sq/libsq/core/kind/detect.go

package kind

import (
	stdj "encoding/json"
	"math/big"
	"slices"
	"strconv"
	"strings"
	"time"

	"github.com/neilotoole/sq/libsq/core/timez"

	"github.com/neilotoole/sq/libsq/core/errz"
	"github.com/neilotoole/sq/libsq/core/stringz"
)

// Detector is used to detect the kind of a stream of values.
// The caller adds values via Sample and then invokes Detect.
type Detector struct {
	kinds    map[Kind]struct{}
	mungeFns map[Kind]MungeFunc
	dirty    bool

	// foundString is set to true if any of the values passed
	// to Detector.Sample had type string.
	foundString bool
}

// NewDetector returns a new instance.
func NewDetector() *Detector {
	return &Detector{
		kinds: map[Kind]struct{}{ //nolint:exhaustive
			Int:      {},
			Float:    {},
			Decimal:  {},
			Bool:     {},
			Time:     {},
			Date:     {},
			Datetime: {},
		},
		mungeFns: map[Kind]MungeFunc{},
	}
}

// Sample adds a sample to the detector.
func (d *Detector) Sample(v any) {
	switch v.(type) {
	case nil:
		// Can't glean any info from nil
		return
	default:
		// Don't know what this, so delete all kinds
		d.retain()
		return
	case float32, float64:
		d.retain(Float, Decimal)
		return
	case int, int8, int16, int32, int64:
		d.retain(Int, Float, Decimal)
		return
	case bool:
		d.retain(Bool)
		return
	case time.Time:
		d.retain(Time, Date, Datetime)
		return
	case stdj.Number:
		// JSON number
		d.foundString = true
		d.retain(Decimal)
		return
	case string:
		// We need to do more work to figure out the kind when
		// we're getting string values
		d.foundString = true
	}

	// We're dealing with a string value, which could a variety
	// of things, such as: "1", "1.0", "true", "11:30".
	d.doSampleString(v.(string))
}

//nolint:gocognit,funlen
func (d *Detector) doSampleString(s string) {
	if s == "" {
		// Can't really do anything useful with this
		return
	}

	var err error

	if d.has(Int) || d.has(Decimal) {
		if strings.ContainsRune(s, '.') {
			// Int cannot contain '.', e.g. "1.0".
			d.delete(Int)
		}

		if strings.ContainsAny(s, "eE") {
			// Int and Decimal cannot contain E or e.
			// Most likely a float, e.g. "6.67428e-11"
			d.delete(Int, Decimal)
		}
	}

	if d.has(Decimal) {
		// If Decimal is still a candidate, check that we can parse it
		if _, _, err = big.ParseFloat(s, 10, 64, 0); err != nil {
			// If s cannot be parsed as a decimal, it also can't
			// be int or float
			d.delete(Decimal, Int, Float)
		} else {
			// s can be parsed as decimal, can't be time
			d.delete(Time, Date, Datetime)
		}
	}

	if d.has(Int) {
		if _, err = strconv.ParseInt(s, 10, 64); err != nil {
			d.delete(Int)
		} else {
			// s can be parsed as int, can't be time
			d.delete(Time, Date, Datetime)
		}
	}

	if d.has(Float) {
		if _, err = strconv.ParseFloat(s, 64); err != nil {
			d.delete(Float)
		} else {
			// s can be parsed as float, can't be time
			d.delete(Time, Date, Datetime)
		}
	}

	if d.has(Bool) {
		if _, err = stringz.ParseBool(s); err != nil {
			d.delete(Bool)
		} else {
			// s can be parsed as bool, can't be time,
			// but still could be a number ("1" == true)
			d.delete(Time, Date, Datetime)
		}
	}

	if d.has(Time) {
		ok, format := detectKindTime(s)
		if !ok {
			// It's not a recognized time format
			d.delete(Time)
		} else {
			// If it's kind.Time, it can't be anything else
			d.retain(Time)

			d.mungeFns[Time] = func(val any) (any, error) {
				if val == nil {
					return nil, nil //nolint:nilnil
				}

				s, ok = val.(string)
				if !ok {
					return nil, errz.Errorf("expected %T to be string", val)
				}

				if s == "" {
					return nil, nil //nolint:nilnil
				}

				var t time.Time
				t, err = time.Parse(format, s)
				if err != nil {
					return nil, errz.Err(err)
				}

				// FIXME: Should time always return the canonical format?
				// return t.Format(format), nil
				return t.Format(time.TimeOnly), nil
			}
		}
	}

	if d.has(Date) {
		ok, format := detectKindDate(s)
		if !ok {
			// It's not a recognized date format
			d.delete(Date)
		} else {
			// If it's kind.Date, it can't be anything else
			d.retain(Date)

			d.mungeFns[Date] = func(val any) (any, error) {
				if val == nil {
					return nil, nil //nolint:nilnil
				}

				s, ok = val.(string)
				if !ok {
					return nil, errz.Errorf("expected %T to be string", val)
				}

				if s == "" {
					return nil, nil //nolint:nilnil
				}

				var t time.Time
				t, err = time.Parse(format, s)
				if err != nil {
					return nil, errz.Err(err)
				}

				// Always return the date in the canonical format.
				return t.Format(time.DateOnly), nil
			}
		}
	}

	if d.has(Datetime) {
		ok, format := detectKindDatetime(s)
		if !ok {
			// It's not a recognized datetime format
			d.delete(Datetime)
		} else {
			// If it's kind.Datetime, it can't be anything else
			d.retain(Datetime)

			// This mungeFn differs from kind.Date and kind.Time in that
			// it returns a time.Time instead of a string
			d.mungeFns[Datetime] = func(val any) (any, error) {
				if val == nil {
					return nil, nil //nolint:nilnil
				}

				s, ok := val.(string)
				if !ok {
					return nil, errz.Errorf("expected %T to be string", val)
				}

				if s == "" {
					return nil, nil //nolint:nilnil
				}

				t, err := time.Parse(format, s)
				if err != nil {
					return nil, errz.Err(err)
				}

				return t, nil
			}
		}
	}
}

// Detect returns the detected Kind. If ambiguous, Text is returned,
// unless all sampled values were nil, in which case Null is returned.
// If the returned mungeFn is non-nil, it can be used to convert input
// values to their canonical form. For example, for Datetime the MungeFunc
// would accept string "2020-06-11T02:50:54Z" and return a time.Time,
// while for Date, the MungeFunc would accept "1970-01-01" or "01 Jan 1970"
// and always return a string in the canonicalized form "1970-01-01".
func (d *Detector) Detect() (kind Kind, mungeFn MungeFunc, err error) {
	if !d.dirty {
		if d.foundString {
			return Text, nil, nil
		}

		// If we haven't filtered any kinds, default to Null.
		return Null, nil, nil
	}

	switch len(d.kinds) {
	case 0:
		return Text, nil, nil
	case 1:
		for k := range d.kinds {
			return k, d.mungeFns[k], nil
		}
	}

	// NOTE: this logic below about detecting the remaining type
	//  is a bit sketchy. If you're debugging this code, it's
	//  probably the case that the code below is faulty.
	if d.has(Time, Date, Datetime) {
		// If all three time types are left, use the most
		// general, i.e. Datetime.
		return Datetime, d.mungeFns[Datetime], nil
	}

	if d.has(Time) {
		return Time, d.mungeFns[Time], nil
	}

	if d.has(Date) {
		return Date, d.mungeFns[Date], nil
	}

	if d.has(Datetime) {
		return Datetime, d.mungeFns[Datetime], nil
	}

	if d.foundString {
		if d.has(Decimal, Int) {
			// If we have to choose between Decimal and Int, we
			// pick Int, as it's stricter.
			return Int, nil, nil
		}

		if d.has(Decimal) {
			return Decimal, nil, nil
		}
	}

	if d.has(Int) {
		return Int, nil, nil
	}

	if d.has(Float) {
		return Float, nil, nil
	}

	if d.has(Bool) {
		return Bool, nil, nil
	}

	return Text, nil, nil
}

// delete deletes each of kinds from d.kinds.
func (d *Detector) delete(kinds ...Kind) {
	d.dirty = true
	for _, k := range kinds {
		delete(d.kinds, k)
	}
}

// retain deletes everything from kd.kinds except items
// contains in the kinds arg. If kinds is empty, d.kinds is
// emptied.
func (d *Detector) retain(kinds ...Kind) {
	d.dirty = true
	for k := range d.kinds {
		if !containsKind(k, kinds...) {
			delete(d.kinds, k)
		}
	}
}

// has returns true if kd.kinds contains each of k.
func (d *Detector) has(kinds ...Kind) bool {
	var ok bool
	for _, k := range kinds {
		if _, ok = d.kinds[k]; !ok {
			return false
		}
	}

	return true
}

func detectKindTime(s string) (ok bool, format string) {
	if s == "" {
		return false, ""
	}

	formats := []string{
		time.TimeOnly,
		"15:04:05 PM",
		"15:04:05PM",
		"15:04:05 pm",
		"15:04:05pm",
		"15:04",
		time.Kitchen,
		"3:04 PM",
		"3:04pm",
		"3:04 pm",
	}
	var err error
	for _, f := range formats {
		if _, err = time.Parse(f, s); err == nil {
			return true, f
		}
	}

	return false, ""
}

func detectKindDate(s string) (ok bool, format string) {
	if s == "" {
		return false, ""
	}

	formats := []string{
		time.DateOnly,
		"02 Jan 2006",
		"2006/01/02",
		"01-02-06",
		"01-02-2006",
		"02-Jan-2006",
		"2-Jan-2006",
		"2-Jan-06",
		"Jan _2, 2006",
		"Jan 2, 2006",
		timez.ExcelLongDate,
		"Mon, January 2, 2006",
		"Mon, Jan 2, 2006",
		"January 2, 2006",
		"_2/Jan/06",
		"2/Jan/06",
	}
	var err error

	for _, f := range formats {
		if _, err = time.Parse(f, s); err == nil {
			return true, f
		}
	}

	return false, ""
}

var datetimeFormats = []string{
	timez.RFC3339NanoZ,
	time.RFC3339Nano,
	time.ANSIC,
	time.UnixDate,
	time.RubyDate,
	time.RFC822,
	time.RFC822Z,
	time.RFC850,
	time.RFC1123Z,
	time.RFC1123,
	timez.DateHourMinuteSecond,
	timez.DateHourMinute,
	timez.ExcelLongDate,
	timez.ExcelDatetimeMDYSeconds,
	timez.ExcelDatetimeMDYNoSeconds,
}

func detectKindDatetime(s string) (ok bool, format string) {
	if s == "" {
		return false, ""
	}

	var err error
	formats := slices.Clone(datetimeFormats)

	for _, f := range formats {
		if _, err = time.Parse(f, s); err == nil {
			return true, f
		}
	}

	return false, ""
}

// Hash generates a hash from the kinds returned by
// the detectors. The detectors should already have
// sampled data.
func Hash(detectors []*Detector) (h string, err error) {
	if len(detectors) == 0 {
		return "", errz.New("no kind detectors")
	}

	kinds := make([]Kind, len(detectors))
	for i := range detectors {
		kinds[i], _, err = detectors[i].Detect()
		if err != nil {
			return "", err
		}
	}

	// TODO: use an actual hash function
	hash := strings.Builder{}
	for i := range kinds {
		if i > 0 {
			hash.WriteRune('|')
		}
		hash.WriteString(kinds[i].String())
	}

	h = hash.String()
	return h, nil
}