// Copyright 2026 International Digital Economy Academy
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
///|
let find_max_short_pattern : Int = 32
///|
let find_max_bruteforce : Int = 64
///|
let find_prime_rk : UInt = 16777619U
///|
/// Returns the offset of the first occurrence of the given bytes substring.
///
/// If the substring is not found, `None` is returned.
pub fn BytesView::find(target : BytesView, pattern : BytesView) -> Int? {
find_index(target, pattern)
}
///|
/// Returns the offset of the first occurrence of the given bytes substring.
///
/// If the substring is not found, `None` is returned.
pub fn Bytes::find(target : Bytes, pattern : BytesView) -> Int? {
target[:].find(pattern)
}
///|
// Handle trivial cases first, use the short-pattern scanner when the pattern is
// small, then use the guarded long-pattern scanner.
fn find_index(target : BytesView, pattern : BytesView) -> Int? {
let target_len = target.length()
let pattern_len = pattern.length()
if pattern_len == 0 {
return Some(0)
}
if pattern_len > target_len {
return None
}
if pattern_len == 1 {
let found = find_byte_from_view(
target,
0,
target_len,
pattern.unsafe_get(0),
)
if found < 0 {
None
} else {
Some(found)
}
} else if pattern_len == target_len {
if target == pattern {
Some(0)
} else {
None
}
} else if pattern_len <= find_max_short_pattern {
find_short(target, pattern)
} else {
find_long_by_byte_scanner(target, pattern)
}
}
///|
// Number of failed first-byte candidates tolerated before the long scanner
// switches to Rabin-Karp. This keeps dense false positives linear-time.
#inline
fn find_cutover(i : Int) -> Int {
4 + i / 16
}
///|
/// Finds the first occurrence of `byte` in the view-relative range
/// `[start, end)`.
///
/// The caller must ensure `0 <= start <= end <= target.length()`. The result is
/// relative to `target`, or `-1` when `byte` does not occur in the range.
///
/// ```mbt check
/// test {
/// let target = b"--abcabc--"[2:8]
/// debug_inspect(target[2:].find(b"b"), content="Some(2)")
/// debug_inspect(target.find(b"x"), content="None")
/// }
/// ```
declare fn find_byte_from_view(
target : BytesView,
start : Int,
end : Int,
byte : Byte,
) -> Int
///|
// Scalar byte scanner for backends without direct linear-memory vector loads.
#cfg(any(target="js", target="wasm-gc"))
fn find_byte_from_view(
target : BytesView,
start : Int,
end : Int,
byte : Byte,
) -> Int {
for pos in start.. Int {
let target_start = target.start_offset()
let found = find_byte_from_bytes(
target.data(),
target_start + start,
target_start + end,
byte,
)
if found < 0 {
-1
} else {
found - target_start
}
}
///|
/// Finds the last occurrence of `byte` in the view-relative range
/// `[start, end)`.
///
/// The caller must ensure `0 <= start <= end <= target.length()`. The result is
/// relative to `target`, or `-1` when `byte` does not occur in the range.
///
/// ```mbt check
/// test {
/// let target = b"--abcabc--"[2:8]
/// debug_inspect(target[:4].rev_find(b"b"), content="Some(1)")
/// debug_inspect(target.rev_find(b"x"), content="None")
/// }
/// ```
declare fn rev_find_byte_from_view(
target : BytesView,
start : Int,
end : Int,
byte : Byte,
) -> Int
///|
// Scalar reverse byte scanner for backends without direct linear-memory vector
// loads.
#cfg(any(target="js", target="wasm-gc"))
fn rev_find_byte_from_view(
target : BytesView,
start : Int,
end : Int,
byte : Byte,
) -> Int {
for pos = end - 1; pos >= start; {
if target.unsafe_get(pos) == byte {
break pos
}
continue pos - 1
} nobreak {
-1
}
}
///|
// Convert view-relative bounds to backing `Bytes` offsets before reverse
// scanning.
