- HSR - Institutional Repository

- HSR - Institutional Repository
bachelor thesis
Test Driven Development
for Eclipse CDT
Thomas Kallenberg, Martin Schwab
June 17, 2011
Supervisor: Prof. Peter Sommerlad
In 2006, Prof. Peter Sommerlad introduced the CUTE unit
testing framework for C++ to promote unit testing in C++
[ISO11]. The objective of this bachelor thesis is to ease practice
of Test-Driven Development (TDD) by providing code generation
in Eclipse CDT when using CUTE. Referring to an undefined entity when writing code results in a parse error. Based on such an
error, a resolution should be provided that generates the missing
code defining the entity. Code generation features three advantages: It speeds up the TDD workflow, allows focus on the actual
program logic and is useful for C++ programmers using CDT
even when not practicing TDD.
Throughout this bachelor thesis, the Eclipse development environment for C++ has been extended by a plug-in that provides
automated code generators. Twelve easy to use quick fixes have
been developed, along with a mechanism to extract definitions to
separate files. In addition, to provide dynamic labels and more
specific problem markers, static code checkers have been extended.
Until recently, CDT’s problem marker have only been used to indicate errors in source code. Combined with the developed plugin, they may be used to generate missing code blocks, too, making
Eclipse CDT and unit testing for C++ more attractive.
As the C++ language has a lot of features, more special cases
have probably to be regarded to make acceptance into CDT or
CUTE possible. A further step for improving TDD support in
CDT is to develop a Move Refactoring which allows to move
generated code to another file.
Executive Summary
Problem
Test-Driven Development aims to lead to better software. First
write a test, then add the actual functionality, refactor and integrate. This process benefits of at least two advantages. First,
new code is not implemented speculatively and second, the written software is tested right away.
Modern programming languages and their Integrated Development Environments (IDE) grew up with these principles. However, long established languages like C++ and most of their IDEs
are not designed to provide support for TDD.
In 2006 Peter Sommerlad introduced a Unit testing framework
for C++ called CUTE. Together with the integration in Eclipse
CDT, an IDE for C++, one big gap towards fundamental support
of TDD for C++ was closed.
1
# include " cute . h "
2
3
4
5
void testSquareArea () {
ASSERT_EQUALS (16 , MySquare (4) . getArea () ) ) ;
}
Listing 1: Testing C++ code with CUTE
However, most code still has to be written manually – a possible
source of errors.
Now if the parser is able to complain about missing code, why
not just complete it? This could be dangerous as there are decision to be taken deliberately by a programmer. Such a decision
could include which constructor should be chosen if there are
multiple ones or do not take basic types as arguments. However,
partial automation is possible without affecting correctness.
1
2
int testDivide () {
int i = divide (10 , 2) ;
3
}
Listing 2: The interface of divide() is defined by calling the
function
In Eclipse CDT it is possible to automatically complete function
calls with autocomplete if the function is declared. However,
generating code such as a function signature matching a function
call is not yet possible.
If such code could be generated in a fast way, a test could be
written and all missing parts could be generated. The programmer could fully concentrate on the program logic.
1
# include " cute . h "
2
3
4
5
int divide ( int , int ) {
return int () ;
}
6
7
8
9
void testDivide () {
ASSERT_EQUALS (5 , divide (10 , 2) ) ;
}
Listing 3: Generated function divide() with default return
statement
Solution
To fix missing code in a fast and correct manner, the plug-in
relies on source code problem markers Quick Fixes and linked
mode editing.
As shown in figure 0.1 markers are hints displayed on the left
side of the source code editor.
Quick Fixes are actions bound to a marker which makes corrections to the source code without a dialog and the changes are
displayed in the same source editor providing the possibility to go
on immediately.
Linked mode, as shown in figure 0.6 is a feature of Eclipse to
edit multiple positions simultaneously. Editing the return type of
a function will produce the same text after the return statement
in the function body.
iii
The source code is analyzed instantly by the Eclipse CDT
parser while typing and different kinds of problems are detected.
Such problems are displayed as markers. At any time, the programmer may hit a shortcut for additional information about the
problem of such a marker and various actions to correct them
in a fast and simple manner are presented. These corrections,
called Quick Fixes, start refactorings. They analyze source code
around the marker and generate code to make the marker and
the problem disappear. In addition, linked mode is started to
edit names and options after the change has been completed. On
the following pages the modifications with examples are presented
which support TDD in the plug-in.
Create Local Variable
In a first step, a local variable is created. Analyzing the code
does not return an informative result since it cannot be decided
what kind of type variable s is.
Figure 0.1.: Local variable s is missing
After creation of the local variable the type of s can be changed
to Geometry::square.
Figure 0.2.: Local variable s has been declared
Creating Namespace
The first marker indicates that the namespace Geometry could not
be found. To resolve this, the Quick Fix to create a namespace
is used.
iv
Figure 0.3.: Namespace Geometry needs to be created
Creating Types
The next marker indicates that the type square does not yet
exist. Using the Create new Type Quick Fix will create type
Square in namespace Geometry.
Figure 0.4.: Type square in namespace Geometry needs to be
created
Creating (Member) Functions
The member function getArea() is still missing. Quick Fix Create
Member Function will create this function type and will return
the default constructed type of the return type. To determine the
necessary return type, the called ASSERT macro is analyzed and
the appropriate type is extracted.
Figure 0.5.: Member function getArea() needs to be created
After creating the function, various options such as const or
the return type can be edited with linked mode.
v
Figure 0.6.: Editing function with linked mode
Creating Constructors
After square has been inserted, more code is still missing. Preferably, the length of a side should be passed in via a constructor.
Changing the default constructor call and adding an argument for
the side length is done manually.
Figure 0.7.: Changing the constructor call
A new marker is displayed informing the user that there is
a call to a constructor that does not exist. Using the Create
Constructor Quick Fix will generate the needed constructor.
Figure 0.8.: Newly created constructor call
Creating Member Variables
The passed in variable i needs to be assigned to a member variable. Adding an initializer list will trigger a new marker indicating
that the name side does not exist.
vi
Figure 0.9.: Create missing member variable
The Create Member Variable solves this problem by creating a
privately declared member variable in the struct.
Figure 0.10.: Editing the type of side with linked mode
Implementing the Functionality
The actual program logic is implemented manually: getArea()
must return the square of side.
Figure 0.11.: Implementing the functionality
How to Use
All these features are provided as one installable plug-in together
with CUTE.
vii
Limitations and Outlook
The developed features are mainly designed for use with creation
of unit tests. For other purposes, it is possible that support for
more language constructs has to be added. Furthermore, static
code analysis is a tradeoff between complexity and speed. As
checkers become more complex, code analysis will become slower.
In future, for large projects, less useful checkers will have to be
turned off to ensure responsiveness of error markers.
In a next step, this plug-in could be bundled with CUTE which
is simple because of the plug-in architecture. At a later time, the
new features may be proposed to the CDT community. Anyway,
Quick Fixes for code generation are a hot topic and new features
will be added to CDT in the near future.
viii
Thanks
We would like to thank our ”Eclipse team” at the Institute For
Software IFS for supporting us with the development of this plugin. First thanks goes to Prof. Peter Sommerlad for supervising
this project. Emanuel Graf integrated our semester thesis into
CDT and helped with architectural issues. Thanks also to Lukas
Felber for having a solution for almost every Eclipse issue, especially in case of tests. Other team members helped with feedback
on documentation and a lot of humor.
Another big thanks goes to our families and friends who also
reviewed documentation and were missed out a little bit during
our bachelor thesis.
Contents
1. Introduction
1.1. Current Situation . . . . . . . . . .
1.1.1. Peeking inside Eclipse JDT
1.2. Motivation . . . . . . . . . . . . . .
1.3. Basic Workflow Example . . . . . .
1.3.1. Generate a Free Function .
1.3.2. Add a Missing Type . . . .
1.3.3. Create a Member Function
1.4. Expected Result . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
2. Analysis
2.1. Where to Put New Code? . . . . . . . . . .
2.2. Generating Types . . . . . . . . . . . . . . .
2.2.1. Generating Classes . . . . . . . . . .
2.2.2. Nested Classes and Namespaces . . .
2.2.3. Namespace or Type . . . . . . . . .
2.2.4. Classes, Structs, Enums and Unions
2.2.5. Detecting Interfaces . . . . . . . . .
2.2.6. Templated classes . . . . . . . . . . .
2.3. Generating Functions . . . . . . . . . . . . .
2.3.1. Function Return Type . . . . . . . .
2.3.2. Determining Parameter Type . . . .
2.3.3. Adding a Free Function . . . . . . .
2.3.4. Adding a Member Function . . . . .
2.4. Changing Arguments . . . . . . . . . . . . .
2.4.1. Removing Arguments . . . . . . . . .
2.4.2. Appending Arguments . . . . . . . .
2.4.3. Swapping Arguments . . . . . . . . .
2.4.4. Changing Types . . . . . . . . . . .
2.5. Generating Variables . . . . . . . . . . . . .
2.6. Changing Visibility . . . . . . . . . . . . . .
x
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1
2
2
3
3
4
6
7
8
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
9
9
10
10
12
12
13
13
14
14
15
16
18
19
22
25
26
26
27
28
29
Contents
2.7. Move Refactorings . . . . . . . . . . . . . . . . . .
2.7.1. Moving Functions . . . . . . . . . . . . . . .
2.7.2. Extracting Types to Separate Source Files .
30
31
31
3. Implementation and Solution
3.1. Approach . . . . . . . . . . . . . . . . . . . . . . .
3.2. Prerequisites . . . . . . . . . . . . . . . . . . . . . .
3.2.1. Source Inspection . . . . . . . . . . . . . . .
3.2.2. Problem Detection . . . . . . . . . . . . . .
3.2.3. Resolutions / Quick Fix . . . . . . . . . . .
3.2.4. Code Analysis and Change . . . . . . . . .
3.2.5. User Guidance . . . . . . . . . . . . . . . .
3.3. Used Extension Points . . . . . . . . . . . . . . . .
3.3.1. Checkers . . . . . . . . . . . . . . . . . . . .
3.3.2. Codan Marker Resolution . . . . . . . . . .
3.3.3. Marker Resolution Generator . . . . . . . .
3.3.4. Commands, Bindings, ActionSets . . . . . .
3.3.5. Quick-Fixes for a Specific messagePattern
3.4. Marker Resolution Refactorings . . . . . . . . . . .
3.4.1. Create Type or Namespace . . . . . . . . .
3.4.2. Append / Remove Argument . . . . . . . .
3.4.3. Create Function . . . . . . . . . . . . . . . .
3.4.4. Create Variable . . . . . . . . . . . . . . . .
3.4.5. Change Visibility . . . . . . . . . . . . . . .
3.4.6. Extract Type and Function . . . . . . . . .
3.5. Architecture . . . . . . . . . . . . . . . . . . . . . .
3.5.1. Package Structure . . . . . . . . . . . . . .
3.5.2. Class Diagram . . . . . . . . . . . . . . . .
3.5.3. Basic Call Flow . . . . . . . . . . . . . . . .
3.5.4. Design Decisions . . . . . . . . . . . . . . .
3.5.5. Important Class Overview . . . . . . . . . .
3.6. Testing . . . . . . . . . . . . . . . . . . . . . . . . .
3.6.1. Unit Tests . . . . . . . . . . . . . . . . . . .
3.6.2. Refactoring Tests . . . . . . . . . . . . . . .
3.6.3. Integration Tests . . . . . . . . . . . . . . .
3.6.4. Tests on Large Projects - Performance . . .
3.7. Known Issues . . . . . . . . . . . . . . . . . . . . .
3.7.1. Checker Performance . . . . . . . . . . . . .
3.7.2. False Positives . . . . . . . . . . . . . . . .
32
32
33
33
33
34
34
35
35
36
37
37
39
39
39
39
39
40
41
42
42
42
43
43
43
45
49
52
53
53
53
54
55
55
55
xi
Contents
3.7.3.
3.7.4.
3.7.5.
3.8. Solved
3.8.1.
3.8.2.
3.8.3.
3.8.4.
3.8.5.
Chaos with Basic Types . . . . . . .
Stopping at Typedef . . . . . . . . .
AST Parser and Syntax Errors . . .
Issues . . . . . . . . . . . . . . . . .
Detecting Constructors . . . . . . . .
Detecting Operators . . . . . . . . .
Deleting Function Creates no Marker
Codan Marker Patch . . . . . . . . .
Create Linked Mode Positions . . . .
4. Conclusions, Interpretation and
4.1. Conclusions . . . . . . . . .
4.2. Interpretation . . . . . . . .
4.2.1. Limitations . . . . .
4.2.2. Personal Review . .
4.2.3. Insights . . . . . . .
4.3. Future Work . . . . . . . .
Future Work
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
55
56
57
58
58
59
59
60
60
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
62
62
63
63
63
65
65
A. User Manual
A.1. Installation of the Plug-in . . . . .
A.2. Building The Source Code . . . . .
A.3. Using the Quick Fix . . . . . . . .
A.3.1. Editing Generated Code . .
A.3.2. Available Quick Fixes . . .
A.4. Extracting Types and Functions .
A.4.1. Target-Destination Conflicts
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
67
67
68
68
69
69
71
71
B. Extension Manual
B.1. Checkers . . . . . . . . .
B.2. Quick Fix . . . . . . . .
B.3. Refactoring . . . . . . .
B.4. Tests . . . . . . . . . . .
B.4.1. Refactoring Tests
B.4.2. Quick Fix Tests .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
73
73
73
74
74
74
74
C. Project Setup
C.1. Configuration Management . . . . . . . . . . . . .
75
75
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
xii
D. Time Management
D.1. Time Budget . . . . . . . . . . . . . . . . . . . . .
