Reprint iOS Engineering adaptation 64-bit experience sharing

After all, still come. Apple has issued an ultimatum to support the 64-bit:

As we announced in October, beginning February 1, new IOS apps submitted to the APP Store must include 64-bit support and is built with the IOS 8 SDK. Beginning June 1, app updates would also need to follow the same requirements.

The adaptation that should have been done has finally begun, and the 64-bit CPU has run a 32-bit program for so long. The previous period of time the company project completed the 64-bit package, this is not so complicated by the numerous non-standard old code stirring not light, summed up a few tip mutual encouragement.

Tips

Reject basic data types and implicit conversions

The first is the basic type, such as the following 4 types in the 32-bit and 64-bit, respectively, how long?

1234	size_t s1 = sizeof(int);size_t s2 = sizeof(long);size_t s3 = sizeof(float);size_t s4 = sizeof(double);

32-bit under: 4, 4, 4, 8;64-bit: 4, 8, 4, 8

(PS: This result with the compiler, for other platforms may not necessarily)

Their length changes may not be our expectation of doubling the length of the 64-bit, so the code for sizeof appears in the program to see two more eyes. And, unless you know exactly what you're doing, you should use the following type instead of the base type:

• Nsinteger int

• Unsigned-Nsuinteger

• Float-CGFloat

• Animation Time-Nstimeinterval

• ...

These are the types defined in the SDK, and most of the time we are dealing with the SDK APIs, which can be used to reduce the impact of type conversions a lot.

Let's say the following code:

1234	NSArray *items = @[@1, @2, @3];for(int i = -1; i < items.count; i++) {    NSLog(@"%d", i);}

As a result, the For loop does not go in at once.

The count of the array is of type Nsuinteger, and 1 is implicitly converted to Nsuinteger when compared to a large number:

1234	(lldb) p i(int) $0 = -1(lldb) p (NSUInteger)i(NSUInteger) $1 = 18446744073709551615

This has nothing to do with 64-bit, to illustrate, this implicit conversion also needs to be careful, be aware of all the operations associated with this variable (assignment, comparison, conversion)

The old-fashioned for loop can be considered written as:

1	for(NSUInteger index = 0; index < items.count; index++) {}

Of course, array traversal is more recommended with for-in or block versions, and comparisons between them can be reviewed in this article.

Using the new enumeration

Similar to the above, enumerations should use the new version of the notation:

123456

typedef NS_ENUM(NSInteger, UIViewAnimationCurve) {    UIViewAnimationCurveEaseInOut,    UIViewAnimationCurveEaseIn,    UIViewAnimationCurveEaseOut,    UIViewAnimationCurveLinear};

Not only can you specify a type for the enumeration value, but the compiler will also give a warning when assigning the wrong type, and there is no reason to use this notation.

Replace format string

When adapting to the 64-bit, did you encounter the following disgusting notation:

12	NSArray *items = @[@1, @2, @3];NSLog(@"数组元素个数：%lu", (unsigned long)items.count);

In general, use NSNumber's @ syntax sugar to solve:

12	NSArray *items = @[@1, @2, @3];NSLog(@"数组元素个数：%@", @(items.count));

Similarly, int to string can also:

12	NSInteger i = 10086;NSString *string = @(i).stringValue;

Of course, this format does not apply if you need to%.2f.

BOOL under the 64-bit

32-bit, BOOL is defined as signed char, and the result of @encode (BOOL) is ' C '

64-bit, BOOL is defined as BOOL, @encode (BOOL) result is ' B '

The more intuitive explanation is:

1234	(lldb) p/t (signed char)7(BOOL) $0 = 0b00000111 (YES)(lldb) p/t (bool)7(bool) $1 = 0b00000001 (YES)

The 32-bit version of BOOL includes the possibility of 256 values and also causes some pits, as this article says. and 64-bit only 0 (NO), 1 (YES) two possible, finally to bool name.

Do not take the ISA pointer directly

The compiler has disabled this use by default, the ISA pointer is the address of class under 32 bits, but with bits mask under 64 bits to remove the real address, if you really need to, use the runtime's Object_getclass and Object_setclass methods. For a 64-bit explanation of ISA, you can read this article

Resolve third-party Lib dependencies and Lipo commands

In the form of the source code in the project of the third-party LIB, as long as the target plus arm64 compile the good.

The disgusting thing is that the static libraries (. A) or the framework, which are dragged directly into the project, need to be re-searched for the package that supports 64-bit. This time can see which is already unmanned maintenance of LIB, it is a substitute for a replacement (for example, I can not find the whole network of the project used in an audio library of the 64-bit package, finally found on a buddy's GitHub, crying to give a star--)

Print supported schemas for mach-o files

How to see if an executable file supports 64-bit?

Use the Lipo-info command, for example, to see Uikit supported schemas:

123	// 当前在Xcode Frameworks目录sunnyxx$ lipo -info UIKit.framework/UIKitArchitectures inthe fat file: UIKit.framework/UIKit are: arm64 armv7s

To see more detailed information you can use Lipo-detailed_info:

12345678910111213141516

sunnyxx$ lipo -detailed_info UIKit.framework/UIKit Fat header in: UIKit.framework/UIKitfat_magic 0xcafebabenfat_arch 2architecture arm64    cputype CPU_TYPE_ARM64    cpusubtype CPU_SUBTYPE_ARM64_ALL    offset 4096    size 16822272    align 2^12 (4096)architecture armv7s    cputype CPU_TYPE_ARM    cpusubtype CPU_SUBTYPE_ARM_V7S    offset 16826368    size 14499840    align 2^12 (4096)

Of course, you can also use the file command:

1234	sunnyxx$ file UIKit.framework/UIKit UIKit.framework/UIKit: Mach-O universal binary with2 architecturesUIKit.framework/UIKit (for architecture arm64):Mach-O 64-bit dynamically linked shared libraryUIKit.framework/UIKit (forarchitecture armv7s):Mach-O dynamically linked shared library arm

The above command applies to the Mach-o file, the static library. A file, the. A file in the framework, and the executable file of your app can be printed down to see.

Merging packages with multiple schemas

If we have mylib-32.a and mylib-64.a, you can use the lipo-create command to merge:

1	sunnyxx$ lipo -create MyLib-32.a MyLib-64.a -output MyLib.a

Will the package become larger after 64-bit support?

Will, support 64-bit, more than a arm64 architecture, in theory, each architecture a set of instructions, but compared to the original will be much more difficult to say, we have increased by about 50%, and heard that will increase by one times.

A lib contains a lot of architecture, will it hit the last bag?

No, if Lib has armv7, armv7s, arm64, i386 architecture, and target architecture chooses armv7s, arm64, then only the binary code of the two schemas specified by link from Lib, The code in the other schemas does not link to the final executable; in turn, a Lib needs to link normally in the emulator environment, as well as to include instructions for the i386 architecture.

Checklist

Finally, make a list of the points of note in the official documentation:

• Do not turn the pointer strongly into an integer

• Unified data types are used throughout the program

• It is important to pay attention to the operation of different types of integers.

• When you need a fixed-length variable, use a fixed-length type such as int32_t, int64_t

• Do not write dead size when using malloc

• Use a formatted string that can fit two schemas at the same time

• Note Function and function pointers (type conversions and mutable parameters)

• Do not directly access the OBJECTIVE-C pointer (ISA)

• Using the built-in synchronization primitives (primitives)

• Do not hardcode the virtual memory page size

• Go Position Independent

Reprint iOS Engineering adaptation 64-bit experience sharing