IRCloggy #git 2007-03-03

Logs Search ←Prev date Next date→ Channels Documentation

Provider of IRC logs since 2005.
WARNING: As Freenode became unjoinable and lost all warnings in topics, we cannot log channels on Freenode anymore.

2007-03-03

arj what happens if I create changes in the repo where I want to fetch from, do I have to rerun the fetch command?00:01
← Oejet left00:05
cehteh yes00:09
arj ok00:10
gitte arj: If that happens often, you can setup a shortcut, so you can say "git fetch other".00:10
cehteh unlike arch, git does all operations locally ...remotes are only used to store locatitions from where to fetch00:10
gitte (Instead of that longer "git fetch ..." line.)00:11
cehteh thats quite diffrent than arch00:11
arj yeah00:11
cehteh .. brb .. again00:12
arj I'm running git 1.4.1, should I have any reason to upgrade to 1.5?00:12
it's the default in ubuntu00:12
cehteh yes .. lots neat stuff00:12
arj can I find a changelog somewherE?00:13
gitte http://repo.or.cz/w/git.git?a=shortlog;h=master00:15
;-)00:15
arj thx00:15
can git push using ftp? the man page doesn't seem to mention it00:16
gitte Not yet.00:21
It can push via SSH, HTTP and RSYNC. In 1.5, you can even open it to pushing via Git's own protocol, but that is without authentication...00:22
Pushing via FTP would be probably just a relatively minor change to http-push, so a dedicated developer should need only a couple of days to make it work.00:26
Anyway, I have to go home now. Ciao,00:31
arj bye thx for the help00:31
gitte arj: You are very welcome!00:31
gitte is even too tired to type a dot correctly, hitting a comma instead. :-( >)00:32
→ benlau joined00:43
→ anholt joined00:54
arj must git be installed on remote host where one pushes?01:03
→ devogon joined01:33
→ Thumper_ joined01:44
→ mugwump joined01:44
→ ianw__ joined01:44
→ zakame joined01:44
→ maio_ joined01:44
→ corecode joined01:44
→ z3ro joined01:44
→ cworth joined01:44
→ tchan joined01:44
→ robfitz joined01:44
→ devogon joined01:44
→ Tv joined01:44
→ halfline joined01:44
→ tko joined01:44
→ meyering joined01:44
→ MadCoder joined01:44
→ Ori_B joined01:44
→ davi joined01:44
→ mndrix_ joined01:44
→ yann joined01:44
→ mtm joined01:44
→ tonyj joined01:44
→ PugMajere joined01:44
→ cods joined01:44
→ Romster joined01:47
linuxmig1ation → linuxmigration01:51
→ orospakr joined02:08
→ kanru joined02:32
→ spearce joined02:58
robinr when does the "encoding" header enter a commit?03:07
spearce when the commit is not UTF-8.03:08
robinr not true03:08
that was what I thought too03:08
spearce ? I thought that was the case, we only insert it into the commit if the commit body was not in UTF-8...03:09
looking at builtin-commit-tree.c, that's what the code says...03:10
robinr does git-cat-file convert to local?03:10
i think not and that means my commit is iso-8859-103:10
without an encoding header03:10
spearce cat-file only specially handles tags and trees with the -p option; otherwise it dumps the raw data.03:11
show/log/whatchanged will convert to local.03:11
robinr then I don't understand03:14
does the same apply for tags?03:15
spearce no. we don't mess with the tag at all.03:16
because the damn thing might be pgp signed...03:16
robinr shouldn't be a problem as long as you decide what is signed, i.e. the raw data03:17
I can sign non-ascii mail03:17
spearce yes, of course. but we don't have an encoding header in the tag format it seems, and there's no way for the user to set one themselves before signing the tag.03:18
robinr I get a warning03:21
yeas p(normal user gets this right) -> 003:23
so jgit has to guess the encoding regardless03:25
the encoding header just reflects the config setting and nothing else03:25
spearce yea... basically i think we might be able to do what git-gui does: read the header and assume utf-8 unless an encoding header is found, if the encoding line is found switch to whatever it says.03:25
robinr you can't assume that, because a default configuration encodes the platform default03:26