D.1.1. Activities . . . . . . . . . . . . . . . . . . .
D.2. Declaration of Originality . . . . . . . . . . . . . .
76
76
77
79
Glossary
81
Bibliography
84
CHAPTER 1. INTRODUCTION
1. Introduction
For this bachelor thesis, a plug-in is developed that supports the
development cycle of Test-Driven Development: Red – Green –
Refactor [Bec03]. Basically, tests are written before the actual
code exists. Following the parse errors, code is inserted to make
the code compile. Yet the unit tests stay red.
Figure 1.1.: example test with two parse errors about missing
code
Now if the parser is able to complain about missing code, why
does it not complete it? This could be dangerous as there are
decisions to be taken deliberately by a programmer. For example
choosing one of multiple constructors or constructors with a nonbasic type as argument. However, partial automation is possible
without affecting correctness.
Automation is the subject of this thesis. It is objected to provide a tool to C++ programmers which helps generating such
missing code. The tool should integrate into Eclipse[Ecl11a] CDT
[Bau10], an extensible and industrial strength development environment for C++.
Readers are guided through this documentation by sample code
written for the CUTE[Som11b] unit testing framework. Throughout the project it is used as reference test environment. Nevertheless, the plug-in that is being developed should also work with
other C++ test frameworks.
1
CHAPTER 1. INTRODUCTION
1.1. Current Situation
Various C++ unit testing frameworks exist for Eclipse CDT, providing mechanisms to generate new test suites and test cases.
However, to get really into the TDD workflow, a lot of code still
has to be written by hand.
The good news is that missing definitions are already detected
[Ecl10a] by code analysis (codan[Ecl11g]). Codan checks the
source code for errors during typing. The product of Codan are
markers on the left site of the editor which indicate possible errors. Such a marker may be associated with a so called Quick
Fix [Eclb] which provides the user with a solution to solve the
particular problem.
Although Codan existing for a while, there are not yet many
Quick Fixes to solve problems. In most cases no help is offered to
generate missing code. In summer 2010, functionality for creating
a local variable[Ecl10b] has been added to the CDT repository1 ,
so a first step has already been taken.
Figure 1.2.: generating a local variable in CDT
1.1.1. Peeking inside Eclipse JDT
Eclipse JDT offers a set of tools for TDD. Figure 1.3 visualizes
how code automation helps writing a JUnit [JUN11] test method
in Java.
However, in the authors’ views, JDT has some usability flaws
such as overloaded user dialogs. In addition there is no guarantee
that a solution in Java is also a good way to solve problems in
1
since v201008041021
2
CHAPTER 1. INTRODUCTION
(a) Parser complains about missing (b) Generate empty method getAncode. Generate new class Uniswer() which returns zero.
verse.
(c) Test fails because the created (d) Tests go green after changing
method returns zero.
the default return value to 42.
Figure 1.3.: TDD workflow realized in Eclipse JDT (Java)
C++. Furthermore, introducing the Java way of thinking should
be avoided for CDT features. It is intended to implement similar
functionality as JDT with best possible usability.
1.2. Motivation
Automating frequent programming steps is helpful. Programmers
should be able to focus on their task instead of typing code all
the time. There are several reasons why TDD should have better
support in Eclipse. First, it makes TDD, Eclipse CDT and CUTE
tests more attractive. Second, the planned functionality is helpful
for other tasks too, so every C++ programmer who uses CDT
may profit. And in the end, working on a large project like
Eclipse is an interesting challenge.
1.3. Basic Workflow Example
At the beginning of TDD there is a test suite. After creating
a suite, a new test function is added and filled with an ASSERT
statement. It specifies the expected result of the future code.
3
CHAPTER 1. INTRODUCTION
Listing 1.1 shows an example of a CUTE test function which
does not compile as the symbol square is undeclared.
1
# include " cute . h "
2
3
4
5
void testSimpleSquare () {
ASSERT_EQUAL (4 , square (2) ) ;
}
6
7
8
9
10
11
12
void runSuite () {
cute :: suite s ;
s . push_back ( CUTE ( testSimpleSquare ) ) ;
cute :: ide_listener lis ;
cute :: makeRunner ( lis ) (s , " The Suite " ) ;
}
13
14
15
16
17
int main () {
runSuite () ;
return 0;
}
Listing 1.1: Symbol square inside testSimpleSquare is missing
In subsequent code listings, repeating test framework code (lines
1-3 and 7-17 in listing 1.1) will be omitted. Starting at the testSimpleSquare test function, examples are illustrating the convenience of code generation during test writing.
1.3.1. Generate a Free Function
First, a test is written and the expected result is defined.
1
2
3
void thisIsATest () {
ASSERT_EQUAL (4 , square (2) ) ;
}
Listing 1.2: defining the designated functionality
The missing function is generated by Eclipse above the test
function.
1
2
3
int square ( int i ) {
return int () ;
}
4
CHAPTER 1. INTRODUCTION
4
5
6
7
void thisIsATest () {
ASSERT_EQUAL (4 , square (2) ) ;
}
Listing 1.3: generated function square with default return value
The compiler now accepts the code and the tests turn red
caused by the default return value of square(). It may now be
replaced by the square of the parameter.
Make the Tests go Green
The last example leaves the programmer with compiling code and
the tests go red because of square() returning the default value of
zero. Remember the TDD mantra[Bec03]: Red, Green, Refactor.
The next step is to make the test go green. So far there is
a default return value int() which corresponds to the needed
return type. In this case the function body is return i*i; which
causes the test to pass. Listing 1.4 shows the code that leads to
a green bar.
1
2
3
int square ( int i ) {
return i * i ;
}
4
5
6
7
void thisIsATest () {
ASSERT_EQUAL (4 , square (2) ) ;
}
Listing 1.4: replaced return value to make tests green
Move the Implementation
As soon as the tests show green, the final step is to move the
newly implemented code to the place where it belongs – the
refactoring step of the TDD lifecycle. For this purpose the Move
Function Refactoring may be used. In the code of listing 1.4 the
function square() should be moved to its own header file.
The Move Function Refactoring will move the function square
to a newly created header file called square.h and leaves the test
code and the implementation separated and clean.
5
CHAPTER 1. INTRODUCTION
1
# include " square . h "
2
3
4
5
void thisIsATest () {
ASSERT_EQUAL (4 , square (2) ) ;
}
Listing 1.5: Test source file: test.cpp
1
2
# ifndef SQUARE_H_
# define SQUARE_H_
3
4
5
6
int square ( int i ) {
return i * i ;
}
7
8
# endif /* _SQUARE_H */
Listing 1.6: Separated header file: square.h
1.3.2. Add a Missing Type
In listing 1.7, the type Universe and its member are missing and
need to be created.
1
2
3
4
void thisIsATest () {
Universe universe ;
ASSERT_EQUAL (42 , universe . getAnswer () ) ;
}
Listing 1.7: The type universe is missing
The missing type is indicated by an error marker. A Quick Fix
is registered to the error marker. As the Quick Fix is applied
by the programmer, a simple struct is generated in front of
thisIsATest(). Listing 1.8 demonstrates the situation after the
missing type has been created.
1
2
struct Universe {
};
3
4
5
void thisIsATest () {
Universe universe ;
6
CHAPTER 1. INTRODUCTION
ASSERT_EQUAL (42 , universe . getAnswer () ) ;
6
7
}
Listing 1.8: A new struct has been created
Since several keywords are available to define a class, such as
class, struct or enum, the desired keyword can be chosen from
an editable proposal menu. See figure 1.4 for an example of linked
mode editing with a proposal menu.
Figure 1.4.: Proposal menu offered by the Eclipse framework.
1.3.3. Create a Member Function
The newly created type Universe is still missing its member
getAnswer(). getAnswer() is still producing a parse error as
it is an unknown member function for type Universe. To resolve
this parse error, another Quick Fix is offered to create an appropriate member function. Having applied the latter, the code
should be in a compiling state as in listing 1.9.
1
2
3
4
5
struct Universe {
int getAnswer ( int i ) {
return int () ;
}
};
6
7
8
9
10
void thisIsATest () {
Universe universe ;
ASSERT_EQUAL (42 , universe . getAnswer () ) ;
}
11
12
13
void runSuite () {
[...]
Listing 1.9: Newly created member function getAnswer()
7
CHAPTER 1. INTRODUCTION
Here, linked mode editing may also be used to quickly edit the
return type after the member function has been created.
1.4. Expected Result
In the end there should be no problems in creating new free or
member functions, new types or namespaces. Visibility on member functions can be changed fast. Adding or removing types from
functions should not be a problem anymore. Moving functions is
supported in a basic way. Classes may be moved to their own
files and an #include statement is inserted where needed.
The result should be easy to install from Eclipse CDT, either
as a standalone plug-in or bundled with the CUTE unit testing
plug-in. It may be also be integrated into the CDT project.
A user manual is created, explaining how to effectively use the
plug-in and to be fully aware of all TDD features offered by the
plug-in. For developers, there is an extension manual explaining
how to easily create new Quick Fixes and refactorings for code
editing.
8
CHAPTER 2. ANALYSIS
2. Analysis
TDD for CDT is a broad topic and a look at Eclipse JDT for
Java reveals that the number of possible code generation features
is immense. The scope of this analysis is reduced to features that
can be considered helpful for Test-Driven Development.
This chapter discusses which features are helpful for TDD and
for which reason, where generated code should be put and lists
special cases that should be kept in mind during implementation.
2.1. Where to Put New Code?
This topic was discussed controversially at the beginning of the
project. The question where to put new code is difficult to answer.
Usually, tests are placed into a separate project. Which effective
project is the project under test? If there is more than one, which
should be preferred?
Standing in front of a hard problem with many solutions, the
idea to use a wizard was close. However, many wizards in CDT
are too overloaded because they try to do too much (look at the
New Class wizard for an example). Therefore, a simpler solution
was looked for.
In a meeting, the idea came up to put the new code above the
current code block. During the design of a new unit test, this has
the following advantages:
• The code compiles because the declaration is above the calling code.
• Switching between tests and implementation is faster because the new code is close to the test function and can be
reviewed immediately.
• No user dialog is required, which shortens the process of
writing tests.
9
CHAPTER 2. ANALYSIS
There are some disadvantages to this solution too. First, if a
function is called from multiple points inside the code, it possible
that the new code requires a separate forward declaration which
is not favorable. However, this is not critical to TDD.
Second, the generated code has to be moved separately. This
can be solved by providing a comfortable feature to move code to
the project under test. A full-featured Move Refactoring would
support such a behavior.
2.2. Generating Types
Each member function corresponds to a certain type. For objectoriented programming, types are fundamental, which is why their
creation is indispensable for this plug-in. Before a parser checker
is able to complain about a missing member function, the corresponding type has to be resolved correctly.
There is a temptation to offer the overloaded Create Class wizard here. However, the programmer is able to continue much
faster if the class is just inserted above the current test function.
Unlike Java, in C++ there is no one-public-class-per-file [Mic05]
limitation.
Later, after having implemented the member functions, the new
type may be moved to another source file using the feature to
extract types described in 2.7.2. The type creation and extraction
process is illustrated in figure 2.2.
2.2.1. Generating Classes
Listing 2.1 shows a basic situation with a missing type. This is the
simplest case of code generation that needs to be implemented.
Code generation needs to create a new type named MyClass above
the test function testNewClass().
1
2
3
4
void testNewClass () {
MyClass c ;
ASSERT_EQUAL ( c . member () , true ) ;
}
Listing 2.1: Creating a simple new type.
10
CHAPTER 2. ANALYSIS
Figure 2.1.: Create a class, implement it and move it
To make the plug-in usable for other purposes, too, classes should
be created out of other situations, too.
For test fixtures as supported by CUTE[Som08] a test is specified inside a struct as in listing 2.2.
1
2
3
4
5
6
struct fix {
void test () {
klass k ;
ASSERT ( k . member () ) ;
}
};
Listing 2.2: Test fixture as supported by CUTE
The class under test, in this case klass, should then be created
outside of the fixture class as shown in listing 2.3.
1
2
3
4
5
struct klass {
boolean member () {
return true ;
}
};
6
7
8
9
10
11
struct fix {
void test () {
klass k ;
ASSERT ( k . member () ) ;
}
11
CHAPTER 2. ANALYSIS
12
};
Listing 2.3: Test fixture as supported by CUTE
2.2.2. Nested Classes and Namespaces
A class may be placed inside a specific namespace as shown in
listing 2.4 and 2.5.
1
2
3
4
void testFoo () {
N :: MyClass c ;
ASSERT_EQUAL ( c . foo () , true ) ;
}
Listing 2.4: Call to an undefined namespace.
After applying a Quick Fix for creating namespaces the code looks
like in listing 2.5
1
2
3
4
5
6
7
namespace N {
struct MyClass {};
}
void testFoo () {
N :: MyClass c ;
ASSERT_EQUAL ( c . foo () , true ) ;
}
Listing 2.5: Creating a class inside a namespace.
If scope of the current selection is inside a class, a newly created
class should be nested to reduce name conflicts.
2.2.3. Namespace or Type
Given the source code in listing 2.6. If X does not exist, it is not
possible to decide whether X should be created as a namespace
or as a type. To solve this dilemma, two quick fixes have to be
offered, one to create a namespace and one that creates a new
type.
1
2
3
void t es t N am e s pa c e Or T y pe () {
X :: MyClass c ;
ASSERT_EQUAL ( c . foo () , true ) ;
12
CHAPTER 2. ANALYSIS
4
}
Listing 2.6: Should X be created as namespace or as type?
