I have code similar to
#define LIST_OF_VARIABLES \
X(value1) \
X(value2) \
X(value3)
as explained in https://en.wikipedia.org/wiki/X_Macro
Now I have the need to make the LIST_OF_VARIABLES configurable at compile time
So it could effectively be e.g.
#define LIST_OF_VARIABLES \
X(default_value1) \
X(cust_value2) \
X(default_value3)
or e.g.
#define LIST_OF_VARIABLES \
X(default_value1) \
X(default_value2) \
X(cust_value3)
depending on some macros previously defined. The LIST_OF_VARIABLES is long and the customizations are relatively small. I would not like to copy the long list for each customization, because that will cause maintenance issues (the DRY principle https://en.wikipedia.org/wiki/Don%27t_repeat_yourself). As a matter of fact the LIST_OF_VARIABLES should be in one file and the
customizations elsewhere (either another file or just -D options in the Makefile)
In pseudo-code I was thinking of something like
#define X(arg) \
#ifdef CUST_##arg \
Y(CUST_##arg) \
#else \
Y(DEFAULT_##arg) \
#endif
And then use the X-macros under the name Y.
But of course that does not work, because a macro cannot contain preprocessor
directives.
What would be a way to achieve this? C is a must (no templates or Boost
macros), gcc specific solutions are acceptable.
I think that what you have to do is along the lines of:
#ifdef USE_DEFAULT_VALUE1
#define X_DEFAULT_VALUE1 X(default_value1)
#else
#define X_DEFAULT_VALUE1 /* omitted */
#endif
#ifdef USE_DEFAULT_VALUE2
#define X_DEFAULT_VALUE2 X(default_value2)
#else
#define X_DEFAULT_VALUE2 /* omitted */
#endif
#ifdef USE_DEFAULT_VALUE3
#define X_DEFAULT_VALUE3 X(default_value3)
#else
#define X_DEFAULT_VALUE3 /* omitted */
#endif
#ifdef USE_CUST_VALUE1
#define X_CUST_VALUE1 X(cust_value1)
#else
#define X_CUST_VALUE1 /* omitted */
#endif
#ifdef USE_CUST_VALUE2
#define X_CUST_VALUE2 X(cust_value2)
#else
#define X_CUST_VALUE2 /* omitted */
#endif
#define LIST_OF_VARIABLES \
X_DEFAULT_VALUE1 \
X_DEFAULT_VALUE2 \
X_DEFAULT_VALUE3 \
X_CUST_VALUE1 \
X_CUST_VALUE2 \
You then need to define USE_DEFAULT_VALUE1 etc as required for the specific configuration you are after.
As long as you always need the items in the same order, this is sufficient. If you need them in different orders, then you conditionally define LIST_OF_VARIABLES in the different sequences.
Answering myself.
With help of the comments I came up with a solution that works and meets most
requirements I had mentioned
With the "main code"
$cat main.c
#ifndef VALUE1
#define VALUE1 value1
#endif
#ifndef VALUE2
#define VALUE2 value2
#endif
#ifndef VALUE3
#define VALUE3 value3
#endif
#define LIST_OF_VARIABLES \
X(VALUE1) \
X(VALUE2) \
X(VALUE3)
and a customization file like
$cat cust1
-DVALUE2=value2cust
the code can be compiled using (GNUmake pseudo syntax)
$(CC) $(CFLAGS) $(shell cat cust1) main.c
Actually having the extra indirection with every value defined on a single
line is good, because it allows commenting the values. That would not have
been possible with the continuation lines in the single LIST_OF_VARIABLES macro.
Edit: Not true. A COMMENT(foo) macro expanding to nothing would have solved that issue, too. (Credit: Got the idea from the answer posted by #Jonathan Leffer.)
