mikeash.com pyblog/friday-qa-2012-06-22-objective-c-literals.html comments

bob - 2014-10-28 23:32:00

Tue, 28 Oct 2014 23:32:00 GMT

Addendum to my previous post: Comparison operators in C (but not C++), like <, >=, ==, etc. also all return an int. So:

NSLog(@"%@", @(42 == 17).class); // prints "__NSCFNumber"

bob - 2014-10-28 23:24:36

Tue, 28 Oct 2014 23:24:36 GMT

One thing to be careful about with boxed expressions is that the underlying "type" of the resulting NSNumber object depends on the compile-time type of the expression boxed, which may sometimes not be what you expect. In Objective-C:



BOOL foo = NO;

NSLog(@"%@", @(foo).class); // prints "__NSCFBoolean"

NSLog(@"%@", @(!foo).class); // prints "__NSCFNumber"

NSLog(@"%@", @((BOOL)!foo).class); // prints "__NSCFBoolean"

Although we normally think of the NOT ("!") of a BOOL to be a BOOL, in C, and hence Objective-C, the result of "!" always has type int. It works fine when we assign it to a BOOL or use it in a comparison, but boxed expressions care a lot about the actual compile-time type, and creates the NSNumber object differently depending on it.

(Note that in C++, "!" returns type bool, so if you rename the file ".mm" to make it Objective-C++, then the above will all print "__NSCFBoolean". This change in behavior by just changing the language is also counterintuitive.)

Ashley Mills - 2014-02-26 14:35:26

Wed, 26 Feb 2014 14:35:26 GMT

I'd be curious to know if there's anyway to use the @{} dictionary initialiser to initialise another class. For example:



NSFetchRequest * fetchRequest = @{@"Entity" : @"SomeEntity", @"attributeName" : @"value"};

which would be equivalent to:



NSFetchRequest * fetchRequest = [NSFetchRequest fetchRequestWithEntityName: @"Entity"];

fetchRequest.predicate = [NSPredicate predicateWithFormat: @"(%K == %@)", @"attributeName", @"value"];

Hoon H. - 2013-03-23 00:37:00

Sat, 23 Mar 2013 00:37:00 GMT

It's also possible to substitute literal classes with `@compatibility_alias`.



#import <Foundation/NSObject.h>



@interface  AAA : NSObject

+ (id)numberWithInt:(int)num;

@end



@implementation AAA

+ (id)numberWithInt:(int)num

{

    return  @"AAAAA!!!";    // Abused type system just to check result.

}

@end



@compatibility_alias NSNumber AAA;

Stuart Carnie - 2013-01-09 16:44:20

Wed, 09 Jan 2013 16:44:20 GMT

@mike,

You are right, the vertical bars would pose a problem when parsing constant expressions.

NSSet *s = @(| @"test", @"one", @"two" |);

No whitespace for @(| and |) would make it more resilient…

Dan - 2012-12-20 16:10:35

Thu, 20 Dec 2012 16:10:35 GMT

I'm so happy I found this. Great post!

rx - 2012-11-29 07:00:14

Thu, 29 Nov 2012 07:00:14 GMT

Never mind, had the old gcc compiler set -- sorry!

rx - 2012-11-29 06:53:58

Thu, 29 Nov 2012 06:53:58 GMT

Hi, could anyone explain to me why this throws a compiler error ("Expected expression before @ token", haha, which one?)



NSArray *myArray = @[@{@"x1":@0,@"y1":@100,@"x2":@100,@"y2":@110}];

Thanks and best regards,

rx

alex gray - 2012-09-29 09:40:19

Sat, 29 Sep 2012 09:40:19 GMT

@Anonymous

just throw the following in a header....

#define $(...) ((NSString *)[NSString stringWithFormat:__VA_ARGS__,nil])

and then use like

[$(@"whatever%ld %@", 2, obj ) stringByRemoving......

Kat - 2012-07-27 22:39:02

Fri, 27 Jul 2012 22:39:02 GMT

Xcode's auto-indentation has trouble with existing Objective-C code. I'm a little disappointed, but not surprised, that it doesn't handle new literals so well yet. :-(

Anonymous - 2012-07-25 19:18:56

Wed, 25 Jul 2012 19:18:56 GMT

Also, Xcode's auto-indentation is pretty weird:



        self.navigationBar.titleTextAttributes = @{

    UITextAttributeFont: [UIFont systemFontOfSize:12.0],

    UITextAttributeTextColor: [UIColor colorWithWhite:0.5 alpha:1.0],

    UITextAttributeTextShadowColor: [UIColor clearColor],

        };

I would expect it at least to look like:



        self.navigationBar.titleTextAttributes = @{

            UITextAttributeFont: [UIFont systemFontOfSize:12.0],

            UITextAttributeTextColor: [UIColor colorWithWhite:0.5 alpha:1.0],

            UITextAttributeTextShadowColor: [UIColor clearColor],

        };

Anonymous - 2012-07-25 19:13:09

Wed, 25 Jul 2012 19:13:09 GMT

Very nice. I was hoping they would add some shortcut for +stringWithFormat: too, like @("foo %d", 1). Oh well.

mikeash - 2012-07-09 14:21:26

Mon, 09 Jul 2012 14:21:26 GMT

How about some emoji syntax?

NSSet *vowels = @💩@"a, @"e", @"i", @"o", @"u"💩;

(That probably won't show up for everybody. In fact, it doesn't show up for me.)

