// 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 // By default, this json support uses base64 encoding for bytes, because you cannot // store and read any arbitrary string in json (only unicode). // However, the user can configre how to encode/decode bytes. // // This library specifically supports UTF-8 for encoding and decoding only. // // Note that the library will happily encode/decode things which are not valid // json e.g. a map[int64]string. We do it for consistency. With valid json, // we will encode and decode appropriately. // Users can specify their map type if necessary to force it. // // We cannot use strconv.(Q|Unq)uote because json quotes/unquotes differently. import ( "encoding/base64" "math" "reflect" "strconv" "time" "unicode" "unicode/utf16" "unicode/utf8" ) //-------------------------------- // jsonLits and jsonLitb are defined at the package level, // so they are guaranteed to be stored efficiently, making // for better append/string comparison/etc. // // (anecdotal evidence from some benchmarking on go 1.20 devel in 20220104) const jsonLits = `"true"false"null"` var jsonLitb = []byte(jsonLits) const ( jsonLitT = 1 jsonLitF = 6 jsonLitN = 12 ) const jsonEncodeUintSmallsString = "" + "00010203040506070809" + "10111213141516171819" + "20212223242526272829" + "30313233343536373839" + "40414243444546474849" + "50515253545556575859" + "60616263646566676869" + "70717273747576777879" + "80818283848586878889" + "90919293949596979899" var jsonEncodeUintSmallsStringBytes = []byte(jsonEncodeUintSmallsString) const ( jsonU4Chk2 = '0' jsonU4Chk1 = 'a' - 10 jsonU4Chk0 = 'A' - 10 ) const ( // If !jsonValidateSymbols, decoding will be faster, by skipping some checks: // - If we see first character of null, false or true, // do not validate subsequent characters. // - e.g. if we see a n, assume null and skip next 3 characters, // and do not validate they are ull. // P.S. Do not expect a significant decoding boost from this. jsonValidateSymbols = true // jsonEscapeMultiByteUnicodeSep controls whether some unicode characters // that are valid json but may bomb in some contexts are escaped during encoeing. // // U+2028 is LINE SEPARATOR. U+2029 is PARAGRAPH SEPARATOR. // Both technically valid JSON, but bomb on JSONP, so fix here unconditionally. jsonEscapeMultiByteUnicodeSep = true // jsonRecognizeBoolNullInQuotedStr is used during decoding into a blank interface{} // to control whether we detect quoted values of bools and null where a map key is expected, // and treat as nil, true or false. jsonNakedBoolNullInQuotedStr = true // jsonManualInlineDecRdInHotZones controls whether we manually inline some decReader calls. // // encode performance is at par with libraries that just iterate over bytes directly, // because encWr (with inlined bytesEncAppender calls) is inlined. // Conversely, decode performance suffers because decRd (with inlined bytesDecReader calls) // isn't inlinable. // // To improve decode performamnce from json: // - readn1 is only called for \u // - consequently, to optimize json decoding, we specifically need inlining // for bytes use-case of some other decReader methods: // - jsonReadAsisChars, skipWhitespace (advance) and jsonReadNum // - AND THEN readn3, readn4 (for ull, rue and alse). // - (readn1 is only called when a char is escaped). // - without inlining, we still pay the cost of a method invocationK, and this dominates time // - To mitigate, we manually inline in hot zones // *excluding places where used sparingly (e.g. nextValueBytes, and other atypical cases)*. // - jsonReadAsisChars *only* called in: appendStringAsBytes // - advance called: everywhere // - jsonReadNum: decNumBytes, DecodeNaked // - From running go test (our anecdotal findings): // - calling jsonReadAsisChars in appendStringAsBytes: 23431 // - calling jsonReadNum in decNumBytes: 15251 // - calling jsonReadNum in DecodeNaked: 612 // Consequently, we manually inline jsonReadAsisChars (in appendStringAsBytes) // and jsonReadNum (in decNumbytes) jsonManualInlineDecRdInHotZones = true jsonSpacesOrTabsLen = 128 // jsonAlwaysReturnInternString = false ) var ( // jsonTabs and jsonSpaces are used as caches for indents jsonTabs, jsonSpaces [jsonSpacesOrTabsLen]byte jsonCharHtmlSafeSet bitset256 jsonCharSafeSet bitset256 ) func init() { var i byte for i = 0; i < jsonSpacesOrTabsLen; i++ { jsonSpaces[i] = ' ' jsonTabs[i] = '\t' } // populate the safe values as true: note: ASCII control characters are (0-31) // jsonCharSafeSet: all true except (0-31) " \ // jsonCharHtmlSafeSet: all true except (0-31) " \ < > & for i = 32; i < utf8.RuneSelf; i++ { switch i { case '"', '\\': case '<', '>', '&': jsonCharSafeSet.set(i) // = true default: jsonCharSafeSet.set(i) jsonCharHtmlSafeSet.set(i) } } } // ---------------- type jsonEncState struct { di int8 // indent per: if negative, use tabs d bool // indenting? dl uint16 // indent level } func (x jsonEncState) captureState() interface{} { return x } func (x *jsonEncState) restoreState(v interface{}) { *x = v.