I actually have another idea. Since we all agree that passing a custom "parser" to iterparse is dubious, IncrementalParser does not have to do that at all. This will make it a much more future-proof API. So its signature can be:
class xml.etree.ElementTree.IncrementalParser(events=None)
The only remaining question is how will iterparse then work based on IncrementalParser (since iterparse does accept a parser). iterparse can just set the parser on IncrementalParser after creating it - it's an internal contract that does not have to be exposed.
This will be better than the current approach in terms of future-proofing.