vendor tsgo

2026-07-09 16:50:43 -04:00
parent c06ea2e5a4
commit 98978e4930
5804 changed files with 1556156 additions and 101 deletions
--- a/tools/tsgo/vendor/github.com/go-json-experiment/json/arshal_time.go
+++ b/tools/tsgo/vendor/github.com/go-json-experiment/json/arshal_time.go
@@ -0,0 +1,795 @@
+// Copyright 2020 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build !goexperiment.jsonv2 || !go1.25
+
+package json
+
+import (
+	"bytes"
+	"cmp"
+	"errors"
+	"fmt"
+	"math"
+	"math/bits"
+	"reflect"
+	"strconv"
+	"strings"
+	"time"
+
+	"github.com/go-json-experiment/json/internal"
+	"github.com/go-json-experiment/json/internal/jsonflags"
+	"github.com/go-json-experiment/json/internal/jsonopts"
+	"github.com/go-json-experiment/json/internal/jsonwire"
+	"github.com/go-json-experiment/json/jsontext"
+)
+
+var (
+	timeDurationType = reflect.TypeFor[time.Duration]()
+	timeTimeType     = reflect.TypeFor[time.Time]()
+)
+
+func makeTimeArshaler(fncs *arshaler, t reflect.Type) *arshaler {
+	// Ideally, time types would implement MarshalerTo and UnmarshalerFrom,
+	// but that would incur a dependency on package json from package time.
+	// Given how widely used time is, it is more acceptable that we incur a
+	// dependency on time from json.
+	//
+	// Injecting the arshaling functionality like this will not be identical
+	// to actually declaring methods on the time types since embedding of the
+	// time types will not be able to forward this functionality.
+	switch t {
+	case timeDurationType:
+		fncs.nonDefault = true
+		marshalNano := fncs.marshal
+		fncs.marshal = func(enc *jsontext.Encoder, va addressableValue, mo *jsonopts.Struct) error {
+			xe := export.Encoder(enc)
+			var m durationArshaler
+			if mo.Flags.Has(jsonflags.FormatTag) {
+				if !m.initFormat(mo.Format) {
+					return newInvalidFormatError(enc, t)
+				}
+			} else if mo.Flags.Get(jsonflags.FormatDurationAsNano) {
+				return marshalNano(enc, va, mo)
+			} else {
+				// TODO(https://go.dev/issue/71631): Decide on default duration representation.
+				var workaround string
+				if mo.Flags.Get(jsonflags.FormatTagSupported) {
+					workaround = "; specify an explicit format"
+				}
+				return newMarshalErrorBefore(enc, t, errors.New("no default representation"+workaround))
+			}
+			if mo.Flags.Get(jsonflags.StringTag) && !m.isNumeric() && !mo.Flags.Get(jsonflags.ReportErrorsWithLegacySemantics) {
+				return newMarshalErrorBefore(enc, t, errInvalidStringTag)
+			}
+
+			m.td, _ = reflect.TypeAssert[time.Duration](va.Value)
+			k := stringOrNumberKind(!m.isNumeric() || xe.Tokens.Last.NeedObjectName() || mo.Flags.Get(jsonflags.StringifyNumbers|jsonflags.StringTag))
+			if err := xe.AppendRaw(k, true, m.appendMarshal); err != nil {
+				if !isSyntacticError(err) && !export.IsIOError(err) {
+					err = newMarshalErrorBefore(enc, t, err)
+				}
+				return err
+			}
+			return nil
+		}
+		unmarshalNano := fncs.unmarshal
+		fncs.unmarshal = func(dec *jsontext.Decoder, va addressableValue, uo *jsonopts.Struct) error {
+			xd := export.Decoder(dec)
+			var u durationArshaler
+			if uo.Flags.Has(jsonflags.FormatTag) {
+				if !u.initFormat(uo.Format) {
+					return newInvalidFormatError(dec, t)
+				}
+			} else if uo.Flags.Get(jsonflags.FormatDurationAsNano) {
+				return unmarshalNano(dec, va, uo)
+			} else {
+				// TODO(https://go.dev/issue/71631): Decide on default duration representation.
