RIPPLES: Tool for Change in Legacy Software - CS Home
Contents
1. The rest of the dependencies of Figure 3 were not used in change propagation f menubar_cb2. argument Rdata default_font_choice In this subtask we want to find how this field of global variable Rdata gets its value The information flows to Rdata from library function XtVaGetApplicationResources which is called by mo_do_gui In this function call variables Rdata and resources are two actual arguments They store the same MOTIF resource nformation in different formats and the function converts the data between these two formats The data flow is directed from resources to Rdata XtVaGetApplicationResource is a library function its source code is not available We are not able to extract this data flow information This edge is added as conceptual dependence by hand Inspection of global variable resources finds that the information is stored as a part of a structure an array Each element in the array describes one XtResource its name it default value etc and defaultFontChoice is one of them Its initial value is Times Regular A conceptual dependence is found between resources and Rdata Both of them are actual arguments of XtVaGetApplicationResources We add a conceptual dependence to the ASDG to represent this connection Appendix C Change propagation We started the change propagation in function mo_get_font_size_from_res which is the one of the located functions see Figure 3 This function maps the font string to the enumerated font type We need to add statement to deal with four new fon
3. deteriorating and the documentation does not reflect the source code Such systems are usually called legacy systems Full comprehension is expensive and almost impossible and this substantially impacts the process of software change As a result usually only minor changes are done during this stage Software change starts with a maintenance request which is typically expressed in terms of domain concepts or program features that have to be modified The requested change may be corrective perfective adaptive or preventive 3 but in almost all cases it is formulated in terms of domain concepts The first part of change process is feature or concept location abbreviated location which has to be done before any actual change can be made to the system Location is a process that maps relevant domain concepts to the software components 7 Location may be easy in small systems that the programmer fully understands but it can be a considerable task in large and complex legacy systems 230 As there are always many different implementations for one concept it is very difficult if not impossible to implement a fully automatic feature locator The maintainer often performs location manually with the help of simple tools such as grep In our previous research 7 we looked for a more systematic process and proposed a computer assisted search of Abstract System Dependence Graph ASDG where the maintainer and the tool
4. edges to screen graph These operations do not change the underlying ASDG 3 Tool RIPPLES Tool RIPPLES extracts ASDG from C code and supports both concept location and change propagation The architecture of RIPPLES is shown in Figure 1 RIPPLES is implemented in Tcl Tk 22 Figure 2 contains the user interface of RIPPLES The tool consists of the following major components Version controller manages the source code archive Project manager handles project information file information and environment setting etc View manager controls zooming layout and event processing Log facility records all operations instrumentation efficiency measurements and undo Efficiency analyzer measures efficiency of RIPPLES see details in section 5 Program analyzer parses the code constructs ASDG and stores it in the database It uses Datrix C C parser 8 Tool RIPPLES supports the following operations Location operations are mark unmark and locate At the beginning of location process all components are unmarked As the location proceeds the maintainer marks interesting components by operation mark that sets one component as a candidate for investigation It can be applied to any component on screen graph The for 232 marks can have different colors Operation unmark is the opposite Operation locate allows the user to label the vertexes as unrelated located or visited Pattern matching fo
5. just created Between these two function calls two functions related to font processing are called mo_get_font_size_from_res and mo_set_fonts Function mo_get_font_size_from_res is called when current window is the top window i e without parent The actual argument is Rdata default_font_choice Rdata is a global variable of type AppData a C structure while default_font_choice is a string field of this C structure Function mo_set_fonts is called when the font is not the mo_regular_fonts Function mo_get_font_size_from_res maps the string font type to enumerated mo_token type For example the string Helvetica Regular is mapped to enumerated constant mo_regular_helvetica There is one enumerated constant defined for each font type Function mo_set_fonts is used to define font properties It defines font properties for the given font type Each property specification is a call to function XmxSetArg one of the arguments is the property name X system function XtSetValues is then called to transform the properties to X widget resources for later GUI interface usage In summary functions mo_get_font_size_from_res and mo_set_fonts are used for font type definition and are located in this search Appendix B Search what default font type is and how it is set In last subtask mo_set_fonts is located It is called by mo_open_window_internal which also calls function mo_get_font_size_from_res with actual
