The rules below are not guidelines or recommendations, but strict rules.Contributions to Android generally will not be accepted if they do notadhere to these rules.
Not all existing code follows these rules, but all new code is expected to.
We follow standard Java coding conventions. We add a few rules:
Sometimes it is tempting to write code that completely ignores an exceptionlike this:
void setServerPort(String value) { try { serverPort = Integer.parseInt(value); } catch (NumberFormatException e) { }}You must never do this. While you may think that your code will neverencounter this error condition or that it is not important to handle it,ignoring exceptions like above creates mines in your code for someone else totrip over some day. You must handle every Exception in your code in someprincipled way. The specific handling varies depending on the case.
Anytime somebody has an empty catch clause they should have acreepy feeling. There are definitely times when it is actually the correctthing to do, but at least you have to think about it. In Java you can't escapethe creepy feeling. -James Gosling
Acceptable alternatives (in order of preference) are:
Throw the exception up to the caller of your method.
void setServerPort(String value) throws NumberFormatException { serverPort = Integer.parseInt(value);}Throw a new exception that's appropriate to your level of abstraction.
void setServerPort(String value) throws ConfigurationException { try { serverPort = Integer.parseInt(value); } catch (NumberFormatException e) { throw new ConfigurationException("Port " + value + " is not valid."); }}Handle the error gracefully and substitute an appropriate value in thecatch {} block.
/** Set port. If value is not a valid number, 80 is substituted. */void setServerPort(String value) { try { serverPort = Integer.parseInt(value); } catch (NumberFormatException e) { serverPort = 80; // default port for server }}Catch the Exception and throw a new RuntimeException. This is dangerous:only do it if you are positive that if this error occurs, the appropriatething to do is crash.
/** Set port. If value is not a valid number, die. */void setServerPort(String value) { try { serverPort = Integer.parseInt(value); } catch (NumberFormatException e) { throw new RuntimeException("port " + value " is invalid, ", e); }}Note that the original exception is passed to the constructor forRuntimeException. If your code must compile under Java 1.3, you will need toomit the exception that is the cause.
Last resort: if you are confident that actually ignoring the exception isappropriate then you may ignore it, but you must also comment why with a goodreason:
/** If value is not a valid number, original port number is used. */void setServerPort(String value) { try { serverPort = Integer.parseInt(value); } catch (NumberFormatException e) { // Method is documented to just ignore invalid user input. // serverPort will just be unchanged. }}Sometimes it is tempting to be lazy when catching exceptions and dosomething like this:
try { someComplicatedIOFunction(); // may throw IOException someComplicatedParsingFunction(); // may throw ParsingException someComplicatedSecurityFunction(); // may throw SecurityException // phew, made it all the way } catch (Exception e) { // I'll just catch all exceptions handleError(); // with one generic handler!}You should not do this. In almost all cases it is inappropriate to catchgeneric Exception or Throwable, preferably not Throwable, because it includesError exceptions as well. It is very dangerous. It means that Exceptions younever expected (including RuntimeExceptions like ClassCastException) end upgetting caught in application-level error handling. It obscures the failurehandling properties of your code. It means if someone adds a new type ofException in the code you're calling, the compiler won't help you realize youneed to handle that error differently. And in most cases you shouldn't behandling different types of exception the same way, anyway.
There are rare exceptions to this rule: certain test code and top-levelcode where you want to catch all kinds of errors (to prevent them from showingup in a UI, or to keep a batch job running). In that case you may catchgeneric Exception (or Throwable) and handle the error appropriately. Youshould think very carefully before doing this, though, and put in commentsexplaining why it is safe in this place.
Alternatives to catching generic Exception:
Catch each exception separately as separate catch blocks after a singletry. This can be awkward but is still preferable to catching all Exceptions.Beware repeating too much code in the catch blocks.
Refactor your code to have more fine-grained error handling, with multipletry blocks. Split up the IO from the parsing, handle errors separately in eachcase.
Rethrow the exception. Many times you don't need to catch the exception atthis level anyway, just let the method throw it.
Remember: exceptions are your friend! When the compiler complains you'renot catching an exception, don't scowl. Smile: the compiler just made iteasier for you to catch runtime problems in your code.
