Should be casting used for setting a bitfield?

Question

Let's suppose, I have a following structure:

struct my_struct {
    uint32_t bf1 : 3;
    uint32_t bf2 : 5;
    uint32_t bf3 : 16;
    uint32_t bf4 : 8;
};

and the following enum:

enum bf1_values {
     Val1 = 0x0;
     Val2 = 0x4;
     Val3 = 0x7;
};

in addition, getter and setter functions for bf1:

uint32_t bf1_getter() {
    return global_struct.bf1; // cast value to (uint32_t)?
}


void bf1_setter(enum bf1_values val) {
    global_struct.bf1 = val; // cast enum to (uint32_t)?
}

Should I use the typecasting in getter and setter functions for safety?

EDIT:

The structure is supposed to be sent to HW.

EDIT2:

What I want to achieve is to be really sure that enum will be correctly written to a bitfield, and correctly read from bitfield.

Type-casting has nothing to do with safety, often it's quite the opposite. — Some programmer dude, Oct 08 '13 at 08:45
@JoachimPileborg ok, but what is the normal/good practice of such code writing? — Alex, Oct 08 '13 at 08:47
Does it compile without warnings? Then you don't need casting. Otherwise *you* have to make sure the code is safe, casting does not help you with that (it only helps you shut up the compiler warnings). — Some programmer dude, Oct 08 '13 at 08:51
Oh and by the way, if the bitfield structure is to be used by hardware, then you should make sure it's packed properly (so no padding between any fields). — Some programmer dude, Oct 08 '13 at 08:52
@JoachimPileborg regarding to packing, I know it. Anyway, thanks. — Alex, Oct 08 '13 at 08:54
if you need to set hw registers or smthn look at unions. have a look at this discussion and scroll to third answer --> [link](http://stackoverflow.com/questions/252552/why-do-we-need-c-unions) — David Tsulaia, Oct 08 '13 at 08:56
@KerrekSB I mean, that value will not overwrite other fields. — Alex, Oct 08 '13 at 09:01
@Alex: That isn't at all what "safety" normally means. Other fields will never be overwritten if you use bitfields in the normal way. — Kerrek SB, Oct 08 '13 at 09:02
`uint32_t bf1 : 3` does this ever make sense? You have a 3-bit wide unsigned integral field, not a 3-bit wide 32-bit-unsigned-integer. — n. m. could be an AI, Oct 08 '13 at 09:06
It's not clear what exactly you want to cast, and to which type. — n. m. could be an AI, Oct 08 '13 at 09:13

Christoph · Accepted Answer · 2013-10-08T09:43:49.330

No need for casting here - the assignments are already 'safe' insofar as a conforming implementation shouldn't corrupt other members. Assuming normal integer overflow semantics apply, the only problematic case is signed overflow, which may raise a signal (but I'm hard pressed to see that happening in practice if the bit field width is smaller than a full word as harware support for overflow detection will be lacking) and is otherwise implementation-defined. This caveat does not apply to your example as the target types are unsigned.

Keep in mind that bit field semantics are largely implementation-defined - even using a type other than int is actually a language extension - and it's up to you to check that the compiler does what you expect it to do on all relevant platforms.

A more portable but also less convenient approach would be to just use an uint32_t and do the bit fiddling manually. IF you don't need that protability, it should be fine as-is.

Should be casting used for setting a bitfield?

1 Answers1