zebra/zebra-chain/src/work/difficulty.rs

534 lines
19 KiB
Rust

//! Block difficulty data structures and calculations
//!
//! The block difficulty "target threshold" is stored in the block header as a
//! 32-bit `CompactDifficulty`. The `block::Hash` must be less than or equal
//! to the `ExpandedDifficulty` threshold, when represented as a 256-bit integer
//! in little-endian order.
//!
//! The target threshold is also used to calculate the `Work` for each block.
//! The block work is used to find the chain with the greatest total work. Each
//! block's work value depends on the fixed threshold in the block header, not
//! the actual work represented by the block header hash.
#![allow(clippy::unit_arg)]
use crate::{block, parameters::Network};
use std::{
cmp::{Ordering, PartialEq, PartialOrd},
convert::TryFrom,
fmt,
iter::Sum,
ops::Add,
ops::Div,
ops::Mul,
};
pub use crate::work::u256::U256;
#[cfg(any(test, feature = "proptest-impl"))]
mod arbitrary;
#[cfg(test)]
mod tests;
/// A 32-bit "compact bits" value, which represents the difficulty threshold for
/// a block header.
///
/// Used for:
/// - checking the `difficulty_threshold` value in the block header,
/// - calculating the 256-bit `ExpandedDifficulty` threshold, for comparison
/// with the block header hash, and
/// - calculating the block work.
///
/// Details:
///
/// This is a floating-point encoding, with a 24-bit signed mantissa,
/// an 8-bit exponent, an offset of 3, and a radix of 256.
/// (IEEE 754 32-bit floating-point values use a separate sign bit, an implicit
/// leading mantissa bit, an offset of 127, and a radix of 2.)
///
/// The precise bit pattern of a `CompactDifficulty` value is
/// consensus-critical, because it is used for the `difficulty_threshold` field,
/// which is:
/// - part of the `BlockHeader`, which is used to create the
/// `block::Hash`, and
/// - bitwise equal to the median `ExpandedDifficulty` value of recent blocks,
/// when encoded to `CompactDifficulty` using the specified conversion
/// function.
///
/// Without these consensus rules, some `ExpandedDifficulty` values would have
/// multiple equivalent `CompactDifficulty` values, due to redundancy in the
/// floating-point format.
#[derive(Clone, Copy, Eq, PartialEq, Serialize, Deserialize)]
pub struct CompactDifficulty(pub(crate) u32);
impl fmt::Debug for CompactDifficulty {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
// There isn't a standard way to show different representations of the
// same value
f.debug_tuple("CompactDifficulty")
// Use hex, because it's a float
.field(&format_args!("{:#010x}", self.0))
// Use expanded difficulty, for bitwise difficulty comparisons
.field(&format_args!("{:?}", self.to_expanded()))
.finish()
}
}
/// An invalid CompactDifficulty value, for testing.
pub const INVALID_COMPACT_DIFFICULTY: CompactDifficulty = CompactDifficulty(u32::MAX);
/// A 256-bit unsigned "expanded difficulty" value.
///
/// Used as a target threshold for the difficulty of a `block::Hash`.
///
/// Details:
///
/// The precise bit pattern of an `ExpandedDifficulty` value is
/// consensus-critical, because it is compared with the `block::Hash`.
///
/// Note that each `CompactDifficulty` value can be converted from a
/// range of `ExpandedDifficulty` values, because the precision of
/// the floating-point format requires rounding on conversion.
///
/// Therefore, consensus-critical code must perform the specified
/// conversions to `CompactDifficulty`, even if the original
/// `ExpandedDifficulty` values are known.
///
/// Callers should avoid constructing `ExpandedDifficulty` zero
/// values, because they are rejected by the consensus rules,
/// and cause some conversion functions to panic.
//
// TODO: Use NonZeroU256, when available
#[derive(Clone, Copy, Eq, PartialEq, Ord, PartialOrd)]
pub struct ExpandedDifficulty(U256);
impl fmt::Debug for ExpandedDifficulty {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
let mut buf = [0; 32];
// Use the same byte order as block::Hash
self.0.to_big_endian(&mut buf);
f.debug_tuple("ExpandedDifficulty")
.field(&hex::encode(&buf))
.finish()
}
}
/// A 128-bit unsigned "Work" value.
///
/// Used to calculate the total work for each chain of blocks.
///
/// Details:
///
/// The relative value of `Work` is consensus-critical, because it is used to
/// choose the best chain. But its precise value and bit pattern are not
/// consensus-critical.
///
/// We calculate work values according to the Zcash specification, but store
/// them as u128, rather than the implied u256. We don't expect the total chain
/// work to ever exceed 2^128. The current total chain work for Zcash is 2^58,
/// and Bitcoin adds around 2^91 work per year. (Each extra bit represents twice
/// as much work.)
