View Source

This page documents the standard adopted for Java code in the Cask project(s).

All committers are expected to follow these standards; Checkstyle or similar (Lynt) is used to check compliance.

CDAP 6.9+

CDAP starting from version 6.8 uses slightly relaxed Google Coding Standards.

CDAP 6.8 and before

Summary

The main things for layout purposes in the standard are:

Indent using two spaces. No tabs.

Braces always go on same line, e.g.

if (x == 5) {
  System.out.println("Hello");
}

rather than

// WRONG
if (x == 5}
{
  System.out.println("Hello");
}

Always add braces, e.g.

if (x == 5) {
  System.out.println("Hello");
}

rather than

// WRONG
if (x == 5)
  System.out.println("Hello");

Introduction

This document describes two types of coding standard:

1. Mandatory standards must be followed at all times.
2. Recommended standards should in general be followed but in particular cases may be omitted where the programmer feels that there is a good reason to do so.

Code that does not adhere to mandatory standards will not pass the automated checks (or a code review if the guideline is not stylistic).

Source files

This section defines the general rules associated with the contents of a Java source file and the order in which the each part should be presented. No rules on programming style, naming conventions or indentation are given here.

Java source files must have a ".java" suffix (this will be enforced by the compiler) [mandatory].
The basename of a Java source file must be the same as the public class defined therein (this will be enforced by the compiler) [mandatory].
Only one class should be defined per source file (except for inner classes and one-shot uses where the non-public class cannot conceivably be used outside of its context) [mandatory].
Source files should not exceed 1500 lines [recommended].
No line in a source file should exceed 120 characters [mandatory].
The sections of a source file should be presented in the following order [mandatory]:
1. Copyright Information
2. File information comment (see rule 7 below).
3. Package name (see rules 1 to 3 in the section 2.1 above and rule 8 below).
4. Imports (see rules 9 to 10 below).
5. Other class definitions.
6. Public class definition.
Every class that is to be released must be a member of a package [mandatory].
1. Rationale: classes that are not explicitly put in a package are placed in the unnamed package by the compiler. Therefore as the classes from many developers will be being placed in the same package the likelihood of a name clash is greatly increased.
All class imports from the same package should be grouped together. A single blank line should separate imports from different packages [recommended]. The actual order of imports must be as follows, with each group ordered lexically:
1. all co.cask
2. all other but non-java
3. all java[x]
No import package.* allowed. Classes must be imported one by one [mandatory].
Use Javadoc tags and use HTML mark-up to enhance the readability of the output files [mandatory].
Every package should provide package-info.html [recommended].

Compilation

This section defines the general rules associated with results of compiling a Java source file.

Compiling a Java source should not result in undefined documentation compilation errors [mandatory].
Compiling a Java source should result in no warnings [mandatory].

Java Elements

This section gives advice on coding the various elements of the Java programming language.

Class definitions

This section gives guidelines for class and interface definitions in Java. The term class in this section is used more broadly to mean class and interface:

