It turns out we already had code that was supposed to fix this (by resolving one alias explicitly when defining another that depended on it), but it was subverted by the fact that the slot aliases weren't being defined until after that code ran. With that fixed, I think we may need to get rid of the need for multi-level alias resolution entirely, which would be a big simplification. I'll explore that idea on DM-3401.

John Swinbank, smallish review for you that should hopefully be somewhat familiar: this fixes the recursive aliases you saw on DM-1674, where were due to the plugin algorithms being initialized before the slots; that ordering subverted code that was supposed to resolve some aliases immediately. I've also added code to check for that condition and throw, and that in turn required changes to a bunch of test code that was also initializing them out of order.

All changes are on meas_base tickets/DM-3400; I'll kick off a CI run now, and there's a chance that will inspire some changes to other packages, but they should be trivial.

Looks good; minor comments on the PRs on meas_base and meas_algorithms, but they're just about tidiness – nothing really substantive.

Wearing my test-all-of-the-things hat, I'd probably have added a unit test which demonstrates that the the exceptions you've aded in InputUtilities.cc really do get thrown when things are set up out of order, but I don't think that's essential.

I've added a unit test - none of the code in InputUtilities.[h,cc] was being tested directly, and this seemed like a good opportunity to add that.

Will merge after one more Jenkin's run; adding the test required another slight change to the test utility code, and I want to make sure that doesn't break anything downstream.

Merged to master. Final CI run did indeed uncover one more trivial change I needed to make, in ip_diffim.