Next article: Rogue Amoeba Is Hiring Again
Previous article: Friday Q&A 2010-12-17: Custom Object Allocators in Objective-C
Tags: c fridayqna macros
The year is almost over, but there's time for one last Friday Q&A before 2011 comes around. For today's post, fellow Amoeba Dan Wineman suggested that I discuss tricks for writing macros in C.
Preprocessor vs Compiler
To properly understand C macros, you must understand how a C program is compiled. In particular, you must understand the different things that happen in the preprocessor and in the compiler.
The preprocessor runs first, as the name implies. It performs some simple textual manipulations, such as:
- Stripping comments.
- Resolving
#includedirectives and replacing them with the contents of the included file. - Evaluating
#ifand#ifdefdirectives. - Evaluating
#defines. - Expading the macros found in the rest of the code according to those
#defines.
Note that the preprocessor largely has no understanding of the text that it processes. There are some exceptions to this. For example, it knows that this is a string, and so does not expand the macro inside it:
#define SOMETHING hello
char *str = "SOMETHING, world!" // nope
#define ONEARG(x) NSLog x
ONEARG((@"hello, %@", @"world"));
#if to check whether a type is defined or not:
// makes no sense
#ifndef MyInteger
typedef int MyInteger
#endif
#ifndef always comes out true even if the MyInteger type is already defined. Type definitions are evaluated as part of the compilation phase, which hasn't even happened yet.
Likewise, there is no need for the contents of a #define to be syntactically correct on their own. It is completely legal, although a poor idea, to create macros like this:
#define STARTLOG NSLog(@
#define ENDLOG , @"testing");
STARTLOG "just %@" ENDLOG
STARTLOG and ENDLOG with their definitions. By the time the compiler comes along to try to make sense of this code, it actually does make sense, and so it compiles as valid code.
A Word of Warning
C macros are at the same time too powerful and not powerful enough. Their somewhat ad-hoc nature makes them dangerous, so treat them with care.
The C preprocessor is nearly Turing-complete. With a simple driver, you can compute any computable function using the preprocessor. However, the contortions required to do this are so bizarre and difficult that they make Turing-complete C++ templates look simple by comparison.
While powerful, they are also very simple. Since macro expansion is a simple textual process, there are pitfalls. For example, operator precedence can be dangerous:
#define ADD(x, y) x+y
// produces 14, not 20
ADD(2, 3) * 4;
#define MULT(x, y) x*y
// produces 14, not 20
MULT(2 + 3, 4);
Evaluating a macro argument multiple times can also lead to unexpected results:
#define MAX(x, y) ((x) > (y) ? (x) : (y))
int a = 0;
int b = 1;
int c = MAX(a++, b++);
// now a = 1, c = 1, and b = 3!
// (a++ > b++ ? a++ : b++)
// b++ gets evaluated twice
Macro Debugging
Macros are code, and like any code, they will have bugs. Macro bugs tend to manifest as weird compiler errors at the site where the macro is used. This can be incredibly confusing.
To reduce confusion, you'll want to look at the file as it appears after preprocessing. This means all of your macros are expanded, and you can see the raw C code that the compiler sees, rather than trying to expand the macro in your head. In Xcode you can do this by selecting Build->Preprocess. The resulting file will generally be very large due to all of the #include directives, but you'll find your code near the end. Find the site where the macro is used, figure out how the code has gone wrong, then modify your macro to make it right.
Multi-Statement Macros
It's common to write a macro that consists of multiple statements. For example, a timing macro:
#define TIME(name, lastTimeVariable) NSTimeInterval now = [[NSProcessInfo processInfo] systemUptime]; if(lastTimeVariable) NSLog(@"%s: %f seconds", name, now - lastTimeVariable); lastTimeVariable = now
- (void)calledALot
{
// do some work
// time it
TIME("calledALot", _calledALotLastTimeIvar);
}
#define is terminated at the end of the line, but by putting \ at the end of the line, you can make the preprocessor continue the definition on the next line:
#define TIME(name, lastTimeVariable) \
NSTimeInterval now = [[NSProcessInfo processInfo] systemUptime]; \
if(lastTimeVariable) \
NSLog(@"%s: %f seconds", name, now - lastTimeVariable); \
lastTimeVariable = now
- (void)calledALot
{
if(...) // only time some calls
TIME("calledALot", _calledALotLastTimeIvar);
}
- (void)calledALot
{
if(...) // only time some calls
NSTimeInterval now = [[NSProcessInfo processInfo] systemUptime];
if(_calledALotLastTimeIvar)
NSLog(@"%s: %f seconds", name, now - _calledALotLastTimeIvar);
_calledALotLastTimeIvar = now;
}
NSTimeInterval now in the if statement is illegal. Even if that worked, only the first statement is subject to the if, and the following lines would run regardless. Not what we wanted!