Finalizers are a way to have a chunk of code executedwhen an object is garbage collected.
Pros: can be handy for doing cleanup, particularly of external resources.
Cons: there are no guarantees as to when a finalizer will be called,or even that it will be called at all.
Decision: we don't use finalizers. In most cases, you can do whatyou need from a finalizer with good exception handling. If you absolutely needit, define a close() method (or the like) and document exactly when thatmethod needs to be called. See InputStream for an example. In this case it isappropriate but not required to print a short log message from the finalizer,as long as it is not expected to flood the logs.
When you want to use class Bar from package foo,thereare two possible ways to import it:
import foo.*;Pros: Potentially reduces the number of import statements.
import foo.Bar;Pros: Makes it obvious what classes are actually used. Makescode more readable for maintainers.
Decision: Use the latter for importing all Android code. An explicitexception is made for java standard libraries (java.util.*, java.io.*, etc.)and unit test code (junit.framework.*)
There are conventions for using Android's Java libraries and tools. In somecases, the convention has changed in important ways and older code might use adeprecated pattern or library. When working with such code, it's okay tocontinue the existing style (see Consistency). Whencreating new components never use deprecated libraries.
Every file should have a copyright statement at the top. Then a packagestatement and import statements should follow, each block separated by a blankline. And then there is the class or interface declaration. In the Javadoccomments, describe what the class or interface does.
/* * Copyright (C) 2010 The Android Open Source Project * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. */package com.android.internal.foo;import android.os.Blah;import android.view.Yada;import java.sql.ResultSet;import java.sql.SQLException;/** * Does X and Y and provides an abstraction for Z. */public class Foo { ...}Every class and nontrivial public method you write must contain aJavadoc comment with at least one sentence describing what the class or methoddoes. This sentence should start with a 3rd person descriptive verb.
Examples:
/** Returns the correctly rounded positive square root of a double value. */static double sqrt(double a) { ...}or
/** * Constructs a new String by converting the specified array of * bytes using the platform's default character encoding. */public String(byte[] bytes) { ...}You do not need to write Javadoc for trivial get and set methods such assetFoo() if all your Javadoc would say is "sets Foo". If the method doessomething more complex (such as enforcing a constraint or having an importantside effect), then you must document it. And if it's not obvious what theproperty "Foo" means, you should document it.
Every method you write, whether public or otherwise, would benefit fromJavadoc. Public methods are part of an API and therefore require Javadoc.
Android does not currently enforce a specific style for writing Javadoccomments, but you should follow the Sun Javadoc conventions.
To the extent that it is feasible, methods should be kept small andfocused. It is, however, recognized that long methods are sometimesappropriate, so no hard limit is placed on method length. If a method exceeds40 lines or so, think about whether it can be broken up without harming thestructure of the program.
Fields should be defined either at the top of the file, or immediately before the methods that use them.
The scope of local variables should be kept to a minimum (EffectiveJava Item 29). By doing so, you increase the readability andmaintainability of your code and reduce the likelihood of error. Each variableshould be declared in the innermost block that encloses all uses of thevariable.
Local variables should be declared at the point they are first used. Nearlyevery local variable declaration should contain an initializer. If you don'tyet have enough information to initialize a variable sensibly, you shouldpostpone the declaration until you do.
One exception to this rule concerns try-catch statements. If a variable isinitialized with the return value of a method that throws a checked exception,it must be initialized inside a try block. If the value must be used outsideof the try block, then it must be declared before the try block, where itcannot yet be sensibly initialized:
// Instantiate class cl, which represents some sort of Set Set s = null;try { s = (Set) cl.newInstance();} catch(IllegalAccessException e) { throw new IllegalArgumentException(cl + " not accessible");} catch(InstantiationException e) { throw new IllegalArgumentException(cl + " not instantiable");}// Exercise the set s.addAll(Arrays.asList(args));But even this case can be avoided by encapsulating the try-catch block in a method:
Set createSet(Class cl) { // Instantiate class cl, which represents some sort of Set try { return (Set) cl.newInstance(); } catch(IllegalAccessException e) { throw new IllegalArgumentException(cl + " not accessible"); } catch(InstantiationException e) { throw new IllegalArgumentException(cl + " not instantiable"); }}...// Exercise the set Set s = createSet(cl);s.addAll(Arrays.asList(args));Loop variables should be declared in the for statement itself unless thereis a compelling reason to do otherwise:
for (int i = 0; i < n; i++) { doSomething(i);}and
for (Iterator i = c.iterator(); i.hasNext(); ) { doSomethingElse(i.next());}The ordering of import statements is:
Android imports
Imports from third parties (com, junit, net, org)
java and javax
To exactly match the IDE settings, the imports should be:
Alphabetical within each grouping, with capital letters before lower case letters (e.g. Z before a).
