use crate::internal::{Err, IResult, Needed};
use crate::error::{ParseError, ErrorKind};
use crate::lib::std::ops::{Range, RangeFrom, RangeFull, RangeTo};
use crate::lib::std::iter::Enumerate;
use crate::lib::std::slice::Iter;
use crate::lib::std::iter::Map;
use crate::lib::std::str::Chars;
use crate::lib::std::str::CharIndices;
use crate::lib::std::str::FromStr;
use crate::lib::std::str::from_utf8;
use memchr;
#[cfg(feature = "alloc")]
use crate::lib::std::string::String;
#[cfg(feature = "alloc")]
use crate::lib::std::vec::Vec;
pub trait InputLength {
#[inline]
fn input_len(&self) -> usize;
}
impl<'a, T> InputLength for &'a [T] {
#[inline]
fn input_len(&self) -> usize {
self.len()
}
}
impl<'a> InputLength for &'a str {
#[inline]
fn input_len(&self) -> usize {
self.len()
}
}
impl<'a> InputLength for (&'a [u8], usize) {
#[inline]
fn input_len(&self) -> usize {
self.0.len() * 8 - self.1
}
}
pub trait Offset {
fn offset(&self, second: &Self) -> usize;
}
impl Offset for [u8] {
fn offset(&self, second: &Self) -> usize {
let fst = self.as_ptr();
let snd = second.as_ptr();
snd as usize - fst as usize
}
}
impl<'a> Offset for &'a [u8] {
fn offset(&self, second: &Self) -> usize {
let fst = self.as_ptr();
let snd = second.as_ptr();
snd as usize - fst as usize
}
}
impl Offset for str {
fn offset(&self, second: &Self) -> usize {
let fst = self.as_ptr();
let snd = second.as_ptr();
snd as usize - fst as usize
}
}
impl<'a> Offset for &'a str {
fn offset(&self, second: &Self) -> usize {
let fst = self.as_ptr();
let snd = second.as_ptr();
snd as usize - fst as usize
}
}
pub trait AsBytes {
fn as_bytes(&self) -> &[u8];
}
impl<'a> AsBytes for &'a str {
#[inline(always)]
fn as_bytes(&self) -> &[u8] {
<str as AsBytes>::as_bytes(self)
}
}
impl AsBytes for str {
#[inline(always)]
fn as_bytes(&self) -> &[u8] {
self.as_ref()
}
}
impl<'a> AsBytes for &'a [u8] {
#[inline(always)]
fn as_bytes(&self) -> &[u8] {
*self
}
}
impl AsBytes for [u8] {
#[inline(always)]
fn as_bytes(&self) -> &[u8] {
self
}
}
macro_rules! as_bytes_array_impls {
($($N:expr)+) => {
$(
impl<'a> AsBytes for &'a [u8; $N] {
#[inline(always)]
fn as_bytes(&self) -> &[u8] {
*self
}
}
impl AsBytes for [u8; $N] {
#[inline(always)]
fn as_bytes(&self) -> &[u8] {
self
}
}
)+
};
}
as_bytes_array_impls! {
0 1 2 3 4 5 6 7 8 9
10 11 12 13 14 15 16 17 18 19
20 21 22 23 24 25 26 27 28 29
30 31 32
}
pub trait AsChar {
#[inline]
fn as_char(self) -> char;
#[inline]
fn is_alpha(self) -> bool;
#[inline]
fn is_alphanum(self) -> bool;
#[inline]
fn is_dec_digit(self) -> bool;
#[inline]
fn is_hex_digit(self) -> bool;
#[inline]
fn is_oct_digit(self) -> bool;
#[inline]
fn len(self) -> usize;
}
impl AsChar for u8 {
#[inline]
fn as_char(self) -> char {
self as char
}
#[inline]
fn is_alpha(self) -> bool {
(self >= 0x41 && self <= 0x5A) || (self >= 0x61 && self <= 0x7A)
}
#[inline]
fn is_alphanum(self) -> bool {
self.is_alpha() || self.is_dec_digit()
