IRCloggy #git 2007-09-29

Provider of IRC logs since 2005.

WARNING: As Freenode became unjoinable and lost all warnings in topics, we cannot log channels on Freenode anymore.

	2007-09-29
MadCoder	heh ohloh is using a webserver I had never heard of before	00:00
	funny	00:01
gitte	gitster: now it worked, oddly enough...	00:01
gitster	nice ;-)	00:02
	heh, I wonder what would happen to that Kudo I gave to the two people that are merged into one ;-)	00:03
Mikachu	i would guess they only keep one of them	00:03
gitte	Actually, I have but rank "1" now.	00:04
gitster	Mikachu: I'd hope so.	00:04
	Don't worry, they will recompute once a day.	00:04
	← ferdy left	00:04
Mikachu	phew	00:04
gitster	Looking at other project's contributor stats is always interesting.	00:04
gitte	20,134 lines changed.	00:05
	Hehe.	00:05
gitster	You can tell whom to hit with a bus to kill your competitor ;-)	00:05
gitte	gitster: you don't have any...	00:05
	competitors.	00:05
MadCoder	gitte: that won't last	00:06
gitster	more seriously, we are a lot flatter than others.	00:06
MadCoder	I'm at rank 9, and I'm not a very big contributor in any very famous project yet	00:06
	maybe except debian	00:06
	though sth unfair is that the svn guy is 4th :P	00:06
gitster	I do not think that rank number means anything.	00:07
gitte	normalperson worked quite hard for that rank, though.	00:07
	← branstrom left	00:07
gitte	(But he's 5th AFAICT)	00:07
MadCoder	by 4th I mean he's at rank 10, 4th in the general list	00:07
	and "the svn guy' is svn upstream	00:08
	not git-svn	00:08
	gitster is only 70 or sth	00:08
gitster	ohloh is owned by svn company so it is not very surprising ;-)	00:08
MadCoder	haha	00:08
	really ?	00:08
gitte	Karl Fogel also wrote CVS, and parts of Emacs.	00:08
MadCoder	well, I a fan of none of the three	00:09
pasky	hmm, ohloh	00:09
MadCoder	used svn a lot, barely CVS (too young) and never emacs :)	00:09
pasky	is that another advogato rehash?	00:09
gitte	I used CVS a lot, and was utterly disappointed by svn.	00:09
	It took me 3 days to create my first svn repository.	00:10
gitster	http://www.collab.net/news/press/2007/ohloh.html	00:10
MadCoder	gitte: well, I spent some time using tla	00:10
	it took me 3 day to remember how to commit	00:11
	:)	00:11
	gitster did too. it was very steep.	00:11
MadCoder	*days damnit	00:11
gitte	Git is the 61st in the project list.	00:11
MadCoder	(and I've all forgotten since)	00:11
gitte	But CVS is 51st.	00:12
gitster	and I recall it was not usable without setting up revision library which was not done by default, just like git is unusable without initial repacking. So in that sense I do understand the worry Linus had about unpacked repositories.	00:12
gitte	And Spring is in more stacks than Git. That POC!	00:12
MadCoder	spring ? what isit ?	00:12
gitster	you are joking, right?	00:13
gitte	MadCoder: you don't want to know.	00:13
spearce	that hasn't sprung into the junk pile yet?	00:13
gitte	About Spring being in more stacks? Nope: http://www.ohloh.net/projects?page=3&sort=stack_count_reverse	00:13
	I actually personally know a few people who love Spring.	00:13
spearce	we only have 3 more to pass it	00:13
MadCoder	oooh java	00:14
gitster	allergic to caffeine?	00:14
gitte	It's also an island, you know?	00:14
gitster	and people from there are javanese. how confusing ;-).	00:15
MadCoder	gitster: have you by any chance had some time too look at my proposal to avoid the NULL ->buf pitfal ?	00:15
	damn, I've received kudos from people I've absolutely no clue who they are	00:15
gitster	Looked good on the initial pass but haven't had chance to look at any patches deeply enough for the past 24 hours.	00:16
MadCoder	np	00:16
	I was afraid the strbuf_slopbuf was freaking you out :)	00:16
gitte	MadCoder: you check for that when realloc()ing, right?	00:17
MadCoder	yes	00:17
gitte	(of course you do.)	00:17
MadCoder	it's even detailed in the commit message	00:17
gitte	Which I did not have the chance to read in detail.	00:17
	gitte was too busy the last few days	00:17
MadCoder	I wasn't blaming you	00:17
	my commits messages tends to be too long nowadays	00:19
gitte	No, they are just right.	00:19
MadCoder	(and my english worse when it's that late)	00:19
gitte	Not as bad as in "a dyslexic man walks into a bra..."	00:19
MadCoder	anyways, as nobody was responding, I was afraid the static slop buf was of too bad taste	00:19
pasky	hmm	00:20
	looks like ohloh can't be used w/o javascript?	00:20
gitte	pasky: hmm back	00:20
pasky	or something?	00:20
gitte	AJAX.	00:20
pasky	at least stack can't	00:20
	bah	00:20
	← GyrosGeier left	00:20
pasky	they should write that somewhere so that I don't waste my time :)	00:20
MadCoder	but that's the most straightforward and lightweight mechanism I've thought of	00:20
spearce	pasky: its web 2.0. you need a "real" browser. :)	00:20
gitster	Is anybody interested in seeing "interactive unstash"?	00:23
gitte	How's that look like?	00:23
MadCoder	and what would it do ?	00:23
gitster	You have a huge change that is complete in your work tree and you would want to split.	00:24
	You stash it away, and then trickle bits in from it and commit after testing in the work tree.	00:24
gitte	Ah, and you get it piece by piece...	00:24
	That's a nice idea.	00:24
gitster	Rince repeat continue until the stash gets empty.	00:24
gitte	But actually, I'd try to enhance git gui to do that...	00:25
	gitte is trying to find the time to clean up that splitting hunks patch.	00:25
gitster	if it were perl/Tk instead of tcl/Tk you could have talked me into it...	00:26
	As I've already done the same thing twice already. Once originally for jit and then for add--interactive.	00:26
MadCoder	hmm I see	00:27
	wasn't git add -i supposed to do that ?	00:27
gitte	MadCoder: not really: git add -i is about index <-> worktree interaction	00:27
	This would be about stash <-> worktree interaction	00:28
gitster	interactive unstash is going the other way.	00:28
MadCoder	hmmm	00:28
gitster	"add -i" is a wrong workflow. What you commit is not something you had in the work tree as a consistent whole.	00:28
gitte	But it would be similar enough that it's probably just an enhancement of git-add--interactive.perl, right?	00:28
gitster	Probably.	00:28
MadCoder	I'm not sure I would have a use for it, when I split a patch for commits, I use my $EDITOR because I've not found anything else I'm really confided with yet	00:28
gitster	But more like "git stash apply -i" ;-)	00:28
MadCoder	oh, seen like this it may be interesting	00:29
	like we discussed some days ago, my workflow is currently:	00:29
	split patch, git add and commit	00:29
	git stash	00:29
	test it builds and whatever else	00:29
	git stash apply	00:29
	got back to the splitting part	00:29
gitte	I have basically two different work flows:	00:30
	← mnemoc left	00:30
MadCoder	but a git stash apply -i makes sense	00:30
	I thought you were proposing git stash -i	00:30
	and that didn't made real sense to me	00:30
	actually git stsah apply -i may be a brilliant idea	00:30
gitte	One (rarer) is to split commits as described in SPLITTING COMMITS in git-rebase.txt	00:30
	And the other is: commit really often, with short descriptive onelines.	00:31
	Then to clean up the patch series, rebase -i and squash quite a lot.	00:31
Mikachu	so you never commit and leave them? :)	00:32
gitte	To test, I rebase -i, and mark all (except the last one) "edit".	00:32
	Umm. No.	00:32
MadCoder	gitte: yeah I try to have this workflow nowadays	00:32
gitte	I commit _really_ often.	00:32
MadCoder	(the commit often and then rebase -i to cleanse history)	00:32
	though sometimes I'm in the middle of a big refactor, and from thread to the needle (not sure it's the proper expression) I endup with a patch touching 1k lines :)	00:33
	and here git stash; git stash apply -i may be really nice	00:35
	"okay, let's pretend we redo that ... sloooowly"	00:35
fonseca	<shameless plug> recent tig supports staging of chunks </>	00:37
gitte	But without being able to split hunks, I think it will not be as useful for me sa with it...	00:37
gitster	fonseca: yeah, I learned about it last night ;-)	00:38
gitte	fonseca: I actually like tig very much...	00:38
	→ orospakr joined	00:38
fonseca	Cool :)	00:38
	MadCoder likes the idea of tig very much	00:40
MadCoder	but I'm just unable to use it	00:40
	brain issue	00:40
REPLeffect_	I have a question about GIT_DIR, and GIT_WORK_TREE.	00:43
fonseca	Heh, that is a new one.	00:43
REPLeffect_	If I set GIT_DIR AND GIT_WORK_TREE to directories in different heirarchies ...	00:43
	should I be able to execute commands from anywhere inside the work tree ...	00:44
gitte	Yes.	00:44
REPLeffect_	and have them correctly discern where I am relative to the root of the work tree?	00:44
gitte	Yes.	00:44
REPLeffect_	it doesn't seem to work with 1.5.3	00:44
	commands at the root work fine.	00:44
gitte	sgrimm: ping!	00:47
	← spearce left	00:54
REPLeffect_	unless For instance. I have a work tree in /wt ...	00:56
gitte	MadCoder: Linus has a good point here, no?	00:56
	REPLeffect_: lemme check	00:56
	→ sbohrer joined	00:56
REPLeffect_	OK	00:57
	(the "unless" in that statement was unintentional)	00:57
gitte	Yes, I can reproduce.	00:58
REPLeffect_	What command did you try?	00:58
gitte	git diff	00:59
REPLeffect_	adds with relative paths don't seem to work either.	00:59
	neither do commits of adds that I do from the root.	00:59
	git status will show an added file, but with a path starting at the root.	00:59
	Hm. I guess git status always shows you the file from the root (even if you are in a relative dir)	01:03
gitte	I think we borked your use case...	01:03
REPLeffect_	hehe.	01:03
	stuff happens.	01:04
gitte	(Meaning _I_ borked it)	01:04
REPLeffect_	Well, I spent a fair amount of time trying to get my head into the code.	01:04
	and, well, I think I didn't succeed very well.	01:04
	or I guess I should just say I didn't succeed.	01:05
	I know the GIT_WORK_TREE stuff is fairly new. So I would expect there to be a few kinks to work out.	01:07
	← jasam left	01:10
	REPLeffect_ → REPLeffect	01:15
	→ spearce joined	01:21
gitte	REPLeffect: I'm somewhat lost... There might be something I missed why the code behaves as it behaves; and I'm too tired to think about that right now. Will attempt a fix tomorrow, okay?	01:25
REPLeffect	hey, that would be awesome.	01:26
	I'll be happy to play the guinea pig if you want some help testing.	01:28
gitte	Bed time! Git night!	01:29
REPLeffect	git night, sleep well.	01:29
	← gitte left	01:29
	← REPLeffect left	01:30
	← herzi left	01:31
	← engla left	01:39
	→ tcoppi\|laptop joined	01:40
	→ GyrosGeier joined	02:02
	→ bicchi joined	02:07
bicchi	I was thinking that git-config should have an option to set the commit editor. like: git config --global user.editor "gedit"	02:08
Mikachu	wouldn't that be core.editor ?	02:09
bicchi	is there such an option already?	02:09
Mikachu	quite	02:09
	"Commands such as commit and tag that lets you edit messages by launching an editor uses the value of this variable when it is set"	02:10
bicchi	ok them. core it is	02:10
Mikachu	you can also set the env vars VISUAL, EDITOR or GIT_EDITOR	02:10
	→ diactitical joined	02:11
bicchi	i guess it would be nice if it integrates with the git-config command	02:11
	← cods left	02:12
	→ glguy joined	02:13
	→ a2c joined	02:14
gitster	Huh? "GIT_EDITOR=vi git config" starts your favorite editor and lets you edit your configuration file?	02:16
