Re: [PATCH v2] filter-branch -d: Export GIT_DIR earlier

* Johannes Schindelin [off-list ref] [09-02-17 16:08]:

On Tue, 17 Feb 2009, Lars Noschinski wrote:

quoted

The improved error handling catches a bug in filter-branch when using
-d pointing to a path outside any git repository:

$ mkdir foo
$ cd foo
$ git init
$ touch bar
$ git add bar
$ git commit -m bar
$ cd ..
$ git clone --bare foo
$ cd foo.git
$ git filter-branch -d /tmp/filter master
fatal: Not a git repository (or any of the parent directories): .git

This could be written as

$ cd .git
$ git filter-branch -d /tmp/bla master

Does not work, as we get another (slightly misleading) error message:

/tmp/foo/.git$ git filter-branch -d /tmp/bar master
fatal: This operation must be run in a work tree
Cannot rewrite branch(es) with a dirty working directory.

But we do not need a bare repository at all to demonstrate this bug, so
we can skip even the 'cd .git'.

Funny, git am -3 reports:

Did you hand edit your patch?
It does not apply to blobs recorded in its index.
Cannot fall back to three-way merge.

Hm, for some reason, format=flowed was enabled. I wonder, that it has
not bitten me earlier.

We try to avoid cd'ing back, by using constructs like this:

(cd drepo &&
 ...
) &&

Ok, can do.

After those two (maybe three) changes and your SOB: ACK.

BTW the reason I wanted to test this thing is that I suspected that you 
meant test_commit instead of make_commit.  But then, I realized that there 
exists a make_commit in t7003... which shares the shortcoming of our 
previous implementation of test_commit in that it adds ambiguities on 
case-insensitive filesystems.

Yeah, I used make_commit to stay consistent with the rest of the file.
I'll change it to test_commit. I think as it does not bite us, it would
be unnecessary code churn to remove the remaining usage of make_commit?

So I _had_ to look who introduced make_commit:

$ git blame -L '/make_commit/,/}/' t/t7003*

Making a fool out of yourself -- priceless.

:)

  - Lars.