Software Engineering (phần 13) ppt

A deri ved class now is def ined, that is, a cl ass identical to the base class in many ' but new features m ay be added and existi ng features renamed,reim plemented, suppressed, or cha

j '

' ;

f

. f unet ion open-di sk-fi l e funeti on open-tape-f i e f unct i on open-di sket t e-f i e

I

I

l

I

I

Di skFi l ecl ass TapeFil ecl ass Di sketteFi l ecl aes

Im plem entation of Imptementation of lmplementation of

tbr a disk file fofa tape file for a diskette file :

( b)

Flgure I#.a Def i ni ii on of base cl ass Fi l ecl oss wi t h deri ved cl ass Dî skFi l ecl css, T opeFi l ecbss, . and Di ske- Fi l ecl oss.

. A deri ved class now is def ined, that is, a cl ass identical to the base class in many

' but new features m ay be added and existi ng features renamed,reim plemented,

suppressed, or changed in other ways Furthermore, these changes m ay be made

j w i thout havi ng an effect on the base class or any other deri ved cl asses However,

) suppose now t hat the base class itself is changed lf t his happens, all deri ved cl asses

1 are changed in t he same way In other words, t he strength of i nherit ance i s that new

, l eaves can be added t o the inheritance tree (or graph, if the impl ementation l anguage

l supoorts m ulti ole inheri tance, as C++ does) w it hout alt erinz anv other cl ass in the

j tree But if an interi or node of the tree is changed in any way, then this change is

. propagated to al1 i descendants (the ii fragil e class probl em'' ) Thus, inheritance is

! another feat ure of object-ori ent ed technol ogy that can have a major positi ve i n:uence

1

on devel opment but a negative i mpact on maint enance.

!

J

p :v:kopM :NT SKILLS

Earli er in this chapter m uch w as sai d about t he skill s needed for m aintenance Fof

,the abi li ty to determ inethe cause of afail ure of alarge product '

i was deem edessenti al But t hi s skil l is notneeded exclusively forproduct m aintenance

!

1

l

.)

I

-k ',

I h

'

'

' I l y RE VERSE ENGI NEERI NG sos I j lt i

I .

: l i s used throughout integration and product t esting. Anot her vi tal skil l i s the abi lity 1

I

. t o funct i m ef f ect i vel y wi t hout adequat e document at i on Agai n, t he document at i on j

' r ar el y icompl et e whi l e i nt egr at i on and product es t ng ar e under way Al so st r es sed j ' 1

j kills w ith regard to specihcation, design, implem entation and integration, 1 ' was that s

and t es ti ng are essenti al for adapti ve and perf ect i ve mai nt enanee These act i vi t es al so i ( 1

' are carr e ou i d t during the devel opm ent process, and each requires specialized ski lls ' l

.

' ln other words, the skil ls a maintenance program m er needs are di ftkrent i n no way I i

j

from those needed by software professionals specializing in other phases of softw are l '

1 producti on.The key point is that a m aintenance program m er m ust not be m erely

skill ed i n a broad variety of areas but highl y skill ed i n 6 111 those areas Although t he j

'

as desi gn or testing, the software m ai ntainer m ust be a special ist in virtually every 1

1

!

1

l

;

! i 1ê T R EV KRSE EN / IN EERIN / ; 7 i l

.1 1

p

E

As has been pointed out,som etim es the only documentati on available tor m ai nte- l

11

nan ce s t he our c code t e f ( Thi s happe ns aI l oo r eque nt l when ma i nt a ni ng 2 ! ! h

Iegacy' A'.v, s /c/ z? ç, that is, software in current use but devel oped some l 5 or 20 years ago, i

if not earlien) Under t hese circumstances, m aint aining t he code can be ext rem el y

dif-b hcult O ne w ay of handling this problem is t o start with the source code and att em pt '

! I

t to recreate the design documents or even the speeihcations.Thi s proeess is call ed , !

r CASE t ool s can assi s wi t h t hi s proeess One of he si mpl est i s a pr et t y pri nt er ' j

!