	← glguy left	02:20
	→ glguy joined	02:22
	← topfunky left	02:24
	← glguy left	02:29
	→ aroben_ joined	02:34
	← aroben_ left	02:36
	→ aroben_ joined	02:36
	← aroben left	02:37
	← aroben_ left	02:38
	→ aroben joined	02:39
	→ rtmfd_ joined	02:46
	← GyrosGeier left	02:48
	→ GyrosGeier joined	02:48
	→ fowlduck joined	02:51
	← bicchi left	02:57
	← rtmfd_icbm left	03:02
	→ elmex_ joined	03:05
	→ rtmfd_icbm joined	03:14
	← rtmfd_ left	03:14
	← rtmfd_icbm left	03:14
	→ rtmfd_icbm joined	03:15
	← elmex left	03:20
	← a2c left	03:22
	← GyrosGeier left	03:22
	→ fhobia joined	03:28
	← tcoppi left	03:32
	→ tcoppi joined	03:34
	← kanru left	03:39
	← madewokherd` left	03:42
diactitical	with git remote add is if you dont get it right the first time, you have to edit the config file right?	03:42
	not that the following question makes sense: but to achieve the equivalent of the INSTALL instructions to fetch man and html, you would use git add --remote	03:44
	erm	03:44
	you would use `git add remote --mirror' right ?	03:44
	i'm trying git init; `git add remote [-t master] --mirror ~/path/to/repo' but that just fails	03:51
	however i get what i want by doing (a=`pwd`; cd ~/path/to-repo; git push $a master:refs/heads/master)	03:56
	i think there may be some benefit from simplification of syntax --- like rsync dtrt with src and dest and it figures out whether to do a push or a pull.	03:58
	not pull: i meant have --fetch/ --pull as an argument	03:59
	← rtmfd_icbm left	04:00
diactitical	current intelligence notwithstanding, i cant find any unifying principle which simplifies or facilitates my understanding in the current scheme	04:00
	← fhobia left	04:01
	← Sho_ left	04:07
	← aroben left	04:07
	← orospakr left	04:09
	→ orospakr joined	04:13
diactitical	after my last command, if i do (cd $a; git checkout -b newbranch; echo changed >> file; git commit; git push ~/path/to-repo newbranch:newbranch) --- there is a huge performance problem	04:16
	it takes minutes to complete	04:17
	the refspec is actually newbranch:refs/heads/newbranch	04:17
	the size of the .git is under 500k	04:17
	it cant just be because of nfs could it	04:19
	this behaviour i'm seeing is on git-maint	04:20
gitwump	git's performance is sometimes quite dependent on a good buffercache	04:20
	(sometimes meaning, for certain operations)	04:20
diactitical	my whole source tree is under 200k	04:21
	< 50 commits	04:21
gitwump	which part is slow?	04:24
diactitical	git push where i'm creating a new branch on the remote	04:24
	it says: Total 34 (delta 14), reused 33 (delta 14)	04:25
	then Unpacking 34 objects...	04:25
	this takes minutes	04:25
	← rkaway left	04:27
spearce	oh, unpacking is decompressing the objects back to disk	04:27
diactitical	I did do git gc --aggressive --prune on both ends before the operatin	04:28
	that would explain it?	04:28
spearce	sounds like its just taking a long time to write the objects to disk	04:28
diactitical	ok (but still a hmm, i'm doing bzips of huge files fast enough)	04:29
	→ rkaway joined	04:29
	diactitical had no luck with git --mirror on git-maint	04:31
diactitical	spearce, maybe its just NFS issues. i'll take back that user experience for now	04:34
spearce	oooh, NFS? yea. i've seen unpacking loose objects suck on at least one freebsd NFS installation. i don't blame freebsd, i blame an overloaded server that was too busy to keep up with its clients. :-)	04:34
	drop your transfer.unpackLimit to a really low number. its default is 100. it can help by only making the .pack and .idx instead of lots of loose objects.	04:35
diactitical	will try that next time!	04:36
	off	04:37
	← diactitical left	04:37
	→ glguy joined	04:38
	← sbohrer left	04:43
	→ mgrimes joined	04:45
	← tcoppi left	04:56
	→ tcoppi joined	04:57
	→ radarsat1 joined	04:59
radarsat1	hello, i have a quick question about git if anyone can help	05:00
	i'm exploring the idea of moving my project to git which is currently under svn	05:00
	the project is organized such that there is a main codebase, and a library	05:00
	i think it might be best to keep these as separate "projects", i.e., each in its own repo	05:01
spearce	probably, especially if the library's development history and release schedule is different from that of the main codebase.	05:01
radarsat1	but what is the best way to inform git of their dependency? i looked around a bit but couldn't find a faq on this	05:01
spearce	use the new git-submodule supported in 1.5.3 (and still being improved in latest versions from git.git)	05:02
radarsat1	in svn, i am used to just treating each subfolder as a separate subrepository, so i just use one big huge repository for everything, and i don't have to worry about this	05:02
	ah, thanks i haven't read about that	05:02
	sounds like that's about what i'm looking for: "Add the given repository as a submodule at the given path to the changeset to be committed next.	05:03
	thanks!	05:03
	ah, sadly it looks like Ubuntu's not up to that version. sigh well i guess i can just compile it myself..	05:04
	cheers :)	05:04
spearce	yea, its pretty easy to get git with git.	05:04
radarsat1	thanks again, i'll pop in again if i have another question.. :) thought it might be faster than bugging the mailing list	05:05
	← radarsat1 left	05:05
	→ udi_trugman joined	05:19
	← Pistahh left	05:25
	← tcoppi\|laptop left	05:28
	← tcoppi left	05:28
	→ tcoppi joined	05:33
	← spearce left	05:34
	← DrNick left	05:59
	→ topspin joined	06:47
	← krawek left	06:47
	→ Evelynn joined	06:48
	← glguy left	07:01
	← Evelynn left	07:09
	← dduncan left	07:27
	← topspin left	07:45
MadCoder	heh linus subscribed to ohloh	07:51
gitster	can't believe it.	07:54
MadCoder	well, someone did for him then :]	07:55
	I just went to see if gitte was ranked already, and I saw that linus has merged his linux and git contributions	07:56
	that'll be fun to see him jump in the top 10 directly ;)	07:56
gitster	Yeah, he used to be two separate entries, the kernel one being somewhere around 70-80 range.	07:57
	I guess he learned about it by reading the #git channel log ;-)	07:59
MadCoder	heh	07:59
	he is _that_ bored to read channel logs ?	07:59
gitster	35 more to go ;-)	07:59
MadCoder	btw that's the sucky part of ohloh, not being able to join identities of people not subscribed but who you know are the same person	08:00
gitster	keeping an irc client open risking getting distracted is way far less efficient; skimming the channel log every three days is a lot more efficient way to stay up to date.	08:00
MadCoder	heh right	08:00
gitster	Yeah, I've discussed with ohloh people in their "forum" about it a few times.	08:01
	The easy thing they could do is,	08:01
	Just register a fake user themselves, have _him_ claim "I am this person" to consolidate the entries.	08:01
MadCoder	and unregister him ?	08:02
gitster	Because they _must_ have an identity dispute procedure already established,	08:02
MadCoder	well I would be surprised you would need such perverts ways to achieve that	08:02
	probably	08:02
gitster	just in case if I claim to be you by mistake.	08:02
	So they do not even have to unregister that fake person. They can just wait until the real one comes and disputes.	08:02
MadCoder	that's an option	08:03
gitster	Then they can transfer the identities to the real owner and retire the fake one.	08:03
	→ cods joined	08:04
gitster	Yay, we are at 165 tied with gitte's unfavorite language ;-)	08:04
MadCoder	heh	08:06
	← drizzd left	08:08
Ilari	What language is that?	08:09
MadCoder	it's not a language but some kind of library or toolkit or sth similar	08:09
	for a language that smells coffee	08:09
gitster	There are a handful entries at 165 stacks last time I checked a few minutes ago.	08:10
	MadCoder: Bzzt.	08:10
	It is my understanding that gitte does not have much against that language. What he dislikes is a Japanese one, not Javanese one.	08:11
	heh, they recomputed and Linus is at rank 43 now.	08:12
MadCoder	oh it's a ruby thing ? I thought it was for java	08:14
	"only" 43 ?	08:16
	heh	08:16
	gitster: heh, welcome in the elite ;)	08:17
gitster	well, I am a bit unhappy that this moved linus a bit above my second mentor X-<.	08:18
	but that's fine because linus is my third mentor anyway ;-)	08:19
MadCoder	the two previous being ?	08:19
	the tar guy and ?	08:19
	→ xiaoxiaoqing joined	08:21
	← xiaoxiaoqing left	08:22
eMBee	good evening	08:24
gitster	good whatevertime.	08:24
	MadCoder: don't worry, my first mentor predates this whole open source craze and nobody would know him.	08:25
MadCoder	heh	08:25
	so you're a dinosaur then	08:25
gitster	depends on how old I was when I was trained by him ;-)	08:26
eMBee	computer related training? or martial arts or something?	08:27
MadCoder	haha	08:27
gitster	joy of hacking.	08:27
	← up_the_irons left	08:28
eMBee	hmm, i think you can teach the joy of hacking to a 5 year old, so yeah, that doesn't make you old	08:31
	→ up_the_irons joined	08:32
	→ roundeye joined	08:38
roundeye	odeley cholos	08:38
	im looking for some good docs on encryption	08:38
MadCoder	and what make you think #git is the place to ask for it ?	08:40
roundeye	why not	08:40
	go tr0ll some one else	08:40
	are you saying that the people in git are nubs?	08:40
gitster	MadCoder: two questions.	08:49
blokkie	how do I name my project and branch ? now it's like this : http://first.h0m3.be/pipermail/svn-grudge/2007-September/000191.html	08:52
gitster	this is quite up to the person who set up the hook script in the repository.	08:53
blokkie	I am	08:54
	← roundeye left	08:54
gitster	Are you using the Andy Parkins script from contrib?	08:54
blokkie	so I need to adjust the config on my git server ?	08:54
	yes	08:54
	my config does not contain [branch]	08:54
gitster	It seems "UNNAMED PROJECT" is taken from ".git/description" (or lack thereof).	08:55
	I am not following your question about "branch". I think the mail look funny because it talks about "UNNAMED PROJECT" but otherwise I do not see anything funny about it.	08:56
	It would say something like "Blokkie's Cool Gadget branch master updated."	08:56
	when the branch 'master' was updated.	08:56
blokkie	ah ok , I adjusted the description , cool	08:57
	now about [branch]	08:57
	I don't have any such thing	08:57
	in my config	08:57
gitster	That has a good side effect of showing that name in the gitweb.	08:57
blokkie	onlt core and hooks	08:57
	← metafollic left	08:57
gitster	What makes you say you need [branch]?	08:57
	← up_the_irons left	09:00
	→ garry joined	09:01
MadCoder	gitster: yes ?	09:01
	garry → up_the_irons	09:02
gitster	(1) what's the rationale behind slop[32]? Not slop[4] nor slop[1] nor slop[100]?	09:02
MadCoder	oh, in fact I think 8 should be enough	09:02
	it's to be sure that if we read a intmax_t it fits	09:02
gitster	(2) without strbuf_init() assigning slop[0] = '\0', does it still guarantee that "people can always assume buf is non NULL and ->buf[0] is '\0'"?	09:03
MadCoder	I don't understand (2)	09:03
	I mean I don't get what you are asking	09:04
gitster	The comment says slop is to guarantee that people can safely assume buf.buf != NULL and ->buf[0] is '\0'. I do not understand the latter guarantee.	09:05
MadCoder	the slop is full of zeroes right ?	09:05
	and after init everyone shares the same	09:05
	→ metafollic joined	09:06
MadCoder	in fact what I mean is that with this thing, after initialization, the "empty" strbuf is not special cased anymore	09:06
