/// Trait used by the deserializer for iterating over input. This is manually /// "specialized" for iterating over `&[u8]`. Once feature(specialization) is /// stable we can use actual specialization. /// /// This trait is sealed and cannot be implemented for types outside of /// `serde_json`. pubtrait Read<'de>: private::Sealed { #[doc(hidden)] fn next(&mutself) -> Result<Option<u8>>; #[doc(hidden)] fn peek(&mutself) -> Result<Option<u8>>;
/// Only valid after a call to peek(). Discards the peeked byte. #[doc(hidden)] fn discard(&mutself);
/// Position of the most recent call to next(). /// /// The most recent call was probably next() and not peek(), but this method /// should try to return a sensible result if the most recent call was /// actually peek() because we don't always know. /// /// Only called in case of an error, so performance is not important. #[doc(hidden)] fn position(&self) -> Position;
/// Position of the most recent call to peek(). /// /// The most recent call was probably peek() and not next(), but this method /// should try to return a sensible result if the most recent call was /// actually next() because we don't always know. /// /// Only called in case of an error, so performance is not important. #[doc(hidden)] fn peek_position(&self) -> Position;
/// Offset from the beginning of the input to the next byte that would be /// returned by next() or peek(). #[doc(hidden)] fn byte_offset(&self) -> usize;
/// Assumes the previous byte was a quotation mark. Parses a JSON-escaped /// string until the next quotation mark using the given scratch space if /// necessary. The scratch space is initially empty. #[doc(hidden)] fn parse_str<'s>(&'s mutself, scratch: &'s mut Vec<u8>) -> Result<Reference<'de, 's, str>>;
/// Assumes the previous byte was a quotation mark. Parses a JSON-escaped /// string until the next quotation mark using the given scratch space if /// necessary. The scratch space is initially empty. /// /// This function returns the raw bytes in the string with escape sequences /// expanded but without performing unicode validation. #[doc(hidden)] fn parse_str_raw<'s>(
&'s mut self,
scratch: &'s mut Vec<u8>,
) -> Result<Reference<'de, 's, [u8]>>;
/// Assumes the previous byte was a quotation mark. Parses a JSON-escaped /// string until the next quotation mark but discards the data. #[doc(hidden)] fn ignore_str(&mutself) -> Result<()>;
/// Assumes the previous byte was a hex escape sequence ('\u') in a string. /// Parses next hexadecimal sequence. #[doc(hidden)] fn decode_hex_escape(&mutself) -> Result<u16>;
/// Switch raw buffering mode on. /// /// This is used when deserializing `RawValue`. #[cfg(feature = "raw_value")] #[doc(hidden)] fn begin_raw_buffering(&mutself);
/// Switch raw buffering mode off and provides the raw buffered data to the /// given visitor. #[cfg(feature = "raw_value")] #[doc(hidden)] fn end_raw_buffering<V>(&mutself, visitor: V) -> Result<V::Value> where
V: Visitor<'de>;
/// Whether StreamDeserializer::next needs to check the failed flag. True /// for IoRead, false for StrRead and SliceRead which can track failure by /// truncating their input slice to avoid the extra check on every next /// call. #[doc(hidden)] const should_early_return_if_failed: bool;
/// Mark a persistent failure of StreamDeserializer, either by setting the /// flag or by truncating the input data. #[doc(hidden)] fn set_failed(&mutself, failed: &mut bool);
}
pubstruct Position { pub line: usize, pub column: usize,
}
/// JSON input source that reads from a std::io input stream. #[cfg(feature = "std")] #[cfg_attr(docsrs, doc(cfg(feature = "std")))] pubstruct IoRead<R> where
R: io::Read,
{
iter: LineColIterator<io::Bytes<R>>, /// Temporary storage of peeked byte.
ch: Option<u8>, #[cfg(feature = "raw_value")]
raw_buffer: Option<Vec<u8>>,
}
/// JSON input source that reads from a slice of bytes. // // This is more efficient than other iterators because peek() can be read-only // and we can compute line/col position only if an error happens. pubstruct SliceRead<'a> {
slice: &'a [u8], /// Index of the *next* byte that will be returned by next() or peek().