This can be solved by putting brackets around the macro definition:
#define TIME(name, lastTimeVariable) \
{ \
NSTimeInterval now = [[NSProcessInfo processInfo] systemUptime]; \
if(_calledALotLastTimeIvar) \
NSLog(@"%s: %f seconds", name, now - _calledALotLastTimeIvar); \
_calledALotLastTimeIvar = now; \
}
- (void)calledALot
{
if(...) // only time some calls
{
NSTimeInterval now = [[NSProcessInfo processInfo] systemUptime];
if(lastTimeVariable)
NSLog(@"%s: %f seconds", name, now - lastTimeVariable);
lastTimeVariable = now;
};
}
In fact, this is a problem. Consider this code:
- (void)calledALot
{
if(...) // only time some calls
TIME("calledALot", _calledALotLastTimeIvar);
else // otherwise do something else
// stuff
}
- (void)calledALot
{
if(...) // only time some calls
{
NSTimeInterval now = [[NSProcessInfo processInfo] systemUptime];
if(_calledALotLastTimeIvar)
NSLog(@"%s: %f seconds", name, now - _calledALotLastTimeIvar);
_calledALotLastTimeIvar = now;
};
else // otherwise do something else
// stuff
}
You could work around this by requiring the user of the macro not to put a semicolon at the end. However, this is highly unnatural and tends to mess with things like automatic code indenting.
A better way to fix it is to wrap the function in a do ... while(0) construct. This construct requires a semicolon at the end, which is exactly what we want. Using while(0) ensures that the loop never really loops, and its contents are only executed once.
#define TIME(name, lastTimeVariable) \
do { \
NSTimeInterval now = [[NSProcessInfo processInfo] systemUptime]; \
if(lastTimeVariable) \
NSLog(@"%s: %f seconds", name, now - lastTimeVariable); \
lastTimeVariable = now; \
} while(0)
if statement and in all other situations. A multi-statement macro should always be wrapped in do ... while(0) for this reason.
This macro defines a variable called now. This is a poor choice of names for a macro variable, because it could conflict with a variable from outside. Imagine the following code, with somewhat poor variable naming:
NSTimeInterval now; // ivar
TIME("whatever", now);
Unfortunately, C does not have a good way to generate unique variable names for this use. The best thing to do is to add a prefix, like you do with Objective-C class names:
#define TIME(name, lastTimeVariable) \
do { \
NSTimeInterval MA_now = [[NSProcessInfo processInfo] systemUptime]; \
if(lastTimeVariable) \
NSLog(@"%s: %f seconds", name, MA_now - lastTimeVariable); \
lastTimeVariable = MA_now; \
} while(0)
String Concatenation
This feature is not strictly part of macros, but it's useful for building macros, so it deserves mention. It's a little-known feature of C that if you put two string literals next to each other in the source code, they get concatenated:
char *helloworld = "hello, " "world!";
// equivalent to "hello, world!"