There should be a blank line between each major grouping (android, com, junit, net, org, java, javax).
Originally there was no style requirement on the ordering. This meant thatthe IDE's were either always changing the ordering, or IDE developers had todisable the automatic import management features and maintain the imports byhand. This was deemed bad. When java-style was asked, the preferred styleswere all over the map. It pretty much came down to our needing to "pick anordering and be consistent." So we chose a style, updated the style guide, andmade the IDEs obey it. We expect that as IDE users work on the code, theimports in all of the packages will end up matching this pattern without anyextra engineering effort.
This style was chosen such that:
The imports people want to look at first tend to be at the top (android)
The imports people want to look at least tend to be at the bottom (java)
Humans can easily follow the style
IDEs can follow the style
The use and location of static imports have been mildly controversialissues. Some people would prefer static imports to be interspersed with theremaining imports, some would prefer them reside above or below all otherimports. Additinally, we have not yet come up with a way to make all IDEs usethe same ordering.
Since most people consider this a low priority issue, just use yourjudgement and please be consistent.
We use 4 space indents for blocks. We never use tabs. When in doubt, beconsistent with code around you.
We use 8 space indents for line wraps, including function calls andassignments. For example, this is correct:
Instrument i = someLongExpression(that, wouldNotFit, on, one, line);and this is not correct:
Instrument i = someLongExpression(that, wouldNotFit, on, one, line);Non-public, non-static field names start with m.
Static field names start with s.
Other fields start with a lower case letter.
Public static final fields (constants) are ALL_CAPS_WITH_UNDERSCORES.
For example:
public class MyClass { public static final int SOME_CONSTANT = 42; public int publicField; private static MyClass sSingleton; int mPackagePrivate; private int mPrivate; protected int mProtected;}Braces do not go on their own line; they go on the same line as the codebefore them. So:
class MyClass { int func() { if (something) { // ... } else if (somethingElse) { // ... } else { // ... } }}We require braces around the statements for a conditional. Except, if theentire conditional (the condition and the body) fit on one line, you may (butare not obligated to) put it all on one line. That is, this is legal:
if (condition) { body(); }and this is legal:
if (condition) body();but this is still illegal:
if (condition) body(); // bad!Each line of text in your code should be at most 100 characters long.
There has been lots of discussion about this rule and the decision remainsthat 100 characters is the maximum.
Exception: if a comment line contains an example command or a literal URLlonger than 100 characters, that line may be longer than 100 characters forease of cut and paste.
Exception: import lines can go over the limit because humans rarely seethem. This also simplifies tool writing.
Annotations should precede other modifiers for the same language element.Simple marker annotations (e.g. @Override) can be listed on the same line withthe language element. If there are multiple annotations, or parameterizedannotations, they should each be listed one-per-line in alphabeticalorder.<
Android standard practices for the three predefined annotations in Java are:
@Deprecated: The @Deprecated annotation must be used whenever the use of the annotatedelement is discouraged. If you use the @Deprecated annotation, you must alsohave a @deprecated Javadoc tag and it should name an alternate implementation.In addition, remember that a @Deprecated method is still supposed towork.
If you see old code that has a @deprecated Javadoc tag, please add the @Deprecated annotation.
@Override: The @Override annotation must be used whenever a method overrides thedeclaration or implementation from a super-class.
For example, if you use the @inheritdocs Javadoc tag, and derive from aclass (not an interface), you must also annotate that the method @Overridesthe parent class's method.
@SuppressWarnings: The @SuppressWarnings annotation should only be used under circumstanceswhere it is impossible to eliminate a warning. If a warning passes this"impossible to eliminate" test, the @SuppressWarnings annotation must beused, so as to ensure that all warnings reflect actual problems in thecode.
