// Bitcoin Hashes Library // Written in 2018 by // Andrew Poelstra // // To the extent possible under law, the author(s) have dedicated all // copyright and related and neighboring rights to this software to // the public domain worldwide. This software is distributed without // any warranty. // // You should have received a copy of the CC0 Public Domain Dedication // along with this software. // If not, see . // // This module is largely copied from the rust-crypto ripemd.rs file; // while rust-crypto is licensed under Apache, that file specifically // was written entirely by Andrew Poelstra, who is re-licensing its // contents here as CC0. //! RIPEMD160 implementation. //! use core::{cmp, str}; use core::convert::TryInto; use core::ops::Index; use core::slice::SliceIndex; use crate::{Error, HashEngine as _}; crate::internal_macros::hash_type! { 160, false, "Output of the RIPEMD160 hash function.", "crate::util::json_hex_string::len_20" } #[cfg(not(fuzzing))] fn from_engine(mut e: HashEngine) -> Hash { // pad buffer with a single 1-bit then all 0s, until there are exactly 8 bytes remaining let data_len = e.length as u64; let zeroes = [0; BLOCK_SIZE - 8]; e.input(&[0x80]); if e.length % BLOCK_SIZE > zeroes.len() { e.input(&zeroes); } let pad_length = zeroes.len() - (e.length % BLOCK_SIZE); e.input(&zeroes[..pad_length]); debug_assert_eq!(e.length % BLOCK_SIZE, zeroes.len()); e.input(&(8 * data_len).to_le_bytes()); debug_assert_eq!(e.length % BLOCK_SIZE, 0); Hash(e.midstate()) } #[cfg(fuzzing)] fn from_engine(e: HashEngine) -> Hash { let mut res = e.midstate(); res[0] ^= (e.length & 0xff) as u8; Hash(res) } const BLOCK_SIZE: usize = 64; /// Engine to compute RIPEMD160 hash function. #[derive(Clone)] pub struct HashEngine { buffer: [u8; BLOCK_SIZE], h: [u32; 5], length: usize, } impl Default for HashEngine { fn default() -> Self { HashEngine { h: [0x67452301, 0xefcdab89, 0x98badcfe, 0x10325476, 0xc3d2e1f0], length: 0, buffer: [0; BLOCK_SIZE], } } } impl crate::HashEngine for HashEngine { type MidState = [u8; 20]; #[cfg(not(fuzzing))] fn midstate(&self) -> [u8; 20] { let mut ret = [0; 20]; for (val, ret_bytes) in self.h.iter().zip(ret.chunks_exact_mut(4)) { ret_bytes.copy_from_slice(&(*val).to_le_bytes()); } ret } #[cfg(fuzzing)] fn midstate(&self) -> [u8; 20] { let mut ret = [0; 20]; ret.copy_from_slice(&self.buffer[..20]); ret } const BLOCK_SIZE: usize = 64; fn n_bytes_hashed(&self) -> usize { self.length } engine_input_impl!