}
#[inline]
fn is_dec_digit(self) -> bool {
self >= 0x30 && self <= 0x39
}
#[inline]
fn is_hex_digit(self) -> bool {
(self >= 0x30 && self <= 0x39) || (self >= 0x41 && self <= 0x46) || (self >= 0x61 && self <= 0x66)
}
#[inline]
fn is_oct_digit(self) -> bool {
self >= 0x30 && self <= 0x37
}
#[inline]
fn len(self) -> usize {
1
}
}
impl<'a> AsChar for &'a u8 {
#[inline]
fn as_char(self) -> char {
*self as char
}
#[inline]
fn is_alpha(self) -> bool {
(*self >= 0x41 && *self <= 0x5A) || (*self >= 0x61 && *self <= 0x7A)
}
#[inline]
fn is_alphanum(self) -> bool {
self.is_alpha() || self.is_dec_digit()
}
#[inline]
fn is_dec_digit(self) -> bool {
*self >= 0x30 && *self <= 0x39
}
#[inline]
fn is_hex_digit(self) -> bool {
(*self >= 0x30 && *self <= 0x39) || (*self >= 0x41 && *self <= 0x46) || (*self >= 0x61 && *self <= 0x66)
}
#[inline]
fn is_oct_digit(self) -> bool {
*self >= 0x30 && *self <= 0x37
}
#[inline]
fn len(self) -> usize {
1
}
}
impl AsChar for char {
#[inline]
fn as_char(self) -> char {
self
}
#[inline]
fn is_alpha(self) -> bool {
self.is_ascii_alphabetic()
}
#[inline]
fn is_alphanum(self) -> bool {
self.is_alpha() || self.is_dec_digit()
}
#[inline]
fn is_dec_digit(self) -> bool {
self.is_ascii_digit()
}
#[inline]
fn is_hex_digit(self) -> bool {
self.is_ascii_hexdigit()
}
#[inline]
fn is_oct_digit(self) -> bool {
self.is_digit(8)
}
#[inline]
fn len(self) -> usize {
self.len_utf8()
}
}
impl<'a> AsChar for &'a char {
#[inline]
fn as_char(self) -> char {
*self
}
#[inline]
fn is_alpha(self) -> bool {
self.is_ascii_alphabetic()
}
#[inline]
fn is_alphanum(self) -> bool {
self.is_alpha() || self.is_dec_digit()
}
#[inline]
fn is_dec_digit(self) -> bool {
self.is_ascii_digit()
}
#[inline]
fn is_hex_digit(self) -> bool {
self.is_ascii_hexdigit()
}
#[inline]
fn is_oct_digit(self) -> bool {
self.is_digit(8)
}
#[inline]
fn len(self) -> usize {
self.len_utf8()
}
}
pub trait InputIter {
type Item;
type Iter: Iterator<Item = (usize, Self::Item)>;
type IterElem: Iterator<Item = Self::Item>;
fn iter_indices(&self) -> Self::Iter;
fn iter_elements(&self) -> Self::IterElem;
fn position<P>(&self, predicate: P) -> Option<usize>
where
P: Fn(Self::Item) -> bool;
fn slice_index(&self, count: usize) -> Option<usize>;
}
pub trait InputTake: Sized {
fn take(&self, count: usize) -> Self;
fn take_split(&self, count: usize) -> (Self, Self);
}
fn star(r_u8: &u8) -> u8 {
*r_u8
}
impl<'a> InputIter for &'a [u8] {
type Item = u8;
type Iter = Enumerate<Self::IterElem>;
type IterElem = Map<Iter<'a, Self::Item>, fn(&u8) -> u8>;
#[inline]
fn iter_indices(&self) -> Self::Iter {
self.iter_elements().enumerate()
}
#[inline]
fn iter_elements(&self) -> Self::IterElem {
self.iter().map(star)
}
#[inline]
fn position<P>(&self, predicate: P) -> Option<usize>
where
P: Fn(Self::Item) -> bool,
{
self.iter().position(|b| predicate(*b))
}
#[inline]
fn slice_index(&self, count: usize) -> Option<usize> {