6. to analyze the behavior of specific components and their interactions After running a program they analyze the event trace and abstract the interaction pattern into various levels of abstraction A visualization tool ISVis is developed for this purpose They use ISVis in a Mosaic case study Horwitz and Reps 11 12 proposed System Dependence Graph to represent programs with multiple procedures They use SDG in program slicing program differencing and program integration Harrold et al 10 developed a method to construct SDG from Abstract Syntax Trees Liang and Harrold 16 also extended SDG to object oriented programs and used it in program slicing Recall and precision are the most commonly used formulas to compute the information retrieval effectiveness 6 The original concept of recall is defined as the ratio of relevant documents retrieved for a given query over the number of relevant documents for that query in the database Precision is defined as the ratio of the number of relevant documents retrieved over the total number of documents retrieved Both recall and precision are between 0 and 1 Researchers have recently used recall and precision in evaluating effectiveness of software tools Koschke etc 15 used them for experimental evaluation of clustering techniques for component recovery Antoniol etc 1 also used recall and precision to evaluate the performance of their tool that maintains traceability li
7. Copyright 2001 IEEE Published in the Proceedings of ICSM 01 IEEE Computer Society Press Los Alamitos CA ISBN 0 7695 1189 9 pp 230 239 Personal use of this material is permitted However permission to reprint republish this material for advertisin g or promotional purposes or for creating new collective works for resale or redistribution to servers or lists or to reuse any copyrighted component of this work in other works must be obtained from the IEEE Contact Manager Copyrights and Permissions IEFE Service Center 445 Hoes Lane P O Box 1331 Piscataway NJ 08855 1331 USA RIPPLES Tool for Change in Legacy Software Kunrong Chen Vaclav Rajlich Department of Computer Science Wayne State University Detroit MI 48202 USA rajlich cs wayne edu Abstract Key parts of software change are concept location and change propagation We introduce a tool RIPPLES that supports both It uses the Abstract System Dependence Graph ASDG of the program enriched by conceptual dependencies A case study of NCSA Mosaic demonstrates the use of the tool Precision and recall are used to evaluate the quality of support provided by RIPPLES 1 Introduction Staged model of software lifecycle 19 partitions software lifecycle into initial software development evolution servicing phase out and close down In this paper we focus on software changes that take place during servicing stage For most systems in this stage the architecture is
8. and types The edges are data flows control flows and define use relationships ASDG is derived from finer granularity System Dependence Graph SDG 11 12 that represents programs on the level of statements The vertexes of SDG are functions variables types arguments and program statements The edges represent control and data dependencies among them The extraction of SDG from the code in procedural languages is discussed in 10 11 12 Sets Def s and Use s contain variables defined or used in a statement s For a function definition a Format_in vertex represents a format argument and for each possible modified format argument an additional Format_out vertex is added For a function call a call vertex is added Actual_in vertexes are created for each actual argument and Actual_out for each possibly modified actual argument The actual and formal arguments are connected by dataflow edges Between Actual_in and Actual_out vertexes of one procedure there may be summary edges which model the transitive data flow inside the procedure call To construct an ASDG we need to delete statement vertexes of SDG If there is no function call in statement s and s does not define or use any variable s and all related edges are deleted If s does define and use variables add data flow edges from each vertex in 231 Use s to each vertex in Def s delete s and related edges If there is function call in s three operations are done bef
