Test of Equality for String Macro - Invalid?

[Jesdisciple]

I'm trying to understand the problem here, and I've reduced it to a minimal test-case, main.c:

#define FOO "foo"

#include <stdio.h>
#include <stdlib.h>

int main(int argc, char** argv) {
#if FOO == 7
	/* ... */
#else
	/* ... */
#endif
	printf("Done\n");
	return (EXIT_SUCCESS);
}

I get this when compiling it:

$ gcc main.c
main.c:7:5: error: token ""foo"" is not valid in preprocessor expressions

??? Am I using an incorrect operator?

EDIT: Oh, I should also note that "Done" is never printed, even if no errors or warning are encountered. Not sure if that's related...

[whiteflags]

Just use strcmp or something. The equality operator is not for strings; at least, it doesn't do a lexicographic comparison.

[tabstop]

You're only allowed a constant behind an #if. "foo"==7 is not a constant, as it requires taking the address of the string "foo".

[Jesdisciple]

I'm surprised the preprocessor doesn't understand more complex comparisons than the compiler... Anyhow, thanks!

[matsp]

I'm surprised the preprocessor doesn't understand more complex comparisons than the compiler... Anyhow, thanks!

And exactly what do you expect "foo" to equal, numerically speaking?

If you want a set of values, you can define a number of constants, and then set your compile-option to one of those constants, e.g.

#define FOO_TYPE 1
#define BAR_TYPE 2
#deifne BAZ_TYPE 3
#define CONFIG_TYPE FOO_TYPE
...

#if CONFIG_TYPE == FOO_TYPE
...
#else if CONFIG_TYPE == BAR_TYPE
...
#else if CONFIG_TYPE == BAZ_TYPE
...
#else 
#error Must define CONFIG_TYPE macro
#endif
...

[Jesdisciple]

I wouldn't expect a numerical value per se, but several numerical values. Given the expansions the preprocessor is capable of, it seems to me that it could expand a string into a number of characters and compare each character against its corresponding value (which would have been 'f' == 7 && 'o' == 0 && 'o' == 0 in this case).

[matsp]

I wouldn't expect a numerical value per se, but several numerical values. Given the expansions the preprocessor is capable of, it seems to me that it could expand a string into a number of characters and compare each character against its corresponding value (which would have been 'f' == 7 && 0 == 'o' && 0 == 'o' in this case).

Assuming your logic is somewhat correct, I'd at least expect the values to be:
'f' == 0x66
'o' == 0x6f
'o' == 0x6f
[Assuming ASCII, but no other character set I know of has 'o' as zero, nor 'f' as 7].

It is POSSIBLE that the preprocessor would understand 'abc' as a numeric constant, although translating that to a decimal number involves knowing whether is is stored as 'a', 'b', 'c', 0 or 0, 'c', 'b', 'a' [or possibly some other permutation of the possible patterns of 0, a, b and c] and what that corresponds to as a decimal value. Tricky, but perhaps possible.

--
Mats

[Jesdisciple]

I was demonstrating what the comparison would expand to, and I reversed the 0s and 'o's until my edit. I didn't assign any values to the characters, just interpreted a comparison between a string and any other value as an incremental comparison of each set of corresponding components (where a number has only one component which is 0-padded as necessary).

[matsp]

Right, so if you use single-quotes, you will get something along those lines - I'm not sure if the C preprocessor accepts single-quoted characters. C itself does - and the corresponding value is an integer.

--
Mats

[Elysia]

I'm surprised the preprocessor doesn't understand more complex comparisons than the compiler... Anyhow, thanks!

The preprocessor is pretty dumb. It wasn't designed for complex things such as comparing strings.
Technically, this should be possible in C++ if you use const and compare the individual elements of the const char array.
In C++0x, it might even be possible to use a const function to do this at compile time.

[Jesdisciple]

Right, so if you use single-quotes, you will get something along those lines - I'm not sure if the C preprocessor accepts single-quoted characters. C itself does - and the corresponding value is an integer.

But that would defeat the purpose described in my original post, because the characters would be stored as separate macros.

In reflection, I realize that the behavior I expected while I was actually typing the code (rather than responding here) was that the preprocessor would actually compare the literal tokens - "foo" (5 characters) and 7 (1 character). I didn't expect it to even be conscious of compiler concepts like types, just literal code.

And for the record, I actually did use the strcmp() solution before the macro, but I wanted to optimize the executable by processing the equality before compilation rather than at runtime.

[Elysia]

Compile-time optimizations is not C's strong side.
C++ is much better in this regard.
I think you've hit a snag.
But perhaps you can hope the compiler optimizes it away if you tell it that it's constant data.

[tabstop]

The C preprocessor does accept single quoted characters as constants.

[matsp]

I have a feeling we are missing something important here - I suspect that the REAL code is doing something a bit different from the posted code, and what has been removed is important to understnad for us to be able to provide PROPER guidance of how you go about this using the C preprocessor [assuming it is possible].

And no, I didn't mean comparing single characters. You CAN have a multicharacter constant like this:

int x = 'foo';

I still haven't figured out whether those type of constants are available in the preprocessor - I think so. They are, however, clearly defined as "implementation dependant", so how this value is stored (e.g. if it is stored as the bytes 0, 'o', 'o', 'f' or some other way, is entirely up to the compiler, and it may not even be the same for two differnet compilers on the same model of processor [although IT PROBABLY IS]).

--
Mats

[tabstop]

I have a feeling we are missing something important here - I suspect that the REAL code is doing something a bit different from the posted code, and what has been removed is important to understnad for us to be able to provide PROPER guidance of how you go about this using the C preprocessor [assuming it is possible].

And no, I didn't mean comparing single characters. You CAN have a multicharacter constant like this:

int x = 'foo';

I still haven't figured out whether those type of constants are available in the preprocessor - I think so. They are, however, clearly defined as "implementation dependant", so how this value is stored (e.g. if it is stored as the bytes 0, 'o', 'o', 'f' or some other way, is entirely up to the compiler, and it may not even be the same for two differnet compilers on the same model of processor [although IT PROBABLY IS]).

--
Mats

Oh indeed. I suppose it's easy to check your particular implementation:

include <stdio.h>
#define FOO 'foo'

int main() {
    printf("%d\n", 'foo');
    printf("%d\n", FOO);

#if FOO == 0x666f6f
    printf("In order\n");
#elif FOO == 0x6f6f66
    printf("Backwards\n");
#else
    printf("Something else\n");
#endif
    return 0;
}

For what it's worth I get "in order" using gcc on the Mac, but I wouldn't extend that very far.