|
I currently have a sizable code base with LinFu that I can't afford to rewrite from scratch, and I want to increase the code coverage rate on existing portions of the library which don't have any coverage at all. Is there any way to apply unit tests to existing code and increase the coverage rate, or is this an exercise in futility?
|
|
|
|
|
Philip Laureano wrote: Is there any way to apply unit tests to existing code
Couldn't you use LinFu to do that? 
I just spent like two minutes looking at your LinFu and that was the first thing I thought of. Also though, I have used NUnit for several years and I find the quote above from you a bit confusing because of course you can apply unit tests to existing code. I must be missing something.
led mike
|
|
|
|
|
led mike wrote: Couldn't you use LinFu to do that? [Confused]
I just spent like two minutes looking at your LinFu and that was the first thing I thought of. Also though, I have used NUnit for several years and I find the quote above from you a bit confusing because of course you can apply unit tests to existing code. I must be missing something.
Hi Mike,
Sorry for the confusion. This question is more of a philosophical question rather than a technical one, so I'll go ahead and rephrase it: "Should unit testing be applied to an existing code base that already works?"
If the rest of the code base changes very little, wouldn't the tests themselves be a case of YAGNI?
led mike wrote: Couldn't you use LinFu to do that?
I suppose LinFu could "dogfood" itself and put the tests everywhere, but that's beside the point. The real question here is that is there really any value (read: value = quality) gained in adding unit tests for code that has already been proven to work on an innumerable amount of occasions?
|
|
|
|
|
Philip Laureano wrote: The real question here is that is there really any value (read: value = quality) gained in adding unit tests for code that has already been proven to work on an innumerable amount of occasions?
If you're never ever going to change the code again - no. If there's the possibility that the code could change - yes. The unit tests will help give you confidence that your code still works post changes, which is (obviously) hugely important.
|
|
|
|
|
Philip Laureano wrote:
If the rest of the code base changes very little, wouldn't the tests themselves be a case of YAGNI?
Why not simply add the tests once you do need to change the code?
As you say, if it won't change then it's pretty pointless.
On the other hand, you might find a bug or two in the existing code if you add tests.
|
|
|
|
|
I hope somebody can offer me some guidance as I'm a little stuck and not sure what exactly to do.
I have a nicely designed system that has utilised a couple of abstract classes which then have concrete classes implementing them.
I'm using the base classes to encapsulate functionality that is applicable across the objects but only use the concrete classes in the application with each concrete class being used in a specific way (so that you wouldn't swap one out for the other)
I've run into an issue unit testing, trying to use [Moq] as it is unable to mock a class that inherits from an abstract base class where the methods and properties are not virtual.
The only way out of the situation is to either create interfaces specific to each of the classes that inherit from the abstract base (which seems pointless to me, since I'd implement an interface, have current concrete class implement the interface and then replace all instances where concrete class was used with interface) or I have to mark the methods and properties of the base class as virtual and cause a performance problem.
I don't like either option and wondered if anybody had come across a similar situation and how they had resolved it?
|
|
|
|
|
I'm confused by the title, if you had been writing this system using TDD you would not have the issue because you would have designed your classes with testability in mind.
Also it's not Architecture vs TDD they aren't conflicting by any means I practice TDD and have found it a good way to design you generally end up with objects that are loosely coupled and because of the code coverage gained you can perform constant refactoring leading to an architecture that evolves more easier
|
|
|
|
|
The issue I'm having is that I am unable to use the testing technology that is available to enable me to adequately test the system by isolating each layer.
To be forced to have to use interfaces where there really is no need has a 'smell' to it.
The system wasn't Test Driven Design, it was designed using normal OOP techniques and it is only trying to use Test Driven Development techniques that I have run into problems.
I have found one solution and that is TypeMock which will happily mock a class that has an abstract base class the only downside being the cost.
Having spent a fair amount of time looking at Mock 'products' it seems to me that for the most part you end up having to sacrifice legitimate design choices or alter code so that it can cause performance penalities (such as marking methods/properties as virtual) in order to be able to test.
