John Parejko has asked me to take the ticket out of reviewed state because I don't think the ticket should be merged when it is so much slower than the previous implementation.

Looking at the code the only difference I can see between the new and the old version is that the obs_base conversion script does not set rootSkyMapName in the config. Maybe that causes the problem?

Thanks for that suggestion, Tim Jenness: I don't understand why `rootSkyMapName` mattered (the skymap config and name was already specified in `config.skymaps`), but it appears to have cut the convert runtime down to about a minute.

That's great. I haven't looked too deeply at the code but I saw a bunch of "if rootSkyMapName is None" shenanigans going on suggesting that different code paths would be involved.

I saw a bunch of "if rootSkyMapName is None" shenanigans going on suggesting that different code paths would be involved.

This is a workaround for the fact that a Gen2 repo can have datasets defined on tract and patch without having a skymap dataset that defines what those mean.  It's a totally ill-formed repository configuration, and apparently we get away with it because people just guess that a skymap dataset in a child repository is the right one for the parent.  That config option sets what skymap to guess when the information simply isn't in the Gen2 repo.

Jenkins failure due to the bug introduced by DM-24285, but I think everything in this ticket is ok. I've merged it to get this done with. If there is a bug hiding because of the problem in DM-24285, let me know and I'll fix it ASAP.