Do Code Clones Matter? Software Engineering Seminar Spring Semester - - PowerPoint PPT Presentation

Do Code Clones Matter?

Software Engineering Seminar Spring Semester 2010 Dan Tecu

09th of March 2 Do Code Clones Matter?

Do Code Clones Matter?

Juergens, E.; Deissenboeck, F.; Hummel, B.; Wagner, S. ICSE'09, May 16-24, 2009, Vancouver, Canada

Paper presented today:

09th of March 3 Do Code Clones Matter?

What is code cloning?

• Code cloning is a primitive form of code reuse, usually by

copying (copy-paste)

• Code cloning may happen intentionally or unintentionally

09th of March 4 Do Code Clones Matter?

Clones and possible consequences

• Clones introduce redundancy and might introduce

dependencies

• Inconsistent changes to cloned code might create faults

and might lead to incorrect behavior

• Inconsistent bug fixing: when a bug found in cloned code is

fixed in only one clone

09th of March 5 Do Code Clones Matter?

Previous work

• Most previous work agrees that cloning poses a problem

for software maintenance [Lague et al. 1997]

• However, there is little information available concerning the

impact of code cloning on software quality

• Some researchers even started to doubt that code cloning

is harmful [Krinke, 2007]

09th of March 6 Do Code Clones Matter?

Research problem

• This paper purpose is to shed some light in this field
• It presents the results of a large case study that was

undertaken to find out:

• 1. whether the clones are changed inconsistently
• 2. whether the inconsistencies are introduced intentionally
• 3. whether the unintentional inconsistencies can represent faults

09th of March 7 Do Code Clones Matter?

Contribution

• A novel suffix-tree based algorithm for detection of

inconsistent clones

• An empirical study showing whether the cloned code is

harmful or not

09th of March 8 Do Code Clones Matter?

Detection of inconsistent clones

• Works on token level
• Identifier names are irrelevant (due to the normalizer)
• Clone groups whose clones overlap with each other are

filtered out

09th of March 9 Do Code Clones Matter?

Detection in detail: suffix trees

• A suffix tree for the word: “wood”
• A path from root to a leaf

describes (uniquely) a suffix; the tree describes all suffixes

• The circles represent nodes, the

rectangles represent leaves

• Each edge to a node is labeled

with a substring of the initial word

• Each edge to a leaf is labeled

with the empty string wood d

• d
• d

09th of March 10 Do Code Clones Matter?

Detection in detail: edit distance

• The edit distance between 2 strings is the minimum

number of operations required to change one string into another

• The allowed operations are:
• 1. Deletion of a character
• 2. Insertion of a character
• 3. Changing a character into another
• The edit distance between “wood” and “floor” is 3

09th of March 11 Do Code Clones Matter?

Detection in detail: the algorithm

• Input parameters: the sequence (composed of n tokens),

maximum edit distance and minimal clone length

• The suffix tree over the input sequence is constructed
• For each suffix an approximate search based on the edit

distance is performed in the tree

09th of March 12 Do Code Clones Matter?

Performance of algorithm

• Worst case complexity is hard to analyze
• Results on Intel Core 2 Duo 2.4 GHz, 3.5 GB RAM, running

Java in a single thread are shown below:

09th of March 13 Do Code Clones Matter?

Study description: study objects

• 5 projects described below, were analyzed:

System Organization Language Age (years) Size (kLOC) A Munich Re C# 6 317 B Munich Re C# 4 454 C Munich Re C# 2 495 D LV 1871 Cobol 17 197 Sysiphus TUM Java 8 281

09th of March 14 Do Code Clones Matter?

Study description: research questions

• RQ1: Are clones changed inconsistently?
• RQ2: Are inconsistent clones created unintentionally?
• RQ3: Can inconsistent clones be indicators for faults in real

systems?

09th of March 15 Do Code Clones Matter?

Study measurements: parameters tuning

• Minimal clone length: 10 statements (for Cobol: 20)
• Maximum edit distance: 5 (for Cobol: 10)
• Maximal inconsistency ratio (the ratio of edit distance and

clone length): 0.2

• Additional constraint: the first 2 statements of two clones

need to be equal

09th of March 16 Do Code Clones Matter?

Study measurements: absolute numbers

09th of March 17 Do Code Clones Matter?

Study measurements: relative numbers

• RQ1 mean value: 0.52
• RQ2 mean value: 0.28
• RQ3 mean value: 0.15

09th of March 18 Do Code Clones Matter?

Fault density

• To answer RQ3, we have to compare the fault density in

inconsistencies against the average fault density

• Fault density in inconsistencies was evaluated in faults/

kLOC

• But, the average fault density in the analyzed systems was

not known

• Typical range for fault density is: 0.1 – 50 faults/kLOC

[Endres and Rombach, 2003]

09th of March 19 Do Code Clones Matter?

RQ3 answered

• Average fault density in inconsistencies: 48.1 faults/kLOC

09th of March 20 Do Code Clones Matter?

Threats to validity

• The development repositories of the systems were not

analyzed (to trace the evolution of inconsistencies)

• Comparison with the actual fault density would have been

better

• The analyzed projects were not sampled randomly
• Majority of the systems is written in C#
• Only 5 systems were analyzed

09th of March 21 Do Code Clones Matter?

Conclusion

• The answer to RQ1 is positive: clones are changed

inconsistently

• The answer to RQ2 is positive: inconsistent clones are

created unintentionally

• The answer to RQ3 is also positive: the average fault in

inconsistent clones is very close to the upper bound of the reported average fault density

• Inconsistent clones can be indicators for faults in real

systems

09th of March 22 Do Code Clones Matter?

References

• A. Endres and D. Rombach. A Handbook of Software and

Systems Engineering. Pearson 2003

• J. Krinke. A study of consistent and inconsistent changes

to code clones. In Proc. WCRE´07. IEEE, 2007

• B. Lague, D. Proulx, J. Mayrand, E. M. Merlo and J.
• Hudepohl. Assessing the benefits of incorporating function

clone detection in a development process. In Proc. ICSM ´97. IEEE, 1997

09th of March 23 Do Code Clones Matter?

Detection in detail: detect procedure

09th of March 24 Do Code Clones Matter?

Detection in detail: search procedure