Class names should start with a capital letter with every subsequent word capitalised, for example: DataProcessor [mandatory].
The name of exception classes should end in the word exception, for example: UnknownMungeException [mandatory].
Class names should in general not be overloaded. For example, defining a class "com.foo.bar.String" should be avoided as there is already a class "java.lang.String" [recommended].
Rationale: adhering to this rule reduces the likelihood of confusion and means that the use of fully qualified class names should not be required.
Class name should not be plural. For example, `ProgramManager` not `ProgramsManager` [recommended].
The definition of the primary class (i.e. the class with the same name as the java file) should start in column 0 of the source file. Inner class definitions should be indented 2 spaces more than their enclosing class [mandatory].
Declare a class as final only if specialization will never be required and improved performance is essential. With modern JVMs there in fact may be no performance advantage. Warning: use of final limits code reuse [mandatory].
For all but simplest classes the following methods should have useful definitions [recommended]:
```
public boolean equals(Object obj)
public int hashCode()
public String toString()
```
The order of presentation of the sections in a class should be [mandatory]:
Variables
Methods Variables
This section gives guidelines for class and instance variable definitions in Java. In this section if a rule uses the term variable rather than instance variable or class variable, then the rule applies to both types of variable.
The order of presentation of variables in a class definition should be [recommended]:
1. private, protected, public: static final variables (aka constant class variables).
2. private, protected, public: static variables (aka class variables).
3. private, protected, public: final variables (aka constant instance variables).
4. private, protected, public: variables (aka instance variables).
  It should be noted that as Javadoc will automatically order variables in a consistent manner, rigid adherence to this rule is not necessary.
Variable modifiers must be presented in the following order: static, final, transient, volatile [mandatory].
The names of static final variables should be upper case with subsequent words prefixed with an underscore [mandatory]. For example:
```
public static final int NOT_FOUND = -1;
```
When a subclass refers to a static final variable defined in a parent class, access should be qualified by specifying the defining class name [mandatory]. For example: use ParentClass.MAX rather than MAX.
The names of variables (other that static final) should start with a lower case letter. Any words that are contained in the rest of the variable name should be capitalised [mandatory]. For example:
```
String name;
String[] childrensNames;
```
Variables must not be named using the so-called Hungarian notation [mandatory]. For example:
```
int nCount = 4; // not allowed
```
Only one variable may be defined per line [mandatory].
Variable declarations should be indented 2 spaces more than their enclosing class [mandatory].
All variables should be preceded by a comment that explains what the variable is for, where it is used and so forth, unless it is obvious. The comment should not be a Javadoc, unless it is a public or protected variable.
All public constants must be preceded by a Javadoc that explains the meaning of the constant [mandatory].
Never declare instance variables as public unless the class is effectively a "struct" [mandatory].
Never give a variable the same name as a variable in a superclass [mandatory].

Ensure that all non-private class variables have sensible values even if no instances have been created (use static initialisers if necessary, i.e. "static { ... }") [mandatory].
Rationale: prevents other objects accessing fields with undefined/unexpected values.

Methods

This section gives guidelines for class and instance method definitions in Java. In this section if a rule uses the term method rather than instance method or class method, then the rule applies to both types of method.

Constructors and finalize methods should follow immediately after the variable declarations [mandatory].
Do not call non-final methods from constructors. This can lead to unexpected results when the class is subclassed. If you must call non-final methods from constructors, document this in the constructor's Javadoc [mandatory]. Note that private implies final.
Methods that are associated with the same area of functionality should be physically close to one another [recommended].
After grouping by functionality, methods should be presented in the following order [recommended]:
private, protected, public: static methods.
private, protected, public: instance methods.
It should be noted that as Javadoc will automatically order methods in a consistent manner, rigid adherence to this rule is not necessary.
Method modifiers should be presented in the following order: abstract, static, final., synchronized [mandatory]
When a synchronized method is overloaded, it should be explicitly synchronized in the subclass [recommended].
Method names should start with a lower case letter with all subsequent words being capitalised [mandatory]. For example:
```
protected int resize(int newSize)
protected void addContentsTo(Container destinationContainer)
```
Methods which get and set values should be named as follows [mandatory]:
```
Type getVariableName()
void setVariableName(Type newValue)
```
Exceptions should be used to report any failure to get or set a value. The "@param" description should detail any assumptions made by the implementation, for example: "Specifying a null value will cause an error to be reported".
Method names should not have redundancy. For example, method name `getSecret` in `SecretManager` class should be avoided. Instead, the method name should be simply `get`[recommended].
Method definitions should be indented 2 spaces more than their enclosing class [mandatory].
All methods should be preceded by a Javadoc comment specifying what the method is for, detailing all arguments, returns and possible exceptions. This comment should be of the following form and be indented to the same level as the method it refers to [mandatory]:
The braces associated with a method should be on a line on their own and be indented to the same level as the method [mandatory]. For example:
```
public void munge() {
    int i;
    // method definition omitted...
}
```
The body of a method should be indented 2 columns further that the opening and closing braces associated with it [mandatory]. See the above rule for an example.
When declaring and calling methods there should be no whitespace before or after the parenthesis [mandatory].
In argument lists there should be no white space before a comma, and only a single space (or newline) after it [mandatory]. For example:
```
public void munge(int depth, String name) {
    if (depth > 0) {
        munge(depth - 1, name);
    }
    // do something
}
```
Wherever reasonable define a default constructor (i.e. one that takes no arguments) so that Class.newInstance() may be used [recommended]. If an instance which was created by default construction could be used until further initialisation has been performed, then all unserviceable requests should cause a runtime exception to be thrown.
The method public static void main() should not be used for test purposes. Instead a test/demo program should be supplied separately. [mandatory].
Public access methods (i.e. methods that get and set attributes) should only be supplied when required [mandatory].
If an instance method has no natural return value, declare it as void rather than using the "return this;" convention [mandatory].

