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
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
use std::arch::wasm32::{u8x16_bitmask, u8x16_eq, u8x16_splat, v128, v128_load, v128_store};
use crate::{
error::ErrorType,
safer_unchecked::GetSaferUnchecked,
stringparse::{handle_unicode_codepoint, ESCAPE_MAP},
Deserializer, Result, SillyWrapper,
};
#[target_feature(enable = "simd128")]
#[allow(
clippy::if_not_else,
clippy::cast_possible_wrap,
clippy::too_many_lines
)]
#[cfg_attr(not(feature = "no-inline"), inline)]
pub(crate) fn parse_str<'invoke, 'de>(
input: SillyWrapper<'de>,
data: &'invoke [u8],
buffer: &'invoke mut [u8],
mut idx: usize,
) -> Result<&'de str> {
use ErrorType::{InvalidEscape, InvalidUnicodeCodepoint};
let input = input.input;
// Add 1 to skip the initial "
idx += 1;
// we include the terminal '"' so we know where to end
// This is safe since we check sub's length in the range access above and only
// create sub sliced form sub to `sub.len()`.
let src = unsafe { data.get_kinda_unchecked(idx..) };
let mut src_i = 0;
let mut len = src_i;
loop {
let v = unsafe {
// v128_load requires no alignment
#[allow(clippy::cast_ptr_alignment)]
v128_load(src.as_ptr().add(src_i).cast::<v128>())
};
let bs_bits = u8x16_bitmask(u8x16_eq(v, u8x16_splat(b'\\')));
let quote_bits = u8x16_bitmask(u8x16_eq(v, u8x16_splat(b'"')));
if (bs_bits.wrapping_sub(1) & quote_bits) != 0 {
// we encountered quotes first. Move dst to point to quotes and exit
// find out where the quote is...
let quote_dist = quote_bits.trailing_zeros();
///////////////////////
// Above, check for overflow in case someone has a crazy string (>=4GB?)
// But only add the overflow check when the document itself exceeds 4GB
// Currently unneeded because we refuse to parse docs larger or equal to 4GB.
////////////////////////
// we advance the point, accounting for the fact that we have a NULl termination
len += quote_dist as usize;
unsafe {
let v =
std::str::from_utf8_unchecked(std::slice::from_raw_parts(input.add(idx), len));
return Ok(v);
}
// we compare the pointers since we care if they are 'at the same spot'
// not if they are the same value
}
if (quote_bits.wrapping_sub(1) & bs_bits) == 0 {
// they are the same. Since they can't co-occur, it means we encountered
// neither.
src_i += 16;
len += 16;
} else {
// Move to the 'bad' character
let bs_dist = bs_bits.trailing_zeros();
len += bs_dist as usize;
src_i += bs_dist as usize;
break;
}
}
let mut dst_i = 0;
// To be more conform with upstream
loop {
let v = unsafe {
// v128_load requires no alignment
#[allow(clippy::cast_ptr_alignment)]
v128_load(src.as_ptr().add(src_i).cast::<v128>())
};
unsafe {
// v128_store requires no alignment
#[allow(clippy::cast_ptr_alignment)]
v128_store(buffer.as_mut_ptr().add(dst_i).cast::<v128>(), v);
};
// store to dest unconditionally - we can overwrite the bits we don't like
// later
let bs_bits = u8x16_bitmask(u8x16_eq(v, u8x16_splat(b'\\')));
let quote_bits = u8x16_bitmask(u8x16_eq(v, u8x16_splat(b'"')));
if (bs_bits.wrapping_sub(1) & quote_bits) != 0 {
// we encountered quotes first. Move dst to point to quotes and exit
// find out where the quote is...
let quote_dist = quote_bits.trailing_zeros();
///////////////////////
// Above, check for overflow in case someone has a crazy string (>=4GB?)
// But only add the overflow check when the document itself exceeds 4GB
// Currently unneeded because we refuse to parse docs larger or equal to 4GB.
////////////////////////
// we advance the point, accounting for the fact that we have a NULl termination
dst_i += quote_dist as usize;
unsafe {
input
.add(idx + len)
.copy_from_nonoverlapping(buffer.as_ptr(), dst_i);
let v = std::str::from_utf8_unchecked(std::slice::from_raw_parts(
input.add(idx),
len + dst_i,
));
return Ok(v);
}
// we compare the pointers since we care if they are 'at the same spot'
// not if they are the same value
}
if (quote_bits.wrapping_sub(1) & bs_bits) != 0 {
// find out where the backspace is
let bs_dist = bs_bits.trailing_zeros();
let escape_char = unsafe { *src.get_kinda_unchecked(src_i + bs_dist as usize + 1) };
// we encountered backslash first. Handle backslash
if escape_char == b'u' {
// move src/dst up to the start; they will be further adjusted
// within the unicode codepoint handling code.
src_i += bs_dist as usize;
dst_i += bs_dist as usize;
let (o, s) = if let Ok(r) =
handle_unicode_codepoint(unsafe { src.get_kinda_unchecked(src_i..) }, unsafe {
buffer.get_kinda_unchecked_mut(dst_i..)
}) {
r
} else {
return Err(Deserializer::error_c(src_i, 'u', InvalidUnicodeCodepoint));
};
if o == 0 {
return Err(Deserializer::error_c(src_i, 'u', InvalidUnicodeCodepoint));
};
// We moved o steps forward at the destination and 6 on the source
src_i += s;
dst_i += o;
} else {
// simple 1:1 conversion. Will eat bs_dist+2 characters in input and
// write bs_dist+1 characters to output
// note this may reach beyond the part of the buffer we've actually
// seen. I think this is ok
let escape_result =
unsafe { *ESCAPE_MAP.get_kinda_unchecked(escape_char as usize) };
if escape_result == 0 {
return Err(Deserializer::error_c(
src_i,
escape_char as char,
InvalidEscape,
));
}
unsafe {
*buffer.get_kinda_unchecked_mut(dst_i + bs_dist as usize) = escape_result;
}
src_i += bs_dist as usize + 2;
dst_i += bs_dist as usize + 1;
}
} else {
// they are the same. Since they can't co-occur, it means we encountered
// neither.
src_i += 16;
dst_i += 16;
}
}
}