BestPractices/Coding Guidelines

2019独角兽企业重金招聘Python工程师标准>>>

BestPractices/Coding Guidelines

Note: This document is a work in progress.

Contents

• 1 Git commit message style
• 2 General Code headers
  • 2.1 Bash/Python
  • 2.2 XML
• 3 General Code Style
• 4 Java
  • 4.1 Checkstyle
    • 4.1.1 Enforcement
    • 4.1.2 @SuppressWarnings
    • 4.1.3 Notes for particular Checks
      • 4.1.3.1 {@inheritDoc} JavaDoc
      • 4.1.3.2 IllegalThrows
      • 4.1.3.3 IllegalCatch
      • 4.1.3.4 System.out
      • 4.1.3.5 Javadoc Paragraph: < p > tag should be preceded with an empty line
    • 4.1.4 Additional Resources
  • 4.2 Class's methods / fields ordering
  • 4.3 FindBugs
• 5 Python
• 6 XML
• 7 YANG

Git commit message style

For Git commit messages we recommend following the OpenStack commit message recommendations.

General Code headers

License and Copyright headers need to exist at the top of all code files. Examples of copyright headers for each language can be seen below.

Note: In case you need multiple Copyright headers simply duplicate the Copyright line for additional copyrights

-== C/C++/Java ===

/*
 * Copyright (c) 2016 <Company or Individual>.  All rights reserved.
 *
 * This program and the accompanying materials are made available under the
 * terms of the Eclipse Public License v1.0 which accompanies this distribution,
 * and is available at http://www.eclipse.org/legal/epl-v10.html
 */

Bash/Python

##############################################################################
# Copyright (c) 2016 <Company or Individual>.  All rights reserved.
#
# This program and the accompanying materials are made available under the
# terms of the Eclipse Public License v1.0 which accompanies this distribution,
# and is available at http://www.eclipse.org/legal/epl-v10.html
##############################################################################

XML

<!--
  Copyright (c) 2016 <Company or Individual>.  All rights reserved.

  This program and the accompanying materials are made available under the
  terms of the Eclipse Public License v1.0 which accompanies this distribution,
  and is available at http://www.eclipse.org/legal/epl-v10.html
-->

General Code Style

• 120 character line length

Java

In General we follow the Google Java Code Style Guide with a few exceptions. See: https://google.github.io/styleguide/javaguide.html

• 4 space indentation
• 120 character line length
• 72 or 80 chars for javadoc
• import ordering (https://wiki.opendaylight.org/view/GettingStarted:_Eclipse_Setup#Import_ordering)
• YangTools Design Guidelines (https://wiki.opendaylight.org/view/YANG_Tools:Design_and_Coding_Guidelines)
• follow JLS modifier ordering (http://checkstyle.sourceforge.net/config_modifier.html)
• do not use underscores (_) in identifiers, as they trigger warnings with JDK8

Checkstyle

All OpenDaylight projects automatically run Checkstyle, as the maven-checkstyle-plugin is declared in the odlparent.

This will only shows warnings in the build log and will not prevent your project from building if code violations are found.

Enforcement

A project can "opt in" to fail for violations of the Checkstyle rules configured in odlparent by adding the following to their pom.xml:

<build>
  <plugins>
    <plugin>
      <groupId>org.apache.maven.plugins</groupId>
      <artifactId>maven-checkstyle-plugin</artifactId>
      <configuration>
        <propertyExpansion>checkstyle.violationSeverity=error</propertyExpansion>
      </configuration>
    </plugin>
  </plugins>
</build>

@SuppressWarnings

If really needed, projects can to override individual Checkstyle rules on a case-by-case basis by using a @SuppressWarnings annotation:

@SuppressWarnings("checkstyle:methodparampad")
public AbstractDataTreeListener (INetvirtSfcOF13Provider provider, Class<T> clazz) {
}

The rule ID (e.g. 'checkstyle:methodparampad' above) is the name of the respective Checkstyle module; these IDs can be found e.g. in the git/odlparent/checkstyle/src/main/resources/odl_checks.xml configuration, or directly on the Checkstyle website from the http://checkstyle.sourceforge.net/checks.html list. For example, for the http://checkstyle.sourceforge.net/config_coding.html#EqualsHashCode rule you would put @SuppressWarnings("checkstyle:EqualsHashCode").

