When you
* are on a branch "foo" that is not "bar",
* have "origin/bar",
* and already have a local branch "bar",
"git checkout -t origin/bar" seems to misbehave.
$ git clone -s git.junio victim-002
$ cd victim-002
$ git branch
* master
$ git checkout -t origin/next
Branch next set up to track remote branch refs/remotes/origin/next.
Switched to a new branch "next"
$ git checkout -t origin/master
fatal: A branch named 'master' already exists.
$ git branch
master
* next
$ git diff --cached --shortstat
60 files changed, 2378 insertions(+), 3412 deletions(-)
$ git diff --cached master
$ exit
The first "checkout -t" is fine. The failed one seems to have already
updated the index and the work tree when it notices that it cannot create
a new branch.
I suspect "-t" does not have to be in effect to trigger this; in other
words, "git checkout -b master origin/master" would have the same issue.
I'm reporting this before digging it further myself, because I may not be
able to diagnose this before I leave for a vacation.
On Sun, 21 Sep 2008, Junio C Hamano wrote:
When you
* are on a branch "foo" that is not "bar",
* have "origin/bar",
* and already have a local branch "bar",
"git checkout -t origin/bar" seems to misbehave.
It's always been the case if the thing that fails is changing the ref for
HEAD, you're left with the index and working tree changed but the ref
unchanged. OTOH, the C conversion lost the early checks that a new branch
name is plausible in advance of trying anything.
commit c36a025b20ac752f8960bc36dcbab98ca1824657
Author: Daniel Barkalow [off-list ref]
Date: Sun Sep 21 14:25:31 2008 -0400
Check early that a new branch is new and valid
If you fail to update refs to change branches in checkout, your index
and working tree are left already updated. We don't have an easy way
to undo this, but at least we can check things that would make the
creation of a new branch fail. These checks were in the shell version,
and were lost in the C conversion.
The messages are from the shell version, and should probably be made nicer.
Signed-off-by: Daniel Barkalow [off-list ref]
diff --git a/builtin-checkout.c b/builtin-checkout.c
index 9377a1c..4497b70 100644
--- a/builtin-checkout.c
+++ b/builtin-checkout.c
@@ -580,6 +580,18 @@ no_reference:
return checkout_paths(source_tree, pathspec);
}
+ if (opts.new_branch) {
+ struct strbuf buf;
+ strbuf_init(&buf, 0);
+ strbuf_addstr(&buf, "refs/heads/");
+ strbuf_addstr(&buf, opts.new_branch);
+ if (!get_sha1(buf.buf, rev))
+ die("git checkout: branch %s already exists", opts.new_branch);
+ if (check_ref_format(buf.buf))
+ die("git checkout: we do not like '%s' as a branch name.", opts.new_branch);
+ strbuf_release(&buf);
+ }
+
if (new.name && !new.commit) {
die("Cannot switch branch to a non-commit.");
}Junio C Hamano, Sun, Sep 21, 2008 10:23:00 +0200:
When you
* are on a branch "foo" that is not "bar",
* have "origin/bar",
* and already have a local branch "bar",
"git checkout -t origin/bar" seems to misbehave.
$ git clone -s git.junio victim-002
$ cd victim-002
$ git branch
* master
$ git checkout -t origin/next
Branch next set up to track remote branch refs/remotes/origin/next.
Switched to a new branch "next"
$ git checkout -t origin/master
fatal: A branch named 'master' already exists.
$ git branch
master
* next
$ git diff --cached --shortstat
60 files changed, 2378 insertions(+), 3412 deletions(-)
$ git diff --cached master
$ exit
The first "checkout -t" is fine. The failed one seems to have already
updated the index and the work tree when it notices that it cannot create
a new branch.
Precisely this (branch already exists) case is easy to handle with a
resolve_ref in builtin-checkout.c:switch_branches.
The other errors will still leave index and working tree in this
state: branch.c:create_branch does not cleanup in case of errors,
it just dies.