(Secti on 5 6), which may help displ ay the code more clearly Other t ools construct l k l 1

-di

agr ams, such as owchar t or UML di agrams, di rect l y t r om t he sour ee code' hes e , j i

1

: Once t he m aintenance team has reconstruct ed the design, t here are two possibil- : 1

; i ti es. One alt ernative i s to at tempt to reconstruct t he specif ications, m odify the recon- ! ' )

-j ; struct ed specif icatit m s to renect the necessary changes,and reimplement the product l '

l

the usual way (W i thin the context of reverse engineering, the usual developm ent j j

process that proceeds from specif icati ons through design to code is call ed jbrward j i

engi neering The process of reverse engineering foll owed by forw ard engineeri ng j

som etimes is called reengi neering.) In practice, reconstruction of the specil ications 1 1,

' l t he reconstructed design is modihed and 1

is an ext remely hard task.M ore trequent y, t l 1 , the modil ied design then is forward engineered.

A related activi ty often pert brmed during the m aint enance phase is restructuring I '

Reverse engineeri ng takes the product from a lower level of abstracti on to a hi gher ! 1

level ot abstraction, tor exam pl e, from code to desi gn Forward engi neeri ng takes the I

hl m a hi gher v el of abs t ct i on o a l ower ev el Res t u ct ur i g howev er , ' )

!

(

l

5@* t H A p T E R 44 * M oinfenante Phase

takes plaee at the same level lt is the process of improving the product wi thout changing its functionality Pretty printing is one form of restructuring' , so is convert i ng

) code f r om unst r uct ur ed t o st r uct ured f orm I n gener al , rest r uct uri ng i per formed t

t make t he sour ce code ( or desi gn or even t he dat abase) easi er o mai nt ai n '

A worse sit uati on occurs i f the source code is lost and the execut able versi on of the product i s all that i s available At f irst si ght it m ight seem t hat the only possi bl e !

l way to recreat e the source code is to use a disassembler t o creat e assembl er code

i

j then buil d a tool (that mi ght be termed a reverse compi ler) to try t o recreat e the

' Or g i inal high-l evel l anguage code A number of virtuall y i nsurmountable problems

accom pany t his approach'

l The nam es of the vari abl es w il l have been l ost as a consequence of t he origi nal com p

2 M any compi lers optimize the code in some w ay, m aking i t extrem ely difhcul t att em pt to recreat e the source code.

3 A construct such as a l oop in the assembler coul d correspond to a num ber of different possible constructs i n the source code.

: ln practi ce, therefore, t he exist ing product is t reated as a black box and reverse

engi neering is used t o deduce the specifications from the behavi or of the current product The reconstructed speci hcati ons are modified as required, and a new versi on

of the product is forward engineered from those specihcati ons.

'!

l

1 W hil e the product is being devel oped, m any m em bers of t he development t eam have

!

! a broad overview of the product as a whole, but as a result of t he rapid personnel

l turnover in the computer industrys it is unlikely that members of the maintenance t eam

;

j during t he m aintenance phase will have been i nvolved in the original devel opment.

j Theref ore, he mai nt ai ner ends o see t he pr oduct as a set of l oosel y r el at ed modul es

1 and generall y is not aw are t hat a change to one modul e may seriously affect one or

1 mor e ot her modul es and hence t he pr oduct as a whol e Even i f t he mai nt ai ner wi shed

l

to understand every aspect of the product, the pressures to tix or to extend the product generall y are such that no time is al lowed for the detailed study needed to achieve this Furtherm ore in many cases, lit tle or no docum entation i s avai labl e to assi st i n ' gaining that understanding One w ay of trying to m inimize this di ffi cul ty is to use ,

rexression testi nc, that is testing the changed product against previous test cases t o ' ensure that it still works correctly.

, For this reason, i t is vi tal to store all test cases, together wi th their expect ed

outcom es in m achine-readable t brm A s a result of changes made t o t he product, certain stored test cases m ay have to be m odil ied For exam pl e, if sal ary w i thholdi ng percentages change as a consequence of tax legislati on, then the correct output from

j

l a payroll product for each test case involving wi thholding will change.Sim il arly, i f

1 sate ll ite observati ons lead to correcti tl ns in the latitude and longit ude of an island,t hen

! the correct output from a product that calcul ates the positi on ot an aircratt using the

1

i l i ' @ k

p

l ?

