump but yes, I think logically a full reachability analysis is the right thing
22:27 Heffalump for inventories and patches anyway
22:28 Heffalump hmm. I wonder why we cache pristine.
22:28 Heffalump oh, I guess it means faster remote gets, as with the other things
22:31 mdiaz__ Yes. Ok, I think that was all I wanted to ask.
22:31 Heffalump you may prefer to work on darcs undo first, particularly if you still feel unsure about this feature
22:34 mdiaz__ And darcs undo is easier?
22:34 Heffalump ah, now that I don't know :-)
22:35 Heffalump but one strategy might be to post to the mailing list about the garbage collection feature and see what people think, and then work on designing darcs undo while that's being discussed
22:36 Heffalump on the other hand that might just mean you get distracted too much
22:37 sm probably the simpler approaches are best.. is blowing away the whole cache so bad ?
22:37 Heffalump we don't need a darcs command to do that :-)
22:37 Heffalump well, I guess in terms of knowing the location, we need something basic
22:37 sm well users shouldn't have to do that manually
22:37 sm darcs should keep itself working
22:38 Heffalump if it's a natural operation to blow the cache away entirely, then I'd return to my earlier point about why we bother to have it at all (and the answer could be that it's not worth it, of course)
22:39 sm not having it would presumably solve all those cache-related errors I see, and the disk usage and expensive backups, in one swoop
22:39 sm I suppose there is some good reason for it though :)
22:39 Heffalump perhaps not for hub
22:39 sm faster repeat gets of the same repo ?
22:40 Heffalump yes, that's the main point
22:40 Heffalump don't your backups respect hardlinks, btw?
22:40 Heffalump I think the actual cache related errors we see would happen even if the fetch was local
22:40 sm I don't know.. it seems to be related to number of files. Linode's backup takes most of the night
22:40 Heffalump it's just that the download is to the cache so we see an error in the cache
22:40 sm ok
22:43 * sm remembers the file count is high because of repos too, not just the cache
22:43 Heffalump :-)
22:48 IbnFirnas joined #darcs
22:58 mdiaz__ About darcs undo... it should undo only the last command?
23:10 IbnFirnas joined #darcs

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