HAL: Do you ever put implementation code in abstract methods?

[dmurray]

I have a fundamental question on HALs and abstract methods that I'd like clarified, please. In my HAL's up to this point I've been using a parent with abstract methods, implementing concrete methods in the concrete product(s) (see here for example), and using dynamic dispatch as appropriate to execute the concrete method. I think this is a standard technique and it works fine for me.

 

Now I have another layer of abstraction I want to implement, but in this case the concrete products share a lot of functionality and often I can get away with just executing the same method for each product. Should I just go ahead and implement this logic in the parent method? If I do, that means I no longer have abstract methods in my parent (right?), so am I breaking some 'good-practice' rules in OOP design? 

 

Or am I just getting bogged down in semantics? 

 

[by the way, I have loads of little niggling questions like these which I may ask as I'm going along in my OOP learning - hope you don't mind.] 

[ak_nz]

You aren't really "breaking" any good-practice rules in OOP design. There is nothing inherently wrong with have common functionality in an abstract parent - as long as you are comfortable that this will always reasonably be the case for all child implementations. If you come up against scenarios where children are over-riding methods just to inhibit parent behavior then you are definitely seeing a code smell - your hierarchy is making assumptions about behavior that are not true for all children.

 

One of the most important guidelines for OOP design is to favor composition over inheritance for exactly this reason - inheritance enforces behavior that isn't always appropriate. The guideline encourages small, focused hierarchies of classes and then using composition where needed to add functionality without impacting the hierarchy.

 

Unfortunately OOP development in LabVIEW is cumbersome compared to other languages, so we can all be forgiven for taking a short-cut where it is simply more expedient.

 

EDIT: If you are new to OOP development, I encourage you to read through these if you haven't already:

• http://butunclebob.com/ArticleS.UncleBob.PrinciplesOfOod
• http://javarevisited.blogspot.co.nz/2012/03/10-object-oriented-design-principles.html

Edited November 16, 2014 by ak_nz

[drjdpowell]

I would say putting common logic in the parent is desirable behavior.   And I don’t mind if some children will want to override the default parent behavior.  For example, I have a project at the moment that uses cameras and “lockinâ€, the extraction of a periodic signal in the image.  Most concrete camera types just override the "Get Image" method, and the parent’s "Get Lockin Image†method uses "Get Imageâ€.  But  one camera model has a built-in lockin mode, so I override “Get Lockin Image†in that case.

[dmurray]

Thanks for the replies, guys. (Thanks also for the links, ak_nz). That (mostly) clarifies things for me. ak_nz, I won't have a problem with children restricting the parent functionality if I do override the method at some point - I guess you're talking about LSP, right?

 

One final question on this: once I put code in my abstract parent method, is it technically no longer called "abstract"? I can see that it doesn't matter functionally, but I just want to get my terminology right at this early stage. 

[ak_nz]

Thanks for the replies, guys. (Thanks also for the links, ak_nz). That (mostly) clarifies things for me. ak_nz, I won't have a problem with children restricting the parent functionality if I do override the method at some point - I guess you're talking about LSP, right?

 

One final question on this: once I put code in my abstract parent method, is it technically no longer called "abstract"? I can see that it doesn't matter functionally, but I just want to get my terminology right at this early stage. 

 

You can have an abstract class that implements some concrete logic. Since it is functionally irrelevant from a compiler perspective the important thing is to communicate the intent of the class to other developers. One technique I use in this scenario is to alter the class icon to a wire-frame cube (rather than a full cube) using GDS which indicates that the class is not intended to be instantiated but exists only to specify a common interface for concrete children and implement some common behavior. 

[JKSH]

As ak_nz and drjdpowell have both said, it is perfectly fine to implement concrete code in an abstract class. This lets you make full use of inheritance to keep your project tidy and organized.

 

 

If you come up against scenarios where children are over-riding methods just to inhibit parent behavior then you are definitely seeing a code smell - your hierarchy is making assumptions about behavior that are not true for all children.

I would say putting common logic in the parent is desirable behavior.   And I don’t mind if some children will want to override the default parent behavior.

 

"Rules" and "best practices" are established to guide you down a tried-and-true path, to maximize your chances of succeeding the first time you try. However, a set of rules will never cover all possibilities so simply following rules will not guarantee you success.

 

As the software designer/architect, it is your job to decide whether or not it makes sense to follow a rule in a particular part of your project. Referring to the two quotes above, I'd say: "It's ok to inhibit parent behaviour if 2 out of 20 children need to do it, but not if 18 out of 20 need to do it. If 10 out of 20 need to do it, well... make a judgement call."

 

 

once I put code in my abstract parent method, is it technically no longer called "abstract"? I can see that it doesn't matter functionally, but I just want to get my terminology right at this early stage. 

 

I'll answer your question in multiple levels.

 

Level 1: The definition of "Abstract"

In OOP, an abstract method is defined as a method which has no implementation. Its purpose is to define an interface for children to follow, and it is the children's job to provide the implementation.

 

An abstract class is a class that contains one or more abstract methods. An abstract class can contain concrete methods too.

 

So yes, if you put code in your parent method, then the parent method is concrete, not abstract. However, this in itself doesn't make the parent class non-abstract.