if self.len() >= count {
Some(count)
} else {
None
}
}
}
impl<'a> InputTake for &'a [u8] {
#[inline]
fn take(&self, count: usize) -> Self {
&self[0..count]
}
#[inline]
fn take_split(&self, count: usize) -> (Self, Self) {
let (prefix, suffix) = self.split_at(count);
(suffix, prefix)
}
}
impl<'a> InputIter for &'a str {
type Item = char;
type Iter = CharIndices<'a>;
type IterElem = Chars<'a>;
#[inline]
fn iter_indices(&self) -> Self::Iter {
self.char_indices()
}
#[inline]
fn iter_elements(&self) -> Self::IterElem {
self.chars()
}
fn position<P>(&self, predicate: P) -> Option<usize>
where
P: Fn(Self::Item) -> bool,
{
for (o, c) in self.char_indices() {
if predicate(c) {
return Some(o);
}
}
None
}
#[inline]
fn slice_index(&self, count: usize) -> Option<usize> {
let mut cnt = 0;
for (index, _) in self.char_indices() {
if cnt == count {
return Some(index);
}
cnt += 1;
}
if cnt == count {
return Some(self.len());
}
None
}
}
impl<'a> InputTake for &'a str {
#[inline]
fn take(&self, count: usize) -> Self {
&self[..count]
}
#[inline]
fn take_split(&self, count: usize) -> (Self, Self) {
(&self[count..], &self[..count])
}
}
pub trait UnspecializedInput {}
pub trait InputTakeAtPosition: Sized {
type Item;
fn split_at_position<P, E: ParseError<Self>>(&self, predicate: P) -> IResult<Self, Self, E>
where
P: Fn(Self::Item) -> bool;
fn split_at_position1<P, E: ParseError<Self>>(&self, predicate: P, e: ErrorKind) -> IResult<Self, Self, E>
where
P: Fn(Self::Item) -> bool;
fn split_at_position_complete<P, E: ParseError<Self>>(&self, predicate: P) -> IResult<Self, Self, E>
where
P: Fn(Self::Item) -> bool;
fn split_at_position1_complete<P, E: ParseError<Self>>(&self, predicate: P, e: ErrorKind) -> IResult<Self, Self, E>
where
P: Fn(Self::Item) -> bool;
}
impl<T: InputLength + InputIter + InputTake + Clone + UnspecializedInput> InputTakeAtPosition for T {
type Item = <T as InputIter>::Item;
fn split_at_position<P, E: ParseError<Self>>(&self, predicate: P) -> IResult<Self, Self, E>
where
P: Fn(Self::Item) -> bool,
{
match self.position(predicate) {
Some(n) => Ok(self.take_split(n)),
None => Err(Err::Incomplete(Needed::Size(1))),
}
}
fn split_at_position1<P, E: ParseError<Self>>(&self, predicate: P, e: ErrorKind) -> IResult<Self, Self, E>
where
P: Fn(Self::Item) -> bool,
{
match self.position(predicate) {
Some(0) => Err(Err::Error(E::from_error_kind(self.clone(), e))),
Some(n) => Ok(self.take_split(n)),
None => Err(Err::Incomplete(Needed::Size(1))),
}
}
fn split_at_position_complete<P, E: ParseError<Self>>(&self, predicate: P) -> IResult<Self, Self, E>
where P: Fn(Self::Item) -> bool {
match self.split_at_position(predicate) {
Err(Err::Incomplete(_)) => Ok(self.take_split(self.input_len())),
res => res,
}
}
fn split_at_position1_complete<P, E: ParseError<Self>>(&self, predicate: P, e: ErrorKind) -> IResult<Self, Self, E>
where P: Fn(Self::Item) -> bool {
match self.split_at_position1(predicate, e) {