NSString *helloworld = @"hello, " @"world!";
NSString *helloworld = @"hello, " "world!";
#define COM_URL(domain) [NSURL URLWithString: @"http://www." domain ".com"];
COM_URL("google"); // gives http://www.google.com
COM_URL("apple"); // gives http://www.apple.com
By placing a
# in front of a parameter name, the preprocessor will turn the contents of that parameter into a C string. For example:
#define TEST(condition) \
do { \
if(!(condition)) \
NSLog(@"Failed test: %s", #condition); \
} while(0)
TEST(1 == 2);
// logs: Failed test: 1 == 2
#define WITHIN(x, y, delta) (fabs((x) - (y)) < delta)
TEST(WITHIN(1.1, 1.2, 0.05));
// logs: Failed test: WITHIN(1.1, 1.2, 0.05)
#define STRINGIFY(x) #x
#define TEST(condition) \
do { \
if(!(condition)) \
NSLog(@"Failed test: %s", STRINGIFY(condition)); \
} while(0)
TEST(WITHIN(1.1, 1.2, 0.05));
// logs: Failed test: (fabs(1.1 - 1.2) < 0.05)
Token Pasting
The preprocessor provides a ## operator to concatenate tokens together. This allows you to build multiple related items in a macro to eliminate redundancy. Writing a ## b produces the single token ab. If a or b is a macro parameter, its content will be used instead. A useless example:
#define NSify(x) NS ## x
NSify(String) *s; // gives NSString
NSMutableArray:
#define ARRAY_ACCESSORS(capsname, lowername) \
- (NSUInteger)countOf ## capsname { \
return [lowername count]; \
} \
\
- (id)objectIn ## capsname ## AtIndex: (NSUInteger)index { \
return [lowername objectAtIndex: index]; \
} \
\
- (void)insertObject: (id)obj in ## capsname ## AtIndex: (NSUInteger)index { \
[lowername insertObject: obj atIndex: index]; \
} \
\
- (void)removeObjectFrom ## capsname ## AtIndex: (NSUInteger)index { \
[lowername removeObjectAtIndex: index]; \
}
// instance variable
NSMutableArray *thingies;
// in @implementation
ARRAY_ACCESSORS(Thingies, thingies)
Like the stringify operator, the concatenation operator won't evaluate macros passed to it without an extra level of indirection:
#define ARRAY_NAME thingies
#define ARRAY_NAME_CAPS Thingies
// incorrectly creates accessors for "ARRAY_NAME_CAPS"
ARRAY_ACCESSORS(ARRAY_NAME_CAPS, ARRAY_NAME)
#define CONCAT(x, y) x ## y
// define ARRAY_ACCESSORS using CONCAT, and the above works
Variable Argument Lists
Imagine that you want to write a logging macro that only logs if a variable is set:
#define LOG(string) \
do { \
if(gLoggingEnabled) \
NSLog(@"Conditional log: %s", string); \
} while(0)
LOG("hello");
Conditional log: hello
NSLog takes a format string and variable arguments. It would be really useful if LOG could do the same:
LOG("count: %d name: %s", count, name);
If you place the magic ... parameter at the end of the macro's parameter list, the macro will accept a variable number of arguments. If you then use the magic __VA_ARGS__ identifier in the macro body, that will be replaced by all of the variable arguments, commas and all. Thus the LOG macro can be made to accept variable arguments like this:
#define LOG(...) \
do { \
if(gLoggingEnabled) \
NSLog(@"Conditional log: " __VA_ARGS__); \
} while(0)
#define LOG(fmt, ...) \
do { \
if(gLoggingEnabled) \
NSLog(@"Conditional log: --- " fmt " ---", __VA_ARGS__); \
} while(0)
LOG("hello"), the NSLog line expands to:
NSLog(@"Conditional log: --- " "hello" " ---", );
To avoid this problem in a completely portable way, you have to go back to taking one parameter, and do fancier tricks. For example, you might construct the user-provided string separately, then combine it into the log:
NSString *MA_logString = [NSString stringWithFormat: __VA_ARGS__]; \
NSLog(@"Conditional log: --- %@ ---", MA_logString);
## operator between the trailing comma and __VA_ARGS__, the preprocessor will eliminate the trailing comma in the case that no variable arguments are provided:
#define LOG(fmt, ...) \
do { \
if(gLoggingEnabled) \
NSLog(@"Conditional log: --- " fmt " ---", ## __VA_ARGS__); \
} while(0)
Magic Identifiers
C provides several built-in identifiers which can be extremely useful when building macros:
__LINE__: a built-in macro that expands to the current line number.__FILE__: another built-in macro that expands to a string literal containing the name of the current source file.__func__: this is an implicit variable which contains the name of the current function as a C string.
