UnicodeDecodeError for non-existing changesets in marked up text

[menno@…]

I am using the Mercurial plugin for the IMAPClient project (​http://imapclient.freshfoo.com).

Someone recently added a ticket containing some sample IMAP protocol output that happened to contain a string that looks like a changeset hash. This changeset doesn't actually exist in the Mercurial repository, causing a UnicodeDecodeError from the Mercurial API for some reason.

This error isn't handled meaning that the ticket was no longer accessible using Trac. I had to edit the Trac DB manually and add literal markup to prevent the traceback so that the ticket could be accessed.

Through experimentation I found that any string in a ticket description that looked like a changeset hash but didn't refer to a changeset would trigger the problem and "break" the ticket (in my user's case it was "f2e30b5f0407").

I'm using the patch attached to avoid this issue. There's probably a better way of dealing with this. I suspect the real issue is with Mercurial itself. It appears to be doing something odd when queried about changesets that don't exist.

For your reference, the traceback I was seeing looked like this:

Traceback (most recent call last):
  File "build/bdist.linux-i686/egg/tracext/hg/backend.py", line 248, in _format_link
  File "build/bdist.linux-i686/egg/tracext/hg/backend.py", line 416, in get_changeset
  File "build/bdist.linux-i686/egg/tracext/hg/backend.py", line 334, in hg_node
  File "/usr/lib/python2.4/site-packages/mercurial-1.3.1-py2.4-linux-i686.egg/mercurial/localrepo.py", line 504, in lookup
    if key in self.dirstate.parents():
UnicodeDecodeError: 'ascii' codec can't decode byte 0x81 in position 4: ordinal not in range(128)

Version info:

• Trac - 0.11.6
• Mercurial - 1.3.1 (old, I know)
• TracMercurial - latest as of 2010-01-03 (r9000)

[Christian Boos]

This problem hints at on of the next big change for the TracMercurial plugin, better handling of the revision descriptors.

See #9724.