Switching to the local branch creates recursion

[Nils Durner]

$ bzr init .
$ bzr commit -m 1 --unchanged
$ bzr commit -m 2 --unchanged
$ bzr switch -r 1

bzr: ERROR: exceptions.RuntimeError: maximum recursion depth exceeded while calling a Python object

$ bzr log

bzr: ERROR: exceptions.RuntimeError: maximum recursion depth exceeded in cmp

[Nils Durner (ndurner)]

To rescue the corrupted branch:
1. echo "Bazaar Branch Format 7 (needs bzr 1.6)" > .bzr/branch/format

2. echo > .bzr/branch/location

3. open .bzr/checkout/dirstate, search for the most recently changed filename and copy the ID from the end of the line (between @ and ^@, starts with the user's email address)

4. enter the last revision number and the ID in .bzr/branch/last-revision:
echo "2 <email address hidden>" > .bzr/branch/last-revision

5. bzr log to check if the branch work again

[John A Meinel (jameinel)]

This isn't because of the '-r' as it can be reproduced with 'bzr switch .'

The issue is that you have a full branch, which you then issue 'bzr switch' which makes us try to change your branch into a checkout. However, since you haven't specified a location, we use the local branch as the target of the switch. Which leaves us with a branch reference pointing at itself (and thus infinite recursion).

We seem to intentionally be changing a regular branch into a branch reference during 'bzr switch', to make it easier for people to switch to 'some other' branch.

1) We could stop doing that entirely. If someone is in a regular branch, require something else in order to turn their tree into a checkout of a 3rd branch.

2) Add a check that the target of the switch is, indeed, a different branch. If it is the same location, refuse.

3) Add a check that if only '-r' is given, then the local tree is already a checkout (has a branch reference) and not a full branch.

[Nils Durner (ndurner)]

What I had originally intended is this: "svn switch -r" is used to
(temporarily) switch the working copy to that particular revision (to see
if a bug was already present back then, for example). Would it make sense
for bzr to up/downgrade the working tree without touching the branch if
just -r is passed (would that be the same as #3 - is a branch with a
working tree technically a checkout with a branch reference?)?

[Alexander Belchenko (bialix)]

Nils Durner пишет:
> What I had originally intended is this: "svn switch -r" is used to

You want `bzr update -r` instead.

--
All the dude wanted was his rug back

[Spundun (spundun)]

https://code.launchpad.net/~spundun/bzr/bzr-1018628

Created a new branch with three blackbox unittests.

I banged my head against the TestCase api for a while and gave up to create a simpler script test case. feel free to port them to the faster api.

[Spundun (spundun)]

Simplified the test where a simple colo branch creation locks the directory path.

                $ mkdir mywork
                $ cd mywork
                $ bzr init
                Created a standalone tree (format: 2a)
                $ echo A > a && bzr add a && bzr commit -m A
                $ bzr switch -b br1
                $ cd ..
                $ mv mywork mywork1
                $ cd mywork1
                $ bzr branches

This should not give error.