> What would the behavior be when run on an older version without support for that placeholder be?

Then it would fail when the format string contained e.g. %(UTF8BOM)s and there was no corresponding attribute in the LogRecord - but that's true of any feature which is introduced in newer Pythons. I get that it would be unexpected in a point release, and that's why I've posted about this on c.l.py (no feedback from there so far).

> I'm okay with adding this but wouldn't be surprised if release managers are not.

Hence the post to python-dev, but no release manager has expressed an opinion yet.

> Perhaps just fixing it nicely in 3.3 is sufficient while documenting the misbehavior as a known issue for 2.7 and 3.2.

This is doable at a user level, except for the fact that the BOM insertion is currently unconditional. If I just remove the BOM insertion in 2.7 and 3.2, then I don't need to do the UTF8BOM placeholder thing in the stdlib; I could just add a cookbook recipe telling users how to do it. I am thinking about a different solution for 3.3 anyway, i.e. adding one or more overridable methods to SysLogHandler.

Since no one has objected on c.l.py about the proposed change (which implied that if there were no objections, the change would happen) Marko may be right that not many people are affected, or care. I'll wait a little while longer, and if no objections are forthcoming I'll remove the BOM insertion in 2.7 and 3.2, add a cookbook recipe for those who need a BOM and leave it at that for those versions.