27 no 1 january 2001 gm mh i c mc e mh e 1fdatec

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: a of aracters in the widths of the windows effects of changes are ªdampedº and attenuate over time, One motivation and complexity is highly informative.) Tuesday, highly variable stand terms such as sizefor the form used in (6) is to distinguish September 10, 13 age is • Today’s Agenda • • • • Tuesday, September 10, 13 Code decay paper results Four types of threats to validity Quiz 4 Empirical study paper by Kemerer and Slaughter Results: (1) The span of changes increases over time? 8 Tuesday, September 10, 13 IEEE TRANSACTIONS ON SOFTWARE ENGINEERING, VOL. 27, NO. 1, JANUARY 2001 G…m; mH ; I† ˆ c > m;c e > mH e 1fDATE…c† P Ig Results: (2) Breakdown of modularity increases over time? ber of MRs touching both m and mH . In the upper the heads show this network view using all the ta through t he end of calendar year 1988 ding to one choice of I ), while the tails of the display the same view at the end of 1987 (an ice of I ). Cway, one can AGEMENT DATA E FROM CHANGE MANsee how relationships among volved through time. In the top panel of Fig. 4, s of wider wo main clusters of roughly a dozen modules e network ever, in the center panel, which displays the ps display a through ows us to the end of 1989 in the locations of the s 1988 data appear here as the locations of the breaking n which .e4, dpoints), these two clusters have mostly e merging process continued and at the end of upper are no longer visible. While the logic of the sters left tions of theintended to see to it that some modules riginally anner that essentially independent of each other, new and n changed ted functionality may have helped to destroy this r of nce. times. ms of the ights in (7) constitute a compromise between odules m nts 9 Fig. 4. Top: NicheWorks view of the modules in one subsystem change data through 1988 to place modules which have been ch at the same time close to one another. Two clusters of modu evident; a module within one of these clusters is often changed to with other modules in the cluster but not with other modules. NicheWorks view of the modules in the top left, this time incorp the change history through 1989. The clusters that appear in the view are converging in on each other. This suggests that the arch that was previously successful in separating the functionality of c l u s t e r s o f m o du l e s i s b r e a k i n g d o w n. B o t t om : T h e b r e a continued and at the end of 1996 there was no suggestion of m clusters of modules. If modules have changed together as a part of the same MR, they were placed to close to each other. w…m; mH † ˆ CHNG…m; mH ; I†;  ; …7† I†to place too close together pairs of modules that d together frequently only because they are here rge numbers of times in total and Tuesday, September 10, 13 P Ig Prmodules' cIsange fhistorys Tf eCDIs t thatodels tpredicted edictive CD h ar e unction . oh bes m quan ify responses to code colored boxes represent frequently changed imp by incl did not ther mod le size o f a source code line. Rainbowdecay. Here we summarize research files, rove modulesuding eicontributeuto fault po d in linking faults or faults using and are intended to decay, module symptoms of risk software to symptomschanges to the measures (metrics) of software complexity (which numbers in the factors numbers of of code predict this suggests that the decline of mod ce their creation. Results: (3) Fault potential, the likelihood of changes to induce faults increases over time the key past, the datesFPWTD of changes (i.e., of negative using the predictive CDIseffort, interval,and FPGLMthe (5). dataofare correlated essentially completely wit in the responses of of these (4) and quality. We Section 5.2 may not be harmful, but sin More complete measured of this fault potentialqualityas in (4): present ages, discussion in two dealing their sizes, and Thus, changes to code are more responsible fo their three such indices, years), and with modeling their ses appears Fault Potenthis are discussed more thoroughly in 4.2.5with effort, In tial work, the authors counted the than the is correlated withcode. span, it is mo...
View Full Document

This document was uploaded on 03/16/2014 for the course EE 360f at University of Texas at Austin.

Ask a homework question - tutors are online