not UTF-803:26
i.e. it is best to ignore the header and is it looks like utf-8 it is, else guess something else03:27
→ rlb joined03:27
robinr s/is it /if it/03:27
just like IRC clients do03:28
spearce heh. :)03:28
robinr I thought the encoding header was there to convince, not confuse03:29
spearce yea, i ignored most of that discussion on the list when it happened. i'm not too interested in encodings and whatnot, i'm a US-ASCII only guy. ;-)03:30
robinr I was trying to argue there. junio thought it we'd fix it five years or so03:31
I had an implementation that asked the user what locale he had.03:31
the NLS stuff03:32
spearce what about i18n.commitencoding? if that is set and the commit doesn't have an encoding header, assume that value? hmm... confusing it is!03:33
robinr much less code too03:33
spearce heh. yoda voices don't work on irc. bummer.03:33
robinr don't they03:34
spearce i hate C compiler vendors. sizeof(unsigned long) == 4 here, but sizeof(unsigned long) == 8 on other systems. yet sizeof(off_t) == 8 here and there. grrrr.03:36
robinr isn't that why there is sizeof?03:36
spearce doesn't help when a million variables in git code were declared long, unsigned long, or (horrors) int when they should have f'in been off_t from the get-go.03:38
robinr so the encoding isn't the only thing wrong03:39
otoh there isn't a helluvalot of code to explicitly get it wrong03:39
spearce :)03:39
robinr the non-encoding stuff, I mean03:40
spearce right03:40
robinr Sure I wan't to see my mistakes, but I'd rather avoid them in the first place03:45
seems java doesn't choke on trying to decode UTF-8 when it isn't03:46
spearce :)03:46
that's koalty kode dere... :)03:47
robinr ?03:47
spearce bad joke. i guess the j2se runtime cannot really be expected to tell if it really is utf8 or not, the only way to really know is if the last character is missing an additional byte i guess.03:48
DrNick well, no, you can look for missing bytes anywhere in the string, or you could look for non-minimal sequences or sequences that encode outside of the Unicode space03:49
→ MenTaLguY joined03:50
MenTaLguY hello03:50
robinr if it can be decoded as UTF-8 it is extreemly likely that it actually is03:51
iso-2022 is the only encding where misdetections can be expected in practice03:52
with the ISO-8859-x encoding it just won't happen03:53
theoretically it could by noone encodes such nonsense03:53
s/by/but/03:53
→ Tv joined04:40
→ iano joined04:49
→ rkaway1 joined05:27
→ linuxmigration joined05:35
→ linuxmigration joined05:38
→ benlau joined06:54
← MenTaLguY left06:55
→ iano joined08:21
→ lyakh joined08:30
→ Oejet joined08:55
→ alley_cat joined09:01
→ Eludias joined09:11
→ chris2 joined09:17
→ nud_ joined10:10
nud_ → nud10:47
→ jeffpc joined12:11
corecode gitte: i don't care if creating patches for fromcvs is harder for people. actually, i am sure that it is the subject and not the language. ruby is easy (proven by a a large amount of available software in ruby), but converting cvs is hard (proven by the fact that there are no more than 5 cvs converters)13:15
robinr the hard part about ruby there is the fact that it is another dependency. From looking at the code it is however a very appropriate language for solving the problem13:35
seems fast too13:35
and fromcvs is the only one so far that I've seen work13:36
ok, haven't checked to cvs2cvn based one13:37
s/to/out the/13:37
matled corecode: well, you could put all stuff needed in one repository and add a makefile to run the stuff in-place instead of installing ruby modules14:06
this would help a lot14:06
and make the repository available as tarball too14:07
corecode pulling ruby-rbtree in?14:07
i'd rather not14:07
matled I did stop when I saw that I'd have to install a ruby module somewhere14:07
and that was after installing hg just to get the source code14:07
and cloning 2 repositories14:08
corecode there you could have used the tarball14:08