+				var workaround string
+				if uo.Flags.Get(jsonflags.FormatTagSupported) {
+					workaround = "; specify an explicit format"
+				}
+				return newUnmarshalErrorBeforeWithSkipping(dec, t, errors.New("no default representation"+workaround))
+			}
+			if uo.Flags.Get(jsonflags.StringTag) && !u.isNumeric() && !uo.Flags.Get(jsonflags.ReportErrorsWithLegacySemantics) {
+				return newUnmarshalErrorBeforeWithSkipping(dec, t, errInvalidStringTag)
+			}
+
+			stringify := !u.isNumeric() || xd.Tokens.Last.NeedObjectName() || uo.Flags.Get(jsonflags.StringifyNumbers|jsonflags.StringTag)
+			var flags jsonwire.ValueFlags
+			td, _ := reflect.TypeAssert[*time.Duration](va.Addr())
+			val, err := xd.ReadValue(&flags)
+			if err != nil {
+				return err
+			}
+			switch k := val.Kind(); k {
+			case 'n':
+				if !uo.Flags.Get(jsonflags.MergeWithLegacySemantics) {
+					*td = time.Duration(0)
+				}
+				return nil
+			case '"':
+				if !stringify {
+					break
+				}
+				val = jsonwire.UnquoteMayCopy(val, flags.IsVerbatim())
+				if err := u.unmarshal(val); err != nil {
+					return newUnmarshalErrorAfter(dec, t, err)
+				}
+				*td = u.td
+				return nil
+			case '0':
+				if stringify {
+					break
+				}
+				if err := u.unmarshal(val); err != nil {
+					return newUnmarshalErrorAfter(dec, t, err)
+				}
+				*td = u.td
+				return nil
+			}
+			return newUnmarshalErrorAfter(dec, t, nil)
+		}
+	case timeTimeType:
+		fncs.nonDefault = true
+		fncs.marshal = func(enc *jsontext.Encoder, va addressableValue, mo *jsonopts.Struct) (err error) {
+			xe := export.Encoder(enc)
+			var m timeArshaler
+			if mo.Flags.Has(jsonflags.FormatTag) {
+				if !m.initFormat(mo.Format) {
+					return newInvalidFormatError(enc, t)
+				}
+			}
+			if mo.Flags.Get(jsonflags.StringTag) && !m.isNumeric() && !mo.Flags.Get(jsonflags.ReportErrorsWithLegacySemantics) {
+				return newMarshalErrorBefore(enc, t, errInvalidStringTag)
+			}
+
+			m.tt, _ = reflect.TypeAssert[time.Time](va.Value)
+			k := stringOrNumberKind(!m.isNumeric() || xe.Tokens.Last.NeedObjectName() || mo.Flags.Get(jsonflags.StringifyNumbers|jsonflags.StringTag))
+			if err := xe.AppendRaw(k, !m.hasCustomFormat(), m.appendMarshal); err != nil {
+				if mo.Flags.Get(jsonflags.ReportErrorsWithLegacySemantics) {
+					return internal.NewMarshalerError(va.Addr().Interface(), err, "MarshalJSON") // unlike unmarshal, always wrapped
+				}
+				if !isSyntacticError(err) && !export.IsIOError(err) {
+					err = newMarshalErrorBefore(enc, t, err)
+				}
+				return err
+			}
+			return nil
+		}
+		fncs.unmarshal = func(dec *jsontext.Decoder, va addressableValue, uo *jsonopts.Struct) (err error) {
+			xd := export.Decoder(dec)
+			var u timeArshaler
+			if uo.Flags.Has(jsonflags.FormatTag) {
+				if !u.initFormat(uo.Format) {
+					return newInvalidFormatError(dec, t)
+				}
+			} else if uo.Flags.Get(jsonflags.ParseTimeWithLooseRFC3339) {
+				u.looseRFC3339 = true
+			}
+			if uo.Flags.Get(jsonflags.StringTag) && !u.isNumeric() && !uo.Flags.Get(jsonflags.ReportErrorsWithLegacySemantics) {
+				return newUnmarshalErrorBeforeWithSkipping(dec, t, errInvalidStringTag)
+			}
+
+			stringify := !u.isNumeric() || xd.Tokens.Last.NeedObjectName() || uo.Flags.Get(jsonflags.StringifyNumbers|jsonflags.StringTag)
+			var flags jsonwire.ValueFlags
+			tt, _ := reflect.TypeAssert[*time.Time](va.Addr())
+			val, err := xd.ReadValue(&flags)
+			if err != nil {
+				return err
+			}
+			switch k := val.Kind(); k {
+			case 'n':
+				if !uo.Flags.Get(jsonflags.MergeWithLegacySemantics) {
+					*tt = time.Time{}
+				}
+				return nil
+			case '"':
+				if !stringify {
+					break
+				}
+				val = jsonwire.UnquoteMayCopy(val, flags.IsVerbatim())
+				if err := u.unmarshal(val); err != nil {
+					if uo.Flags.Get(jsonflags.ReportErrorsWithLegacySemantics) {
+						return err // unlike marshal, never wrapped
+					}
+					return newUnmarshalErrorAfter(dec, t, err)
+				}
+				*tt = u.tt
+				return nil
+			case '0':
+				if stringify {
+					break
+				}
+				if err := u.unmarshal(val); err != nil {
+					if uo.Flags.Get(jsonflags.ReportErrorsWithLegacySemantics) {
+						return err // unlike marshal, never wrapped
+					}
+					return newUnmarshalErrorAfter(dec, t, err)
+				}
+				*tt = u.tt
+				return nil
+			}
+			return newUnmarshalErrorAfter(dec, t, nil)
+		}
+	}
+	return fncs
+}
+
+type durationArshaler struct {
+	td time.Duration
+
+	// base records the representation where:
+	//   - 0 uses time.Duration.String
+	//   - 1e0, 1e3, 1e6, or 1e9 use a decimal encoding of the duration as
+	//     nanoseconds, microseconds, milliseconds, or seconds.