When a @SuppressWarnings annotation is necessary, it must be prefixed witha TODO comment that explains the "impossible to eliminate" condition. Thiswill normally identify an offending class that has an awkward interface. Forexample:
// TODO: The third-party class com.third.useful.Utility.rotate() needs generics @SuppressWarnings("generic-cast")List<String> blix = Utility.rotate(blax);When a @SuppressWarnings annotation is required, the code should berefactored to isolate the software elements where the annotation applies.
Treat acronyms and abbreviations as words in naming variables, methods, and classes. The names are much more readable:
| Good | Bad |
|---|---|
| XmlHttpRequest | XMLHTTPRequest |
| getCustomerId | getCustomerID |
| class Html | class HTML |
| String url | String URL |
| long id | long ID |
Both the JDK and the Android code bases are very inconsistent with regardsto acronyms, therefore, it is virtually impossible to be consistent with thecode around you. Bite the bullet, and treat acronyms as words.
For further justifications of this style rule, see Effective JavaItem 38 and Java Puzzlers Number 68.
Use TODO comments for code that is temporary, a short-term solution, orgood-enough but not perfect.
TODOs should include the string TODO in all caps, followed by a colon:
// TODO: Remove this code after the UrlTable2 has been checked in.and
// TODO: Change this to use a flag instead of a constant.If your TODO is of the form "At a future date do something" make sure thatyou either include a very specific date ("Fix by November 2005") or a veryspecific event ("Remove this code after all production mixers understandprotocol V7.").
While logging is necessary it has a significantly negative impact onperformance and quickly loses its usefulness if it's not kept reasonablyterse. The logging facilities provides five different levels of logging. Beloware the different levels and when and how they should be used.
ERROR: This level of logging should be used when something fatal has happened,i.e. something that will have user-visible consequences and won't berecoverable without explicitly deleting some data, uninstalling applications,wiping the data partitions or reflashing the entire phone (or worse). Thislevel is always logged. Issues that justify some logging at the ERROR levelare typically good candidates to be reported to a statistics-gatheringserver.
WARNING: This level of logging should used when something serious and unexpectedhappened, i.e. something that will have user-visible consequences but islikely to be recoverable without data loss by performing some explicit action,ranging from waiting or restarting an app all the way to re-downloading a newversion of an application or rebooting the device. This level is alwayslogged. Issues that justify some logging at the WARNING level might also beconsidered for reporting to a statistics-gathering server.
INFORMATIVE:This level of logging should used be to note that something interesting tomost people happened, i.e. when a situation is detected that is likely to havewidespread impact, though isn't necessarily an error. Such a condition shouldonly be logged by a module that reasonably believes that it is the mostauthoritative in that domain (to avoid duplicate logging by non-authoritativecomponents). This level is always logged.
DEBUG:This level of logging should be used to further note what is happening on thedevice that could be relevant to investigate and debug unexpected behaviors.You should log only what is needed to gather enough information about what isgoing on about your component. If your debug logs are dominating the log thenyou probably should be using verbose logging.
This level will be logged, evenon release builds, and is required to be surrounded by an if (LOCAL_LOG) or if(LOCAL_LOGD) block, where LOCAL_LOG[D] is defined in your class orsubcomponent, so that there can exist a possibility to disable all suchlogging. There must therefore be no active logic in an if (LOCAL_LOG) block.All the string building for the log also needs to be placed inside the if(LOCAL_LOG) block. The logging call should not be re-factored out into amethod call if it is going to cause the string building to take place outsideof the if (LOCAL_LOG) block.
There is some code that still says if(localLOGV). This is considered acceptable as well, although the name isnonstandard.
VERBOSE:This level of logging should be used for everything else. This level will onlybe logged on debug builds and should be surrounded by an if (LOCAL_LOGV) block(or equivalent) so that it can be compiled out by default. Any string buildingwill be stripped out of release builds and needs to appear inside the if (LOCAL_LOGV) block.
Notes:
Within a given module, other than at the VERBOSE level, anerror should only be reported once if possible: within a single chain offunction calls within a module, only the innermost function should return theerror, and callers in the same module should only add some logging if thatsignificantly helps to isolate the issue.