hgweb serves tars14:08
matled the webinterface was quite confusing for me14:08
corecode click on head rev, then on bz214:08
i could arrange that you simply can put the other modules in the workdir and uses them from there14:09
maybe supply a makefile which fetches the versions14:09
and extracs them14:09
matled http://ww2.fs.ei.tum.de/~corecode/hg/fromcvs where to click?14:10
corecode either on manifest14:10
or on the changeset id14:10
matled there.. at the top14:10
corecode yes14:10
matled I wouldn't have found it without help14:11
corecode well, if you want a tarball14:11
you need a revision you want it from14:11
so the natural thing would be to click the revision id14:11
changeset id14:11
and then go bz214:11
matled sure, if you know where it is you probably find it..14:12
corecode i wouldn't know where it could be placed to make it more obvious14:12
→ mountie joined14:13
matled http://ww2.fs.ei.tum.de/~corecode/hg/fromcvs?cs=a4da2672cb53 e.g. here I see all the information in this 'table' above the diff14:13
and wouldn't search the top for a tarball14:13
well, I didn't14:13
corecode :)14:13
matled -corecode-hg-fromcvs-a4da2672cb53.tar.bz2 hooray, filenames beginning with '-' :)14:14
corecode haha14:15
cehteh touch '$(rm -rf .)'14:21
→ robfitz joined14:39
corecode hm.14:48
maybe more14:48
no14:49
that's good14:49
:)14:49
ya ya14:49
cehteh: when would that trigger?14:49
cehteh prolly never ;)14:50
well .. maybe with cut'n'paste14:51
→ raalkml joined15:21
→ iano joined15:43
ruskie hmm how can one do a shalow clone with git 1.5?16:21
gitte corecode: *lol* You just proved the point about "harder on other people" for me... Thank you :-)16:22
ruskie: a shallow clone is only possible when both sides include support for it (1.5 is safe, methinks).16:23
In that case, you have to provide a "--depth <n>" argument to "git clone" or "git fetch".16:23
The "<n>" is the depth in the ancestor chain.16:24
ruskie so 1 would just get the latest?16:24
gitte I think "0" would just get the latest. I don't remember. Try it!16:24
ruskie ahh can't work over http :(16:25
gitte Yes, that's right.16:25
ruskie guess I'll go complain to the ppl that maintain that to fix up their repo then... it fails mid clone due to a missing object...16:25
gitte There is not much we can do there: Over HTTP you usually fetch the complete packs anyway.16:25
ruskie thanks16:26
gitte ruskie: could also be that the information generated for HTTP (AKA dumb transports) is not updated properly.16:28
You can check yourself:16:28
Browse the repository with a webbrowser, look into "objects/info/pack/, fetch all these files and stash them into your local ".git/objects/pack/".16:29
If http-fetch still complains about missing objects, they are really missing...16:29
→ BlindGlobe joined16:30
→ z3ro joined16:39
→ ferdy joined16:44
→ kukks joined16:46
→ GyrosGeier joined17:02
corecode gitte: i don't get you17:05
→ wereHamster joined17:14
wereHamster is there any way to make git-push chown the files to a certain group and make them world-unreadable (750)?17:15
→ kumbayo joined17:17
corecode umask?17:18
kumbayo hi, is there specific channel for stgit topics, or are these discussed here too?17:19
→ _jeremy joined17:20
→ DrNick2 joined17:27
kumbayo since nobody has problems with me asking stgit questions here :-) i will ask my question17:28
newer versions of stgit do not allow a : in the pathname. is this intentional? i guess so, because of the new rev:path/to/file syntax for git-rev-parse17:30
pathname -> patch name17:30
→ kumbayo joined17:46
cehteh gitte: nack ... git updated and fsck still segfaults .. but i think the repository is fucked up anyways17:48
wereHamster git-push: Pushing an empty <src> allows you to delete the <dst> ref from the remote repository. - what is an 'empty <src>'? 'git-push ... :refs/heads/to-delete' doesn't work17:53
gitte kumbayo: I don't know StGIT, so I cannot help you, unfortunately. But if you send your question to the Git list, I am sure it will get answered.18:16
