Exception diaper just raises about raygun4py HOT 3 CLOSED

The reason for this is to prevent a circular reference which the GC cannot handle efficiently. Please see this issue for more details: #26 (and the official documentation of this behavior in the warning at https://docs.python.org/2/library/sys.html#sys.exc_info).

I'll add a code comment for those lines in the next release as it does look odd if the reader is unfamiliar with that behavior (it is rather counter-intuitive).

from raygun4py.

I think I understand what the code is trying to accomplish, but I still think it's broken. As I understand the docs you linked to, the recommended pattern for avoiding the circular reference is more something like this:

def some_exception_handler():
    exc_type, exc_val, exc_tb = sys.exc_info()
    try:
        do_stuff_with_exc_info(exc_type, exc_val, exc_tb)
    finally:
        del exc_type, exc_val, exc_tb

The biggest distinction of note is that deling the references to the exc_info() values is not what needs to be guarded by the try clause, it's the cleanup action that needs to take place should anything you are doing with the exc_info values in the try clause throw any kind of exception.

del some_var doesn't throw exceptions when the name some_var is in scope, which it plainly is with just a quick glance at the code, so you don't need to wrap that in a try clause. As it sits now, the only way your except clause will be triggered is via some bug in the python interpreter. Not a very likely scenario.

Also, even if you were concerned about del exc_type, exc_value, exc_traceback somehow failing in the try clause, the following two pieces of code are equivalent in python:

try:
    do_something()
except Exception:
    raise

and

do_something()

In other words, if you have nothing but raise in your except clause, you might as well not wrap the try clause at all, because you're throwing the error up to your caller regardless.

from raygun4py.

Hi Don,

I'm inclined to agree with the analysis above, had a quick attempt during the last release at reproducing the actual behavior which didn't work. In order to confirm the impact of this change it looks like I'll need to create a proper repro and profile the GC to see the circular reference. This will allow us to verify removing the raise/altering the logic. As the code is running in production with no (reported) issues as this time we have a policy on not making breaking changes without evidence, so I'll be revisiting this issue and hopefully fixing the logic once I have time to create the repro.

from raygun4py.