Figure 2.2.: Both, namespace and type creation is offered
2.2.4. Classes, Structs, Enums and Unions
C++ offers multiple keywords for declaring types. Is is up to
the programmer to choose between the keywords class, struct,
enum and union. To reduce visibility issues, struct could be used
as default, as its members are public by default. However, this
is a matter of coding style. An option is to provide a selection
of keywords from which the user may choose from. No support
needs to be implemented for union since it is not necessary in
C++. In C++, type members are already non-virtual by default
which supersedes the use of the union keyword. [Mic11]
2.2.5. Detecting Interfaces
It is theoretically possible to gain information about the interface
a new class has to provide. In listing 2.7, a situation is shown
where a class should be created which satisfies a certain interface.
1
2
3
4
5
void foo ( Listener x ) { /* ... */ }
void testSomeFunction () {
MyClass c ;
ASSERT_EQUAL ( foo ( c ) , true ) ;
}
Listing 2.7: Interfaces
13
CHAPTER 2. ANALYSIS
2.2.6. Templated classes
The creation of templated classes should be allowed by using a
class with it’s actual template parameters.
1
2
3
void function () {
A < int > a ;
}
Listing 2.8: Use of a templated class which does not exist
A is not defined yet so it should be complained about the missing type. Creating the class A will result in the following code
seen in 2.9.
1
2
3
template < typename T >
struct A {
};
4
5
6
7
void function () {
A < int > a ;
}
Listing 2.9: Templated class was created
This should also work for multiple template argumentss as for
non-typed template parameter seen in 2.10
1
2
3
template < typename T , int i >
struct A {
};
4
5
6
7
void function () {
A < int , 5 > a ;
}
Listing 2.10: Class A with non-typed template parameter i
2.3. Generating Functions
During (member) function creation, several aspects have to be
considered. A new function needs a correct signature, has to be
inserted with an appropriate visibility and const correctness[Sut09]
14
CHAPTER 2. ANALYSIS
should be regarded. A set of such constraints for new functions
are discussed in the following sections.
2.3.1. Function Return Type
To be most supportive, suitable return types should be detected
automatically. Hence, before the function is created, the function
call is analyzed to gather information about the return type. The
chosen return type may be edited in linked mode right after the
function was inserted.
Procedures
In case of procedures, void is assumed as return type. A function
is considered a procedure if no calculation is done with its return
value. Listing 2.11 shows an example of a missing procedure.
1
2
3
4
void test () {
procedure () ;
ASSERT ( p r o c e d u r e _ w a s _ s u c c e s s f u l ) ;
}
Listing 2.11: A procedure which should have a void return type
Basic Data Types
In case of basic types, it is always legal to return an integer.
However, if for example a function is called inside an if(...)
statement, it is more intuitive to create a function returning a
bool instead. In case of other basic data types, loss of precision
is possible. In addition, std::string should be preferred over
const char[] as return type.
Binary Expressions
In case of binary operators, the type of the sibling operand can
be used. There may be other possible types, especially in case
of inheritance, but the primary objective is to produce compiling
code. A selection of possible types could be presented in a proposal menu after function creation. For simplicity, only the closest
15
CHAPTER 2. ANALYSIS
type should be used while linked mode could give the option to
choose another type manually.
An example of a binary operator is the assignment operator in
listing 2.12.
1
int x = function () ;
Listing 2.12: Simple assignment
This leaves only the option of an int as return type – the type of
the left-hand side of the assignment.
If the sibling operand cannot be resolved to a type too, the
correct return type can be searched in the expression’s parent.
The type is looked up from inside out until there is an assignment
or another expression which reveals the desired data type.
Another case could be a cast like this:
1
int x = static_cast < int >( function () ) ;
Listing 2.13: Determining function return type in case of a cast
Since this does not indicate a direct return type of function(),
it could return anything that could be converted to an int.
Unary Expressions
Unary expressions are similar to binary expressions. However,
because the operator itself does not make clear what data type is
used, the parent node has to be searched for the correct return
type.
2.3.2. Determining Parameter Type
Similar to the function return type, parameter types have to be
determined too. Parameter types may be looked up based on the
unresolved function call. In listing 2.14, based on the argument
42, a parameter of type int is created.
1
2
void foo ( int i ) {
}
3
4
5
void testAnswer () {
foo (42) ;
16
CHAPTER 2. ANALYSIS
6
}
Listing 2.14: Determining function parameter types
As in the above example, determining the parameter type for
literal nodes is straightforward. In case of a function call, the
corresponding return type of the function has to be copied. For
example, to create the function myFunction() in listing 2.15, the
signature of mulitply needs to be inspected to determine the
second parameter type.
1
double multiply ( double first , double second ) ;
2
3
4
5
void testMyFunction () {
myFunction (1 , mulitiply () ) ;
}
Listing 2.15: Looking up the type of a function
Parameter Names
When a function with parameters is created, each parameter
should have an intuitive and unique name to ensure the code
compiles. For basic data types, the first letter of the type name,
e.g. i for int, is taken. If multiple parameters are created with
the same type, the next free letter is chosen as shown in listing
2.16.
1
2
3
void foo ( int const & i , int const & j ,
int const & k , double const & d ) {
}
4
5
6
7
void testAnswer () {
foo (42 , 43 , 44 , 45.0) ;
}
Listing 2.16: letters used for basic type parameters
If a function is called with a symbol as argument (like function
calls, variable names), the name of the symbol in lowercase is
taken. In case of name collisions, either an incrementing number
is appended to the symbol name, or the next free letter is chosen
17
CHAPTER 2. ANALYSIS
in case of basic types (like i for int, then j, k, l, etc...).
Shadowing of variables should be avoided and the next number
should be chosen that does not conflict with another symbol.
Listing 2.17 shows an example that covers all discussed parameter names and collisions.
1
2
3
void foo ( double const & d , double const & e ,
int & e1 , int & delta , int & delta1 ) {
}
4
5
6
7
8
void testAnswer () {
int e , delta ;
foo (42.0 , 43.0 , e , delta , delta ) ;
}
Listing 2.17: avoiding name conflicts for parameter names
2.3.3. Adding a Free Function
A new function definition shall be created with parameters that
match the data type of the calling functions arguments as discussed in sections 2.3.2 and 2.3.2. The insertion point of the new
function is one of the following:
1. Above the parent function definition if the current scope is
not inside a type definition. Example:
1
2
3
4
5
void freefunction () {
}
void test () {
freefunction () ;
}
Listing 2.18: free function created above a function call
2. Above the outermost type definition if current scope is located inside a type definition. Example:
1
2
3
4
namespace N {
void freefunction () {
}
class A {
18
CHAPTER 2. ANALYSIS
struct B {
void test () {
freefunction () ;
}
};
5
6
7
8
9
10
11
};
}
Listing 2.19: free function created above type definitions
3. Above the current expression statement else. Example:
1
2
3
4
int freefunction () {
return int () ;
}
int x = freefunction () ;
Listing 2.20: free function created above an assignment
Inside a type definition, both options, Create Free Function and
Create Member Function, must be available.
2.3.4. Adding a Member Function
After a class has been created, member functions may be added
to it. The system must deal with the following cases:
• One or more (see [ISO11] 3.2, §5: One Definition Rule)
class definitions exist: add the new member function to all
definitions. Example:
19
CHAPTER 2. ANALYSIS
1
2
struct MyClass ;
struct MyClass {
3
4
5
8
9
10
11
2
3
};
struct MyClass {
6
7
1
4
5
6
};
void testMember () {
MyClass m ;
m . member () ;
}
7
8
9
10
11
Listing 2.21: before
struct MyClass ;
struct MyClass {
void member () {}
};
struct MyClass {
void member () {}
};
void testMember () {
MyClass m ;
m . member () ;
}
Listing 2.22: after
• Only a class declaration exists: replace the declaration with
a class definition containing the new member function. Example:
1
struct MyClass ;
2
2
3
4
5
6
7
1
3
void testMember () {
MyClass m ;
m . member () ;
}
Listing 2.23: before
4
5
6
7
struct MyClass {
void member () {}
};
void testMember () {
MyClass m ;
m . member () ;
}
Listing 2.24: after
In case of multiple declarations, one of the occurences is replaced.
• No declaration exists: do not offer a resolution in this case.
The class first needs to be declared properly to be able to
add members. Even if it is theoretically possible to create
both at a time, class definition and the requested member
inside, do not offer this. A resolution should only do one
thing at a time. Example:
20
CHAPTER 2. ANALYSIS
1
2
3
4
void t es tM em be rF un ct io n () {
MyClass m ;
m . member () ;
}
Listing 2.25: MyClass needs to be created before
members may be added
The system should be able to handle static member function
calls as well.
Editing Assistance
After creation of the new member, its declaration specifier, parameter types, parameter names and if available its return type
may be edited in linked mode. Assistance for editing the function
name may be left away because it is definitly non-ambiguous. An
example is shown in figure 2.3.
Figure 2.3.: Assisted editing in linked mode after function has
been created. Green: the exit position
Visibility
Members of a class are private by default and members of a struct
are public by default[ISO11]. If the member is created based on
a function call outside of the scope of the target class, it must
have public visibility. Otherwise it cannot not be accessed after
creation. Listing 2.26 describes an example of a generated member
function that needs public visibility to be accessible.
21
CHAPTER 2. ANALYSIS
1
2
3
4
5
6
7
8
class A {
void member () {
}
};
void testPublicMember () {
A a;
a . member () ;
}
Listing 2.26: A member that needs public visibility
Else, if a member is created from within the class it should be
private. An example is shown in listings 2.27 and 2.28. Existing
visibility labels should be reused if available.
1
2
3
4
5
class ATest {
public :
void testMember () {
member () ;
}
1
2
3
4
5
6
6
7
7
8
8
9
};
9
class ATest {
public :
void testMember () {
member () ;
}
private :
void member () {
}
};
Listing 2.27: before
Listing 2.28: after
2.4. Changing Arguments
Whenever a function is called with the wrong set of arguments, a
resolution may be offered to fix the problem. This situation occurs
if the arguments of a function call do not match the parameters
of an existing function with the same name. An example of a
function call with wrong arguments is described in listing 2.29.
1
2
3
4
void badArgs () { ... }
void testBadArgs () {
badArgs (42) ;
}
Listing 2.29: a function call with non-matching arguments
22
CHAPTER 2. ANALYSIS
The situations of non-matching function arguments may be categorized as follows:
• Calling function has more arguments than the declared function (see listing 2.29 above)
• Calling function has less arguments than the required declared function:
1
2
3
4
void badArgs ( int i ) { ... }
void testBadArgs () {
badArgs () ;
}
Listing 2.30: less arguments than parameters
• The number of arguments and parameters is equal but types
do not match:
1
2
3
4
void badArgs ( int i ) { ... }
void testBadArgs () {
badArgs ( " no Intege rArgum ent " ) ;
}
Listing 2.31: wrong arguments
In this case it is important to be aware that fundamental
types (int, char, double, etc... as described in [ISO11]
3.9.1) are not considered a type mismatch, so the following
code is compilable:
1
2
3
4
void badArgs ( int i ) { ... }
void testBadArgs () {
badArgs (4.2) ;
}
Listing 2.32: wrong arguments
Depending on the situation, arguments need to be added, removed or changed. For all cases, the option to create the missing
function and hence overload the existing function should always
be available.
23
CHAPTER 2. ANALYSIS
In case of function overloading, it is possible that the system
needs to handle more than one candidate function. In this case,
multiple resolutions may be displayed. The number of displayed
resolutions should be limited to a number of overviewable items.
To avoid a combinatorial explosion, to reduce the complexity
and to make the offered resolutions more foreseeable, combinations of the above three cases, listings 2.30, 2.31 and 2.32 should
not result in a resolution offered to the programmer. Take listing 2.33 as an example where no resolution to append arguments
should be offered. The only Quick Fix offered here is the create
member function.
1
void foo ( int i , int j ) { ... }
2
3
4
5
void testResolve () {
foo ( " string " ) ;
}
Listing 2.33: Add Argument plus Change Argument needed at the
same time
At last, a decision has to be made whether the arguments of
the calling function or the parameters of the function declaration
should be changed. On the one hand, changing parameters would
better support TDD because during TDD a system is designed.
On the other hand, changing parameters potentially breaks the
code and needs to be done carefully. Based on these arguments,
it was decided to only offer changing arguments.
If both changing parameters and arguments are implemented,
the resolutions offered to the user may look like in figure 2.4, a
screenshot taken from Eclipse JDT.
Figure 2.4.: Parameter Quick Fix examples from Eclipse JDT
24
CHAPTER 2. ANALYSIS
2.4.1. Removing Arguments
Whenever removing arguments can lead to a well-defined function call, an appropriate resolution may be offered to do so. An
example is shown in listing 2.34.
1
void foo ( int i ) { ... }
2
3
4
5
void testResolve () {
foo (42 , 43.0) ;
}
Listing 2.34: remove 43.0 to match foo(int)
What if the second argument were an integer too? It is not clear
which one to remove. In this case, remove the argument at the
end, assuming that backmost arguments are of less importance.
An example is described in listings 2.35 and 2.36.
1
2
3
4
5
void foo ( int i ) {
}
void test () {
foo (1 , 2 , 3) ;
}
Listing 2.35: before
1
2
3
4
5
void foo ( int i ) {
}
void test () {
foo (1) ;
}
Listing 2.36: after
A resolution is offered for each overload that matches at least
one argument and may be reached by removing an arbitrary number of arguments. Listing 2.37 shows an example where no resolution should be displayed because cropping arguments will never
lead to a correct function call.