cehteh: Still with the same object?18:16
cehteh yes18:17
gitte wereHamster: why not make your repository directory chmod 0770?18:17
cehteh gitte: latter i have some time and i'll investigate it by myself then18:18
gitte cehteh: I suggest starting with an _empty_ repository, and just cp'ing the object into that.18:18
That's how I tested here.18:18
Just to make sure that it is not some nasty side effect of something completely (apparently) unrelated...18:19
cehteh gitte: yeah you seen my valgrind trace .. there is a uninitialized access before that which likely damages some datastructure18:19
actually thats the point where to look18:20
kumbayo gitte: it looks like more patch names are considered invalid (/\;()), and i just used the : in my older patches18:20
cehteh since you confirmed that accessing that object works when trying alone .. the bug is somewhere ahead18:20
i look later .. are you here in 1-2 hours?18:21
kumbayo and now i cant even stg remove the ones with the invalid names, but stg rename to a valid name still works :-)18:21
gitte cehteh: Probably.18:21
cehteh oki18:21
rlb If you have a project with major lines of development (say 1.0 vs 2.0, etc.), what are the arguments for having separate repositories as compared to multiple branches within the same repository?18:48
→ TommyBJ joined18:53
TommyBJ Is there a web project managenemt system avalible for GIT. Much like the trac for svn ?18:54
→ jeffpc joined18:59
→ spuk joined19:05
→ raalkml_ joined19:13
GyrosGeier rlb, there are none19:22
DrNick2 → DrNick19:22
GyrosGeier rlb, use branches19:22
rlb GyrosGeier: so if you're actively working on both, you'd just have two clones, each with a different current branch?19:23
I suppose the master branch might just be the main line of development...19:25
kblin if you use branches, you of course want two checkouts to work in ;)19:26
rlb kblin: then you'd just push each to the other (or to a central repo) if you wanted to be able to merge back and forth from time to time?19:28
wereHamster if I want to pull from one repository, and push to another, I'll end up with two refs with the same name, 'master' in the repo where I fetch from, and 'master' in the repo where I push to..19:28
kblin well, actually I'd assume that a 1.0 branch is different enough from a 2.0 branch that I could just cherry-pick the occasional patch I want to merge19:29
rlb kblin: oh right -- I didn't mean a "full merge".19:29
GyrosGeier rlb, I usually work on one or the other branch, not both at a time19:30
kblin I was talking about a working dir per branch to avoid recompiling19:30
GyrosGeier rlb, so I cg-switch when I want to work on the other19:30
kblin, can be done, but I recompile my stuff fairly often anyway19:30
kblin GyrosGeier: but if you keep switching from 1.0 to 2.0, you don't usually need full rebuilds all the time19:31
I use git for wine, wine needs 15 minutes for a full build with ccache switched on, usually19:31
GyrosGeier well19:32
kblin if I switch branches often, I can't drink enough coffee to bridge that19:32
rlb In the case where you're only managing a ./debian dir, then multiple trees works better since the source in .. (which isn't managed by git) will vary.19:32
GyrosGeier you can work with shared object storage and multiple checkouts then19:32
I agree that it does make sense in the case of wine19:32
kblin but of course I don't mind for the 5-file python project I have :)19:33
GyrosGeier exactly19:33
kblin so I guess your mileage may vary19:33
GyrosGeier in most of my projects I end up writing very generic classes19:34
that get their own testcases and are subsequently split out into a separate project19:34
typically I have around 30 classes in a project, of which 20 are exception classes that only have two methods or so19:35
so each project builds in under ten seconds19:35
everything that is larger than that is a candidate for generalisation of concepts19:36