Err(Err::Incomplete(_)) => if self.input_len() == 0 {
Err(Err::Error(E::from_error_kind(self.clone(), e)))
} else {
Ok(self.take_split(self.input_len()))
}
res => res,
}
}
}
impl<'a> InputTakeAtPosition for &'a [u8] {
type Item = u8;
fn split_at_position<P, E: ParseError<Self>>(&self, predicate: P) -> IResult<Self, Self, E>
where
P: Fn(Self::Item) -> bool,
{
match (0..self.len()).find(|b| predicate(self[*b])) {
Some(i) => Ok((&self[i..], &self[..i])),
None => Err(Err::Incomplete(Needed::Size(1))),
}
}
fn split_at_position1<P, E: ParseError<Self>>(&self, predicate: P, e: ErrorKind) -> IResult<Self, Self, E>
where
P: Fn(Self::Item) -> bool,
{
match (0..self.len()).find(|b| predicate(self[*b])) {
Some(0) => Err(Err::Error(E::from_error_kind(self, e))),
Some(i) => Ok((&self[i..], &self[..i])),
None => Err(Err::Incomplete(Needed::Size(1))),
}
}
fn split_at_position_complete<P, E: ParseError<Self>>(&self, predicate: P) -> IResult<Self, Self, E>
where P: Fn(Self::Item) -> bool {
match (0..self.len()).find(|b| predicate(self[*b])) {
Some(i) => Ok((&self[i..], &self[..i])),
None => Ok(self.take_split(self.input_len())),
}
}
fn split_at_position1_complete<P, E: ParseError<Self>>(&self, predicate: P, e: ErrorKind) -> IResult<Self, Self, E>
where P: Fn(Self::Item) -> bool {
match (0..self.len()).find(|b| predicate(self[*b])) {
Some(0) => Err(Err::Error(E::from_error_kind(self, e))),
Some(i) => Ok((&self[i..], &self[..i])),
None => {
if self.len() == 0 {
Err(Err::Error(E::from_error_kind(self, e)))
} else {
Ok(self.take_split(self.input_len()))
}
},
}
}
}
impl<'a> InputTakeAtPosition for &'a str {
type Item = char;
fn split_at_position<P, E: ParseError<Self>>(&self, predicate: P) -> IResult<Self, Self, E>
where
P: Fn(Self::Item) -> bool,
{
match self.find(predicate) {
Some(i) => Ok((&self[i..], &self[..i])),
None => Err(Err::Incomplete(Needed::Size(1))),
}
}
fn split_at_position1<P, E: ParseError<Self>>(&self, predicate: P, e: ErrorKind) -> IResult<Self, Self, E>
where
P: Fn(Self::Item) -> bool,
{
match self.find(predicate) {
Some(0) => Err(Err::Error(E::from_error_kind(self, e))),
Some(i) => Ok((&self[i..], &self[..i])),
None => Err(Err::Incomplete(Needed::Size(1))),
}
}
fn split_at_position_complete<P, E: ParseError<Self>>(&self, predicate: P) -> IResult<Self, Self, E>
where P: Fn(Self::Item) -> bool {
match self.find(predicate) {
Some(i) => Ok((&self[i..], &self[..i])),
None => Ok(self.take_split(self.input_len()))
}
}
fn split_at_position1_complete<P, E: ParseError<Self>>(&self, predicate: P, e: ErrorKind) -> IResult<Self, Self, E>
where P: Fn(Self::Item) -> bool {
match self.find(predicate) {
Some(0) => Err(Err::Error(E::from_error_kind(self, e))),
Some(i) => Ok((&self[i..], &self[..i])),
None => {
if self.len() == 0 {
Err(Err::Error(E::from_error_kind(self, e)))
} else {
Ok(self.take_split(self.input_len()))
}
},
}
}
}
#[derive(Debug, PartialEq)]
pub enum CompareResult {