gitster	After init everybody shares the same and, ah, after init ->len is 0 so nobody is supposed to write into there without first _grow() which would allocate the real thing.	09:06
MadCoder	->buf holds a NUL terminated buffer even with alloc == 0	09:06
	exactly	09:06
gitster	ok, I see.	09:06
thiago_home	shared_empty	09:06
MadCoder	though there may be people that would do a ->buf[0] = '\0' and this still works	09:07
gitster	Then that makes me confused about your first answer.	09:07
MadCoder	(people doing that would be wrong but ...)	09:07
gitster	rtrim does that.	09:07
thiago_home	I would say that you cannot access buf[0] if len == 0	09:07
MadCoder	thiago_home: you can access buf[len] in a strbuf	09:08
	and it's always \0	09:08
	that's the whole point of strbuf's	09:08
thiago_home	hmm... ok	09:08
	thiago_home is too used to Qt	09:08
thiago_home	you can't access str[str.length()]	09:08
MadCoder	gitster: well maybe my first answer is stupid	09:08
	maybe a 1 char buffer is enough	09:08
gitster	I do not mind wasting 31 bytes but I want a good rationale behind any magic numbers.	09:09
MadCoder	well, that sounded clever at the time	09:09
	and now I do not see how to justify it	09:09
	so it's probably dumb	09:09
	8 would preserve good alignement though	09:10
gitster	Ok, let's shrink it to 1 and see what breaks ;-)	09:10
MadCoder	but I suppose the C compiler already does that	09:10
gitster	That's a good rationale.	09:10
thiago_home	char[n] has an alignment of 1	09:10
gitster	yes.	09:10
MadCoder	thiago_home: I meant the address of the array	09:10
thiago_home	you can't do that with a char array	09:11
MadCoder	anyway we're not supposed to read more than 1 byte in it	09:11
gitster	I do not think you nor compiler need to align char[].	09:11
thiago_home	if you want to force the alignment, you need to have a field with a higher alignment constraint	09:11
MadCoder	so alignment hardly matters anyway	09:11
thiago_home	yep. Why force it?	09:11
gitster	let's make it slopbuf[1] and see what breaks ;-)	09:11
MadCoder	gitster: so well, char[1] may be totally okay	09:11
gitster	I do not think there should be none.	09:12
	→ mmagnon joined	09:12
MadCoder	in fact s/may/will/	09:12
gitster	s/do not//;	09:12
	← mgrimes left	09:20
gittish	someone updated some example scripts and added error checks to open for pipes, i.e. open X,"git .\|" \|\| die, but didn't add the test at the more likely place to see and error, ie. at close()	09:35
	← tko left	09:36
	→ kumbayo joined	09:37
gitster	typical mistake.	09:37
	→ kanru joined	09:37
	→ zecke joined	09:40
	→ alley_cat joined	09:41
	→ tko joined	09:45
udi_trugman	no	09:48
	→ engla joined	09:55
gittish	no?	09:55
	→ cortilap joined	10:01
	→ drizzd joined	10:02
	← alley_cat left	10:09
	→ alley_cat joined	10:10
	← CIA-11 left	10:15
	← G0SUB left	10:20
	← zecke left	10:26
	→ Lash\| joined	10:28
	→ CIA-11 joined	10:29
	← Lash\| left	10:33
	→ zecke joined	10:37
	← kumbayo left	10:40
engla	git rev-parse --branches prints the shas to all branch heads	10:41
	when I do git rev-parse --short --branches it writes them all out in short form and _then_ exits with "fatal: Needed a single revision"	10:42
	→ nud_ joined	10:42
Auris-	kinda like cryptsetup which complains and segfaults when run as non-root.	10:44
	← zecke left	10:55
	← eMBee left	11:01
nud_	cworth: great introduction to git in your latest blogpost	11:03
	cworth: anyway, there is a little typo in there, in 2.7.1: "If any of the GITAUTHORNAME, GITAUTHOREMAIL, GITCOMMITTERNAME, or GITCOMMITEREMAIL..." <-- those are missing the underscores	11:03
	→ eMBee joined	11:08
	← CIA-11 left	11:08
	→ Lash\| joined	11:08
	→ branstrom joined	11:12
	→ CIA-11 joined	11:12
eMBee	can i use the oprions to git-diff with git-cherry?	11:14
	options	11:14
MadCoder	no	11:19
	git-cherry cannot take more than 4 arguments	11:20
	though you can probably pipe the output to git-rev-list --stdin with appropriate options	11:22
	which would not help silly me	11:22
	git-diff-tree --stdin rather	11:22
eMBee	well, i want to influence the revisions that git-cherry produces	11:23
MadCoder	oh	11:23
	well no git-cherry doesn't do that	11:23
	though it would definitely make sense	11:24
eMBee	or rather, i want to compare two trees, but only find differences that are not caused by missing files in one tree	11:24
MadCoder	ask for the feature on the list ? :)	11:24
	eMBee is not on the list yet, should make a note of that	11:25
eMBee	any workaround suggestion?	11:26
MadCoder	well, I'm in need for substantial food, and it deprives me from my neurons efficiency (which isn't that high usually, so imagine right now)	11:26
	eMBee: if there is a way to tell git log / rev-list/... I want all the patches that touches at least one other file than "your file"	11:27
	then you can make the intersection of the one git chery gives and those	11:28
	I'm not sure if git log/rev-list have this kind of expressivity though	11:28
	(or you can list every file in the repository except these)	11:28
eMBee	hmm, i don't know what the file would be though	11:29
	i could approach this from a different angle and look at the time the change was commited	11:30
	in this case i know that the commits happened at the same time	11:30
	(background is that i am comparing two cvs trees, one of which has files missing)	11:30
	as a result ALL commits in the trees will be different, but i want only those commits which are different by more than the missing files	11:32
	← Lash\| left	11:37
	← branstrom left	11:39
	→ zecke joined	11:46
	→ Lash\| joined	11:47
	→ branstrom joined	11:49
gittish	create two branches, one for each tree, delete the to-be-ignored subtrees and compare	11:51
	not very practical if you want to compare over and over with different trees	11:52
eMBee	i don't want to compare trees but commits	11:52
	erm	11:53
	there seems to be tweo uses for the term tree here	11:53
	i am talking about the commit/branch tree not the file/directory tree	11:53
	one branch has A--B--C--D and another A'--B'--D' where A and A' differ only by some files, now i want to find that C which has no equivalent C'	11:55
	← janm left	11:55
eMBee	right now i am attemting a brute-force comparison of the commit timestamps, because i know they are identical for A and A' ...	11:56
	← nud_ left	11:58
eMBee	does git-rev-log --all give me all commits in the whole repo?	12:01
vmiklos	git-rev-log?	12:02
	→ Oejet joined	12:03
	→ Evelynn joined	12:03
eMBee	erm git-rev-list	12:05
	← zecke left	12:05
	→ Sho_ joined	12:05
eMBee	sorry, got mixed up, was looking at git-log which made me look at git-rev-list for options, got them merged in my mind into git-rev-log :-)	12:06
	→ mneisen joined	12:06
vmiklos	git-rev-list (by default) will only list commits from the current branch i think. so no, it won't list all commits	12:07
eMBee	which is why i am asking about the --all option	12:07
	the manpage describes it as: Pretend as if all the refs in $GIT_DIR/refs/ are listed on the command line as <commit>.	12:08
vmiklos	ah, my bad	12:08
	then yes, it seems so	12:08
eMBee	that seems to be enough to make it list all commits, but i am not sure if it really means that	12:08
	→ janm joined	12:13
	→ chris2 joined	12:14
	→ nud joined	12:14
	→ Yuuhi joined	12:23
	→ zecke joined	12:23
	← fowlduck left	12:27
	→ madewokherd joined	12:29
	→ Beket_ joined	12:39
	→ zecke_ joined	12:39
	← zecke left	12:51
	→ kristoffer joined	12:54
	→ FunkeeMonk joined	12:57
	→ zecke joined	12:57
gittish	eMBee: a commit implies exactly one tree. so when comparing two tress, you are actually comparing two trees.	12:58
	(update) eMBee: a commit implies exactly one tree. so when comparing two commits, you are actually comparing two trees.	12:58
	the only difference is that a tree can refer to any tree, but a commit implies the top level only	12:59
eMBee	right, my point was that i am not so much interested in the state of the tree but in the changes the commit introduces	13:02
gittish	still the same thing	13:03
	→ bentob0x joined	13:04
eMBee	not quite: two commits can be different but the changes they introduce could be the same	13:04
gittish	man git-cherry	13:05
	could that be something for you?	13:05
eMBee	no, already been there, problem is i need to find a specific set of differences (ignoring missing files in one tree)	13:07
gittish	is that your real problem or a hypotetical partial solution to your problem?	13:08
	← zecke_ left	13:09
gittish	what you are asking for in not available in the current version of git	13:09
eMBee	my real problem is that i have 7 branches of a cvs repository, where the branches have been created by copying the whole repository. the branches represent the stable releases. i pulled all those into git, and now i want to create a merged repo that shows branches at the actual branchpoint. problm is in the newer branches files have been deleted from the repo (maybe they forgot to copy the attic or something) so on importint into git the commits wil	13:11
	← Beket_ left	13:11
eMBee	now what i am trying to find out is if missing files is the ONLY thing where the repos differ, or if there are other differences as well (and there are)	13:11
gittish	i mentioned something above that doesn't scale well, but for seven branches it is quite feasible	13:13
	seven branches and a one-time event	13:13
eMBee	well, i pretty much need to compare all the commits	13:14
gittish	filter-branch then	13:14
	replicates your branches commit by commit removing stuff	13:14
eMBee	hmm	13:15
	that could be interesting	13:15
	← CIA-11 left	13:15
eMBee	although i'd rather add stuff, but that should work too	13:15
	← FunkeeMonk left	13:16
gittish	that would be harder	13:16
eMBee	you are right	13:16
	but that's what i need to do in the end	13:17
	hmm	13:17
	this will take more thinking	13:18
	there are more than 10 years of history	13:18
gittish	hmm, you won't even have the ability to do commit-by-commit compare, I think	13:19
	those copied repoes reflect the true history of those branches. They did delete files.	13:20
eMBee	yes, well, the trouble is deleting the files resulted in deleting the history of those files	13:21
gittish	that would be a a problem if you want to merge	13:22
	(or well compare)	13:22
eMBee	that is the problem indeed	13:22
	which i am trying to solve	13:22
gittish	otoh, what happens if you do?	13:23
eMBee	compare? well all commits are different, rather hopeless without more intelligent heuristics	13:23
gittish	so do you need to compare commit by commit, or will the tips solve your problem in practice	13:24
eMBee	i am not sure yet	13:25
gittish	another version is to rebase those copies onto a version that has all files to begin with	13:25
eMBee	yes, hmm	13:26
	indeed, that might do something	13:26
gittish	i doubt you'll see a clean solution, but hopefully one that doesn't suck very much	13:26
eMBee	but the rebase would only give me the initial file not it's history	13:27
	its	13:27
gittish	the copied branch didn't change it, so there is no further history on that branch	13:27
eMBee	well, i do know where the branch points are pretty much, and i did create a merged tree by importing from cvs onto the branchpoint, however i found that some commits got lost somewhere in the process	13:28
gittish	you want a "git knit" command	13:28
eMBee	:-)	13:28
	→ CIA-11 joined	13:29
eMBee	the copied branch deleted the file and its history, so if i rebase i need to rebase at the branchpoint, not rebase the while branch, i did that	13:30
