Mike Pinkerton
Greg Miller
Dave MacLachlan

This style guide contains many details that are initially hidden from view. They are marked by the triangle icon, which you see here on your left. Click it now. You should see "Hooray" appear below.

Hooray! Now you know you can expand points to get more details. Alternatively, there's an "expand all" at the top of this document.

Objective-C is a very dynamic, object-oriented extension of C. It's designed to be easy to use and read, while enabling sophisticated object-oriented design. It is the primary development language for new applications on Mac OS X and the iPhone.

Cocoa is one of the main application frameworks on Mac OS X. It is a collection of Objective-C classes that provide for rapid development of full-featured Mac OS X applications.

Apple has already written a very good, and widely accepted, coding guide for Objective-C. Google has also written a similar guide for C++. This Objective-C guide aims to be a very natural combination of Apple's and Google's general recommendations. So, before reading this guide, please make sure you've read:

Note that all things that are banned in Google's C++ guide are also banned in Objective-C++, unless explicitly noted in this document.

The purpose of this document is to describe the Objective-C (and Objective-C++) coding guidelines and practices that should be used for all Mac OS X code. Many of these guidelines have evolved and been proven over time on other projects and teams. Open-source projects developed by Google conform to the requirements in this guide.

Google has already released open-source code that conforms to these guidelines as part of the Google Toolbox for Mac project (abbreviated GTM throughout this document). Code meant to be shared across different projects is a good candidate to be included in this repository.

Note that this guide is not an Objective-C tutorial. We assume that the reader is familiar with the language. If you are new to Objective-C or need a refresher, please read The Objective-C Programming Language.

They say an example is worth a thousand words so let's start off with an example that should give you a feel for the style, spacing, naming, etc.

An example header file, demonstrating the correct commenting and spacing for an @interface declaration

  //  GTMFoo.h
  //  FooProject
  //
  //  Created by Greg Miller on 6/13/08.
  //  Copyright 2008 Google, Inc. All rights reserved.
  //

  #import <Foundation/Foundation.h>

  // A sample class demonstrating good Objective-C style. All interfaces,
  // categories, and protocols (read: all top-level declarations in a header)
  // MUST be commented. Comments must also be adjacent to the object they're
  // documenting.
  //
  // (no blank line between this comment and the interface)
  @interface GTMFoo : NSObject {
   @private
    NSString *foo_;
    NSString *bar_;
  }

  // Returns an autoreleased instance of GMFoo. See -initWithString: for details
  // about the argument.
  + (id)fooWithString:(NSString *)string;

  // Designated initializer. |string| will be copied and assigned to |foo_|.
  - (id)initWithString:(NSString *)string;

  // Gets and sets the string for |foo_|.
  - (NSString *)foo;
  - (void)setFoo:(NSString *)newFoo;

  // Does some work on |blah| and returns YES if the work was completed
  // successfuly, and NO otherwise.
  - (BOOL)doWorkWithString:(NSString *)blah;

  @end

An example source file, demonstating the correct commenting and spacing for the @implementation of an interface. It also includes the reference implementations for important methods like getters and setters, init, and dealloc.

  //
  //  GTMFoo.m
  //  FooProject
  //
  //  Created by Greg Miller on 6/13/08.
  //  Copyright 2008 Google, Inc. All rights reserved.
  //

  #import "GTMFoo.h"


  @implementation GTMFoo

  + (id)fooWithString:(NSString *)string {
    return [[[self alloc] initWithString:string] autorelease];
  }

  // Must always override super's designated initializer.
  - (id)init {
    return [self initWithString:nil];
  }

  - (id)initWithString:(NSString *)string {
    if ((self = [super init])) {
      foo_ = [string copy];
      bar_ = [[NSString alloc] initWithFormat:@"hi %d", 3];
    }
    return self;   
  }

  - (void)dealloc {
    [foo_ release];
    [bar_ release];
    [super dealloc];
  }

  - (NSString *)foo {
    return foo_;
  }

  - (void)setFoo:(NSString *)newFoo {
    [foo_ autorelease];
    foo_ = [newFoo copy];   
  }

  - (BOOL)doWorkWithString:(NSString *)blah {
    // ...
    return NO;
  }

  @end

Use only spaces, and indent 2 spaces at a time.

We use spaces for indentation. Do not use tabs in your code. You should set your editor to emit spaces when you hit the tab key.

Each line of text in your code should be at most 80 characters long.

Even though Objective-C tends to be a more verbose language than C++, to aid in the interoperability with that guide, we have decided to keep the limit at 80 columns as well. It's easier to live with than you might expect.

