'brew update' fails the second time

[xose]

'brew update' seems to fail the second time it is run on a new homebrew installation.

Factory install of Lion and Xcode 4.1:

$ /usr/bin/ruby -e "$(curl -fsSL https://raw.github.com/gist/323731)"
==> This script will install:
/usr/local/bin/brew
/usr/local/Library/Formula/...
/usr/local/Library/Homebrew/...

Press enter to continue
==> /usr/bin/sudo /bin/mkdir /usr/local
==> /usr/bin/sudo /bin/chmod o+w /usr/local
==> Downloading and Installing Homebrew...
==> /usr/bin/sudo /bin/chmod o-w /usr/local
==> Installation successful!
Now type: brew help

$ brew update
Initialized empty Git repository in /usr/local/.git/
remote: Counting objects: 40879, done.
remote: Compressing objects: 100% (18250/18250), done.
remote: Total 40879 (delta 24812), reused 36251 (delta 22037)
Receiving objects: 100% (40879/40879), 5.65 MiB | 971 KiB/s, done.
Resolving deltas: 100% (24812/24812), done.
From http://github.com/mxcl/homebrew
 * branch            HEAD       -> FETCH_HEAD
From http://github.com/mxcl/homebrew
 * branch            master     -> FETCH_HEAD
Already up-to-date.

$ brew update
error: The following untracked working tree files would be overwritten by checkout:

[... very long list of ALL homebrew files ...]

Aborting
Error: Failure while executing: git checkout -q master 

This seems to happen only on new installs, 'brew update' still works fine on my other ones.

I fixed it by running 'git reset' in /usr/local.

[jacknagel]

Dang, I thought this was finally fixed in cbcd4a3.

...I will investigate.

[jacknagel]

I guess cbcd4a3 probably needs to be reverted, though now I don't understand what was causing the earlier (unstaged local changes) bug.

[jacknagel]

There's a reverting commit here: jacknagel/homebrew@27dd197

However it doesn't really fix anything in that it will only reintroduce the old behavior, which was also broken.

I don't have anything better to do, so I will continue to work on this in my free time, but considering how confident I was that the '--soft' patch would work it would be good to have some fresh eyes take a look at this, so if anyone sees what the real issue is here, please chime in and suggest a fix.

ping @mxcl