(); } macro_rules! round( ($a:expr, $b:expr, $c:expr, $d:expr, $e:expr, $x:expr, $bits:expr, $add:expr, $round:expr) => ({ $a = $a.wrapping_add($round).wrapping_add($x).wrapping_add($add); $a = $a.rotate_left($bits).wrapping_add($e); $c = $c.rotate_left(10); }); ); macro_rules! process_block( ($h:expr, $data:expr, $( round1: h_ordering $f0:expr, $f1:expr, $f2:expr, $f3:expr, $f4:expr; data_index $data_index1:expr; roll_shift $bits1:expr; )* $( round2: h_ordering $g0:expr, $g1:expr, $g2:expr, $g3:expr, $g4:expr; data_index $data_index2:expr; roll_shift $bits2:expr; )* $( round3: h_ordering $h0:expr, $h1:expr, $h2:expr, $h3:expr, $h4:expr; data_index $data_index3:expr; roll_shift $bits3:expr; )* $( round4: h_ordering $i0:expr, $i1:expr, $i2:expr, $i3:expr, $i4:expr; data_index $data_index4:expr; roll_shift $bits4:expr; )* $( round5: h_ordering $j0:expr, $j1:expr, $j2:expr, $j3:expr, $j4:expr; data_index $data_index5:expr; roll_shift $bits5:expr; )* $( par_round1: h_ordering $pj0:expr, $pj1:expr, $pj2:expr, $pj3:expr, $pj4:expr; data_index $pdata_index1:expr; roll_shift $pbits1:expr; )* $( par_round2: h_ordering $pi0:expr, $pi1:expr, $pi2:expr, $pi3:expr, $pi4:expr; data_index $pdata_index2:expr; roll_shift $pbits2:expr; )* $( par_round3: h_ordering $ph0:expr, $ph1:expr, $ph2:expr, $ph3:expr, $ph4:expr; data_index $pdata_index3:expr; roll_shift $pbits3:expr; )* $( par_round4: h_ordering $pg0:expr, $pg1:expr, $pg2:expr, $pg3:expr, $pg4:expr; data_index $pdata_index4:expr; roll_shift $pbits4:expr; )* $( par_round5: h_ordering $pf0:expr, $pf1:expr, $pf2:expr, $pf3:expr, $pf4:expr; data_index $pdata_index5:expr; roll_shift $pbits5:expr; )* ) => ({ let mut bb = $h; let mut bbb = $h; // Round 1 $( round!(bb[$f0], bb[$f1], bb[$f2], bb[$f3], bb[$f4], $data[$data_index1], $bits1, 0x00000000, bb[$f1] ^ bb[$f2] ^ bb[$f3]); )* // Round 2 $( round!(bb[$g0], bb[$g1], bb[$g2], bb[$g3], bb[$g4], $data[$data_index2], $bits2, 0x5a827999, (bb[$g1] & bb[$g2]) | (!bb[$g1] & bb[$g3])); )* // Round 3 $( round!(bb[$h0], bb[$h1], bb[$h2], bb[$h3], bb[$h4], $data[$data_index3], $bits3, 0x6ed9eba1, (bb[$h1] | !bb[$h2]) ^ bb[$h3]); )* // Round 4 $( round!(bb[$i0], bb[$i1], bb[$i2], bb[$i3], bb[$i4], $data[$data_index4], $bits4, 0x8f1bbcdc, (bb[$i1] & bb[$i3]) | (bb[$i2] & !bb[$i3])); )* // Round 5 $( round!(bb[$j0], bb[$j1], bb[$j2], bb[$j3], bb[$j4], $data[$data_index5], $bits5, 0xa953fd4e, bb[$j1] ^ (bb[$j2] | !bb[$j3])); )* // Parallel rounds: these are the same as the previous five // rounds except that the constants have changed, we work // with the other buffer, and they are applied in reverse // order. // Parallel Round 1 $( round!(bbb[$pj0], bbb[$pj1], bbb[$pj2], bbb[$pj3], bbb[$pj4], $data[$pdata_index1], $pbits1, 0x50a28be6, bbb[$pj1] ^ (bbb[$pj2] | !bbb[$pj3])); )* // Porallel Round 2 $( round!(bbb[$pi0], bbb[$pi1], bbb[$pi2], bbb[$pi3], bbb[$pi4], $data[$pdata_index2], $pbits2, 0x5c4dd124, (bbb[$pi1] & bbb[$pi3]) | (bbb[$pi2] & !bbb[$pi3])); )* // Parallel Round 3 $( round!(bbb[$ph0], bbb[$ph1], bbb[$ph2], bbb[$ph3], bbb[$ph4], $data[$pdata_index3], $pbits3, 0x6d703ef3, (bbb[$ph1] | !bbb[$ph2]) ^ bbb[$ph3]); )* // Parallel Round 4 $( round!(bbb[$pg0], bbb[$pg1], bbb[$pg2], bbb[$pg3], bbb[$pg4], $data[$pdata_index4], $pbits4, 0x7a6d76e9, (bbb[$pg1] & bbb[$pg2]) | (!bbb[$pg1] & bbb[$pg3])); )* // Parallel Round 5 $( round!