That is done by receive-pack; it chdir()s into the repository and does its thing, and the hooks are called from there; I'd expect cwd to be the repository ('.git'), GIT_DIR to be dot ('.'). I think doing the "unset GIT_DIR" to be the first thing if you want to access some other repository is documented somewhere but if not please send a patch to document it. As to file descriptors, I think duping the output to original stderr might make sense, but I do not know what breaks, so interested ...

And Junio C Hamano writes: - Perhaps printf '%s' "$blah" to futureproof, instead of just - saying printf "$blah"? Definitely more likely to stay correct. I'm just lazy. - <rant mode on> - While I do not have anything aginst system without "sane" echo, - I really do not like it. Not your solution, but *having to do* - something like that. Um, you think the wrong one is sane. ;) Why should echo in Bourne shell take C-shell arguments? (I blame AT&T for their crappy ...

On AIX, there is no -n option to the system's echo. Instead, it needs the '\c' control character. We could replace echo -n "foo" with echo -e "foo\c" but printf is recommended by most man pages. Tested on AIX 5.3, Solaris 8, and Debian. Signed-off-by: E. Jason Riedy <ejr@cs.berkeley.edu> --- git-am.sh | 2 +- git-applypatch.sh | 2 +- git-bisect.sh | 2 +- git-status.sh | 4 ++-- 4 files changed, 5 insertions(+), 5 ...

True but the original convert-objects.c defines _XOPEN_SOURCE on all platforms not just AIX. -

I agree with your approach of least impact for now. Thanks. -

Here's a list of things that are missing, wrong, or badly formatted in the documentation (all checked against the man pages): git-checkout-index: SYNOPSIS: --index --quiet --force --all --no-create (How do we put these in the SYNOPSIS?) (My bad, forgot to mention this in an earlier patch.) git-read-tree: --trivial SYNOPSIS: ...

Then something like this? --- diff --git a/pack-objects.c b/pack-objects.c index a62c9f8..c27fee5 100644 --- a/pack-objects.c +++ b/pack-objects.c @@ -296,7 +296,7 @@ static int type_size_sort(const struct o return -1; if (a->size > b->size) return 1; - return a < b ? -1 : (a > b); + return memcmp(a->sha1, b->sha1, 20); } struct unpacked { -

It's perfectly correct. If the same list was to be passed to create_sorted_list() twice it will come out exactly the same the second time as it did the first. The only thing to remark on is that the return above could be written as below instead: return a - b; -- Andreas Ericsson andreas.ericsson@op5.se OP5 AB www.op5.se Tel: +46 8-230225 Fax: +46 8-230231 -

Ah, right you are. Mildly relatedly, notice that a<b can be valid when a-b<0 is not. ptrdiff_t is not required to be big enough to hold the difference! (A highly dubious part of the C standard.) M. -

Dear diary, on Tue, Dec 06, 2005 at 10:19:18PM CET, I got a letter Quite on the contrary, your GIT is too old - at least 0.99.9 is required as noted in the INSTALL file. -- Petr "Pasky" Baudis Stuff: http://pasky.or.cz/ VI has two modes: the one in which it beeps and the one in which it doesn't. -

Oh wow. Exactly correct. Here's Jon building up the new Ubuntu box, and thinking, "I'll be using git, so I don't need to install RCS or CVS anymore." :-) After installing RCS, here's what we get: jdl@ubuntu:/usr/src/git-core$ git reset --hard jdl jdl@ubuntu:/usr/src/git-core$ git pull . origin Trying really trivial in-index merge... fatal: Merge requires file-level merging Nope. Merging HEAD with be61db922a230ae2638c27c071ee4b8c98f01f72 Merging: ...

So, I was surprised by this display this morning. (I swear I'm trying to help! :-) I grabbed new (7:30 CST Dec 6) top-of-git, built and installed it. I then "git checkout jdl" and "git pull . origin" to bring my branch up to date. It looked like this just before the pull into branch "jdl": jdl@ubuntu:/usr/src/git-core$ git show-branch jdl master origin * [jdl] Added documentation for few missing options. ! [master] git-merge-one-file: resurrect leading path creation. ! [origin] ...

Duly noted. Tim. "I've worked myself up from nothing to a state of supreme poverty." -- Groucho Marx -

I tried build-testing the latest version of git, and the tests reported this error: --------- *** t6021-merge-criss-cross.sh *** * FAIL 1: prepare repository echo "1 2 3 4 5 6 7 8 9" > file && git add file && git commit -m "Initial commit" file && git branch A && git branch B && git checkout A && echo "1 2 3 4 5 6 ...