As an example, consider this logging macro:
#define LOG(fmt, ...) NSLog(fmt, ## __VA_ARGS__)
#define LOG(fmt, ...) NSLog(@"%s:%d (%s): " fmt, __FILE__, __LINE__, __func__, ## __VA_ARGS__)
LOG("something happened");
MyFile.m:42 (MyFunction): something happened
LOG statements throughout your code and the log output will automatically contain the file name, line number, and function name of where each log statement was placed.
Compound Literals
This is another item that's not really part of macros, but is really useful for building macros. Compound literals are a new feature in C99. They allow you to to create literals (that is, a constant expression of a given value, like 42 or "hello") of any type, not just built-in types.
The syntax for compound literals is a bit odd, but not hard. It looks like this:
(type){ initializer }
// regular variable and initializer
NSPoint p = { 1, 100 };
DoSomething(p);
// compound literal
DoSomething((NSPoint){ 1, 100 });
NSArray *array = [NSArray arrayWithObjects: (id []){ @"one", @"two", @"three" } count: 3];
#define ARRAY(num, ...) [NSArray arrayWithObjects: (id []){ __VA_ARGS__ } count: num]
NSArray *array = ARRAY(3, @"one", @"two", @"three");
As you probably know, the sizeof operator gives you the size of a particular object or type in bytes. When given an array, it reports the size of the entire array. By dividing this size by the size of a single element, you get the number of elements in the array:
#define ARRAY(...) [NSArray
arrayWithObjects: (id []){ __VA_ARGS__ }
count: sizeof((id []){ __VA_ARGS__ }) / sizeof(id)]
NSArray *array = ARRAY(@"one", @"two", @"three");
#define IDARRAY(...) (id []){ __VA_ARGS__ }
#define IDCOUNT(...) (sizeof(IDARRAY(__VA_ARGS__)) / sizeof(id))
#define ARRAY(...) [NSArray arrayWithObjects: IDARRAY(__VA_ARGS__) count: IDCOUNT(__VA_ARGS__)]
Let's make a similar one for dictionaries. NSDictionary doesn't have a method that exactly corresponds to what I want, so I'll make this macro call a small helper fuction that does some more work:
#define DICT(...) DictionaryWithIDArray(IDARRAY(__VA_ARGS__), IDCOUNT(__VA_ARGS__) / 2)
NSDictionary to create the dictionary:
NSDictionary *DictionaryWithIDArray(id *array, NSUInteger count)
{
id keys[count];
id objs[count];
for(NSUInteger i = 0; i < count; i++)
{
keys[i] = array[i * 2];
objs[i] = array[i * 2 + 1];
}
return [NSDictionary dictionaryWithObjects: objs forKeys: keys count: count];
}
NSDictionary *d = DICT(@"key", @"value", @"key2", @"value2");
typeofThis is a
gcc extension, not part of standard C, but it's extremely useful. It works like sizeof, except instead of providing the size, it provides the type. If you give it an expression, it evaluates to the type of that expression. If you give it a type, it just regurgitates that type.
Note that for maximum compatibility, it's best to write it as __typeof__. The plain typeof keyword is disabled in some gcc modes to avoid conflicts.
Let's take a look at that faulty MAX macro from the beginning of this article:
#define MAX(x, y) ((x) > (y) ? (x) : (y))
#define MAX(x, y) (^{ \
int my_localx = (x); \
int my_localy = (y); \
return my_localx > my_localy ? (my_localx) : (my_localy); \
}())
int, the macro doesn't work correctly for float, long long, or other types that don't quite fit.
Using __typeof__, this macro can be built to be completely generic:
#define MAX(x, y) (^{ \
__typeof__(x) my_localx = (x); \
__typeof__(y) my_localy = (y); \
return my_localx > my_localy ? (my_localx) : (my_localy); \
}())
__typeof__ is a purely compile-time construct, the extra use of the macro parameters does not cause them to be evaluated twice. You can use a similar trick to create a pointer to any value you want:
#define POINTERIZE(x) ((__typeof__(x) []){ x })
NSValue object:
#define BOX(x) [NSValue valueWithBytes: POINTERIZE(x) objCType: @encode(__typeof__(x))]
gcc provides two built-in functions which can be useful for building macros.
