Like I said, if I could go back, I’d get rid of checked exceptions entirely.
-JZ From: Mike Drob <[email protected]> Reply: Mike Drob <[email protected]>> Date: August 1, 2014 at 2:54:47 PM To: Jordan Zimmerman <[email protected]>> Cc: [email protected] <[email protected]>> Subject: Re: Exception throwing So why declare that we throw exceptions instead of just throwing everything as a RuntimeException (or subclass thereof)? On Fri, Aug 1, 2014 at 2:49 PM, Jordan Zimmerman <[email protected]> wrote: -1 (binding) If I could go back I’d remove all checked exceptions entirely. I don’t think there’s an objective answer here - it comes down to personal preference, etc. I don’t see much value in touching nearly every file in the library in order to do this. We’ve had maybe 2 or 3 requests in the many years that Curator has exists. This suggests that the overwhelming majority accept the current exception semantics. If you can point to an actual bug that this causes that would be helpful. -Jordan From: Mike Drob <[email protected]> Reply: [email protected] <[email protected]>> Date: August 1, 2014 at 2:32:07 PM To: [email protected] <[email protected]>> Subject: Exception throwing I'd like to revisit the discussion around always throwing Exception in the API. There were two JIRA issues - CURATOR-135 and CURATOR-29 - that touch on this subject, but I think there is a good conversation to be had. I understand the suggestions that if an exception is thrown, we are in a bad state, regardless of the type of exception. However, throwing Exception comes with some unfortunate Java baggage... By declaring thrown Exception, we force consumers to also catch RuntimeExceptions instead of letting them propagate as they normally would. In some cases, the calling code may need to attempt roll-back semantics, or retry outside of what Curator provides, or something else that we haven't thought of. I'd like to propose replacing much of the thrown Exception methods with CuratorException. This will still carry the connotation that it doesn't matter what kind of exception we encounter, they're all bad. It will also be backwards compatible with the previous API, since users will still be able to catch Exception in their calling code. And it has the advantage of separating checked exceptions from unchecked ones, so that users don't unintentionally catch something unrelated. Thoughts? I tried looking for more details behind the design decision to always throw Exception, but wasn't able to find them. If they're already documented, I'd love to be pointed at the wiki or site, or a mailing list thread will do in a pinch. Thanks, Mike