#cfg(any(target="native", target="wasm"))
fn rev_find_byte_from_view(
target : BytesView,
start : Int,
end : Int,
byte : Byte,
) -> Int {
let target_start = target.start_offset()
let found = rev_find_byte_from_bytes(
target.data(),
target_start + start,
target_start + end,
byte,
)
if found < 0 {
-1
} else {
found - target_start
}
}
///|
#cfg(any(target="native", target="wasm"))
fn Bytes::v128_load(self : Bytes, offset : Int) -> V128 {
v128_load(unsafe_from_bytes(self), offset)
}
///|
// SIMD byte scanner for linear-memory backends. The vector loop scans 16 bytes
// at a time, then the scalar tail handles the remaining bytes.
//
// Returns the first index of `byte` in `data[start..end)`, or -1 if absent.
#cfg(any(target="native", target="wasm"))
fn find_byte_from_bytes(
data : Bytes,
start : Int,
end : Int,
byte : Byte,
) -> Int {
// Broadcast the needle to all 16 lanes so a single vector compare tests 16
// haystack bytes at once:
// byte_v = [byte, byte, byte, ..., byte] (lanes 15..0)
let byte_v = i8x16_splat(byte)
// Vector loop. `for` is an expression here: `tail_start` is bound to the
// value the loop yields. It yields in one of two ways:
// - `return` on a hit, exiting the whole function; or
// - falling through the `pos + 16 <= end` guard (a full chunk no longer
// fits), running `nobreak` whose `pos` becomes `tail_start`.
let tail_start = for pos = start; pos + 16 <= end; {
// Load 16 bytes, compare lanewise (0xFF on match / 0x00 on miss), then
// gather each lane's high bit into a 16-bit integer: bit i set <=> lane i
// matched.
//
// chunk : .. 62 72 61 77 6E 20 61 78 20 (searching 0x61 'a')
// eq : .. 00 00 FF 00 00 00 FF 00 00
// mask : 0b...0100_0100 = 0x0044 (bits 2 and 6 set)
let mask = i8x16_bitmask(i8x16_eq(data.v128_load(pos), byte_v))
if mask != 0 {
// ctz = index of the lowest set bit = first matching lane, which yields
// first-occurrence order. 0x0044 -> ctz 2 -> absolute index `pos + 2`.
return pos + mask.ctz()
}
continue pos + 16
} nobreak {
pos
}
// Scalar tail: fewer than 16 bytes remain, so scan them one at a time.
for pos in tail_start.. Int {
let byte_v = i8x16_splat(byte)
let head_end = for pos = end; pos - 16 >= start; {
let chunk_start = pos - 16
let mask = i8x16_bitmask(i8x16_eq(data.v128_load(chunk_start), byte_v))
if mask != 0 {
return chunk_start + 31 - mask.clz()
}
continue chunk_start
} nobreak {
pos
}
for pos = head_end - 1; pos >= start; {
if data.unsafe_get(pos) == byte {
break pos
}
continue pos - 1
} nobreak {
-1
}
}
///|
/// Finds the first occurrence of a short, non-empty `pattern` in `target`.
///
/// The caller must ensure the pattern has at least two bytes, does not exceed
/// the short-pattern limit, and is no longer than `target`.
///
/// ```mbt check
/// test {
/// let target = b"--abcabc--"[2:8]
/// debug_inspect(target.find(b"bca"), content="Some(1)")
/// debug_inspect(target.find(b"abd"), content="None")
/// }
/// ```
declare fn find_short(target : BytesView, pattern : BytesView) -> Int?
///|
// Scalar short-pattern scanner: find matches for the last byte, then verify
// the unchecked prefix range for each candidate.
#cfg(any(target="js", target="wasm-gc"))
fn find_short(target : BytesView, pattern : BytesView) -> Int? {
let target_len = target.length()
let pattern_len = pattern.length()
let target_start = target.start_offset()
let pattern_start = pattern.start_offset()
let last_offset = pattern_len - 1
let last_byte = pattern.unsafe_get(last_offset)
for anchor_pos = last_offset; anchor_pos < target_len; {
let found = find_byte_from_view(target, anchor_pos, target_len, last_byte)
if found < 0 {
break None
}
// The last byte already matched; verify the prefix range only.
let candidate = found - last_offset
if target
.data()
.unsafe_range_equal(
pattern.data(),
self_off=target_start + candidate,
other_off=pattern_start,
len=last_offset,
) {
break Some(candidate)
}
continue found + 1
} nobreak {
None
}
}
///|
// Native/wasm short-needle path: use the SIMD prefilter to skip most false
// candidates, then compare the unchecked middle range only on hits.
