Re: [PATCH] teach update-index --refresh about --data-unchanged

[Pete Wyckoff <pw@padd.com>]

pw@padd.com wrote on Wed, 03 Nov 2010 18:02 -0400:
> jrnieder@gmail.com wrote on Wed, 03 Nov 2010 13:36 -0500:
> > On Wed, Nov 03, 2010 at 10:37:40AM -0700, Junio C Hamano wrote:
> > > Pete Wyckoff <pw@padd.com> writes:
> > 
> > >> When a repository has been copied with rsync, or cloned using
> > >> a volume manager, the index can be incorrect even though the
> > >> data is unchanged.  This new flag tells update-index --refresh
> > >> that it is not necessary to reread the data contents.
> > >
> > > I know our traditional attitude towards the plumbing commands have been
> > > "give them long enough rope and let users hang themselves", but this
> > > particular rope feels a bit too long for my taste.
> > 
> > Pete, I think you mentioned the possibility of a special-case tool for
> > contrib/ that just updates the inode, device number, and ctime fields?
> > That sounds a little less worrying to use, as plumbing.
> 
> Yes, I'd been discussing with Jonathan offlist that I too agree
> this is a bit bizarre for general use.  Especially when thinking
> about how to explain the new flag in the manpage, and how
> interactions with other command-line options should work.
> 
> I wrote a short C program to use existing functions from cache.h
> to read and write the index, updating the entries by hand.  Once
> I figure out how to build it nicely, I'll submit for contrib/.  I
> haven't figured out how to include the useful bits of the
> top-level Makefile (like SSL setting, which SHA1, -lpthread,
> etc.).

Turns out this was dead simple with dulwich, a python interface
to git internals:  http://samba.org/~jelmer/dulwich .  I got rid
of the C version.

The script uses dulwich to generate a list of index entries,
stats each file and updates the entries, then writes it back
out.  I won't submit it to contrib, since it is so trivial and
because it depends on dulwich.

		-- Pete