	now i need to find those lost commits, which requires looking at the original repos	13:30
	→ _saserr_ joined	13:30
engla	the discussion is above my head, but you are going to rewrite the history to put in the missing history in all branches? terrific	13:31
	← _saserr_ left	13:33
	→ asdx joined	13:34
eMBee	maybe an example can help: consider A--B--C--D--E--F and A'--B'--C'--E'--G, two histories, created by copying at the point of E. now i am trying to create a merged history that has A--B--C--D--E<(branch here into F and G)	13:34
	→ nikodemus joined	13:34
	← nud left	13:34
eMBee	but notice how those histories have D' missing	13:34
	← nikodemus left	13:34
eMBee	commits like that i am trying to find	13:35
	→ nud joined	13:35
gittish	I think I understand, but I'm not convinced it is worth it. Interesting problem though.	13:35
eMBee	well, the critical issue is that some of those commits are remove of files	13:36
	in the head of the newst branch i have files that should have been deleted years ago	13:36
	i need to find where they have been deleted and add that delete commit into the history	13:37
	other irregularities are not so critical, but this part is, as at least the heads of all branches need to be correct	13:38
gittish	hmm, git-merge -i and give it all your commits in those branches	13:38
	rebase -i	13:38
eMBee	yes, after i found where the files need to be deleted	13:38
	im still at the stage of figuring out how to find that	13:39
	once i found them, branch, delete and rebase are clear steps	13:39
gittish	rebase -i will cherry pick, but skip if a commit effectively introduces no change because it has already been applied	13:40
	→ sbohrer joined	13:40
gittish	but it won't tell you how many times it will stop beforehand	13:41
eMBee	the whole repo has about 26000 different commit timestamps	13:42
gittish	good luck....	13:43
eMBee	hehe	13:43
	hmm 29000	13:44
gittish	i'd skip fixing the history and just fix the tips. Even if you think you fixed it, can you really verify that what you did is correct?	13:44
eMBee	well, i believe finding out when those files have been deleted should not be to hard	13:45
	and give a reasonableapproximation	13:45
	to find what i did is correcti i can check out at each release (which is tagged), and compare with the actual release, or a checkout from cvs	13:46
	should be able to script that	13:46
	i want to do this in preparation of getting the project to switch to git, the better the job the more chance i have of convincing them it is a good idea	13:47
	well, thanks for the support so far. it is 2am here (and we got daylight savings coming up tonight, so it's actually 3) time to go to sleep	13:49
	→ G_SabinoMullane joined	13:50
	← Lash\| left	13:51
	← CIA-11 left	14:05
	→ krawek joined	14:11
	← bentob0x left	14:12
	→ CIA-11 joined	14:19
	→ _ry joined	14:33
	← _ry left	14:35
	→ kumbayo joined	14:39
	→ rtmfd_icbm joined	14:40
	← agoode left	14:51
	→ bdiego joined	14:58
	← nud left	15:03
	→ nud joined	15:08
	→ mmass0nnet joined	15:08
	← segher_ left	15:10
	← m8t left	15:11
	mmass0nnet → m8t	15:11
	← branstrom left	15:12
	→ branstrom joined	15:19
	→ cdcarter joined	15:20
	← cdcarter left	15:22
	← alley_cat left	15:35
	→ alley_cat joined	15:36
	← drizzd left	15:37
	→ drizzd joined	15:37
	← CIA-11 left	15:39
	→ nessundorma joined	15:45
	→ robfitz_ joined	15:45
	→ mithraic joined	15:46
madduck	http://blog.madduck.net/vcs/2007.09.29_uninformed-slagger	15:46
	thank you, git people...	15:47
	you made writing this post a joy!	15:47
	→ radiomark joined	15:52
	← mithraic left	15:53
	→ mithraic joined	15:54
mithraic	So I'm thinking of migrating my svn home-dir to git, mainly for the more transparent rename support, which i anticipate making it much easier to work naturally.	15:55
	Any particular gotchas that folks think I should be aware of? Managing 'subprojects' seems to be the biggest sticking point thus far.	15:56
	← robfitz left	15:58
	→ CIA-11 joined	15:59
Mikachu	mithraic: don't accidentally type git-clean	16:03
mithraic	:)	16:03
	I'm still sorting out: if I want different bits to have different access privileges, they need to be in separate repositories, correct?	16:04
	→ Nindi joined	16:08
Nindi	If I do git-log --pretty=oneline the output is	16:09
	6e0ba847b5193aca66cd8e048f5b61da3fdc01ff	16:09
	1ca8c07ccf47bb39bc6e54fa0254ff46fa64aed5	16:09
	73b4def933c53fb0c32e811b4b6d784b0a3177dd	16:09
	then if I do	16:09
	git-bundle create bundle e755..6e0b	16:09
	I get fatal: Refusing to create empty bundle.	16:10
	can anyone explain what I am doing wrong ?	16:10
nessundorma	Nindi, what does 'git log e755..6e0b' say?	16:12
	← jlh left	16:14
	→ jlh joined	16:14
	← CIA-11 left	16:21
	← bdiego left	16:24
madduck	mithraic: http://lists.madduck.net/listinfo/vcs-home btw	16:26
mithraic	lovely, thanks!	16:27
madduck	mithraic: also, http://albatross.madduck.net/pipermail/vcs-home/2007-September/000095.html	16:27
	and about submodules: http://albatross.madduck.net/pipermail/vcs-home/2007-September/000096.html	16:27
mithraic	yeah, that's the one i'm reading :)	16:28
madduck	mithraic: colgit still has quite a bunch of issues	16:30
	and i really want to rewrite it in python, i think	16:30
	so if you're interested, i am looking forward to working with you.	16:30
mithraic	Thanks, I'll take a look at it, and consider my needs.	16:31
	→ bremac joined	16:31
mithraic	i have some crappy collection of shell scripts to manage a fleet of subversion homedirs at the moment.	16:31
	helping put together something for git would be great.	16:31
	→ CIA-11 joined	16:38
	→ dous joined	16:40
engla	Git, the information manager from hell: Deltifying 666 objects...	16:41
	← bremac left	16:43
	→ bdiego joined	16:52
	← dous_ left	16:57
	→ mithraic_ joined	17:01
	← G_SabinoMullane left	17:05
MadCoder	gitster: morning ;)	17:05
	→ Beket joined	17:07
cworth	nud: Thanks for the fix. The source had underscores I'm quite sure, but the markup (markdown) must be eating them. I'll fix those soon.	17:08
Ilari	cworth: Also, '<' and '>' around one email address seemed eaten...	17:14
	← CIA-11 left	17:14
cworth	Ilari: Thanks, I'll check that too.	17:14
	← mithraic left	17:17
Ilari	cworth: It seems that it appears properly in code blocks, but not otherwise.	17:18
	← nessundorma left	17:18
	← mneisen left	17:21
	→ bentob0x joined	17:23
	Insount_away → Insount	17:26
	← sheskar left	17:27
	→ CIA-11 joined	17:28
	→ Lash\| joined	17:33
	← Beket left	17:36
	← orospakr left	17:39
	← CIA-11 left	17:40
	← alley_cat left	17:46
	→ CIA-11 joined	17:50
	← radiomark left	17:51
	→ alley_cat joined	17:57
	← mithraic_ left	17:58
	→ mithraic joined	17:58
	→ orospakr joined	18:07
jaalto	After pull, what does this command mean:	18:17
	.	18:17
	Warning: No merge candidate found because value of config option	18:17
	"branch.master.merge" does not match any remote branch fetched.	18:17
	No changes.	18:17
	.	18:17
	→ iabervon joined	18:18
iabervon	Anybody knowledgable about submodules?	18:19
vmiklos	jaalto you need something like git config branch.master.merge refs/heads/master	18:20
jaalto	vmiklos: what does that do?	18:20
vmiklos	it means that you want to merge the remote repo's master branch to your local master branch	18:21
	→ arwed joined	18:21
	← alley_cat left	18:22
	→ lcapitulino joined	18:23
jaalto	vmiklos: Did that 'git config', but it still gives the same message	18:24
Mikachu	maybe you didn't create the local branch with --track	18:25
jaalto	I didn't. How can I proceed?	18:26
vmiklos	try git config branch.master.remote = origin	18:26
	→ Dodji joined	18:28
jaalto	vmiklos: Good. That made it. Thanks	18:28
	vmiklos: If you could spare a moment to explain what this exactly means:	18:29
	.	18:29
	[branch "master"]	18:29
	merge = refs/heads/master	18:29
	remote = vc	18:29
	.	18:29
	(I pull from alias "vc", not "origin")	18:29
iabervon	When you're on the branch "master", the default remote to fetch/pull/push is "vc", and the default thing to merge is "master" from that remote.	18:30
jaalto	iabervon: Clear. Thank you	18:30
	← kumbayo left	18:35
	→ dduncan joined	18:37
	← arwed left	18:39
	→ arwed joined	18:43
	← chris2 left	18:46
	← dous left	18:48
	→ dous joined	18:49
	→ KirinDave_ joined	18:56
	KirinDave_ → KirinDave	18:57
	← CIA-11 left	19:02
	→ mnemoc joined	19:04
	→ CIA-11 joined	19:14
	← lcapitulino left	19:18
	← drizzd left	19:21
	← KirinDave left	19:21
	← CIA-11 left	19:24
	→ IRCMonkey joined	19:26
	→ alley_cat joined	19:27
	← arwed left	19:27
	→ kelvie joined	19:27
IRCMonkey	Hi, if I want to access (read, git-cat-file), a previous version of a given file; I can do a git-checkout of the related commit; but if I just just cat the file, how could I do?	19:30
vmiklos	do you want to checkout a previous tree or get the contents of a given blob?	19:30
IRCMonkey	If I have the sha1 of a given commit, how can I find the sha of a given file in this commit?	19:31
iabervon	git cat-file blob <revision>:<filename>, if I understand you correctly.	19:31
IRCMonkey	get the content of a given blob.	19:31
vmiklos	or just git show <commit>:<path>	19:31
IRCMonkey	You did (understant correctly)	19:32
	fine	19:32
Ilari	IRCMonkey: Depending on what exactly you want do do with it, there may be easier ways....	19:32
IRCMonkey	show 1ea17ef25:Makefile, is this ok?	19:33
	→ CIA-10 joined	19:33
IRCMonkey	to see a previous version of my Makefile	19:33
Ilari	IRCMonkey: Yes, looks sane.	19:33
IRCMonkey	This way seems fine for the moment.	19:34
	thanks, see you	19:34
Ilari	IRCMonkey: 9 hex commit specification starts to break down at about 256K objects...	19:35
	IRCMonkey: Because at that point, collisions become likely and afterwards number of collisions increases quite fast.	19:35
IRCMonkey	Ilari: (back again) "9 hex specification", it meens "giving only the first 9 hex char of the sha1? Is that it?; By the way, if there would be a collision between two sha1, could it not check if they are equal or not? Like we do usually with a hash function?	19:46
Ilari	IRCMonkey: Yes, abbrevating commit ID with 9 first hex digits. The for full 40 digits, the breakdown point is approximately 1Y objects...	19:48
	← alley_cat left	19:50
Ilari	IRCMonkey: Cryptographic hashes usually don't have collision checks.	19:52
IRCMonkey	Ilari: I saw somewere a function to calculate the git-sha of a given file	19:54
	because sha1 myfile, wont give me the correct answer	19:54
iabervon	git hash-object does it; there's a header and stuff.	19:55
	→ drizzd joined	19:56
	→ kristoffer_ joined	19:56
IRCMonkey	echo "toto"\|git-hash-object --stdin ok	19:58
	ilari: could it be possible to do the processing git does to convert a file into an object by hand?	20:00
	And what about the converse processing?	20:00
	← bdiego left	20:01
Ilari	IRCMonkey: The code to do it isn't terribly complex (as long one is talking about loose objects).	20:01
IRCMonkey	Yes loose objects	20:01
	for instance can you do some thing with "gunzip", on a bolb ?	20:02
Ilari	IRCMonkey: Basically, you need to convert to canonical format (by prepending headers) and then run it through zlib deflate compression.	20:02
	IRCMonkey: Which is _NOT_ the same as gzip compression (diffrent framing format).	20:02
IRCMonkey	If I calculate the length of the header (how), and prepend it...	20:03
	not the same, the same library?	20:03
	zlib	20:03
	But I need to do it in "c"	20:03
iabervon	It's the same compression method, but gzip puts different things on the very beginning and end.	20:04