#cfg(any(target="native", target="wasm"))
fn find_short(target : BytesView, pattern : BytesView) -> Int? {
let target_len = target.length()
let pattern_len = pattern.length()
let last = target_len - pattern_len
let target_start = target.start_offset()
let pattern_start = pattern.start_offset()
let candidate_end = target_start + last + 1
let first = pattern.unsafe_get(0)
let last_offset = pattern_len - 1
let last_byte = pattern.unsafe_get(last_offset)
// First and last bytes are matched by the prefilter; only the middle range
// needs verification. `len=0` (two-byte patterns) trivially compares equal.
let middle_len = last_offset - 1
for pos = 0; pos <= last; {
let found = find_short_candidate_from_bytes(
target.data(),
target_start + pos,
candidate_end,
first,
last_offset,
last_byte,
)
if found < 0 {
break None
}
let candidate = found - target_start
if target
.data()
.unsafe_range_equal(
pattern.data(),
self_off=found + 1,
other_off=pattern_start + 1,
len=middle_len,
) {
break Some(candidate)
}
continue candidate + 1
} nobreak {
None
}
}
///|
// Scalar reverse candidate scanner: match first and last bytes before scalar
// verification of the unchecked middle range.
#cfg(any(target="js", target="wasm-gc"))
fn rev_find_short_candidate_from_view(
target : BytesView,
start : Int,
candidate_end : Int,
first : Byte,
last_offset : Int,
last : Byte,
) -> Int {
for pos = candidate_end - 1; pos >= start; {
if target.unsafe_get(pos) == first &&
target.unsafe_get(pos + last_offset) == last {
break pos
}
continue pos - 1
} nobreak {
-1
}
}
///|
// SIMD short-pattern prefilter: match first and last bytes before scalar
// verification of the unchecked middle range.
#cfg(any(target="native", target="wasm"))
fn find_short_candidate_from_bytes(
data : Bytes,
start : Int,
candidate_end : Int,
first : Byte,
last_offset : Int,
last : Byte,
) -> Int {
let first_v = i8x16_splat(first)
let last_v = i8x16_splat(last)
let tail_start = for pos = start; pos + 16 <= candidate_end; {
let mask = v128_and(
i8x16_eq(data.v128_load(pos), first_v),
i8x16_eq(data.v128_load(pos + last_offset), last_v),
)
let bits = i8x16_bitmask(mask)
if bits != 0 {
return pos + bits.ctz()
}
continue pos + 16
} nobreak {
pos
}
for pos in tail_start.. Int {
let first_v = i8x16_splat(first)
let last_v = i8x16_splat(last)
let head_end = for pos = candidate_end; pos - 16 >= start; {
let chunk_start = pos - 16
let mask = v128_and(
i8x16_eq(data.v128_load(chunk_start), first_v),
i8x16_eq(data.v128_load(chunk_start + last_offset), last_v),
)
let bits = i8x16_bitmask(mask)
if bits != 0 {
return chunk_start + 31 - bits.clz()
}
continue chunk_start
} nobreak {
pos
}
for pos = head_end - 1; pos >= start; {
if data.unsafe_get(pos) == first &&
data.unsafe_get(pos + last_offset) == last {
break pos
}
continue pos - 1
} nobreak {
-1
}
}
///|
/// Finds the last occurrence of a non-empty, multi-byte `pattern` in `target`.
///
/// The caller must ensure the pattern has at least two bytes and is shorter
/// than `target`.
///
/// ```mbt check
/// test {
/// let target = b"--abcabc--"[2:8]
/// debug_inspect(target.rev_find(b"abc"), content="Some(3)")
/// debug_inspect(target.rev_find(b"abd"), content="None")
/// }
/// ```
declare fn rev_find_by_byte_scanner(
target : BytesView,
pattern : BytesView,
) -> Int?
///|
// Scalar reverse scanner for all multi-byte patterns. It searches candidate
// positions by first/last byte and verifies only the middle range on hits.
