Yes, I guess those statements about error handling do imply the behavior I'm seeing. They don't immediately make me think of that behavior, though. I'll try to figure out of I can make it clearer and try a doc patch if so.

Since Try::Tiny says it exists to avoid the mistakes people make using eval for this kind of thing it does seem like using it again inside the the finally block as needed is what the authors had in mind...should be amusing.

(I only had one place in the code I'm actually working on where I'd messed up use of try blocks, but it was good to run through and check all the ones I'd added to be sure.)

You're probably doing something far more complicated than you ought to if your finally block can throw an exception. I wouldn't do anything more than logging an error or setting a flag to restart a computation.

So finally blocks aren't in fact good ways to make sure cleanup gets performed? That's kind of advertised as their main purpose -- but DBIx, for example, reports most errors with exceptions (and I like that style in general).