(bbb[$pf0], bbb[$pf1], bbb[$pf2], bbb[$pf3], bbb[$pf4], $data[$pdata_index5], $pbits5, 0x00000000, bbb[$pf1] ^ bbb[$pf2] ^ bbb[$pf3]); )* // Combine results bbb[3] = bbb[3].wrapping_add($h[1]).wrapping_add(bb[2]); $h[1] = $h[2].wrapping_add(bb[3]).wrapping_add(bbb[4]); $h[2] = $h[3].wrapping_add(bb[4]).wrapping_add(bbb[0]); $h[3] = $h[4].wrapping_add(bb[0]).wrapping_add(bbb[1]); $h[4] = $h[0].wrapping_add(bb[1]).wrapping_add(bbb[2]); $h[0] = bbb[3]; }); ); impl HashEngine { fn process_block(&mut self) { debug_assert_eq!(self.buffer.len(), BLOCK_SIZE); let mut w = [0u32; 16]; for (w_val, buff_bytes) in w.iter_mut().zip(self.buffer.chunks_exact(4)) { *w_val = u32::from_le_bytes(buff_bytes.try_into().expect("4 byte slice")) } process_block!(self.h, w, // Round 1 round1: h_ordering 0, 1, 2, 3, 4; data_index 0; roll_shift 11; round1: h_ordering 4, 0, 1, 2, 3; data_index 1; roll_shift 14; round1: h_ordering 3, 4, 0, 1, 2; data_index 2; roll_shift 15; round1: h_ordering 2, 3, 4, 0, 1; data_index 3; roll_shift 12; round1: h_ordering 1, 2, 3, 4, 0; data_index 4; roll_shift 5; round1: h_ordering 0, 1, 2, 3, 4; data_index 5; roll_shift 8; round1: h_ordering 4, 0, 1, 2, 3; data_index 6; roll_shift 7; round1: h_ordering 3, 4, 0, 1, 2; data_index 7; roll_shift 9; round1: h_ordering 2, 3, 4, 0, 1; data_index 8; roll_shift 11; round1: h_ordering 1, 2, 3, 4, 0; data_index 9; roll_shift 13; round1: h_ordering 0, 1, 2, 3, 4; data_index 10; roll_shift 14; round1: h_ordering 4, 0, 1, 2, 3; data_index 11; roll_shift 15; round1: h_ordering 3, 4, 0, 1, 2; data_index 12; roll_shift 6; round1: h_ordering 2, 3, 4, 0, 1; data_index 13; roll_shift 7; round1: h_ordering 1, 2, 3, 4, 0; data_index 14; roll_shift 9; round1: h_ordering 0, 1, 2, 3, 4; data_index 15; roll_shift 8; // Round 2 round2: h_ordering 4, 0, 1, 2, 3; data_index 7; roll_shift 7; round2: h_ordering 3, 4, 0, 1, 2; data_index 4; roll_shift 6; round2: h_ordering 2, 3, 4, 0, 1; data_index 13; roll_shift 8; round2: h_ordering 1, 2, 3, 4, 0; data_index 1; roll_shift 13; round2: h_ordering 0, 1, 2, 3, 4; data_index 10; roll_shift 11; round2: h_ordering 4, 0, 1, 2, 3; data_index 6; roll_shift 9; round2: h_ordering 3, 4, 0, 1, 2; data_index 15; roll_shift 7; round2: h_ordering 2, 3, 4, 0, 1; data_index 3; roll_shift 15; round2: h_ordering 1, 2, 3, 4, 0; data_index 12; roll_shift 7; round2: h_ordering 0, 1, 2, 3, 4; data_index 0; roll_shift 12; round2: h_ordering 4, 0, 1, 2, 3; data_index 9; roll_shift 15; round2: h_ordering 3, 4, 0, 1, 2; data_index 5; roll_shift 9; round2: h_ordering 2, 3, 4, 0, 1; data_index 2; roll_shift 11; round2: h_ordering 1, 2, 3, 4, 0; data_index 14; roll_shift 7; round2: h_ordering 0, 1, 2, 3, 4; data_index 11; roll_shift 13; round2: h_ordering 4, 0, 1, 2, 3; data_index 8; roll_shift 12; // Round 3 round3: h_ordering 3, 4, 0, 1, 2; data_index 