Testing private functions in classes

85 次查看(过去 30 天)
I'm wondering what the best way is to write unit tests for private functions and properties in classes. Say for example, I have a class that represents a ball launched as a projectile:
class ball
%
properties (Access = private)
initialSpeed = 25; %m/s
acceleration = -10; %m/s^2
end
%
methods
function distTravelled = getDistanceTravelled(obj, timeElapsed)
% Estimate total distance traveled by the ball
% (I know this is a bad approximation)
currSpeed = obj.getCurrSpeed(timeElapsed);
distTravelled = 0.5 * (obj.initialSpeed + currSpeed) * timeElapsed;
end
end
%
methods (Access = private)
function currSpeed = getCurrSpeed(obj, timeElapsed)
%Calculate current speed of ball
currSpeed = obj.initialSpeed + obj.acceleration * timeElapsed;
end
end
end
How should I write a test to check that the values for acceleration or that the value returned by the method getCurrSpeed is accurate? Should I just allow access to the testing functions?

采纳的回答

per isakson
per isakson 2017-10-19
编辑:per isakson 2017-10-20
First goggle "test private method" and read about why you should not do it (and a few ways to do it).
One way (for handle classes only) is to include the test in the class itself.
>> mc = MyClass
mc =
MyClass with no properties.
>> mc.test_private
Private mysort is running
>>
where
classdef MyClass < matlab.unittest.TestCase
%
methods ( Test )
function test_private( this )
this.mysort
end
end
methods ( Access = private )
function mysort( this )
fprintf( 'Private mysort is running\n' )
end
end
end
A better workaround based on localfunctions (added 19 hours later)
>> mc = MyClass;
>> fh = mc.get_local_function_handle;
>> fh = fh{1}
fh =
@the_tricky_algorithm_
>> fh( mc )
ans =
144
>>
where in one mfile
classdef MyClass
properties
val = 12;
end
methods
function fh = get_local_function_handle( ~ )
fh = localfunctions();
end
end
methods ( Access = private )
function my_private_method( this )
fprintf( 'Private mysort is running\n' )
the_tricky_algorithm_( this )
end
end
end
function out = the_tricky_algorithm_( this )
out = this.val .* this.val;
end
  3 个评论
Walter Roberson
Walter Roberson 2017-10-20
When I worked in industry, we tested everything.
Later I worked with some people developing medical devices; I'm pretty sure their applications would have been rejected if they didn't test everything.
JWTay
JWTay 2017-10-20
Good to know! I am trying to develop a "best practices" in regards to my coding so I appreciate the info.

请先登录,再进行评论。

更多回答(4 个)

Wil Koenen
Wil Koenen 2018-7-3
If you're using matlab.unittest.TestCase for your test, but you don't want matlab.unittest.TestCase to be a superclass of the class under test, you can grant access to private functions by providing a list of classes instead of the private keyword. Example:
classdef classUnderTest
...
methods ( Access = { ?classUnderTest, ?matlab.unittest.TestCase } )
...
end
end
  2 个评论
Martin Lechner
Martin Lechner 2020-6-2
编辑:Martin Lechner 2020-6-2
This solution works perfect. Thanks.
I enabled explictly my test class but this included my unit test class in the compiler output. Your solution to enable access for all matlab.unittest.TestCase doesn't include the test classes in the compiler output.

请先登录,再进行评论。


Sean de Wolski
Sean de Wolski 2017-10-19
编辑:Sean de Wolski 2017-10-19
I would test it through the front door with a known set of inputs and expected outputs. This is the beautiful thing about classes in that you can mask the implementation from the outside world and change it later if necessary.
e.g.:
In Unit Test
b = ball
distTravelled = getDistanceTravelled(b, 20);
testCase.verifyEqual(distTravelled, whatever_is_right_for20);
distTravelled = getDistanceTravelled(b, 200);
testCase.verifyEqual(distTravelled, whatever_is_right_for200);
testCase.verifyError(@()getDistanceTravelled(b, -2), 'ball:NoNegativeTime');
You can use the Code Coverage Plugin for a test runner to make sure that you're exciting all of the lines of the implementation and that should be fine.
  1 个评论
JWTay
JWTay 2017-10-20
That is my practice in general.
I guess that the reason I wanted to test some of the private functions is to make sure that the calculations are correct, and this is not directly reflected in the output (i.e. the results of the private methods are used to influence some decision-making in the code). I could come up with model input/expected output data but when it breaks, it might not be easy to trace which function was giving the wrong answers.

请先登录,再进行评论。


Sanjana Ramakrishnan
编辑:per isakson 2017-10-19
Refer the below link for an example of writing MATLAB unit tests relevant to your case:
Please note that testing private functions is just the same as testing any other function in a class.
You can design your own testing strategies as per your requirement.
  1 个评论
per isakson
per isakson 2017-10-19
"testing private functions is just the same as testing any other function in a class" Is that really so?

请先登录,再进行评论。


Tom Hawkins
Tom Hawkins 2020-2-12
Would you be happy to define your properties/methods as Protected, rather than Private? If so you could then create a subclass of your class as part of your test suite, for example:
classdef BallWithAccessToProtected < ball
methods
function obj = BallWithAccessToProtected(varargin)
obj@ball(varargin{:})
end
function SetProtectedProperty(obj, prop, value)
obj.(prop) = value;
end
function v = GetProtectedProperty(obj, prop)
v = obj.(prop);
end
end
end
This subclass exposes the protected properties via its SetProtectedProperty and GetProtectedProperty methods, for example:
>> b = BallWithAccessToProtected(someArgs);
>> b.GetProtectedProperty('initialSpeed')
ans =
25
You can figure out how to extend that to accessing methods if you need to.
Protected means that users of your class's 'official' API can't access those properties and methods, but someone who's willing to subclass it can (as we've done above) - but remember that some languages like Python don't even have a mechanism for restricting access to internal properties and methods of a class, only a naming convention to show which ones you shouldn't really use.
  1 个评论
Wil Koenen
Wil Koenen 2020-2-13
In your example, you could leave out the constructor, because MATLAB supplies a default constructor that does the same (reference).
When a subclass does not define a constructor, the default constructor passes its inputs to the direct superclass constructor. This behavior is useful when there is no need for a subclass to define a constructor, but the superclass constructor does require input arguments.

请先登录,再进行评论。

类别

Help CenterFile Exchange 中查找有关 Construct and Work with Object Arrays 的更多信息

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by