We recognize that this rule is controversial, but so much existing code already adheres to it, and we feel that consistency is important.

You can make violations easier to spot in Xcode by going to Xcode > Preferences > Text Editing > Show page guide.

One space should be used between the - or + and the return type, and no spacing in the parameter list except between parameters.

Methods should look like this:

- (void)doSomethingWithString:(NSString *)theString { ... }

The spacing before the asterisk is optional. When adding new code, be consistent with the surrounding file's style.

If you have too many parameters to fit on one line, giving each its own line is preferred. If multiple lines are used, align each using the : before the parameter.

- (void)doSomethingWith:(GTMFoo *)theFoo rect:(NSRect)theRect interval:(float)theInterval { ... }

The @public and @private access modifiers should be indented by 1 space.

This is similar to public, private, and protected in C++.

@interface MyClass : NSObject { @public ... @private ... } @end

Format exceptions with each @ label on its own line and a space between the @ label and the opening brace '{', as well as between the @catch and the caught object declaration.

If you must use Obj-C exceptions (please see Exceptions for reasons why you should not be using exceptions) format them as follows:

@try { foo(); } @catch (NSException *ex) { bar(ex); } @finally { baz(); }

Naming rules are very import in maintainable code. Objective-C method names tend to be very long, but this has the benefit that a block of code can almost read like prose, thus rendering many comments unnecessary.

When writing pure Objective-C code, we mostly follow standard Objective-C naming rules. These naming guidelines may differ significantly from those outlined in the C++ style guide. For example, Google's C++ style guide recommends the use of underscores between words in variable names, whereas this guide recommends the use of intercaps, which is standard in the Objective-C community.

When writing Objective-C++, however, things are not so cut and dry. Many projects need to implement cross-platform C++ APIs with some Objective-C or Cocoa, or bridge between a C++ back-end and a native Cocoa front-end. This leads to situations where the two guides are directly at odds.

Our solution is that the style follows that of the method/function being implemented. If you're in an @implementation block, use the Objective-C naming rules. If you're implementing a method for a C++ class, use the C++ naming rules. This avoids the situation where instance variable and local variable naming rules are mixed within a single function, which would be a serious detriment to readability.

File names should reflect the name of the class implementation that they contain -- including case. Follow the convention that your project uses.

File extensions should be as follows:

`.h`	C/C++/Objective-C header file
`.m`	Objective-C implementation file
`.mm`	Objective-C++ implementation file
`.cc`	Pure C++ implementation file
`.c`	C implementation file

File names for categories should include the name of the class being extended, e.g. GTMNSString+Utils.h or GTMNSTextView+Autocomplete.h

Within a source file, Objective-C++ follows the style of the function/method you're implementing.

In order to minimize clashes between the differing naming styles when mixing Cocoa/Objective-C and C++, follow the style of the method being implemented. If you're in an @implementation block, use the Objective-C naming rules. If you're implementing a method for a C++ class, use the C++ naming rules.

// file: cross_platform_header.h class CrossPlatformAPI { public: ... int DoSomethingPlatformSpecific(); // impl on each platform private: int an_instance_var_; }; // file: mac_implementation.mm #import "cross_platform_header.h" // A typical Objective-C class, using Objective-C naming. @interface MyDelegate : NSObject { @private int instanceVar_; CrossPlatformAPI* backEndObject_; } - (void)respondToSomething:(id)something; @end @implementation MyDelegate - (void)respondToSomething:(id)something { // bridge from Cocoa through our C++ backend instanceVar_ = backEndObject->DoSomethingPlatformSpecific(); NSString* tempString = [NSString stringWithInt:instanceVar_]; NSLog(@"%@", tempString); } @end // The platform-specific implementation of the C++ class, using // C++ naming. int CrossPlatformAPI::DoSomethingPlatformSpecific() { NSString* temp_string = [NSString stringWithInt:an_instance_var_]; NSLog(@"%@", temp_string); return [temp_string intValue]; }

Class names (along with category and protocol names) should start as uppercase and use mixed case to delimit words.

In application-level code, prefixes on class names should generally be avoided. Having every single class with same prefix impairs readability for no benefit. When designing code to be shared across multiple applications, prefixes are acceptable and recommended (e.g. GTMSendMessage).

Category names should start with a 2 or 3 character prefix identifying the category as part of a project or open for general use. The category name should incorporate the name of the class it's extending.