- 1

1.1* METRI CS FoR THE MAI NTENANCE PHM E 5@7 ' l ?

l !

j coordinates of that island will changecorrespondingl y Depending on the mai ntenance l I performed, some val id test cases will become invalid But the computations that need l '

t o be made t o correct t he stored test cases essentially are the same as woul d have to l

be made to set up new test data for checki ng that the maintenance has been correctly l i

perform ed

.No additi onal work therefore i s i nvolved in maintaining the f il e of test i 1 1

lt can be argued that regression testing is a wast e of tim e because regression l y

t es t g r equi r s he compl et e pr odu ct be et e t d a gai ns t hos t of es t cas es , mos t ) 1 1

of whi ch apparentl y have not hing t o do wi th the modul es modihed in the course of l 1 1

1 : product m aintenance.The word apparently i s cri tical in the previous sent ence The j. ! d

angers of unwi tting side effects of mai ntenance (t hat i s the introducti on of regression J i I

i, 2

-faults) are t oo great for that argument to hold wat er' , regressi on t esti ng is an essenti al l

1 : ! 1

'

( i

j

'

1 : j

1 1

: !

! l

l i unr easonabl e t o expect mai nt enance pr ogrammers to keep t ack manual l y of he r j

: various revision numbers and assi gn the next revision num ber each ti m e a modul e is r )

1

updated Unl ess the operating system incorporates versi on control, a version control j

: tool such as t he UNIX tools sccs (source code control system) (Rochkind, 19751, rcs i 1

'

(revision control syst em) (Ti chy, l 9851, or cvs (concurrent versi ons system) (Louki des j

t

.

revisi ons are updated appropriately A conf iguration control tool is needed A typical

l 1 exampl e of a commerci al ool CCC ( ehange and conf i gur at i on cont r ol ) Even i t he j j

at the very least a bt lil d tool must be used i n conjunction with a version cont rol t ool ! i

j ) A

not her cat egor y of CASE t ool s vi rt ual l y ess ent i al dur i ng t he mai nt enance phase i j j l

a f ul Sec t r t cki on ng 6.7 described some categori ool ha t ke e ps cor d of es of por CASE tools that can assist t ed a ul t not ye t i e d in reverse ( ! l ( l I engineering and reengi neeri ng Exam pl es of such tools that assist by creating visual 1

) di

spl ays of he st r uct ur e of t he pr oduct i ncl ude Bat t emap, Teamwork, and Bachman : j l R

eengineering Product Set.

M aintenance i s dif hcul t and t - rustrati ng The very l east that m anagem ent can do

i s to provide the m ai ntenance team with the t ool s needed for efficient and effeetive r

pr O

' k

1

j .1

1*

I l

î ) The acti viti es of the maintenance phase essent ial ly are specif ication

plement ation, int egration, t est ing, and docum ent at ion Therefore, t he met ri cs that j j

l .

1

soe t u A p T : R 11 * M oi ndenante Phose '

measure these acti viti es are equally appli cable to the m aint enance phase For

exam-pl e, he com pl exity metri cs of Sect ion 14 8 2 are relevant to maintenance, i n that a modul e w i th high com pl exi ty i s a likely candidate for inducing a regression faul t Particular care m ust be taken in m odifying such a modul e '

In additi on, m etri cs speci fic t o the mai ntenance phase incl ude m easures relati ng

to software fault reports, such as the tot al number of faul ts reported and cl assi hcati ons

of those faul ts by severit y and type ln addition i nform ati on regarding the current status of the faul t reports i s needed For exam pl e there is a considerable difference between having 13 cri tical faults reported and f ixed during 2002 and having onl y 2 criti cal faults reported during that year with nei ther of them f ixed.

A INTENA NtK P HASE

1

' A num ber of faults have been seeded in the source code of the case study.

1

'

Det ecting and correcting t hree of them is l eft as an exercise (Problems 16.1 1 '

$

i

A major chal lenge of the maintenance phase is expl ained in detail i n the Just in

Case You W anted to Know box on page 1 0 In bri efp cl assi cal softw are

engineer-i ng can be descri bed as deb t elolll nellt -tl l el l-mal ' ntel lance That is l irst the product is developed from scratch and deli vered t o t he cli ent From then on it is maintained This development-then-mai ntenance m odel is consistent wi th the det ini tions of IEEE Standard 610 2: A11 software producti on activi ties carri ed out before the software

is del ivered to the cl ient are consi dered to be devel opm ent and a1 l activi ties after

delivery consti tute maintenance IIEEE 610 2 19901.

