Merge lp:~hid-iwata/bzr/fix-mojibake into lp:bzr

[IWATA Hidetaka]

Workaround for #777996 and #909152

Unreadable error message can be a huge blocker for newbie users.
This patch may not be perfect, but I think this is better than now.

[Martin Packman (gz)]

This is close to the right fix for this (longstanding and very annoying) issue.

I really think we want to do the decoding inside trace.report_exception rather than wrapping the stream passed to allow it to be stupid, what do you think?

[IWATA Hidetaka (hid-iwata)]

All stderr output must be encoded to terminal-encoding, even if stderr.encoding is None. (qbzr requires such behavior)
inside trace.report_exception, err_file can be any other file-like-object and I don't know which encoding to choose.
This is the reason why I use wrapped stream.

Unmerged revisions

6545. By IWATA Hidetaka on 2012-08-09

Consider decoding error.

6544. By IWATA Hidetaka on 2012-07-26

Fix mojibake of localized error message.

=== modified file 'bzrlib/commands.py'
--- bzrlib/commands.py	2012-03-13 17:25:29 +0000
+++ bzrlib/commands.py	2012-08-09 21:38:46 +0000
@@ -918,6 +918,13 @@
         os.close(pffileno)
         os.remove(pfname)
 
+def _err_file():
+    """
+    Get file like object which exception is written to.
+    """
+    return trace.EncodedOutputStream(sys.stderr,
+                                     osutils.get_terminal_encoding(),
+                                     'replace')
 
 def exception_to_return_code(the_callable, *args, **kwargs):
     """UI level helper for profiling and coverage.
@@ -932,7 +939,7 @@
         # used to handle AssertionError and KeyboardInterrupt
         # specially here, but hopefully they're handled ok by the logger now
         exc_info = sys.exc_info()
-        exitcode = trace.report_exception(exc_info, sys.stderr)
+        exitcode = trace.report_exception(exc_info, _err_file())
         if os.environ.get('BZR_PDB'):
             print '**** entering debugger'
             tb = exc_info[2]
@@ -1252,7 +1259,7 @@
     except Exception, e:
         if (isinstance(e, (OSError, IOError))
             or not getattr(e, 'internal_error', True)):
-            trace.report_exception(sys.exc_info(), sys.stderr)
+            trace.report_exception(sys.exc_info(), _err_file())
             return 3
         else:
             raise

=== modified file 'bzrlib/trace.py'
--- bzrlib/trace.py	2012-02-01 13:24:42 +0000
+++ bzrlib/trace.py	2012-08-09 21:38:46 +0000
@@ -610,6 +610,28 @@
             # to the file so there's no danger of getting into a loop here.
             mutter("Logging record unformattable: %s %% %s", msg, args)
 
+class EncodedOutputStream(object):
+    def __init__(self, stream, encoding=None, errors='strict'):
+        self.stream = stream
+        if encoding is None:
+            encoding = getattr(stream, "encoding", "ascii")
+        self.encoding = encoding
+        self.errors = errors
+
+    def write(self, text):
+        if isinstance(text, unicode):
+            self.stream.write(text.encode(self.encoding, self.errors))
+        else:
+            try:
+                text = text.decode("utf-8")
+                self.stream.write(text.encode(self.encoding, self.errors))
+            except UnicodeError:
+                self.stream.write(text)
+
+    def __getattr__(self, name):
+        """Redirect to self.stream"""
+        return getattr(self.stream, name)
+
 
 class Config(object):
     """Configuration of message tracing in bzrlib.