1
void foo ( int i ) { ... }
2
3
4
5
void test () {
foo ( " noInt " , " noInt " , " noInt " ) ;
}
Listing 2.37: at least one matching argument needed
25
CHAPTER 2. ANALYSIS
2.4.2. Appending Arguments
Default arguments may be added to the function call to match
an overload of a function:
1
void foo ( int i ) { ... }
2
3
4
5
void testResolve () {
foo () ;
}
Listing 2.38: add a default argument 0 to the call to match
foo(int)
To reduce the number of combinations, Append Argument is
only offered if current arguments match the first parameters of an
overload. In this case, a valid function call may result by adding
new arguments.
Adding default arguments may be difficult in case of complex
constructors or when using pointers. However, in this case it
would be already useful to insert a variable name at the right
place. Based on this name, a local variable may be created in a
further step. An example of a nested constructor which results in
a complex default argument is shown in listings 2.39 and 2.40.
1
2
3
struct A { A ( int ) ; };
struct B { B ( A ) ;
};
struct C { C ( B ) ;
};
4
5
6
7
1
2
3
struct A { A ( int ) ; };
struct B { B ( A ) ;
};
struct C { C ( B ) ;
};
4
void testNest () {
C var ;
}
5
6
7
void testNest () {
C var ( B ( A (0.0) ) ) ;
}
Listing 2.39: missing argument Listing 2.40: added argument
2.4.3. Swapping Arguments
If a called function cannot be resolved but has an overload candidate which accepts the same number of parameters but in the
wrong order, reordering arguments is an option. The result of a
resolution on 2.41 shall look like 2.42.
26
CHAPTER 2. ANALYSIS
1
2
3
4
5
6
7
void foo ( A a , B b ) {
}
void testSwap () {
A a;
B b;
foo (b , a ) ;
}
Listing 2.41: before swap
1
2
3
4
5
6
7
void testSwap () {
A a;
B b;
foo (a , b ) ;
}
Listing 2.42: after swap
If the order of the arguments differs in more than two places,
the answer for correct ordering can be ambiguous. In ambiguous
situations, no resolution should be offered. See listing 2.43 for an
example.
1
2
3
4
5
6
void func (A , B , B , A ) ;
void testSwap () {
A a1 , a2 ;
B b1 , b2 ;
foo ( b1 , a1 , a2 , b2 ) ;
}
Listing 2.43: ambiguous argument swapping
It is arguable whether Swap Argument really supports TDD as
the calling code is adapted according to the declaration instead
of the other way around. Its implementation is optional.
2.4.4. Changing Types
Given an unresolved function call with correct number of arguments, wrong types and swapping not possible, a resolution may
be offered that changes the function call to an appropriate overload.
1
void foo ( Foo f , Bar b ) { ... }
2
3
4
5
void testResolve () {
foo (1 , 2.3) ;
}
Listing 2.44: Wrong argument order for foo()
27
CHAPTER 2. ANALYSIS
A resolution should fix the code to the function foo with types
Foo and Bar. The resulting code should look like in 2.45.
1
void foo ( Foo f , Bar b ) { ... }
2
3
4
5
void testResolve () {
foo ( Foo () , Bar () ) ;
}
Listing 2.45: Correct arguments to foo()
For this functionality too, it is questionable whether it is helpful
for TDD since it changes the caller instead of the callee. However,
it would be comfortable for updating tests in other situations.
2.5. Generating Variables
In this context, variables may be looked at as if they were functions without parameters. They have to be visible if they are
called outside of their defining class and they need a certain type.
Local variables are a special case with no visibility issues. Create Local Variable, proposed by Tomasz Wesolowski[Wes11], was
the first Quick Fix in CDT that was using the Codan infrastructure. However, it lacks macro support which is essential for
working with the CUTE framework. In addition, basic types are
only created as integers which is annoying if it were obvious that
a more precise data type could have been chosen.
Member variables called from outside of the class, need to be
publicly accessible as shown in listing 2.46.
1
2
3
4
class Type {
public :
int memberVariable ;
};
5
6
7
8
9
void test () {
Type t ;
t . memberVariable = 42;
}
Listing 2.46: Creation of a public member variable
28
CHAPTER 2. ANALYSIS
If the calling function is defined inside the same class, the new
member variable should be private (”shy code” [Tho05]) as shown
in listing 2.47.
1
2
3
4
5
6
7
struct Type {
void internalMember () {
this - > p r i v a t e M e m b e r V a r i a b l e = 42;
}
private :
int p r i v a t e M e m b e r V a r i a b l e ;
};
Listing 2.47: Creation of a private member variable
2.6. Changing Visibility
The code in listing 2.48 does call a not visible member function
in a type.
1
2
3
4
class Type {
void foo () {
}
};
5
6
7
8
9
void test () {
Type t ;
t . foo () ;
}
Listing 2.48: Member function foo() is not visible
This should produce an error and a Quick Fix should be displayed for changing the visibility of foo to public resulting in
listing 2.49.
1
2
3
4
5
class Type {
public :
void foo () {
}
};
6
7
void test () {
29
CHAPTER 2. ANALYSIS
Type t ;
t . foo () ;
8
9
10
}
Listing 2.49: Member function foo() was changed to public
2.7. Move Refactorings
In most cases, new code is placed right above the test function
definition. After the code has been implemented, it should be
moved to an appropriate source file.
The type to be extracted is chosen by the position of the cursor.
Corresponding to the scope where the cursor is placed the type
is extracted and the header file is named accordingly.
In the case of namespaces the namespace is not moved but the
namespace is created around the moved class in the newly created
header file shown in listing 2.52.
1
2
3
4
5
6
7
8
1
namespace N {
klass k {
| // <- cursor
};
klass j {
};
}
Listing 2.50: before extract
1
2
3
4
2
3
4
# include " klass . h "
namespace N {
// klass has been
// moved to klass . h
5
6
7
8
klass j {
};
}
Listing 2.51: after extract
namespace N {
klass k {
};
}
Listing 2.52: extracted klass.h (without include guards)
However in nested types the class definition which is most close
to the cursor is extracted as listings 2.53 and 2.54 show.
30
CHAPTER 2. ANALYSIS
1
2
3
4
5
6
class a {
class b {
class c {
}; | // <- cursor
};
};
Listing 2.53: before extract
1
2
3
4
5
6
# include " c . h "
class a {
class b {
// moved to c . h
};
};
Listing 2.54: after extract
2.7.1. Moving Functions
Newly created code inside a test file sooner or later needs to be
moved to a place where it belongs inside the project being tested.
This is why a Move Function Refactoring is needed that offers
moving the created code to another source file.
If code is moved, it is important to introduce an include preprocessor statement in the test file if needed.
2.7.2. Extracting Types to Separate Source Files
This feature basically takes the code of an type implementation
and moves it into a new source file. The file name equals the type
name. A problem is that the target project has to be chosen by
someone. A new class file along the test source files is not very
convenient since implementation project and and test project are
separated and the implementation would be in the wrong place.
31
CHAPTER 3. IMPLEMENTATION AND SOLUTION
3. Implementation and
Solution
This chapter describes how the analyzed features were implemented in Eclipse CDT.
A separate Eclipse plug-in was chosen in favor of changing and
patching CDT directly. This has the advantage that the plug-in
may be installed via an update site even if it is not integrated into
CDT. To reduce integration effort, code was developed against a
recent CVS tag of Eclipse CDT.
3.1. Approach
First features were added by registering Quick Fixes at the codanMarkerResolution extension point. All names inside the AST
that resolve to a IProblemBinding were already discovered and
reported by Codan checkers. This enabled a fast start for implementing Create Type, Create Local Variable and Create Free
Function.
Later, the logic that produces the problem markers, the Codan checkers, were extended to report more information needed
inside the Quick Fixes. For Quick Fixes with a dynamic number
of possible resolutions, like for Appending Arguments in case of
overloading, other solutions had to be implemented.
To be able to provide a larger set of Quick Fixes, multiple
features were developed iteratively and in parallel. This way, not
every possible special case is covered but it is more realistic to
provide a complete set of code generation Quick Fixes for TDD.
32
CHAPTER 3. IMPLEMENTATION AND SOLUTION
3.2. Prerequisites
What techniques are needed to implement the discussed features?
First, there has to be an infrastructure that allows to analyze
the code in form of an abstract syntax tree. Second, some functionality is needed that scans this structure for potential errors.
Those errors have to be made visible to the programmer in some
way. Last, functionality is needed to offer a resolution at the error
location.
3.2.1. Source Inspection
The C++ parser of Eclipse CDT provides an abstract syntax tree
(AST) that can be traversed using the visitor pattern. The nodes
inside this tree must not be edited directly. Instead, a rewriter is
provided to alter code based on the nodes of an AST.
3.2.2. Problem Detection
Eclipse CDT already contains an infrastructure to scan the AST
for nodes that are in some error state. The AST itself provides
information about names that could not be resolved by trying
to match names to their underlying definition and declaration.
The Codan subproject has built-in checkers which collect errors
and produce a warning inside the source code by underlining the
problematic text passages. Additionally red markers were placed
at the affected line of code to indicate to the user that there is a
problem in this section.
Figure 3.1.: A checker reports multiple problems
Figure 3.2.: A checker reports multiple problems
33
CHAPTER 3. IMPLEMENTATION AND SOLUTION
3.2.3. Resolutions / Quick Fix
If a problem has been found and there is a solution for it Codan
displays a yellow light behind the red marker to indicate that
there is a problem.
Figure 3.3.: A light indicates that there is a solution to a found
problem
The terms Resolution and Quick Fix describe the same idea.
Resolution is the term used inside CDT source code and Quick
Fix is the name presented at the user interface.
Such a resolution can be called by hitting Ctrl-1 if the cursor
is at the right place in the source code.
Figure 3.4.: Multiple resolution to a missing name
3.2.4. Code Analysis and Change
If the programmer decides to apply such a Quick Fix on a certain
source selection, a mechanism is needed for analyzing what code
has to be generated or changed and at which position it should
be inserted. Such an analysis is done using visitors on the AST
to collect information about a situation and in which context the
Quick Fix is called and trying to generate the expected result.
Figure 3.5.: Resolution to create a type is displayed
34
CHAPTER 3. IMPLEMENTATION AND SOLUTION
Changes to the code are done carefully. Impacts of a Quick
Fix are kept as local as possible. This has several reasons. First,
mistakes are kept local and second, the programmer does not
need to switch the source code file to control or correct the newly
generated code. The programmer can stay in the test file until he
integrates his changes permanently. Until then there is no need to
change the source file, minimizing context switch and maximizing
focus on the current work.
Figure 3.6.: A new type was generated above the test
3.2.5. User Guidance
After code has been generated, some situations require the programmer to review the changes or to take a decision, like choosing
a meaningful name for a newly created variable. To avoid user
dialog, linked mode of Eclipse jFace[Ecl11d] may be used that
allows to toggle between specific text selections. linked mode is
probably best known from the Rename Refactoring both in CDT
and JDT.
Linked mode is explained in more detail in section 3.8.5.
3.3. Used Extension Points
Extension points [Ecl11f] provide a way to extend Eclipse without
changing its source code. Instead of hard-coding changes, new
features can be added by adding an entry inside the plugin.xml
file. Using a certain interface allows communication between the
extension point and the new feature.
Extension Points were used to add Quick Fixes, checkers and
the key binding for the Extract Type / Function features. They
are explained in the following sections.
35
CHAPTER 3. IMPLEMENTATION AND SOLUTION
3.3.1. Checkers
To add a new checker to the Codan infrastructure, first the specific checker name must be registered at the Extension Point:
org.eclipse.cdt.codan.ui.codanMarkerResolution. This is done
via the plugin.xml in the plug-in project. In addition it must
be specified there what kind of problem the checker reports, his
name and the severity. (error or warning).
Figure 3.7.: Registering a checker at the extension point
1
2
3
4
5
6
7
8
9
10
11
12
< extension point = " org . eclipse . cdt . codan . core .
checkers " >
< checker
class = " ch . hsr . eclipse . cdt . codan . checkers .
MissingOperatorChecker "
id = " ch . hsr . eclipse . cdt . codan . checkers .
missingoperatorchecker "
name = " Missing Operator Checker " >
< problem
defaultSeverity = " Error "
id = " ch . hsr . eclipse . cdt . codan . checkers .
MissingOperatorResolutionProblem_HSR "
name = " Operator could not be resolved " >
</ problem >
</ checker >
</ extension >
Listing 3.1: Registering a checker with Codan extension point
For the logic of the checkers, the class AbstractIndexAstChecker
was extended. The checker processes the AST with a visitor and
collects information about context and tries to find out problems
by either resolving names or more complex methods.
36
CHAPTER 3. IMPLEMENTATION AND SOLUTION
3.3.2. Codan Marker Resolution
New Quick Fixes are added by implementing the ICodanMarkerResolution interface. Like new checkers, the implementing class
must be registered at the plugin.xml configuration file in the
plugin project.
Figure 3.8.: Registering a marker resolution at the extension point
Each Quick Fix is registered to a certain problem id which a
checker reports. The plugin.xml file acts as the adapter between
checkers and Quick Fixes.
1
2
3
4
5
6
< extension point = " org . eclipse . cdt . codan . ui .
codanMarkerResolution ">
< resolution
class = " ch . hsr . eclipse . cdt . createtype .
C rea te Ty pe Qu ic kF ix "
problemId = " ch . hsr . eclipse . cdt . codan . checkers .
TypeResolutionProblem_HSR ">
</ resolution >
</ extension >
Many problem ids that are interesting for offering Quick Fixes
are listed and reported in the class ProblemBindingChecker.
Because the checker class cannot not be altered directly, all Codan bindings are disabled inside the Activator class and replaced
by problem ids ending in ” HSR”. All problem ids are referenced
in the ProblemIdCollection class located in the plug-in package
for convenience.
3.3.3. Marker Resolution Generator
The resolution generator of Codan has a problem. It does not
allow a dynamic number of the same Quick Fix to be displayed
for a single reported problem. Why would anyone need this?
