Conditional compilation depending on sizeof(double)

Question

Is there a macro (#define ...) somewhere in the bowels of Arduino IDE or core header files which tells me at compile time(!), if the double data type for a given platform/board is a true double precision floating point number (64 bit width) or "only" an alias for a single precision float (32 bit width)?

With Arduino, both can be the case. 8-bit AVR boards have double equal to float, ARM-based boards (like Arduino Zero) support "true" double precision.

I am working on a library which implements some bit-banging of floating point numbers (both single and double precision). Due to the low-level nature of the problem, there must be separate functions for float and double. I would like to do somthing like that:

uint32_t function_for_float(float f) {
 // function body here...
}
#if defined(HAS_64_BIT_DOUBLE)
uint64_t function_for_double(double d) {
 // function body here...
}
#endif

That way, a compile time error would be generated if somebody tried to use the "double" version on platforms which do not support the "true" double data type. I want the compiler error to be generated, because function_for_double() will do silently "the wrong thing" when fed a single precision float, leading potentially to difficult-to-debug problems.

Right now, I am defining a macro constant for every relevant board I own, like this:

#if defined(ARDUINO_SAMD_ZERO) \
 || defined(SAMD_FEATHER_M0_EXPRESS) \
 || defined(ADAFRUIT_FEATHER_M4_EXPRESS)
 #define HAS_64_BIT_DOUBLE
#endif

But this is ugly and not future-proof (new boards can/will become available...).

Any suggestions?

• Nick Gammon: This is not really an Arduino question. Arduino uses g++ to compile, so your question really is "can g++ do this?" Judging by this - not really. You can assert if double isn't 8 bytes (I tested that) but not make a choice, unless there is another tricky way of doing it.
• Andy: @NickGammon: Thanks for the link. I actually have a static_assert(sizeof(double) == 8) in the code of my "double function". However, then the complete library .cpp file will not compile on platforms where double ist only four bytes. This is what brought me to think about conditional complilation in the first place. I would like to generate a compiler error only if sombody tries to use the double version when he/she shouldn't. Somewhere in Arduino there surely must be definied that double is equal to float? I hope(d) to use this as a hook somehow...

Answer 1

The double data type is almost universally 8-bytes long, so you could simply

#if !__AVR__
# define HAS_64_BIT_DOUBLE
#endif

Gcc provides a more specific macro though: __SIZEOF_DOUBLE__, which has the same value as sizeof(double). But I do not know whether it is provided by other compilers.

• Andy: Perfect! I will make use macros __SIZEOF_DOUBLE__ and __SIZEOF_DOUBLE__, even if they are gcc/g++ specific. The Arduino IDE uses g++ as its backend, anyway, so this is "good enough" for me.
• Andy: Just verified that __SIZEOF_DOUBLE__ == 4 on an Arduino Pro Mini (AVR 328p) and __SIZEOF_DOUBLE__ == 8 on an Adafruit Feather M0 (ARM Cortex-M0+ / SAMD21). Thanks a lot! I have selected your answer as the correct one.