For example, if we want to create a category on NSString for parsing, we would put the category in a file named GTMNSString+Parsing.h, and the category itself would be named GTMStringParsingAdditions (yes, we know the file name and the category name do not match, but this file could have many separate categories related to parsing). Methods in that category should share the prefix (gtm_myCategoryMethodOnAString:) in order to prevent collisions in Objective-C which only has a single namespace. If the code isn't meant to be shared and/or doesn't run in a different address-space, the method naming isn't quite as important.

Method names should start as lowercase and then use mixed case. Each named parameter should also start as lowercase.

The method name should read like a sentence if possible, meaning you should choose parameter names that flow with the method name. (e.g. convertPoint:fromRect: or replaceCharactersInRange:withString:). See Apple's Guide to Naming Methods for more details.

Accessor methods should be named the same as the variable they're "getting", but they should not be prefixed with the word "get". For example: - (id)getDelegate; // AVOID - (id)delegate; // GOOD

This is for Objective-C methods only. C++ method names and functions continue to follow the rules set in the C++ style guide.

Variables names start with a lowercase and use mixed case to delimit words. Class member variables have trailing underscores. For example: myLocalVariable, myInstanceVariable_. Members used for KVO/KVC bindings may begin with a leading underscore iff use of Objective-C 2.0's @property isn't allowed.

Do not use Hungarian notation for syntactic attributes, such as the static type of a variable (int or pointer). Give as descriptive a name as possible, within reason. Don't worry about saving horizontal space as it is far more important to make your code immediately understandable by a new reader. For example:

int w; int nerr; int nCompConns; tix = [[NSMutableArray alloc] init]; obj = [someObject object]; p = [network port]; int numErrors; int numCompletedConnections; tickets = [[NSMutableArray alloc] init]; userInfo = [someObject object]; port = [network port];

Instance variables are mixed case and should be suffixed with a trailing underscore, e.g. usernameTextField_. However, we permit an exception when binding to a member variable using KVO/KVC and Objective-C 2.0 cannot be used (due to OS release constraints). In this case, it is acceptable to prefix the variable with an underscore, per Apple's accepted practices for key/value naming. If Objective-C 2.0 can be used, @property and @synthesize provide a solution that conforms to the naming guidelines.

Constant names (#defines, enums, const local variables, etc.) should start with a lowercase k and then use mixed case to delimit words, i.e. kInvalidHandle, kWritePerm.

Though a pain to write, they are absolutely vital to keeping our code readable. The following rules describe what you should comment and where. But remember: while comments are very important, the best code is self-documenting. Giving sensible names to types and variables is much better than using obscure names and then trying to explain them through comments.

When writing your comments, write for your audience: the next contributor who will need to understand your code. Be generous — the next one may be you!

Remember that all of the rules and conventions listed in the C++ Style Guide are in effect here, with a few additional points, below.

Start each file with a copyright notice, followed by a description of the contents of the file.

Every file should contain the following items, in order:

a copyright statement (for example, Copyright 2008 Google Inc.)
a license boilerplate. Choose the appropriate boilerplate for the license used by the project (for example, Apache 2.0, BSD, LGPL, GPL)

If you make significant changes to a file that someone else originally wrote, add yourself to the author line. This can be very helpful when another contributor has questions about the file and needs to know whom to contact about it.

Every interface, category, and protocol declaration should have an accompanying comment describing its purpose and how it fits into the larger picture.

// A delegate for NSApplication to handle notifications about app // launch and shutdown. Owned by the main app controller. @interface MyAppDelegate : NSObject { ... } @end

If you have already described an interface in detail in the comments at the top of your file feel free to simply state "See comment at top of file for a complete description", but be sure to have some sort of comment.

Additionally, each method in the public interface should have a comment explaining its function, arguments, return value, and any side effects.

Document the synchronization assumptions the class makes, if any. If an instance of the class can be accessed by multiple threads, take extra care to document the rules and invariants surrounding multithreaded use.

Use vertical bars to quote variable names and symbols in comments rather than quotes or naming the symbol inline.

This helps eliminate ambiguity, especially when the symbol is a common word that might make the sentence read like it was poorly constructed. E.g. for a symbol "count":

// Sometimes we need |count| to be less than zero.

or when quoting something which already contains quotes

// Remember to call |StringWithoutSpaces("foo bar baz")|

Make the pointer ownership model as explicit as possible when it falls outside the most common Objective-C usage idioms.

Instance variable pointers to objects derived from NSObject are presumed to be retained, and should be documented as weak if they are not retained by the class. However, instance variables which are labeled as IBOutlets are presumed to not be retained by the class, and should be documented as strong if the class does retain them.