#[derive(Clone, Copy, Default, Eq, PartialEq, Ord, PartialOrd)]
pub struct Work(u128);
impl Work {
/// Return the inner `u128` value.
pub fn as_u128(self) -> u128 {
self.0
}
}
impl fmt::Debug for Work {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
// There isn't a standard way to show different representations of the
// same value
f.debug_tuple("Work")
// Use hex, because expanded difficulty is in hex.
.field(&format_args!("{:#x}", self.0))
// Use decimal, to compare with zcashd
.field(&format_args!("{}", self.0))
// Use log2, to compare with zcashd
.field(&format_args!("{:.5}", (self.0 as f64).log2()))
.finish()
}
}
impl CompactDifficulty {
/// CompactDifficulty exponent base.
const BASE: u32 = 256;
/// CompactDifficulty exponent offset.
const OFFSET: i32 = 3;
/// CompactDifficulty floating-point precision.
const PRECISION: u32 = 24;
/// CompactDifficulty sign bit, part of the signed mantissa.
const SIGN_BIT: u32 = 1 << (CompactDifficulty::PRECISION - 1);
/// CompactDifficulty unsigned mantissa mask.
///
/// Also the maximum unsigned mantissa value.
const UNSIGNED_MANTISSA_MASK: u32 = CompactDifficulty::SIGN_BIT - 1;
/// Calculate the ExpandedDifficulty for a compact representation.
///
/// See `ToTarget()` in the Zcash Specification, and `CheckProofOfWork()` in
/// zcashd.
///
/// Returns None for negative, zero, and overflow values. (zcashd rejects
/// these values, before comparing the hash.)
#[allow(clippy::unwrap_in_result)]
pub fn to_expanded(self) -> Option<ExpandedDifficulty> {
// The constants for this floating-point representation.
// Alias the struct constants here, so the code is easier to read.
const BASE: u32 = CompactDifficulty::BASE;
const OFFSET: i32 = CompactDifficulty::OFFSET;
const PRECISION: u32 = CompactDifficulty::PRECISION;
const SIGN_BIT: u32 = CompactDifficulty::SIGN_BIT;
const UNSIGNED_MANTISSA_MASK: u32 = CompactDifficulty::UNSIGNED_MANTISSA_MASK;
// Negative values in this floating-point representation.
// 0 if (x & 2^23 == 2^23)
// zcashd rejects negative values without comparing the hash.
if self.0 & SIGN_BIT == SIGN_BIT {
return None;
}
// The components of the result
// The fractional part of the floating-point number
// x & (2^23 - 1)
let mantissa = self.0 & UNSIGNED_MANTISSA_MASK;
// The exponent for the multiplier in the floating-point number
// 256^(floor(x/(2^24)) - 3)
//
// The i32 conversion is safe, because we've just divided self by 2^24.
let exponent = i32::try_from(self.0 >> PRECISION).expect("fits in i32") - OFFSET;
// Normalise the mantissa and exponent before multiplying.
//
// zcashd rejects non-zero overflow values, but accepts overflows where
// all the overflowing bits are zero. It also allows underflows.
let (mantissa, exponent) = match (mantissa, exponent) {
// Overflow: check for non-zero overflow bits
//
// If m is non-zero, overflow. If m is zero, invalid.
(_, e) if (e >= 32) => return None,
// If m is larger than the remaining bytes, overflow.
// Otherwise, avoid overflows in base^exponent.
(m, e) if (e == 31 && m > u8::MAX.into()) => return None,
(m, e) if (e == 31 && m <= u8::MAX.into()) => (m << 16, e - 2),
(m, e) if (e == 30 && m > u16::MAX.into()) => return None,
(m, e) if (e == 30 && m <= u16::MAX.into()) => (m << 8, e - 1),
// Underflow: perform the right shift.
// The abs is safe, because we've just divided by 2^24, and offset
// is small.
(m, e) if (e < 0) => (m >> ((e.abs() * 8) as u32), 0),
(m, e) => (m, e),
};
// Now calculate the result: mantissa*base^exponent
// Earlier code should make sure all these values are in range.
let mantissa: U256 = mantissa.into();
let base: U256 = BASE.into();
let exponent: U256 = exponent.into();
let result = mantissa * base.pow(exponent);
if result == U256::zero() {
// zcashd rejects zero values, without comparing the hash
None
} else {
Some(result.into())
}
}
/// Calculate the Work for a compact representation.
///
/// See `Definition of Work` in the [Zcash Specification], and
/// `GetBlockProof()` in zcashd.