This @SuppressWarnings("checkstyle:...") should in practice be very very rarely needed. Please put a comment explaining why you need to suppress a Checkstyle warning into the code for other to understand, for example:

@Override
@SuppressWarnings("checkstyle:EqualsHashCode") // In this particular case an equals without hashCode is OK because
public boolean equals(Object obj) {            // [explain!] (I'm a certified grown up and know what I'm doing.)

Please contact odlparent-dev@lists.opendaylight.org if you feel a Checkstyle rule is too strict in general and should be reviewed.

The Evolving Checkstyle page documents how to test changes to Checkstyle rules.

Notes for particular Checks

{@inheritDoc} JavaDoc

This JavaDoc is useless visual noise that hinders code readability, it is not required to put this, and has no impact, as JavaDoc does this by default:

/**
    * {@inheritDoc}
    */
  @Override // (or on a constructor)

The only case where {@inheritDoc} is useful is when you actually have additional Java documentation. Default JavaDoc interprets as replace the parent's doc. If you truly want the full text of the parent to be copy/pasted by JavaDoc in addition to your additional one, then use:

/**
    * {@inheritDoc}
    * For this specific implementation ... 
    */
  @Override // (or on a constructor)

See also https://github.com/sevntu-checkstyle/sevntu.checkstyle/issues/467 & http://tornorbye.blogspot.ch/2005/02/inheriting-javadoc-comments.html

IllegalThrows

Instead of declaring "throws Exception", in almost all cases you should instead throw a custom existing or new ODL Exception. Instead of an unchecked exception (unchecked = extends RuntimeException; if you must for some technical reason, but should be rare, and avoided), it's recommended to use a custom module specific checked exception (checked = extends Exception); which can wrap a caught RuntimeException, if needed.

In order to avoid proliferation of many kinds of checked Exception subtypes for various particular nitty gritty things which could possibly go wrong, note that it in ODL is perfectly OK & recommended to have just ONE checked exception for a small given functional ODL module (subsystem), and throw that from all of it's API methods. This makes sense because a typical caller wouldn't want do anything different - what you are "bubbling up" is just that one of the operations which one module asked another ODL module to do couldn't be performed. This avoids having multiple throws for each exception in API methods, and having problems with extendibility due to having to add more exceptions to the "throws" declaration of API methods.

The exception for "throws Exception" may be a main() method where it's customary to let anything propagate to the CLI, or @Test testSomething() throws Exception where it's acceptable (Checkstyle does NOT flag this particular use of "throws Exception" in @Test methods).

IllegalCatch

The IllegalCatch violation should almost never be suppressed in regular "functional" code - normal code should only catch specific sub classes of the checked Exception, and never any generic and/or unchecked exceptions.

In the old pre-Java 7 days, some people used "catch (Exception e)" to "save typing" as a shorthand for having to catch a number of possibly thrown types of checked exceptions declared with "throws" by a method within the try block. Nowadays, since Java 7, using a multi-catch block is the right way to do this. In addition to being "nicer" to read because it's clearer, much more importantly than, a multi-catch does not also accidentally catch RuntimeException, as catch (Exception e) would. Catching RuntimeException such as NullPointerException & Co. is typically not what the developer who used "catch (Exception e)" as shorthand intended.

If a catch (Exception e) is used after a try { } block which does not call any methods declaring that they may throw checked exceptions with their throws clause (perhaps not anymore, after code was changed), then that catch may really have been intended to catch any possible RuntimeException instead? In that case, if there exceptionally really is a particular reason to want to "do something and recover from anything that could possibly go wrong, incl. NullPointerException, IndexOutOfBoundsException, IllegalArgumentException, IllegalStateException & Co.", then it is clearer to just catch (RuntimeException e) instead of catch (Exception e). Before doing this, double check that this truly is the intention of that code, by having a closer look at code called within the try, and see if that called code couldn't simply be made more robust.

