Re: Unintended commit?

[David Miller <davem@davemloft.net>]

From: Linus Torvalds <torvalds@osdl.org>
Date: Mon, 23 Oct 2006 10:13:52 -0700 (PDT)

> Indeed. Thanks for noticing.
> 
> David, can you pls explain?

Totally my mistake.

A PCI change a few days ago blows up Sparc64 so I kept putting in that
drivers/pci/quirks.c workaround in order for me to be able to test the
current tree.

I accidently commited it, again sorry.  Greg KH has a patch coming to
you soon which will move that VGA code back into x86/x86_64/IA64
specific areas and will fix the sparc64 problem properly.

I should make some modifications to my workflow so that I don't do
this again when I need to make local unrelated changes in order to
test the current tree.

When I'm grinding out a patch actively in my tree I go "git diff >diff"
and then I use a script called "mkcf" which runs diffstat on the
diff and gives me a file list, it's ugly, but here it is:

#!/bin/sh
diffstat -p 1 $* | grep -v changed | awk ' { print $1 } '

So I end up going:

1) edit files
2) git diff >diff
3) read diff
4) git commit $(mkcf diff)

Usually step #3 catches local changes I'm not intending to commit
and I just edit those out, and therefore they never end up being
committed.

Perhaps it would be cool if you could tell GIT "Look, I know I have
a change to foo.c, but it's a local hack and please act like it's
not there unless I try to do an operation where ignoring that change
is impossible, such as a merge."

The idea is that things like "git diff", "git update-index" etc.
would not show the change, but things like a "git pull" that would
reference the file with such changes would conflict.