Every developer runs into the dreaded nil object error: NoMethodError
in Ruby,
AttributeError
in Python, and NullPointerException
in Java. These errors are
one of the largest sources of
bugs. Even most static
languages allow nil objects to silently pass as any type of object, and it’s
just as easy in Ruby to let a nil object slip into seemingly well-factored code.
If this is such a widespread and well-understood problem, why aren’t there best
practices and solutions to solve it? There are actually many patterns and
solutions for avoiding nil object errors, but you don’t see them too often in
the Ruby community.
Don’t return nil, don’t pass nil
The most direct and bullet-proof way to avoid nil object errors is to avoid nil objects. Sometimes nil is passed around simply as a convenient “other” value or placeholder, and this is an abuse of the concept of nil. Finding and eliminating these abuses isn’t too hard, a program often needs to handle the concept of “nothing” in some way. Although nil is the most basic way to handle this case, it’s often not the best.
Handling the concept of “nothing”
As an example, let’s say you’re writing a software as a service application. SaaS applications frequently need to track which users are members of which projects. Some areas require administrator permissions, and you’d probably end up with code similar to this:
class User < ActiveRecord::Base
has_many :memberships
def admin_of?(project)
membership_for(project).admin?
end
def membership_for(project)
memberships.where(:project_id => project).first
end
end
You can probably spot the issue with this snippet. If the user isn’t a member at
all, membership_for
will return nil and you’ll get a NoMethodError
. We can
easily change admin_of?
to check for a missing membership, but that’s not
really the problem with this code.
The problem with nil is that it’s hard to know when to expect it. As a developer
coming onto this project, or as a developer coming back to a piece of code after
a while, it’s hard to know which methods might return nil. A method named
membership_for
sounds like it should return a Membership
, not nil. Unless
you dive through every method you call (and every method those methods call, and
so on), you can’t tell if a nil might be returned somewhere down the line.
Rather than adding the nil check in this one place and continuing to program in
paranoid fear, let’s look at some possible solutions besides returning nil
instead of a Membership.
Exceptions
One alternative to returning nil is to raise an exception in the “nothing” case.
The implicit contract of the membership
method is that it will return a
Membership
object, so it makes sense to throw an exception in the cases that
we can’t:
class User < ActiveRecord::Base
class NoMembership < StandardError; end
def membership_for(project)
memberships.where(:project_id => project).first or
raise NoMembership
end
end
In this case, we don’t expect that you’ll ever call membership_for
(or
admin_of?
) for a user that isn’t a member, so we can simply throw an exception
in the unexpected case. The exception allows us to specifically handle the
unexpected case of a missing membership (whereas NoMethodError
is too generic
to rescue), and it’s much easier to debug if it should happen in production.
Exceptions are useful when a method might not have a value to return, but nothing useful can be done unless a return value exists.
However, in a real SaaS, we’ll need to ask about permissions for users that aren’t members of projects, so raising an exception isn’t acceptable.
Special nil objects
Another option that I rarely see in Ruby code is the special nil object. A nil object can be written to handle situations that a method needs to handle “nothing” without returning an object that violates the implicit contract of the method:
class User < ActiveRecord::Base
def membership_for(project)
memberships.where(:project_id => project).first or
NilMembership.new
end
end
class NilMembership
def admin?
false
end
end
With the NilMembership
instance, the admin_of?
method doesn’t need to worry
about the return value. It will usefully respond to admin?
whether or not the
admin is a member.
Nil objects are useful when a method may need to return nothing and the returning method clearly knows how to handle the “nothing” edge case.
Unfortunately, this still won’t work for our real SaaS use case, because we’ll need to handle situations where a user is a member but isn’t an admin.
Maybe
Going back to our original code, now with an added member_of?
method:
class User < ActiveRecord::Base
has_many :memberships
def admin_of?(project)
membership_for(project).admin?
end
def member_of?(project)
membership_for(project).present?
end
def membership_for(project)
memberships.where(:project_id => project).first
end
end
The member_of?
method works fine, but admin_of?
is back to raising
NoMethodError
for missing memberships. We still don’t want to just check for
nil in that one client method, so let’s look at another option: the maybe
pattern.
This pattern is a little more heavy-handed and requires some setup. Here’s a simple implementation:
class Object
def maybe
Some.new(self)
end
end
class NilClass
def maybe
None.new
end
end
class Some
def initialize(object)
@object = object
end
def get
@object
end
def present?
true
end
def blank?
false
end
end
class None
class Unwrapped < StandardError; end
def get
raise Unwrapped
end
def present?
false
end
def blank?
true
end
end
And here’s how we’d use it in our application:
class User < ActiveRecord::Base
has_many :memberships
def admin_of?(project)
membership = membership_for(project)
if membership.present?
membership.get.admin?
else
false
end
end
def member_of?(project)
membership_for(project).present?
end
def membership_for(project)
memberships.where(:project_id => project).first.maybe
end
end
The idea behind the maybe pattern is to force client methods to unwrap the
return value before using it. It’s impossible to be unaware that
membership_for
might return nothing, because it never returns an object that
acts like a Membership
. You could argue that this is a violation of the
implicit contract based on the method’s name, but the return value at least
prevents a developer from missing possible edge cases.
The maybe pattern is useful for situations that you need to return nothing but there’s no obvious edge case. In those situations, it makes sense to force consumer methods to check for nothing and handle edge cases on their own.
Return nothing without returning nil
Which pattern you choose (and there are other patterns out there) depends on your situation, but it’s likely you shouldn’t choose to return nil. Nil is a sad shell of an object, and it’s just not polite to hand a developer a nil when he or she asked for a Membership. Try helping your fellow developers out and find something else to return.