+	//   - 8601 uses ISO 8601
+	base uint64
+}
+
+func (a *durationArshaler) initFormat(format string) (ok bool) {
+	switch format {
+	case "units":
+		a.base = 0
+	case "sec":
+		a.base = 1e9
+	case "milli":
+		a.base = 1e6
+	case "micro":
+		a.base = 1e3
+	case "nano":
+		a.base = 1e0
+	case "iso8601":
+		a.base = 8601
+	default:
+		return false
+	}
+	return true
+}
+
+func (a *durationArshaler) isNumeric() bool {
+	return a.base != 0 && a.base != 8601
+}
+
+func (a *durationArshaler) appendMarshal(b []byte) ([]byte, error) {
+	switch a.base {
+	case 0:
+		return append(b, a.td.String()...), nil
+	case 8601:
+		return appendDurationISO8601(b, a.td), nil
+	default:
+		return appendDurationBase10(b, a.td, a.base), nil
+	}
+}
+
+func (a *durationArshaler) unmarshal(b []byte) (err error) {
+	switch a.base {
+	case 0:
+		a.td, err = time.ParseDuration(string(b))
+	case 8601:
+		a.td, err = parseDurationISO8601(b)
+	default:
+		a.td, err = parseDurationBase10(b, a.base)
+	}
+	return err
+}
+
+type timeArshaler struct {
+	tt time.Time
+
+	// base records the representation where:
+	//   - 0 uses RFC 3339 encoding of the timestamp
+	//   - 1e0, 1e3, 1e6, or 1e9 use a decimal encoding of the timestamp as
+	//     seconds, milliseconds, microseconds, or nanoseconds since Unix epoch.
+	//   - math.MaxUint uses time.Time.Format to encode the timestamp
+	base   uint64
+	format string // time format passed to time.Parse
+
+	looseRFC3339 bool
+}
+
+func (a *timeArshaler) initFormat(format string) bool {
+	// We assume that an exported constant in the time package will
+	// always start with an uppercase ASCII letter.
+	if len(format) == 0 {
+		return false
+	}
+	a.base = math.MaxUint // implies custom format
+	if c := format[0]; !('a' <= c && c <= 'z') && !('A' <= c && c <= 'Z') {
+		a.format = format
+		return true
+	}
+	switch format {
+	case "ANSIC":
+		a.format = time.ANSIC
+	case "UnixDate":
+		a.format = time.UnixDate
+	case "RubyDate":
+		a.format = time.RubyDate
+	case "RFC822":
+		a.format = time.RFC822
+	case "RFC822Z":
+		a.format = time.RFC822Z
+	case "RFC850":
+		a.format = time.RFC850
+	case "RFC1123":
+		a.format = time.RFC1123
+	case "RFC1123Z":
+		a.format = time.RFC1123Z
+	case "RFC3339":
+		a.base = 0
+		a.format = time.RFC3339
+	case "RFC3339Nano":
+		a.base = 0
+		a.format = time.RFC3339Nano
+	case "Kitchen":
+		a.format = time.Kitchen
+	case "Stamp":
+		a.format = time.Stamp
+	case "StampMilli":
+		a.format = time.StampMilli
+	case "StampMicro":
+		a.format = time.StampMicro
+	case "StampNano":
+		a.format = time.StampNano
+	case "DateTime":
+		a.format = time.DateTime
+	case "DateOnly":
+		a.format = time.DateOnly
+	case "TimeOnly":
+		a.format = time.TimeOnly
+	case "unix":
+		a.base = 1e0
+	case "unixmilli":
+		a.base = 1e3
+	case "unixmicro":
+		a.base = 1e6
+	case "unixnano":
+		a.base = 1e9
+	default:
+		// Reject any Go identifier in case new constants are supported.
