>> [Comp Sci Dept]
>> [R J Botting]
>> [New Bibliographic Items]
|| [Site Search]
|| [Bibliography Search]
Mon May 16 15:19:05 PDT 2005
- Nicholas E Gold & Andrew M Mohan & Paul J Layzell
- Spatial Complexity Metrics: An Investigation of Utility
- IEEE Trans Software Engineering V31n3(Mar 2005)pp203-212
- =EXPERIENCE METRICS 30 COBOL TECHNICAL LoC
- Spatial Complexity Metrics allow for the distance between references to identifiers.
- Two out of three such metrics are no better than Lines of Code at predicting complexity on this sample.
- Sample is of single file COBOL programs, many versions.
- A metric worth study: Douce Basic(1999). Sum[each call] (number of lines of code from call to definition)
- Hakan Erdogmas & Maurizio Morisio & Marco Torchiano
- On the Effectiveness of the Test-First Approach to Programming
- IEEE Trans Software Engineering V31n3(Mar 2005)pp226-237
- =EXPERIMENT TESTING AGILE TFD TDD
- Summarizes previous studies: test-first makes no difference or improves quality and may or may not improve productivity.
- This paper has a controlled experiment comparing test-first with test-last development of a series of user stories.
- Test-first programmers broke stories into a series of tests and added one test and then added code to make program pass test. Test last coded complete story and then carried out all tests.
- Experiment: 40PCs Java JUnit ECLIPSE CVS, 35 3rd year volunteer students, 11 dropped out.
- Quality measure via hidden acceptance tests.
- Productivity in terms of number of stories per unit time.
- Theories: Test-first may tend to increase the number of tests, the quality, and productivity. However more tests anyway may increase quality and productivity.
- My translation of their statistics Interesting/significant Results:
- quality >= 0.55 + tests/10. NOTE: not a linear equation but an inequality.
- More tests increase the minimum number of test passed in the acceptance testing. Other factors can improve quality as well as tests.
- productivity= 0.255 + 0.659 tests.
- Test-first programmers did more tests on average.
- Test-first programmers where more productive and produced the same quality code.
- 30% dropped out -- "morbidity"