// Copyright (c) 2012-2020 Ugorji Nwoke. All rights reserved. // Use of this source code is governed by a MIT license found in the LICENSE file. package codec import "io" // decReader abstracts the reading source, allowing implementations that can // read from an io.Reader or directly off a byte slice with zero-copying. type decReader interface { // readx will return a view of the []byte if decoding from a []byte, OR // read into the implementation scratch buffer if possible i.e. n < len(scratchbuf), OR // create a new []byte and read into that readx(n uint) []byte readb([]byte) readn1() byte readn2() [2]byte // readn3 will read 3 bytes into the top-most elements of a 4-byte array readn3() [4]byte readn4() [4]byte readn8() [8]byte // readn1eof() (v uint8, eof bool) // // read up to 8 bytes at a time // readn(num uint8) (v [8]byte) numread() uint // number of bytes read // skip any whitespace characters, and return the first non-matching byte skipWhitespace() (token byte) // jsonReadNum will include last read byte in first element of slice, // and continue numeric characters until it sees a non-numeric char // or EOF. If it sees a non-numeric character, it will unread that. jsonReadNum() []byte // jsonReadAsisChars will read json plain characters (anything but " or \) // and return a slice terminated by a non-json asis character. jsonReadAsisChars() []byte // skip will skip any byte that matches, and return the first non-matching byte // skip(accept *bitset256) (token byte) // readTo will read any byte that matches, stopping once no-longer matching. // readTo(accept *bitset256) (out []byte) // readUntil will read, only stopping once it matches the 'stop' byte (which it excludes). readUntil(stop byte) (out []byte) } // ------------------------------------------------ type unreadByteStatus uint8 // unreadByteStatus goes from // undefined (when initialized) -- (read) --> canUnread -- (unread) --> canRead ... const ( unreadByteUndefined unreadByteStatus = iota unreadByteCanRead unreadByteCanUnread ) // -------------------- type ioDecReaderCommon struct { r io.Reader // the reader passed in n uint // num read l byte // last byte ls unreadByteStatus // last byte status b [6]byte // tiny buffer for reading single bytes blist *bytesFreelist bufr []byte // buffer for readTo/readUntil } func (z *ioDecReaderCommon) reset(r io.Reader, blist *bytesFreelist) { z.blist = blist z.r = r z.ls = unreadByteUndefined z.l, z.n = 0, 0 z.bufr = z.blist.check(z.bufr, 256) } func (z *ioDecReaderCommon) numread() uint { return z.n } // ------------------------------------------ // ioDecReader is a decReader that reads off an io.Reader. // // It also has a fallback implementation of ByteScanner if needed. type ioDecReader struct { ioDecReaderCommon br io.ByteScanner x [64 + 48]byte // for: get struct field name, swallow valueTypeBytes, etc } func (z *ioDecReader) reset(r io.Reader, blist *bytesFreelist) { z.ioDecReaderCommon.reset(r, blist) z.br, _ = r.(io.ByteScanner) } func (z *ioDecReader) Read(p []byte) (n int, err error) { if len(p) == 0 { return } var firstByte bool if z.ls == unreadByteCanRead { z.ls = unreadByteCanUnread p[0] = z.l if len(p) == 1 { n = 1 return } firstByte = true p = p[1:] } n, err = z.r.Read(p) if n > 0 { if err == io.EOF && n == len(p) { err = nil // read was successful, so postpone EOF (till next time) } z.l = p[n-1] z.ls = unreadByteCanUnread } if firstByte { n++ } return } func (z *ioDecReader) ReadByte() (c byte, err error) { if z.br != nil { c, err = z.br.ReadByte() if err == nil { z.l = c z.ls = unreadByteCanUnread } return } n, err := z.Read(z.b[:1]) if n == 1 { c = z.b[0] if err == io.EOF { err = nil // read was successful, so postpone EOF (till next time) } } return } func (z *ioDecReader) UnreadByte() (err error) { if z.br != nil { err = z.br.UnreadByte() if err == nil { z.ls = unreadByteCanRead } return } switch z.ls { case unreadByteCanUnread: z.ls = unreadByteCanRead case unreadByteCanRead: err = errDecUnreadByteLastByteNotRead case