3; roll_shift 11; round3: h_ordering 2, 3, 4, 0, 1; data_index 10; roll_shift 13; round3: h_ordering 1, 2, 3, 4, 0; data_index 14; roll_shift 6; round3: h_ordering 0, 1, 2, 3, 4; data_index 4; roll_shift 7; round3: h_ordering 4, 0, 1, 2, 3; data_index 9; roll_shift 14; round3: h_ordering 3, 4, 0, 1, 2; data_index 15; roll_shift 9; round3: h_ordering 2, 3, 4, 0, 1; data_index 8; roll_shift 13; round3: h_ordering 1, 2, 3, 4, 0; data_index 1; roll_shift 15; round3: h_ordering 0, 1, 2, 3, 4; data_index 2; roll_shift 14; round3: h_ordering 4, 0, 1, 2, 3; data_index 7; roll_shift 8; round3: h_ordering 3, 4, 0, 1, 2; data_index 0; roll_shift 13; round3: h_ordering 2, 3, 4, 0, 1; data_index 6; roll_shift 6; round3: h_ordering 1, 2, 3, 4, 0; data_index 13; roll_shift 5; round3: h_ordering 0, 1, 2, 3, 4; data_index 11; roll_shift 12; round3: h_ordering 4, 0, 1, 2, 3; data_index 5; roll_shift 7; round3: h_ordering 3, 4, 0, 1, 2; data_index 12; roll_shift 5; // Round 4 round4: h_ordering 2, 3, 4, 0, 1; data_index 1; roll_shift 11; round4: h_ordering 1, 2, 3, 4, 0; data_index 9; roll_shift 12; round4: h_ordering 0, 1, 2, 3, 4; data_index 11; roll_shift 14; round4: h_ordering 4, 0, 1, 2, 3; data_index 10; roll_shift 15; round4: h_ordering 3, 4, 0, 1, 2; data_index 0; roll_shift 14; round4: h_ordering 2, 3, 4, 0, 1; data_index 8; roll_shift 15; round4: h_ordering 1, 2, 3, 4, 0; data_index 12; roll_shift 9; round4: h_ordering 0, 1, 2, 3, 4; data_index 4; roll_shift 8; round4: h_ordering 4, 0, 1, 2, 3; data_index 13; roll_shift 9; round4: h_ordering 3, 4, 0, 1, 2; data_index 3; roll_shift 14; round4: h_ordering 2, 3, 4, 0, 1; data_index 7; roll_shift 5; round4: h_ordering 1, 2, 3, 4, 0; data_index 15; roll_shift 6; round4: h_ordering 0, 1, 2, 3, 4; data_index 14; roll_shift 8; round4: h_ordering 4, 0, 1, 2, 3; data_index 5; roll_shift 6; round4: h_ordering 3, 4, 0, 1, 2; data_index 6; roll_shift 5; round4: h_ordering 2, 3, 4, 0, 1; data_index 2; roll_shift 12; // Round 5 round5: h_ordering 1, 2, 3, 4, 0; data_index 4; roll_shift 9; round5: h_ordering 0, 1, 2, 3, 4; data_index 0; roll_shift 15; round5: h_ordering 4, 0, 1, 2, 3; data_index 5; roll_shift 5; round5: h_ordering 3, 4, 0, 1, 2; data_index 9; roll_shift 11; round5: h_ordering 2, 3, 4, 0, 1; data_index 7; roll_shift 6; round5: h_ordering 1, 2, 3, 4, 0; data_index 12; roll_shift 8; round5: h_ordering 0, 1, 2, 3, 4; data_index 2; roll_shift 13; round5: h_ordering 4, 0, 1, 2, 3; data_index 10; roll_shift 12; round5: h_ordering 3, 4, 0, 1, 2; data_index 14; roll_shift 5; round5: h_ordering 2, 3, 4, 0, 1; data_index 1; roll_shift 12; round5: h_ordering 1, 2, 3, 4, 0; data_index 3; roll_shift 13; round5: h_ordering 0, 1, 2, 3, 4; data_index 8; roll_shift 14; round5: h_ordering 4, 0, 1, 2, 3; data_index 11; roll_shift 11; round5: h_ordering 3, 4, 0, 1, 2; data_index 6; roll_shift 8; round5: h_ordering 