engla	but the hash is of the uncompressed blob or unc blob + headers?	20:04
IRCMonkey	uncompressed, yes	20:04
engla	iabervon: how do you convince gzip to unpack an object manually?	20:04
Ilari	IRCMonkey: These headers contain <objecttype><space><size><nul>, where space is ASCII SPACE, nul is ASCII character 0, objecttype is one of 'blob', 'tree', 'commit', 'tag' and size is size of object expressed as ASCII number.	20:04
iabervon	I don't think gzip can be convinced.	20:04
IRCMonkey	<nul> : stop at the first nul char	20:05
	Ilary: zlib, not hard to use?, nothing special about it? That would be interesting to actually try.	20:07
Ilari	> echo -en "blob 0\x00" \| sha1sum	20:08
	e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 -	20:08
	echo -n \| git-hash-object --stdin	20:08
	e69de29bb2d1d6434b8b29ae775ad8c2e48c5391	20:08
	← mnemoc left	20:08
IRCMonkey	Ilary: that's cool	20:09
engla	Ilari: so headers are included in the hash. Which is probably sane..	20:09
	← mutex left	20:10
	→ DrNick joined	20:10
Ilari	engla: One would need only to preturb object ID calculation with object type. The size does not really help much.	20:10
IRCMonkey	Convincing.	20:10
engla	Ilari: absolutely	20:11
Ilari	IRCMonkey: Actually, there are two loose object formats. The older (the one that is used) is just zlib compressed stream of canonical object representation.	20:11
engla	what is the other format? where is that used?	20:12
Ilari	engla: It is not used. Write support was removed (read support remains).	20:12
iabervon	The type has to be in the hash; otherwise, if you did "git cat-file commit HEAD > name; git add name" it would be unhappy trying to store a blob and a commit with the same hash.	20:13
engla	exactly	20:14
	I saw that git has code to detect collisions	20:14
	but since everything else is in place the only possibility is a sha1 collision	20:15
	we don't even have a test for that code path :)	20:15
IRCMonkey	engla; has it code to detect collision?	20:15
iabervon	Someday, somebody will find a sha1 collision, and we can use it...	20:15
	← kristoffer left	20:16
IRCMonkey	engla: that's why I asked a way to calulate the sha : git-hash-object, to make the test myself	20:17
engla	index-pack.c: 339	20:18
	in branch maint	20:18
	die("SHA1 COLLISION FOUND WITH %s !", sha1_to_hex(sha1));	20:18
	:)	20:18
Ilari	engla: Not all SHA-1 collisions are useful to construct colliding git objects.	20:19
engla	exactly	20:19
Ilari	But should SHA-1 get broken to same level as MD5...	20:19
	← z3ro left	20:20
	→ z3ro joined	20:21
IRCMonkey	engla: I don't understant: is there actually a collision test if I git-update-index a file?	20:21
Ilari	IRCMonkey: That collision check is AFAIK used only for fetches.	20:22
mithraic	Can someone devise a file whose first line is the SHA-1 hash of the entire file?	20:23
	That'd be a nice trick.	20:23
Ilari	mithraic: That would be very difficult. Much more difficult than collision...	20:23
mithraic	a kind of quine.	20:24
engla	I don't know where that function is used. but it just checks if we have that file already, and does a sanity check of size, type etc.	20:24
Ilari	mithraic: It is however possible to construct program that prints SHA-1 of its source code...	20:24
mithraic	without actually running the sha-1 algorithm on the source code?	20:25
Ilari	mithraic: That actually extends to machine language, so it is possible to write ELF file that prints its own SHA-1 hash. Without reading itself.	20:26
	→ evan joined	20:26
evan	why the heck is gitweb stripping all spaces?	20:26
	and emitting that it doesn't understand	20:26
IRCMonkey	Also, to be sure a move will be commited correctly, we can calculate the hash of the object after the move and compare with the one of the same object before. that must be the same.	20:27
Ilari	mithraic: Actually, it internally runs SHA-1 on reconstructed copy of itself.	20:27
mithraic	i see. so it's a classic self-quoting combined with a sha1 implementation. interesting.	20:27
Ilari	mithraic: All turing complete languages with unrestricted printing operators have quines.	20:29
Nindi	If I do git-log --pretty=oneline the output is	20:34
	6e0ba847b5193aca66cd8e048f5b61da3fdc01ff	20:34
	1ca8c07ccf47bb39bc6e54fa0254ff46fa64aed5	20:34
	73b4def933c53fb0c32e811b4b6d784b0a3177dd	20:34
	then if I do	20:34
	git-bundle create bundle e755..6e0b	20:34
	I get fatal: Refusing to create empty bundle.	20:34
	can anyone explain what I am doing wrong ?	20:34
	← evan left	20:34
Ilari	Nindi: What does 'git log e755..6e0b' return?	20:34
	← Evelynn left	20:35
Nindi	one sec and I will find out	20:35
	Ilari: It shows me the cirrect commits	20:35
Ilari	Maybe you can't use non-ref as upper bound in bundles...	20:35
	← drizzd left	20:35
Nindi	ilari that could be true becausgit-bundle create bundle e755..6e0be	20:36
	git-bundle create bundle e755..HEAD works fine	20:36
	but why ?	20:37
Ilari	Nindi: Bundles have refs too which pin commits inside bundle...	20:37
Nindi	Ilari: I don't underrstand that	20:38
Ilari	Nindi: Look at the created bundle file using something like 'less' (the part before the binary gunk begins).	20:38
Nindi	ilaria : ok	20:39
	am looking .. I can see two SHA IDs	20:39
	→ alley_cat joined	20:40
Ilari	Nindi: One of them has 'HEAD' after it, hasn't it?	20:42
Nindi	ilari : no -73b4def933c53fb0c32e811b4b6d784b0a3177dd cmmit 3 into 1	20:42
	762e5a8dd8efa5498c5dadb8ee596fa6c1471eb1 refs/heads/master	20:42
	it only has	20:42
	-73b4def933c53fb0c32e811b4b6d784b0a3177dd cmmit 3 into 1	20:42
	762e5a8dd8efa5498c5dadb8ee596fa6c1471eb1 refs/heads/master	20:42
Ilari	Nindi: Ah, of course. It resolves symrefs.	20:43
Nindi	cmmit 3 into 1 was y 3rd commit message	20:43
	ilaria : could you elaborate	20:43
Ilari	Nindi: It interpretted 'e755..HEAD' as 'e755...refs/heads/master', because HEAD is 'symbolic link' to refs/heads/master...	20:44
	← IRCMonkey left	20:45
Nindi	ok , but that still doesn't explai Why I can't bundle all the cvommits between to SHAIDs	20:45
Ilari	Try running 'git ls-remote foo.bundle'.	20:45
	(Yes, bundles do have remote branches).	20:46
madduck	bundles?	20:46
	is that a new feature?	20:46
Nindi	it gives me	20:46
	762e5a8dd8efa5498c5dadb8ee596fa6c1471eb1 refs/heads/master	20:46
	← Romster left	20:48
Ilari	madduck: Not new one: Author date is 20070222 005914 UTC...	20:48
	Nindi: Meaning that the bundle has one ref, named refs/heads/master, which points to 762e5a8dd8efa5498c5dadb8ee596fa6c1471eb1.	20:48
madduck	Ilari: ah yeah, now i remember.	20:49
Nindi	Ilari: ok ... are you saying the the upper bound must be a ref ?	20:49
	→ Romster joined	20:49
Ilari	Nindi: Yes. Upper bounds given apparently must be refs.	20:49
Nindi	ilari : why ??	20:50
	Ilari: is there a rational reason for that ?	20:50
	: ilar so taggin should work	20:50
Ilari	Nindi: Lower bounds don't have to be. The upper bounds need all to be named in the bundle.	20:51
	Nindi: In your case, there is one upper bound named 'refs/heads/master'.	20:51
Nindi	but thhe SHA IDs are ARE names	20:51
Ilari	Nindi: But there must be reference name too...	20:52
Mikachu	does it have to do with that you can only git-fetch named refs and not random sha1s?	20:52
	← alley_cat left	20:52
Ilari	Mikachu: Exactly.	20:52
Nindi	ok Tagging worked	20:53
Ilari	Nindi: Even tracking branches (or even more exotic kinds of refs) can be used as upper bounds.	20:54
Nindi	I tagged a random commit .. and git-bundle create bundle 654...someTag worked	20:54
	← cortilap left	20:54
	→ mutex joined	20:56
Ilari	Nindi: IIRC, there was some bug with bundling annotated/signed tags fixed some time ago...	20:57
	Nindi: First release containing fix was 1.5.3-rc5.	20:59
Nindi	Ilari: what exactly was the bug ?	21:00
Ilari	Nindi: As far as I can tell, It treated the annotated/signed tag as simple tag.	21:01
Nindi	Ilari: Am sorry .. but am quite new to git, so what exactly does that mean ... and what does it mean in terms of my problem ?	21:03
Ilari	Nindi: It is not related to that can't-use-numeric-IDs-for-upper-bounds problem	21:03
	→ Ilari_ joined	21:04
	← Ilari left	21:04
	← Ilari_ left	21:04
	→ Ilari joined	21:05
	Ilari hit the wrong button...	21:05
	→ clausen joined	21:07
clausen	th	21:07
	hi all	21:08
	→ mutex_ joined	21:08
Ilari	Nindi: In practice it meant that if commit was signed via the tag, the signature was lost, as well as other explanation introduced via that tag.	21:08
clausen	is there an easy way to start a new repository from some pre-existing files	21:08
	with "push/pull"	21:08
	(the new repository should be "bare")	21:08
Ilari	clausen: 'git init' 'git add .' 'git commit' (run in the directory with files).	21:08
clausen	Ilari, that wouldn't be bare	21:09
	I have been doing this:	21:09
Ilari	clausen: Then create new bare repository (git --bare init --shared) and push the created repostory there.	21:09
clausen	ah	21:09
	great	21:09
	will git automatically know where to push/pull from?	21:10
Ilari	clausen: Or if it is updated only by you, you can omit the --shared.	21:10
clausen	what does --shared do?	21:10
	ah, just permissions	21:10
Ilari	clausen: Sets shared mode, in which git forces group access flags to match user access flags in files it creates inside repo.	21:10
	clausen: A way to override umask.	21:11
clausen	yeah, I just rtfm :)	21:11
Ilari	clausen: It might be easier first to create remote for the new bare repo (git remote add) and then 'git push --all'.	21:12
	clausen: Actually, 'git push <remotename> --all'.	21:12
	← CIA-10 left	21:13
Ilari	clausen: Then one can update what's there with 'git push <remotename>', or 'git push <remotename> <branch>' to update/add branch.	21:14
clausen	here's another way to think about it:	21:14
	I have a full repository	21:14
	I make a bare clone of it	21:14
	(git-clone --bare project project.git)	21:14
	and then I want to make push/pull default to "project.git"	21:14
Ilari	clausen: In 'project', add project.git as remote using 'git remote add foo ../project.git', where foo is name of remote.	21:15
	clausen: The default used by push/pull is 'origin' unless overriden per-branch.	21:16
clausen	Ilari, does that make it default?	21:17
	I guess I should just try it :)	21:18
	ah, thanks	21:19
	I think many / most people would like a setup like I suggest	21:19
	it seems like it should be in an FAQ or something	21:19
jaalto	When I merge, and get conclicts -- it is possible to say "replace" from the the merge source?	21:20
	→ gitte joined	21:20
jaalto	That is, I don't want to preserve local changes, but the file woudl be pulled from the merge source	21:20
Ilari	jaalto: IIRC, :3:path/to/file is expression for SHA-1 of version of file from branch being merged.	21:20
	← Lash\| left	21:21
jaalto	What's that :3: thing=	21:21
Ilari	jaalto: Meaning look for file from index stage 3.	21:21
jaalto	ok	21:21
Ilari	jaalto: Normally 1 is ancestor, 2 is yours, 3 is theirs. But in one place (rebase), 2 and 3 seem to be swapped...	21:22
	← mutex left	21:23
jaalto	So I run (after conflict message):	21:23
	git checkout HEAD -- :3:path/to/file	21:23
Ilari	jaalto: I don't think that will work... More like 'git show :3:path/to/file >path/to/file'...	21:24
engla	how suitable is the bundle format for transferring git repositories? Wouldn't that be the "native" way to compress and store a whole repository	21:25
jaalto	Ilari: ok	21:25
Ilari	engla: It doesn't store config (including remotes) or reflogs...	21:25
	engla: But otherwise it stores the exact objects.	21:26