unreadByteUndefined: err = errDecUnreadByteNothingToRead default: err = errDecUnreadByteUnknown } return } func (z *ioDecReader) readn2() (bs [2]byte) { z.readb(bs[:]) return } func (z *ioDecReader) readn3() (bs [4]byte) { z.readb(bs[1:]) return } func (z *ioDecReader) readn4() (bs [4]byte) { z.readb(bs[:]) return } func (z *ioDecReader) readn8() (bs [8]byte) { z.readb(bs[:]) return } func (z *ioDecReader) readx(n uint) (bs []byte) { if n == 0 { return } if n < uint(len(z.x)) { bs = z.x[:n] } else { bs = make([]byte, n) } _, err := readFull(z.r, bs) halt.onerror(err) z.n += uint(len(bs)) return } func (z *ioDecReader) readb(bs []byte) { if len(bs) == 0 { return } _, err := readFull(z.r, bs) halt.onerror(err) z.n += uint(len(bs)) } func (z *ioDecReader) readn1() (b uint8) { b, err := z.ReadByte() halt.onerror(err) z.n++ return } func (z *ioDecReader) readn1eof() (b uint8, eof bool) { b, err := z.ReadByte() if err == nil { z.n++ } else if err == io.EOF { eof = true } else { halt.onerror(err) } return } func (z *ioDecReader) jsonReadNum() (bs []byte) { z.unreadn1() z.bufr = z.bufr[:0] LOOP: i, eof := z.readn1eof() if eof { return z.bufr } if isNumberChar(i) { z.bufr = append(z.bufr, i) goto LOOP } z.unreadn1() return z.bufr } func (z *ioDecReader) jsonReadAsisChars() (bs []byte) { z.bufr = z.bufr[:0] LOOP: i := z.readn1() z.bufr = append(z.bufr, i) if i == '"' || i == '\\' { return z.bufr } goto LOOP } func (z *ioDecReader) skipWhitespace() (token byte) { LOOP: token = z.readn1() if isWhitespaceChar(token) { goto LOOP } return } func (z *ioDecReader) readUntil(stop byte) []byte { z.bufr = z.bufr[:0] LOOP: token := z.readn1() z.bufr = append(z.bufr, token) if token == stop { return z.bufr[:len(z.bufr)-1] } goto LOOP } func (z *ioDecReader) unreadn1() { err := z.UnreadByte() halt.onerror(err) z.n-- } // ------------------------------------ type bufioDecReader struct { ioDecReaderCommon c uint // cursor buf []byte } func (z *bufioDecReader) reset(r io.Reader, bufsize int, blist *bytesFreelist) { z.ioDecReaderCommon.reset(r, blist) z.c = 0 if cap(z.buf) < bufsize { z.buf = blist.get(bufsize) } else { z.buf = z.buf[:0] } } func (z *bufioDecReader) readb(p []byte) { var n = uint(copy(p, z.buf[z.c:])) z.n += n z.c += n if len(p) != int(n) { z.readbFill(p, n, true, false) } } func readbFillHandleErr(err error, must, eof bool) (isEOF bool) { if err == io.EOF { isEOF = true } if must && !(eof && isEOF) { halt.onerror(err) } return } func (z *bufioDecReader) readbFill(p0 []byte, n uint, must, eof bool) (isEOF bool, err error) { // at this point, there's nothing in z.buf to read (z.buf is fully consumed) var p []byte if p0 != nil { p = p0[n:] } var n2 uint if len(p) > cap(z.buf) { n2, err = readFull(z.r, p) if err != nil { isEOF = readbFillHandleErr(err, must, eof) return } n += n2 z.n += n2 // always keep last byte in z.buf z.buf = z.buf[:1] z.buf[0] = p[len(p)-1] z.c = 1 return } // z.c is now 0, and len(p) <= cap(z.buf) var n1 int LOOP: // for len(p) > 0 && z.err == nil { z.buf = z.buf[0:cap(z.buf)] n1, err = z.r.Read(z.buf) n2 = uint(n1) if n2 == 0 && err != nil { isEOF = readbFillHandleErr(err, must, eof) return } err = nil z.buf = z.buf[:n2] z.c = 0 if len(p) > 0 { n2 = uint(copy(p, z.buf)) z.c = n2 n += n2 z.n += n2 p = p[n2:] if len(p) > 0 { goto LOOP } if z.c == 0 { z.buf = z.buf[:1] z.buf[0] = p[len(p)-1] z.c = 1 } } return } func (z *bufioDecReader) readn1() (b byte) { if z.c >= uint(len(z.buf)) { z.readbFill(nil, 0, true, false) } b = z.buf[z.c] z.c++ z.n++ return } func (z *bufioDecReader) readn1eof() (b byte, eof bool) { if z.c >= uint(len(z.buf)) { eof, _ = z.readbFill(nil, 0, true, true) if eof { return } } b = z.buf[z.c] z.c++ z.n++ return } func (z *bufioDecReader) unreadn1() { if z.c == 0 { halt.onerror(errDecUnreadByteNothingToRead) } z.c-- z.n-- } func (z *bufioDecReader) readn2() (bs [2]byte) { z.readb(bs[:]) return } func (z *bufioDecReader) readn3() (bs [4]byte) { z.readb(bs[1:]) return } func (z *bufioDecReader) readn4() (bs [4]byte) { z.readb(bs[:]) return } func (z *bufioDecReader) readn8() (bs [8]byte) { z.readb(bs[:]) return } func (z *bufioDecReader) readx(n uint) (bs []byte) { if n == 0 { // return } else if z.c+n <= uint(len(z.buf)) { bs = z.buf[z.c : z.c+n] z.n += n z.c += n } else { bs = make([]byte, n) // n no longer used - can reuse n = uint(copy(bs, z.buf[z.c:])) z.n += n z.c += n z.readbFill(bs, n, true, false) } return } func (z *bufioDecReader) jsonReadNum() (bs []byte) { z.unreadn1() z.bufr = z.bufr[:0] LOOP: i, eof := z.readn1eof() if eof { return z.bufr } if isNumberChar(i) { z.bufr = append(z.bufr, i) goto LOOP } z.unreadn1() return z.bufr } func (z *bufioDecReader) jsonReadAsisChars() (bs []byte) { z.bufr = z.bufr[:0] LOOP: i := z.readn1() z.bufr = append(z.bufr, i) if i == '"' || i == '\\' { return z.bufr } goto LOOP } func (z *bufioDecReader) skipWhitespace() (token byte) { i := z.c LOOP: if i < uint(len(z.buf)) { // inline z.skipLoopFn(i) and refactor, so cost is within inline budget token = z.buf[i] i++ if isWhitespaceChar(token) { goto LOOP } z.n += i - 2 - z.c z.c = i return } return z.skipFillWhitespace() } func (z *bufioDecReader) skipFillWhitespace() (token byte) { z.n += uint(len(z.buf)) - z.c var i, n2 int var err error for { z.c = 0 z.buf = z.buf[0:cap(z.buf)] n2, err = z.r.Read(z.buf) if n2 == 0 { halt.onerror(err) } z.buf = z.buf[:n2] for i, token = range z.buf { if !isWhitespaceChar(token) { z.n += (uint(i) - z.c) - 1 z.loopFn(uint(i + 1)) return } } z.n += uint(n2) } } func (z *bufioDecReader) loopFn(i uint) { z.c = i } func (z *bufioDecReader) readUntil(stop byte) (out []byte) { i := z.c LOOP: if i < uint(len(z.buf)) { if z.buf[i] == stop { z.n += (i - z.c) - 1 i++ out = z.buf[z.c:i] z.c = i goto FINISH } i++ goto LOOP } out = z.readUntilFill(stop) FINISH: return out[:len(out)-1] } func (z *bufioDecReader) readUntilFill(stop byte) []byte { z.bufr = z.bufr[:0] z.n += uint(len(z.buf)) - z.c z.bufr = append(z.bufr, z.buf[z.c:]...) for { z.c = 0 z.buf = z.buf[0:cap(z.buf)] n1, err := z.r.Read(z.buf) if n1 == 0 { halt.onerror(err) } n2 := uint(n1) z.buf = z.buf[:n2] for i, token := range z.buf { if token == stop { z.n += (uint(i) - z.c) - 1 z.bufr = append(z.bufr, z.buf[z.c:i+1]...) z.loopFn(uint(i + 1)) return z.bufr } } z.bufr = append(z.bufr, z.buf...) z.n += n2 } } // ------------------------------------ // bytesDecReader is a decReader that reads off a byte slice with zero copying // // Note: we do not try to convert index'ing out of bounds to an io.EOF. // instead, we let it bubble up to the exported Encode/Decode method // and recover it as an io.EOF. // // see panicValToErr(...) function in helper.go. type bytesDecReader struct { b []byte // data c uint // cursor } func (z *bytesDecReader) reset(in []byte) { z.b = in[:len(in):len(in)] // reslicing must not go past capacity z.c = 0 } func (z *bytesDecReader) numread() uint { return z.c } // Note: slicing from a non-constant start position is more expensive, // as more computation is required to decipher the pointer start position. // However, we do it only once, and it's better than reslicing both z.b and return value. func (z *bytesDecReader) readx(n uint) (bs []byte) { x := z.c + n bs = z.b[z.c:x] z.c = x return } func (z *bytesDecReader) readb(bs []byte) { copy(bs, z.readx(uint(len(bs)))) } // MARKER: do not use this - as it calls into memmove (as the size of data to move is unknown) // func (z *bytesDecReader) readnn(bs []byte, n uint) { // x := z.c // copy(bs, z.b[x:x+n]) // z.c += n // } // func (z *bytesDecReader) readn(num uint8) (bs [8]byte) { // x := z.c + uint(num) // copy(bs[:], z.b[z.c:x]) // slice z.b completely, so we get bounds error if past // z.c = x // return // } // func (z *bytesDecReader) readn1() uint8 { // z.c++ // return z.b[z.c-1] // } func (z *bytesDecReader) readn1() (v uint8) { v = z.b[z.c] z.c++ return } // MARKER: for readn{2,3,4,8}, ensure you slice z.b completely so we get bounds error if past end. func (z *bytesDecReader) readn2() (bs [2]byte) { // copy(bs[:], z.b[z.c:z.c+2]) bs[1] = z.b[z.c+1] bs[0] = z.b[z.c] z.c += 2 return } func (z *bytesDecReader) readn3() (bs [4]byte) { // copy(bs[1:], z.b[z.c:z.c+3]) bs = okBytes3(z.b[z.c : z.c+3]) z.c += 3 return } func (z *bytesDecReader) readn4() (bs [4]byte) { // copy(bs[:], z.b[z.c:z.c+4]) bs = okBytes4(z.b[z.c : z.c+4]) z.c += 4 return } func (z *bytesDecReader) readn8() (bs [8]byte) { // copy(bs[:], z.b[z.c:z.c+8]) bs = okBytes8(z.b[z.c : z.c+8]) z.c += 8 return } func (z *bytesDecReader) jsonReadNum() []byte { z.c-- i := z.c LOOP: if i < uint(len(z.b)) && isNumberChar(z.b[i]) { i++ goto LOOP } z.c, i = i, z.c return z.b[i:z.c] } func (z *bytesDecReader) jsonReadAsisChars() []byte { i := z.c LOOP: token := z.b[i] i++ if token == '"' || token == '\\' { z.c, i = i, z.c return z.b[i:z.c] } goto LOOP } func (z *bytesDecReader) skipWhitespace() (token byte) { i := z.c LOOP: token = z.b[i] if isWhitespaceChar(token) { i++ goto LOOP } z.c = i + 1 return } func (z *bytesDecReader) readUntil(stop byte) (out []byte) { i := z.c LOOP: if z.b[i] == stop { out = z.b[z.c:i] z.c = i + 1 return } i++ goto LOOP } // -------------- type decRd struct { mtr bool // is maptype a known type? str bool // is slicetype a known type? be bool // is binary encoding js bool // is json handle jsms bool // is json handle, and MapKeyAsString cbor bool // is cbor handle bytes bool // is bytes reader bufio bool // is this a bufioDecReader? rb bytesDecReader ri *ioDecReader bi *bufioDecReader decReader } // From out benchmarking, we see the following in terms of performance: // // - interface calls // - branch that can inline what it calls // // the if/else-if/else block is expensive to inline. // Each node of this construct costs a lot and dominates the budget. // Best to only do an if fast-path else block (so fast-path is inlined). // This is irrespective of inlineExtraCallCost set in $GOROOT/src/cmd/compile/internal/gc/inl.go // // In decRd methods below, we delegate all IO functions into their own methods. // This allows for the inlining of the common path when z.bytes=true. // Go 1.12+ supports inlining methods with up to 1 inlined function (or 2 if no other constructs). // // However, up through Go 1.13, decRd's readXXX, skip and unreadXXX methods are not inlined. // Consequently, there is no benefit to do the xxxIO methods for decRd at this time. // Instead, we have a if/else-if/else block so that IO calls do not have to jump through // a second unnecessary function call. // // If golang inlining gets better and bytesDecReader methods can be inlined, // then we can revert to using these 2 functions so the bytesDecReader // methods are inlined and the IO paths call out to a function. // // decRd is designed to embed a decReader, and then re-implement some of the decReader // methods using a conditional branch. We only override the ones that have a bytes version // that is small enough to be inlined. We use ./run.sh -z to check. // Right now, only numread and readn1 can be inlined. func (z *decRd) numread() uint { if z.bytes { return z.rb.numread() } else if z.bufio { return z.bi.numread() } else { return z.ri.numread() } } func (z *decRd) readn1() (v uint8) { if z.bytes { // MARKER: manually inline, else this function is not inlined. // Keep in sync with bytesDecReader.readn1 // return z.rb.readn1() v = z.rb.b[z.rb.c] z.rb.c++ } else { v = z.readn1IO() } return } func (z *decRd) readn1IO() uint8 { if z.bufio { return z.bi.readn1() } return z.ri.readn1() } type devNullReader struct{} func (devNullReader) Read(p []byte) (int, error) { return 0, io.EOF } func (devNullReader) Close() error { return nil } func readFull(r io.Reader, bs []byte) (n uint, err error) { var nn int for n < uint(len(bs)) && err == nil { nn, err = r.Read(bs[n:]) if nn > 0 { if err == io.EOF { // leave EOF for next time err = nil } n += uint(nn) } } // do not do this below - it serves no purpose // if n != len(bs) && err == io.EOF { err = io.ErrUnexpectedEOF } return } var _ decReader = (*decRd)(nil)