why git-reset needed after "cp -a" of a git repo?

why git-reset needed after "cp -a" of a git repo?

[Erez Zadok]

Apologies if this had been discussed before: I wasn't able to find this in
the ML archives.

Sometimes I copy a whole git archive using cp -a, for experimental reasons,
or otherwise; sometimes I rsync several git repos between remote and local
computers to make access faster (it's often faster to rsync two git repos
than to re-clone or deal with merge conflicts).

However, I noticed that after I copy a git repo (using v1.5.2.2), the index
entries are all out of sync, and I need to run git-reset.  Why?  What's in
the index file that changes after a cp -a or rsync that git depends on?  Is
it atime's and if so, aren't they copied by cp -a or rsync?  If it depends
on atime's, what happens if I mount my filesystem with noatime?  Or does
git's index depends on inode numbers which change after a cp -a?  (BTW, I
tried a variety of rsync options and none helped.)  I also briefly looked at
the source code and wasn't able to find the answer.

So, is there a way to efficiently copy a git repo on a local or remote host
w/o having to rerun git-reset afterwards?

Thanks,
Erez.

Re: why git-reset needed after "cp -a" of a git repo?

[Jon Loeliger]

On Wed, 2007-08-22 at 13:57, Erez Zadok wrote:

> However, I noticed that after I copy a git repo (using v1.5.2.2), the index
> entries are all out of sync, and I need to run git-reset.  Why?  What's in
> the index file that changes after a cp -a or rsync that git depends on?  Is
> it atime's and if so, aren't they copied by cp -a or rsync?  If it depends
> on atime's, what happens if I mount my filesystem with noatime?  Or does
> git's index depends on inode numbers which change after a cp -a?  (BTW, I
> tried a variety of rsync options and none helped.)  I also briefly looked at
> the source code and wasn't able to find the answer.
> 
> So, is there a way to efficiently copy a git repo on a local or remote host
> w/o having to rerun git-reset afterwards?


Use "git update-index --refresh" to update time stamps in the index.

jdl

Re: why git-reset needed after "cp -a" of a git repo?

[Linus Torvalds]

On Wed, 22 Aug 2007, Erez Zadok wrote:
> 
> However, I noticed that after I copy a git repo (using v1.5.2.2), the index
> entries are all out of sync, and I need to run git-reset.  Why?  What's in
> the index file that changes after a cp -a or rsync that git depends on?  Is
> it atime's and if so, aren't they copied by cp -a or rsync?

ctime/mtime and inode numbers too.

If you use hardlinks to copy the working tree, *and* you reset ctime 
afterwards, you'd be ok. But basically, git tries to be *really* anal in 
noticing any possible change to the inode, so anything it can do to notice 
that the index file might be stale, it does.

But you don't need to do a "git reset", you're actually better off just 
doing a "git status" instead. That will refresh the index.

		Linus

Re: why git-reset needed after "cp -a" of a git repo?

[Erez Zadok]

In message <alpine.LFD.0.999.0708221208090.30176@woody.linux-foundation.org>, Linus Torvalds writes:
> 
> 
> On Wed, 22 Aug 2007, Erez Zadok wrote:
> > 
> > However, I noticed that after I copy a git repo (using v1.5.2.2), the index
> > entries are all out of sync, and I need to run git-reset.  Why?  What's in
> > the index file that changes after a cp -a or rsync that git depends on?  Is
> > it atime's and if so, aren't they copied by cp -a or rsync?
> 
> ctime/mtime and inode numbers too.
> 
> If you use hardlinks to copy the working tree, *and* you reset ctime 
> afterwards, you'd be ok. But basically, git tries to be *really* anal in 
> noticing any possible change to the inode, so anything it can do to notice 
> that the index file might be stale, it does.
> 
> But you don't need to do a "git reset", you're actually better off just 
> doing a "git status" instead. That will refresh the index.
> 
> 		Linus

Thanks for the info and tips.  It's a good idea of course to detect any
possible changes, but I wonder if for those of us who know what they're
doing (i.e., living on the edge :-), there could be an option to ignore
inode numbers and just depend on good 'ol ctime/mtime (as other tools like
make do).

If such an option might be deemed useful, I'm willing to take a crack at it.

Erez.

Re: why git-reset needed after "cp -a" of a git repo?

[J. Bruce Fields]

On Wed, Aug 22, 2007 at 03:19:23PM -0400, Erez Zadok wrote:
> Thanks for the info and tips.  It's a good idea of course to detect any
> possible changes, but I wonder if for those of us who know what they're
> doing (i.e., living on the edge :-), there could be an option to ignore
> inode numbers and just depend on good 'ol ctime/mtime (as other tools like
> make do).

I don't think there's any need for that.  You can always just run
git-update-index --refresh or just git-status to refresh the stat data
that's kept in the index.  That miscellaneous stat data (ctime, mtime,
inode number) is just there as an optimization, so git-diff doesn't have
to read every file.  It can be tossed out and regenerated any time.

--b.