index: usize, #[cfg(feature = "raw_value")]
raw_buffering_start_index: usize,
}
/// JSON input source that reads from a UTF-8 string. // // Able to elide UTF-8 checks by assuming that the input is valid UTF-8. pubstruct StrRead<'a> {
delegate: SliceRead<'a>, #[cfg(feature = "raw_value")]
data: &'a str,
}
// Prevent users from implementing the Read trait. mod private { pubtrait Sealed {}
}
#[cfg(feature = "std")] impl<R> IoRead<R> where
R: io::Read,
{ /// Create a JSON input source to read from a std::io input stream. /// /// When reading from a source against which short reads are not efficient, such /// as a [`File`], you will want to apply your own buffering because serde_json /// will not buffer the input. See [`std::io::BufReader`]. /// /// [`File`]: std::fs::File pubfn new(reader: R) -> Self {
IoRead {
iter: LineColIterator::new(reader.bytes()),
ch: None, #[cfg(feature = "raw_value")]
raw_buffer: None,
}
}
}
#[cfg(feature = "std")] impl<R> private::Sealed for IoRead<R> where R: io::Read {}
fn decode_hex_escape(&mutself) -> Result<u16> { let a = tri!(next_or_eof(self)); let b = tri!(next_or_eof(self)); let c = tri!(next_or_eof(self)); let d = tri!(next_or_eof(self)); match decode_four_hex_digits(a, b, c, d) {
Some(val) => Ok(val),
None => error(self, ErrorCode::InvalidEscape),
}
}
if !forbid_control_characters { self.index += memchr::memchr2(b'"', b'\\', rest).unwrap_or(rest.len()); return;
}
// We wish to find the first byte in range 0x00..=0x1F or " or \. Ideally, we'd use // something akin to memchr3, but the memchr crate does not support this at the moment. // Therefore, we use a variation on Mycroft's algorithm [1] to provide performance better // than a naive loop. It runs faster than equivalent two-pass memchr2+SWAR code on // benchmarks and it's cross-platform, so probably the right fit. // [1]: https://groups.google.com/forum/#!original/comp.lang.c/2HtQXvg7iKc/xOJeipH6KLMJ
#[cfg(fast_arithmetic = "64")] type Chunk = u64; #[cfg(fast_arithmetic = "32")] type Chunk = u32;
/// The big optimization here over IoRead is that if the string contains no /// backslash escape sequences, the returned &str is a slice of the raw JSON /// data so we avoid copying into the scratch space. fn parse_str_bytes<'s, T, F>(
&'s mut self,
scratch: &'s mut Vec<u8>,
validate: bool,
result: F,
) -> Result<Reference<'a, 's, T>> where
T: ?Sized + 's,
F: for<'f> FnOnce(&'s Self, &'f [u8]) -> Result<&'f T>,
{ // Index of the first byte not yet copied into the scratch space. letmut start = self.index;
loop { self.skip_to_escape(validate); ifself.index == self.slice.len() { return error(self, ErrorCode::EofWhileParsingString);
} matchself.slice[self.index] {
b'"' => { if scratch.is_empty() { // Fast path: return a slice of the raw JSON without any // copying. let borrowed = &self.slice[start..self.index]; self.index += 1; return result(self, borrowed).map(Reference::Borrowed);
} else {
scratch.extend_from_slice(&self.slice[start..self.index]); self.index += 1; return result(self, scratch).map(Reference::Copied);
}
}
b'\\' => {
scratch.extend_from_slice(&self.slice[start..self.index]); self.index += 1;
tri!(parse_escape(self, validate, scratch));
start = self.index;
}
_ => { self.index += 1; return error(self, ErrorCode::ControlCharacterWhileParsingString);
}
}
}
}
}
fn position(&self) -> Position { self.position_of_index(self.index)
}
fn peek_position(&self) -> Position { // Cap it at slice.len() just in case the most recent call was next() // and it returned the last byte. self.position_of_index(cmp::min(self.slice.len(), self.index + 1))
}
fn parse_str<'s>(&'s mutself, scratch: &'s mut Vec<u8>) -> Result<Reference<'a, 's, str>> { self.delegate.parse_str_bytes(scratch, true, |_, bytes| { // The deserialization input came in as &str with a UTF-8 guarantee, // and the \u-escapes are checked along the way, so don't need to // check here.
Ok(unsafe { str::from_utf8_unchecked(bytes) })
})
}
/// Parses a JSON escape sequence and appends it into the scratch space. Assumes /// the previous byte read was a backslash. fn parse_escape<'de, R: Read<'de>>(
read: &mut R,
validate: bool,
scratch: &mut Vec<u8>,
) -> Result<()> { let ch = tri!(next_or_eof(read));
/// Parses a JSON \u escape and appends it into the scratch space. Assumes `\u` /// has just been read. #[cold] fn parse_unicode_escape<'de, R: Read<'de>>(
read: &mut R,
validate: bool,
scratch: &mut Vec<u8>,
) -> Result<()> { letmut n = tri!(read.decode_hex_escape());
// Non-BMP characters are encoded as a sequence of two hex escapes, // representing UTF-16 surrogates. If deserializing a utf-8 string the // surrogates are required to be paired, whereas deserializing a byte string // accepts lone surrogates. if validate && n >= 0xDC00 && n <= 0xDFFF { // XXX: This is actually a trailing surrogate. return error(read, ErrorCode::LoneLeadingSurrogateInHexEscape);
}
loop { if n < 0xD800 || n > 0xDBFF { // Every u16 outside of the surrogate ranges is guaranteed to be a // legal char.
