peco.peco/matchers.go

package peco

import (
	"fmt"
	"os/exec"
	"regexp"
	"sort"
	"strings"
)

// Global var used to strips ansi sequences
var reANSIEscapeChars = regexp.MustCompile("\x1B\\[(?:[0-9]{1,2}(?:;[0-9]{1,2})?)*[a-zA-Z]")

// Function who strips ansi sequences
func stripANSISequence(s string) string {
	return reANSIEscapeChars.ReplaceAllString(s, "")
}

// Match defines the interface for matches. Note that to make drawing easier,
// we have a DidMatch and NoMatch types instead of using []Match and []string.
type Match interface {
	Buffer() string // Raw buffer, may contain null
	Line() string   // Line to be displayed
	Output() string // Output string to be displayed after peco is done
	Indices() [][]int
}

type matchString struct {
	buf         string
	sepLoc      int
	displayLine string
}

func newMatchString(v string, enableSep bool) *matchString {
	m := &matchString{
		v,
		-1,
		"",
	}
	if !enableSep {
		return m
	}

	// XXX This may be silly, but we're avoiding using strings.IndexByte()
	// here because it doesn't exist on go1.1. Let's remove support for
	// 1.1 when 1.4 comes out (or something)
	for i := 0; i < len(m.buf); i++ {
		if m.buf[i] == '\000' {
			m.sepLoc = i
		}
	}
	return m
}

func (m matchString) Buffer() string {
	return m.buf
}

func (m matchString) Line() string {
	if m.displayLine != "" {
		return m.displayLine
	}

	if i := m.sepLoc; i > -1 {
		m.displayLine = stripANSISequence(m.buf[:i])
	} else {
		m.displayLine = stripANSISequence(m.buf)
	}
	return m.displayLine
}

func (m matchString) Output() string {
	if i := m.sepLoc; i > -1 {
		return m.buf[i+1:]
	}
	return m.buf
}

// NoMatch is actually an alias to a regular string. It implements the
// Match interface, but just returns the underlying string with no matches
type NoMatch struct {
	*matchString
}

// NewNoMatch creates a NoMatch struct
func NewNoMatch(v string, enableSep bool) *NoMatch {
	return &NoMatch{newMatchString(v, enableSep)}
}

// Indices always returns nil
func (m NoMatch) Indices() [][]int {
	return nil
}

// DidMatch contains the actual match, and the indices to the matches
// in the line
type DidMatch struct {
	*matchString
	matches [][]int
}

// NewDidMatch creates a new DidMatch struct
func NewDidMatch(v string, enableSep bool, m [][]int) *DidMatch {
	return &DidMatch{newMatchString(v, enableSep), m}
}

// Indices returns the indices in the buffer that matched
func (d DidMatch) Indices() [][]int {
	return d.matches
}

// Matcher interface defines the API for things that want to
// match against the buffer
type Matcher interface {
	// Match takes in three parameters.
	//
	// The first chan is the channel where cancel requests are sent.
	// If you receive a request here, you should stop running your query.
	//
	// The second is the query. Do what you want with it
	//
	// The third is the buffer in which to match the query against.
	Match(chan struct{}, string, []Match) []Match
	String() string

	// This is fugly. We just added a method only for CustomMatcner.
	// Must think about this again
	Verify() error
}

// These are used as keys in the config file
const (
	IgnoreCaseMatch    = "IgnoreCase"
	CaseSensitiveMatch = "CaseSensitive"
	RegexpMatch        = "Regexp"
)

// RegexpMatcher is the most basic matcher
type RegexpMatcher struct {
	enableSep bool
	flags     []string
	quotemeta bool
}

// CaseSensitiveMatcher extends the RegxpMatcher, but always
// turns off the ignore-case flag in the regexp
type CaseSensitiveMatcher struct {
	*RegexpMatcher
}

// IgnoreCaseMatcher extends the RegexpMatcher, and always
// turns ON the ignore-case flag in the regexp
type IgnoreCaseMatcher struct {
	*RegexpMatcher
}

// CustomMatcher spawns a new process to filter the buffer
// in peco, and uses the output in its Stdout to figure
// out what to display
type CustomMatcher struct {
	enableSep bool
	name      string
	args      []string
}

// NewCaseSensitiveMatcher creates a new CaseSensitiveMatcher
func NewCaseSensitiveMatcher(enableSep bool) *CaseSensitiveMatcher {
	m := &CaseSensitiveMatcher{NewRegexpMatcher(enableSep)}
	m.quotemeta = true
	return m
}

// NewIgnoreCaseMatcher creates a new IgnoreCaseMatcher
func NewIgnoreCaseMatcher(enableSep bool) *IgnoreCaseMatcher {
	m := &IgnoreCaseMatcher{NewRegexpMatcher(enableSep)}
	m.flags = []string{"i"}
	m.quotemeta = true
	return m
}

// NewRegexpMatcher creates a new RegexpMatcher
func NewRegexpMatcher(enableSep bool) *RegexpMatcher {
	return &RegexpMatcher{
		enableSep,
		[]string{},
		false,
	}
}

// Verify always returns nil
func (m *RegexpMatcher) Verify() error {
	return nil
}

// NewCustomMatcher creates a new CustomMatcher
func NewCustomMatcher(enableSep bool, name string, args []string) *CustomMatcher {
	return &CustomMatcher{enableSep, name, args}
}