Where instance variables are pointers to CoreFoundation, C++, and other non-Objective-C objects, they should always be documented in comments as strong or weak. Be mindful that support for automatic C++ objects encapsulated in Objective-C objects is disabled by default, as described here.

Examples of strong and weak documentation: @interface MyDelegate : NSObject { @private IBOutlet* okButton_; // normal NSControl IBOutlet* myContextMenu_; // manually-loaded menu (strong) AnObjcObject* doohickey_; // my doohickey MyController* controller_; // so we can send msgs back (weak, owns me) // non-NSObject pointers... CWackyCPPClass* wacky_; // some cross-platform object (strong) CFDictionaryRef* dict_; // (strong) } @end

strong: The object will be retain'd by this class
weak: The object will be not be retain'd by this class (e.g. a delegate).

Member variables should be declared @private.

@interface MyClass : NSObject { @private id myInstanceVariable_; } // public accessors, setter takes ownership - (id)myInstanceVariable; - (void)setMyInstanceVariable:(id)theVar; @end

Comment and clearly identify your designated initializer.

It is important for those who might be subclassing your class that the designated initializer be clearly identified. That way, they only need to subclass a single initializer (of potentially several) to guarantee their subclass' initializer is called. It also helps those debugging your class in the future understand the flow of initialization code if they need to step through it.

When writing a subclass that requires an init... method, make sure you override the superclass' designated initializer.

If you fail to override the superclass' designated initializer, your initializer may not be called in all cases, leading to subtle and very difficult to find bugs.

Don't initialize variables to 0 or nil in the init method; it's redundant.

All memory for a newly allocated object is initialized to 0 (except for isa), so don't clutter up the init method by re-initializing variables to 0 or nil.

Keep your class simple; avoid "kitchen-sink" APIs. If a method doesn't need to be public, don't make it so. Use a private category to prevent cluttering the public header.

Unlike C++, Objective-C doesn't have a way to differentiate between public and private methods — everything is public. As a result, avoid placing methods in the public API unless they are actually expected to be used by a consumer of the class. This helps reduce the likelihood they'll be called when you're not expecting it. This includes methods that are being overridden from the parent class. For internal implementation methods, use a category defined in the implementation file as opposed to adding them to the public header.

// GTMFoo.m #import "GTMFoo.h" @interface GTMFoo (PrivateDelegateHandling) - (NSString *)doSomethingWithDelegate; // Declare private method @end @implementation GTMFoo(PrivateDelegateHandling) ... - (NSString *)doSomethingWithDelegate { // Implement this method } ... @end

Before Objective-C 2.0, if you declare a method in the private @interface, but forget to implement it in the main @implementation, the compiler will not object. (This is because you don't implement these private methods in a separate category.) The solution is to put the functions within an @implementation that specifies the category.

If you are using Objective-C 2.0, you should instead declare your private category using a class extension, for example:

@interface GMFoo () { ... }

which will guarantee that the declared methods are implemented in the @implementation section by issuing a compiler warning if they are not.

Again, "private" methods are not really private. You could accidentally override a superclass's "private" method, thus making a very difficult bug to squash. In general, private methods should have a fairly unique name that will prevent subclasses from unintentionally overriding them.

Finally, Objective-C categories are a great way to segment a large @implementation section into more understandable chunks and to add new, application-specific functionality to the most appropriate class. For example, instead of adding "middle truncation" code to a random object in your app, make a new category on NSString).

Include root frameworks over individual files.

While it may seem tempting to include individual system headers from a framework such as Cocoa or Foundation, in fact it's less work on the compiler if you include the top-level root framework. The root framework is generally pre-compiled and can be loaded much more quickly. In addition, remember to use #import rather than #include.

#import <Foundation/Foundation.h> // good #import <Foundation/NSArray.h> // avoid #import <Foundation/NSString.h> ...

When creating new temporary objects, autorelease them on the same line as you create them rather than a separate release later in the same method.

While ever so slightly slower, this prevents someone from accidentally removing the release or inserting a return before it and introducing a memory leak. E.g.:

// AVOID (unless you have a compelling performance reason) MyController* controller = [[MyController alloc] init]; // ... code here that might return ... [controller release]; // BETTER MyController* controller = [[[MyController alloc] init] autorelease];

Assignment of objects follows the autorelease then retain pattern.

When assigning a new object to a variable, one must first release the old object to avoid a memory leak. There are several "correct" ways to handle this. We've chosen the "autorelease then retain" approach because it's less prone to error. Be aware in tight loops it can fill up the autorelease pool, and may be slightly less efficient, but we feel the tradeoffs are acceptable.