9. ces found in the location process were added to the ASDG One is between resources and Rdata Both variables hold the same information default font type but each in different format Both of them are actual arguments of function call XtVaGetApplicationResources in function mo_do_gui The other conceptual dependency is between menubar_cb and menuspec For each font related 234 menu item in menuspec there is one corresponding call back function in menubar_cb The concept default font type consists of the following components resources Rdata mo_get_font_size_from_res mo_token mo_set_fonts menubar_cb and menuspec They are interconnected by both explicit and conceptual dependencies see Figure 2 4 2 Propagation Our task is to add four font types Times Tiny Helvetica tiny New Century tiny and Lucida Bright tiny to Mosaic The change begun from one of the located components mo_get_font_size_from_res see Figure 3 This function maps the default font string to enumerated font type Then the mo_token enumerated type was changed to add four types to it The third change was to mo_set_fonts and it added statements that set properties for the four new enumerated font types Function menubar_cb was changed to add the new call back functions Finally the menu content was changed to add the four menu items to the menu In total five modifications were made to the source co
10. de Some of the located components that contained the concept did not need any change More details about the change are in Appendix C 4 3 Observations from the case study During the case study the authors used both domain and programming knowledge to guide the search and the propagation They also used Mosaic naming convention where a component name derives from domain or functionality The comments in the source code have also turned out to be useful To assess the help provided by RIPPLES a comparison was drawn with an earlier case study of Mosaic done by hand 7 During that case study the authors found that the search can become very complex It is not uncommon to forget how and why we got to one specific component and as a result the backtracking posed a particular problem RIPPLES provided us with an overall picture of the process that is very helpful It also provided a lot of information in choosing a vertex to investigate As a result the second case study was finished in a much shorter period 5 Measuring effectiveness of RIPPLES For each location or propagation task the following four numbers are retrieved the number of edges mo_open_window_internal oor ey mo_get_font_size_from_res menubar eb menuspec mo_fill_ window o_make_document_view_menub Figure 3 Screen graph for propagation changed components are highlighted suggested by RIPPLES the number of
11. der and D Webster Program Understanding and the Concept Assignment Problem Communications of the ACM 37 No 5 pp 72 83 May 1994 3 S Bohner and R Arnold An Introduction to Software 237 4 5 6 7 8 9 10 11 12 13 14 15 16 17 Change Impact Analysis Software Change Impact Analysis IEEE Computer Society 1996 R W Bowdidge W G Griswold Automated Support for Encapsulating Abstract Data Types SIGSOFT Software Engineering Notes vol 19 no 5 pp 97 110 Dec 1994 F P Brooks Jr The Computer Scientist as Toolsmith If Computer Graphics Vol 28 pp 281 287 November 1994 J A Capon Elementary Statistics for Social Sciences 1988 Wadworth publishing K Chen V Rajlich Case Study of Feature Location Using Dependence Graph Proc of the IWPC 2000 pp 241 249 Datrix C C Java parser web site by Bell Canada http www iro umontreal ca labs gelo datrix home page htm site Graphviz web by AT amp T research http www research att com sw tools graphviz M J Harrold B Malloy and G Rothermel Efficient Construction of Program Dependence Graphs Proc of the ACM SIGSOFT International Symposium on Software Testing and Analysis ISSTA 93 pp 160 170 S Horwitz T Reps and D Binkley Interprocedural Slicing Using Dependence Graphs ACM Trans Programming Languages and Sy
12. dges suggested by RIPPLES The number of relevant edges Recall Theoreticd precision The number of relevant edges suggested by RIPPLES a The number of edges suggested by RIPPLES Realistic precision The number of relevant edges suggested by RIPPLES The number of edges suggested by RIPPLES and investigated A recall rate is the percentage of relevant edges suggested by RIPPLES Of the three measures it is the 235 most important one Low recall rate means that the programmer must search the whole system for the missing edges He she has to use some other ways to find the edges for example read the source code or use some utility tool such as grep The following situation may cause a low recall rate the ASDG analyzer fails to extract some edges the two vertexes are connected by a conceptual dependence the maintainer uses an inappropriate expansion strategy during location To solve the low recall rate problem we should work on all three above aspects find a powerful and suitable ASDG analyzer understand the system better and adopt proper expansion strategy during location A low precision means that RIPPLES recommends irrelevant edges The difference between theoretical and realistic precision depends on the judgement of the user The continuations that are a part of the theoretical precision and not of realistic precision are only a small part of the programmer s work because they can be ruled out quickly and hence th