 

 

Level 2: The (lack of) necessity of "Abstract"

From your original question, I gather that you're wondering if you need to keep your parent "abstract" in order to abide by some set of best practices, and your latest post suggests that you're no longer worried about this.

 

To reaffirm: Many examples of inheritance show you an abstract common ancestor implemented by multiple children. However, there are no rules that say you must have an abstract common ancestor. You are free to have a fully concrete (and usable) class as your common ancestor if it makes sense to do so.

 

 

Level 3: The (lack of the) concept of "Abstract" in LabVIEW

Since you're interested in formal definitions and terminologies, I'll throw this in: LabVIEW does not support abstract methods in the traditional sense (yet?).

 

As I said in Level 1, an abstract method has no implementation by definition. However, what people commonly refer to as "abstract VI"s do in fact have an implementation: This implementation simply does nothing. "Having no implementation" is not the same as "Having an implementation that does nothing". LabVIEW doesn't let you create a VI with no implementation.

 

In other languages, your program cannot be compiled if you inherit from an abstract class but don't provide implementations for all of its abstract methods. This can be a useful way to remind developers to implement them. LabVIEW doesn't do this -- a LabVIEW program will happily run even if you call an "abstract VI". To work around this, what I did in a recent project was to implement pop-up dialogs in my "abstract VIs". This way, I'll get a message at runtime if I forget to implement something in a new child class a few months down the track.

 

EDIT: LabVIEW programmers fulfil the purpose of abstract functions by doing 2 things: (i) Have a VI that does nothing, and/or (ii) select the "Require descendent classes to override this dynamic dispatch VI" option in the class properties dialog.

 

See this post for more: http://stackoverflow.com/questions/391483/what-is-the-difference-between-an-abstract-function-and-a-virtual-function (Note that NI explicitly rejected the term "virtual" because it's confusing to talk about non-virtual-Virtual-Instruments, so they picked "dynamic dispatch" instead)

 

At the end of the day, knowing terminologies is useful (and important!) when discussing concepts, but it shouldn't affect how you design your project.

 

All the best!

Edited November 18, 2014 by JKSH

[smithd]

 To work around this, what I did in a recent project was to implement pop-up dialogs in my "abstract VIs". This way, I'll get a message at runtime if I forget to implement something in a new child class a few months down the track.

 

I've found this to be critical, even for functions where I require overrides. There are situations (for example if you are debugging and add a conditional disable but forget to wire things through, or if you do something similar with a for loop that runs 0 times) where a class wire can get invalidated. If the wire type is that of the parent class (common), you can still get an instance of that parent class and you will never know whats going on unless you throw an error. Its pretty horrifying to realize you've spent half a day debugging a billion reentrant VIs on cRIO targets just because you dropped down one disable structure and forgot to wire through the enabled case.

 

tl;dr: throw errors in parent functions which shouldn't be called. Its a good idea.

[shoneill]

The idea with the dialogs is good assuming you don't have the option active that any overrides of the abstract method are required to call the parent method!

[ak_nz]

As ak_nz and drjdpowell have both said, it is perfectly fine to implement concrete code in an abstract class. This lets you make full use of inheritance to keep your project tidy and organized.

 

 

 

"Rules" and "best practices" are established to guide you down a tried-and-true path, to maximize your chances of succeeding the first time you try. However, a set of rules will never cover all possibilities so simply following rules will not guarantee you success.

 

As the software designer/architect, it is your job to decide whether or not it makes sense to follow a rule in a particular part of your project. Referring to the two quotes above, I'd say: "It's ok to inhibit parent behaviour if 2 out of 20 children need to do it, but not if 18 out of 20 need to do it. If 10 out of 20 need to do it, well... make a judgement call."

 

Yes, this is an interesting design problem that inevitably ends with a different answer for each developer - what percentage deviation from LSP are you comfortable with? It is a slippery slope - so be careful. And document the limitations so that the next developer to follow you understands the trade-offs you made - they might follow your example in a way you didn't expect. As the quote goes:

 

“Code as if the next guy to maintain your code is a homicidal maniac who knows where you live.â€

 

 

 

tl;dr: throw errors in parent functions which shouldn't be called. Its a good idea.

 

I do this for hierarchies where a child class doesn't support a particular function - the default parent behavior is to generate an "Unsupported Operation" error. Still I personally only do this if I think the number of operations or number of classes in the hierarchy won't really increase (ie. percentage deviation from LSP is small). If I find that I have multiple levels of undo in multiple operations in a hierarchy I wonder how maintainable this codebase will be by the worst critic - another independent developer.

Edited November 18, 2014 by ak_nz

[dmurray]

Thank for the further insights, in particular JKSH for your post. I think your link to stackoverflow.com summarizes it for me as regards my OP:

 

An abstract function can have no functionality. You're basically saying, any child class MUST give their own version of this method, however it's too general to even try to implement in the parent class.

A virtual function, is basically saying look, here's the functionality that may or may not be good enough for the child class. So if it is good enough, use this method, if not, then override me, and provide your own functionality.

 

I'm a little bit wary about the comments regarding children inhibiting parent behavior, as I thought that definitely was bad practice. But I haven't come across any situation where I need to do that so I won't worry about that for now anyway. Thanks also for the insights in throwing exceptions when a method won't be over-ridden. That's something I was thinking about, so you've saved me a question. Cheers, all.