Regarding Stuart's syntax idea, I don't think it's workable. The problem is not the beginning, but the end, which results in ambiguity since | can be an operator or a set-end marker. For example:

NSSet *pathological = @| 0 | -1 | -2;

Of course, this is ultimately invalid code, but you can't know that until well after the actual parsing. There are probably ways to come up with an ambiguous case that is valid, but I don't quite have the energy to figure it out.

jamie - 2012-07-08 16:19:37

Sun, 08 Jul 2012 16:19:37 GMT

Stuart: I see. Are you sure I can't interest you in a nice unicode character pair? Seems like just the job for the black lenticular brackets :)

NSSet *vowels = @【@"a","@"e",@"i",@"o",@"u"】;

Stuart carnie - 2012-07-06 23:07:54

Fri, 06 Jul 2012 23:07:54 GMT

jamie: The Clang parser is a recursive descent parser, so this would work just fine as you have pretty fine-grained control. When the parser hits an @ symbol in an expression, Clang calls the Parser::ParseObjCAtExpression (https://github.com/llvm-mirror/clang/blob/4d3db4eb6caa49a7cdbfe1798728ce4b23cd0b53/lib/Parse/ParseObjc.cpp#L2019) you can simply add to the switch "if the next symbol following an @ is "|" start parsing an NSSet literal.

Cheers

jamie - 2012-07-04 06:59:23

Wed, 04 Jul 2012 06:59:23 GMT

Stuart: I dunno if that set syntax would work without significantly rewriting the Objective-C grammar, since the single bar is going to get eaten by the parser as part of the assignment-expression. Looks like a job for a digraph.

Stuart Carnie - 2012-06-25 05:56:23

Mon, 25 Jun 2012 05:56:23 GMT

Good stuff, Mike. Out of interest, I created a 3-part series in adding my own literal, NSURL to Clang. http://aussiebloke.blogspot.com/2012/06/llvm-clang-hacking-part-1.html

NSSet is interesting. Would be easy to play with an experimental literal syntax for NSSet, like:

NSSet *theSet = @| @"one", @"two", @"three" |

Jordan Rose - 2012-06-23 22:47:05

Sat, 23 Jun 2012 22:47:05 GMT

Ken: Oops, yes, I take it back. I thought you could declare these to return any type you wanted, but it actually produces an error if it's not an Objective-C type. (You can pick something other than id, though, just like any other covariant override.)

(How embarrassing, considering I work on Clang!)

Ken Thomases - 2012-06-23 18:46:39

Sat, 23 Jun 2012 18:46:39 GMT

Jordan: I suspect if you're using ARC and you make those methods return non-objects, you'll be burned quite badly.

Jordan Rose - 2012-06-23 03:28:03

Sat, 23 Jun 2012 03:28:03 GMT

As a note, the return types of -objectAtIndexedSubscript: and -objectForKeyedSubcript: do not have to be id. There probably aren't that many times where this will actually come in handy, but it's there if you want it.

Yuji Tachikawa - 2012-06-23 02:53:38

Sat, 23 Jun 2012 02:53:38 GMT

Thank you very much for the nice article. It reminded me of a very similar syntax WebScript (a scripting version of Objective-C WebObjects once had) had, see Apple's documentation, http://developer.apple.com/legacy/mac/library/documentation/LegacyTechnologies/WebObjects/WebObjects_3.1/DevGuide/WebScript/CreatingObjects.html#REF65852

I never used WebScript or WebObjects to start with. I wonder how they were implemented.

By the way, WebScript also had a "modern" function call/definition syntax, http://developer.apple.com/legacy/mac/library/documentation/LegacyTechnologies/WebObjects/WebObjects_3.1/DevGuide/WebScript/ModernSyntax.html#REF54790 .

Cédric Luthi - 2012-06-22 22:11:55

Fri, 22 Jun 2012 22:11:55 GMT

The subscripting compatibility shim is actually part of ARCLite. The __ARCLite__load function (which oddly enough is called by the objc runtime as if it were a +load method) takes care of dynamically adding the four subscript methods with class_addMethod. The implementations of these methods simply call the equivalent non-subscript methods. E.g. the objectAtIndexedSubscript: method is implemented like this:



id __arclite_objectAtIndexedSubscript(NSArray *self, SEL _cmd, NSUInteger idx)

{

    return [self objectAtIndex:idx];

}

mikeash - 2012-06-22 17:45:38

Fri, 22 Jun 2012 17:45:38 GMT

Eimantas: It's not possible to create your own literal syntax without hacking on the compiler, or possibly using the new C++11 stuff in conjunction with Objective-C++.

Goffredo Marocchi - 2012-06-22 16:26:06

Fri, 22 Jun 2012 16:26:06 GMT

Mike, your detailed article and Objective-C insights are a sight to behold. You do the language and Cocoa community a great service exploring "under the hood" topics. Thank you!

Eimantas - 2012-06-22 15:46:43

Fri, 22 Jun 2012 15:46:43 GMT

How about an article on how to create your own ObjC literal?

mikeash - 2012-06-22 14:48:45

Fri, 22 Jun 2012 14:48:45 GMT

The compiler generates objects of class __NSCFConstantString by default now, although that can be changed with a flag. But of course NSConstantString still has to live forever, since older binaries reference it.

Remy "Psy" Demarest - 2012-06-22 14:36:25

Fri, 22 Jun 2012 14:36:25 GMT

Are you sure NSConstantString is still used by the runtime for string literals?

Also, the C++ initializers are ran after the ObjC runtime +load methods so if you're using NSArray it will probably be properly initialized when you're trying to make a literal. Don't you think?