13. e font types menubar_cb needs to add call back functions for all new menu items The actual modification is to add the following statements in file src gui menubar c before original line 709 ee case mo_tiny_times case mo_tiny_helvetica case mo_tiny_newcentury case mo_tiny_lucidabright Finally the four font types must be added to menu content specification Function menubar_cb and menuspec are connected with conceptual dependence Global variable menuspec is used to specify menu content It is an array of structures and each of them corresponds to one pull down menu in the menu bar Options is one of them Global variable opts_menuspec specifies the content in the Options pull down menu It also is an array of structures and each of them corresponds to one menu item in the menu Font is one of them Global variable fnts_menuspec specifies the content in the Font menu item The actual change is made to fnts_menuspec the following statements are added before original line 1001 1005 1009 and 1013 in file src gui menubar c lt Times Tiny a menubar_cb mo_tiny_times lt Helvetica Tiny b menubar_cb mo_tiny_helvetica lt New Century Tiny d menubar_cb mo_tiny_newcentury lt Lucida Bright Tiny mo_tiny_lucidabright In conclusion we made five changes to the system One conceptual dependency and three explicit dependecies are used in the change propagation
14. ey do not impact adversely the programmer s productivity Therefore they are treated differently in this analysis The following factors could results in low realistic precision the maintainer does not have sufficient domain knowledge of the system the maintainer does not have an overall search skill the maintainer always chooses to expand to all neighbors Analysis of the location process reveals that both control flow and data flow expansions are used in the location The data flow expansion was used when we dealt with a comprehension of specific data items while top down control flow expansion was used when we tried to understand a specific functionality or algorithm For subtask one RIPPLES provides a total of 99 edges It provides all 4 edges taken in the search and all other edges are ruled out immediately So the recall rate is 4 4 100 theoretical precision rate is 4 99 4 04 and realistic precision rate is 4 4 100 In the second subtask RIPPLES provides 48 edges Only 13 of them are investigated and all other edges are excluded immediately In this subtask the recall rate is 13 13 100 Theoretical precision is 13 48 27 08 and realistic precision is 13 13 100 In the third subtask 3 steps are taken all of them are suggested by RIPPLES There are 94 possible choices So the recall is 100 the theoretical precision is 3 94 3 19 the realistic precision is 3 3 100 There are a total of 7 relevant steps in the
15. f Hierarchical System Structures IEEE Transactions on Systems Man and Cybernetics Volume 11 February 1981 pp 109 125 22 Tcl Tk resource web site http www scriptics com 23 N Wilde M Buckellew H Page V Rajlich A Case Study of Feature Location in Unstructured Legacy Fortran Code Proc Fifth European Conf On Software Maintenance and Reengineering IEEE Computer Society Press 2001 68 76 24 N Wilde and T Gust Locating user functionality in old code Proc of Conference on Software Maintenance 1992 Orlando Florida 1992 pp 200 205 25 N Wilde Michael Scully Software Reconnaissance Mapping Program Features to Code Software Maintenance Research and Practice Vol 7 pp 49 62 1995 26 Z Yu V Rajlich Hidden Dependencies In Program Comprehension and Change Propagation Proc of Proc of International Workshop on Program Comprehension 2001 IEEE Computer Society Press pp 201 210 Appendix A Locate components that determine font properties Function mo_open_window calls mo_make_window and mo_load_window_text The functionality of mo_make_window is to make a new window from scratch It creates an X window shell and then calls mo_open_window_internal Function mo_open_window_internal calls memory allocation function malloc to create a structure of type mo_window the internal structure for each window and calls function mo_fill_window to fill the window
16. fourth subtask They are picked up for investigation from 110 possible choices of which we investigated nine Two edges are found irrelevant so we backtrack two steps The recall rate is 100 theoretical precision is 7 110 6 36 and the realistic precision is 7 9 77 78 For the entire location we moved 22 times from one function to another through function call moved through data flow 5 times accessed data type information twice So a total of 29 unique edges are investigated of all 5779 edges in the system of which 27 are relevant to our task RIPPLES suggests a total of 351 edges for investigation Of the 1662 functions and global variables in Mosaic we visited only 23 Six components are located in this process as a part of the concept implementation For the whole location process recall rate is 27 27 100 theoretical precision is 27 351 7 69 and realistic precision is 27 29 93 10 Analysis of the propagation process Five components are changed and four edges are involved Among the four edges one is conceptual edge There are totally 71 edges suggested by RIPPLES and only the four are investigated So the recall rate with conceptual dependencies is 4 4 100 without them it is 3 4 75 The theoretical precision is 4 71 5 63 and realistic precision is 100 6 Other work Our work builds on several efforts Rajlich and Bennett 19 proposed a staged model for the software life cycle During evolution stage s
