pub unsafe trait BitOrder: 'static {
// Required method
fn at<R>(index: BitIdx<R>) -> BitPos<R>
where R: BitRegister;
// Provided methods
fn select<R>(index: BitIdx<R>) -> BitSel<R>
where R: BitRegister { ... }
fn mask<R>(
from: impl Into<Option<BitIdx<R>>>,
upto: impl Into<Option<BitEnd<R>>>,
) -> BitMask<R>
where R: BitRegister { ... }
}
Expand description
§In-Element Bit Ordering
This trait manages the translation of semantic bit indices into electrical positions within storage elements of a memory region.
§Usage
bitvec
APIs operate on semantic index counters that exist in an abstract
memory space independently of the real memory that underlies them. In order to
affect real memory, bitvec
must translate these indices into real values. The
at
function maps abstract index values into their corresponding real
positions that can then be used to access memory.
You will likely never call any of the trait functions yourself. They are used by
bitvec
internals to operate on memory regions; all you need to do is provide
an implementation of this trait as a type parameter to bitvec
data structures.
§Safety
BitOrder
is unsafe to implement because its translation of index to position
cannot be forcibly checked by bitvec
itself, and an improper implementation
will lead to memory unsafety errors and unexpected collisions. The trait has
strict requirements for each function. If these are not upheld, then the
implementation is considered undefined at the library level and its use may
produce incorrect or undefined behavior during compilation.
You are responsible for running verify_for_type
or verify
in your test
suite if you implement BitOrder
.
§Implementation Rules
Values of this type are never constructed or passed to bitvec
functions. Your
implementation does not need to be zero-sized, but it will never have access to
an instance to view its state. It may refer to other global state, but per the
rules of at
, that state may not change while any bitvec
data structures are
alive.
The only function you need to provide is at
. Its requirements are listed in
its trait documentation.
You may also choose to provide implementations of select
and mask
. These
have a default implementation that is correct, but may be unoptimized for your
implementation. As such, you may replace them with a better version, but your
implementation of these functions must be exactly equal to the default
implementation for all possible inputs.
This requirement is checked by the verify_for_type
function.
§Verification
The verify_for_type
function verifies that a BitOrder
implementation is
correct for a single BitStore
implementor, and the verify
function runs
verify_for_type
on all unsigned integers that implement BitStore
on a
target. If you run these functions in your test suite, they will provide
detailed information if your implementation is incorrect.
§Examples
Implementations are not required to remain contiguous over a register, and may have any mapping they wish as long as it is total and bijective. This example swizzles the high and low halves of each byte.
use bitvec::{
order::BitOrder,
index::{BitIdx, BitPos},
mem::BitRegister,
};
pub struct HiLo;
unsafe impl BitOrder for HiLo {
fn at<R>(index: BitIdx<R>) -> BitPos<R>
where R: BitRegister {
unsafe { BitPos::new_unchecked(index.into_inner() ^ 4) }
}
}
#[test]
#[cfg(test)]
fn prove_hilo() {
bitvec::order::verify::<HiLo>();
}
Once a BitOrder
implementation passes the test suite, it can be freely used as
a type parameter in bitvec
data structures. The translation takes place
automatically, and you never need to look at this trait again.
Required Methods§
sourcefn at<R>(index: BitIdx<R>) -> BitPos<R>where
R: BitRegister,
fn at<R>(index: BitIdx<R>) -> BitPos<R>where
R: BitRegister,
Translates a semantic bit index into a real bit position.
This function is the basis of the trait, and must adhere to a number of requirements in order for an implementation to be correct.
§Type Parameters
R
: The memory element type that the index and position govern.
§Parameters
index
: A semantic bit-index within someR
element.
§Returns
The real position of the indexed bit within an R
element. See the
BitPos
documentation for what these positions are considered to mean.
§Requirements
This function must satisfy the following requirements for all possible
input and output values, for all possible R
type parameters:
- Totality: The implementation must be able to accept every input in
BitIdx::<R>::range_all()
, and produce someBitPos
value for each. - Bijection: There must be an exactly one-to-one correspondence between input and output values. No input index may choose its output from a set of more than one position, and no output position may be produced by more than one input index.
- Purity: The translation from index to position must be consistent for
the lifetime of at least all data structures in the program. This
function may refer to global state, but that state must be
immutable while any
bitvec
data structures exist, and must not be used to violate the totality or bijection requirements. - Validity: The produced
BitPos
value must be within the valid range of its type. This is enforced byBitPos::new
, but not by the unsafe constructorBitPos::new_unchecked
.
Provided Methods§
sourcefn select<R>(index: BitIdx<R>) -> BitSel<R>where
R: BitRegister,
fn select<R>(index: BitIdx<R>) -> BitSel<R>where
R: BitRegister,
Produces a single-bit selection mask from a bit-index.
This is an optional function: it is implemented as, and must always be
exactly identical to, BitOrder::at(index).select()
. If your ordering
has a faster implementation, you may provide it, but it must be exactly
numerically equivalent.
sourcefn mask<R>(
from: impl Into<Option<BitIdx<R>>>,
upto: impl Into<Option<BitEnd<R>>>,
) -> BitMask<R>where
R: BitRegister,
fn mask<R>(
from: impl Into<Option<BitIdx<R>>>,
upto: impl Into<Option<BitEnd<R>>>,
) -> BitMask<R>where
R: BitRegister,
Produces a multi-bit selection mask from a range of bit-indices.
This is an optional function: it is implemented as, and must always be
exactly identical to,
BitIdx::range(from, upto).map(BitOrder::select).sum()
. If your
ordering has a faster implementation, you may provide it, but it must be
exactly numerically equivalent.
§Parameters
from
: The inclusive starting value of the indices being selected. Defaults toBitIdx::MIN
.upto
: The exclusive ending value of the indices being selected. Defaults toBitEnd::MAX
.
§Returns
A selection mask with all bit-positions corresponding to from .. upto
selected.