2, 3, 4, 0, 1; data_index 15; roll_shift 5; round5: h_ordering 1, 2, 3, 4, 0; data_index 13; roll_shift 6; // Porallel Round 1; par_round1: h_ordering 0, 1, 2, 3, 4; data_index 5; roll_shift 8; par_round1: h_ordering 4, 0, 1, 2, 3; data_index 14; roll_shift 9; par_round1: h_ordering 3, 4, 0, 1, 2; data_index 7; roll_shift 9; par_round1: h_ordering 2, 3, 4, 0, 1; data_index 0; roll_shift 11; par_round1: h_ordering 1, 2, 3, 4, 0; data_index 9; roll_shift 13; par_round1: h_ordering 0, 1, 2, 3, 4; data_index 2; roll_shift 15; par_round1: h_ordering 4, 0, 1, 2, 3; data_index 11; roll_shift 15; par_round1: h_ordering 3, 4, 0, 1, 2; data_index 4; roll_shift 5; par_round1: h_ordering 2, 3, 4, 0, 1; data_index 13; roll_shift 7; par_round1: h_ordering 1, 2, 3, 4, 0; data_index 6; roll_shift 7; par_round1: h_ordering 0, 1, 2, 3, 4; data_index 15; roll_shift 8; par_round1: h_ordering 4, 0, 1, 2, 3; data_index 8; roll_shift 11; par_round1: h_ordering 3, 4, 0, 1, 2; data_index 1; roll_shift 14; par_round1: h_ordering 2, 3, 4, 0, 1; data_index 10; roll_shift 14; par_round1: h_ordering 1, 2, 3, 4, 0; data_index 3; roll_shift 12; par_round1: h_ordering 0, 1, 2, 3, 4; data_index 12; roll_shift 6; // Parallel Round 2 par_round2: h_ordering 4, 0, 1, 2, 3; data_index 6; roll_shift 9; par_round2: h_ordering 3, 4, 0, 1, 2; data_index 11; roll_shift 13; par_round2: h_ordering 2, 3, 4, 0, 1; data_index 3; roll_shift 15; par_round2: h_ordering 1, 2, 3, 4, 0; data_index 7; roll_shift 7; par_round2: h_ordering 0, 1, 2, 3, 4; data_index 0; roll_shift 12; par_round2: h_ordering 4, 0, 1, 2, 3; data_index 13; roll_shift 8; par_round2: h_ordering 3, 4, 0, 1, 2; data_index 5; roll_shift 9; par_round2: h_ordering 2, 3, 4, 0, 1; data_index 10; roll_shift 11; par_round2: h_ordering 1, 2, 3, 4, 0; data_index 14; roll_shift 7; par_round2: h_ordering 0, 1, 2, 3, 4; data_index 15; roll_shift 7; par_round2: h_ordering 4, 0, 1, 2, 3; data_index 8; roll_shift 12; par_round2: h_ordering 3, 4, 0, 1, 2; data_index 12; roll_shift 7; par_round2: h_ordering 2, 3, 4, 0, 1; data_index 4; roll_shift 6; par_round2: h_ordering 1, 2, 3, 4, 0; data_index 9; roll_shift 15; par_round2: h_ordering 0, 1, 2, 3, 4; data_index 1; roll_shift 13; par_round2: h_ordering 4, 0, 1, 2, 3; data_index 2; roll_shift 11; // Parallel Round 3 par_round3: h_ordering 3, 4, 0, 1, 2; data_index 15; roll_shift 9; par_round3: h_ordering 2, 3, 4, 0, 1; data_index 5; roll_shift 7; par_round3: h_ordering 1, 2, 3, 4, 0; data_index 1; roll_shift 15; par_round3: h_ordering 0, 1, 2, 3, 4; data_index 3; roll_shift 11; par_round3: h_ordering 4, 0, 1, 2, 3; data_index 7; roll_shift 8; par_round3: h_ordering 3, 4, 0, 1, 2; data_index 14; roll_shift 6; par_round3: h_ordering 2, 3, 4, 0, 1; data_index 6; roll_shift 6; par_round3: h_ordering 1, 2, 3, 4, 0; data_index 9; roll_shift 14; par_round3: h_ordering 0, 1, 2, 3, 4; data_index 