17. have alternating and complementary roles The result of the process is the set of components that implement the concept The components implementing the concept are often related to each other by data flow or control flow or definition reference relationship and it is a goal of static program analysis to extract these relationships as accurately as possible However in some instances there are additional dependencies that we call conceptual dependencies Two components in a conceptual dependence have a connection that is not discovered by static code analysis yet both participate in the same concept and a change in one may require a change in the other One example from our case study is the connection between the menu item and its call back function In our case study conceptual dependencies play a significant role They have to be discovered manually during location and added to the ASDG The maintainer starts software modifications after concepts are located and conceptual dependencies are established Modification of a component may cause the system to be temporarily inconsistent as the require provide relationships between the changed component and its neighbors are no longer valid To fix this secondary changes are introduced in the neighbors but they may cause new inconsistencies This situation continues until the system becomes consistent again 18 This process is called change propagation abbreviated propagation In
18. ic variables The search proceeds towards the origin of the values Forward data flow strategy is the opposite and the programmer is checking for the destination of the values Default strategy expands selected vertex by all neighbors 2 2 ASDG in change propagation Marks can also be used to support propagation process The marks on vertexes can be unmarked default and initial value candidate and changed The edges can be unmarked default forward or backward The mark on an edge means the direction of the change propagation During propagation ASDG changes as the maintainer adds or deletes components or edges from the system 2 3 Screen graph To support change in large and comp lex system only User Interface Project Manager shana fa ee hange Text Editor User Manager Check for Change Mark Propagate Annotate Version Controller Source Code Archive Program Analyzer Database Manager Data Storage Figure 1 The Architecture of RIPPLES a part of the ASDG is displayed on the screen This part of the ASDG is called screen graph Several operations manipulate it Operation expand adds the neighbors of selected component and the related edges to screen graph Operation compress hides neighbors of this component and all related edges from screen graph Operation hide hides a component and all its edges Its counterpart display adds any hidden component and its
19. ies to the ASDG First subtask is to find the function that opens a new window It is based on the fact that when we open a new browser window it reads default font settings from the parent window For this we adopt top down strategy The location starts from function main We found that function mo_open_window is used to open a new window Second subtask is to find how the font properties are specified in the new window We again adopted top down strategy and started from mo_open_window Appendix A describes this subtask in more detail We found that function mo_set_fonts sets the properties for each font type Third subtask is to find what the default type is and how and where it is set We adopted backward data flow strategy The search started from function mo_set_fonts We found that the default font is specified as string Times Regular in global variable resources of type XtResources This subtask is described in Appendix B Fourth subtask is to find the connection between the font related menu items and the font settings In the Mosaic menu bar the pull down menu Options has an item Font which sets the font type In order to find this in the code we adopted top_down strategy and started from mo_open_window We found that the menubar_cb is used to specify the callback function for font menu operation Function mo_set_fonts is called to set the properties for each type Conceptual dependen