Ensure that non-private static methods behave sensibly if no instances of the defining class have been created [mandatory].

Expressions

This section defines the rules to be used for Java expressions:

Unary operators should not be separated from their operand by white space [mandatory].
Embedded ++ or – operators should only be used when it improves code clarity [recommended]. This is rare.
Extra parenthesis should be used in expressions to improve their clarity [recommended].
The logical expression operand of the "?:" (ternary) operator must be enclosed in parenthesis. If the other operands are also expressions then they should also be enclosed in parenthesis [mandatory]. For example:
```
biggest = (a > b) ? a : b;
complex = (a + b > 100) ? (100 * c) : (10 * d);
```
Nested "?:" (ternary) operators can be confusing and should be avoided [mandatory].
Use of the binary "," operator (the comma operator) should be avoided [mandatory]. Putting all the work of a for loop on a single line is not a sign of great wisdom and talent.
If an expression is too long for a line (i.e. extends beyond column 120) then it should be split after the lowest precedence operator near the break [mandatory]. For example:
```
if ((state == NEED_TO_REPLY) ||
    (state == REPLY_ACK_TIMEOUT)) {
    // (re)send the reply and enter state WAITING_FOR_REPLY_ACK
}
```
Furthermore if an expression requires to be split more than once, then the split should occur at the same logical level, if possible.

All binary and ternary operators,with the exception of ".", should be separated from their operands by a space [mandatory].

Statements

Simple Statements

This section defines the general rules for simple Java statements:

There must only be one statement per line [mandatory].
In general local variables should be named in a similar manner to instance variables [recommended].
More than one temporary variable may be declared on a single line provided no initialisers are used [mandatory]. For example:
```
int j, k = 10, l;  // Incorrect!
int j, l;          // Correct
int k = 10;
```
A null body for a while, for, if, etc. should be documented so that it is clearly intentional [mandatory].
Keywords that are followed by a parenthesised expression (such as while, if, etc) should be separated from the open bracket by a single space [mandatory]. For example:
```
if (a > b) {
    munge();
}
```

In method calls, there should be no spaces before or after the parentheses [mandatory]. For example:

munge (a, 10);    // Incorrect!
munge(a, 10);     // Correct.

Compound Statements

This section defines the general rules associated with compound statements in Java:

The body of a compound statement should be indented by 2 spaces more than the enclosing braces [mandatory]. See the following rule for an example.
The braces associated with a compound statement should be on their own line and be indented to the same level as the surrounding code [mandatory]. For example:
```
if ((length >= LEN_BOX) && (width >= WID_BOX)) {
    int i;
    // Statements omitted...
}
```
If the opening and closing braces associated with a compound statement are further than 20 lines apart then the closing brace should annotated as follows [mandatory]:
```
for (int j = 0; j < SIZE; j++) {
    lotsOfCode();
} // end for
```
All statements associated with an if or if-else statement should be made compound by the use of braces [mandatory]. For example:
```
if (a > b){
    statement();
} else {
    statement1();
    statement2();
}
```
The case labels in a switch statement should be on their own line and indented by a further 2 spaces. The statements associated with the label should be indented by 2 columns more than the label and not be enclosed in a compound statement. [mandatory]. For example:
```
switch (tState) {
    case NOT_RUNNING:
        start();
        break;

    case RUNNING: // FALLTHROUGH
    default:
        monitor();
        break;
}
```
In switch statements - the statements associated with all cases should terminate with a statement which explicitly determines the flow of control, for example break [recommended].
In switch statements - fall through should be avoided wherever possible, however if it is unavoidable it must be commented with "// FALLTHROUGH" [mandatory].