+		if strings.TrimFunc(format, isLetterOrDigit) == "" {
+			return false
+		}
+		a.format = format
+	}
+	return true
+}
+
+func (a *timeArshaler) isNumeric() bool {
+	return int(a.base) > 0
+}
+
+func (a *timeArshaler) hasCustomFormat() bool {
+	return a.base == math.MaxUint
+}
+
+func (a *timeArshaler) appendMarshal(b []byte) ([]byte, error) {
+	switch a.base {
+	case 0:
+		format := cmp.Or(a.format, time.RFC3339Nano)
+		n0 := len(b)
+		b = a.tt.AppendFormat(b, format)
+		// Not all Go timestamps can be represented as valid RFC 3339.
+		// Explicitly check for these edge cases.
+		// See https://go.dev/issue/4556 and https://go.dev/issue/54580.
+		switch b := b[n0:]; {
+		case b[len("9999")] != '-': // year must be exactly 4 digits wide
+			return b, errors.New("year outside of range [0,9999]")
+		case b[len(b)-1] != 'Z':
+			c := b[len(b)-len("Z07:00")]
+			if ('0' <= c && c <= '9') || parseDec2(b[len(b)-len("07:00"):]) >= 24 {
+				return b, errors.New("timezone hour outside of range [0,23]")
+			}
+		}
+		return b, nil
+	case math.MaxUint:
+		return a.tt.AppendFormat(b, a.format), nil
+	default:
+		return appendTimeUnix(b, a.tt, a.base), nil
+	}
+}
+
+func (a *timeArshaler) unmarshal(b []byte) (err error) {
+	switch a.base {
+	case 0:
+		// Use time.Time.UnmarshalText to avoid possible string allocation.
+		if err := a.tt.UnmarshalText(b); err != nil {
+			return err
+		}
+		// TODO(https://go.dev/issue/57912):
+		// RFC 3339 specifies the grammar for a valid timestamp.
+		// However, the parsing functionality in "time" is too loose and
+		// incorrectly accepts invalid timestamps as valid.
+		// Remove these manual checks when "time" checks it for us.
+		newParseError := func(layout, value, layoutElem, valueElem, message string) error {
+			return &time.ParseError{Layout: layout, Value: value, LayoutElem: layoutElem, ValueElem: valueElem, Message: message}
+		}
+		switch {
+		case a.looseRFC3339:
+			return nil
+		case b[len("2006-01-02T")+1] == ':': // hour must be two digits
+			return newParseError(time.RFC3339, string(b), "15", string(b[len("2006-01-02T"):][:1]), "")
+		case b[len("2006-01-02T15:04:05")] == ',': // sub-second separator must be a period
+			return newParseError(time.RFC3339, string(b), ".", ",", "")
+		case b[len(b)-1] != 'Z':
+			switch {
+			case parseDec2(b[len(b)-len("07:00"):]) >= 24: // timezone hour must be in range
+				return newParseError(time.RFC3339, string(b), "Z07:00", string(b[len(b)-len("Z07:00"):]), ": timezone hour out of range")
+			case parseDec2(b[len(b)-len("00"):]) >= 60: // timezone minute must be in range
+				return newParseError(time.RFC3339, string(b), "Z07:00", string(b[len(b)-len("Z07:00"):]), ": timezone minute out of range")
+			}
+		}
+		return nil
+	case math.MaxUint:
+		a.tt, err = time.Parse(a.format, string(b))
+		return err
+	default:
+		a.tt, err = parseTimeUnix(b, a.base)
+		return err
+	}
+}
+
+// appendDurationBase10 appends d formatted as a decimal fractional number,
+// where pow10 is a power-of-10 used to scale down the number.