(jsonEncState) } type jsonEncDriver struct { noBuiltInTypes h *JsonHandle // se interfaceExtWrapper // ---- cpu cache line boundary? jsonEncState ks bool // map key as string is byte // integer as string typical bool rawext bool // rawext configured on the handle s *bitset256 // safe set for characters (taking h.HTMLAsIs into consideration) // buf *[]byte // used mostly for encoding []byte // scratch buffer for: encode time, numbers, etc // // RFC3339Nano uses 35 chars: 2006-01-02T15:04:05.999999999Z07:00 // MaxUint64 uses 20 chars: 18446744073709551615 // floats are encoded using: f/e fmt, and -1 precision, or 1 if no fractions. // This means we are limited by the number of characters for the // mantissa (up to 17), exponent (up to 3), signs (up to 3), dot (up to 1), E (up to 1) // for a total of 24 characters. // -xxx.yyyyyyyyyyyye-zzz // Consequently, 35 characters should be sufficient for encoding time, integers or floats. // We use up all the remaining bytes to make this use full cache lines. b [48]byte e Encoder } func (e *jsonEncDriver) encoder() *Encoder { return &e.e } func (e *jsonEncDriver) writeIndent() { e.e.encWr.writen1('\n') x := int(e.di) * int(e.dl) if e.di < 0 { x = -x for x > jsonSpacesOrTabsLen { e.e.encWr.writeb(jsonTabs[:]) x -= jsonSpacesOrTabsLen } e.e.encWr.writeb(jsonTabs[:x]) } else { for x > jsonSpacesOrTabsLen { e.e.encWr.writeb(jsonSpaces[:]) x -= jsonSpacesOrTabsLen } e.e.encWr.writeb(jsonSpaces[:x]) } } func (e *jsonEncDriver) WriteArrayElem() { if e.e.c != containerArrayStart { e.e.encWr.writen1(',') } if e.d { e.writeIndent() } } func (e *jsonEncDriver) WriteMapElemKey() { if e.e.c != containerMapStart { e.e.encWr.writen1(',') } if e.d { e.writeIndent() } } func (e *jsonEncDriver) WriteMapElemValue() { if e.d { e.e.encWr.writen2(':', ' ') } else { e.e.encWr.writen1(':') } } func (e *jsonEncDriver) EncodeNil() { // We always encode nil as just null (never in quotes) // so we can easily decode if a nil in the json stream ie if initial token is n. e.e.encWr.writestr(jsonLits[jsonLitN : jsonLitN+4]) } func (e *jsonEncDriver) EncodeTime(t time.Time) { // Do NOT use MarshalJSON, as it allocates internally. // instead, we call AppendFormat directly, using our scratch buffer (e.b) if t.IsZero() { e.EncodeNil() } else { e.b[0] = '"' b := fmtTime(t, time.RFC3339Nano, e.b[1:1]) e.b[len(b)+1] = '"' e.e.encWr.writeb(e.b[:len(b)+2]) } } func (e *jsonEncDriver) EncodeExt(rv interface{}, basetype reflect.Type, xtag uint64, ext Ext) { if ext == SelfExt { e.e.encodeValue(baseRV(rv), e.h.fnNoExt(basetype)) } else if v := ext.ConvertExt(rv); v == nil { e.EncodeNil() } else { e.e.encode(v) } } func (e *jsonEncDriver) EncodeRawExt(re *RawExt) { // only encodes re.Value (never re.Data) if re.Value == nil { e.EncodeNil() } else { e.e.encode(re.Value) } } var jsonEncBoolStrs = [2][2]string{ {jsonLits[jsonLitF : jsonLitF+5], jsonLits[jsonLitT : jsonLitT+4]}, {jsonLits[jsonLitF-1 : jsonLitF+6], jsonLits[jsonLitT-1 : jsonLitT+5]}, } func (e *jsonEncDriver) EncodeBool(b bool) { e.e.encWr.writestr( jsonEncBoolStrs[bool2int(e.ks && e.e.c == containerMapKey)%2][bool2int(b)%2]) } // func (e *jsonEncDriver) EncodeBool(b bool) { // if e.ks && e.e.c == containerMapKey { // if b { // e.e.encWr.writestr(jsonLits[jsonLitT-1 : jsonLitT+5]) // } else { // e.e.encWr.writestr(jsonLits[jsonLitF-1 : jsonLitF+6]) // } // } else { // if b { // e.e.encWr.writestr(jsonLits[jsonLitT : jsonLitT+4]) // } else { // e.e.encWr.writestr(jsonLits[jsonLitF : jsonLitF+5]) // } // } // } func (e *jsonEncDriver) encodeFloat(f float64, bitsize, fmt byte, prec int8) { var blen uint if e.ks && e.e.c == containerMapKey { blen = 2 + uint(len(strconv.AppendFloat(e.b[1:1], f, fmt, int(prec), int(bitsize)))) // _ = e.b[:blen] e.b[0] = '"' e.b[blen-1] = '"' e.e.encWr.writeb(e.b[:blen]) } else { e.e.encWr.writeb(strconv.AppendFloat(e.b[:0], f, fmt, int(prec), int(bitsize))) } } func (e *jsonEncDriver) EncodeFloat64(f float64) { if math.IsNaN(f) || math.IsInf(f, 0) { e.EncodeNil() return } fmt, prec := jsonFloatStrconvFmtPrec64(f) e.encodeFloat(f, 64, fmt, prec) } func (e *jsonEncDriver) EncodeFloat32(f float32) { if math.IsNaN(float64(f)) || math.IsInf(float64(f), 0) { e.EncodeNil() return } fmt, prec := jsonFloatStrconvFmtPrec32(f) e.encodeFloat(float64(f), 32, fmt, prec) } func (e *jsonEncDriver) encodeUint(neg bool, quotes bool, u uint64) { // copied mostly from std library: strconv // this should only be called on 64bit OS. // const smallsString = jsonEncodeUintSmallsString var ss = jsonEncodeUintSmallsStringBytes // typically, 19 or 20 bytes sufficient for decimal encoding a uint64 // var a [24]byte var a = e.b[0:24] var i = uint(len(a)) if quotes { i-- setByteAt(a, i, '"') // a[i] = '"' } // u guaranteed to fit into a uint (as we are not 32bit OS) var is uint var us = uint(u) for us >= 100 { is = us % 100 * 2 us /= 100 i -= 2 setByteAt(a, i+1, byteAt(ss, is+1)) setByteAt(a, i, byteAt(ss, is)) // a[i+1] = smallsString[is+1] // a[i+0] = smallsString[is+0] } // us < 100 is = us * 2 i-- setByteAt(a, i, byteAt(ss, is+1)) // a[i] = smallsString[is+1] if us >= 10 { i-- setByteAt(a, i, byteAt(ss, is)) // a[i] = smallsString[is] } if neg { i-- setByteAt(a, i, '-') // a[i] = '-' } if quotes { i-- setByteAt(a, i, '"') // a[i] = '"' } e.e.encWr.writeb(a[i:]) } func (e *jsonEncDriver) EncodeInt(v int64) { quotes := e.is == 'A' || e.is == 'L' && (v > 1<<53 || v < -(1<<53)) || (e.ks && e.e.c == containerMapKey) if cpu32Bit { if quotes { blen := 2 + len(strconv.AppendInt(e.b[1:1], v, 10)) e.b[0] = '"' e.b[blen-1] = '"' e.e.encWr.writeb(e.b[:blen]) } else { e.e.encWr.writeb(strconv.AppendInt(e.b[:0], v, 10)) } return } if v < 0 { e.encodeUint(true, quotes, uint64(-v)) } else { e.encodeUint(false, quotes, uint64(v)) } } func (e *jsonEncDriver) EncodeUint(v uint64) { quotes := e.is == 'A' || e.is == 'L' && v > 1<<53 || (e.ks && e.e.c == containerMapKey) if cpu32Bit { // use strconv directly, as optimized encodeUint only works on 64-bit alone if quotes { blen := 2 + len(strconv.AppendUint(e.b[1:1], v, 10)) e.b[0] = '"' e.b[blen-1] = '"' e.e.encWr.writeb(e.b[:blen]) } else { e.e.encWr.writeb(strconv.AppendUint(e.b[:0], v, 10)) } return } e.encodeUint(false, quotes, v) } func (e *jsonEncDriver) EncodeString(v string) { if e.h.StringToRaw { e.EncodeStringBytesRaw(bytesView(v)) return } e.quoteStr(v) } func (e *jsonEncDriver) EncodeStringBytesRaw(v []byte) { // if encoding raw bytes and RawBytesExt is configured, use it to encode if v == nil { e.EncodeNil() return } if e.rawext { iv := e.h.RawBytesExt.ConvertExt(v) if iv == nil { e.EncodeNil() } else { e.e.encode(iv) } return } slen := base64.StdEncoding.EncodedLen(len(v)) + 2 // bs := e.e.blist.check(*e.buf, n)[:slen] // *e.buf = bs bs := e.e.blist.peek(slen, false) bs = bs[:slen] base64.StdEncoding.Encode(bs[1:], v) bs[len(bs)-1] = '"' bs[0] = '"' e.e.encWr.writeb(bs) } // indent is done as below: // - newline and indent are added before each mapKey or arrayElem // - newline and indent are added before each ending, // except there was no entry (so we can have {} or []) func (e *jsonEncDriver) WriteArrayStart(length int) { if e.d { e.dl++ } e.e.encWr.writen1('[') } func (e *jsonEncDriver) WriteArrayEnd() { if e.d { e.dl-- e.writeIndent() } e.e.encWr.writen1(']') } func (e *jsonEncDriver) WriteMapStart(length int) { if e.d { e.dl++ } e.e.encWr.writen1('{') } func (e *jsonEncDriver) WriteMapEnd() { if e.d { e.dl-- if e.e.c != containerMapStart { e.writeIndent() } } e.e.encWr.writen1('}') } func (e *jsonEncDriver) quoteStr(s string) { // adapted from std pkg encoding/json const hex = "0123456789abcdef" w := e.e.w() w.writen1('"') var i, start uint for i < uint(len(s)) { // encode all bytes < 0x20 (except \r, \n). // also encode < > & to prevent security holes when served to some browsers. // We optimize for ascii, by assumining that most characters are in the BMP // and natively consumed by json without much computation. // if 0x20 <= b && b != '\\' && b != '"' && b != '<' && b != '>' && b != '&' { // if (htmlasis && jsonCharSafeSet.isset(b)) || jsonCharHtmlSafeSet.isset(b) { if e.s.isset(s[i]) { i++ continue } // b := s[i] if s[i] < utf8.RuneSelf { if start < i { w.writestr(s[start:i]) } switch s[i] { case '\\', '"': w.writen2('\\', s[i]) case '\n': w.writen2('\\', 'n') case '\r': w.writen2('\\', 'r') case '\b': w.writen2('\\', 'b') case '\f': w.writen2('\\', 'f') case '\t': w.writen2('\\', 't') default: w.writestr(`\u00`) w.writen2(hex[s[i]>>4], hex[s[i]&0xF]) } i++ start = i continue } c, size := utf8.DecodeRuneInString(s[i:]) if c == utf8.RuneError && size == 1 { // meaning invalid encoding (so output as-is) if start < i { w.writestr(s[start:i]) } w.writestr(`\uFFFD`) i++ start = i continue } // U+2028 is LINE SEPARATOR. U+2029 is PARAGRAPH SEPARATOR. // Both technically valid JSON, but bomb on JSONP, so fix here *unconditionally*. if jsonEscapeMultiByteUnicodeSep && (c == '\u2028' || c == '\u2029') { if start < i { w.writestr(s[start:i]) } w.writestr(`\u202`) w.writen1(hex[c&0xF]) i += uint(size) start = i continue } i += uint(size) } if start < uint(len(s)) { w.writestr(s[start:]) } w.writen1('"') } func (e *jsonEncDriver) atEndOfEncode() { if e.h.TermWhitespace { var c byte = ' ' // default is that scalar is written, so output space if e.e.c != 0 { c = '\n' // for containers (map/list), output a newline } e.e.encWr.writen1(c) } } // ---------- type jsonDecState struct { rawext bool // rawext configured on the handle tok uint8 // used to store the token read right after skipWhiteSpace _ bool // found null _ byte // padding bstr [4]byte // scratch used for string \UXXX parsing // scratch buffer used for base64 decoding (DecodeBytes in reuseBuf mode), // or reading doubleQuoted string (DecodeStringAsBytes, DecodeNaked) buf *[]byte } func (x jsonDecState) captureState() interface{} { return x } func (x *jsonDecState) restoreState(v interface{}) { *x = v.(jsonDecState) } type jsonDecDriver struct { noBuiltInTypes decDriverNoopNumberHelper h *JsonHandle jsonDecState // se interfaceExtWrapper // ---- cpu cache line boundary? d Decoder } func (d *jsonDecDriver) descBd() (s string) { panic("descBd unsupported") } func (d *jsonDecDriver) decoder() *Decoder { return &d.d } func (d *jsonDecDriver) ReadMapStart() int { d.advance() if d.tok == 'n' { d.checkLit3([3]byte{'u', 'l', 'l'}, d.d.decRd.readn3()) return containerLenNil } if d.tok != '{' { d.d.errorf("read map - expect char '%c' but got char '%c'", '{', d.tok) } d.tok = 0 return containerLenUnknown } func (d *jsonDecDriver) ReadArrayStart() int { d.advance() if d.tok == 'n' { d.checkLit3([3]byte{'u', 'l', 'l'}, d.d.decRd.readn3()) return containerLenNil } if d.tok != '[' { d.d.errorf("read array - expect char '%c' but got char '%c'", '[', d.tok) } d.tok = 0 return containerLenUnknown } // MARKER: // We attempted making sure CheckBreak can be inlined, by moving the skipWhitespace // call to an explicit (noinline) function call. // However, this forces CheckBreak to always incur a function call if there was whitespace, // with no clear benefit. func (d *jsonDecDriver) CheckBreak() bool { d.advance() return d.tok == '}' || d.tok == ']' } func (d *jsonDecDriver) ReadArrayElem() { const xc uint8 = ',' if d.d.c != containerArrayStart { d.advance() if d.tok != xc { d.readDelimError(xc) } d.tok = 0 } } func (d *jsonDecDriver) ReadArrayEnd() { const xc uint8 = ']' d.advance() if d.tok != xc { d.readDelimError(xc) } d.tok = 0 } func (d *jsonDecDriver) ReadMapElemKey() { const xc uint8 = ',' if d.d.c != containerMapStart { d.advance() if d.tok != xc { d.readDelimError(xc) } d.tok = 0 } } func (d *jsonDecDriver) ReadMapElemValue() { const xc uint8 = ':' d.advance() if d.tok != xc { d.readDelimError(xc) } d.tok = 0 } func (d *jsonDecDriver) ReadMapEnd() { const xc uint8 = '}' d.advance() if d.tok != xc { d.readDelimError(xc) } d.tok = 0 } func (d *jsonDecDriver) readDelimError(xc uint8) { d.d.errorf("read json delimiter - expect char '%c' but got char '%c'", xc, d.tok) } // MARKER: checkLit takes the readn(3|4) result as a parameter so they can be inlined. // We pass the array directly to errorf, as passing slice pushes past inlining threshold, // and passing slice also might cause allocation of the bs array on the heap. func (d *jsonDecDriver) checkLit3(got, expect [3]byte) { d.tok = 0 if jsonValidateSymbols && got != expect { d.d.errorf("expecting %s: got %s", expect, got) } } func (d *jsonDecDriver) checkLit4(got, expect [4]byte) { d.tok = 0 if jsonValidateSymbols && got != expect { d.d.errorf("expecting %s: got %s", expect, got) } } func (d *jsonDecDriver) skipWhitespace() { d.tok = d.d.decRd.skipWhitespace() } func (d *jsonDecDriver) advance() { if d.tok == 0 { d.skipWhitespace() } } func (d *jsonDecDriver) nextValueBytes(v []byte) []byte { v, cursor := d.nextValueBytesR(v) decNextValueBytesHelper{d: &d.d}.bytesRdV(&v, cursor) return v } func (d *jsonDecDriver) nextValueBytesR(v0 []byte) (v []byte, cursor uint) { v = v0 var h = decNextValueBytesHelper{d: &d.d} dr := &d.d.decRd consumeString := func() { TOP: bs := dr.jsonReadAsisChars() h.appendN(&v, bs...) if bs[len(bs)-1] != '"' { // last char is '\', so consume next one and try again h.append1(&v, dr.readn1()) goto TOP } } d.advance() // ignore leading whitespace cursor = d.d.rb.c - 1 // cursor starts just before non-whitespace token switch d.tok { default: h.appendN(&v, dr.jsonReadNum()...) case 'n': d.checkLit3([3]byte{'u', 'l', 'l'}, d.d.decRd.readn3()) h.appendS(&v, jsonLits[jsonLitN:jsonLitN+4]) case 'f': d.checkLit4([4]byte{'a', 'l', 's', 'e'}, d.d.decRd.readn4()) h.appendS(&v, jsonLits[jsonLitF:jsonLitF+5]) case 't': d.checkLit3([3]byte{'r', 'u', 'e'}, d.d.decRd.readn3()) h.appendS(&v, jsonLits[jsonLitT:jsonLitT+4]) case '"': h.append1(&v, '"') consumeString() case '{', '[': var elem struct{} var stack []struct{} stack = append(stack, elem) h.append1(&v, d.tok) for len(stack) != 0 { c := dr.readn1() h.append1(&v, c) switch c { case '"': consumeString() case '{', '[': stack = append(stack, elem) case '}', ']': stack = stack[:len(stack)-1] } } } d.tok = 0 return } func (d *jsonDecDriver) TryNil() bool { d.advance() // we shouldn't try to see if quoted "null" was here, right? // only the plain string: `null` denotes a nil (ie not quotes) if d.tok == 'n' { d.checkLit3([3]byte{'u', 'l', 'l'}, d.d.decRd.readn3()) return true } return false } func (d *jsonDecDriver) DecodeBool() (v bool) { d.advance() // bool can be in quotes if and only if it's a map key fquot := d.d.c == containerMapKey && d.tok == '"' if fquot { d.tok = d.d.decRd.readn1() } switch d.tok { case 'f': d.checkLit4([4]byte{'a', 'l', 's', 'e'}, d.d.decRd.readn4()) // v = false case 't': d.checkLit3([3]byte{'r', 'u', 'e'}, d.d.decRd.readn3()) v = true case 'n': d.checkLit3([3]byte{'u', 'l', 'l'}, d.d.decRd.readn3()) // v = false default: d.d.errorf("decode bool: got first char %c", d.tok) // v = false // "unreachable" } if fquot { d.d.decRd.readn1() } return } func (d *jsonDecDriver) DecodeTime() (t time.Time) { // read string, and pass the string into json.unmarshal d.advance() if d.tok == 'n' { d.checkLit3([3]byte{'u', 'l', 'l'}, d.d.decRd.readn3()) return } d.ensureReadingString() bs := d.readUnescapedString() t, err := time.Parse(time.RFC3339, stringView(bs)) d.d.onerror(err) return } func (d *jsonDecDriver) ContainerType() (vt valueType) { // check container type by checking the first char d.advance() // optimize this, so we don't do 4 checks but do one computation. // return jsonContainerSet[d.tok] // ContainerType is mostly called for Map and Array, // so this conditional is good enough (max 2 checks typically) if d.tok == '{' { return valueTypeMap } else if d.tok == '[' { return valueTypeArray } else if d.tok == 'n' { d.checkLit3([3]byte{'u', 'l', 'l'}, d.d.decRd.readn3()) return valueTypeNil } else if d.tok == '"' { return valueTypeString } return valueTypeUnset } func (d *jsonDecDriver) decNumBytes() (bs []byte) { d.advance() dr := &d.d.decRd if d.tok == '"' { bs = dr.readUntil('"') } else if d.tok == 'n' { d.checkLit3([3]byte{'u', 'l', 'l'}, dr.readn3()) } else { if jsonManualInlineDecRdInHotZones { if dr.bytes { bs = dr.rb.jsonReadNum() } else { bs = dr.ri.jsonReadNum() } } else { bs = dr.jsonReadNum() } } d.tok = 0 return } func (d *jsonDecDriver) DecodeUint64() (u uint64) { b := d.decNumBytes() u, neg, ok := parseInteger_bytes(b) if neg { d.d.errorf("negative number cannot be decoded as uint64") } if !ok { d.d.onerror(strconvParseErr(b, "ParseUint")) } return } func (d *jsonDecDriver) DecodeInt64() (v int64) { b := d.decNumBytes() u, neg, ok := parseInteger_bytes(b) if !ok { d.d.onerror(strconvParseErr(b, "ParseInt")) } if chkOvf.Uint2Int(u, neg) { d.d.errorf("overflow decoding number from %s", b) } if neg { v = -int64(u) } else { v = int64(u) } return } func (d *jsonDecDriver) DecodeFloat64() (f float64) { var err error bs := d.decNumBytes() if len(bs) == 0 { return } f, err = parseFloat64(bs) d.d.onerror(err) return } func (d *jsonDecDriver) DecodeFloat32() (f float32) { var err error bs := d.decNumBytes() if len(bs) == 0 { return } f, err = parseFloat32(bs) d.d.onerror(err) return } func (d *jsonDecDriver) DecodeExt(rv interface{}, basetype reflect.Type, xtag uint64, ext Ext) { d.advance() if d.tok == 'n' { d.checkLit3([3]byte{'u', 'l', 'l'}, d.d.decRd.readn3()) return } if ext == nil { re := rv.