Ok,
Incomplete,
Error,
}
pub trait Compare<T> {
fn compare(&self, t: T) -> CompareResult;
fn compare_no_case(&self, t: T) -> CompareResult;
}
impl<'a, 'b> Compare<&'b [u8]> for &'a [u8] {
#[inline(always)]
fn compare(&self, t: &'b [u8]) -> CompareResult {
let pos = self.iter().zip(t.iter()).position(|(a, b)| a != b);
match pos {
Some(_) => CompareResult::Error,
None => {
if self.len() >= t.len() {
CompareResult::Ok
} else {
CompareResult::Incomplete
}
}
}
}
#[inline(always)]
fn compare_no_case(&self, t: &'b [u8]) -> CompareResult {
let len = self.len();
let blen = t.len();
let m = if len < blen { len } else { blen };
let reduced = &self[..m];
let other = &t[..m];
if !reduced.iter().zip(other).all(|(a, b)| match (*a, *b) {
(0..=64, 0..=64) | (91..=96, 91..=96) | (123..=255, 123..=255) => a == b,
(65..=90, 65..=90) | (97..=122, 97..=122) | (65..=90, 97..=122) | (97..=122, 65..=90) => *a | 0b00_10_00_00 == *b | 0b00_10_00_00,
_ => false,
}) {
CompareResult::Error
} else if m < blen {
CompareResult::Incomplete
} else {
CompareResult::Ok
}
}
}
impl<'a, 'b> Compare<&'b str> for &'a [u8] {
#[inline(always)]
fn compare(&self, t: &'b str) -> CompareResult {
self.compare(AsBytes::as_bytes(t))
}
#[inline(always)]
fn compare_no_case(&self, t: &'b str) -> CompareResult {
self.compare_no_case(AsBytes::as_bytes(t))
}
}
impl<'a, 'b> Compare<&'b str> for &'a str {
#[inline(always)]
fn compare(&self, t: &'b str) -> CompareResult {
let pos = self.chars().zip(t.chars()).position(|(a, b)| a != b);
match pos {
Some(_) => CompareResult::Error,
None => {
if self.len() >= t.len() {
CompareResult::Ok
} else {
CompareResult::Incomplete
}
}
}
}
#[inline(always)]
fn compare_no_case(&self, t: &'b str) -> CompareResult {
let pos = self
.chars()
.zip(t.chars())
.position(|(a, b)| a.to_lowercase().zip(b.to_lowercase()).any(|(a, b)| a != b));
match pos {
Some(_) => CompareResult::Error,
None => {
if self.len() >= t.len() {
CompareResult::Ok
} else {
CompareResult::Incomplete
}
}
}
}
}
pub trait FindToken<T> {
fn find_token(&self, token: T) -> bool;
}
impl<'a> FindToken<u8> for &'a [u8] {
fn find_token(&self, token: u8) -> bool {
memchr::memchr(token, self).is_some()
}
}
impl<'a> FindToken<u8> for &'a str {
fn find_token(&self, token: u8) -> bool {
self.as_bytes().find_token(token)
}
}
impl<'a, 'b> FindToken<&'a u8> for &'b [u8] {
fn find_token(&self, token: &u8) -> bool {
memchr::memchr(*token, self).is_some()
}
}
impl<'a, 'b> FindToken<&'a u8> for &'b str {
fn find_token(&self, token: &u8) -> bool {
self.as_bytes().find_token(token)
}
}
impl<'a> FindToken<char> for &'a [u8] {
fn find_token(&self, token: char) -> bool {
for i in self.iter() {
if token as u8 == *i {
return true;
}
}
false
}
}
impl<'a> FindToken<char> for &'a str {
fn find_token(&self, token: char) -> bool {
for i in self.chars() {
if token == i {
return true;
}
}
false
}
}
pub trait FindSubstring<T> {
