@Eric: ok, I was just asking given the "unusual" situation 2.7 is (i.e. a very long support series), but it's perfectly fine not to backport the feature.

@Charles-François: I changed a bit the logic: I check for 'None' at first, since it signals that we don't want to change the uid/gid, and then I recheck for None after _get_uid/gid, and only there raise a ValueError (is there a better exception for cases like this?) - what do you think about it?

@Éric: re rewrap: I kinda find that almost every core dev has his opinion on this :) some don't want to rewrap, others are fine if you're touching that paragraph, and so on. Anyhow, to be on the safe-side, I un-rewrapped the first paragraph of os.chown().

@Gregory: I'd like to first concentrate on the "single path" function, then we'll look at the "tree" version.

Thanks a lot to everyone for the review/comment/inspiration: attached you can find the updated patch. If something still needs a fix, I'd be happy to work on it.