Perl 6 - the future is here, just unevenly distributed

IRC log for #darcs, 2015-10-19

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

All times shown according to UTC.

Time Nick Message
00:38 Riastradh joined #darcs
01:47 ilbot3 joined #darcs
01:47 Topic for #darcs is now http://darcs.net/ | logs: http://irclog.perlgeek.de/darcs/ | darcs 2.10.0 is out http://darcs.net/Releases/2.10
03:59 PEB` joined #darcs
05:33 Guest58751 joined #darcs
08:04 gal_bolle joined #darcs
09:15 Guest58751 joined #darcs
12:59 Big_G joined #darcs
15:59 mdl joined #darcs
15:59 mdl i accidentally did "darcs unrecord" and it blew away all my committed work.  the relevant patches are still in _darcs/patches.  is there a way to reapply them?
16:03 dolio mdl: What happens if you do a get from that to a new directory?
16:04 mdl i get the same blown-away repo
16:05 dolio Ah. Bummer.
16:18 mdl :/
16:30 byorgey joined #darcs
16:30 byorgey joined #darcs
16:30 mdl joined #darcs
16:30 jackhill joined #darcs
16:32 sm darcs unrecord shouldn't do that
16:32 sm aren't the changes still in your working copy
16:33 mdl yes, but they're no longer in patch sets, they're just a ton of changes in my working copy
16:33 sm did you unrecord a whole lot of commits ?
16:33 mdl yeah
16:33 sm aha. ouch
16:34 sm I'm not sure that can be recovered. Heffalump might know
16:34 mdl in git i'd just rebase to the previous mark on the reflog
16:34 mdl i have no idea what to do here
16:34 mdl it's stupid that i can't just "darcs reapply" a patch in _darcs/patches/ or something
16:34 mdl since...the patch is right there
16:34 sm well, there must be a way then
16:51 mdl fortunately, i realized i had a clone on another machine with all the patches save for one, so i just pulled it, then re-recorded the remaining one patch (the contents of my working copy)
16:58 Riastradh joined #darcs
17:01 sm \o/
17:40 notdan 1/whois mdl
17:40 notdan oops, sry
17:42 notdan [1maybe Riastrad^
17:42 notdan dammit, what's wrong with my keyboard
17:43 notdan Maybe it is possible to create a patch bundle by hand.. that would be tonns of work tho :(
17:43 notdan we should have ununrecord :)
17:46 Heffalump mdl: otherwise you can edit the patch back into the inventory by hand, but you have to know the right order
17:46 Heffalump I don't think patches in the repository store their own context (it's implied by the inventory)
17:47 Heffalump notdan: darcs undo was targeted at this, but it didn't get finished
17:52 mdl Heffalump: i did know the order, but i wasn't sure where to even start.  thanks, if i screw up again that'll be handy
17:53 mdl s/where to even start/what files would need to be changed/
17:58 Heffalump mdl: the starting point is _darcs/hashed_inventory
18:01 Heffalump put the patches back in that following the format given - the hashes are just the patch file names and you can get the headers for each patch from the individual patch files
18:01 Heffalump it's in forward order starting from an inventory stored in _darcs/inventories, but hopefully you woiuuldn't need to go into those
18:01 Heffalump then you'll need to use 'darcs repair' to sort out the 'pristine' part
18:02 Heffalump (obviously, do this in a copy of th edie directory, it's fiddly!)
18:09 pointfree https://www.reddit.com/r/compsci/comments/3pdd5g/pijul_combines_speed_of_git_usability_of_darcs/
18:12 sm "We beat darcs 2.10 and Mercurial 3.5 on our performance contest" - now you're talking! :)
18:14 sm pointfree: is this your project ?
18:24 pointfree sm, It's Florent's project. (Although I do have a project of my own to do distributed version control with regular expressions only and no patch positional logic -- ie, version control on the regular language rung of the Chomsky ladder only. Mine is based on ctre https://github.com/gritzko/ctre by Victor Grishchenko)
18:27 notdan there should be a conference on version control
18:27 notdan with people giving tutorials on their formalisms
18:36 pointfree Perhaps we need new projects (and fresh branding) to spawn a renaissance in alternative version control.
18:39 mdl left #darcs
18:51 Heffalump FWIW, we've talked about integrating the pijul algorithm into darcs
18:51 Heffalump but yes, I imagine there are trade-offs in branding etc - e.g. that's why camp was a fresh start
18:53 dolio Is it just faster, or does it store more information about the actual history?
18:53 Heffalump which, pijul or camp?
18:53 dolio pijul
18:53 Heffalump pijul primarily has a better conflict algorithm, AFAIK
18:54 Heffalump I don't know what makes it faster, though I suspect a mixture of being in OCaml and a fresh implementation.
19:00 dolio Well, that might explain constant factors, but something else seems likely to explain their exponential graphs. Unless they're all exponential and darcs just has a worse exponential.
19:00 Heffalump exponential graphs?
19:01 dolio https://pijul.org/Performance
19:01 dolio I guess they are all exponential curves.
19:02 Heffalump AIUI they are quadratic in something to do with the conflict
19:02 Heffalump hmm, that ever-growing log-factor looks quite scary
19:03 Heffalump also, if they are just graphing a linear sequence of changes, I don't see why they aren't all constant
19:03 dolio That first one doesn't look like a quadratic to me, at least.
19:04 dolio The performance vs. number of lines looks like it could be quadratic.
19:06 dolio I guess the first one is performance vs. number of commits?
19:06 Heffalump but performance of what? Why should a VCS slow down for a simple record operation based on the number of existing commits?
19:07 dolio I don't know. You're right that it doesn't make much sense.
19:07 dolio I mean, it really makes no sense for something that's just taking a snapshot to be exponential for that.
19:10 dolio Actually, does the file grow exponentially according to what they're doing?
19:11 dolio I think it might.
19:11 dolio For each line, add a line with 10% probability and delete line with 7% probability.
19:12 dolio That might make a file of size around 1.03^n, where n is number of steps.
19:13 dolio I think that explains it.
19:15 dolio So the size of the change is growing exponentially, as well, and everything looks quadratic in change size, but darcs has worse constant factors.
19:35 Heffalump makes sense

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