However the approach does not yet meet the following requirements I hadn't mentioned
no ugly boilerplate code (all these #ifndef lines are not really nice)
customization should also make it possible to drop default values from the
list altogether or add completely new values (yes, this could probably be
done with some ugly dummy code already now)
So not really satisfied yet with my own answer. Need to think about the
approach from the Dr. Dobbs article a bit more, maybe that can be used.
Open for better answers.
Given further context, it appears you want to be able to cherry pick individual values from your list at compile time. I think you might be interested in a preprocessor switch, which can accomplish what you're using preprocessor conditionals for with a lot less boilerplate.
Generic preprocessor switch
Here's a brief framework:
#define GLUEI(A,B) A##B
#define GLUE(A,B) GLUEI(A,B)
#define SECONDI(A,B,...) B
#define SECOND(...) SECONDI(__VA_ARGS__,,)
#define SWITCH(NAME_, PATTERN_, DEFAULT_) SECOND(GLUE(NAME_,PATTERN_), DEFAULT_)
SWITCH macro usage
Invoke SWITCH(MY_PREFIX_,SPECIFIC_IDENTIFIER,DEFAULT_VALUE) to expand everything that is not a matching pattern to DEFAULT_VALUE. Things that are a matching pattern can expand to whatever you map them to.
To create a matching pattern, define an object like macro called MY_PREFIX_SPECIFIC_IDENTIFIER, whose replacement list consists of a single comma followed by the value you want the SWITCH to expand to in this case.
The magic here is simply that SWITCH builds a hidden token, giving it a chance to expand (well, in this implementation SECOND's indirection is also significant), and inject a new second argument to SECOND if it's defined. Nominally this new token isn't defined; in such cases, it simply becomes the first argument to SECOND, which just discards it, never to be seen again.
For example, given the above macros:
#define CONTRACT_IDENTIFIER_FOR_DEFAULT , overridden_id_for_default
#define CONTRACT_IDENTIFIER_FOR_SIGNED , overridden_id_for_signed
SWITCH(CONTRACT_IDENTIFIER_FOR_, DRAFT , draft )
SWITCH(CONTRACT_IDENTIFIER_FOR_, DRAWN , drawn )
SWITCH(CONTRACT_IDENTIFIER_FOR_, PROOFED , proofed )
SWITCH(CONTRACT_IDENTIFIER_FOR_, DELIVERED , delivered )
SWITCH(CONTRACT_IDENTIFIER_FOR_, SIGNED , signed )
SWITCH(CONTRACT_IDENTIFIER_FOR_, FULFILLED , fulfilled )
SWITCH(CONTRACT_IDENTIFIER_FOR_, DEFAULT , default )
...will expand to:
draft
drawn
proofed
delivered
overridden_id_for_signed
fulfilled
overridden_id_for_default
Decorated X Macros
Assuming you wish to give your values names, and simply replace cherry picked values from the command line, you can make use of SWITCH to do something like this:
#define VARVALUE(N_,V_) SWITCH(VALUE_FOR_, N_, V_)
#define LIST_OF_VARIABLES \
X(VARVALUE(value1, default_value1)) \
X(VARVALUE(value2, default_value2)) \
X(VARVALUE(value3, default_value3))
The VARVALUE macros will be applied first in this form. To override a specific value, you can define your pattern matcher using either a #define:
#define VALUE_FOR_value2 , custom_value2
...or on the command line/makefile:
CFLAGS += -DVALUE_FOR_value2=,custom_value2
Disable/insertion using switch macro
To support disabling individual items safely, nest two switches and add an EAT macro to catch the entry:
#define EAT(...)
#define SELECT_ITEM_MACRO_FOR_STATE_ON , X
#define X_IF_ENABLED(N_, V_) \
SWITCH(SELECT_ITEM_MACRO_FOR_STATE_, SWITCH(ENABLE_VALUE_, N_, ON), EAT) \
(SWITCH(VALUE_FOR_, N_, V_))
#define LIST_OF_VARIABLES \
X_IF_ENABLED(value1, default_value1) \
X_IF_ENABLED(value2, default_value2) \
X_IF_ENABLED(value3, default_value3)
Just as before, individual macros can be overridden using VALUE_FOR_valuex pattern macros, but this also allows disabling items using ENABLE_VALUE_valuex macros, which can be set to anything but ,ON to disable that item.