(*RawExt) re.Tag = xtag d.d.decode(&re.Value) } else if ext == SelfExt { d.d.decodeValue(baseRV(rv), d.h.fnNoExt(basetype)) } else { d.d.interfaceExtConvertAndDecode(rv, ext) } } func (d *jsonDecDriver) decBytesFromArray(bs []byte) []byte { if bs != nil { bs = bs[:0] } d.tok = 0 bs = append(bs, uint8(d.DecodeUint64())) d.tok = d.d.decRd.skipWhitespace() // skip(&whitespaceCharBitset) for d.tok != ']' { if d.tok != ',' { d.d.errorf("read array element - expect char '%c' but got char '%c'", ',', d.tok) } d.tok = 0 bs = append(bs, uint8(chkOvf.UintV(d.DecodeUint64(), 8))) d.tok = d.d.decRd.skipWhitespace() // skip(&whitespaceCharBitset) } d.tok = 0 return bs } func (d *jsonDecDriver) DecodeBytes(bs []byte) (bsOut []byte) { d.d.decByteState = decByteStateNone d.advance() if d.tok == 'n' { d.checkLit3([3]byte{'u', 'l', 'l'}, d.d.decRd.readn3()) return nil } // if decoding into raw bytes, and the RawBytesExt is configured, use it to decode. if d.rawext { bsOut = bs d.d.interfaceExtConvertAndDecode(&bsOut, d.h.RawBytesExt) return } // check if an "array" of uint8's (see ContainerType for how to infer if an array) if d.tok == '[' { // bsOut, _ = fastpathTV.DecSliceUint8V(bs, true, d.d) if bs == nil { d.d.decByteState = decByteStateReuseBuf bs = d.d.b[:] } return d.decBytesFromArray(bs) } // base64 encodes []byte{} as "", and we encode nil []byte as null. // Consequently, base64 should decode null as a nil []byte, and "" as an empty []byte{}. d.ensureReadingString() bs1 := d.readUnescapedString() slen := base64.StdEncoding.DecodedLen(len(bs1)) if slen == 0 { bsOut = []byte{} } else if slen <= cap(bs) { bsOut = bs[:slen] } else if bs == nil { d.d.decByteState = decByteStateReuseBuf bsOut = d.d.blist.check(*d.buf, slen) bsOut = bsOut[:slen] *d.buf = bsOut } else { bsOut = make([]byte, slen) } slen2, err := base64.StdEncoding.Decode(bsOut, bs1) if err != nil { d.d.errorf("error decoding base64 binary '%s': %v", bs1, err) } if slen != slen2 { bsOut = bsOut[:slen2] } return } func (d *jsonDecDriver) DecodeStringAsBytes() (s []byte) { d.d.decByteState = decByteStateNone d.advance() // common case - hoist outside the switch statement if d.tok == '"' { return d.dblQuoteStringAsBytes() } // handle non-string scalar: null, true, false or a number switch d.tok { case 'n': d.checkLit3([3]byte{'u', 'l', 'l'}, d.d.decRd.readn3()) return nil // []byte{} case 'f': d.checkLit4([4]byte{'a', 'l', 's', 'e'}, d.d.decRd.readn4()) return jsonLitb[jsonLitF : jsonLitF+5] case 't': d.checkLit3([3]byte{'r', 'u', 'e'}, d.d.decRd.readn3()) return jsonLitb[jsonLitT : jsonLitT+4] default: // try to parse a valid number d.tok = 0 return d.d.decRd.jsonReadNum() } } func (d *jsonDecDriver) ensureReadingString() { if d.tok != '"' { d.d.errorf("expecting string starting with '\"'; got '%c'", d.tok) } } func (d *jsonDecDriver) readUnescapedString() (bs []byte) { // d.ensureReadingString() bs = d.d.decRd.readUntil('"') d.tok = 0 return } func (d *jsonDecDriver) dblQuoteStringAsBytes() (buf []byte) { checkUtf8 := d.h.ValidateUnicode d.d.decByteState = decByteStateNone // use a local buf variable, so we don't do pointer chasing within loop buf = (*d.buf)[:0] dr := &d.d.decRd d.tok = 0 var bs []byte var c byte var firstTime bool = true for { if firstTime { firstTime = false if dr.bytes { bs = dr.rb.jsonReadAsisChars() if bs[len(bs)-1] == '"' { d.d.decByteState = decByteStateZerocopy return bs[:len(bs)-1] } goto APPEND } } if jsonManualInlineDecRdInHotZones { if dr.bytes { bs = dr.rb.jsonReadAsisChars() } else { bs = dr.ri.jsonReadAsisChars() } } else { bs = dr.jsonReadAsisChars() } APPEND: _ = bs[0] // bounds check hint - slice must be > 0 elements buf = append(buf, bs[:len(bs)-1]...) c = bs[len(bs)-1] if c == '"' { break } // c is now '\' c = dr.readn1() switch c { case '"', '\\', '/', '\'': buf = append(buf, c) case 'b': buf = append(buf, '\b') case 'f': buf = append(buf, '\f') case 'n': buf = append(buf, '\n') case 'r': buf = append(buf, '\r') case 't': buf = append(buf, '\t') case 'u': rr := d.appendStringAsBytesSlashU() if checkUtf8 && rr == unicode.ReplacementChar { d.d.errorf("invalid UTF-8 character found after: %s", buf) } buf = append(buf, d.bstr[:utf8.EncodeRune(d.bstr[:], rr)]...) default: *d.buf = buf d.d.errorf("unsupported escaped value: %c", c) } } *d.buf = buf d.d.decByteState = decByteStateReuseBuf return } func (d *jsonDecDriver) appendStringAsBytesSlashU() (r rune) { var rr uint32 var csu [2]byte var cs [4]byte = d.d.decRd.readn4() if rr = jsonSlashURune(cs); rr == unicode.ReplacementChar { return unicode.ReplacementChar } r = rune(rr) if utf16.IsSurrogate(r) { csu = d.d.decRd.readn2() cs = d.d.decRd.readn4() if csu[0] == '\\' && csu[1] == 'u' { if rr = jsonSlashURune(cs); rr == unicode.ReplacementChar { return unicode.ReplacementChar } return utf16.DecodeRune(r, rune(rr)) } return unicode.ReplacementChar } return } func jsonSlashURune(cs [4]byte) (rr uint32) { for _, c := range cs { // best to use explicit if-else // - not a table, etc which involve memory loads, array lookup with bounds checks, etc if c >= '0' && c <= '9' { rr = rr*16 + uint32(c-jsonU4Chk2) } else if c >= 'a' && c <= 'f' { rr = rr*16 + uint32(c-jsonU4Chk1) } else if c >= 'A' && c <= 'F' { rr = rr*16 + uint32(c-jsonU4Chk0) } else { return unicode.ReplacementChar } } return } func (d *jsonDecDriver) nakedNum(z *fauxUnion, bs []byte) (err error) { // Note: nakedNum is NEVER called with a zero-length []byte if d.h.PreferFloat { z.v = valueTypeFloat z.f, err = parseFloat64(bs) } else { err = parseNumber(bs, z, d.h.SignedInteger) } return } func (d *jsonDecDriver) DecodeNaked() { z := d.d.naked() d.advance() var bs []byte switch d.tok { case 'n': d.checkLit3([3]byte{'u', 'l', 'l'}, d.d.decRd.readn3()) z.v = valueTypeNil case 'f': d.checkLit4([4]byte{'a', 'l', 's', 'e'}, d.d.decRd.readn4()) z.v = valueTypeBool z.b = false case 't': d.checkLit3([3]byte{'r', 'u', 'e'}, d.d.decRd.readn3()) z.v = valueTypeBool z.b = true case '{': z.v = valueTypeMap // don't consume. kInterfaceNaked will call ReadMapStart case '[': z.v = valueTypeArray // don't consume. kInterfaceNaked will call ReadArrayStart case '"': // if a string, and MapKeyAsString, then try to decode it as a bool or number first bs = d.dblQuoteStringAsBytes() if jsonNakedBoolNullInQuotedStr && d.h.MapKeyAsString && len(bs) > 0 && d.d.c == containerMapKey { switch string(bs) { // case "null": // nil is never quoted // z.v = valueTypeNil case "true": z.v = valueTypeBool z.b = true case "false": z.v = valueTypeBool z.b = false default: // check if a number: float, int or uint if err := d.nakedNum(z, bs); err != nil { z.v = valueTypeString z.s = d.d.stringZC(bs) } } } else { z.v = valueTypeString z.s = d.d.stringZC(bs) } default: // number bs = d.d.decRd.jsonReadNum() d.tok = 0 if len(bs) == 0 { d.d.errorf("decode number from empty string") } if err := d.nakedNum(z, bs); err != nil { d.d.errorf("decode number from %s: %v", bs, err) } } } //---------------------- // JsonHandle is a handle for JSON encoding format. // // Json is comprehensively supported: // - decodes numbers into interface{} as int, uint or float64 // based on how the number looks and some config parameters e.g. PreferFloat, SignedInt, etc. // - decode integers from float formatted numbers e.g. 1.27e+8 // - decode any json value (numbers, bool, etc) from quoted strings // - configurable way to encode/decode []byte . // by default, encodes and decodes []byte using base64 Std Encoding // - UTF-8 support for encoding and decoding // // It has better performance than the json library in the standard library, // by leveraging the performance improvements of the codec library. // // In addition, it doesn't read more bytes than necessary during a decode, which allows // reading multiple values from a stream containing json and non-json content. // For example, a user can read a json value, then a cbor value, then a msgpack value, // all from the same stream in sequence. // // Note that, when decoding quoted strings, invalid UTF-8 or invalid UTF-16 surrogate pairs are // not treated as an error. Instead, they are replaced by the Unicode replacement character U+FFFD. // // Note also that the float values for NaN, +Inf or -Inf are encoded as null, // as suggested by NOTE 4 of the ECMA-262 ECMAScript Language Specification 5.1 edition. // see http://www.ecma-international.org/publications/files/ECMA-ST/Ecma-262.pdf . // // Note the following