Note to $self -- here are some tidbits ToDo: o --help doesn't work on: git-hash-object git-init-db git-write-tree o There is an odd <any-file-on-the-filesystem> argment named in the git-hash-object documentation. Would <file> be better and more consistent? Or is there some subtlety that I am missing? o Someone might want to check my --template=<dir> description on git-init-db, or make it clearer. o Work towards moving uniform description of all ...

More $ shell prompts in examples. Minor English grammar improvements. Added a few "See Also"s. Use back-ticks on more command examples. Signed-off-by: Jon Loeliger <jdl@freescale.com> --- Documentation/git-checkout-index.txt | 78 +++++++++++++++++++++------------- Documentation/git-init-db.txt | 27 ++++++++++-- Documentation/git-prune-packed.txt | 11 +++-- Documentation/git-read-tree.txt | 74 ++++++++++++++++++-------------- Documentation/git-update-index.txt ...

Are you suggesting it to be done like this? --- git diff diff --git a/object.c b/object.c index 427e14c..dac5c92 100644 --- a/object.c +++ b/object.c @@ -82,7 +82,12 @@ static int compare_object_pointers(const { const struct object * const *pa = a; const struct object * const *pb = b; - return *pa - *pb; + if (*pa == *pb) + return 0; + else if (*pa < *pb) + return -1; + else + return 1; } void set_object_refs(struct object *obj, struct object_refs *refs) -

And Junio C Hamano writes: - I'd like to throw in another Makefile patch to catch both 5.8 and - 5.9 for this, but would appreciate if people with various vintage - of Solaris boxes can give some inputs before doing that. Anyone else seeing the qsort in object.c trash memory? The input _looks_ correct. I only see this on Solaris (8), but I've never had a problem with qsort there. I think we have a Solaris 9 box around here somewhere, but probably only partially 9. The admins here ...

> Are you suggesting it to be done like this? Precisely. Morten -

Using the git-* commands means that tab completion works, which is good for impatient people with bad memories, like myself :) Mike -

The specific git-diff-XXX command example was longer. In general, there is not much difference, except that I was admonished to avoid the git-XXX in my howto. I suppose that makes sense if the git-XXX programs are moved out of $prefix/bin, leaving only $prefix/bin/git. Jeff -

When you clone you need to get all the objects. If the objects aren't packed on the remote end (which is usually the case), the git protocol will pack them for you which takes quite some time for a large repo and puts the server under considerable stress. When you pull incrementally the native git protocol does some figuring In general, rsync:// should be the fastest protocol to clone over. Why you got the numbers listed above I have no idea, but I'm guessing the rsync daemon ...

How does e.g. git-diff differ from git diff? I thought they were equivalent.--b. -

Has that been the general git user experience? While there are certainly disadvantages to the http transport, I didn't think that performance was one of them. For comparison, I ran some clones this morning/afternoon with the following results: Clone of kernel.org/pub/scm/git/git.git http real 0m24.204s real 0m25.750s real 0m24.094s user 0m5.870s user 0m5.530s user 0m5.350s sys 0m0.660s sys 0m0.710s sys ...

The (latest as of now) code for cogito is in cogito's contrib/ directory. You can get stg,cogito,git,gitk completion code from my repo with: cg clone http://www.hawaga.org.uk/gitcompletion.git (I posted a note about it the other day on this git list, entitled 'bash completions code for git+porcelain) -- Ben • ベン • Бэн • 벤 • 班明 http://www.hawaga.org.uk/ben/ My email is high latency but best way to contact me. Alternatively, SMS number(s) at above URL. -

Ah, that is what you meant. I agree. -

Thanks. I've had a chance to try the one in the "master" branch on cygwin myself today and it seemed to work OK. I did a partial build on Solaris 5.9 as well. -

I've considered this one after hearing what you meant by your previous comment, but I am slightly in favor of keeping this die(); what would trigger this is a programmer error, and the extra error message from die() makes it obvious, although I agree that it is unclean from purist point of view. This one I'd agree with 100%. -

By the way, by "fun" I do not mean "joke" at all. You earlier said such newsgroups cannot be distributed, but I would say why not. They will be moderated groups and patches will go to the "moderator" who applies the patches to the branches. Reading such a newsgroup would be like going through "commitdiff" link on gitweb one commit after another, but is more efficient and would match the workflow of people who review lots of patches on mailing lists. -

And then, it could be taught to apply posted patches, which could be even more fun.. :-) -

"Ask and it shall be given you" ;-) In case it was not obvious, I was talking about an NNTP server, not a client. You would probably have to keep a mapping between sequence number and commit object name in each "newsgroup", which is the only nontrivial part. -

Yes, I see that for day-to-day development the gitk view is optimal. I'll see if I can do something similar, but web-based. Artem. -

And the reason why special NNTP server is needed is, I guess, that plain NNTP server will not be able to cope with all the diffs stored in message database? Well, but then, the newsgroups will not be distributable.. Or is there any other reason? -