20. ions can be easily ruled out The important research goal is to improve both precision and recall Both will be improved by extracting more relevant information by program analysis An intriguing question is posed by conceptual dependencies They are the dependencies that are not retrieved by current static analysis techniques yet they play an indispensable role in the change propagation The conceptual dependencies we encountered can be classified into two categories In the first category dependence involves data flow within library functions Due to unavailability of source code of these functions it is impossible to retrieve the dependence For example in our case study the conceptual edge between resources and Rdata could not be recovered by static analysis because the code of library function XtVaGetApplicationResources is not available In the other category the two components are only connected in the application domain An example in our case study is the dependency between menubar_cb and menuspec At this moment it is not clear what if any analysis would automatically discover such a dependency Further study of conceptual dependencies is our future goal 8 References 1 G Antoniol G Canfora G Casazza A De Lucia E Merlo Tracing Object Oriented Code into Functional Requirements Proc of International Workshop for Program Comprehension IWPC 2000 pp 79 86 2 T Biggerstaff B Mitban
21. nalyzer 4 Case Study of Change in MOSAIC We conducted a case study of a change consisting of both location and propagation using RIPPLES The Ripples C ripples4 usr windows mosaicnew mosaic5 _prj Project Edit Setting Tool Debug Help mo_get_font_size_from_res mo_try_to_grok_menubar mo_open_window_ internal o_make_document_view_menub Sian EE A A mo seios d xmxSetArg xtSet Values mo_fill_window menuspec xmxSet alues m ca Function mo_open_window id 838 of srcfgui c es _ sil Figure 2 RIPPLES screen graph for location Located Components Are Highlighted we studied was web browser NCSA Mosaic 2 5 for X Windows 17 and the task was to add one new font size to the standard set of fonts This case study is a self observatory case study done by the first author of the paper Prior to the case study he used Mosaic and that gave him rudimentary domain knowledge He then read the user manual and on line help to deepen the domain level comprehension This preparation was sufficient for the case study Domain knowledge NCSA Mosaic web browser supports four font families Times Helvetica New Century and Lucida Bright Three font sizes are provided for each font family regular small or large This means that with all combination there are 12 font types available The default font is Times Regular Mosaic environment defines the following 17 properties and each of them
22. nks between source code and free text documents Software browsers 13 20 extract and graphically represent program dependencies However they do not support the process of the change and do not support marks that the process of change requires Wilde et al 23 investigated a FORTRAN legacy system from the early 70 s and found that it has many properties different from Mosaic of early 90 s The data flows still play a prominent role in the location while control flows become less useful Graph layout algorithms relevant to RIPPLES are Sygiyama 21 Graphviz 9 and Star 4 7 Conclusions and future work The architecture and functionality of tool RIPPLES have been described in this paper The case study of NCSA Mosaic shows that it can work on practical systems Mosaic has been widely used by researchers as a subject for the research of maintenance tools and techniques We believe it is a representative of early 90 s C language legacies and there is a lot of legacy code with similar characteristics With the usage of screen graph we believe that RIPPLES is able to support the servicing of many large systems Precision and recall are important indicators of the tool quality Of the two recall is more important for the programmer it is important to get the information about all possible continuations Compared to that low precision that provides additional unused continuations is not such a big problem because those continuat
23. oftware engineers fully comprehend the system and extend the functionality of the system to satisfy new requirements In servicing stage only small repairs will be done to the entire system RIPPLES is aimed for servicing stage where full comprehension is not expected Biggerstaff et al 2 defined and investigated the concept location called concept assignment in their paper To perform concept assignment a prior knowledge of the specific domain a plausible reasoning etc are needed As concepts and program are not in the same level of abstraction human input is necessary Their conclusion is that totally automated tool for concept assignment is probably impossible but some degree of automation is helpful 236 In our previous research 7 we studied location using the dependence graphs The case study reported in 7 identified the requirements for RIPPLES Wilde et al 24 25 developed a program feature location technology called Software Reconnaissance The technology is based on the analysis of test cases The instrumented program is tested with two sets of test cases one set of test cases with the feature and the other set without The feature location is done by analysis of the two sets of event traces A tool Recon2 supports this technique Jerding and Rugaber 14 use both static and dynamic analysis in program understanding Static analysis is used to extract the system architecture and dynamic analysis is used
