After looking at this what I've done is add formatterClasses to the convertRepo configuration file and that formatter is then passed all the way down to TargetFileHandler where it is added to the FileDataset. Seems to work for cpBias/cpFlat.

Jim Bosch would you mind taking a look. Most of the changes are adding a formatter parameter all the way up the call stack. It looks like I could also remove the cpBias/cpDark hard coding in BuilderTargetInput if we wanted (essentially the same changes). The main question there is whether the config file states an override FileHandler class for a dataset type name, or defines a set that indicates multi extension file handler should be used.

Looks good!

Jim Bosch did you have an opinion on the hardcoding of cpBias/cpDark for choosing the multi extension handler?

What was done on the branches was what I expected to see, and certainly seems reasonable.  I don't think I understood what the alternative was that you were describing, but if what you've done was also what you preferred, maybe I don't have to.

I haven't done anything about this line: https://github.com/lsst/obs_base/blob/master/python/lsst/obs/base/gen2to3/repoWalker/builders.py#L212

My question is really whether we care enough for me to add similar API arguments to pass down a handler class from above (as done for the formatter) to remove this DECam-ism from obs_base. If the answer is yes then I imagine the config option should be a mapping from dataset type name to handler class.

Ah, right.  Sorry I'd forgotten that that was different.  I'm okay with leaving it, just because I think this code will be irrelevant before it becomes a maintenance problem.  Fine to fix it, too, of course, and I agree the solution would be similar to what you've already done elsewhere on this ticket.