Programming, website development forum

DaniWeb Programming
and Web Development Community

Revenue Sharing

Welcome to TechnicalTalk.Net forum! It seems that you are not a member yet. To interact with other members and to post on the forum. Get a free account now

Earn money on this forum

Advertise with us...

TechnicalTalk » Programming and Web Development » Other programming languages » Formal Specification and Property Based Testing

Quick Links:

Want to earn money on forum? visit --> Earn money on forum

How do you unit test your software?

Poll - This thread contains a Poll. First mark your answer from the options and then reply with your justification.

Question:

How do you unit test your software?

JUnit (xUnit) Tests	1 (50%)
Ad Hoc (self defined) Tests	0 (0%)
Property Based Test	1 (50%)
I dont test my code	0 (0%)

Total Voters: 2

Pages: [1] Go Down

		Email this topic \| Print
0 Members and 1 Guest are viewing this topic.

Formal Specification and Property Based Testing

xplo99

Reputation Power: 1

New Poster

Posts: 2

Topics: 1

Formal Specification and Property Based Testing

September 02, 2010, 09:31:34 PM

I have a question that has been bugging me ever since I was asked to write unit tests using a specific (and popular framework)? What am I testing for. "Well, I am testing to see whether my software is correct". Yes, but what is your definition of correct?

I am not going to write a paper here but I hope you can see what I am getting at. Researching this topic will lead to the subject of Formal Specification and Property Based Testing which is a method of testing software against properties extracted from formal specification. I was wondering how many of you are familiar with this testing methodology. How may of you test using an xUnit based framework? And what are some or your thoughts are on the subject?

polas

Reputation Power: 17

Code Guru

Gender:

Posts: 1403

Topics: 86

Re: Formal Specification and Property Based Testing

September 06, 2010, 12:16:05 PM

An excellent topic - probably the best one I have seen for a long time

I think of correctness in terms of the formal (axiomatic) definition. "Partial Correctness" guarantees that the software (or unit being tested) behaves correctly, "total correctness" guarantees partial correctness (it behaves correctly) AND that it terminates.

I have used JUnit quite a lot - good for unit testing but often you need other testing too to test the integration of all the units of functionality together rather than just unit testing.

I have done a reasonable amount of formal specification before (memories of Z), it is OK but I don't think as useful in the "real world" practical application

Mesham Type Oriented Parallel Programming Language, Free online technical support

xplo99

Reputation Power: 1

New Poster

Posts: 2

Topics: 1

Re: Formal Specification and Property Based Testing

September 06, 2010, 04:09:18 PM

After digging a little deeper it seemed, at first, that I was thinking about testing all wrong. It seemed that the main goal of testing is to find errors, not to verify anything is correct (Kaner, 1988). With current computing technology it is impossible to show a program to be correct. There are too many combinations of input, paths, etc. to be tested even for a small "unit" of code.

While, I have not the experience to formulate too forceful of an opinion -- I do not believe this to be entirely true. I agree somewhat with polas and "Partial Correctness". For example, say that I have a sorting algorithm that I wish to test. I know that sorting a list twice should yield the same result as sorting it once. This property is called indempotency (O'Sullivan et. al. , 2008) and should be true for a class of operators that includes at least some sorting routines. So if I write a sorting algorithm then I can reason about a test as follow:

sort(myList) == sort(sort(myList)); //Test for idempotentcy

If this holds for all lists of order-able objects then we may say that our sorting routine, at very least, has the property of idempotentcy as it should. Now we can't test for all lists, as stated above, but with a nice random object generator we can test lots of them. Tests like this can help when refactoring.

This is a kind of test for correctness. I think that it yields valuable results. However there are a multitude of other reasons why it is better to test for errors and bugs than to test for correctness. Psychological studies dictates that you will not find as many bugs if you think you are testing for correctness as opposed to searching for errors (Kaner, 1988). In fact, who wants to tear apart their own code -- so to speak? Subconsciously, the programmer does not want his code to fail so he will not trash his code with vigorous testing. This is why I think JUnit like frameworks are not very useful for "self" testing. I think xUnit frameworks are best in the hands of a dedicated tester who is thinking objectively. I think that to test your own code programmers should use property based test like the one shown above that require large quantities of randomly selected values.

Anyway, I wonder what everyone else thinks.

Email this topic | Print

Pages: [1] Go Up

Jump to:

Home | Advertise | Forum Help | Privacy Policy | Contact us

Powered by SMF 1.1.16 | SMF © 2011, Simple Machines

Google visited last this page May 05, 2012, 08:09:11 AM