behaviour differences vs std-library encoding/json package: // - struct field names matched in case-sensitive manner type JsonHandle struct { textEncodingType BasicHandle // Indent indicates how a value is encoded. // - If positive, indent by that number of spaces. // - If negative, indent by that number of tabs. Indent int8 // IntegerAsString controls how integers (signed and unsigned) are encoded. // // Per the JSON Spec, JSON numbers are 64-bit floating point numbers. // Consequently, integers > 2^53 cannot be represented as a JSON number without losing precision. // This can be mitigated by configuring how to encode integers. // // IntegerAsString interpretes the following values: // - if 'L', then encode integers > 2^53 as a json string. // - if 'A', then encode all integers as a json string // containing the exact integer representation as a decimal. // - else encode all integers as a json number (default) IntegerAsString byte // HTMLCharsAsIs controls how to encode some special characters to html: < > & // // By default, we encode them as \uXXX // to prevent security holes when served from some browsers. HTMLCharsAsIs bool // PreferFloat says that we will default to decoding a number as a float. // If not set, we will examine the characters of the number and decode as an // integer type if it doesn't have any of the characters [.eE]. PreferFloat bool // TermWhitespace says that we add a whitespace character // at the end of an encoding. // // The whitespace is important, especially if using numbers in a context // where multiple items are written to a stream. TermWhitespace bool // MapKeyAsString says to encode all map keys as strings. // // Use this to enforce strict json output. // The only caveat is that nil value is ALWAYS written as null (never as "null") MapKeyAsString bool // _ uint64 // padding (cache line) // Note: below, we store hardly-used items e.g. RawBytesExt. // These values below may straddle a cache line, but they are hardly-used, // so shouldn't contribute to false-sharing except in rare cases. // RawBytesExt, if configured, is used to encode and decode raw bytes in a custom way. // If not configured, raw bytes are encoded to/from base64 text. RawBytesExt InterfaceExt } func (h *JsonHandle) isJson() bool { return true } // Name returns the name of the handle: json func (h *JsonHandle) Name() string { return "json" } func (h *JsonHandle) desc(bd byte) string { return string(bd) } func (h *JsonHandle) typical() bool { return h.Indent == 0 && !h.MapKeyAsString && h.IntegerAsString != 'A' && h.IntegerAsString != 'L' } func (h *JsonHandle) newEncDriver() encDriver { var e = &jsonEncDriver{h: h} // var x []byte // e.buf = &x e.e.e = e e.e.js = true e.e.init(h) e.reset() return e } func (h *JsonHandle) newDecDriver() decDriver { var d = &jsonDecDriver{h: h} var x []byte d.buf = &x d.d.d = d d.d.js = true d.d.jsms = h.MapKeyAsString d.d.init(h) d.reset() return d } func (e *jsonEncDriver) resetState() { e.dl = 0 } func (e *jsonEncDriver) reset() { e.resetState() // (htmlasis && jsonCharSafeSet.isset(b)) || jsonCharHtmlSafeSet.isset(b) // cache values from the handle e.typical = e.h.typical() if e.h.HTMLCharsAsIs { e.s = &jsonCharSafeSet } else { e.s = &jsonCharHtmlSafeSet } e.rawext = e.h.RawBytesExt != nil e.di = int8(e.h.Indent) e.d = e.h.Indent != 0 e.ks = e.h.MapKeyAsString e.is = e.h.IntegerAsString } func (d *jsonDecDriver) resetState() { *d.buf = d.d.blist.check(*d.buf, 256) d.tok = 0 } func (d *jsonDecDriver) reset() { d.resetState() d.rawext = d.h.RawBytesExt != nil } func jsonFloatStrconvFmtPrec64(f float64) (fmt byte, prec int8) { fmt = 'f' prec = -1 fbits := math.Float64bits(f) abs := math.Float64frombits(fbits &^ (1 << 63)) if abs == 0 || abs == 1 { prec = 1 } else if abs < 1e-6 || abs >= 1e21 { fmt = 'e' } else if noFrac64(fbits) { prec = 1 } return } func jsonFloatStrconvFmtPrec32(f float32) (fmt byte, prec int8) { fmt = 'f' prec = -1 // directly handle Modf (to get fractions) and Abs (to get absolute) fbits := math.Float32bits(f) abs := math.Float32frombits(fbits &^ (1 << 31)) if abs == 0 || abs == 1 { prec = 1 } else if abs < 1e-6 || abs >= 1e21 { fmt = 'e' } else if noFrac32(fbits) { prec = 1 } return } var _ decDriverContainerTracker = (*jsonDecDriver)(nil) var _ encDriverContainerTracker = (*jsonEncDriver)(nil) var _ decDriver = (*jsonDecDriver)(nil) var _ encDriver = (*jsonEncDriver)(nil)