push_wtf8_codepoint(n as u32, scratch); return Ok(());
}
// n is a leading surrogate, we now expect a trailing surrogate. let n1 = n;
if tri!(peek_or_eof(read)) == b'u' {
read.discard();
} else { returnif validate {
read.discard();
error(read, ErrorCode::UnexpectedEndOfHexEscape)
} else {
push_wtf8_codepoint(n1 as u32, scratch); // The \ prior to this byte started an escape sequence, so we // need to parse that now. This recursive call does not blow the // stack on malicious input because the escape is not \u, so it // will be handled by one of the easy nonrecursive cases.
parse_escape(read, validate, scratch)
};
}
let n2 = tri!(read.decode_hex_escape());
if n2 < 0xDC00 || n2 > 0xDFFF { if validate { return error(read, ErrorCode::LoneLeadingSurrogateInHexEscape);
}
push_wtf8_codepoint(n1 as u32, scratch); // If n2 is a leading surrogate, we need to restart.
n = n2; continue;
}
// This value is in range U+10000..=U+10FFFF, which is always a valid // codepoint. let n = ((((n1 - 0xD800) as u32) << 10) | (n2 - 0xDC00) as u32) + 0x1_0000;
push_wtf8_codepoint(n, scratch); return Ok(());
}
}
/// Adds a WTF-8 codepoint to the end of the buffer. This is a more efficient /// implementation of String::push. The codepoint may be a surrogate. #[inline] fn push_wtf8_codepoint(n: u32, scratch: &mut Vec<u8>) { if n < 0x80 {
scratch.push(n as u8); return;
}
scratch.reserve(4);
// SAFETY: After the `reserve` call, `scratch` has at least 4 bytes of // allocated but uninitialized memory after its last initialized byte, which // is where `ptr` points. All reachable match arms write `encoded_len` bytes // to that region and update the length accordingly, and `encoded_len` is // always <= 4. unsafe { let ptr = scratch.as_mut_ptr().add(scratch.len());
let encoded_len = match n { 0..=0x7F => unreachable!(), 0x80..=0x7FF => {
ptr.write(((n >> 6) & 0b0001_1111) as u8 | 0b1100_0000); 2
} 0x800..=0xFFFF => {
ptr.write(((n >> 12) & 0b0000_1111) as u8 | 0b1110_0000);
ptr.add(1)
.write(((n >> 6) & 0b0011_1111) as u8 | 0b1000_0000); 3
} 0x1_0000..=0x10_FFFF => {
ptr.write(((n >> 18) & 0b0000_0111) as u8 | 0b1111_0000);
ptr.add(1)
.write(((n >> 12) & 0b0011_1111) as u8 | 0b1000_0000);
ptr.add(2)
.write(((n >> 6) & 0b0011_1111) as u8 | 0b1000_0000); 4
} 0x11_0000.. => unreachable!(),
};
ptr.add(encoded_len - 1)
.write((n & 0b0011_1111) as u8 | 0b1000_0000);
scratch.set_len(scratch.len() + encoded_len);
}
}
/// Parses a JSON escape sequence and discards the value. Assumes the previous /// byte read was a backslash. fn ignore_escape<'de, R>(read: &mut R) -> Result<()> where
R: ?Sized + Read<'de>,
{ let ch = tri!(next_or_eof(read));
match ch {
b'"' | b'\\' | b'/' | b'b' | b'f' | b'n' | b'r' | b't' => {}
b'u' => { // At this point we don't care if the codepoint is valid. We just // want to consume it. We don't actually know what is valid or not // at this point, because that depends on if this string will // ultimately be parsed into a string or a byte buffer in the "real" // parse.
fn decode_four_hex_digits(a: u8, b: u8, c: u8, d: u8) -> Option<u16> { let a = HEX1[a as usize] as i32; let b = HEX0[b as usize] as i32; let c = HEX1[c as usize] as i32; let d = HEX0[d as usize] as i32;
let codepoint = ((a | b) << 8) | c | d;
// A single sign bit check. if codepoint >= 0 {
Some(codepoint as u16)
} else {
None
}
}
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.