Pytorch-lightning: Improve Exception Handling

[awaelchli]

🚀 Code Quality Improvement

I came across this a few times already:

try:
   # something
except Exception:
   # something

It is the worst possible way to handle Exceptions. It is better to catch the specific exception or at least log a message.

Alternatives

None. Sooner or later someone has to deal with this anyway :)

[Borda]

@awaelchli so what do you propose to have more exact Exceptions like ImportError?

[awaelchli]

Yes exactly, the specific exception we try to catch.
Also, sometimes we don't need exception handling at all:

try: 
    x = myobject.attribute
except Exception
    x = 0

can be replaced with
x = getattr(myobject, "attribute", 0)

[awaelchli]

Another code smell we frequently see (in many places in the code):

try:
    from some_package import some_module
except ImportError:
    SOMEPACKAGE_AVAILABLE = False
else:
    SOMEPACKAGE_AVAILABLE = True

Better:

catch ModuleNotFoundError instead

Even better (and shorter):

import importlib
SOMEPACKAGE_AVAILABLE = importlib.util.find_spec("some_package") is not None

[Borda]

About the import error, I remember that thare is something twisted about python version and avalanche exceptions, I guess that py3.6 doesn't have the one suggest (need to check)

Also is there a simple way how to import jist a function from a package that you are not sure if it is installed?

[awaelchli]

Also is there a simple way how to import jist a function from a package that you are not sure if it is installed?

I think the simplest is this:

if importlib.util.find_spec("some_package") is not None:
    from some_package import my_function

because when we import the function in Python it always has to import the whole package anyway.

[stale[bot]]

This issue has been automatically marked as stale because it hasn't had any recent activity. This issue will be closed in 7 days if no further activity occurs. Thank you for your contributions, Pytorch Lightning Team!