///
/// Returns None if the corresponding ExpandedDifficulty is None.
/// Also returns None on Work overflow, which should be impossible on a
/// valid chain.
///
/// [Zcash Specification]: https://zips.z.cash/protocol/protocol.pdf#workdef
pub fn to_work(self) -> Option<Work> {
let expanded = self.to_expanded()?;
Work::try_from(expanded).ok()
}
}
impl TryFrom<ExpandedDifficulty> for Work {
type Error = ();
fn try_from(expanded: ExpandedDifficulty) -> Result<Self, Self::Error> {
// We need to compute `2^256 / (expanded + 1)`, but we can't represent
// 2^256, as it's too large for a u256. However, as 2^256 is at least as
// large as `expanded + 1`, it is equal to
// `((2^256 - expanded - 1) / (expanded + 1)) + 1`, or
let result = (!expanded.0 / (expanded.0 + 1)) + 1;
if result <= u128::MAX.into() {
Ok(Work(result.as_u128()))
} else {
Err(())
}
}
}
impl From<ExpandedDifficulty> for CompactDifficulty {
fn from(value: ExpandedDifficulty) -> Self {
value.to_compact()
}
}
impl ExpandedDifficulty {
/// Returns the difficulty of the hash.
///
/// Used to implement comparisons between difficulties and hashes.
///
/// Usage:
///
/// Compare the hash with the calculated difficulty value, using Rust's
/// standard comparison operators.
///
/// Hashes are not used to calculate the difficulties of future blocks, so
/// users of this module should avoid converting hashes into difficulties.
pub(super) fn from_hash(hash: &block::Hash) -> ExpandedDifficulty {
U256::from_little_endian(&hash.0).into()
}
/// Returns the easiest target difficulty allowed on `network`.
///
/// See `PoWLimit` in the Zcash specification.
pub fn target_difficulty_limit(network: Network) -> ExpandedDifficulty {
let limit: U256 = match network {
/* 2^243 - 1 */
Network::Mainnet => (U256::one() << 243) - 1,
/* 2^251 - 1 */
Network::Testnet => (U256::one() << 251) - 1,
};
// `zcashd` converts the PoWLimit into a compact representation before
// using it to perform difficulty filter checks.
//
// The Zcash specification converts to compact for the default difficulty
// filter, but not for testnet minimum difficulty blocks. (ZIP 205 and
// ZIP 208 don't specify this conversion either.) See #1277 for details.
ExpandedDifficulty(limit)
.to_compact()
.to_expanded()
.expect("difficulty limits are valid expanded values")
}
/// Calculate the CompactDifficulty for an expanded difficulty.
///
/// See `ToCompact()` in the Zcash Specification, and `GetCompact()`
/// in zcashd.
///
/// Panics:
///
/// If `self` is zero.
///
/// `ExpandedDifficulty` values are generated in two ways:
/// * conversion from `CompactDifficulty` values, which rejects zeroes, and
/// * difficulty adjustment calculations, which impose a non-zero minimum
/// `target_difficulty_limit`.
///
/// Neither of these methods yield zero values.
pub fn to_compact(self) -> CompactDifficulty {
// The zcashd implementation supports negative and zero compact values.
// These values are rejected by the protocol rules. Zebra is designed so
// that invalid states are not representable. Therefore, this function
// does not produce negative compact values, and panics on zero compact
// values. (The negative compact value code in zcashd is unused.)
assert!(self.0 > 0.into(), "Zero difficulty values are invalid");
// The constants for this floating-point representation.
// Alias the constants here, so the code is easier to read.
const UNSIGNED_MANTISSA_MASK: u32 = CompactDifficulty::UNSIGNED_MANTISSA_MASK;
const OFFSET: i32 = CompactDifficulty::OFFSET;
// Calculate the final size, accounting for the sign bit.
// This is the size *after* applying the sign bit adjustment in `ToCompact()`.
let size = self.0.bits() / 8 + 1;
// Make sure the mantissa is non-negative, by shifting down values that
// would otherwise overflow into the sign bit
let mantissa = if self.0 <= UNSIGNED_MANTISSA_MASK.into() {
// Value is small, shift up if needed
self.0 << (8 * (3 - size))
} else {
// Value is large, shift down
self.0 >> (8 * (size - 3))
};
// This assertion also makes sure that size fits in its 8 bit compact field
assert!(
size < (31 + OFFSET) as _,
"256^size (256^{}) must fit in a u256, after the sign bit adjustment and offset",
size
);
let size = u32::try_from(size).expect("a 0-6 bit value fits in a u32");
assert!(
mantissa <= UNSIGNED_MANTISSA_MASK.into(),
"mantissa {:x?} must fit in its compact field",
mantissa
);
let mantissa = u32::try_from(mantissa).expect("a 0-23 bit value fits in a u32");
if mantissa > 0 {
CompactDifficulty(mantissa + (size << 24))
} else {
// This check catches invalid mantissas. Overflows and underflows
// should also be unreachable, but they aren't caught here.
