Although we only expect Fuchsia to run on little endian (LE) CPU architectures, we still need to consider big endian (BE) issues. This doc explains:
A lot of peripheral hardware defines multi-byte BE values which must be converted.
Network byte order is BE. SCSI data structures are BE.
Even if Fuchsia never runs on a BE CPU (which it might someday, at least in theory), some of our code may be ported to a BE CPU.
Any time we define a multi-byte value, we create the possibility that another platform may want to write or read that value, and our code (which is open source) may be ported to that platform in order to do this.
Many modules do not need to do anything about endian issues; their data will only be interpreted by a single CPU running Fuchsia.
For those which might be ported to other OS's, or whose data might be exported by any channel:
Suggested style in C or C++ is to add
#include <endian.h> ... static_assert(__BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__);
either in every file, or accompanied by a comment explaining which files are not BE compatible.
(It's OK to not do anything, but better to make it explicit that the code is not BE compatible.)
In structures that are inherently endian, it's best to include macros that “convert” little-endian data to CPU endianness; this is a form of self-documenting code. Of course big-endian data should always use the macros.
Best style is to use the LE16 .. BE64 macros from endian.h, which should be available everywhere including DDK.
#include <endian.h> ... hw_le_struct.int_field = LE32(program_int); program_long = BE64(hw_be_struct.long_field);
To access multi-byte values in a byte buffer, use this crate. To convert integer values, use these methods.
