autoflake
autoflake copied to clipboard
a ‘pass’ statement in an intentionally empty function with a docstring is not useless
not all pass
statements occurring in places where they are not technically required are useless.
consider for instance a command line app written using the popular click
library, featuring some subcommands (see docs). for the sake of argument, consider a hypothetical git
-like application supporting a command like git remote add
, which would look like this:
@click.group()
def remote_group(...):
pass
@remote_group.command():
def add(...):
... # implementation goes here
now, let's add docstrings, which in the case of click
, also appear in --help
output:
@click.group()
def remote_group(...):
"""
Manage set of tracked repositories.
"""
pass
@remote_group.command()
def add(...):
"""
Add a new remote.
"""
... # implementation goes here
technically, the presence of a docstring means the pass
is not needed to make it valid python. but the pass
shows the intent to the reader. without it i would immediately wonder if someone forget to write some code, or if some git merge went wrong. in this specific example, click
actually supports shared code in functions decorated with click.group()
. the presence of pass
signals that this function does not need that, and is intended as an empty body.
so, in short: not all pass
statements that are strictly technically unnecessary are to be considered ‘useless’.
my suggestion would be to treat the ‘function body, with only a docstring and a pass statement’ as an extra special case, and always leave the pass
statements in place.
I just came across this tool and was very exited. But this is a showstopper.
Can you make it configurable to remove 'useless' pass
statements or not? If I run autoflake with --remove-all-unused-imports
I expect to not make any other changes except to the imports.
This also happens if you use methods decorated with @abc.abstractmethod
: they will typically have a docstring but no implementation.
Sometimes people use ellipsis ...
in empty functions to mark it as implementation missing, though not sure is that more of a hack, as it has varying semantics depending on where it's used.
Still, I actually prefer using ellipsis in these cases, as it stands out more clear that method/function doesn't have implementation, and autoflake doesn't seem to complain about it either.
...
reads as a placeholder for code that needs to be written, or that it's a ~‘template’~ ‘blueprint’, e.g. in .pyi
stub files.
pass
in an exception class body means it's intentionally empty. another use case is the click
example and its rationale in my earlier comment.
Good explanation, thanks!
Will this be fixed?