unreachable!("converted CompactDifficulty values must be valid")
}
}
}
impl From<U256> for ExpandedDifficulty {
fn from(value: U256) -> Self {
ExpandedDifficulty(value)
}
}
impl From<ExpandedDifficulty> for U256 {
fn from(value: ExpandedDifficulty) -> Self {
value.0
}
}
impl Sum<ExpandedDifficulty> for ExpandedDifficulty {
fn sum<I: Iterator<Item = ExpandedDifficulty>>(iter: I) -> Self {
iter.map(|d| d.0).fold(U256::zero(), Add::add).into()
}
}
impl<T> Div<T> for ExpandedDifficulty
where
T: Into<U256>,
{
type Output = ExpandedDifficulty;
fn div(self, rhs: T) -> Self::Output {
ExpandedDifficulty(self.0 / rhs)
}
}
impl<T> Mul<T> for ExpandedDifficulty
where
U256: Mul<T>,
<U256 as Mul<T>>::Output: Into<U256>,
{
type Output = ExpandedDifficulty;
fn mul(self, rhs: T) -> ExpandedDifficulty {
ExpandedDifficulty((self.0 * rhs).into())
}
}
impl PartialEq<block::Hash> for ExpandedDifficulty {
/// Is `self` equal to `other`?
///
/// See `partial_cmp` for details.
fn eq(&self, other: &block::Hash) -> bool {
self.partial_cmp(other) == Some(Ordering::Equal)
}
}
impl PartialOrd<block::Hash> for ExpandedDifficulty {
/// `block::Hash`es are compared with `ExpandedDifficulty` thresholds by
/// converting the hash to a 256-bit integer in little-endian order.
///
/// Greater values represent *less* work. This matches the convention in
/// zcashd and bitcoin.
fn partial_cmp(&self, other: &block::Hash) -> Option<Ordering> {
self.partial_cmp(&ExpandedDifficulty::from_hash(other))
}
}
impl PartialEq<ExpandedDifficulty> for block::Hash {
/// Is `self` equal to `other`?
///
/// See `<ExpandedDifficulty as PartialOrd<block::Hash>::partial_cmp`
/// for details.
fn eq(&self, other: &ExpandedDifficulty) -> bool {
other.eq(self)
}
}
impl PartialOrd<ExpandedDifficulty> for block::Hash {
/// How does `self` compare to `other`?
///
/// See `<ExpandedDifficulty as PartialOrd<block::Hash>::partial_cmp`
/// for details.
#[allow(clippy::unwrap_in_result)]
fn partial_cmp(&self, other: &ExpandedDifficulty) -> Option<Ordering> {
Some(
// Use the canonical implementation, but reverse the order
other
.partial_cmp(self)
.expect("difficulties and hashes have a total order")
.reverse(),
)
}
}
impl std::ops::Add for Work {
type Output = PartialCumulativeWork;
fn add(self, rhs: Work) -> PartialCumulativeWork {
PartialCumulativeWork::from(self) + rhs
}
}
#[derive(Clone, Copy, Debug, Default, PartialEq, Eq, PartialOrd, Ord)]
/// Partial work used to track relative work in non-finalized chains
pub struct PartialCumulativeWork(u128);
impl From<Work> for PartialCumulativeWork {
fn from(work: Work) -> Self {
PartialCumulativeWork(work.0)
}
}
impl std::ops::Add<Work> for PartialCumulativeWork {
type Output = PartialCumulativeWork;
fn add(self, rhs: Work) -> Self::Output {
let result = self
.0
.checked_add(rhs.0)
.expect("Work values do not overflow");
PartialCumulativeWork(result)
}
}
impl std::ops::AddAssign<Work> for PartialCumulativeWork {
fn add_assign(&mut self, rhs: Work) {
*self = *self + rhs;
}
}
impl std::ops::Sub<Work> for PartialCumulativeWork {
type Output = PartialCumulativeWork;
fn sub(self, rhs: Work) -> Self::Output {
let result = self.0
.checked_sub(rhs.0)
.expect("PartialCumulativeWork values do not underflow: all subtracted Work values must have been previously added to the PartialCumulativeWork");
PartialCumulativeWork(result)
}
}
impl std::ops::SubAssign<Work> for PartialCumulativeWork {
fn sub_assign(&mut self, rhs: Work) {
*self = *self - rhs;
}
}