The first is __builtin_types_compatible_p. You pass two types to this function (__typeof__ comes in handy here) and it produces 1 if the two types are "compatible" (roughly, if they're equal) and 0 if they aren't.
The second is __builtin_choose_expr. This works like the C standard ?: operator, except that the predicate must be a compile-time constant, and the type of the entire __builtin_choose_expr expression is the type of whichever branch gets chosen; the two branches are not required to be similar types.
This allows you to write macros which do different things depending on the type of the argument. As an example, here's a macro which turns an expression into an NSString, and tries to make the output as useful as possible:
// make the compiler treat x as the given type no matter what
#define FORCETYPE(x, type) *(type *)(__typeof__(x) []){ x }
#define STRINGIFY(x) \
__builtin_choose_expr( \
__builtin_types_compatible_p(__typeof__(x), NSRect), \
NSStringFromRect(FORCETYPE(x, NSRect)), \
\
__builtin_choose_expr( \
__builtin_types_compatible_p(__typeof__(x), NSSize), \
NSStringFromSize(FORCETYPE(x, NSSize)), \
\
__builtin_choose_expr( \
__builtin_types_compatible_p(__typeof__(x), NSPoint), \
NSStringFromPoint(FORCETYPE(x, NSPoint)), \
\
__builtin_choose_expr( \
__builtin_types_compatible_p(__typeof__(x), SEL), \
NSStringFromSelector(FORCETYPE(x, SEL)), \
\
__builtin_choose_expr( \
__builtin_types_compatible_p(__typeof__(x), NSRange), \
NSStringFromRange(FORCETYPE(x, NSRange)), \
\
[NSValue valueWithBytes: (__typeof__(x) []){ x } objCType: @encode(__typeof__(x))] \
)))))
FORCETYPE macro. Even though the code branch to follow is chosen at compile time, unused branches still have to be valid code. The compiler won't accept NSStringFromRect(42) even though that branch will never be chosen. By pointerizing the value and then casting it before dereferencing it, it ensures that the code will compile. The cast is invalid for everything but the one branch that is taken, but it doesn't need to be valid for any of the others anyway.
X-Macros
This is something I've never used, but is interesting enough that it deserves mention. X-macros are a way of defining a macro in terms of another macro, which are then redefined multiple times to give that macro new meaning. This is confusing, so here's an example:
#define MY_ENUM \
MY_ENUM_MEMBER(kStop) \
MY_ENUM_MEMBER(kGo) \
MY_ENUM_MEMBER(kYield)
// create the actual enum
enum MyEnum {
#define MY_ENUM_MEMBER(x) x,
MY_ENUM
#undef MY_ENUM_MEMBER
};
// stringification
const char *MyEnumToString(enum MyEnum value)
{
#define MY_ENUM_MEMBER(x) if(value == (x)) return #x;
MY_ENUM
#undef MY_ENUM_MEMBER
}
// destringification
enum MyEnum MyEnumFromString(const char *str)
{
#define MY_ENUM_MEMBER(x) if(strcmp(str, #x) == 0) return x;
MY_ENUM
#undef MY_ENUM_MEMBER
// default value
return -1;
}
Conclusion
C macros are complicated and powerful. If you use them, you must be extremely careful not to abuse them. However, in some situations they can be incredibly useful, and, when used correctly, these tips and tricks can help you create macros which make your code easier to write and easier to read.
That's it for 2010. Come back next year (in two weeks) for the next Friday Q&A. As always, if you have a topic that you would like to see covered here, send it in!
Comments:
You could've used it for MAX and avoided the (less portable!) block, and improved your generated code at the same time.
OSC: I debated whether to include the
({}) construct, and ultimately decided not to. While portability for that construct is indeed better at the moment, it's incredibly special-purpose, not good for much else than writing macros. Semantically, the block technique does the exact same thing using a more general-purpose (and, I hope, longer-lived) construct. The poor generated code is unfortunate, but I hope that the compilers will eventually learn to optimize out such direct block invocations.NSTimeInterval lastTimeVariable##_now = ...({}) construct rather useful even outside of macros, generally when I need a throwaway variable that I want to scope nicely (I use this often when tweaking the .frame property of views).