In a chain of modules, other than at the VERBOSE level, when alower-level module detects invalid data coming from a higher-level module, thelower-level module should only log this situation to the DEBUG log, and onlyif logging provides information that is not otherwise available to the caller.Specifically, there is no need to log situations where an exception is thrown(the exception should contain all the relevant information), or where the onlyinformation being logged is contained in an error code. This is especiallyimportant in the interaction between the framework and applications, andconditions caused by third-party applications that are properly handled by theframework should not trigger logging higher than the DEBUG level. The onlysituations that should trigger logging at the INFORMATIVE level or higher iswhen a module or application detects an error at its own level or coming froma lower level.
When a condition that would normally justify some logging islikely to occur many times, it can be a good idea to implement somerate-limiting mechanism to prevent overflowing the logs with many duplicatecopies of the same (or very similar) information.
Losses of network connectivity are considered common and fullyexpected and should not be logged gratuitously. A loss of network connectivitythat has consequences within an app should be logged at the DEBUG or VERBOSElevel (depending on whether the consequences are serious enough and unexpectedenough to be logged in a release build).
A full filesystem on a filesystem that is acceessible to or onbehalf of third-party applications should not be logged at a level higher thanINFORMATIVE.
Invalid data coming from any untrusted source (including anyfile on shared storage, or data coming through just about any networkconnections) is considered expected and should not trigger any logging at alevel higher then DEBUG when it's detected to be invalid (and even thenlogging should be as limited as possible).
Keep in mind that the + operator, when used on Strings,implicitly creates a StringBuilder with the default buffer size (16characters) and potentially quite a few other temporary String objects, i.e.that explicitly creating StringBuilders isn't more expensive than relying onthe default '+' operator (and can be a lot more efficient in fact). Also keepin mind that code that calls Log.v() is compiled and executed on releasebuilds, including building the strings, even if the logs aren't beingread.
Any logging that is meant to be read by other people and to beavailable in release builds should be terse without being cryptic, and shouldbe reasonably understandable. This includes all logging up to the DEBUGlevel.
When possible, logging should be kept on a single line if itmakes sense. Line lengths up to 80 or 100 characters are perfectly acceptable,while lengths longer than about 130 or 160 characters (including the length ofthe tag) should be avoided if possible.
Logging that reports successes should never be used at levelshigher than VERBOSE.
Temporary logging that is used to diagnose an issue that'shard to reproduce should be kept at the DEBUG or VERBOSE level, and should beenclosed by if blocks that allow to disable it entirely at compile-time.
Be careful about security leaks through the log. Privateinformation should be avoided. Information about protected content mustdefinitely be avoided. This is especially important when writing frameworkcode as it's not easy to know in advance what will and will not be privateinformation or protected content.
System.out.println() (or printf() for native code) shouldnever be used. System.out and System.err get redirected to /dev/null, so yourprint statements will have no visible effects. However, all the stringbuilding that happens for these calls still gets executed.
The golden rule of logging is that your logs may notunnecessarily push other logs out of the buffer, just as others may not pushout yours.
Our parting thought: BE CONSISTENT. If you're editing code, take a fewminutes to look at the code around you and determine its style. If they usespaces around their if clauses, you should too. If their comments have littleboxes of stars around them, make your comments have little boxes of starsaround them too.
The point of having style guidelines is to have a common vocabulary ofcoding, so people can concentrate on what you're saying, rather than on howyou're saying it. We present global style rules here so people know thevocabulary. But local style is also important. If code you add to a a filelooks drastically different from the existing code around it, it throwsreaders out of their rhythm when they go to read it. Try to avoid this.
When naming test methods, you can use an underscore to seperate what isbeing tested from the specific case being tested. This style makes it easierto see exactly what cases are being tested.
For example:
testMethod_specificCase1 testMethod_specificCase2void testIsDistinguishable_protanopia() { ColorMatcher colorMatcher = new ColorMatcher(PROTANOPIA) assertFalse(colorMatcher.isDistinguishable(Color.RED, Color.BLACK)) assertTrue(colorMatcher.isDistinguishable(Color.X, Color.Y))}联系客服