37
CHAPTER 3. IMPLEMENTATION AND SOLUTION
Having overloaded functions with different argument and a wrong
function call with no arguments.
1
void func ( int ) {}
2
3
void func ( int , int ) {}
4
5
6
7
void call () {
func () ;
}
Listing 3.2: Overloaded functions and a wrong function call
Here two Quick Fixes need to be displayed, first for the function
with one and second the function with two arguments. This is
not possible with the standard Codan way of registering a single
instance of a Quick Fix to a number of same problem ids.
Figure 3.9.: Registering a marker resolution at the extension point
org.eclipse.ui.ide.markerResolution is used to implement
a separate resolution generator for Appending Argument. This is
needed to display multiple times the same Quick Fix .
1
2
3
4
5
6
< extension point = " org . eclipse . ui . ide .
markerResolution " >
< markerResolutionGenerator
class = " ch . hsr . eclipse . cdt . addArgument .
AddArgumentQFGenerator "
markerType = " org . eclipse . cdt . codan . core .
codanProblem " >
</ markerResolutionGenerator >
</ extension >
Listing 3.3: Registering a resolution generator
38
CHAPTER 3. IMPLEMENTATION AND SOLUTION
3.3.4. Commands, Bindings, ActionSets
These are the default extension points for adding commands to
menus and key bindings in Eclipse.
3.3.5. Quick-Fixes for a Specific messagePattern
Optionally, the xml attribute messagePattern can be used to
offer more specific Quick Fixes for the same problem id depending
on the marker message.
1
2
3
4
< resolution ...
messagePattern = " Function ’* ’ is undefined " >
</ resolution >
</ extension >
This feature was not used but could be considered if the plugin is integrated into CDT. Be aware that the messagePattern
attribute of a problem id has another meaning.
3.4. Marker Resolution Refactorings
Behind every Quick Fix, a class derived from refactoring classes
does change the source code. However Marker Resolution Refactorings are not really refactorings but are the same infratstructure
to rewrite the source code in the editor. These refactoring classes
contain the logic of the Quick Fixes and are the heart of the developed plug-in. This section introduces the implemented marker
resolution refactorings.
3.4.1. Create Type or Namespace
Weather a type or a namespace should be created is pre-determined
in some situations. Hence, it is possible that both Quick Fixes
are offered to the user.
This is also discussed in section 2.2.3
3.4.2. Append / Remove Argument
Append Argument is special because it is the only Quick Fix that
is offered multiple times for a single problem marker. Hence, the
39
CHAPTER 3. IMPLEMENTATION AND SOLUTION
refactoring has to know which of these Quick Fixes was applied.
To find out which overload should be used, each Quick Fix is
aware of the candidate number it is linked to. The refactoring
retrieves the sorted list of candidates itself independently and uses
the passed in number to find the same candidate again.
Figure 3.10.: Offering three times the same Quick Fix type
3.4.3. Create Function
This is the most complex set of Quick Fixes. Different refactorings are needed for free, member, operator, constructor and
static functions. The latter function types mostly differ in their
way of creating the newly inserted function definition. This logic
has been wrapped in a strategy pattern and classes have been
grouped into three packages: common, strategy and Quick Fix
classes. Figure 3.11 describes the package structure with the different function creation strategies.
The different Quick Fix classes are responsible for showing an
appropriate image and label and each declare which strategy
should be used. Figure 3.12 visualizes the role of the strategy
during the calculation of changes inside the refactoring object.
The FunctionCreationStrategy is the most common of the
strategies and is used for creating member, non-member and indirectly also for static functions. For creating operator and constructor function definitions, different functionality was required
each.
40
CHAPTER 3. IMPLEMENTATION AND SOLUTION
Figure 3.11.: Package overview for the create function features
3.4.4. Create Variable
Create Local Variable was introduced by the CDT community
at the time of the start of this thesis. Nevertheless, the Quick
Fix was continued since it has better return type recognition for
basic types (boolean, double, etc.) and linked mode was added
for convenient change of type. However a linked mode position
to change the name was not implemented since the name was
already chose by writing it before Creating Variable.
For creating member variables another refactoring was developed but placed into the same package since it also creates variables.
41
CHAPTER 3. IMPLEMENTATION AND SOLUTION
Figure 3.12.: Sequence diagram of the function creation process
3.4.5. Change Visibility
This refactoring is not directly useful for TDD but the visibility
logic was needed for other refactorings.
3.4.6. Extract Type and Function
This refactoring is the only one which is not invoked through a
Quick Fix but via a key combination (Alt-Shift-P was free, stands
for extract type). The extract refactoring does both, extract types
and functions, since behavior is almost the same.
The extract refactoring supports moving functions and types
into a separate header file. However, it is only a small step
towards a good Move Refactoring which is required for convenient
TDD.
3.5. Architecture
A lot of architecture is already predetermined by Eclipse, CDT
and Codan. This section discusses the overall architecture of the
42
CHAPTER 3. IMPLEMENTATION AND SOLUTION
developed plug-in.
3.5.1. Package Structure
The classes are structured into several packages of separated interests. Classes used by more than one parts were moved to the
top level package.
All checkers were put into a single package as detecting errors
and correcting them are two different parts.
The Quick Fixes which represent the user interface part of the
plug-in, the logic that decides whether to provide a Quick Fix and
the actual logic for changing the source code are located together
in a single package for each provided Quick Fix .
3.5.2. Class Diagram
Each Quick Fix consists of at least two classes, extending MarkerResolutionRefactoring or RefactoringLinkedModeQuickFix
respectively. Latter uses the ChangeRecorder to find out the positions to be shown in linked mode based on the changes generated
by the refactoring class.
Separate of the Quick Fixes is the extract functionality which
is registered to a key binding instead of a problem.
Checkers all define a visitor that handles problems on the AST.
Each visitor extends class AbstractResolutionProblemVisitor
which helps filering problem nodes.
Figure 3.13 shows all important classes of the plug-in without
helper classes.
3.5.3. Basic Call Flow
Whenever the user makes changes inside the C++ source code
editor, a Codan reconciling job is scheduled. After 50ms of pause
without typing, this job starts. After the Abstract Syntax Tree
has been updated, the Codan reconciler job runs all registered
checkers, reporting problems to the central CodanMarkerResolutionRegistry. See figure 3.14. A list of all markers and resolutions
is requested by the editor and displayed appropriately. This process is visualized in figure 3.15.
43
CHAPTER 3. IMPLEMENTATION AND SOLUTION
Figure 3.13.: Reduced class diagram of the entire plug-in
Checkers produce a problem marker which can be resolved by
applying a Quick Fix . The Quick Fix in turn performs a certain
refactoring, depending on the problem. Inside the refactoring the
affected problem node is analyzed, a new node is created and a
suitable insertion point is determined.
The refactoring returns a set of changes which is not yet performed. Changes are actually performed by the Quick Fix class
inside a ChangeRecorder. The information gained by the ChangeRecorder is used to calculate offsets of newly inserted code for
use with linked mode editing. See figure 3.16.
44
CHAPTER 3. IMPLEMENTATION AND SOLUTION
Figure 3.14.: The AST is updated after code changes, then checkers are invoked
3.5.4. Design Decisions
The choice of architecture is critical to performance and influences
the flexibility of the Quick Fixes. The following sections discuss
implications of the existing and chosen design.
More Logic for Codan Checkers
The need for more logic inside checkers actually rises from the
nature of Quick Fixes. The method isApplicable() of the Quick
Fix is called by the user interface for every displayed marker and
therefore needs to be very responsive. Otherwise the GUI is frozen
until it can display all the labels. This also means that access
to the AST is not realistic because it is too slow to retrieve
information about the situation. Therefore all information needed
to display in the label of the Quick Fix is passed in inside the
marker.
45
CHAPTER 3. IMPLEMENTATION AND SOLUTION
Figure 3.15.: Registered Quick Fixes are searched for each problem marker
Passing Information to the Quick-Fix
A Quick Fix needs to know at least two details: First, if it is
applicable and second, what its label is for a specific marker.
This information has to be passed to the Quick Fix as it should
not access the AST itself.
A checker has three ways for passing information to a Quick
Fix. Either the checker reports more granular problems, the messagePattern attribute is used or more information is passed via
the problem markers argument list. Because the marker message
is displayed to the user, it should not be used to transport internal details. Multiple problem ids for the same problem are
confusing so it was decided to pass internal information via the
open marker parameter list. Listing 3.4 shows how problems are
reported inside a checker.
1
2
3
4
5
6
A b s t r a c t I n d e x A s t C h e c k e r . reportProblem (
< Error ID > ,
< IASTNode to be marked with red underline > ,
< hover message to be displayed > ,
[ user defined arguments , ...]
);
46
CHAPTER 3. IMPLEMENTATION AND SOLUTION
Figure 3.16.: User applies a Quick Fix
Listing 3.4: Reporting a problem inside a checker
Other Benefits of Extended Checkers
Checkers with more logic have other benefits too. As the first
argument of a reported problem is used as the hover message of
a marker, the checker is able to chose a more specific message.
In addition, because the checker has access to the AST, it can
already decide about the needed resolution strategy which allows
to produce a more user friendly Quick Fix text.
Checkers cannot pass whole ASTNodes to the Quick Fix because
if the AST is destroyed after a marker has been created, the
marker still holds a reference to an ASTNode which does not exist
anymore. This means that even if the AST is not valid anymore,
47
CHAPTER 3. IMPLEMENTATION AND SOLUTION
Figure 3.17.: Hover messages are defined by the problem markers
checker
there exists a marker with indirect references to the AST which
cannot be freed by the garbage collector.
To prevent such a memory leak and access to an invalid AST,
all arguments appended to a reported problem are serialized and
passed as a string to the marker.
This has some consequences to the whole plug-in. Every problematic node found in a checker needs the be searched again in
the same AST. Since most nodes have a selection, a position in
the text of the source file, they can be retrieved again relatively
easy. However passing additional information like other nodes in
the context of the problematic node is not possible and should be
avoided, because all non-serializable Information is lost between
detecting the problem and reacting to it.
Keeping this in mind, gathering too much information in the
checker is useless as long as it cannot be passed to the Quick Fix.
Also it makes it hard for sophisticated resolution generators to
generate appropriate solutions to a problem. A solution maybe
depends on the situation where the problem occurred. This leaves
no other option than passing at least some of the context information as a string to the marker to search and generate a resolution.
Single QuickFix for Multiple Problems
The design of the Codan resolution generator implies that only
one Quick Fix exists for a whole class of problems with the same
problem id. Since the apply() method of a Quick Fix takes a
marker as argument, one Quick Fix is basically enough.
However, the getLabel() method, as defined in the Eclipse
framework (not CDT or Codan), does not take a marker as an
argument. This makes it uncomfortable to display the name of a
missing function or variable in the label. This is another problem from the fact that there is just a single instance of a Quick
48
CHAPTER 3. IMPLEMENTATION AND SOLUTION
Fix created by Codan. If every resolution would be newly created from a resolution generator, a label could be passed in the
constructor of that
It is possible to back up the marker in the last isApplicable()
call. This way, specific labels may be displayed even with only
one Quick Fix per problem type. This is definitly a hack and fails
if isApplicable() is not called directly before getLabel().
MarkerResolutionGenerator for Append Argument
Add Argument needed its own marker resolution generator. Why?
The problem with the resolution generator of Codan is, it can only
produce one Quick Fix per problem. In case of multiple overloads,
different Add Argument Quick Fixes have to be displayed. This
could be solved by reporting the problem several times for each
overload.
Disabled Codan Problems
Since a separate plug-in was developed and Codan checkers were
duplicated to change their behavior, the old checkers are disabled.
This prevents from the issue that some problems got reported
twice. It is done inside the Activator class of the plug-in. There
a listener waits for Codan to be loaded and then disables all
Codan problems since checkers can only be disabled if all there
to be reported problems were disabled. From this time on only
the self written checkers are active and report problems.
3.5.5. Important Class Overview
Here is a short description about some classes used in the implementation and their duties.
Checkers
To write a checker, an AST-visitor[Gam95] is defined that resolves
the binding for every IASTName inside the AST. If the binding cannot be resolved properly, the returned binding is of the
type IProblemBinding, containing additional information about
the problem. Depending on the type of the problem binding,
49
CHAPTER 3. IMPLEMENTATION AND SOLUTION
problem markers with different problem ids are reported. A more
detailed description is found in section 3.3.1 how checkers work.
QuickFixes
Each resolution action for a problem is implemented as a separate Quick Fix. This is the simple most way to decide which
problem is solved by which Quick Fix, statically defined in the
plugin.xml. The Quick Fix class itself is the glue between the
reported problems and the logic to correct the problem. As a side
effect it displays information about the correction to be done. Due
to the fact that it is not straight forward to dynamically display
information about the Quick Fix in the label various tricks were
applied to this class. See section 3.5.4 how this is done.
CodanArgument
To achieve a proper refactoring some information is needed from
the marker. Markers do support open argument lists to pass
additional information. This information, no matter what kind
of object, will be serialized. See section 3.3.1 for this marker
problem.
A lot of problems came up with retrieving information with
the wrong position from the argument map of the marker. To
avoid this the CodanArgument class was created. It serves as
an adapter before and after serialization. Passing information to
CodanArgument will put the information in the correct order.To
retrieve information from a marker, the marker is passed to the
constructor of CodanArguments through which the information of
the marker can be fetched.
Refactorings
The logic to correct the problem is done in a refactoring. Early
versions of some refactorings used a text replacement method but
were removed from our package since this is a quick and dirty
way of correcting a problem.