+func appendDurationBase10(b []byte, d time.Duration, pow10 uint64) []byte {
+	b, n := mayAppendDurationSign(b, d)            // append sign
+	whole, frac := bits.Div64(0, n, uint64(pow10)) // compute whole and frac fields
+	b = strconv.AppendUint(b, whole, 10)           // append whole field
+	return appendFracBase10(b, frac, pow10)        // append frac field
+}
+
+// parseDurationBase10 parses d from a decimal fractional number,
+// where pow10 is a power-of-10 used to scale up the number.
+func parseDurationBase10(b []byte, pow10 uint64) (time.Duration, error) {
+	suffix, neg := consumeSign(b, false)                     // consume sign
+	wholeBytes, fracBytes := bytesCutByte(suffix, '.', true) // consume whole and frac fields
+	whole, okWhole := jsonwire.ParseUint(wholeBytes)         // parse whole field; may overflow
+	frac, okFrac := parseFracBase10(fracBytes, pow10)        // parse frac field
+	hi, lo := bits.Mul64(whole, uint64(pow10))               // overflow if hi > 0
+	sum, co := bits.Add64(lo, uint64(frac), 0)               // overflow if co > 0
+	switch d := mayApplyDurationSign(sum, neg); {            // overflow if neg != (d < 0)
+	case (!okWhole && whole != math.MaxUint64) || !okFrac:
+		return 0, fmt.Errorf("invalid duration %q: %w", b, strconv.ErrSyntax)
+	case !okWhole || hi > 0 || co > 0 || neg != (d < 0):
+		return 0, fmt.Errorf("invalid duration %q: %w", b, strconv.ErrRange)
+	default:
+		return d, nil
+	}
+}
+
+// appendDurationISO8601 appends an ISO 8601 duration with a restricted grammar,
+// where leading and trailing zeroes and zero-value designators are omitted.
+// It only uses hour, minute, and second designators since ISO 8601 defines
+// those as being "accurate", while year, month, week, and day are "nominal".
+func appendDurationISO8601(b []byte, d time.Duration) []byte {
+	if d == 0 {
+		return append(b, "PT0S"...)
+	}
+	b, n := mayAppendDurationSign(b, d)
+	b = append(b, "PT"...)
+	n, nsec := bits.Div64(0, n, 1e9)  // compute nsec field
+	n, sec := bits.Div64(0, n, 60)    // compute sec field
+	hour, min := bits.Div64(0, n, 60) // compute hour and min fields
+	if hour > 0 {
+		b = append(strconv.AppendUint(b, hour, 10), 'H')
+	}
+	if min > 0 {
+		b = append(strconv.AppendUint(b, min, 10), 'M')
+	}
+	if sec > 0 || nsec > 0 {
+		b = append(appendFracBase10(strconv.AppendUint(b, sec, 10), nsec, 1e9), 'S')
+	}
+	return b
+}
+
+// daysPerYear is the exact average number of days in a year according to
+// the Gregorian calendar, which has an extra day each year that is
+// a multiple of 4, unless it is evenly divisible by 100 but not by 400.
+// This does not take into account leap seconds, which are not deterministic.
+const daysPerYear = 365.2425
+
+var errInaccurateDateUnits = errors.New("inaccurate year, month, week, or day units")
+
+// parseDurationISO8601 parses a duration according to ISO 8601-1:2019,
+// section 5.5.2.2 and 5.5.2.3 with the following restrictions or extensions:
+//
+//   - A leading minus sign is permitted for negative duration according
+//     to ISO 8601-2:2019, section 4.4.1.9. We do not permit negative values
+//     for each "time scale component", which is permitted by section 4.4.1.1,
+//     but rarely supported by parsers.
+//
+//   - A leading plus sign is permitted (and ignored).
+//     This is not required by ISO 8601, but not forbidden either.
+//     There is some precedent for this as it is supported by the principle of
+//     duration arithmetic as specified in ISO 8601-2-2019, section 14.1.
+//     Of note, the JavaScript grammar for ISO 8601 permits a leading plus sign.
+//
+//   - A fractional value is only permitted for accurate units
+//     (i.e., hour, minute, and seconds) in the last time component,
+//     which is permissible by ISO 8601-1:2019, section 5.5.2.3.
+//
+//   - Both periods ('.') and commas (',') are supported as the separator
+//     between the integer part and fraction part of a number,
+//     as specified in ISO 8601-1:2019, section 3.2.6.
+//     While ISO 8601 recommends comma as the default separator,
+//     most formatters use a period.
+//
+//   - Leading zeros are ignored. This is not required by ISO 8601,
+//     but also not forbidden by the standard. Many parsers support this.