The developm ent-then-m aint enance model is unrealisti c today:

l The cli ent' s requi rem ents frequentl y change before the product is deli vered.

'

; 2 Faul ts often have to be hxed before t he product is del ivered.

3 Developm ent from scratch i s becom ing i ncreasingl y rare Instead, much of t he ( product is buil t out of reused com ponents, and these reused com ponents fre-' quentl y need to be m odil ied before they can be utili zed i n t he new product.

For t hese three reasons a product al most al ways has to be modi fied before

de-Ii very, and the acti viti es performed when carrying out those modihcations are

indis-1 tingui shabl e from those of classi cal maintenance.

1 The r eal i ti es of mai nt enance ar e capt ur ed i n I SO/I EC St andar d 12207 whi ch

i st ates that the m aintenance process is activat ed when ç' software undergoes modi

-j '

I '

j ' FoR FURTHER READING 5@@ 1 :

I h

cations to code and associ ated documentation due to a problem or the need for j

improvement or adaptat on g ,

ook forward to the day when a greater proporti on ot t he sottware engineering j '

i t accepts t hat S: maint enance' ' is an acti vity performed throughout the soft- l 1 commun y

l 1 ware life cycl e.This will I ead to major i mprovement s, not just to the maintenance 1 , j

1 1

l 1 1

l I

;

I I

l i

' I

E

i .I @

ai nt enance i s an i mpor t ant and chal l engi ng s of t war e act i vi t y ( Sect i ons 16.1 and t j

16 2) This is ill ustrat ed by means of the case study of Section l 6 3 Issues relating t o 1 i

' i tenance are described (Section 16 4), i ncl udi ng the probl em ' 1 1

of repeat ed maint enance (Secti on 16 4 4) The mai ntenance of object-oriented soft- y !

ware is discussed in Secti on l.5. The skil ls t hat a m ai ntenance programm er needs j

:

are t he same as those of a devel oper' , the difference i s that a developer can specialize )

in One asl lect Of the software process, whereas t he maintainer must be an expert in all ( I j

aspects of software producti on (Section 16 6) A descripti on of reverse engineering ' 1

1

is gi ven in Section l

.7 Next foll ows a descripti on of testing during the mai ntenance j g

phase (Secti on 16 8) and CASE tools for mai ntenance (Secti on 16 9) M etri cs for the ! 1 1

m aintenance phase are di scussed in Section 1 6 0 M aint enance of the Air G ourm et case study, ci ted in Section l 6 1 is I eft as an exercise The chapter concl udes with

a discussion of the chall enges of the maintenance phase (Section l6 12).

1

Acl assi c s ource of i nf or mati on on mai nt enance i ( Li ent z and Swanson, 1 9801 Usef ul 2 1

material on maintenance can be found in (Longst reet, 19901 gBasili, 19901 presents I j j j

;

an int eresti ng vi ew of m aintenance as a reuse process.Regression testing is anal yzed j ! )

I ' :

i n (Rot hermel and Harrold, 19961 , predicting its cost effectiveness is descri bed i n i (Rosenblum and W eyuker 19971 Regression t esting in an i ndustrial environment is

iscussed in (Onoma, Tsai, Poonawala, and Suganuma, 19981 Planning of reengi- 2 d

neeri ng i s cover ed i n ( Sneed, l 9951, one of t he many ar t cl es on reengi neer i ng i n j

the January 1995 issue of IEEE Soft ware Costs and benefits of reengineering are i

di scuss ed i n ( Adol ph, 19961 ( Charet t e Adams , and Whi t e, 19971 des cri be r sk man- j

agement wi thin t he framework of maint enance And (von M ayrhauser and Vana, i 1

' 1.