Similarly, one way to add support for inserting values is to flip the idea:
#define ADD_ITEM_MACRO_FOR_STATE_EAT , EAT
#define X_IF_ADDED(N_) \
SWITCH(ADD_ITEM_MACRO_FOR_STATE_, SWITCH(VALUE_FOR_, N_, EAT), X) \
(SECOND(GLUE(VALUE_FOR_,N_)))
#define LIST_OF_VARIABLES \
X_IF_ENABLED(value1, default_value1) \
X_IF_ENABLED(value2, default_value2) \
X_IF_ENABLED(value3, default_value3) \
X_IF_ADDED(value4) \
X_IF_ADDED(value5) \
X_IF_ADDED(value6)
...this allows you to define VALUE_FOR_value4 as a a pattern macro, but by default will expand to nothing.
Summary
The framework supporting setting, removing, or inserting values winds up being:
#define GLUEI(A,B) A##B
#define GLUE(A,B) GLUEI(A,B)
#define SECONDI(A,B,...) B
#define SECOND(...) SECONDI(__VA_ARGS__,,)
#define SWITCH(NAME_, PATTERN_, DEFAULT_) SECOND(GLUE(NAME_,PATTERN_), DEFAULT_)
#define EAT(...)
#define SELECT_ITEM_MACRO_FOR_STATE_ON , X
#define X_IF_ENABLED(N_, V_) \
SWITCH(SELECT_ITEM_MACRO_FOR_STATE_, SWITCH(ENABLE_VALUE_, N_, ON), EAT) \
(SWITCH(VALUE_FOR_, N_, V_))
#define ADD_ITEM_MACRO_FOR_STATE_EAT , EAT
#define X_IF_ADDED(N_) \
SWITCH(ADD_ITEM_MACRO_FOR_STATE_, SWITCH(VALUE_FOR_, N_, EAT), X) \
(SECOND(GLUE(VALUE_FOR_,N_)))
Given this framework, your list macro would be comprised of a series of X(value), X_IF_ENABLED(name,default_value), and/or X_IF_ADDED(name) values, where:
X(value) can be used to always insert a call to the X macro with value
X_IF_ENABLED(name,default_value) will call X with default_value, allowing you to override the default based on name.
X_IF_ADDED(name) will provide an "empty slot" with name, which will do nothing unless you override that slot.
Overriding slots is done by defining VALUE_FOR_name to expand to ,replacement. Disabling enabled slots is done by defining ENABLE_VALUE_name to expand to ,OFF.
Demo showing change, removal, addition using command line
I am asking this question from an educational/hacking point of view, (I wouldn't really want to code like this).
Is it possible to implement a while loop only using C preprocessor directives. I understand that macros cannot be expanded recursively, so how would this be accomplished?