engla	true. It only stores all objects and the ref at the tip.	21:26
	it can be unbundled in an empty repository. Then you check it out with the funny message "warning, you seem to be on a branch yet to be born"	21:27
	→ CIA-11 joined	21:27
iabervon	You should be able to clone it, although that would leave an odd remotes config.	21:28
	(Not to say that you can clone it, but...)	21:28
clausen	this didn't work: git-remote add origin ../project.git; git-pull	21:34
	it fails:	21:35
	Warning: No merge candidate found because value of config option	21:35
	"branch.master.merge" does not match any remote branch fetched.	21:35
Ilari	clausen: 'git config branch.master.merge refs/heads/master' 'git config branch.master.remote origin' (if I recall correctly).	21:36
	→ moh_ joined	21:36
	→ segher_ joined	21:37
gitte	Or do the more correct thing: git checkout -b master origin/master	21:37
clausen	Ilari, that worked... but isn't there an easier way?!	21:37
	← moh left	21:38
Ilari	clausen: What gitte said above	21:38
clausen	why -b master?	21:41
gitte	'cause you are technically branching off a new local branch.	21:41
clausen	that would "create" a new branch ,right?	21:41
	weird	21:41
gitte	Why weird?	21:41
	I assume you did not clone that repo, right?	21:42
clausen	yes, I did	21:42
gitte	Otherwise you just overwrote "origin".	21:42
clausen	basically, I want to use git like cvs	21:42
	I want to create a central repository that I push/pull from	21:43
	it seems non-trivial to do this	21:43
	I suppose I haven't played with branches in git	21:43
PerlJam	clausen: what makes you think it's non-trivial?	21:49
	clausen: just that you don't know how?	21:49
clausen	I don't know :)	21:50
	it certainly isn't documented	21:50
Mikachu	if you clone a bare repo, then clone a non-bare repo from that, the new repo should have origin all set up to push/pull from the bare repo	21:50
	s/clone/clone to/g	21:51
clausen	Mikachu, good point, but that's still non-trivial :)	21:51
	(it's still the simplest way I know. thanks :)	21:51
Mikachu	well, you're not supposed to start from a non-bare repo with no origin i guess :)	21:51
clausen	what does "repo with no origin" mean?	21:52
Mikachu	btw, i think you would want to move all refs from refs/remotes/origin to refs/heads in the to-be-central repo before you clone from it again	21:52
	one that you just git-init	21:52
clausen	ah	21:53
	so, how else do you start?	21:53
Mikachu	i guess the other way is to git-init the bare repo first and clone from that	21:53
	feel free to stop me if i'm saying crazy things, anyone :)	21:54
gitte	clausen: I have to admit that I wanted to suggest to you to stick with cvs, since you want to use git like cvs.	21:54
	clausen: but your suggestion that it is non-trivial suggests that you have not read the tutorials...	21:55
clausen	no way	21:55
	cvs sucks	21:55
	git completley dominates it	21:55
	(eg: commitng locally is great)	21:55
gitte	So why do you want to stick with cvs _mindset_?	21:55
clausen	I haven't found a clear description of the git "mindset"	21:55
vmiklos	even after reading the user manual? :)	21:56
gitte	(Actually, you did not really... in CVS you do not checkout a project into the same directory from which you imported it...)	21:56
	← Oejet left	21:56
gitte	So whenever you clone from somewhere, the config is all set up for pulling.	21:57
	No magical git incantation needed.	21:57
clausen	vmiklos, the manual is rather uninformative about day-to-day usage of git	21:58
	vmiklos, perhaps it would be good if it told some stories :)	21:58
gitte	I guess your problem was that you actually pushed to a freshly initiated "upstream", and thought that it was set up to pull from there, too.	21:58
vmiklos	clausen have you read the "everyday with git" document, too?	21:58
iabervon	We've never gotten around to writing a script for "reverse clone", so you have to do the ~5 commands by hand.	21:58
gitte	Hey iabervon! Thanks for the builtin fetch...	21:59
vmiklos	clausen http://www.kernel.org/pub/software/scm/git/docs/everyday.html <- quite useful to start	21:59
clausen	vmiklos, yes	21:59
	gitster thanks daniel too.	21:59
iabervon	You're welcome.	21:59
clausen	vmiklos, I didn't find it very helpful	21:59
	vmiklos, I would prefer it if it were written in the style of "a day in the life of a kernel hacker" or something	22:00
gitster	Is clausen a kernel hacker?	22:00
clausen	no :)	22:01
	(I did some kernel hacking about 5 years ago, but that's ancient history)	22:01
vmiklos	parted:)	22:01
clausen	you can stalk me here: www.econ.upenn.edu/~clausen	22:01
	that too :)	22:01
vmiklos	and the pagerank wizard;)	22:01
clausen	(although, I must confess that parted isn't the most brilliant program ever...)	22:01
gitster	I am not particularly interested in what you are. I asked it because such a document exists (by Greg KH -- ask google) and I suspect it might be more useful to the kernel people than users on other projects.	22:02
clausen	I'm actually using git to write academic economics articles	22:02
gitte	clausen: so how about this: you describe what you want to do, we help you, and you write that document you wished for?	22:02
Mikachu	it's impossible to read that green link text on the gray background	22:02
gitster	I would freely admit that the documentation set we currently have are definitely geared towards majority of audiences we have picked up recently.	22:04
clausen	gitte, simplify "git-clone project project.git; rm -rf project; git-clone project.git project"	22:04
	(well, that's the immediate concern)	22:04
PerlJam	clausen: you know, you're the first person I've heard who is using git to track changes to articles (I'm doing the same myself)	22:04
gitster	The oldest part are mostly for people who hack git, and then newer parts are for people who hack their code _and_ are willing to customize git for their taste while git hasn't really matured.	22:04
clausen	I haven't used any branches, although that probably would come in handy	22:05
Mikachu	are the articles in a text format?	22:05
gitte	clausen: I think that the two "git config" lines that "git pull" suggests are not only descriptive, but they also help you _know_ what you can do.	22:05
	clausen: namely, adding other upstreams.	22:06
	PerlJam: I do that, too. Even articles I did not write myself: Even articles written in OpenOffice)	22:06
clausen	Mikachu, latex	22:06
gitte	clausen: I do not think that this is a common use case, but we could easily add a "--push" option to git-remote: "git remote add --push origin <url>"	22:09
clausen	gitte, how do you usually expect people to start a new project?	22:09
gitte	I do "git init".	22:09
	That's it.	22:10
clausen	do you have 1 computer?	22:10
	← CIA-11 left	22:10
gitte	ATM yes.	22:10
clausen	would you just do git-init if you worked from 3 different computers?	22:10
PerlJam	clausen: he'd do it at least once :)	22:11
clausen	(eg: students often work from home or school)	22:11
gitte	For backups, I initialise empty repositories, and I _do_ add them with "git remote add <nick> <url>"	22:11
clausen	PerlJam, how do you sync up?	22:11
gitte	And because I have several sites to pull from, I usually call "git pull <nick> <branch>"	22:11
	Just to be sure that I pull from the correct machine, the correct branch.	22:12
PerlJam	depends on which side I'm on. either push or pull :)	22:12
gitte	Without having to check everytime what the config says.	22:12
clausen	I'm not sure I undestand	22:13
gitte	clausen: I am paranoid.	22:13
clausen	I have 1 copy of the master (on www.econ.upenn.edu)	22:13
gitte	I have backups on three different computers.	22:13
clausen	if I screw up, it fails nicely	22:13
gitte	(Three in addition to _mine_)	22:13
PerlJam	I do wish there was a "git sync" or something that would make two repositories identical.	22:13
gitte	And I actually _work_ on them, too.	22:13
PerlJam	Maybe something like that exists and I just don't know about it though	22:13
clausen	if I forget to git-commit or git-push or git-pull, git gives me nice error messages to remind me	22:14
gitte	PerlJam: huh? Define the semantics first.	22:14
clausen	git-pull; git-push ?	22:14
Mikachu	like that git-mirror thing someone mentioned the other day?	22:14
	or cp -a? :)	22:15
clausen	gitte, so, is there a way my configuration can screw up?	22:15
gitte	clausen: so with the different computers, I say _exactly_ from where I want to pull, and which branch. Makes things much more clear in my head.	22:15
clausen	with my approach, I don't need to think	22:16
gitte	git-mirror is about having a bare repo tracking another repo.	22:16
	clausen: oh.	22:16
	clausen: what was that quote about that?	22:16
	gitte goes and searches it...	22:16
clausen	well, I'm asking if there's any way "not thinking" can screw things up :)	22:16
engla	git clone --mirror perhaps?	22:16
clausen	btw, I haven't found greg kh's git howto on google	22:17
gitte	"not thinking" always screws things up.	22:18
vmiklos	git-mirror? :)	22:18
clausen	is it this? http://cs343-spr0607.stanford.edu/index.php/Writeups:Greg_Kroah-Hartman	22:18
vmiklos	didn't you mean git remote add --mirror?	22:18
gitte	Well, I did not find the quote, but here is another one: "Human beings are not only unique in their ability to think, but also in their unwillingness to do so"	22:18
clausen	gitte, for example, I like logging in non-priviledged, because it means I can't break my machine	22:18
	gitte, it means I don't need to think as much	22:18
gitte	vmiklos: git-mirror predated "git remote add --mirror" quite some.	22:18
clausen	gitte, when I'm roote, I fear my machine, and have to think	22:19
gitte	vmiklos: and "--mirror" is not in any official release yet.	22:19
Mikachu	clausen: if you find anyone who logs in as root, hit them on the head for me	22:19
clausen	gitte, it's nice to not have to think	22:19
	→ bish0p joined	22:19
vmiklos	hm	22:19
	$ git mirror	22:19
clausen	Mikachu, my argument is, "not having to think" is a virtue	22:19
vmiklos	git: 'mirror' is not a git-command	22:19
Mikachu	btw, all files that you care about are owned by your user, the things root can destroy are fixable by inserting a live cd and installing some packages again	22:19
clausen	Mikachu, and unix's design of not being root is good for this reason (among others :)	22:19
gitte	clausen: I'm sorry, but I will always discourage people from "not thinking". _I_ think it is wrong.	22:20
	vmiklos: git-mirror is a script by pasky, IIRC	22:20
vmiklos	ah	22:20
	vmiklos googles	22:20
clausen	gitte: when you download email, do you "think" about where your mail server? your network connection? your router?	22:20
engla	clausen: if something happends you can mostly recover from it easily. Like using git reset to undo a wrong pull/merge	22:20
Mikachu	vmiklos: http://thread.gmane.org/gmane.comp.version-control.git/27726	22:21
	that's the link i saw the other day in here anyway	22:21
gitte	clausen: in any case, I think it is not really warranted to call three commands to set up your particular needs on _one_ machine "complicated"	22:21
clausen	engla, that's my experience... git is very friendly when you forget to git-push, git-pull, or git-commit	22:21
vmiklos	Mikachu thanks:)	22:21
	so this is something hot new:)	22:21
gitte	clausen: remember, if you clone from your "upstream" (which will be the case for _all_ except the first one), it is all set up by git-clone.	22:21
Mikachu	vmiklos: actually it's an old patch that never got merged :)	22:21