19971 describe vari ous mechanisms of program comprehensi on for maintenance I

of large-scal e products A profi le of successful reengi neeri ng projects is given in ' (Teng Jeong, and Grover, 19981 M aint enance of information systems is descri bed in

(Bisbal, Lawless, W u, and Grimson, 19991 The use ot met ri cs wi thin the cont ext ot 1 ! maintainabili ty is di scussed in (Banker, Datar, Kemerer and Zwieg, l 993, and Henry, 1

q H

j '

I

51@ t u A p T : R 14 @ M oi nfenonie Phase

The impact on mai ntenance of t he use of obj ects is described i n gl- lenry and 16.1 Humphrey, 1 990, and M ancl and Havanas 1 9901 M aint enance of speciéc

object-ori ent ed products is descri bed i n (Lejter M eyers, and Rei ss, l 992, and W il de, j;.j Mat thews, and Hui tt, l 9931.

Papers on software m aintenance appear in t he M ay l 994 issue of Comm unications

q/' t he ACM The Jul y/August 1 998 issue of IEEE Stftware contains a number of ( arti cles on l egacy systems, especi all y gRugaber and W hite, l 9981 The proceedi ngs jw

.j

the annual Conference on Software M aint enance are a broadly based source of

int - ormati on on aIl aspects of m aintenance.

1f.1

!;.j

1d.1

P RO BU M S

i 16.1 W hy do you think that t he m ist ake i s frequently nlade of considering software m ai

n-tenance to be inferior to software development?

16.2 Consi der a product that determ ines whether a com puter is vi rus free.Describe why ( such a product i s I ikel y to have m ul tipl e variations of many of i ts modul es.W hat ar e

I the i m pli cations for maintenance? How can the resul ting probl em s be solved?

'

. 16.3 Repeat Problem l 6 2 for the automat ed l ibrary circul ati on syst em of Problem 8 7 '

:

j 16 4 Repeat Problem l 6 2 for t he product of Problem 8.8 that checks whether a bank ' .

stat em ent i s correct.

: 16.5 Repeat Problem 16.2 for t he automat ed tell er machine of Probl em 8 9.

i 1ô 6 You are the manager in charge of m ai ntenance in a Iarge software organizati on

.What

i qualiti es do you Iook for when hi ring new employees?

'

j

l 1 6.7 w hat are the implicati ons of m ai ntenance for a one-person software producti on

1 8 You have bee n asked t o build a computeri zed fault report fil e. W hat sort of data woul d

I you st ore in t he fil e? W hat sort of queries could be answered by your tool' ? W hat sort

1 of queri es coul d not be an swered by yt lur tool ?

1f.9 You recei ve a memo from the vi ce-presi dent for soft ware m aintenance of Ye Olde ;

Fashioned Software Corporation (Pr( )bIem 15 6) pointing out t hat, for the foreseeabl e

futures Olde Fashioned w i ll have to maintain t ens of m ill ions ofli nes of CO BO L code and aski ng your advice with regard to CASE tool s for such maint enance W hat do

16.10 (Term Project) The free accommodati on and transportati on provided by the Joini ng '

Children w i th their Fam il ies subcomm itt ee of Broadl ands Area Chil dren's Hospi tal

(Appendix A) currentl y is restri cted to chi ldren whose parents li ve withi n a 500-mi1 e I

j radius of Broadlands Now this restriction is t o be lifted W hat changes need t o be

i made to your product of Problem 15.10? Com pare your answer to the answer you

i

I gave to Probl em I l

jj j j j

.

j j

I l '

1ô.1 1 (Case Study) Correct the implementati on of Secti on l 5 3 so that once a seat has been ' 1

l assigned t o a passenger it cannot subsequently be assigned to another passenger 1

'

'

ever, there are pl aces in the impl ementation of Section l 5 3 where the program m er !

ë seem s to have t hought that there were only 12 kinds Fix the im plem entation appro- i

j .

I : j l

'

.13 (Case Study) The implementation of Section 15.13 does not check t hat, when t he user j l

! inputs m eal qual ity, the value is in the range 1 through 5 Fix this faul t l (

!

2

16.14 (Case Study) lmprove t he aestheti c appearance of a11 the reports i n the i mplementati on l 1

1 ' of section 15.13 by adjusti ng the ali gnment of the various components j ! ,

! 1 6 1 5 ( Case St udy) Repl ace t he menu- dr i ven i nput rout i nes n t he i mpl ement at i on of Sect i on ' ' !

