Merge lp:~gz/bzr-builddeb/decode_changelog_for_commit_message_853664 into lp:bzr-builddeb

[Martin Packman]

This branch makes builddeb set the commit message correctly again from non-ascii debian/changelog files.

With Bazaar 2.4 there's a new hook to specify the commit message fully rather than just using a template, which builddeb takes advantage of when it's available. The old template hook point expects a byte string, and decodes the result from the user encoding. The new hook point doesn't do any decoding so returning non-ascii bytes results in bzr internals complaining.

According to Debian Policy, "The entire changelog must be encoded in UTF-8." so it should be safe to just decode without need for any fancy error handling:
<http://www.debian.org/doc/debian-policy/ch-source.html#s-dpkgchangelog>

Note the old hook is actually subtly broken, it relies on the program being run in a locale with a UTF-8 encoding or it would break (but in an obvious and fixable way, so it's not that important).

[Jelmer Vernooij (jelmer)]

urgh, I cry every time I see the two space indentation..

Thanks for the fix.

review: Approve

=== modified file '__init__.py'
--- __init__.py	2011-06-24 11:37:57 +0000
+++ __init__.py	2011-09-29 17:56:29 +0000
@@ -123,7 +123,8 @@
     bugs_fixed = find_bugs_fixed([changes], commit.work_tree.branch)
     commit.builder._revprops["bugs"] = "\n".join(bugs_fixed)
 
-    return debian_changelog_commit_message(commit, start_message)
+    # Debian Policy Manual states that debian/changelog must be UTF-8
+    return changes.decode("utf-8")
 
 
 def changelog_merge_hook_factory(merger):

=== modified file 'tests/blackbox/test_builddeb.py'
--- tests/blackbox/test_builddeb.py	2011-07-19 15:35:48 +0000
+++ tests/blackbox/test_builddeb.py	2011-09-29 17:56:29 +0000
@@ -171,4 +171,25 @@
     self.assertPathExists('pre-export')
     self.assertInBuildDir(['pre-build', 'post-build'])
 
+  def test_utf8_changelog(self):
+    from bzrlib.plugins import builddeb
+    from bzrlib.msgeditor import hooks
+    if "set_commit_message" not in hooks:
+      self.skip("No set_commit_message hook in bzrlib this old")
+    hooks.install_named_hook("set_commit_message",
+      builddeb.debian_changelog_commit, "Test builddeb set commit msg hook")
+    tree = self.make_unpacked_source()
+    # The changelog is only used for commit message when it already exists and
+    # is then changed, so need to clear it, commit, then set the contents.
+    open("debian/changelog", "w").close()
+    tree.commit("Prepare for release", rev_id="prerel")
+    c = self.make_changelog()
+    c.add_change("")
+    c.add_change("  *  \xe2\x80\xa6and another thing")
+    self.write_changelog(c, "debian/changelog")
+    self.run_bzr(['commit'])
+    branch = tree.branch
+    self.assertEqual(u"*  test build\n*  \u2026and another thing\n",
+      branch.repository.get_revision(branch.last_revision()).message)
+
 # vim: ts=2 sts=2 sw=2

=== modified file 'tests/test_commit_message.py'
--- tests/test_commit_message.py	2011-06-13 23:01:14 +0000
+++ tests/test_commit_message.py	2011-09-29 17:56:29 +0000
@@ -115,3 +115,16 @@
         self.assertEqual(commit.builder._revprops, 
                         {'bugs': 'https://launchpad.net/bugs/1234 fixed\n'
                                  'https://launchpad.net/bugs/4321 fixed'})
+
+    def test_set_message_returns_unicode(self):
+        wt = self.make_branch_and_tree(".")
+        self.build_tree(['a', 'debian/', 'debian/changelog'])
+        wt.add(['debian/', 'debian/changelog'])
+        wt.commit("one")
+        self.set_changelog_content("  * \xe2\x80\xa6real fix this time\n")
+        wt.add(['a'])
+        wt.lock_read()
+        self.addCleanup(wt.unlock)
+        commit = self._Commit(wt)
+        self.assertEqual(debian_changelog_commit(commit, None),
+                u"\u2026real fix this time\n")