clausen	gitte: I think it's complicated	22:22
	→ aroben joined	22:22
vmiklos	eh:)	22:22
	vmiklos read 2007 instead of 2006	22:22
clausen	gitte, I probably wouldn't have thought of it myself	22:22
	(since it's the first thing I tried doing, I didn't have much intuition about how things worked...)	22:22
gitte	clausen: if you think that those two "git config" lines (that are suggested by the failing pull!) are complicated, I fear I cannot help you.	22:23
clausen	how are they suggested by the failing pull?	22:24
	→ CIA-11 joined	22:24
clausen	firstly, I haven't learned about branches yet	22:24
	I think it's unreasonable to expect new users to learn branches at the start	22:24
	(the git-pull failure message talks about branches)	22:25
Mikachu	branches are very fundamental when talking about pushing and pulling	22:25
clausen	AFAIK, I only have 1 branch	22:25
gitte	clausen: basically, what you did is not usually what you do as a new user of git.	22:25
clausen	so, they aren't fundamental in my case	22:25
gitte	What you do is _clone_ from somebody.	22:26
clausen	why would a new user clone from somebody?	22:26
gitte	Not set up an upstream, and then set that up to be pulled from.	22:26
clausen	are you claiming it's uncommon for a new user to start a new project?	22:26
gitte	clausen: because you do not set up new repos everyday.	22:26
	clausen: I'm not claiming that.	22:27
	clausen: I am claiming that setting up an _upstream_ is not what a new user of git does right away.	22:27
	And I really have to caution here: if you set up an upstream, but do not understand branching, you will not be happy.	22:27
clausen	I think it should be typical	22:27
	if you have many computers to work from	22:27
	(eg: laptop + desktop)	22:27
gitte	For example, you might still have the impression that you all work on the same branch.	22:28
	Well, here's some news for you: you don't.	22:28
clausen	can you give an example of some semantics I wouldn't expect?	22:28
gitte	Face it: most new git users are cloning an existing project.	22:28
	clausen: try to check out a remote branch (i.e. origin/master in your case).	22:29
	You'll be in for a surprise.	22:29
clausen	I've been using this configuration for about a month with no trouble, pushing and pulling 100s of times	22:29
gitte	You did not "git checkout origin/master".	22:30
clausen	correct	22:30
gitte	That was what _I_ was saying. You'd be surprised.	22:30
	Actually, you can try that right now (if you have no uncommitted changes)	22:30
clausen	obviously, I wouldn't try it on my working repositories...	22:31
gitte	You could.	22:31
	It's safe.	22:31
	It's git.	22:31
clausen	I trust you, but I'll do it on a toy repo anyway :p	22:31
nud	git makes it easy to fix a broken repo ;-)	22:31
	gitte used to run git-fsck --full twice a day for a while; now it is only once a week	22:32
clausen	(that way, I don't need to think!)	22:32
nud	(at least when brokenness is related to branches and heads)	22:32
Mikachu	i think it's a bit too easy to delete a tag :)	22:32
gitte	clausen: if you keep saying that you don't want to think, I'll stop... ;-)	22:33
clausen	I don't want to think about losing/repairing data :)	22:33
gitwump	clausen, out of interest, which tools did you use previously, eg did you use darcs or arch?	22:33
gitte	gitwump: heh, nice nick!	22:34
clausen	gitwump, I just kept a tarball of every version	22:34
pasky	repo 13877 0.0 0.0 3296 1052 ? SN Sep29 0:00 \| \| \_ git-peek-remote git //git.icculus.org mikachu/openbox	22:34
Mikachu	:O	22:34
	pasky wishes again that git would time out in some sane fashion	22:34
	→ rodserling joined	22:34
clausen	gitwump, I used cvs a bit, but screwed up too much	22:34
Mikachu	i think that server is currently a bit down	22:34
gitwump	ok. Interestingly, before BitKeeper, that was Linus' only VCS too :)	22:34
	gitte, thanks	22:34
pasky	sometimes these hanging gits are blocking the mirroring process for days	22:34
clausen	gitwump, in particular, my internet access is very spotty, so cvs didn't work in that case	22:35
gitwump	yeah, that's a big issue with centralised systems for those on the fringe of the interner	22:35
	-r+t	22:35
gitte	gitwump: remove "read", add "sticky"?	22:36
gitwump	it's another irc form for "s/r/t/" but applies to wherever it needs to	22:36
Mikachu	i like s/interner/internet/ since it takes less parsing to understand what you mean	22:37
mithraic	Can I have nested submodules? Documentation is still a little scarce on the ground.	22:38
Ilari	mithraic: Yes.	22:39
gitte	mithraic: Why shouldn't you?	22:39
	pasky: it'd involve a patch to xread() in git-compat-util.h, I guess...	22:39
mithraic	Well, I'd hope so, but wanted reassurance.	22:39
iabervon	Nested submodules work, but they're unweildy.	22:39
clausen	gitte, can you please tell me how git-checkout can surprise me?	22:40
pasky	gitte: yeah, probably	22:40
	← janm left	22:40
gitte	pasky: you'd have to select() with a given timeout, and in the timeout case, die().	22:40
pasky	or just alarm()	22:40
gitte	clausen: it'd make you think, I fear ;-)	22:41
clausen	gitte, that's what I'm scared of :p	22:41
	bring it on!	22:41
mithraic	and is it reasonable to create a submodule 'in-situ', rather than creating it elsewhere and then referencing it from there?	22:41
clausen	ah, I see	22:41
gitte	pasky: alarm() is not so good, methinks... you'll have to _unset_ the timer in case of success, with setitimer().	22:42
	pasky: so maybe better be explicit about it with select(), no?	22:42
iabervon	I don't think there's a script for turning a directory into a submodule.	22:43
Ilari	mithraic: Yes. Just create git repository, commit something there, and then git-add the directory.	22:43
clausen	git-checkout took me off the "master" branch?	22:43
mithraic	Great! So this makes partial checkouts pretty feasible, then!	22:43
gitte	iabervon: that's right; the beginnings of it are in filter-branch --subdirectory-filter, though.	22:44
Ilari	mithraic: Actually, that is pretty much the only submodule-related user-level operation that works in 1.5.2...	22:44
gitte	clausen: exactly, because "git checkout" is about switching branches.	22:44
	(In general.)	22:44
clausen	but, if you ignore branches, that will never happen :)	22:45
gitte	And your statement "I'm not working with multiple branches" is provably wrong... for each repo you have, you have an independent set of branches.	22:45
	So whenever you clone, you branch off.	22:45
Mikachu	that's a slightly confusing thing about git, half the commands are overloaded with unrelated functions for convenience :)	22:45
clausen	how so?	22:45
gitte	clausen: whenever you make local changes, you branch off.	22:45
	clausen: that's true for _every_ SCM; we are just honest about it.	22:46
clausen	what does git-push do then?	22:46
gitte	Mikachu: if you want simple, why not go mercurial? ducks	22:46
pasky	bah	22:46
	:)	22:46
Mikachu	i'm not saying it's bad or that i want something else, just that it's confusing :)	22:46
pasky	git should be simple	22:46
gitte	clausen: git-push updates a remote branch to the state of your local branch.	22:46
	pasky: I find it simple.	22:46
pasky	many people don't	22:47
	I don't, for one :)	22:47
gitte	pasky: but there are people who say "it should be simpler"... nevertheless, most seem to be incapable of proposing sane simpler semantics...	22:47
clausen	I suppose I should just play around with branches to learn about them	22:47
	here's a proposal:	22:47
	git-push and git-pull should have an option called --set-default that sets the default places to push/pull from	22:48
engla	git should continue to strive to be simpler	22:48
Mikachu	i think it would perhaps be a bit helpful if the "plumbing" commands were either in /usr/libexec out of the normal path, or named git-plumbing-*, or something	22:48
clausen	(origin/master ?)	22:48
iabervon	origin/master is the branch that stores the latest information you have about origin's master branch.	22:49
	← kristoffer_ left	22:49
gitster	clausen: there already is "setting the default places to push/pull from".	22:50
	Read about branch..origin and branch..merge options.	22:50
Mikachu	he doesn't want to know about branches or git-config, it seems	22:50
clausen	gitster, IMHO there should be no git-config command	22:50
gitwump	clausen, git-push and git-pull are having their operation rethought for 1.6, so now is a good time to participate in the ML discussion	22:50
engla	using --track with checkout should do that I think?	22:51
gitster	You just need to learn what people have already worked hard to prepare for you, before proposing something new and useful.	22:51
	And I do not particularly think "I'll play with branches a bit more" is useful.	22:51
gitte	clausen: there should be no git-config command? That is really not thinking... ;-)	22:52
gitster	If you have two repositories and pushing-fetching between them, then you are already playing with branches.	22:52
Mikachu	git-dwim should be the only git command	22:52
gitster	Heh, why then use git? The only command you need is "write-perfect-paper-or-code-for-me".	22:52
gitwump	with darcs, arch, bzr, they call it branches when you make a complete copy of the repo	22:53
gitster	You use the output of that command. No revision control necessary ;-)	22:53
gitwump	git-config most of the time feels more like plumbing	22:54
	not really a user command. It doesn't know what keys are valid for a start, so it doesn't really do much checking of its arguments	22:55
	which is arguably how it should be	22:55
mithraic	Ummm, permission to ask a very dumb question? Once I've done the no-no of git-pushing to a repository that has a working copy, how do i update that working copy?	22:55
gitster	If you did not have any precious local change, "git reset --hard HEAD".	22:55
gitwump	mithraic: use git diff to check you didn't have any local changes, then you can git reset --hard	22:56
mithraic	thanks kindly :)	22:56
Ilari	What about 'git set-tracked-branch' for that one case? Or it is just too rare to matter?	22:56
gitwump	well it might be an option to branch or pull	22:56
	eg git pull origin branch --set-default	22:57
Ilari	gitwump: Yes, that is probably better.	22:57
gitwump	People already think there are too many commands :)	22:57
Mikachu	more options to existing commands are fine i think :)	22:57
gitte	gitster: wouldn't "git read-tree -u -m HEAD" been better?	22:58
	← Nindi left	22:59
gitte	gitwump: FWIW the idea is not new: http://article.gmane.org/gmane.comp.version-control.git/19337	22:59
Mikachu	isn't that git-remote add?	23:00
gitster	I do not quite get that whole thing. Are you guys complaining that there is no easy way to change what a "git remote add" sets up?	23:01
clausen	why is git-checkout called that?	23:01
	wouldn't a better name be git-branch-select?	23:02
gitster	You "checkout a branch".	23:02
	Learn the lingo of the land.	23:02
Mikachu	you can checkout things that are not branches too	23:02
clausen	git-branch-checkout?	23:02
	ah	23:02
	thanks	23:02
gitster	And it is not unique to git -- cvs checkout, svn checkout, etc.	23:03
engla	cogito had "switch" (?)	23:03
	sounds good to switch branch. I do that a lot	23:03
gitster	That's the oddball and it died because of its "uniqueness" ;-)	23:03
clausen	gitster, "cvs checkout" is more like "git pull"	23:04
Mikachu	switching branches is switch in svn too iirc, checkout means to clone branch tip and check it out :)	23:04