Proliferation of catch (Exception or RuntimeException e) { LOG.error("It failed", e); } in regular "functional" code is a symptom of a missing abstraction in framework code; e.g. an Abstract interface implementation helper with correct default error handling, so that functional code does not have to repeat this over and over again. For example:

1. Instead of catch(Exception) after a try { close(anAutoCloseable) } just use AutoCloseables.closeOrWarn(anAutoCloseable) introduced in https://git.opendaylight.org/gerrit/#/c/44145/

Sometimes developers also simply don't see that an existing framework API intends implementations to simply propagate their errors up to them. For example, for Exception handling in:

1. OsgiCommandSupport's doExecute(), the right thing to do is... nothing. The parent doExecute() method declaration throws Exception, and that is intentional by the Good People of Karaf. Thefore, catch(Exception) in a OsgiCommandSupport's doExecute is not required - in this case it's perfectly OK to just let any error "propagate" upwards automatically. If doExecute() calls other private methods of an OsgiCommandSupport implementation, then it is perfectly OK to make those methods declare "throws SomeException" too, and not "handle" them yourself.

1. Callable's call() passed to a DataStoreJobCoordinator enqueueJob(), the right thing to do is... nothing, do not catch (Exception) but let it propagate. If it's useful to "augment" the exception message with more custom details which are available inside Callable's call(), then the right thing to do is to catch (Exception e) { throw new YourProjectApiException("Failed to ... for {}", aDetail, e); } and, exceptionally, use @SuppressWarnings("checkstyle:IllegalCatch").

1. org.opendaylight.infrautils.inject.AbstractLifecycle's start() and stop() methods, again the right thing to do is... nothing, do not catch any Exception but let it propagate.

Here are some cases where catch(Excepion) is almost always wrong, and a respective @SuppressWarnings almost never acceptable:

1. In Tests code you typically just "@Test testSomething() throws (Some)Exception" instead of catch, or uses @Test(expected = ReadFailedException.class). One rare case we have seen where it's justified is a @Test(expected = ReadFailedException.class) with catch (Exception e) throw e.getCause().

1. In one time "setup" (initialization) kind of code. For example, catch for a DataBroker registerDataChangeListener makes little sense - it's typically much better to let a failure to register a data change listener "bubble up" then continue, even if logged, and have users wonder why the listener isn't working much later.

Only in lower-level "Framework" kind of code, catch (Exception e) is sometimes justified / required, and thus @SuppressWarnings("checkstyle:IllegalCatch") acceptable.

System.out

The RegexpSingleLineJava "Line contains console output" and "Line contains printStackTrace" should NEVER be suppressed.

In src/main code, System.out.println has no place, ever (it should probably be a LOG.info; and System.err probably a LOG.error).

In src/main code related to OSGi/Karaf/Felix console output, use the respective correct API to obtain the console to write to; e.g. CommandSession getConsole() which returns the PrintStream to use instead of System.out. This isn't just "cleaner", but System.out is a bug - because console output from commands using System.out would not work e.g. over an SSH session into Karaf.

In src/test code, there should be no need to write things out - the point of a test is to assert something. During development of a test it's sometimes handy to print things to the console to see what's going on instead of using the debugger, but this should be removed in final code, for clarity, and non-verbose test execution. If you must, do you a Logger even in a test - just like in src/main code. This also makes it easier to later move code such as helper methods from test to production code.

 

Javadoc Paragraph: < p > tag should be preceded with an empty line

Checkstyle (rightfully) flags this kind of JavaDoc up as not ideal for readability:

/** 
   * Utilities for...
   * <p>This...

and you can address this either like this:

/** 
   * Utilities for...
   *
   * <p>This...

or like this:

/** 
   * Utilities for...
   * <p/>
   * This...

Different ODL developers agree to disagree on which of the above is more readable.