If you want to implement a while loop, you will need to use recursion in the preprocessor. The easiest way to do recursion is to use a deferred expression. A deferred expression is an expression that requires more scans to fully expand:
#define EMPTY()
#define DEFER(id) id EMPTY()
#define OBSTRUCT(id) id DEFER(EMPTY)()
#define EXPAND(...) __VA_ARGS__
#define A() 123
A() // Expands to 123
DEFER(A)() // Expands to A () because it requires one more scan to fully expand
EXPAND(DEFER(A)()) // Expands to 123, because the EXPAND macro forces another scan
Why is this important? Well when a macro is scanned and expanding, it creates a disabling context. This disabling context will cause a token, that refers to the currently expanding macro, to be painted blue. Thus, once its painted blue, the macro will no longer expand. This is why macros don't expand recursively. However, a disabling context only exists during one scan, so by deferring an expansion we can prevent our macros from becoming painted blue. We will just need to apply more scans to the expression. We can do that using this EVAL macro:
#define EVAL(...) EVAL1(EVAL1(EVAL1(__VA_ARGS__)))
#define EVAL1(...) EVAL2(EVAL2(EVAL2(__VA_ARGS__)))
#define EVAL2(...) EVAL3(EVAL3(EVAL3(__VA_ARGS__)))
#define EVAL3(...) EVAL4(EVAL4(EVAL4(__VA_ARGS__)))
#define EVAL4(...) EVAL5(EVAL5(EVAL5(__VA_ARGS__)))
#define EVAL5(...) __VA_ARGS__
Next, we define some operators for doing some logic(such as if, etc):
#define CAT(a, ...) PRIMITIVE_CAT(a, __VA_ARGS__)
#define PRIMITIVE_CAT(a, ...) a ## __VA_ARGS__
#define CHECK_N(x, n, ...) n
#define CHECK(...) CHECK_N(__VA_ARGS__, 0,)
#define NOT(x) CHECK(PRIMITIVE_CAT(NOT_, x))
#define NOT_0 ~, 1,
#define COMPL(b) PRIMITIVE_CAT(COMPL_, b)
#define COMPL_0 1
#define COMPL_1 0
#define BOOL(x) COMPL(NOT(x))
#define IIF(c) PRIMITIVE_CAT(IIF_, c)
#define IIF_0(t, ...) __VA_ARGS__
#define IIF_1(t, ...) t
#define IF(c) IIF(BOOL(c))
Now with all these macros we can write a recursive WHILE macro. We use a WHILE_INDIRECT macro to refer back to itself recursively. This prevents the macro from being painted blue, since it will expand on a different scan(and using a different disabling context). The WHILE macro takes a predicate macro, an operator macro, and a state(which is the variadic arguments). It keeps applying this operator macro to the state until the predicate macro returns false(which is 0).
#define WHILE(pred, op, ...) \
IF(pred(__VA_ARGS__)) \
( \
OBSTRUCT(WHILE_INDIRECT) () \
( \
pred, op, op(__VA_ARGS__) \
), \
__VA_ARGS__ \
)
#define WHILE_INDIRECT() WHILE
For demonstration purposes, we are just going to create a predicate that checks when number of arguments are 1:
#define NARGS_SEQ(_1,_2,_3,_4,_5,_6,_7,_8,N,...) N
#define NARGS(...) NARGS_SEQ(__VA_ARGS__, 8, 7, 6, 5, 4, 3, 2, 1)
#define IS_1(x) CHECK(PRIMITIVE_CAT(IS_1_, x))
#define IS_1_1 ~, 1,
#define PRED(x, ...) COMPL(IS_1(NARGS(__VA_ARGS__)))
Next we create an operator, which we will just concat two tokens. We also create a final operator(called M) that will process the final output:
#define OP(x, y, ...) CAT(x, y), __VA_ARGS__
#define M(...) CAT(__VA_ARGS__)
Then using the WHILE macro:
M(EVAL(WHILE(PRED, OP, x, y, z))) //Expands to xyz
Of course, any kind of predicate or operator can be passed to it.
Take a look at the Boost preprocessor library, which allows you to write loops in the preprocessor, and much more.
You use recursive include files. Unfortunately, you can't iterate the loop more than the maximum depth that the preprocessor allows.
