The --allow-conflicts option for pull is mentioned in the manual, but the 
behavior is not documented. It seems to be it could mean multiple things:

1. It's the same as pulling, getting conflict markers and then during a 
"revert -a".
2. It always resolves to one side of the conflict, but leaves it as 
"whatsnew"
3. It always resolves to the other side of the conflict, but leaves the 
changes as "whatsnew".
4. It is exactly like the behavior with conflict markers, except the markers 
are present. (Both sides of the conflicts might be present and unrecorded, 
but with no markers). 

A simple sentence to or two to clarify the behavior would be helpful.

From a test, it appears the behavior is the same as pulling with conflict 
markers, and then doing a revert -a. I believe this means that conflicts 
are resolved in the favor of the changes being pulled in.