I haven't come up with any answers and am more than happy to hear other peoples opinions about this.
|
|
|
|
|
Mock objects are designed to be used up front. Attempting to apply them at the end of a development is pretty much a futile issue.
|
|
|
|
|
Can you please elaborate on why?
In a well designed system with good separation of concerns it should be easy to use Mock objects, is it not simply that generally the technology we have in .Net to support mock objects is not up to the task?
|
|
|
|
|
Lowest of the Low wrote: In a well designed system with good separation of concerns it should be easy to use Mock objects,
Well - if you've used techniques such as Inversion of Control then, yes, Mock objects should be fairly easy to use. However, most systems are developed against timescales where good practices like this go out the window. Those that don't tend to have been developed with TDD up front, so the actual mocking has been done well beforehand.
|
|
|
|
|
Lowest of the Low wrote: or I have to mark the methods and properties of the base class as virtual and cause a performance problem.
What kind of application are you building?
I'm willing to bet money that you will not notice any difference at all if you go virtual unless you are building some extremely processor intensive algorithm stuff.
Anyway, there are other mock frameworks that allow you to do what you want.
(eg, Typemock , but that one is commercial)
There are also two kinds of mock usage:
1) You use the mock as a stub just to return test data for you.
Eg, you want your DAL class to return some dummy list of customers to your business layer.
2) You do interaction tests, you want to see that classes communicate correctly.
Eg, you might want to test that a business class calls your logger class under certain conditions.
(http://martinfowler.com/articles/mocksArentStubs.html)
The first case is is easy to do with mockframeworks that requires virtual or interfaces.
Since if you need a stub you need to create some sort of interface anyway..
So in this aspect Pete is completely right.
In the 2nd case, you want to make sure that a certain call flow occurss under a certain condition.
In this case you do need to have actual code that interacts, and the mock framework should simply ensure that certain methods were or were not called.
This can also be done with virtual/iface mocks, but they are not always enough, eg, they can not ensure that a static method was called.
So in this aspect you are better off using a mock framework that can intercept any type of call.
(One could argue that static methods are ugly and you should design your api so that you do not need to test if a static/private/non virtual method was called, but that is a whole different story  )
|
|
|
|
|
The application is fairly standard, its a web application broken down into 3 layers Presentation, Business Logic and DataAccess.
Within the data access layer we have an abstract base class that encapsulates standard db access and then we have concrete classes to encapsulate the data access for different parts of the application to different databases. Within the Business logic layer a class will use a specific concrete classes to perform their data access.
Inside the business logic layer we have an abstract class that again encapsulates standard functionality used by the concrete classes representing entities.
I'm only using mock objects in relation to testing the business logic layer and am using a mixture of the 2 types of mock you describe (not in the same test).
I have been using the community edition of TypeMock and it does exactly what I want but the problem is that I am unable to use code coverage inside VSTS with the community edition and it is doubtful that the company will pay for the licence which is why I have had to look into other Mock frameworks which has lead me to run into the issue of not being able to mock.
In relation to your type 1 mock how would you go about using an interface when the class you are attempting to mock is unique within the app? Or am I just being dense here and you would simply replace the class interface with an explict interface and implement that in the class? Even if you do this it just 'feels' wrong having to create interfaces where a concrete class would work (from an application point of view)
I would value your thoughts.
|
|
|
|
|
Ok but in such case, you won't suffer from any perf problems at all if you mark your methods/props as virtual
The overhead of virtual calls is non existing when you bring html rendering and database access into the picture.
Lowest of the Low wrote: In relation to your type 1 mock how would you go about using an interface when the class you are attempting to mock is unique within the app? Or am I just being dense here and you would simply replace the class interface with an explict interface and implement that in the class? Even if you do this it just 'feels' wrong having to create interfaces where a concrete class would work (from an application point of view)
Well if you use them as stubs because your concrete class is not yet written (kind of the purpose of stubs)
Then you would create an interface, maybe not because the app absolutely needs it in the end, but because _you_ need it to be able to continue to work before the stubbed class is actually implemented...