The proper way to do this and how it is done now is trying
to collect all needed information from the context of the AST,
50
CHAPTER 3. IMPLEMENTATION AND SOLUTION
creating new nodes, insert them into the AST and let the AST
infrastructure rewrite the code.
Quick Fix Generator
The static assignment between problems and Quick Fixes may be
more memory efficient but raises a big problem. See section 3.5.4
With the AddArgumentQFGenerator it is shown how dynamically
a Quick Fix is put in context with a problem.
Strategies
Creating the different kinds of functions, like operators or static
functions would produce a lot of duplicated code as already mentioned in 3.4.3. For Example operators can be implemented as a
member or as non-member function. The usual implementation
approach to have such shared logik did not produce the expected
result of a good readable code. It was decided to use a strategy
based implementation which results in returning the various kinds
of functions: Operators, static functions or normal functions. The
actual refactoring then retrieves the created function an places it
in the correct member or non-member context.
ChangeRecorder
To enable linked mode, exact positions of the newly inserted code
elements have to be known. ChangeRecorder is responsible for
determining these positions. Before this class was created, several
means to find the correct locations were analyzed:
1. Search the document for the inserted name
2. Guess positions with the knowledge that code is inserted at
a certain offset
3. Save a reference of the newly inserted element and read its
location after insertion
4. Wait for the AST to be reconciled and search for the affected
node
5. Analyze generated changes
51
CHAPTER 3. IMPLEMENTATION AND SOLUTION
The first is the fastest way to get the locations, however it is
obvious that this solution will break as soon as another element
has the same name. Item two was the first implemented approach
which worked well in first place. However this solution was too
unstable because the insertion point was too variable. The third
approach failed due to the fact that newly inserted nodes are
disposed after insertion and new nodes (with new references) are
created by the parser when the AST is reconciled. Searching for
the new node inside the reconciled AST was also tried. However,
this solution introduced a delay of aound half a second before
linked mode could be started which is not acceptable.
In the end, analyzing the generated changes was the last alternative. It introduces an indirect coupling to the ChangeGenerator. However, generated code must stick to syntax rules, so only
whitespaces may change. Only one feature complicates extracting
the changes: To avoid that ChangeGeneratorVisitor reformats
the whole file, generated changes are reduced to the smallest necessary text edit. Most complexity of ChangeRecorder origins from
the structure of these text edits. For example, figure 3.18 highlights the contents of a change during an overload of a member
function.
Figure 3.18.: Reduced text changes provided by ChangeGenerator
3.6. Testing
This project could not have been created without the use of tests.
About 230 tests ensure correct functionality of the implemented
features. This way, even at the end of the project, refactorings
could be safely executed. In this section, the different kinds of
tests run throughout the project are discussed.
52
CHAPTER 3. IMPLEMENTATION AND SOLUTION
3.6.1. Unit Tests
Unit tests profit from the advantage of being able to run without
an Eclipse instance. They run through instantly, however not all
functionality could have been tested this way.
Unit tests were used to ensure correct positions for linked mode
editing. The changes created by the refactorings are subject to
change and the mechanism that analyzes changes needs to be
flexible especially in connection with whitespaces.
3.6.2. Refactoring Tests
Refactoring tests are ensuring correctness of the Quick Fix logic.
Some of the tests are explicitly written to test compatibility with
CUTE macros. To avoid importing all CUTE headers, a simplified
CUTE structure was used together with the possibility to put
them as external files and include them in the various tests if
they need it. Earlier without the possibility of the external file
test infrastructure, every definition of the specific cute macro was
included in every test which needed the macro. Using the external
file infrastructure reduced the test code for about 400 lines of
code.
Later in the project it was noticed that these refactoring tests
were very slow and slowed down the development a little bit.
Waiting 120 second for tests to complete is no option since it
leads to the behavior of postponing the tests.
However the project was analyzed with the Testing and Performing Tools Platform [Ecl11e]. It was discovered that opening
a document in the IDE to retrieve the marker and the selection
is very inefficient and slow. The code was refactored so that the
document was created with the content of the tests and the selection was extracted directly from the Marker, rather than using
the document.
With this refactoring, tests were sped up by about 30 percent.
3.6.3. Integration Tests
For each Quick Fix, an integration test checks for a specific code
example whether
53
CHAPTER 3. IMPLEMENTATION AND SOLUTION
• markers of the appropriate type are generated
• the positions of the markers match the actual position of
the problem
• the message of the marker is correct
• the Quick Fix message is correct
• the correct code is generated
• there is a displayed image in the Quick Fix label
Also later in the project the integration tests ran too slow.
They were examined and it was detected that there is no need to
run the indexer[Ecl11c] before every check on the marker position
or a correct marker message. Running the indexer once before all
tests is enough.
3.6.4. Tests on Large Projects - Performance
To test how the implemented checkers and Quick Fixes behave in
a bigger project, Webkit[web10] was analyzed.
Webkit
Despite some rumors from mail on the CDT-mailing list[Esk11],
Codan behaved quite performant in this test.
Various problems and bugs were found and fixed. Webkit makes
intensive use of friends. To prevent a lot of false-positive messages a feature to not complain about accessing private members
from outside if it is a friend class was implemented.
False-positive messages all about constructors were detected
and could be fixed after the test. The problem resides in the
fact that a declaration of a variable in a function can be constructor call weather a declaration inside type definition is not a
constructor call.
However some limitation were discovered in context with templates. With multiple nested instances of vectors or other data
structures using iterators and operators. What CDT does is trying to instantiate the templated classes during ”coding time” and
54
CHAPTER 3. IMPLEMENTATION AND SOLUTION
tries to resolve functions to see if they match. This functionality
is needed for the auto completion but does only work to a certain
level of template nesting. The Operator checker were not able
to detect the defined operators on such a nested templated data
structure like vector<vector<pair.
3.7. Known Issues
A lot of time was invested to avoid false positive on checkers and
the generation of wrong code. But since no software is perfect
and time is limited some issues remain unresolved.
3.7.1. Checker Performance
Static analysis is most helpful if it is done while typing. However,
for big projects, checking the code for errors consumes too much
time. Checkers may be turned off at the code analysis preferences
page.
3.7.2. False Positives
It is possible that checkers are generating false positives - markers
for problems that do not exist. Even if the compiler is able to
build the code without errors, false positives will place red markers on affected source lines. If a specific checker is producing false
positives, it may be turned off using the code analysis preferences
page and a bug should be reported in Bugzilla[Ecl11b].
3.7.3. Chaos with Basic Types
During analysing code of types it was detected that in CDT there
seems to be a chaos how types like int, float or double are
defined. Especially with arguments it is important to compare
two basic types.
One way of the definition is in the class IASTSimpleDeclSpecifier. There, basic types are defined as constant integers. Then
there is the interface IBasicType which defines a new enum called
Kind in which all the basic types are defined again. BasicType defines the kind of basic types for example for parameters
55
CHAPTER 3. IMPLEMENTATION AND SOLUTION
(ICPPParameter). BasicType was once used to return the same
integer value as IASTSimpleDeclSpecifier but this getType()
method is marked as deprecated.
So if one wants to compare a ICPPParameter with a IASTSimpleDeclSpecifier there is no chance not to run into a marked
as deprecated warning.
3.7.4. Stopping at Typedef
CDT has good support for detecting and handling typedef. For
example overwriting the output operator operator<<() as a free
ouput operator, the stream to show the information usually is
std::cout. Now the type of std::cout is defined as a std::ostream.
However, std::ostream is a typedef which looks like this.
1
typedef basic_ostream < char > ostream ;
Listing 3.5: typedef of std::ostream
This means when detecting the return type in a statement
that looks like the following listing 3.6 it would insert a basic
ostream<char> for the operator<<() while digging out the
required return type.
1
2
3
4
void test () {
A a;
std :: cout << a ;
}
Listing 3.6: Instance of type a should be send to the output
For sure, nobody does that and it looks ugly.
So, the easiest solution is to stop at the typedef and not dig
deeper to the actual type.
This however brings up another problem. Consider the following code:
1
2
struct A {
};
3
4
typedef A X ;
5
6
void test () {
56
CHAPTER 3. IMPLEMENTATION AND SOLUTION
X a;
Y a = a . giveMeA () ;
7
8
9
}
Listing 3.7: Typedef’ed class
Now when creating the giveMeA() function it expects a return
type of A. But considering the typedef between the definition and
the actual call, the result looks much more like the listing 3.8.
1
2
3
4
5
struct A {
X giveMeA () {
return * this ;
}
};
6
7
typedef A X ;
8
9
10
11
12
void test () {
X a;
Y a = a . giveMeA () ;
}
Listing 3.8: Typedef’ed class returning itself
This dilemma was detected quite late and remained unresolved
at the time of documenting.
3.7.5. AST Parser and Syntax Errors
When creating a whole templated class by writing the class name
and the actual parameter type as template argument there is no
problem in detecting the class name and the template parameter.
See section 2.2.6
1
2
3
void test () {
A < int > a ;
}
Listing 3.9: Instantiating a templated class which does not exist
This however can be detected and a templated class can be
created without a problem. See section ??
57
CHAPTER 3. IMPLEMENTATION AND SOLUTION
If on the other side an instantiation of the following code in
3.10 is written, the Parser is unable to detect that this is a class
instantiation with an non-type parameter. In this case the AST
is build as: Problematic AST name ’A’ operator<5 operator>.
1
2
3
void test () {
A <5 > a ;
}
Listing 3.10: Instantiating a templated class with non-type
parameter
Even stranger, if the class is defined as in listing 3.11, the
parser has no problem detecting that it is a non-type parameter
template instantiation. This problem remained unresolved.
1
2
template < int i >
class A { ... };
Listing 3.11: Class defined with a non-typed template argument
3.8. Solved Issues
This section explains some problem encountered during the project
which could be solved.
3.8.1. Detecting Constructors
Detecting constructors and its callers is a little different than detecting unresolved normal functions or variables. Since there is
no AST element identifying a constructor call, the AST must be
carefully analyzed. To find a matching constructor which is already defined, there can be used a function called findImplitlyCalledConstructor() in the class CPPSemantics. This function
takes a Declarator as an argument. However, having no constructor found does not mean that this is no constructor defined.
Declarators do appear a lot in the AST. Therefore, the AST is
checked on a specific combination of some AST elements which do
only appear in constructor call. Then this combination is reported
as call to a undefined constructor.
58
CHAPTER 3. IMPLEMENTATION AND SOLUTION
3.8.2. Detecting Operators
Like detecting constructors, detecting operators is a non-trivial
business but a little easier.
Generally all expressions in an AST are analyzed. If such an
expressions is ether of type IASTUnaryExpression of IASTBinaryExpression, then the implicit names of the expression is
retrieved. The following code in listing 3.8.2 and 3.8.2 shows the
relation between expressions and implicit names.
1
2
3
4
void test () {
A a;
a ++;
}
Such a expression is actually nothing else than:
1
2
3
4
5
void test () {
A a;
a . operator ++( int ) ;
}
Since the definition of a operator looks like:
1
2
3
4
struct A {
A operator ++( int ) {
}
};
Such a implicit name is only found if an operator already exists.
If no operator is found the name of the operator is copied with
the help from the class OverloadableOperator and reported.
3.8.3. Deleting Function Creates no Marker
Detecting problems with missing names and other issues depends
on resolving the corresponding binding for a name. Resolving
names uses the index to check the bindings. If the indexer settings
are not set correctly problems like missing functions were not
detected. The indexer needs to be automatically updated.
59
CHAPTER 3. IMPLEMENTATION AND SOLUTION
3.8.4. Codan Marker Patch
While writing the Quick Fix for creating new functions, odd behavior was detected when macros were used. The reported problem marker position included the whole macro instead of the pure
function name. This way it was difficult to determine the name of
the function to be created. To find the correct name, the problem
binding had to be searched again inside the Quick Fix . To solve
this, it had to be found out why codan was reporting the wrong
positions.
The ProblemBindingChecker of Codan relied on the getRawSignature() method of IASTNode, to determine the name of an
affected node. However, if a node is contained inside a macro,
getRawSignature() returns the text of the macro node instead
of the actual name of the node. To solve the problem, getSimpleID() method of IName was used to get the name of the
node. Another solution would have been to change all getRawSignature() methods which could have been very laborious.
The Problem has been reported in Bug 341089[Kal11] and a
patch has been filed.
3.8.5. Create Linked Mode Positions
Whenever code is generated, the programmer may adapt some
parts of the new code: Function parameters may be renamed, the
return type may be changed or a class may be chosen in place of
a struct. To enable this functionality a mechanism called linked
mode is provided by Eclipse JFace.
Linked mode has three powerful features: First, it allows to
edit identical text parts simultaneously. Second, it allows to jump
between important edit positions. Last, an optional exit, position
may be defined to determine the place where the editors cursor is
place after the return key has been pressed.
Gathering the Text Positions
It is difficult to find the right positions of the newly inserted
nodes in a timely manner. A straightforward solution would be
to use the AST. However, the AST first has to recover from the
perfromed changes and the linked mode has to be available to
60
CHAPTER 3. IMPLEMENTATION AND SOLUTION
Figure 3.19.: Two linked mode positions and an exit position
the user immediately. Guessing the number of whitespaces would
not provide a stable solution. To have immediate access to the
positions, the text changes provided by the rewriter[Gra06] are
analyzed to determine the linked mode positions.
The problem has been solved with a class ChangeRecorder that
wraps the original changes provided by the rewriter. Using the
size of the inserted text and the name of the inserted node, the
ChangeRecorder is able to calculate the code positions for the
linked mode.
61
CHAPTER 4. CONCLUSIONS, INTERPRETATION AND
FUTURE WORK
4. Conclusions,
Interpretation and Future
Work
A lot of features have been implemented, problems have been
encountered and experiences have been collected throughout this
thesis. This chapter summarizes what has been reached and where
future potential is possible.