It turns out that C++ templates are Turing Complete and can be used in similar ways. Check out Generative Programming
I use meta-template programming for this purpose, its fun once you get a hang of it. And very useful at times when used with discretion. Because as mentioned its turing complete, to the point where you can even cause the compiler to get into an infinite loop, or stack-overflow! There is nothing like going to get some coffee just to find your compilation is using up 30+ gigabytes of memory and all the CPU to compile your infinite loop code!
well, not that it's a while loop, but a counter loop, nonetheless the loop is possible in clean CPP (no templates and no C++)
#ifdef pad_always
#define pad(p,f) p##0
#else
#define pad0(p,not_used) p
#define pad1(p,not_used) p##0
#define pad(p,f) pad##f(p,)
#endif
// f - padding flag
// p - prefix so far
// a,b,c - digits
// x - action to invoke
#define n0(p,x)
#define n1(p,x) x(p##1)
#define n2(p,x) n1(p,x) x(p##2)
#define n3(p,x) n2(p,x) x(p##3)
#define n4(p,x) n3(p,x) x(p##4)
#define n5(p,x) n4(p,x) x(p##5)
#define n6(p,x) n5(p,x) x(p##6)
#define n7(p,x) n6(p,x) x(p##7)
#define n8(p,x) n7(p,x) x(p##8)
#define n9(p,x) n8(p,x) x(p##9)
#define n00(f,p,a,x) n##a(pad(p,f),x)
#define n10(f,p,a,x) n00(f,p,9,x) x(p##10) n##a(p##1,x)
#define n20(f,p,a,x) n10(f,p,9,x) x(p##20) n##a(p##2,x)
#define n30(f,p,a,x) n20(f,p,9,x) x(p##30) n##a(p##3,x)
#define n40(f,p,a,x) n30(f,p,9,x) x(p##40) n##a(p##4,x)
#define n50(f,p,a,x) n40(f,p,9,x) x(p##50) n##a(p##5,x)
#define n60(f,p,a,x) n50(f,p,9,x) x(p##60) n##a(p##6,x)
#define n70(f,p,a,x) n60(f,p,9,x) x(p##70) n##a(p##7,x)
#define n80(f,p,a,x) n70(f,p,9,x) x(p##80) n##a(p##8,x)
#define n90(f,p,a,x) n80(f,p,9,x) x(p##90) n##a(p##9,x)
#define n000(f,p,a,b,x) n##a##0(f,pad(p,f),b,x)
#define n100(f,p,a,b,x) n000(f,p,9,9,x) x(p##100) n##a##0(1,p##1,b,x)
#define n200(f,p,a,b,x) n100(f,p,9,9,x) x(p##200) n##a##0(1,p##2,b,x)
#define n300(f,p,a,b,x) n200(f,p,9,9,x) x(p##300) n##a##0(1,p##3,b,x)
#define n400(f,p,a,b,x) n300(f,p,9,9,x) x(p##400) n##a##0(1,p##4,b,x)
#define n500(f,p,a,b,x) n400(f,p,9,9,x) x(p##500) n##a##0(1,p##5,b,x)
#define n600(f,p,a,b,x) n500(f,p,9,9,x) x(p##600) n##a##0(1,p##6,b,x)
#define n700(f,p,a,b,x) n600(f,p,9,9,x) x(p##700) n##a##0(1,p##7,b,x)
#define n800(f,p,a,b,x) n700(f,p,9,9,x) x(p##800) n##a##0(1,p##8,b,x)
#define n900(f,p,a,b,x) n800(f,p,9,9,x) x(p##900) n##a##0(1,p##9,b,x)
#define n0000(f,p,a,b,c,x) n##a##00(f,pad(p,f),b,c,x)
#define n1000(f,p,a,b,c,x) n0000(f,p,9,9,9,x) x(p##1000) n##a##00(1,p##1,b,c,x)
#define n2000(f,p,a,b,c,x) n1000(f,p,9,9,9,x) x(p##2000) n##a##00(1,p##2,b,c,x)