+//
+//   - Lowercase designators are supported. This is not required by ISO 8601,
+//     but also not forbidden by the standard. Many parsers support this.
+//
+// If the nominal units of year, month, week, or day are present,
+// this produces a best-effort value and also reports [errInaccurateDateUnits].
+//
+// The accepted grammar is identical to JavaScript's Duration:
+//
+//	https://tc39.es/proposal-temporal/#prod-Duration
+//
+// We follow JavaScript's grammar as JSON itself is derived from JavaScript.
+// The Temporal.Duration.toJSON method is guaranteed to produce an output
+// that can be parsed by this function so long as arithmetic in JavaScript
+// does not use a largestUnit value higher than "hours" (which is the default).
+// Even if it does, this will do a best-effort parsing with inaccurate units,
+// but report [errInaccurateDateUnits].
+func parseDurationISO8601(b []byte) (time.Duration, error) {
+	var invalid, overflow, inaccurate, sawFrac bool
+	var sumNanos, n, co uint64
+
+	// cutBytes is like [bytes.Cut], but uses either c0 or c1 as the separator.
+	cutBytes := func(b []byte, c0, c1 byte) (prefix, suffix []byte, ok bool) {
+		for i, c := range b {
+			if c == c0 || c == c1 {
+				return b[:i], b[i+1:], true
+			}
+		}
+		return b, nil, false
+	}
+
+	// mayParseUnit attempts to parse another date or time number
+	// identified by the desHi and desLo unit characters.
+	// If the part is absent for current unit, it returns b as is.
+	mayParseUnit := func(b []byte, desHi, desLo byte, unit time.Duration) []byte {
+		number, suffix, ok := cutBytes(b, desHi, desLo)
+		if !ok || sawFrac {
+			return b // designator is not present or already saw fraction, which can only be in the last component
+		}
+
+		// Parse the number.
+		// A fraction is allowed for the accurate units in the last part.
+		whole, frac, ok := cutBytes(number, '.', ',')
+		if ok {
+			sawFrac = true
+			invalid = invalid || len(frac) == len("") || unit > time.Hour
+			if unit == time.Second {
+				n, ok = parsePaddedBase10(frac, uint64(time.Second))
+				invalid = invalid || !ok
+			} else {
+				f, err := strconv.ParseFloat("0."+string(frac), 64)
+				invalid = invalid || err != nil || len(bytes.Trim(frac[len("."):], "0123456789")) > 0
+				n = uint64(math.Round(f * float64(unit))) // never overflows since f is within [0..1]
+			}
+			sumNanos, co = bits.Add64(sumNanos, n, 0) // overflow if co > 0
+			overflow = overflow || co > 0
+		}
+		for len(whole) > 1 && whole[0] == '0' {
+			whole = whole[len("0"):] // trim leading zeros
+		}
+		n, ok := jsonwire.ParseUint(whole)         // overflow if !ok && MaxUint64
+		hi, lo := bits.Mul64(n, uint64(unit))      // overflow if hi > 0
+		sumNanos, co = bits.Add64(sumNanos, lo, 0) // overflow if co > 0
+		invalid = invalid || (!ok && n != math.MaxUint64)
+		overflow = overflow || (!ok && n == math.MaxUint64) || hi > 0 || co > 0
+		inaccurate = inaccurate || unit > time.Hour
+		return suffix
+	}
+
+	suffix, neg := consumeSign(b, true)
+	prefix, suffix, okP := cutBytes(suffix, 'P', 'p')
+	durDate, durTime, okT := cutBytes(suffix, 'T', 't')
+	invalid = invalid || len(prefix) > 0 || !okP || (okT && len(durTime) == 0) || len(durDate)+len(durTime) == 0
+	if len(durDate) > 0 { // nominal portion of the duration
+		durDate = mayParseUnit(durDate, 'Y', 'y', time.Duration(daysPerYear*24*60*60*1e9))
+		durDate = mayParseUnit(durDate, 'M', 'm', time.Duration(daysPerYear/12*24*60*60*1e9))
+		durDate = mayParseUnit(durDate, 'W', 'w', time.Duration(7*24*60*60*1e9))
+		durDate = mayParseUnit(durDate, 'D', 'd', time.Duration(24*60*60*1e9))
+		invalid = invalid || len(durDate) > 0 // unknown elements
+	}
+	if len(durTime) > 0 { // accurate portion of the duration
+		durTime = mayParseUnit(durTime, 'H', 'h', time.Duration(60*60*1e9))
+		durTime = mayParseUnit(durTime, 'M', 'm', time.Duration(60*1e9))
+		durTime = mayParseUnit(durTime, 'S', 's', time.Duration(1e9))
+		invalid = invalid || len(durTime) > 0 // unknown elements
+	}
+	d := mayApplyDurationSign(sumNanos, neg)
+	overflow = overflow || (neg != (d < 0) && d != 0) // overflows signed duration
+
+	switch {
+	case invalid:
+		return 0, fmt.Errorf("invalid ISO 8601 duration %q: %w", b, strconv.ErrSyntax)
+	case overflow:
+		return 0, fmt.Errorf("invalid ISO 8601 duration %q: %w", b, strconv.ErrRange)
+	case inaccurate:
+		return d, fmt.Errorf("invalid ISO 8601 duration %q: %w", b, errInaccurateDateUnits)
+	default:
+		return d, nil
+	}
+}
+
+// mayAppendDurationSign appends a negative sign if n is negative.
