Contents conflicts in checkouts end up in .OTHER being added to the repository

[Mary Gardiner]

Scenario:

I am maintaining a spreadsheet in a bzr checkout. I make incompatible changes in the local copy against the master branch.

Actual behaviour:

$ bzr init /tmp/mainline
[create spreadsheet test.gnumeric attached here as test.gnumeric.BASE]
$ bzr add test.gnumeric
added test.gnumeric
$ bzr ci -m "Test"
Committing to: /tmp/mainline/
added test.gnumeric
Committed revision 1.
$ bzr co /tmp/mainline /tmp/checkout
[edit spreadsheet /tmp/checkout/test.gnumeric attached here as test.gnumeric.OTHER]
$ cd /tmp/checkout
$ bzr ci -m "test change" --local
Committing to: /tmp/checkout/
modified test.gnumeric
Committed revision 2.
$ cd ../mainline
[edit spreadsheet incompatibly, attached here as test.gnumeric.THIS]
$ bzr ci -m "Mainline change"
Committing to: /tmp/mainline/
modified test.gnumeric
Committed revision 2.
$ cd ../checkout
$ bzr update
 M test.gnumeric
All changes applied successfully.
RM test.gnumeric => test.gnumeric.OTHER
Contents conflict in test.gnumeric
1 conflicts encountered.
Updated to revision 2.
Your local commits will now show as pending merges with 'bzr status', and can be committed with 'bzr commit'.
$ bzr st
renamed:
  test.gnumeric => test.gnumeric.OTHER
modified:
  test.gnumeric.OTHER
unknown:
  test.gnumeric.BASE
  test.gnumeric.THIS
conflicts:
  Contents conflict in test.gnumeric
pending merges:
  Mary Gardiner 2008-12-12 test change
$ ls
test.gnumeric.BASE test.gnumeric.OTHER test.gnumeric.THIS

I don't know what the expected behaviour actually is, but problems with this are:

1. It has .THIS and .OTHER the wrong way around: .THIS in the checkout actually contains the mainline's test.gnumeric, and .OTHER contains the checkout's locally committed test.gnumeric
2. it assumes without proof that .OTHER is the version of the file I want
3. it actually renames the file to test.gnumeric.OTHER in the metadata. I can't imagine when this would be useful behaviour: I can't think of a scenario when I am going to want to refer to this file as "test.gnumeric.OTHER" henceforth. I am likely to want the file to continue to be named test.gnumeric.

I will also attach both mainline and checkout to this bug.

[Andrew Bennetts (spiv)]

One part of the problem here is that THIS and OTHER are poor labels. MASTER and CHECKOUT would be more intuitive, I think.

[Andrew Bennetts (spiv)]

Here's a bash script that automates the test case in the original report. It uses gzipped text files to make binary files, rather than gnumeric.

#!/bin/bash
set -e
bzr init /tmp/mainline
echo 'base' | gzip -c > /tmp/mainline/test.gz
bzr add /tmp/mainline/test.gz
(cd /tmp/mainline && bzr ci -m "Test")

bzr co /tmp/mainline /tmp/checkout
echo 'checkout' | gzip -c > /tmp/checkout/test.gz
(cd /tmp/checkout && bzr ci -m "test change" --local)

echo 'mainline' | gzip -c > /tmp/mainline/test.gz
(cd /tmp/mainline && bzr ci -m "Mainline change")

cd /tmp/checkout
bzr update
bzr st

[Martin Pool (mbp)]

This may be something Vincent recently fixed, or is working on?