24. ore deleting statement s First add data flow edges from each vertex in the Use s to each vertex in Def s Second add acalling edge from the calling function to the called function Third remove Actual_in and Actual_out vertexes and related edges Add new summary edge for transitive data flow if there is dataflow through Actual_in or Actual_out vertexes This process allows us to base the extraction of ASDG on SDG algorithms ASDG is used for both concept location and change propagation 2 1 ASDG in concept location Location is a computer assisted search process 7 It is a step by step process where in each step one comp onent is selected for investigation To facilitate the process marks are set on vertexes and edges A vertex can be unmarked default and initial value candidate for investigation visited promising component further investigation in this direction is needed located part of the result and unrelated to the feature Marks on edges denote search direction Three possible values are unmarked default forward same as edge direction and reverse The programmer can choose one of several search strategies Top down strategy expands selected vertexes by adding called functions Bottomup strategy is the opposite of top down strategy and expands selected vertexes by adding calling functions Backward data flow strategy is employed when functionality of the system depends on specific values in specif
25. r vertex names is also provided The programmer specifies the vertex name and the name of the file it belongs to Wild card any length of character combination or any one character are allowed in the query All vertexes whose names match the description will be part of the result Propagation is supported by operations change add component delete component and skip The first three operations change the actual ASDG of the system Operation add component adds a new component to both ASDG and screen graph Operation delete component deletes an existing component and all its edges Operation change can apply to any marked component on screen graph After the maintainer changes the source code the operation prompts the maintainer to input the edge s that are deleted or added ASDG and screen graph is updated accordingly After the change all neighbors of the changed component are marked Operation skip is invoked when the marked component is visited and does not change The programmer has an option either to stop the change i e to simply erase the mark or to propagate it to all neighbors i e to mark all neighbors Conceptual dependencies are added to ASDG during the location process Operation add conceptual dependence is for this purpose The system displays the conceptual dependence using dashed lines Conceptual dependencies have to be inserted by the programmer as they are not extracted from the program by the a
26. relevant edges the number of relevant edges suggested by RIPPLES and the number of edges suggested by RIPPLES and investigated The first of these numbers is dependent purely on RIPPLES while the remaining three numbers are produced by the programmer using it As such they are dependent on the user and his her experience Together they measure how well does RIPPLES support the human user Using these four numbers we computed the recall theoretical precision and realistic precision using formulas of 6 with an additional slight modification In particular we make a distinction between theoretical and realistic precision The theoretical precision is based on all possible continuations suggested by RIPPLES Since RIPPLES is conducting static analysis of the program it suggests a large number of possible continuations of the search some of which can be ruled out immediately without reading the code of the components in question If we do not consider the continuations that can be ruled out immediately we get the realistic precision To rule out some continuations can be done very effectively For example when looking for a substantial concept like font type we can immediately rule out all trivial functions We can also rule out all functions that obviously deal with other concepts Hence we are justified in making a distinction between theoretical and realistic precisions The formulas are the following The number of relevant e