I'm surprised you made no mention of
__PRETTY_FUNCTION__ when talking about the similar identifiers. It's especially helpful for Obj-C.
#define MAX(x, y) (^{ \
__typeof__(x) my_localx = (x); \
__typeof__(y) my_localy = (y); \
return my_localx > my_localy ? my_localx : my_localy; \
}())
Also, I prefer to use another GCC extension (statements as expressions), which avoids using blocks:
#define MAX(x, y) ({ \
__typeof__(x) my_localx = (x); \
__typeof__(y) my_localy = (y); \
my_localx > my_localy ? my_localx : my_localy; \
})
Kevin Ballard: Why can't you just use a completely standard block to get throwaway variables?
Chris Suter: Thanks for the MAX tip, how embarrassing. Fixed now. For statement expressions, I discussed my reasons for preferring blocks in my previous comment, but it's not a particularly strong preference.
You can define a name for the '…' argument. For example, if you want to use 'args' instead of __VA_ARGS__, you can define your macro like this:
#define IDARRAY(args...) (id []){ args }
And something you should keep in mind when you want to write macro: wherever you can use an inline function instead of a macro, choose the former. There are as fast as a macro (http://gcc.gnu.org/onlinedocs/gcc/Inline.html), but don't have all the pitfalls described in this article, and as they provide more informations to the compiler (type of argument), they are easier to 'debug' too.
I've used this exactly once, when I had code that took an array of booleans and (among other things) called a function on it and then deinterlaced the elements, and I wanted to reuse that code to process floats instead, with the function varying only by name (what could make me switch from bools to floats? Simple: going from hard decoding to soft decoding in a channel coding simulation). How to avoid duplicating that code (I wanted to have both available at runtime)? You can't have a function that can manipulate either bools or floats, the only way I found was to use a mega-macro.
The builtin macros such as __LINE__ and __FILE__ are useful for building assert statements.
Of course, I wouldn't try to discourage someone else from using macros (unless I was going to be working with them).
objs[i] = array[i * 2];
keys[i] = array[i * 2 + 1];
... or at least if we are going with the language "dictionary with objects and keys", hence objects are listed before keys.
object, key ordering to be broken. It's backwards from how other modern languages do it and IMO confusing as a result. Switching the order for this macro was deliberate. Of course, if you prefer it the other way around, it's an easy change as you note.int a = 0;
int b = 1;
int c = MAX(a++, b++);
// now a = 1, c = 1, and b = 3!
// (a++ > b++ ? a++ : b++)
// b++ gets evaluated twice
I think the value of c is 2, and I have verified in my computer(fedora 14,i686)
For simple multi-statement macros I just use the comma operator instead of do {} while(0) - probably for no good reason other than it's shorter. Eg:
#define DO_TWO_THINGS do_one_thing(), do_the_other_thing
That works in most cases I think - have I missed anything obvious?
#ifndef DebugLog_h
#define DebugLog_h
#if DEBUG
#define _FILE_ (strrchr(__FILE__, '/') ? strrchr(__FILE__, '/') + 1 : __FILE__)
#define DLog(...) do{\
printf("[%s:%d]", _FILE_, __LINE__);\
NSString *_S_ = [NSString stringWithFormat:__VA_ARGS__];\
printf("%s\n",[_S_ cStringUsingEncoding:NSUTF8StringEncoding]);\
}while(0);
#else
#define DLog(...)
#endif
#endif
this will shorten the file name to just the files name and not the path included and will get rid of all the NSLog prefix stuff
I'm not quite sure how I could do that using
#if __DATE__ > FIXBYDATE...
I think it could be useful. Any pointers?
Comments RSS feed for this page
Add your thoughts, post a comment:
Spam and off-topic posts will be deleted without notice. Culprits may be publicly humiliated at my sole discretion.
I’ve been seeing X-macros a lot recently while trying to get SpiderMonkey building in a sane (by OS X standards) way. They’re used for obvious things like keywords and interpreter opcodes, but also for error message tables.