11; roll_shift 12; par_round3: h_ordering 4, 0, 1, 2, 3; data_index 8; roll_shift 13; par_round3: h_ordering 3, 4, 0, 1, 2; data_index 12; roll_shift 5; par_round3: h_ordering 2, 3, 4, 0, 1; data_index 2; roll_shift 14; par_round3: h_ordering 1, 2, 3, 4, 0; data_index 10; roll_shift 13; par_round3: h_ordering 0, 1, 2, 3, 4; data_index 0; roll_shift 13; par_round3: h_ordering 4, 0, 1, 2, 3; data_index 4; roll_shift 7; par_round3: h_ordering 3, 4, 0, 1, 2; data_index 13; roll_shift 5; // Parallel Round 4 par_round4: h_ordering 2, 3, 4, 0, 1; data_index 8; roll_shift 15; par_round4: h_ordering 1, 2, 3, 4, 0; data_index 6; roll_shift 5; par_round4: h_ordering 0, 1, 2, 3, 4; data_index 4; roll_shift 8; par_round4: h_ordering 4, 0, 1, 2, 3; data_index 1; roll_shift 11; par_round4: h_ordering 3, 4, 0, 1, 2; data_index 3; roll_shift 14; par_round4: h_ordering 2, 3, 4, 0, 1; data_index 11; roll_shift 14; par_round4: h_ordering 1, 2, 3, 4, 0; data_index 15; roll_shift 6; par_round4: h_ordering 0, 1, 2, 3, 4; data_index 0; roll_shift 14; par_round4: h_ordering 4, 0, 1, 2, 3; data_index 5; roll_shift 6; par_round4: h_ordering 3, 4, 0, 1, 2; data_index 12; roll_shift 9; par_round4: h_ordering 2, 3, 4, 0, 1; data_index 2; roll_shift 12; par_round4: h_ordering 1, 2, 3, 4, 0; data_index 13; roll_shift 9; par_round4: h_ordering 0, 1, 2, 3, 4; data_index 9; roll_shift 12; par_round4: h_ordering 4, 0, 1, 2, 3; data_index 7; roll_shift 5; par_round4: h_ordering 3, 4, 0, 1, 2; data_index 10; roll_shift 15; par_round4: h_ordering 2, 3, 4, 0, 1; data_index 14; roll_shift 8; // Parallel Round 5 par_round5: h_ordering 1, 2, 3, 4, 0; data_index 12; roll_shift 8; par_round5: h_ordering 0, 1, 2, 3, 4; data_index 15; roll_shift 5; par_round5: h_ordering 4, 0, 1, 2, 3; data_index 10; roll_shift 12; par_round5: h_ordering 3, 4, 0, 1, 2; data_index 4; roll_shift 9; par_round5: h_ordering 2, 3, 4, 0, 1; data_index 1; roll_shift 12; par_round5: h_ordering 1, 2, 3, 4, 0; data_index 5; roll_shift 5; par_round5: h_ordering 0, 1, 2, 3, 4; data_index 8; roll_shift 14; par_round5: h_ordering 4, 0, 1, 2, 3; data_index 7; roll_shift 6; par_round5: h_ordering 3, 4, 0, 1, 2; data_index 6; roll_shift 8; par_round5: h_ordering 2, 3, 4, 0, 1; data_index 2; roll_shift 13; par_round5: h_ordering 1, 2, 3, 4, 0; data_index 13; roll_shift 6; par_round5: h_ordering 0, 1, 2, 3, 4; data_index 14; roll_shift 5; par_round5: h_ordering 4, 0, 1, 2, 3; data_index 0; roll_shift 15; par_round5: h_ordering 3, 4, 0, 1, 2; data_index 3; roll_shift 13; par_round5: h_ordering 2, 3, 4, 0, 1; data_index 9; roll_shift 11; par_round5: h_ordering 1, 2, 3, 4, 0; data_index 11; roll_shift 11; ); } } #[cfg(test)] mod tests { #[test] #[cfg(feature = "alloc")] fn test() { use crate::{Hash, HashEngine, ripemd160}; #[derive(Clone)] struct Test { input: &'static str, output: Vec, output_str: &'static str, } let tests = vec![ // Test messages from FIPS 180-1 Test { input: "abc", output: vec![ 0x8e, 0xb2, 0x08, 0xf7, 0xe0, 0x5d, 0x98, 0x7a, 0x9b, 0x04, 0x4a, 0x8e, 0x98, 0xc6, 0xb0, 0x87, 0xf1, 0x5a, 0x0b, 0xfc, ], output_str: "8eb208f7e05d987a9b044a8e98c6b087f15a0bfc" }, Test { input: "abcdbcdecdefdefgefghfghighijhijkijkljklmklmnlmnomnopnopq", output: vec![ 0x12, 0xa0, 0x53, 0x38, 0x4a, 0x9c, 0x0c, 0x88, 0xe4, 0x05, 0xa0, 0x6c, 0x27, 0xdc, 0xf4, 0x9a, 0xda, 0x62, 0xeb, 0x2b, ], output_str: "12a053384a9c0c88e405a06c27dcf49ada62eb2b" }, // Examples from wikipedia Test { input: "The quick brown fox jumps over the lazy dog", output: vec![ 0x37, 0xf3, 0x32, 0xf6, 0x8d, 0xb7, 0x7b, 0xd9, 0xd7, 0xed, 0xd4, 0x96, 0x95, 0x71, 0xad, 0x67, 0x1c, 0xf9, 0xdd, 0x3b, ], output_str: "37f332f68db77bd9d7edd4969571ad671cf9dd3b", }, Test { input: "The quick brown fox jumps over the lazy cog", output: vec![ 0x13, 0x20, 0x72, 0xdf, 0x69, 0x09, 0x33, 0x83, 0x5e, 0xb8, 0xb6, 0xad, 0x0b, 0x77, 0xe7, 0xb6, 0xf1, 0x4a, 0xca, 0xd7, ], output_str: "132072df690933835eb8b6ad0b77e7b6f14acad7", }, ]; for test in tests { // Hash through high-level API, check hex encoding/decoding let hash = ripemd160::Hash::hash(test.input.as_bytes()); assert_eq!(hash, test.output_str.parse::().expect("parse hex")); assert_eq!(&hash[..], &test.output[..]); assert_eq!(&hash.to_string(), &test.output_str); // Hash through engine, checking that we can input byte by byte let mut engine = ripemd160::Hash::engine(); for ch in test.input.as_bytes() { engine.input(&[*ch]); } let manual_hash = ripemd160::Hash::from_engine(engine); assert_eq!(hash, manual_hash); assert_eq!(hash.into_inner()[..].as_ref(), test.output.as_slice()); } } #[cfg(feature = "serde")] #[test] fn ripemd_serde() { use serde_test::{Configure, Token, assert_tokens}; use crate::{ripemd160, Hash}; static HASH_BYTES: [u8; 20] = [ 0x13, 0x20, 0x72, 0xdf, 0x69, 0x09, 0x33, 0x83, 0x5e, 0xb8, 0xb6, 0xad, 0x0b, 0x77, 0xe7, 0xb6, 0xf1, 0x4a, 0xca, 0xd7, ]; let hash = ripemd160::Hash::from_slice(&HASH_BYTES).expect("right number of bytes"); assert_tokens(&hash.compact(), &[Token::BorrowedBytes(&HASH_BYTES[..])]); assert_tokens(&hash.readable(), &[Token::Str("132072df690933835eb8b6ad0b77e7b6f14acad7")]); } } #[cfg(bench)] mod benches { use test::Bencher; use crate::{Hash, HashEngine, ripemd160}; #[bench] pub fn ripemd160_10(bh: &mut Bencher) { let mut engine = ripemd160::Hash::engine(); let bytes = [1u8; 10]; bh.iter( || { engine.input(&bytes); }); bh.bytes = bytes.len() as u64; } #[bench] pub fn ripemd160_1k(bh: &mut Bencher) { let mut engine = ripemd160::Hash::engine(); let bytes = [1u8; 1024]; bh.iter( || { engine.input(&bytes); }); bh.bytes = bytes.len() as u64; } #[bench] pub fn ripemd160_64k(bh: &mut Bencher) { let mut engine = ripemd160::Hash::engine(); let bytes = [1u8; 65536]; bh.iter( || { engine.input(&bytes); }); bh.bytes = bytes.len() as u64; } }