Question

So I’m serializing a C data structure for cross-platform use, and I want to make sure I’m recording my floating point numbers in a cross-platform manner.

I had been planning on just doing

 char * pos;
 /*...*/
 *((double*) pos) = dataStructureInstance->fieldWithOfTypeDouble;
 pos += sizeof(double);

But I wasn’t sure that the bytes would be recorded in the char * array in the IEEE 754 interchange format. I’ve been bitten by cross-platform issues before (endian-ness and whatnot). Is there anything I need to do to a double to get the bytes in interchange format?

Answer 1

C implementations are not required to use the IEEE 754 representation at all, so any solution based on directly accessing the binary representation of a double is not going to be genuinely portable anyway.

If you assume that you are only going to be working with systems that do represent doubles in the IEEE 754 format, and endianness is the only worry (and you assume that a big-endian sequence of bytes vs. a little-endian sequence of bytes are the only options!)… then you could perhaps probe endianness at run-time using some value with a known binary representation (in the same way that you could test an integer by looking to see if 0x12345678 was stored with 0x12 or 0x78 as the first byte).

(If you’re really concerned with portability, I’m not sure that there’s any better option than using sprintf() and strtod(), and assuming that you may not necessarily have any more accuracy than the absolute minimum required by he relevant specification, e.g. section 5.2.4.2.2 of the C99 spec.)