Where does self.message get set on the Revision object?

[Alexander Belchenko]

Jelmer Vernooij пишет:
> On Tue, Jun 02, 2009 at 12:51:31PM -0500, John Arbash Meinel wrote:
>>>> But I'll comment that this Revision object has probably not been
>>>> serialized to disk. And my bet is that it is generated from something
>>>> like bzr-svn.
> 
>>> I believe I stated that initially, but yes, it's coming from bzr-svn.
>>> I ran the client directly against the Subversion repository (*the*
>>> Subversion repository), and ran into a commit that didn't have a log
>>> message.  If it's bzr-svn's fault (and it sounds like you're hinting
>>> that it is), then I can look to make the change there so that it
>>> provides a default empty message.  If it's truly expected that
>>> self.message can be None, get_summary() should definitely be updated,
>>> and depending on it's output, qbzr might need to be fixed to.  I'm
>>> just trying to figure out where I need to focus my energy. :-)
> 
>>> I'll definitely send submit a patch and a test for get_summary().  But
>>> if anyone knows the answer as to whether it's acceptable for .message
>>> to be None, I'll submit a patch to bzr-svn as well, if it's considered
>>> to be broken in that respect.
>> 1) bzrlib internals seem to always set Revision.message to the empty
>> string, rather than None
>> 2) Everything but LineLogFormatter seems to wrap a:
>>   if not rev.message:
>> 	# say something about "no message"
>>   else:
>> 	message = XXX
>> Which means that they handle None or '' equally.
>> 3) I don't really care whether Revision.message is None or ''. There at
>> least is an argument that bzr-svn => bzr => will end up as an empty
>> string, so arguably bzr-svn should encode that in the beginning.
>> Otherwise "SVNRevision == Revision" would fail because None != ''.
> 
>> I would probably say that Revision.get_summary() should return an empty
>> string, and that LogLineFormatter should format that into "(no message)"
>> if it wants.
> 
>> And thus QBzr can also format '' => (no message) or whatever other
>> string it wants to use. I prefer that to get_summary() doing the
>> formatting, because QBzr can then know if it is a truly empty message
>> and change to do something like use a grayed-out message for that
>> revision, etc. If you return "(no message)" then QBzr needs to guess
>> whether there is no message, or whether someone explictly chose to do:
> 
>>   bzr commit -m "(no message)"
> 
> It would be nice if rev.message = None was supported.  
> Subversion distinguishes between the commit message not being set at
> all and it being empty. It would be nice if the fact that the commit
> message was not set at all could be kept when importing into Bazaar
> rather than "cast" to an empty message. 
> 
> We should add some tests to see that the serializer deals with
> None properly, whether we decide to allow it or not.

IIUC from John analysis if there is no serialized message item, it should deserialized as None.
Does it right?