The interfaces adds value in form of testability which make you able to develop faster and you get an app that is easier to maintain, even if they don't add much value inside the executing application itself.
modified on Monday, June 9, 2008 8:01 AM
|
|
|
|
|
class A
{
public:
...
AddTo(CList &L);
private:
}
For the bad, A is couple with CList.
For the good, A knows the knowledge to fill CList. But I think it's not necesary for A to know CList.
Is it better to write a gobal function, like:
bool FillList(CList &L, const class &A);
|
|
|
|
|
Why do you want to write a function for that ? CList probably has a function to do that (are you talking about MFC CList?). If not, then you should put this function in CList.
|
|
|
|
|
For polymophism,
class A {
public:
virtual void Fill(CList &L);
}
class B : pubic A
{
public:
virtual void Fill(CList &L);
}
class C : pubic A
{
public:
virtual void Fill(CList &L);
}
If I get a object refereced by A *p, I may call p->FillList(L).
In another sense, the object knows what to add to the list, it's a reason to add such a memeber function.
The bad thing is that they are coupled.
|
|
|
|
|
The list handles perfeclty adding polymorphic objects to itself. So, what's the problem ? Why do you absolutely want to have a function in class A ? That doesn't make a lot of sense to me. Of course, your list should store pointers to A, then you can add B or C objects to the list and keep the power of polymorphism.
|
|
|
|
|
I haven't say it clearly.
The case is that class B or C has a lot a members to add to the list.
class B : public class A
{
public:
virtual void AddToList(CList &L);
private:
long m_var1;
long m_var2;
long m_var3;
}
|
|
|
|
|
followait wrote: The case is that class B or C has a lot a members to add to the list.
 Ooookayy. That was not clear at all from the begining: you don't want to add the object itself to the list but the object should add some elements in the list. In that case, I would make it a member function of the class.
|
|
|
|
|
Class A only needs to have a method AddToList.
I am visualizing from example that List is going to contain only pointer of Class A. Then base class object can point derived class object so basically this method not need to make virtual.
Even though if list is going to have copy of object then need to implement copy constructor & assignment operator in each derived class to make appropriate copy.
But from my point of view class responsibilities should not be mixed up it should be some controller class responsibility to collect object in list rather then Class A itself. So writing a separate method in any other class is better from design-wise. Or make a another class derived by CList and create a method e.g say: "CollectDataInList(A *a)" or have this method in class which is containing CList object.
Akash
|
|
|
|
|
Hi People,
I have an ACE placed on a user control.
If no matching results are found for the criteria supplied I display --> No match found. Please select the item that matches your field of expertise.
This is a bit late and maybe less intuitive.
What I would like is to place a non selectable item in the result list at the very bottom. It would say --> Please select the item that matches your field of expertise.
The result list would then look like this below.
Finance
Clerk
Developer
Trader
Retail
etc
--------------------
Please select the item that matches your field of expertise
Is this possible at all?
Many thanks
|
|
|
|
|
wantToLearn wrote: Is this possible at all?
Yes.
|
|
|
|
|
We are developing a smart client application which consists of shell application with Deck workspace.
Hence we are showing all the UI elements (views) in an overlapped manner. In a particular view, we have an explorer bar which loads group of text editors and combo box editors controls and Save/Cancel buttons. I have given the tab order in a sequence (0 1 2…). But while traversing through that particular view, have noticed that after cancel button the tab order going to next view instead of focusing the initial field in the same view (page).
Thank you.
Regards,
Divya
|
|
|
|
|
Welcome to CodeProject ![Beer | [beer]](https://www.codeproject.com/script/Forums/Images/beer.gif)
Take some time to look through the forum list and try to find one that more specifically matches your problem. It seems your problem has nothing to do with Design or Architecture. Many of the forums are specific to languages and/or platforms and it seems that is the problem you are dealing with, how the tab control works and perhaps the IDE you are working with.
led mike
|
|
|
|
General 
News 
Suggestion 
Question 
Bug 
Answer 
Joke 
Praise 
Rant 
Admin 
Submit an article or tip
