While the general principle is correct, I'm not sure we should ban cases like this:

try:

    do_something_specific

except Exception as e:

    log.debug("Couldn't do something: %s", e)

    do_something_in_another_way

where overall failure in the case of an unexpected exception is undesirable.

do_something_specific has a handful of known exceptions it can raise. If something else gets raised, something strange is probably going wrong. If an exception outside those defined for do_something_specific is raised, how do you have any idea if do_something_in_another_way could possibly work?

We have a number of cases where we want to make a "best effort" to do something, and continue if not. Plotting code is typical. In such cases I think it is appropriate to log and move on.

We also have a lot of code that does not document what exceptions it may raise. And even if all our code did document the exceptions it could raise, that documentation is likely to go out of date as our code changes. It can be very tricky to maintain such documentation because code calls other code that calls other code...

I agree that logging exceptions is much better than silently ignoring them. I am much less sanguine that we can always catch just a few specific exceptions. Sometimes we certainly can. But for complex code it is often safest to cast a broader net, which in Python usually means catching Exception.

Having just been burned by this again just now in makeCoaddTempExp (whose run contains a for loop with 4 "catch and mask all exceptions" blocks!), I'm going to try to make a list of these cases. At absolutely bare minimum, they should be modified to respect doRaise.

Just voting here that I support doing something about this