Puppet

Most likely, this means building a stack printing service that sends everything to the logs, rather than just printing on stdout as is done now.

This also means going through all of the code and fixing the existing calls to print traces.

It might make sense to, at the same time, refactor our error handling a bit – we currently don’t create specific error classes for the various places we raise errors, and if we did, we could probably have the error creation automatically produce stack traces as appopriate. This would itself be a good bit of work so should probably be considered separately, but it’s at least worth considering.