Additional Resources

• Checkstyle http://checkstyle.sourceforge.net/
• Maven: https://github.com/checkstyle/checkstyle/blob/master/src/main/resources/google_checks.xml
• Eclipse: https://code.google.com/p/google-styleguide/source/browse/trunk/eclipse-java-google-style.xml
• IntelliJ: https://code.google.com/p/google-styleguide/source/browse/trunk/intellij-java-google-style.xml

Class's methods / fields ordering

Ordering based on modifiers. This is based on visibility and mutability:

public static final fields 
static final fields 
private static final fields 

private final fields 
private non-final fields 
private volatile fields 

private constructors 
public constructors 
static factory methods 

public methods 
static methods 
private methods 

The first group should be very strict, with the exception of FieldUpdaters, which should be private static final, but defined just above the volatile field they access. The reason for that is they are tied via a string literal name.

The second group is less clear-cut and depends on how instances are created -- there are times when juggling the order makes it easier to understand what is going on (e.g. co-locating a private static method with static factory method which uses it).

The third group is pretty much free-for-all. The goal is to group things so that people do not have scroll around to understand the code flow. Public methods are obviously entry-points, hence are mostly glanced over by users.

Overall this has worked really well so far because;

• the first group gives a 10000-foot overview of what is going on in the class, its footprint and references to other classes,
• the second group gives instantiation entry-points, useful for examining who creates the objects and how
• the third group is implementation details -- for when you really need to dive into the details.

Note this list does not include non-private fields. The reason is that public fields should be forbidden, as should be any mutable non-private fields. The reason for that is they are a nightmare to navigate when attempting to reason about object lifecycle.

Same reasoning applies to inner class, keep them close to the methods which use them so that the class is easy to read. If the inner class needs to be understood before the methods that operate on it, place it before them, otherwise (especially if it's an implementation detail) after them. That's when an inner class is appropriate of course.

FindBugs

OpenDaylight project builds can automatically run FindBugs, as the findbugs-maven-plugin is declared in the odlparent. Contrary to Checkstyle, this is not enable by default. A project can "opt in" to fail for violations of the FindBugs rules configured in odlparent by adding the following to their pom.xml:

<build>
   <plugins>
     <plugin>
       <groupId>org.codehaus.mojo</groupId>
       <artifactId>findbugs-maven-plugin</artifactId>
       <configuration>
         <failOnError>true</failOnError>
       </configuration>
     </plugin>
   </plugins>
 </build>

TODO Document correct <dependency> to use for annotations (com.google.code.findbugs:jsr305 VS com.google.code.findbugs:annotations), and see if <scope>provided needs to be repeated in projects using it or if having it on odlparent, where it's not yet, will be sufficient)

Python

PEP8 is the Python standard that should be followed when coding any Python code with the following exceptions.

• 120 character line length

To automate pep8 scanning we recommend using a tox configuration as follows:

tox.ini

[tox]
envlist = pep8
#skipsdist = true  # Command only available in tox 1.6.0

[testenv:pep8]
deps = flake8
commands = flake8

[flake8]
max-line-length = 120

Unfortunately the version of tox installed in the Jenkins build infra does not support the skipdist parameter which is necessary if your project does not have a setup.py file. A workaround is to create a minimal setup.py file as follows:

setup.py

# Workaround for tox missing parameter 'skipsdist=true' which was not
# introduced until tox 1.6.0

import setuptools

setuptools.setup(name='project-name')

XML

• use self-closing <tags when="whenever possible"/>
• include proper namespace/model/version declarations
• TBD

YANG

• Do not use underscores ('_') in identifiers. JDK 9 is on track to making underscores forbidden in identifiers, which means we will need to map them and it is not going to be pleasant :-(
• Each declaration needs to have either a description or a reference to a defintion document (like an IETF draft)
• Use typedefs to declare concepts. An UUID is typeless, so each instance should have its scope, so we know its applicability domain. 'type string' should only be used to things like free-form comments and similar. Please attach a 'units' statement whenever possible.
• TBD

转载于:https://my.oschina.net/leblancs/blog/795661