27. specifies how Mosaic displays the corresponding content of html document Font ItalicFont FixedBoldFont HeaderlFont BoldFont FixedFont FixedItalicFont Header2Font Header3Font Header4Font Header5Font Header6Font AddressFont PlainFont PlainBoldFont PlainItalicFont SupSubFont For example the Header2Font property defines how to display a level 2 header A new top level window uses default fonts The user can change the current fonts at any time All twelve types are available for selection New windows inherit the font types from their parent window For our case study the maintenance request is to add a new font size Tiny That means four new font types will be added Times Tiny Helvetica Tiny New Century Tiny and Lucida Bright Tiny As the font type setting is hard coded in the system we need to change the source code In our case study we first locate the concept font type then we change the code and finally we propagate the change 4 1 Location Our task is to locate where the font type is determined and find how the properties are set and how the menu font operations manipulate these setting It is a search process and the relevant part of ASDG is depicted in Figure 2 We divided location into four subtasks for easier manageability First subtask starts from the function main Each following subtask starts where the previous subtask finished Also during the search we add conceptual dependenc
28. stems Vol 12 No 1 Jan 1990 pp 26 60 S Horwitz T Reps The Use of Program Dependence Graphs in Software Engineering Proc of the 14 International Conference on Software Engineering ICSE May 1992 Imagix web site http www imagix com D Jerding and S Rugaber Using Visualization for Architectural Localization and Extraction Proceedings of the Fourth Working Conference on Reverse Engineering October 1997 the Netherlands IEEE Computer Society Press pp 56 65 R Koschke T Eisenbarth A Framework for Experimental Evaluation of Clustering Techniques Proc of International Workshop on Program Comprehension 2000 IEEE Computer Society Press pp 201 210 D Liang M J Harrold Slicing Objects Using System Dependence Graphs Proc of International Conference of Software Maintenance ICSMY 1998 November 1998 pp 358 367 site source codes Mosaic web and documents http www ncsa uiuc edu SDG Software Mosaic 18 V Rajlich A Model for Change Propagation Based on Graph Rewriting Proc of ICSM 1997 pp 84 91 19 V Rajlich K Bennett A Staged Model for the Software Life Cycle IEEE Computer July 2000 pp 66 71 20 V Rajlich N Damaskinos P Linos W Khorshid VIFOR A Tool for Software Maintenance Software Practice and Experience Vol 20 1 January 1990 67 77 21 K Sugiyama S Tagawa and M Toda Methods for Visual Understanding o
29. t types This is accomplished by addition of the following if statements to the appropriate places if strstr lowerfontstr tiny NULL return mo_tiny_times if strstr lowerfontstr tiny NULL return mo_tiny_helvetica if strstr lowerfontstr tiny NULL return mo_tiny_newcentury if strstr lowerfontstr tiny NULL return mo_tiny_lucidabright This requires a change in enumerated type mo_token The following four enumerate constants will be added to it mo_tiny_times mo_tiny_helvetica mo_tiny_newcentury mo_tiny_lucidabright The third changed component is mo_set_fonts add statements to set properties for the four new enumerated font types Function mo_set_fonts is connected with mo_token by a dependence The following statements are added to mo_set_fonts before original line 392 in src gui menubar c case mo_tiny_times XmxSetArg XtNfont wrapFont adobe times medium 239 XmxSetArg WbNplainitalicF ont wrapFont adobe XmxSetValues win gt scrolled_win win gt font_family 0 break Each above function call to XmxSetArg defines one of the 17 properties of the font type For example the Header2Font property is defined as adobe times bold r normal 15 Similar changes are made for the remaining font types Function menubar_cb defines call back functions for all menu items After calling mo_set_fonts to define properties for all four new enumerat
30. this paper we describe a new tool RIPPLES developed for servicing legacy systems It supports both location and propagation and combines automatic static code analysis with human intelligence as advocated by Brooks 5 Brooks claimed that intelligence amplifying systems can at any given level of available systems technology beat AI systems That is a machine and a mind can beat a mind imitating machine working by itself Based on that philosophy RIPPLES is a tool that helps the programmers in their maintenance work rather than replacing them It is the task of the programmers not of the tool to acquire all necessary domain and programming knowledge The programmers also make all decisions choose the starting component make changes determine the end of the process etc The tool only assists in these tasks In section 2 we describe Abstract System Dependency Graphs Section 3 describes the tool RIPPLES Section 4 describes case study of a change in NCSA Mosaic Section 5 analyzes the efficiency of the tool Section 6 lists other work in the area Section 7 concludes the paper The appendixes report selected parts of the case study in more detail 2 Abstract system dependence graph The basic data structure of tool RIPPLES is abstract system dependence graph ASDG 7 26 ASDG represents dependencies among software components For C programming language the vertexes are functions function arguments and global variables
Download Pdf Manuals
Related Search