#cfg(any(target="js", target="wasm-gc"))
fn rev_find_by_byte_scanner(target : BytesView, pattern : BytesView) -> Int? {
let target_len = target.length()
let pattern_len = pattern.length()
let last = target_len - pattern_len
let target_start = target.start_offset()
let pattern_start = pattern.start_offset()
let first = pattern.unsafe_get(0)
let last_offset = pattern_len - 1
let last_byte = pattern.unsafe_get(last_offset)
// First and last bytes are matched by the candidate scanner; only the middle
// range needs verification. `len=0` (two-byte patterns) trivially compares
// equal.
let middle_len = last_offset - 1
for candidate_end = last + 1; candidate_end > 0; {
let found = rev_find_short_candidate_from_view(
target, 0, candidate_end, first, last_offset, last_byte,
)
if found < 0 {
break None
}
if target
.data()
.unsafe_range_equal(
pattern.data(),
self_off=target_start + found + 1,
other_off=pattern_start + 1,
len=middle_len,
) {
break Some(found)
}
continue found
} nobreak {
None
}
}
///|
// Native/wasm reverse scanner for all multi-byte patterns. It uses the SIMD
// first/last byte prefilter and verifies only the middle range on hits.
#cfg(any(target="native", target="wasm"))
fn rev_find_by_byte_scanner(target : BytesView, pattern : BytesView) -> Int? {
let target_len = target.length()
let pattern_len = pattern.length()
let last = target_len - pattern_len
let target_start = target.start_offset()
let pattern_start = pattern.start_offset()
let first = pattern.unsafe_get(0)
let last_offset = pattern_len - 1
let last_byte = pattern.unsafe_get(last_offset)
// First and last bytes are matched by the prefilter; only the middle range
// needs verification. `len=0` (two-byte patterns) trivially compares equal.
let middle_len = last_offset - 1
for candidate_end = last + 1; candidate_end > 0; {
let found = rev_find_short_candidate_from_bytes(
target.data(),
target_start,
target_start + candidate_end,
first,
last_offset,
last_byte,
)
if found < 0 {
break None
}
let candidate = found - target_start
if target
.data()
.unsafe_range_equal(
pattern.data(),
self_off=found + 1,
other_off=pattern_start + 1,
len=middle_len,
) {
break Some(candidate)
}
continue candidate
} nobreak {
None
}
}
///|
// Linear-time fallback for long needles. Hash matches are verified to avoid
// returning false positives from collisions.
fn find_rabin_karp_from(
target : BytesView,
pattern : BytesView,
start : Int,
) -> Int? {
fn hash(bytes : BytesView, start : Int, length : Int) -> UInt {
for i in 0.. UInt {
for pow = 1U, square = find_prime_rk, exp = length; exp > 0; {
let next_pow = if exp % 2 == 1 { pow * square } else { pow }
continue next_pow, square * square, exp / 2
} nobreak {
pow
}
}
let pattern_len = pattern.length()
let target_len = target.length()
if start + pattern_len > target_len {
return None
}
let target_start = target.start_offset()
let pattern_start = pattern.start_offset()
let pattern_hash = hash(pattern, 0, pattern_len)
let pow = prime_pow(pattern_len)
let hash = hash(target, start, pattern_len)
if hash == pattern_hash &&
target
.data()
.unsafe_range_equal(
pattern.data(),
self_off=target_start + start,
other_off=pattern_start,
len=pattern_len,
) {
return Some(start)
}
for index = start + pattern_len, hash = hash; index < target_len; {
let hash = hash * find_prime_rk + target.unsafe_get(index).to_uint()
let hash = hash - pow * target.unsafe_get(index - pattern_len).to_uint()
let candidate = index - pattern_len + 1
if hash == pattern_hash &&
target
.data()
.unsafe_range_equal(
pattern.data(),
self_off=target_start + candidate,
other_off=pattern_start,
len=pattern_len,
) {
break Some(candidate)
}
continue index + 1, hash
} nobreak {
None
}
}
///|
// Long-needle scanner: start with first-byte candidates for common fast cases,
// but switch to Rabin-Karp after enough failed verifications.