In switch statements - a default case must be present and should always be the last case [mandatory].

General

This section gives general rules to be followed when programming in Java:

When comparing objects for equivalence use the method equals() and not the == operator. The only exceptions to this are static final objects that are being used as constants and interned Strings [mandatory].
In general labelled break and continue statements should be avoided [recommended]. This is due to the complex flow of control, especially when used with try/finally blocks.
Unless some aspect of an algorithm relies on it, then loops count forward [mandatory]. For example:
```
for (int j = 0; j < size; j++) {
    // Do something interesting
}
```

Use local variables in loops [recommended]. For example:

ArrayList clone = (ArrayList) listeners.clone();
final int size = clone.size();
for (int j = 0; j < size; j++) {
    System.out.println(clone.elementAt(j));
}

Anonymous inner classes should define no instance variables and be limited to three single line methods. Inner classes that declare instance variables or have more complex methods should be named [mandatory].

Use final local variables where possible to help avoid errors in code [recommended]. For example:

public void foo() {
    final int x = dataSource.getCount();
    // do things with x
    // ...
}

ThreadLocal

ThreadLocal is a nice Java construct that allows you to isolate state across different threads (https://docs.oracle.com/javase/7/docs/api/java/lang/ThreadLocal.html). However, uncareful usage of ThreadLocal (or InheritableThreadLocal) can easily lead to memory leaks. As a rule of thumb, they are safe to use if the thread executing the code is short lived. Otherwise, care must be made to remove the ThreadLocal once it is no longer needed.

The longer explanation is that each thread contains a Map of all the ThreadLocal objects created by any code executed by that Thread. For example:

CDAP > Java Coding Standards > threadlocal.png

These objects are garbage collected if the thread goes away. However, if thread is long running (for example, it is used in a thread pool), these objects will never get garbage collected. In the example above, every time an instance of ExampleClass is created, a new entry will be added to the ThreadLocalMap inside the current Thread. If the thread never exits, memory will leak until the process dies. See for an example of this in the past.

If you are not absolutely certain that the Thread using the ThreadLocal is short-lived, you must be sure to remove the ThreadLocal once it is no longer needed. Much like closing an InputStream in a finally block, classes that use a ThreadLocal must have a way to release all their resources. In the example below, a close() method is added to the ExampleClass that removes the ThreadLocal. After this, we can see that the Map in the ThreadLocal no longer has the objects.

CDAP > Java Coding Standards > threadlocal_clean.png

Another option if possible is to make the ThreadLocal static so that there is only instance per thread.

Exceptions

This section gives general guidance on the use of exceptions when programming in Java.

try/catch blocks should be laid out like any other compound statement [mandatory]. For example:

try {
    String str = someStrings[specifiedIndex];
} catch (IndexOutOfBoundsException ex) {
    // The user specified an incorrect index, better take
    // some remedial action.
}

When an exception is caught but ignored then a comment should be supplied explaining the rationale [mandatory]. For example:

try {
    propertySet.setProperty("thingy", new Integer(10));
} catch (UnknownPropertyException ignore) {
    // This exception will never occur as "thingy" definitely exists
}

All exceptions that are likely to be thrown by a method should be documented, except if they are runtime exceptions (note: the compiler will not enforce catch blocks for runtimes even if they are mentioned in the throws clause) [mandatory]. For example:
```
/* Comment snippet:
 * @exception IllegalValueException Thrown if values is null or 
 *     any of the integers it contains is null.
 */
private Integer sum(Integer[] values) throws IllegalValueException
```
InterruptedException should be caught and should not be ignored. It should set the correct state of the thread or status is checked before rethrowing.