#define n3000(f,p,a,b,c,x) n2000(f,p,9,9,9,x) x(p##3000) n##a##00(1,p##3,b,c,x)
#define n4000(f,p,a,b,c,x) n3000(f,p,9,9,9,x) x(p##4000) n##a##00(1,p##4,b,c,x)
#define n5000(f,p,a,b,c,x) n4000(f,p,9,9,9,x) x(p##5000) n##a##00(1,p##5,b,c,x)
#define n6000(f,p,a,b,c,x) n5000(f,p,9,9,9,x) x(p##6000) n##a##00(1,p##6,b,c,x)
#define n7000(f,p,a,b,c,x) n6000(f,p,9,9,9,x) x(p##7000) n##a##00(1,p##7,b,c,x)
#define n8000(f,p,a,b,c,x) n7000(f,p,9,9,9,x) x(p##8000) n##a##00(1,p##8,b,c,x)
#define n9000(f,p,a,b,c,x) n8000(f,p,9,9,9,x) x(p##9000) n##a##00(1,p##9,b,c,x)
#define n00000(f,p,a,b,c,d,x) n##a##000(f,pad(p,f),b,c,d,x)
#define n10000(f,p,a,b,c,d,x) n00000(f,p,9,9,9,9,x) x(p##10000) n##a##000(1,p##1,b,c,d,x)
#define n20000(f,p,a,b,c,d,x) n10000(f,p,9,9,9,9,x) x(p##20000) n##a##000(1,p##2,b,c,d,x)
#define n30000(f,p,a,b,c,d,x) n20000(f,p,9,9,9,9,x) x(p##30000) n##a##000(1,p##3,b,c,d,x)
#define n40000(f,p,a,b,c,d,x) n30000(f,p,9,9,9,9,x) x(p##40000) n##a##000(1,p##4,b,c,d,x)
#define n50000(f,p,a,b,c,d,x) n40000(f,p,9,9,9,9,x) x(p##50000) n##a##000(1,p##5,b,c,d,x)
#define n60000(f,p,a,b,c,d,x) n50000(f,p,9,9,9,9,x) x(p##60000) n##a##000(1,p##6,b,c,d,x)
#define n70000(f,p,a,b,c,d,x) n60000(f,p,9,9,9,9,x) x(p##70000) n##a##000(1,p##7,b,c,d,x)
#define n80000(f,p,a,b,c,d,x) n70000(f,p,9,9,9,9,x) x(p##80000) n##a##000(1,p##8,b,c,d,x)
#define n90000(f,p,a,b,c,d,x) n80000(f,p,9,9,9,9,x) x(p##90000) n##a##000(1,p##9,b,c,d,x)
#define cycle5(c1,c2,c3,c4,c5,x) n##c1##0000(0,,c2,c3,c4,c5,x)
#define cycle4(c1,c2,c3,c4,x) n##c1##000(0,,c2,c3,c4,x)
#define cycle3(c1,c2,c3,x) n##c1##00(0,,c2,c3,x)
#define cycle2(c1,c2,x) n##c1##0(0,,c2,x)
#define cycle1(c1,x) n##c1(,x)
#define concat(a,b,c) a##b##c
#define ck(arg) a[concat(,arg,-1)]++;
#define SIZEOF(x) (sizeof(x) / sizeof((x)[0]))
void check5(void)
{
int i, a[32769];
for (i = 0; i < SIZEOF(a); i++) a[i]=0;
cycle5(3,2,7,6,9,ck);
for (i = 0; i < SIZEOF(a); i++) if (a[i] != 1) printf("5: [%d] = %d\n", i+1, a[i]);
}
Here's an abuse of the rules that would get it done legally. Write your own C preprocessor. Make it interpret some #pragma directives the way you want.
I found this scheme useful when the compiler got cranky and wouldn't unroll certain loops for me
#define REPEAT20(x) { x;x;x;x;x;x;x;x;x;x;x;x;x;x;x;x;x;x;x;x;}
REPEAT20( val = pleaseconverge(val) );
But IMHO, if you need something much more complicated than that, then you should write your own pre-preprocessor. Your pre-preprocessor could for instance generate an appropriate header file for you, and it is easy enough to include this step in a Makefile to have everything compile smoothly by a single command. I've done it.