Emit `const` function for bitfields in bindings

[Luro02]

Motivations

• Would you like to implement this feature? maybe

While working on my PR for esp-idf-hal esp-rs/esp-idf-hal#548 I noticed that I am unable to create a const for the used rmt_symbol_word_t wrapper.

This is unfortunate, because I would have to use lazy_static instead or leave them in the respective functions as local variables:

fn byte_to_symbols(byte: u8) -> impl IntoIterator<Item = Symbol> {
    let ws2812_zero = Symbol::new_with(
        RMT_LED_STRIP_RESOLUTION_HZ,
        PinState::High,
        Duration::from_micros(300),
        PinState::Low,
        Duration::from_micros(900),
    )
    .unwrap();
    let ws2812_one = Symbol::new_with(
        RMT_LED_STRIP_RESOLUTION_HZ,
        PinState::High,
        Duration::from_micros(900),
        PinState::Low,
        Duration::from_micros(300),
    )
    .unwrap();

    (0..8)
        .map(|i| 0x80 >> i)
        .map(move |bitmask| (byte & bitmask) != 0)
        .map(move |is_one| if is_one { ws2812_one } else { ws2812_zero })
}

It is a relatively minor inconvenience, but given that the generated code like this one

    #[inline]
    pub fn set_level0(&mut self, val: u16) {
        unsafe {
            let val: u16 = ::core::mem::transmute(val);
            self._bitfield_1.set(15usize, 1u8, val as u64)
        }
    }

could be a const function, I am making this issue in the hopes that this can be achieved by simply setting a flag when generating the bindings.

[Luro02]

From what I can tell, the problem is that bindgen uses the BitfieldUnit which is generic over T and uses traits to implement the functions for some T:

https://github.com/rust-lang/rust-bindgen/blob/main/bindgen/codegen/bitfield_unit.rs

I assume the T is always some [T; N], in which case one might be able to implement a const function (one would have to reimplement the bitfield and instruct bindgen to not emit it: https://github.com/rust-lang/rust-bindgen/blob/50ec3ee0d9436ff0d6e50c5ff59bf05ef137b386/book/src/using-bitfields.md?plain=1#L12).

If this is not worth the effort, feel free to close the issue, it doesn't look trivial.