4.1. Conclusions
It was encouraging that even during implementation, Quick Fixes
were already helpful for writing new test cases. However, what
has been done cannot be regarded as ”complete” TDD support.
There is still a lot of work to do and functionality which is not
obviously supportive for TDD is also needed to provide a full set
of Quick Fixes for writing software test-first. For some features,
it is difficult to decide whether they support a test-first or testlast approach, as the discussion around Change Argument versus
Change Parameter showed. In the end, every programmer will
use the provided Quick Fixes in its own way.
As JDT demonstrates, there is far more left to do than the
few Quick Fixes implemented during this bachelor thesis. As the
choice of implemented features has impact on code quality and on
checker execution time, future projects should think about what
should be implemented and what not. At this point, we would
like to refer to the ”assignment-to-itself-checker”.
Execution speed during manual and automated testing was satisfying, even for large projects like Webkit. However complexity
grows linearly as more checkers will be added and at some point,
one will have to decide which checkers are really needed.
62
CHAPTER 4. CONCLUSIONS, INTERPRETATION AND
FUTURE WORK
The TDD for CDT plug-in supplements the Toggle Function
Definition [KS10, Som11a] which has been developed in 2010.
Both features in combination are a powerful tool for fast development.
4.2. Interpretation
This section reflects the opinion of the authors about the developed project.
4.2.1. Limitations
There was a big potential for supporting special cases that will
never be used. Therefore, focus has been set on implementing
basic functionality for a broad set of Quick Fixes . Special cases
may be added later if they are needed. Hence, it is possible that
checkers will produce false positives or false negatives for some
code.
Add Argument has been artificially limited by inserting ” ” instead of a default constructor even if it was possible to determine
such a default value. We think, this simplification is needed because the programmer should program deliberately. In Java, the
Quick Fixes that insert null references is a dangerous feature and
we are convinced that the selection and design of Quick Fixes has
influence on clean code.
At last, the fact that own checkers have been implemented that
in some cases duplicate functionality from Codan will make integration into CDT harder. CDT community has to be convinced
that these checkers are needed to provide meaningful labels for
both markers and Quick Fixes .
4.2.2. Personal Review
We remember that this section was the first one we have read
when we were browsing references. Following a short and personal
summary of the authors for future CDT plug-in thesis students.
63
CHAPTER 4. CONCLUSIONS, INTERPRETATION AND
FUTURE WORK
Martin Schwab
Working with Eclipse is challenging because it is such a large
project. A lot of functionality already exists at some place and
may be reused. In some cases, e.g. the definition lookup mechanism, code is available but bound together with other functionality so it cannot be reused. This inspired a lot of discussions
about code smells, refactoring and architecture which was very
interesting.
Although after a while, adding new Quick Fixes was pure work,
new challenges had to be solved throughout the project, especially
with Add Argument.
Personally, I enjoyed working with legacy code over designing
a new system from scratch. Eclipse is decorated with concrete
implementations of patterns and a lot can be learnt by browsing
its code. Although there is still is a long way to go to have
functionality like in JDT, I would be happy if with this project a
first stone had been laid down for future Quick Fixes in CDT.
Thomas Kallenberg
As in the semester thesis half a year ago, the productive environment with my team member and the good relationship with
the supervisor and the IFS team made fun. Having a deeper
look inside the CDT team through the mailing list [?] gave an
extra motivated feeling to be part of the development cycle of the
Eclipse CDT. As we heard that our semester thesis was merged
in CDT 8 we were much more motivated to bring good results in
this thesis too.
What came out was mostly hard work. The most difficult part
was to connect the multiple parts which were already available in
Eclipse to a unit we needed. But unlike the semester thesis we
could compare to the JDT infrastructure providing code assistance
for Java and learn from their mistakes. Very funny to see the
selection algorithm behind JDT proposals and how primitive it
is working in some cases. But much more shocking if you think
about how much more functionality and logic is needed to provide
better results.
64
CHAPTER 4. CONCLUSIONS, INTERPRETATION AND
FUTURE WORK
4.2.3. Insights
For future projects, in this section we collected positive and negative experiences throughout this bachelor thesis.
Co-location was encouraged by HSR by providing fast development machines and a big desk. This was very productive and
written project management could be reduced to a minimum.
Redmine was used to track issues and as information radiator
for project supervisors. We have spent as few time as possible
with this tool, and the mechanism to refer to a certain issue
inside the commit message (e.g. refs #123) was a big help to
reach this goal of avoiding double work. What we would definitely
not do is writing documentation inside the wiki because all has
to be moved to the actual documentation at a later time. We
tried to begin the documentation earlier this time, but we had to
rewrite a lot of it, so writing the documentation in the beginning
provides no bigger benefit.
Git was used for revision control and this time, only a master and one development branch were used constantly. By using
rebase, a lot of merge commits could be avoided, resulting in a
cleaner repository history.
At last, for any project that we will work with in future, we
will follow the guideline Cheap desk, nice chair from [Bec03].
4.3. Future Work
For TDD support, a subset of possible Quick Fixes was selected
and implemented. However, a lot more Quick Fixes could be used
for daily programming and also for TDD. For interested students,
this is a field with a lot of potential and it is a thankful work
because it is really useful.
Concerning this plug-in, some work will be needed to bundle it
with CUTE and more work to integrate it into CDT. An interesting statistical work that was not done during this project due to
will be to measure the time saved by the Quick Fixes . A striking
difference in speed was noticed during writing tests, however, no
numbers have been collected to be able to promote a speed-up of
twenty percent. Even if no time is saved, the plug-in was worth
its effort because of less typing. It will be interesting to follow up
65
CHAPTER 4. CONCLUSIONS, INTERPRETATION AND
FUTURE WORK
future improvements for TDD and Quick Fixes.
66
APPENDIX A. USER MANUAL
A. User Manual
This chapter explains the installation and usage of the plug-in for
Eclipse.
A.1. Installation of the Plug-in
This plug-in was developed for Eclipse CDT v8.0.0 using Eclipse
Indigo M5. Follow these instructions to use the plug-in:
• Get a recent Eclipse Indigo (3.7) at
http://download.eclipse.org/eclipse/downloads/
• Install a recent CDT v8.0.0 build, available at
http://download.eclipse.org/tools/cdt/builds/8.0.0/
• Install the Tdd4Cdt plug-in using the update site at
http://sinv-56042.edu.hsr.ch/updatesite
• Optionally, the CUTE unit testing framework may be installed using the update site at
http://sinv-56042.edu.hsr.ch/downloads/cute/
Please notice that this is a pre-release of CUTE. The publicly available version did not yet run on Indigo at the time
of this writing.
Be aware that the update site is hosted on a virtual server that
will be shut down at the end of the semester. A copy of the
update site is available on the attached CD. Eclipse update site
supports local paths: use the path
file:/mnt/cdrom/tdd/ch.hsr.eclipse.update/target/site
67
APPENDIX A. USER MANUAL
A.2. Building The Source Code
To build the source code, Maven 3 is needed, available at
http://maven.apache.org/download.html.
Copy the git repository stored on the attached CD and enter the
directory ch.hsr.eclipse.parent. To build the project, run
1
mvn clean install
After a successful build, the plug-in may be installed using the
update site from the ch.hsr.eclipse.update/target/site directory.
Documentation is available in directory ch.hsr.eclipse.doc.
To build the documentation, run
1
mvn clean latex : latex
A.3. Using the Quick Fix
Quick Fixes are available whenever a bug (A.1a) is displayed together with a Quick Fix lamp (A.1b). If the mouse pointer is held
above the bug symbol for a moment, a hover message (A.1c)lists
all problems at the current line.
(a) Bug
(b) Bug with Quick Fix
(c) Hover message
Figure A.1.: Display of markers and Quick Fixes
Quick Fixes are context sensitive: Before calling the Quick Fix
menu, place the cursor on the element to be fixed. The cursor
automatically jumps to the nearest problem on the current line.
To open the menu with a list of Quick Fixes , press the Ctrl-1
key combination or select Quick Fix from the ”Edit” menu in
Eclipse. The Quick Fix is applied by pressing the return key or
with a double-click on the appropriate Quick Fix.
68
APPENDIX A. USER MANUAL
A.3.1. Editing Generated Code
Some Quick Fixes offer the option to choose another symbol name
or a more specific return type. For this purpose, linked mode
editing is started after the code has been generated. See figure
D.4 for an example of linked mode editing. The tabulator key is
Figure A.2.: Usage of linked mode after code generation.
used to switch to the next marked position whereas the escape key
exits linked mode at the current selection. Pressing the return key
at any time jumps back to the initial position which is indicated
by a green bar.
Whenever focus is lost or a region outside of a linked text is
edited, linked Mode exits automatically. Further Quick Fixes may
be applied without having to exit linked mode.
A.3.2. Available Quick Fixes
The following Quick Fixes are available. Each Quick Fix is shown
together with a code example.
• Create new types and namespaces
Figure A.3.: Choice between a new class and a new namespace
69
APPENDIX A. USER MANUAL
Figure A.4.: Dropdown list on a newly created type definition
• Add member functions, -variables, constructors and operators
1
2
3
4
5
6
struct square {};
void test () {
square s1 (2) , s2 (3) ;
s1 . size = s2 . size ;
ASSERT_EQUAL (16 , s1 + s2 ) ;
}
• Create Local Variable
1
2
3
void test () {
ASSERT_EQUAL (16 , square1 ) ;
}
• Add arguments to match a certain candidate function
1
2
3
4
5
double add ( int first , square second ) ;
double add ( double first , double second ) ;
void test () {
ASSERT_EQUAL (16 , add (8) ) ;
}
• Change visibility of type members
1
2
3
4
5
6
7
class square {
bool privateMember () ;
};
void test () {
square s ;
ASSERT ( s . privateMember () ) ;
}
70
APPENDIX A. USER MANUAL
A.4. Extracting Types and Functions
The Extract Type or Function feature may be executed with the
key Alt-Shift-P. To be applicable, the current selection has to
be inside a type or function definition.
1
2
3
4
5
6
7
8
struct square {
square ( double s ) : side ( s ) {}
int getPerimeter () { // Current cursor position
return side * side ;
}
private :
int side ;
};
9
10
11
12
13
void test () {
square s ;
ASSERT (8 , s . getPerimeter () ) ;
}
Listing A.1: Code before type extraction
1
# include " square . h "
2
3
4
5
6
void test () {
square s ;
ASSERT (8 , s . getPerimeter () ) ;
}
Listing A.2: Code after class square has been extracted to a file
Member functions cannot be extracted separately. Their corresponding outmost type definition is extracted instead. Namespaces are not extracted. Listings A.1 and A.2 show a before and
after example of type extraction.
A.4.1. Target-Destination Conflicts
Types and functions are always extracted to a header file with the
same name as the affected declaration. If a header file with the
same name already exists, the user is asked whether she wants
71
APPENDIX A. USER MANUAL
Figure A.5.: Overwrite dialog if destination file already exists
to overwrite the old file. Moving a declaration into a header file
with another name is not implemented.
By accepting this message, the whole header file is overwritten.
72
B. Extension Manual
This chapter describes how to write new Quick Fixes for instant
source code correction and how to test them.
B.1. Checkers
Using the normal way of Codan checkers should be fine. This
means extending the AbstractIndexAstChecker class and process the AST with a visitor. If one should react on a problem
with a unresolved name it is probably best to modify the ProblemBindingChecker and integrate them into the official CDT.
The glue between a reported problem and the appropriated
Quick Fix is the problem id. In the plugin.xml the binding of a
Quick Fix and the problem id is made.
For more information on how to configure Checkers and register
them refer to section 3.3.2 and 3.3.1.
B.2. Quick Fix
Writing a Quick Fix has been made easy by extending the RefactoringLinkedModeQuickFix and implementing the appropriate
methods. Most important is the getRefactoring() method where
the actual code refactoring is created.
Second there is the configureLinkedMode() method were groups
and positions of the linked mode can be set to edit the code later.
To add a group add the position and the length of the element.
The delivered ChangeRecorder helps to detect changed the codeparts after editing.
If a list of proposals should be presented to the programmer,
there is the addProposal() method in the passed in LinkedModeInformation.
73
APPENDIX B. EXTENSION MANUAL
B.3. Refactoring
The returned refactoring in the Quick Fix is of the type MarkerResolutionRefactoring which is a CRefactoring2 with some
small changes.
B.4. Tests
Tests are important in this area where small changes can produce
wrong code. So there are two kinds of tests.
B.4.1. Refactoring Tests
First there are Refactoring tests. Extending MarkerResolutionRefactoringTest provides an infrastructure to test code which
produces markers and applies the appropriate Quick Fix. The
code and the expected outcome is written in a .rts file.
B.4.2. Quick Fix Tests
In contrast to the refactoring tests, the Quick Fix tests do a more
detailed level of testing. Here various other properties can be
tested like the marker length or the marker text.
74
APPENDIX C. PROJECT SETUP
C. Project Setup
C.1. Configuration Management
The configuration management was exactly the same as in the
semester thesis [KS10] with minor differences between some versions of the used software.
75
APPENDIX D. TIME MANAGEMENT
D. Time Management
Focus in this chapter is the time management and presents some
charts explaining where the time was spent and by whom.
D.1. Time Budget
Compared to the previous semester thesis [KS10] one change in
the organization was done to prevent the issue thesis where time
was not logged for three or four weeks. Since the Redmine project
management tool [red10] is able to analyze a commit message,
commit messages can be linked to an issue. Putting the spent
time on a issue in that commit message together with the issue
number it was possible to massively reduce logging effort. Hence
time logging could be done while working on issues directly without the risk of postponing it indefinitely. In the last week some
time was taken on credit since this chapter has to be written and
the time chart was freezed.