- (void)setFoo:(GMFoo *)aFoo { [foo_ autorelease]; // Won't dealloc if |foo_| == |aFoo| foo_ = [aFoo retain]; }

Setters taking an NSString, should always copy the string it accepts.

Never just retain the string. This avoids the caller changing it under you without your knowledge. Don't assume that because you're accepting an NSString that it's not actually an NSMutableString.

- (void)setFoo:(NSString *)aFoo { [foo_ autorelease]; foo_ = [aFoo copy]; }

Don't @throw Objective-C exceptions, but you should be prepared to catch them from third-party or OS calls.

We do compile with -fobjc-exceptions (mainly so we get @synchronized), but we don't @throw. Use of @try, @catch, and @finally are allowed when required to properly use 3rd party code or libraries. If you do use them please document exactly which methods you expect to throw.

Do not use the NS_DURING, NS_HANDLER, NS_ENDHANDLER, NS_VALUERETURN and NS_VOIDRETURN macros unless you are writing code that needs to run on MacOS 10.2 or before.

Also be aware when writing Objective-C++ code that stack based objects are not cleaned up when you throw an Objective-C exception. Example:

class exceptiontest { public: exceptiontest() { NSLog(@"Created"); } ~exceptiontest() { NSLog(@"Destroyed"); } }; void foo() { exceptiontest a; NSException *exception = [NSException exceptionWithName:@"foo" reason:@"bar" userInfo:nil]; @throw exception; } int main(int argc, char *argv[]) { GMAutoreleasePool pool; @try { foo(); } @catch(NSException *ex) { NSLog(@"exception raised"); } return 0; }

will give you:

2006-09-28 12:34:29.244 exceptiontest[23661] Created 2006-09-28 12:34:29.244 exceptiontest[23661] exception raised

Note that the destructor for a never got called. This is a major concern for stack based smartptrs such as shared_ptr and linked_ptr, as well as any STL objects that you may want to use. Therefore it pains us to say that if you must use exceptions in your Objective-C++ code, use C++ exceptions whenever possible. You should never re-throw an Objective-C exception, nor are stack based C++ objects (such as std::string, std::vector etc.) allowed in the body of any @try, @catch, or @finally blocks.

Use nil checks for logic flow only.

Use nil checks for logic flow of the application, not for crash prevention. Sending a message to a nil object is handled by the Objective-C runtime. If the method has no return result, you're good to go. However if there is one, there may be differences based on runtime architecture, return size, and OS X version (see Apple's documentation for specifics).

Note that this is very different from checking C/C++ pointers against NULL, which the runtime does not handle and will cause your application to crash. You still need to make sure you do not dereference a NULL pointer.

Be careful with BOOL return values and mixing bool with BOOL. Avoid comparing directly with YES.

Only use the constants YES and NO for BOOL return values and assignments. Common mistakes include casting an array's size or a pointer value as a BOOL, which depending on the value of the last byte, could still result in a NO value.

Also be careful when interoperating with C++ and its bool type. Don't just assume that YES and NO map directly to true and false. You can use the ternary operator to succinctly map between them.

bool b = myBOOL ? true : false; // to go one way... myBOOL = b ? YES : NO; // ... or the other

Finally, don't directly compare BOOL variables directly with YES. Not only is is harder to read for those well-versed in C, the first point above demonstrates that return values may not always be what you expect.

BOOL great = [foo isGreat]; if (great == YES) // ...be great! BOOL great = [foo isGreat]; if (great) // ...be great!

Delegate objects should not be retained.

A class that implements the delegate pattern should:

Have an instance variable named delegate_ to reference the delegate.
Thus, the accessor methods should be named delegate and setDelegate:.
The delegate_ object should not be retained.

Separate the model from the view. Separate the controller from the view and the model. Use @protocols for callback APIs.

Separate model from view: don't build assumptions about the presentation into the model or data source. Keep the interface between the data source and the presentation abstract. Don't give the model knowledge of its view. (A good rule of thumb is to ask yourself if it's possible to have multiple presentations, with different states, on a single instance of your data source.)
Separate controller from view and model: don't put all of the "business logic" into view-related classes; this makes the code very unusable. Make controller classes to host this code, but ensure that the controller classes don't make too many assumptions about the presentation.
Use a protocol for callback APIs where all the methods must be implemented. Use a category (or an "informal protocol") when not all the methods need to be implemented.

Mike Pinkerton
Greg Miller
Dave MacLachlan