fn find_substring(&self, substr: T) -> Option<usize>;
}
impl<'a, 'b> FindSubstring<&'b [u8]> for &'a [u8] {
fn find_substring(&self, substr: &'b [u8]) -> Option<usize> {
let substr_len = substr.len();
if substr_len == 0 {
Some(0)
} else if substr_len == 1 {
memchr::memchr(substr[0], self)
} else if substr_len > self.len() {
None
} else {
let max = self.len() - substr_len;
let mut offset = 0;
let mut haystack = &self[..];
while let Some(position) = memchr::memchr(substr[0], haystack) {
offset += position;
if offset > max {
return None;
}
if &haystack[position..position + substr_len] == substr {
return Some(offset);
}
haystack = &haystack[position + 1..];
offset += 1;
}
None
}
}
}
impl<'a, 'b> FindSubstring<&'b str> for &'a [u8] {
fn find_substring(&self, substr: &'b str) -> Option<usize> {
self.find_substring(AsBytes::as_bytes(substr))
}
}
impl<'a, 'b> FindSubstring<&'b str> for &'a str {
fn find_substring(&self, substr: &'b str) -> Option<usize> {
self.find(substr)
}
}
pub trait ParseTo<R> {
fn parse_to(&self) -> Option<R>;
}
impl<'a, R: FromStr> ParseTo<R> for &'a [u8] {
fn parse_to(&self) -> Option<R> {
from_utf8(self).ok().and_then(|s| s.parse().ok())
}
}
impl<'a, R: FromStr> ParseTo<R> for &'a str {
fn parse_to(&self) -> Option<R> {
self.parse().ok()
}
}
pub trait Slice<R> {
#[inline(always)]
fn slice(&self, range: R) -> Self;
}
macro_rules! impl_fn_slice {
( $ty:ty ) => {
fn slice(&self, range:$ty) -> Self {
&self[range]
}
}
}
macro_rules! slice_range_impl {
( [ $for_type:ident ], $ty:ty ) => {
impl<'a, $for_type> Slice<$ty> for &'a [$for_type] {
impl_fn_slice!( $ty );
}
};
( $for_type:ty, $ty:ty ) => {
impl<'a> Slice<$ty> for &'a $for_type {
impl_fn_slice!( $ty );
}
}
}
macro_rules! slice_ranges_impl {
( [ $for_type:ident ] ) => {
slice_range_impl! {[$for_type], Range<usize>}
slice_range_impl! {[$for_type], RangeTo<usize>}
slice_range_impl! {[$for_type], RangeFrom<usize>}
slice_range_impl! {[$for_type], RangeFull}
};
( $for_type:ty ) => {
slice_range_impl! {$for_type, Range<usize>}
slice_range_impl! {$for_type, RangeTo<usize>}
slice_range_impl! {$for_type, RangeFrom<usize>}
slice_range_impl! {$for_type, RangeFull}
}
}
slice_ranges_impl! {str}
slice_ranges_impl! {[T]}
macro_rules! array_impls {
($($N:expr)+) => {
$(
impl InputLength for [u8; $N] {
#[inline]
fn input_len(&self) -> usize {
self.len()
}
}
impl<'a> InputLength for &'a [u8; $N] {
#[inline]
fn input_len(&self) -> usize {
self.len()
}
}
impl<'a> Compare<[u8; $N]> for &'a [u8] {
#[inline(always)]
fn compare(&self, t: [u8; $N]) -> CompareResult {
self.compare(&t[..])
}
#[inline(always)]
fn compare_no_case(&self, t: [u8;$N]) -> CompareResult {
self.compare_no_case(&t[..])
}
}
impl<'a,'b> Compare<&'b [u8; $N]> for &'a [u8] {
#[inline(always)]
fn compare(&self, t: &'b [u8; $N]) -> CompareResult {
self.compare(&t[..])
}
#[inline(always)]
fn compare_no_case(&self, t: &'b [u8;$N]) -> CompareResult {
self.compare_no_case(&t[..])
