This is an attempt to write a minimal-foreknowledge failing test for CLJ-1633. By minimal-foreknowledge, I mean a test that fails in the presence of the bug, but which one could imagine writing without intimate knowledge of the details of the bug. I suspect that looking for tests like this is a good way to find gaps in test coverage, and produce tests that will uncover novel regressions later on.
Approach: Generate a single list of operations that could be performed on a sequence, changing that sequence. Make two copies of that operation list, and insert what should be identity-preserving operations into each. Run the two lists of operations and verify that the final results are the same.
With CLJ-1633 unfixed, we get this output: