Interesting thread. Thanks for launching it.
I wrestled with this recently and, while I know a lot of monks more experienced than me would recommend just die-ing on error, I decided to use a softer approach. I created a few error-handling methods for my class that can be called as follows:
$obj->set_err( 'Something went boom' );
my $msg = $obj->err_str;
my $trace = $obj->err_trace; # error msg + full stack trace
The main reasons I went with this approach are:
- I didn't want to wrap nearly every method call with an eval
- I consider very few of the possible errors to be fatal-worthy
- I wanted the methods to have the ability to pass information back to the caller as a warning even if the method succeeds (e.g., over-writing data may or may not be intentional)
In short, I wanted my class to be able to warn (or carp
) that something might
be fishy (e.g., with the input data) without die
-ing all the time, and I didn't want the behavior of the class to overly-influence how I wrote the calling program. (I work in research and it is common to write a one-off to explore an idea; in some cases this means using data that may cause the class methods to throw errors that I want to ignore, and continue processing without having to work around fatal exceptions.)
I realize this may be a choice that goes against accepted conventions, and over time (as my modules mature and become more battle-tested) I may change the way errors are handled. For now, though, this approach works well.