rlb GyrosGeier: I was thinking about just having one repo that both of the checkouts push? to to keep it up to date -- either that, or if I rarely cherry-pick, then just do it directly from one repo to the other. How would a shared object store work?19:40
(thanks for the help BTW)19:40
GyrosGeier rlb, the shared object store is realised with "alternates"19:46
I think there is a HOWTO somewhere, but I've never needed it so far so I cannot tell you much about it19:46
rlb GyrosGeier: it seems like I might be fine with just separate working dirs for each version and pull/push between them (or use a central common repo)...19:49
GyrosGeier rlb, yes, that should be fine19:59
rlb, the common object storage is good for saving disk space19:59
rlb GyrosGeier: ok, thanks -- disk space shouldn't be relevant in this case.20:03
gitte rlb: try "git clone -n -l -s wine-main/ wine-branch/"20:05
(This assumes that you have a wine working dir & repo in wine-main/.)20:06
rlb gitte: ok, so looking at the manpage -- I see how that might be really fast. I assume you'll have to be careful not to move wine-main...20:06
gitte If you want to branch from "master", there is no need to do "-n".20:07
rlb: you can move wine-main. But you have to adjust the alternates and remotes...20:07
rlb gitte: right.20:07
(That's one thing I'm really enjoying about git. You can just edit various files when you need to make changes...)20:08
gitte: anyway, thanks.20:09
gitte: I'll probably just let the repos be independent for now.20:09
gitte rlb: Hey, it's my pleasure!20:13
gitte is quite happy how the ':/blablub' thing works out in practice!20:15
→ anholt joined20:30
→ fork0_ joined20:35
→ cworth joined20:51
← wereHamster left21:02
cehteh gitte: ... tomorrow21:10
gitte ... okay.21:12
rlb Is master treated specially enough by the commands that it's somehow "awkward" to work on other branches regularly, or is that not an issue?21:57
→ merlyn joined22:03
merlyn OK, I think I have a grafting situation22:03
A B C D E F G H => old devel22:04
A B1 C1 => new devel22:04
now I want the changes between C and H regrafted onto C122:04
to make A B1 C1 D1 etc22:04
Oh - C1 came from git-merge -s ours on B1 C22:04
if that helps22:05
anyone around for the challenge?22:05
→ [L]ash joined22:12
[L]ash hi all22:13
is possible to move a file from a branch to another ?22:14
or merge only a file and not all the branch22:14
gitte merlyn: sounds like a job for rebase, not for graft.22:14
gitte has to be away for half an hour...22:15
→ gitster joined22:27
gitte Lash: yes, that's possible.22:43
merlyn I figured it out22:43
careful edits. :)22:43
lots of backups22:43
but I got it22:44
gitte But since it is not necessarily what you want, it is not _that_ obvious.22:44
[L]ash gitte, how ?22:44
gitte The easiest way to get a file from another branch into the current branch is "git checkout other-branch file-name"22:45
Of course, this is a copy, not a move.22:45
merlyn: glad you figured it out. Did you see my script, edit-patch-series?22:46
It would only need a single option, "--rebase", and a few lines, to make it easy what you wanted.22:46
[L]ash i write move but i intend copy22:47
sorry22:47
gitte Lash: no problem. I thought so much.22:47
[L]ash gitte, an it ,erge also the file if it exist right?22:48
gitte merlyn: with this option, it would be as simple as switching to old devel, and saying "git edit-patch-series start C1".22:48
merlyn: BTW I am glad that you are innocent ;-)22:48
merlyn: Would you be interested in the "--rebase" feature?22:57
→ jeffpc joined23:07
merlyn well, "conviction erased" is not the same as "innocent" :)23:57
and rather than a rebase, what I decided was to leave the history as-is and call it a merge.23:58
it'll be interesting for historical reasons, I guesss.23:58
gitte okay.23:59
Note that you can do that also with a rebase, just after branching.23:59

Logs Search ←Prev date Next date→ Channels Documentation