+func mayAppendDurationSign(b []byte, d time.Duration) ([]byte, uint64) {
+	if d < 0 {
+		b = append(b, '-')
+		d *= -1
+	}
+	return b, uint64(d)
+}
+
+// mayApplyDurationSign inverts n if neg is specified.
+func mayApplyDurationSign(n uint64, neg bool) time.Duration {
+	if neg {
+		return -1 * time.Duration(n)
+	} else {
+		return +1 * time.Duration(n)
+	}
+}
+
+// appendTimeUnix appends t formatted as a decimal fractional number,
+// where pow10 is a power-of-10 used to scale up the number.
+func appendTimeUnix(b []byte, t time.Time, pow10 uint64) []byte {
+	sec, nsec := t.Unix(), int64(t.Nanosecond())
+	if sec < 0 {
+		b = append(b, '-')
+		sec, nsec = negateSecNano(sec, nsec)
+	}
+	switch {
+	case pow10 == 1e0: // fast case where units is in seconds
+		b = strconv.AppendUint(b, uint64(sec), 10)
+		return appendFracBase10(b, uint64(nsec), 1e9)
+	case uint64(sec) < 1e9: // intermediate case where units is not seconds, but no overflow
+		b = strconv.AppendUint(b, uint64(sec)*uint64(pow10)+uint64(uint64(nsec)/(1e9/pow10)), 10)
+		return appendFracBase10(b, (uint64(nsec)*pow10)%1e9, 1e9)
+	default: // slow case where units is not seconds and overflow would occur
+		b = strconv.AppendUint(b, uint64(sec), 10)
+		b = appendPaddedBase10(b, uint64(nsec)/(1e9/pow10), pow10)
+		return appendFracBase10(b, (uint64(nsec)*pow10)%1e9, 1e9)
+	}
+}
+
+// parseTimeUnix parses t formatted as a decimal fractional number,
+// where pow10 is a power-of-10 used to scale down the number.
+func parseTimeUnix(b []byte, pow10 uint64) (time.Time, error) {
+	suffix, neg := consumeSign(b, false)                     // consume sign
+	wholeBytes, fracBytes := bytesCutByte(suffix, '.', true) // consume whole and frac fields
+	whole, okWhole := jsonwire.ParseUint(wholeBytes)         // parse whole field; may overflow
+	frac, okFrac := parseFracBase10(fracBytes, 1e9/pow10)    // parse frac field
+	var sec, nsec int64
+	switch {
+	case pow10 == 1e0: // fast case where units is in seconds
+		sec = int64(whole) // check overflow later after negation
+		nsec = int64(frac) // cannot overflow
+	case okWhole: // intermediate case where units is not seconds, but no overflow
+		sec = int64(whole / pow10)                     // check overflow later after negation
+		nsec = int64((whole%pow10)*(1e9/pow10) + frac) // cannot overflow
+	case !okWhole && whole == math.MaxUint64: // slow case where units is not seconds and overflow occurred
+		width := int(math.Log10(float64(pow10)))                                // compute len(strconv.Itoa(pow10-1))
+		whole, okWhole = jsonwire.ParseUint(wholeBytes[:len(wholeBytes)-width]) // parse the upper whole field
+		mid, _ := parsePaddedBase10(wholeBytes[len(wholeBytes)-width:], pow10)  // parse the lower whole field
+		sec = int64(whole)                                                      // check overflow later after negation
+		nsec = int64(mid*(1e9/pow10) + frac)                                    // cannot overflow
+	}
+	if neg {
+		sec, nsec = negateSecNano(sec, nsec)
+	}
+	switch t := time.Unix(sec, nsec).UTC(); {
+	case (!okWhole && whole != math.MaxUint64) || !okFrac:
+		return time.Time{}, fmt.Errorf("invalid time %q: %w", b, strconv.ErrSyntax)
+	case !okWhole || neg != (t.Unix() < 0):
+		return time.Time{}, fmt.Errorf("invalid time %q: %w", b, strconv.ErrRange)
+	default:
+		return t, nil
+	}
+}
+
+// negateSecNano negates a Unix timestamp, where nsec must be within [0, 1e9).
+func negateSecNano(sec, nsec int64) (int64, int64) {
+	sec = ^sec               // twos-complement negation (i.e., -1*sec + 1)
+	nsec = -nsec + 1e9       // negate nsec and add 1e9 (which is the extra +1 from sec negation)
+	sec += int64(nsec / 1e9) // handle possible overflow of nsec if it started as zero
+	nsec %= 1e9              // ensure nsec stays within [0, 1e9)
+	return sec, nsec
+}
+
+// appendFracBase10 appends the fraction of n/max10,
+// where max10 is a power-of-10 that is larger than n.
+func appendFracBase10(b []byte, n, max10 uint64) []byte {
+	if n == 0 {
+		return b
+	}
+	return bytes.TrimRight(appendPaddedBase10(append(b, '.'), n, max10), "0")
+}
+
+// parseFracBase10 parses the fraction of n/max10,
+// where max10 is a power-of-10 that is larger than n.
+func parseFracBase10(b []byte, max10 uint64) (n uint64, ok bool) {
+	switch {
+	case len(b) == 0:
+		return 0, true
+	case len(b) < len(".0") || b[0] != '.':
+		return 0, false
+	}
+	return parsePaddedBase10(b[len("."):], max10)
+}
+
+// appendPaddedBase10 appends a zero-padded encoding of n,
+// where max10 is a power-of-10 that is larger than n.
+func appendPaddedBase10(b []byte, n, max10 uint64) []byte {
+	if n < max10/10 {
+		// Formatting of n is shorter than log10(max10),
+		// so add max10/10 to ensure the length is equal to log10(max10).
+		i := len(b)
+		b = strconv.AppendUint(b, n+max10/10, 10)
+		b[i]-- // subtract the addition of max10/10
+		return b
+	}
+	return strconv.AppendUint(b, n, 10)
+}
+
+// parsePaddedBase10 parses b as the zero-padded encoding of n,
+// where max10 is a power-of-10 that is larger than n.
+// Truncated suffix is treated as implicit zeros.
+// Extended suffix is ignored, but verified to contain only digits.
+func parsePaddedBase10(b []byte, max10 uint64) (n uint64, ok bool) {
+	pow10 := uint64(1)
+	for pow10 < max10 {
+		n *= 10
+		if len(b) > 0 {
+			if b[0] < '0' || '9' < b[0] {
+				return n, false
+			}
+			n += uint64(b[0] - '0')
+			b = b[1:]
+		}
+		pow10 *= 10
+	}
+	if len(b) > 0 && len(bytes.TrimRight(b, "0123456789")) > 0 {
+		return n, false // trailing characters are not digits
+	}
+	return n, true
+}
+
+// consumeSign consumes an optional leading negative or positive sign.
+func consumeSign(b []byte, allowPlus bool) ([]byte, bool) {
+	if len(b) > 0 {
+		if b[0] == '-' {
+			return b[len("-"):], true
+		} else if b[0] == '+' && allowPlus {
+			return b[len("+"):], false
+		}
+	}
+	return b, false
+}
+
+// bytesCutByte is similar to bytes.Cut(b, []byte{c}),
+// except c may optionally be included as part of the suffix.
+func bytesCutByte(b []byte, c byte, include bool) ([]byte, []byte) {
+	if i := bytes.IndexByte(b, c); i >= 0 {
+		if include {
+			return b[:i], b[i:]
+		}
+		return b[:i], b[i+1:]
+	}
+	return b, nil
+}
+
+// parseDec2 parses b as an unsigned, base-10, 2-digit number.
+// The result is undefined if digits are not base-10.
+func parseDec2(b []byte) byte {
+	if len(b) < 2 {
+		return 0
+	}
+	return 10*(b[0]-'0') + (b[1] - '0')
+}