fn find_long_by_byte_scanner(target : BytesView, pattern : BytesView) -> Int? {
let target_len = target.length()
let pattern_len = pattern.length()
let last = target_len - pattern_len
let target_start = target.start_offset()
let pattern_start = pattern.start_offset()
let first_byte = pattern.unsafe_get(0)
for pos = 0, failures = 0; pos <= last; {
let found = find_byte_from_view(target, pos, target_len, first_byte)
if found < 0 || found > last {
break None
}
// The first byte already matched; verify the remaining suffix range.
if target
.data()
.unsafe_range_equal(
pattern.data(),
self_off=target_start + found + 1,
other_off=pattern_start + 1,
len=pattern_len - 1,
) {
break Some(found)
}
let failures = failures + 1
if failures > find_max_bruteforce || failures > find_cutover(found) {
break find_rabin_karp_from(target, pattern, found + 1)
}
continue found + 1, failures
} nobreak {
None
}
}
///|
/// Returns the offset of the last occurrence of the given
/// bytes substring. If the substring is not found, `None` is returned.
pub fn BytesView::rev_find(target : BytesView, pattern : BytesView) -> Int? {
let target_len = target.length()
let pattern_len = pattern.length()
if pattern_len == 0 {
return Some(target_len)
}
if pattern_len > target_len {
return None
}
if pattern_len == 1 {
let found = rev_find_byte_from_view(
target,
0,
target_len,
pattern.unsafe_get(0),
)
if found < 0 {
None
} else {
Some(found)
}
} else if pattern_len == target_len {
if target == pattern {
Some(0)
} else {
None
}
} else {
rev_find_by_byte_scanner(target, pattern)
}
}
///|
/// Returns the offset of the last occurrence of the given
/// bytes substring. If the substring is not found, `None` is returned.
pub fn Bytes::rev_find(target : Bytes, pattern : BytesView) -> Int? {
target[:].rev_find(pattern)
}
///|
/// Returns true if this bytes view starts with the given prefix.
pub fn BytesView::has_prefix(self : BytesView, prefix : BytesView) -> Bool {
let prefix_len = prefix.length()
self.length() >= prefix_len && self[:prefix_len] == prefix
}
///|
/// Returns true if this bytes starts with the given prefix.
pub fn Bytes::has_prefix(self : Bytes, prefix : BytesView) -> Bool {
self[:].has_prefix(prefix)
}
///|
/// Returns true if this bytes view ends with the given suffix.
pub fn BytesView::has_suffix(self : BytesView, suffix : BytesView) -> Bool {
let self_len = self.length()
let suffix_len = suffix.length()
self_len >= suffix_len && self[self_len - suffix_len:] == suffix
}
///|
/// Returns true if this bytes ends with the given suffix.
pub fn Bytes::has_suffix(self : Bytes, suffix : BytesView) -> Bool {
self[:].has_suffix(suffix)
}
///|
/// Removes the given prefix from the view if it exists.
///
/// Returns `Some(suffix)` if the view starts with the given prefix.
/// Returns `None` otherwise.
pub fn BytesView::chop_prefix(
self : BytesView,
prefix : BytesView,
) -> BytesView? {
let prefix_len = prefix.length()
if self.length() >= prefix_len && self[:prefix_len] == prefix {
Some(self[prefix_len:])
} else {
None
}
}
///|
/// Removes the given prefix from the bytes if it exists.
pub fn Bytes::chop_prefix(self : Bytes, prefix : BytesView) -> BytesView? {
self[:].chop_prefix(prefix)
}
///|
/// Removes the given suffix from the view if it exists.
///
/// Returns `Some(prefix)` if the view ends with the given suffix.
/// Returns `None` otherwise.
pub fn BytesView::chop_suffix(
self : BytesView,
suffix : BytesView,
) -> BytesView? {
let self_len = self.length()
let suffix_len = suffix.length()
if self_len >= suffix_len && self[self_len - suffix_len:] == suffix {
Some(self[:self_len - suffix_len])
} else {
None
}
}
///|
/// Removes the given suffix from the bytes if it exists.
pub fn Bytes::chop_suffix(self : Bytes, suffix : BytesView) -> BytesView? {
self[:].chop_suffix(suffix)
}