[PATCH] bzr-0.0.7 selftest failure on Cygwin

[John Whitley]

Hi all,

bzr selftest fails on Cygwin in bzrlib/selftest/blackbox.py, in  
test_ignore_patterns().  The issue is that the .bzrignore file is  
written with the 'wt' text-mode flag.  For Cygwin, this forces  
behavior that causes line-ending translation ('\n' ==> '\r\n').  The  
assert comparison then opens the file with binary-mode flags ('rb'),  
which implies that the system's line-ending must be '\n'.  The patch  
below changes these two assert tests to use universal line-ending  
support ('rU'), which then allows this (and all other) selftests to  
pass.

This issue raises some hairy questions regarding cross-platform  
source control semantics for line-endings.  My team has had a few  
battles with CVS regarding line-ending changes confusing updates and  
merges.

Cygwin is an annoying point in this space, since writing with 'wt'  
forces Windows text-mode write semantics... which produces a file  
whose line endings don't match os.linesep (Cygwin's os.linesep =  
'\n', at least for a Unix-mode installation...)

-- John



*** modified file 'bzrlib/selftest/blackbox.py'
--- bzrlib/selftest/blackbox.py
+++ bzrlib/selftest/blackbox.py
@@ -112,7 +112,7 @@
          self.assertEquals(list(b.unknowns()), ['foo.blah'])
          self.runbzr('ignore *.blah')
          self.assertEquals(list(b.unknowns()), [])
-        assert file('.bzrignore', 'rb').read() == '*.blah\n'
+        assert file('.bzrignore', 'rU').read() == '*.blah\n'
          # 'ignore' works when then .bzrignore file already exists
          file('garh', 'wt').write('garh')
@@ -120,7 +120,7 @@
          assert self.backtick('bzr unknowns') == 'garh\n'
          self.runbzr('ignore garh')
          self.assertEquals(list(b.unknowns()), [])
-        assert file('.bzrignore', 'rb').read() == '*.blah\ngarh\n'
+        assert file('.bzrignore', 'rU').read() == '*.blah\ngarh\n'
      def test_revert(self):
          self.runbzr('init')