gitster	clausen: I suspect you are thinking about "cvs update".	23:04
clausen	gitster, oops. I meant "git clone" I suppose	23:04
gitte	clausen: that git-config thing you hate so much helps you here: git config --global alias.branch-select checkout	23:04
gitster	clausen: Ah, Ok, it is more like "git clone".	23:04
	gitte: but then you can also "branch-select a file out of a different commit".	23:05
Mikachu	even if it's not the same as the checkout from cvs and svn i think it's an appropriate name for what it does	23:05
gitster	We usually say "checkout that file from this commit".	23:05
	So try to resist the urge of "improving" things before you understand how the world currently works. It will not help you.	23:06
gittish	the inconsitency there is that if I leave out the path it changes HEAD	23:06
gitster	gittish: that's not "inconsistency" -- it is just "use of overloaded word".	23:06
	You either "checkout the branch", or "checkout the path out of the branch".	23:06
	The verb checkout is working on different objective noun in these two sentences.	23:07
gittish	the only difference is an extra parameter	23:07
	normally that would mean "optional"	23:07
gitster	The only difference is "the path out of the".	23:07
clausen	gitster, I think trying to improve things is a good way to figure out how the works	23:08
gittish	I know what it means, and you know what I mean. I still think it is confusing.	23:08
mithraic	Okay, so with nested submodules, it looks like I need to manually recurse and do a git-submodules init && git-submodules-update.	23:09
gitster	That's because the word "checkout" in human language is already overloaded. We could call the former operation "git frotz" and the latter operation "git nitfol" if that would make you feel better ;-)	23:09
gittish	:)	23:09
iabervon	I think it's a bit odd that you can't change branches using "git branch".	23:10
mithraic	Yeah, that kind of pulls against intuition.	23:11
gitster	Yeah, and it is very odd that you have to say "git branch -d"; more natural "git chainsaw" does not work.	23:11
iabervon	Well, everything else you might want to do with branches can be done with "git branch", so it's a logical place to look.	23:12
gitster	well if we do not have "git checkout" and have you "git branch --switch", we would please only people whose first (and only) SCM is git.	23:12
clausen	does git-push only push the selected branch?	23:13
iabervon	(and the "git chainsaw" alias is easy to make, while connecting an existing function to an existing program)	23:13
gitster	Doesn't git-push(1) describe that enough?	23:13
	If you tell it what to push, it pushes them and nothing else.	23:13
Mikachu	iabervon: git-branch doesn't do anything else with the work tree so i think it would be strange if it had one subcommand that did that	23:14
	it only juggles around thnigs in .git/refs	23:14
gitster	If you don't there are set of rules to choose the default things to be pushed.	23:14
iabervon	FWIW, users of other SCMs are too afraid of branches to want the feature at all.	23:14
clausen	gitster, no. git-push(1) just says "update remote refs using local refs"	23:14
	gitster, it doesn't say which local refs	23:14
	(you can have "refs" -- whatever they are -- on many branches, right?)	23:14
Mikachu	surely it mentions almost at the top "Note: If no explicit refspec is found, then all the heads that exist both on the local side and on the remote side are updated."	23:15
gittish	a branch is a ref	23:15
gitster	I did not ask "doesn't the first line of git-push(1) tell everything?". Obviously there is not enough room.	23:15
Mikachu	maybe scrolling down counts as thinking	23:16
	gitster goes back to 1.5.3.3	23:16
gitte	gitster: one question before you leave...	23:16
clausen	gitster, the DESCRIPTION part of git-push(1) is literally 3 sentences	23:16
Mikachu	back to the future?	23:16
gittish	"description" fits on the first page, unless you are running on an old cash register or something	23:16
gitte	gitster: what did you mean by "gitlink" feature?	23:17
	gitster: is that the core functionality behind submodules?	23:17
gitster	bind html branch out of git.git as a subproject to WinGit.	23:17
clausen	(isn't it common for DESCRIPTION to be several pages?)	23:17
gitte	gitster: ah, but do I not need a symlink then?	23:17
	gitster: forget what I just said. alternates are enough.	23:18
Mikachu	clausen: if you wonder what the command does when the optional paremeter 'refspec' isn't given, isn't it natural to scroll down to the description of that option and see what it says?	23:18
gitte	git-dont-make-me-think? ;-)	23:18
gitster	I am actually not _actively_ suggesting to use the subproject for WinGit at this stage, yet.	23:18
iabervon	clausen: only if you don't have OPTIONS as a separate section.	23:18
	→ edalytical joined	23:18
clausen	ah, fair enough	23:19
gitte	gitster: and I maintain that it would be actively wrong to have a subproject for WinGit.	23:19
	gitster: but having the html branch checked out as a submodule makes sense.	23:19
gitster	the difference between the two escapes me, sorry.	23:19
gitte	gitster: WinGit is just the installer (without dev env)	23:20
	→ dash__ joined	23:20
gitster	Can somebody add "git push" without any parameters to the Examples section of git-push.txt please?	23:20
gitte	gitster: so ATM we have everything needed for it in /share/WinGit/	23:20
Mikachu	but Examples is at the bottom, surely nobody scrolls down that far ;)	23:21
gitster	gitte: ok, then I probably meant mingwgit or gitme or whatever it is called these days.	23:21
gitte	gitster: I'm sorry... lemme explain in one sentence:	23:21
edalytical	hi everyone, does "git status" respect the configuration variable core.excludesfile?	23:21
clausen	what's the difference between git-push and git-push --all?	23:22
gitte	gitster: msysgit.git contains the complete dev env, with "git" as a submodule. The scripts to make the installers are in /share/WinGit/, /share/msysGit/ and /share/GitMe/.	23:22
gitster	Mikachu: quite the contrary. Lazy and overconfident people including me are known to skip the description section and go right to the Examples section, expecting that it is enough to learn from examples.	23:22
clausen	oops	23:22
Mikachu	clausen: presumably it pushes even refs that aren't already in the remote	23:22
clausen	I misread	23:22
	thanks	23:22
gitte	Mikachu: I agree with gitster here; I often only read the examples section.	23:22
gitster	So I would not blame people not reading descriptions section and wonder why things do not work as they expect. I do that to other people's software.	23:23
kreaturr	is there a way to pick merge or cherry pick between branches - but just for one file?	23:24
gitster	But descriptions and options sections are exactly for these people, when the law of the land is _NOT_ what they thought was, to find it out.	23:24
	kreaturr: if you are doing "just for one file", that is not a merge nor a cherry pick.	23:24
	Having said that, "git diff ...that-branch -- this-file \| git apply --index" would probably be what you are looking for, to grab the change that-branch made to this-file and apply it to your current branch.	23:26
clausen	is this synposis correct?	23:26
Mikachu	kreaturr: you could cherry-pick -n and git-add only the file, then reset --hard	23:26
clausen	SYNOPSIS	23:26
	git-pull <options> <repository> <refspec>...	23:26
	shouldn't it be:	23:26
	SYNOPSIS	23:26
	git-pull [<options>] [<repository>] <refspec>...	23:26
Mikachu	possibly gitster's command is more foolproof	23:26
gitte	kreaturr: you can write the contents in the different revisions (branch1, branch2, merge base) into files and use git merge-file.	23:26
iabervon	clausen: they're actually all optional, and repository is required if you have a refspec.	23:27
kreaturr	well, "one file" is an example for a limited subset. thanks for the suggestions, those should work! :)	23:27
clausen	git-pull [<options>] [<repository> <refspec>...] ??	23:27
gitte	So git-pull [<options>] [<repository> [<refspec>...]]	23:27
Mikachu	i think if there is only one <> inside a [], just [options] is the standard notation	23:27
gitster	gitte: yeah, except everybody understands ... is 0 or more so [] around it is unnecessary.	23:28
gitte	kreaturr: I like Mikachu's suggestion better than mine; please ignore mine.	23:28
Mikachu	if we are being pedantic	23:28
	gitte is not pedantic and goes back to real work...	23:28
iabervon	I thought ... was 1 or more, in general.	23:29
Mikachu	git-pull [options] [<repository <refspec>...] then?	23:30
	although i guess that is incorrect if you can say just git-pull remote-foo	23:30
gitster	I think that was what clausen suggested. To avoid 1 or more vs 0 or more confusion, it would not hurt to have [] around refspec...	23:30
Mikachu	the only input i really have is to not write [<foo>] :)	23:31
	← dash_ left	23:32
Mikachu	gitster: speaking of pedantic, did you miss the mail i sent about the type in the git-svnimport manpage? http://git.mikachu.ath.cx/?p=git.git;a=commitdiff;h=typo :)	23:33
	actually... that should say "higher than the" i think	23:34
	and specifies	23:34
	→ Fullmoon joined	23:34
Mikachu	the typo stands out so much i didn't even read the sentence before :P	23:34
	i guess i can proof read the whole page just as well	23:35
	would you say "an SVN repository" instead of "a SVN repository"?	23:35
	← engla left	23:36
Mikachu	are double spaces ignored in asciidoc?	23:36
edalytical	I figured it out, core.excludesfile can't be set to an arbitrary file it has to be ~/.gitignore	23:37
	is this a bug?	23:38
iabervon	Are you sure you aren't overriding it with a different config file?	23:39
	← dooglus left	23:41
gittish	does the whitespace=strip option to apply strip all trailing whitespace or just the ones introduced by the patch?	23:41
	ah, it says there..	23:42
clausen	does git-clone copy all branches?	23:42
gitte	edalytical: I don't believe that.	23:42
	← elmex_ left	23:42
gittish	core.excludesfile can be any file	23:43
edalytical	I had it pointing at a different file and my where still shown in 'git status'	23:43
	changed it to ~/.gitignore and it worked as expected	23:44
Mikachu	gitster: okay i've updated it with some more fixes now, so it's less trivial	23:44
clausen	(it didn't copy all of my branches when I tried it)	23:44
gittish	edalytical: I have it set to another file in a number of repos	23:45
Mikachu	clausen: it does, but it puts them in remotes/origin/* instead of refs/heads/*	23:45
	clausen: see git branch -a	23:45
edalytical	ok, i'll check and make sure I didn't make a mistake	23:45
iabervon	edalytical: If you have it as something else, and you do "git config --list", does it look right?	23:45
edalytical	iabervon: it looks right and it also works	23:49
iabervon	Oh, okay, then.	23:50
edalytical	I must of had a typo in the path	23:50
Mikachu	that always happens to me, i make some silly mistake and then declare there's a bug :)	23:50
edalytical	should an error be reported if the path doesn't exist?	23:51
gitte	edalytical: is an error reported if .git/info/excludes does not exist?	23:58
Mikachu	it's possibly not quite the same	23:58
gitte	why?	23:59
	Remember, you can set it in ~/.gitconfig	23:59
iabervon	One would assume that, if you specify a non-empty filename, it's because you intend to use it.	23:59
gitte	Again: you can set it in ~/.gitconfig	23:59

Logs Search ←Prev date Next date→

Channels Documentation