I : 15

1& 1f (Readings in Software Engi neeri ng) Your i nstruct orwil l distri bute copi es of (Rugaber :: and Whi te, 19981 W hat do you see as the bi ggest challenges of mai ntai ning l egacy j j

'

l

'

j '

'

7

I CaS I COW it he Tar Pi t Reengi neer i ng a Legacy Sys t em, ' ' 1

! IEEE Softb vare 13 (May l 996) pp 4 1-47 : ! l I '

(Banker, Dat aq Kemerer, and Zwi eg, 1 9931 R D BANKER, S M DATAR, C F KEMERER, ' I AND D ZwIEG, t t soft ware Compl exi t y and Maintenance Costs, ' Communicati ons ofthe 5

ACM 36 (November 1993), pp 8 1-94 : 1

( Basi li

, 19901 V R BASILI , t ' Viewing Mai nt enance as Reuse-ori ent ed Software j j

(Bi sbal, Lawl ess, W u, and Gri msons 1 9991 J BISBAL, D LAw l - Ess, B W I J, J GRI MSON, '

tt l wegacy lnformati on Syst ems: lssues and Di recti onss ' IEEE Soft ware 16 l 1

g

(Sept ember/october l 999), pp I 03-1 1 l '

1 i tcharett e

,Adam s, and W hi t e, 19971 R N CHARETTE, K M ADAMS ANn M B W HITE j ) i

' Managing Risk in Sof tware Mai nt enance, ' IEEE s' ofrwwr c 14 (May/lune 1 997), 1 l l 1 i

gl - lenry and Humphr ey, 19901 S M HENRV AND M HUMPHREY, ' ' A Cont roll ed Experiment

:' t o Eval uat e Maintai nabil it y of Object-ori ented Software, '' Proceedings oft he IEEE k

! i

. Confer ence on sk// wt zrc Maintenance, San Di ego, CA, November 1 990, pp 258-65 j g

gl denry, Henry, Kafur a, and M atheson 19941 J HENRY, S HENRY, D KAFI JRA, AND '

L M A' rHssoN, ' ' lmprovi ng Soft ware M ai nt enance at M arti n M ariett a, ' IEEE Softbvare I

11 (JuI y 19

1

(IEEE 6 l0 2, 9901 ' ' A Gl ossary of Sof tware Engineering Termi nol ogys ' IEEE 610 12- 1 990, i 2

lnst it ut e of El ectri cal and El ectroni c Engineers, New York, 1990 ' û

: SO/ I EC 1 2207: 995, nfor mat i on Technol ogy sof t war e ' i J

I SO/ I EC l 2207, 1 9951

' l nt er nat i onal Or gani zat i on f or St andar di zat i on, I nt ernat i onal j

El ect r ot echni cal Commi s si on, Geneva, 1995 j l l )

(Li entz and Swanson, 19801 B P LI ENTZ ANt ) E B SwAxsop k, Software Maintenance I

anagement: zt Study o ft he Mai ntenance c/ -t l pp/ pr dr cr Appli cati ons Softv vare i n 487 i

Dat a Processing O r ganizati ons Addison-W esley, Readi ng, M A, l 980 (

'

j '

!

j '

1 l

!

1

i

A R R œ M o : x ( k

'?

1

J

A IR * @ U RM ET QA SE STU PT :

:

t pIp P R@ T@ TT PE

1 The C rapi d prot otype for the Ai r Gourmet case study is avail able on the World Wi de Web at

-.mhheacom/engcs/compsci /schach.

! ' ( -1

1 ,

i

;

E

!

:

:

)

1

; :

i

l

' :

r j

i

t p

l

)

I

-j )

' j

; ;

! j

A IR * @ URM KT QA S: sTu py r ; l

p !

!

.

j j j

Yhe Java rapi d prototype for t he Ai r Gourmet case study is avail abl e on the World W i de Web at j 1 1

j I

l

j '

l

i !

? j

! ;

!

'

j

!

1 r (

l

j

! I

2 l

'

l

1

!

!

-!

521

f

j '