Misleading/unexpected output when failed to load a checkpoint during normal workflow
When a checkpoint fails to load, eg. simply because it wasn't computed before (normal workflow), a logger.error
message is still displayed, and a backtrace is printed.
This is misleading and implies that something went wrong. We should make this display nicer and
- only print the backtrace on debug (
-g
) mode; - restrict the cases upon which the error message is displayed to those that are legitimately an error, or were explicitly requested by the user (
--checkpoint FOO
) and still failed with whatever error.