}
}
impl FindToken<u8> for [u8; $N] {
fn find_token(&self, token: u8) -> bool {
memchr::memchr(token, &self[..]).is_some()
}
}
impl<'a> FindToken<&'a u8> for [u8; $N] {
fn find_token(&self, token: &u8) -> bool {
memchr::memchr(*token, &self[..]).is_some()
}
}
)+
};
}
array_impls! {
0 1 2 3 4 5 6 7 8 9
10 11 12 13 14 15 16 17 18 19
20 21 22 23 24 25 26 27 28 29
30 31 32
}
pub trait ExtendInto {
type Item;
type Extender: Extend<Self::Item>;
#[inline]
fn new_builder(&self) -> Self::Extender;
#[inline]
fn extend_into(&self, acc: &mut Self::Extender);
}
#[cfg(feature = "alloc")]
impl ExtendInto for [u8] {
type Item = u8;
type Extender = Vec<u8>;
#[inline]
fn new_builder(&self) -> Vec<u8> {
Vec::new()
}
#[inline]
fn extend_into(&self, acc: &mut Vec<u8>) {
acc.extend(self.iter().cloned());
}
}
#[cfg(feature = "alloc")]
impl ExtendInto for &[u8] {
type Item = u8;
type Extender = Vec<u8>;
#[inline]
fn new_builder(&self) -> Vec<u8> {
Vec::new()
}
#[inline]
fn extend_into(&self, acc: &mut Vec<u8>) {
acc.extend(self.iter().cloned());
}
}
#[cfg(feature = "alloc")]
impl ExtendInto for str {
type Item = char;
type Extender = String;
#[inline]
fn new_builder(&self) -> String {
String::new()
}
#[inline]
fn extend_into(&self, acc: &mut String) {
acc.push_str(self);
}
}
#[cfg(feature = "alloc")]
impl ExtendInto for &str {
type Item = char;
type Extender = String;
#[inline]
fn new_builder(&self) -> String {
String::new()
}
#[inline]
fn extend_into(&self, acc: &mut String) {
acc.push_str(self);
}
}
#[cfg(feature = "alloc")]
impl ExtendInto for char {
type Item = char;
type Extender = String;
#[inline]
fn new_builder(&self) -> String {
String::new()
}
#[inline]
fn extend_into(&self, acc: &mut String) {
acc.push(*self);
}
}
pub trait ToUsize {
#[inline]
fn to_usize(&self) -> usize;
}
impl ToUsize for u8 {
#[inline]
fn to_usize(&self) -> usize {
*self as usize
}
}
impl ToUsize for u16 {
#[inline]
fn to_usize(&self) -> usize {
*self as usize
}
}
impl ToUsize for usize {
#[inline]
fn to_usize(&self) -> usize {
*self
}
}
#[cfg(any(target_pointer_width = "32", target_pointer_width = "64"))]
impl ToUsize for u32 {
#[inline]
fn to_usize(&self) -> usize {
*self as usize
}
}
#[cfg(target_pointer_width = "64")]
impl ToUsize for u64 {
#[inline]
fn to_usize(&self) -> usize {
*self as usize
}
}
pub trait ErrorConvert<E> {
fn convert(self) -> E;
}
impl<I> ErrorConvert<(I, ErrorKind)> for ((I, usize), ErrorKind) {
fn convert(self) -> (I, ErrorKind) {
((self.0).0, self.1)
}
}
impl<I> ErrorConvert<((I, usize), ErrorKind)> for (I, ErrorKind) {
fn convert(self) -> ((I, usize), ErrorKind) {
((self.0, 0), self.1)
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn test_offset_u8() {
let s = b"abcd123";
let a = &s[..];
let b = &a[2..];
let c = &a[..4];
let d = &a[3..5];
assert_eq!(a.offset(b), 2);
assert_eq!(a.offset(c), 0);
assert_eq!(a.offset(d), 3);
}
#[test]
fn test_offset_str() {
let s = "abcřèÂßÇd123";
let a = &s[..];
let b = &a[7..];
let c = &a[..5];
let d = &a[5..9];
assert_eq!(a.offset(b), 7);
assert_eq!(a.offset(c), 0);
assert_eq!(a.offset(d), 5);
}
}