Figure D.1.: Used time for the first part
Figure D.2.: Used time for second part
76
APPENDIX D. TIME MANAGEMENT
Figure D.3.: Used time by team members
D.1.1. Activities
Most time was spent on implementation and documentation. Bugfixing took some time too. Administration took quite some time
too since the process of creating an issue in the wiki, estimating
the required time to complete, chooseing a category and logging
time afterwards was quite time consuming.
Figure D.4.: Used time by categorie
After all the team thinks that time management with Redmine is quite time consuming. Planing over more than one week
and even visualize it is nearly impossible. Here a spreadsheet or
something that is faster to administrate would be a better choice
because switching between the two web pages, the visualization
and the issue manipulation page of Redmine is tedious.Many bugs
in the visualization graph prevent it from using it properly as a
77
APPENDIX D. TIME MANAGEMENT
planing and visualization tool resulting in loosing time without
any benefit.
78
APPENDIX D. TIME MANAGEMENT
D.2. Declaration of Originality
Ich erkläre hiermit,
• dass ich die vorliegende Arbeit selber und ohne fremde Hilfe
durchgeführt habe, ausser derjenigen, welche explizit in der
Aufgabenstellung erwähnt ist oder mit dem Betreuer schriftlich vereinbart wurde,
• dass ich sämtliche verwendeten Quellen erwähnt und gemäss
gängigen wissenschaftlichen Zitierregeln korrekt angegeben
habe.[HSR11]
Ort, Datum:
,
Thomas Kallenberg
Martin Schwab
79
Task Agreement for Bachelor Thesis: TDD for CDT
During the bachelor thesis in spring semester 2011, a plug-in for Eclipse CDT is developed. It has to provide several features that facilitate the process of Test-Driven
Development (TDD), especially writing CUTE tests. This is achieved by offering functions to generate appropriate code where code is missing. In the TDD life cycle of Red
– Green – Refactor – Integrate, this project is dedicated to speed up the process of
satisfying the compiler with syntactically correct code.
Current Situation
Test-Driven Development has become a very useful and important software development process in
modern software development. Fast and safe TestDriven Development needs support by the development environment. Eclipse JDT, a development environment for Java does offer efficient and intuitive
support for TDD. The horizon of this bachelor thesis is to develop TDD functionality for Eclipse CDT.
Test automation in Eclipse CDT is already possible using a separate plug-in like CUTE. This plug-in
supports generating new test code. Adapting or creating code starting from parse errors as is useful for
TDD is not yet supported. The figure on the right
visualizes which code may already be generated by
Green: Code may be generated.
CDT.
Red: No automation available.
Basic Objectives
Whenever a user writes a test for a program element that does not exist yet, a parse
error is produced and the code is underlined by a red waved line by the IDE. Starting
from this problem marker which is already part of the functionality in CDT 8, it shall
be possible to generate the missing code using a simple interface, ideally a keystroke.
The task may be split up into the following subtasks:
• Automation of (member) function creation
• Automation of class creation (in current file)
• Offer above as quick-fixes for corresponding error markers
• Provide technical documentation with a user guide
Also included in the basic objectives is support for changing the signature so the following
items should work.
1
Glossary
activator a class inside Eclipse plug-ins that controls the lifecycle
of plug-ins, see OSGi[Ecla].
autocomplete context sensitive text assistance feature to complete previously defined words.
candidate in case of an abmiguous function call, a proposal of a
function with the same name but other arguments.
commit incremental change to a project under version control.
compiler program that transforms source code into machine executable representation.
constructor as defined in the C++ specification [ISO11] 12.1 .
CUTE C++ unit test framework developed by Prof. Peter Sommerlad (supervisor of this thesis).
Eclipse CDT C/C++ Development Tooling, an Integrated Development Environment for C++.
friend a function or class that is given permission to use the
private and protected meber names from the class.[ISO11]
11.4 .
garbage collector automatic memory management, tool for freeing unused memory.
header file a file with extension .h or .hpp that contains forward
declarations that may be embedded in source files.
implicit name is used to resolve uses of implicit bindings, such
as overloaded operators.[K+ 11].
81
Glossary
initializer list a way to initalize variable in classes as defined in
the C++ specification [ISO11] 18.9 .
Integrated Development Environment smart text editor with tools
optimized for software development.
keywords reserved character sequence in programming language,
for C++ defined in [ISO11], appendix A.1.
linked mode editing a jFace feature which simplifies editing groups
of text.
macro is a preprocessor directive as defined in the C++ specification [ISO11] 16.3 .
member function a member function, as defined in the C++
specification [ISO11] 17.6.4.5 .
namespace container for declarations in C++, as defined in the
C++ specification [ISO11] 7.3 .
operator similar to functions, may be overloaded as described in
the C++ specification [ISO11] 13.5 .
overloading as defined in the C++ specification [ISO11] 13 .
parameter value that is accepted by a function declaration in
case of a call.
parse error even with correct syntax, an error can occur because
of wrong semantic.
parser tranforms source code into other representation formats.
plug-in used to extend an existing system with new functionality,
Eclipse CDT in this case.
pointer as defined in the C++ specification [ISO11] 8.3.1 .
problem marker are displayed inside an Eclipse editor, usually
with a red bullet. Markers are registered to a certain location in code and may be underlined with color, by default
yellow for warnings and red for errors.
82
Glossary
proposal menu editable dropdown list, displayed during linked
mode editing, for quick access of predefined options.
quick fix context-sensitive menu which may be called by pressing
Ctrl-1 or via Edit → Quick Fix .
refactoring functionality for modifcation of source code. In this
context, a refactoring is a subclass of CDT CRefactoring2
class but not a refactoring in the sense of [Fow99].
rewriter is used to make modify the source code using the AST.
scope as defined in the C++ specification [ISO11] 3.3 .
serialize process of expressing a data structure as a stream of
data.
shadowing occurs when a variable declared within a certain scope
(decision block, method, or inner class) has the same name
as a variable declared in an outer scope. This outer variable
is said to be shadowed [Wik10b].
static code checker tool that scans an Abstract Syntax Tree for
potential problems. It is called static because it does not
rely on compiled source.
symbol an abstraction, tokens of which may be marks or a configuration of marks which form a particular pattern [Wik10a].
template arguments as defined in the C++ specification [ISO11]
14.3 .
template parameter as defined in the C++ specification [ISO11]
14.1 .
templated class a family of classes or functions or an alias for a
family of types. [ISO11] 14 .
Test-Driven Development a development guideline which promotes designing software test-first.
83
Glossary
unit testing defines fine granular constraints for source code to
ensure expected behavior.
wizard modal dialog collecting user decisions for complex tasks.
84
Bibliography
[Bau10]
Sebastian Bauer. Eclipse für C/C++ Programmierer.
dpunkt, 11 2010.
[Bec03]
Beck, Kent. Test-Driven Development by Example.
Pearson Education, Inc., 2003.
[Ecla]
Eclipse.
Equinox
http://www.eclipse.org/equinox/.
[Eclb]
Eclipse SDK Documentation.
Platform Plug-in
Developer Guide – Contributing Marker Resolution.
http://help.eclipse.org/helios/
topic/org.eclipse.platform.doc.isv/guide/
[Online; accessed
wrkAdv markerresolution.htm.
01-March-2011].
[Ecl10a]
Eclipse Bugs.
Bug 309760: Provide checker for
name resolution problems found by indexer. https:
//bugs.eclipse.org/bugs/show bug.cgi?id=309760,
2010. [Online; accessed 08-March-2011].
–
OSGi.
[Ecl10b] Eclipse Bugs. Bug 319196: Quick fixes for creating a
variable. https://bugs.eclipse.org/bugs/show bug.
cgi?id=319196, 2010. [Online; accessed 08-March2011].
[Ecl11a]
Eclipse. Eclipse. http://www.eclipse.org/, 2011.
[Online; accessed 15-June-2011].
[Ecl11b] Eclipse. Eclipse Bugzilla. http://bugs.eclipse.org,
2011. [Online; accessed 15-June-2011].
[Ecl11c]
Eclipse.
Eclipse CDT Indexer.
http:
//help.eclipse.org/indigo/topic/org.eclipse.
cdt.doc.isv/reference/extension-points/
85
Bibliography
org eclipse cdt core CIndexer.html, 2011. [Online;
accessed 15-June-2011].
[Ecl11d] Eclipse. Eclipse JFace. http://wiki.eclipse.org/
index.php/JFace, 2011. [Online; accessed 15-June2011].
[Ecl11e]
Eclipse. Eclipse Test and Performance Tools Platform
(TPTP). www.eclipse.org/tptp, 2011. [Online; accessed 14-June-2011].
[Ecl11f]
Eclipse. Extension Point. http://www.eclipse.org/
resources/?category=Extension%20points,
2011.
[Online; accessed 15-June-2011].
[Ecl11g]
Eclipse CDT Wiki. CDT/designs/StaticAnalysis. http:
//wiki.eclipse.org/CDT/designs/StaticAnalysis,
2011. [Online; accessed 08-March-2011].
[Esk11]
Jesper Eskilson.
[cdt-dev] Codan performance.
http://dev.eclipse.org/mhonarc/lists/cdt-dev/
msg21516.html, 2011. [Online; accessed 15-June-2011].
[Fow99]
Martin Fowler. Refactoring, Improving the Design of
Existing Code. Addison-Wesley, 1999.
[Gam95] Gamma, Erich; Helm, Richard; Johnson, Ralph E. Design Patterns – Elements of Reusable Object-Oriented
Software. Addison-Wesley Professional Computing Series. Prentice Hall, 12 1995.
[Gra06]
Leo Graf, Emnuel; Buettiker. C++ Refactoring Support
for Eclipse CDT. http://leo.freeflux.net/blog/
archiv/c-refactoring-support-fuer-eclipse-cdt.
html, 2006. [Online; accessed 15-June-2011].
[HSR11] Hochschule für Technik Rapperswil HSR.
Muster
einer Erklaerung ist integrierender Bestandteil der
SA BA DA. HSR internal document: Muster einer
Erklaerung ist integrierender Bestandteil der SA BA
DA.doc, 2011. [Online; accessed 15-June-2011].
86
Bibliography
[ISO11]
WG21 ISO/IEC. IS 14882: Programming Languages –
C++. International Organization for Standardization,
Geneva, Switzerland, March 2011.
[JUN11] JUnit.org – Resources for Test Driven Development.
http://www.junit.org/, 2011. [Online; accessed 08March-2011].
[K+ 11]
Mike Kucera et al.
CDT source code, class
org.eclipse.cdt.core.dom.ast.IASTImplicitName
in project org.eclipse.cdt.core, 2011. SVN version
v201105161139.
[Kal11]
Thomas Kallenberg.
Bug 341089 – Codan does
not handle missing functions in macros correctly.
https://bugs.eclipse.org/bugs/show bug.
cgi?id=341089, 2011. [Online; accessed 15-June-2011].
[KS10]
Thomas Kallenberg and Martin Schwab.
One
touch
C++
code
automation
for
Eclipse
CDT.
http://eprints3.hsr.ch/158/1/
OneTouchToggleRefactoring.pdf, December 2010.
[Online; accessed 06-June-2011].
[Mic05]
Sun Microsystems.
Top Level Type Declarations. http://java.sun.com/docs/books/jls/third
edition/html/packages.html#26783, 2005. [Online;
accessed 15-June-2011].
[Mic11]
Microsoft Developer Network.
MSDN Library - C++ Language Reference - Unions.
http://msdn.microsoft.com/en-us/library/
5dxy4b7b%28v=VS.100%29.aspx, 2011.
[Online;
accessed 21-Mai-2011].
[red10]
Redmine – Project Management Web Application (Version 0.9.3-1). http://www.redmine.org/, 2010. [Online; accessed 22-December-2010].
[Som08]
Prof. Peter Sommerlad.
C++ Refactoring and
TDD with Eclipse CDT.
http://accu.org/
index.php/conferences/accu conference 2008/
87
Bibliography
accu2008 sessions#C++%20Refactoring%20and%
20TDD%20with%20Eclipse%20CDT, 2008.
[Online;
accessed 15-June-2011].
[Som11a] Prof. Peter Sommerlad.
Agile C++ Through
Advanced IDE Features.
http://accu.org/
index.php/conferences/accu conference 2011/
accu2011 sessions#Agile%20C++%20through%
20Advanced%20IDE%20Features, April 2011.
[Online; accessed 15-June-2011].
[Som11b] Sommerlad, Prof. Peter. CUTE – C++ Unit Testing
Easier. http://www.cute-test.com/, 2011. [Online;
accessed 01-March-2011].
[Sut09]
Herb Sutter. Const-Correctness. http://gotw.ca/
gotw/006.htm, 2009. [Online; accessed 15-June-2011].
[Tho05]
Andrew Thomas, David; Hunt. The Pragmatic Programmer – From Journeyman to Master. AddisonWesley Longman, Amsterdam, 01 2005.
[web10]
The WebKit Open Source Project (Version SVN head:
r71799). http://webkit.org/, 2010. [Online; accessed
22-December-2010].
[Wes11]
Tomasz Wesolowski. Bug 319196 - Quick fixes for creating a variable. https://bugs.eclipse.org/bugs/
show bug.cgi?id=319196, 2011. [Online; accessed 10June-2011].
[Wik10a] Wikipedia. Symbol (formal). http://en.wikipedia.
org/wiki/Symbol %28formal%29, 2010. [Online; accessed 15-June-2011].
[Wik10b] Wikipedia.
Variable Shadowing.
http://en.
wikipedia.org/wiki/Variable shadowing, 2010. [Online; accessed 15-June-2011].
88
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement