When attempting to combined HSC PDR1 (DM-13666) data for WIDE and DEEP, the following Butler construction does not work correctly:

butler = dafPersistence.Butler(inputs=[{'root': '/datasets/hsc/repo/rerun/DM-13666/WIDE/'}, {'root': '/datasets/hsc/repo/rerun/DM-13666/DEEP/'}])

One clue is that this prints the following INFO:

CameraMapper INFO: Loading exposure registry from /datasets/hsc/repo/registry.sqlite3

CameraMapper INFO: Loading calib registry from /datasets/hsc/repo/CALIB/calibRegistry.sqlite3

CameraMapper INFO: Loading Posix exposure registry from /datasets/hsc/repo/rerun/DM-13666/DEEP

CameraMapper INFO: Loading calib registry from /datasets/hsc/repo/CALIB/calibRegistry.sqlite3

CameraMapper INFO: Loading exposure registry from /datasets/hsc/repo/registry.sqlite3

CameraMapper INFO: Loading calib registry from /datasets/hsc/repo/CALIB/calibRegistry.sqlite3

CameraMapper INFO: Loading calib registry from /datasets/hsc/repo/CALIB/calibRegistry.sqlite3

And there should not be anything with Posix registries here. (note that if you reverse the order of the inputs, then the WIDE gets loaded as a Posix registry)

Upon deeper inspection, the problem is logged to debug in getParentRegsitry (sic): https://github.com/lsst/daf_persistence/blob/1e7c341a1a884cf06fc74a946a126c1f72f56360/python/lsst/daf/persistence/butler.py#L570 (and following).
I believe that on the first call to _setParentRegistry() this argument is `None` (not sure why), but on subsequent calls (after repoData.repo = Repository(repoData) is called) then it's okay.

I checked what would happen if instead of kicking out instead it calls parentRepoData.repo = Repository(parentRepoData) and that appears to work, though I'm afraid of the side-effects. (Although it does pass all local scons tests).