// Verify checks to see that the executable given to CustomMatcher
// is actual found and is executable via exec.LookPath
func (m *CustomMatcher) Verify() error {
	if _, err := exec.LookPath(m.args[0]); err != nil {
		return err
	}
	return nil
}

func regexpFor(q string, flags []string, quotemeta bool) (*regexp.Regexp, error) {
	reTxt := q
	if quotemeta {
		reTxt = regexp.QuoteMeta(q)
	}

	if flags != nil && len(flags) > 0 {
		reTxt = fmt.Sprintf("(?%s)%s", strings.Join(flags, ""), reTxt)
	}

	re, err := regexp.Compile(reTxt)
	if err != nil {
		return nil, err
	}
	return re, nil
}

func (m *RegexpMatcher) queryToRegexps(query string) ([]*regexp.Regexp, error) {
	queries := strings.Split(strings.TrimSpace(query), " ")
	regexps := make([]*regexp.Regexp, 0)

	for _, q := range queries {
		re, err := regexpFor(q, m.flags, m.quotemeta)
		if err != nil {
			return nil, err
		}
		regexps = append(regexps, re)
	}

	return regexps, nil
}

func (m *RegexpMatcher) String() string {
	return "Regexp"
}

func (m *CaseSensitiveMatcher) String() string {
	return "CaseSensitive"
}

func (m *IgnoreCaseMatcher) String() string {
	return "IgnoreCase"
}

func (m *CustomMatcher) String() string {
	return m.name
}

// sort related stuff
type byStart [][]int

func (m byStart) Len() int {
	return len(m)
}

func (m byStart) Swap(i, j int) {
	m[i], m[j] = m[j], m[i]
}

func (m byStart) Less(i, j int) bool {
	return m[i][0] < m[j][0]
}

// Match does the heavy lifting, and matches `q` against `buffer`.
// While it is doing the match, it also listens for messages
// via `quit`. If anything is received via `quit`, the match
// is halted.
func (m *RegexpMatcher) Match(quit chan struct{}, q string, buffer []Match) []Match {
	results := []Match{}
	regexps, err := m.queryToRegexps(q)
	if err != nil {
		return results
	}

	// The actual matching is done in a separate goroutine
	iter := make(chan Match, len(buffer))
	go func() {
		// This protects us from panics, caused when we cancel the
		// query and forcefully close the channel (and thereby
		// causing a "close of a closed channel"
		defer func() { recover() }()

		// This must be here to make sure the channel is properly
		// closed in normal cases
		defer close(iter)

		// Iterate through the lines, and do the match.
		// Upon success, send it through the channel
		for _, match := range buffer {
			ms := m.MatchAllRegexps(regexps, match.Line())
			if ms == nil {
				continue
			}
			iter <- NewDidMatch(match.Buffer(), m.enableSep, ms)
		}
		iter <- nil
	}()

MATCH:
	for {
		select {
		case <-quit:
			// If we recieved a cancel request, we immediately bail out.
			// It's a little dirty, but we focefully terminate the other
			// goroutine by closing the channel, and invoking a panic in the
			// goroutine above

			// There's a possibility that the match fails early and the
			// cancel happens after iter has been closed. It's totally okay
			// for us to try to close iter, but trying to detect if the
			// channel can be closed safely synchronously is really hard
			// so we punt it by letting the close() happen at a separate
			// goroutine, protected by a defer recover()
			go func() {
				defer func() { recover() }()
				close(iter)
			}()
			break MATCH
		case match := <-iter:
			// Receive elements from the goroutine performing the match
			if match == nil {
				break MATCH
			}

			results = append(results, match)
		}
	}
	return results
}

// MatchAllRegexps matches all the regexps in `regexps` against line
func (m *RegexpMatcher) MatchAllRegexps(regexps []*regexp.Regexp, line string) [][]int {
	matches := make([][]int, 0)

	allMatched := true
Match:
	for _, re := range regexps {
		match := re.FindAllStringSubmatchIndex(line, -1)
		if match == nil {
			allMatched = false
			break Match
		}

		for _, ma := range match {
			start, end := ma[0], ma[1]
			for _, m := range matches {
				if start >= m[0] && start < m[1] {
					continue Match
				}

				if start < m[0] && end >= m[0] {
					continue Match
				}
			}
			matches = append(matches, ma)
		}
	}

	if !allMatched {
		return nil
	}

	sort.Sort(byStart(matches))

	return matches
}

// Match matches `q` aginst `buffer`
func (m *CustomMatcher) Match(quit chan struct{}, q string, buffer []Match) []Match {
	if len(m.args) < 1 {
		return []Match{}
	}

	results := []Match{}
	if q == "" {
		for _, match := range buffer {
			results = append(results, NewDidMatch(match.Buffer(), m.enableSep, nil))
		}
		return results
	}

	// Receive elements from the goroutine performing the match
	lines := []Match{}
	matcherInput := ""
	for _, match := range buffer {
		matcherInput += match.Line() + "\n"
		lines = append(lines, match)
	}
	args := []string{}
	for _, arg := range m.args {
		if arg == "$QUERY" {
			arg = q
		}
		args = append(args, arg)
	}
	cmd := exec.Command(args[0], args[1:]...)
	cmd.Stdin = strings.NewReader(matcherInput)

	// See RegexpMatcher.Match() for explanation of constructs
	iter := make(chan Match, len(buffer))
	go func() {
		defer func() { recover() }()
		defer func() {
			if p := cmd.Process; p != nil {
				p.Kill()
			}
			close(iter)
		}()
		b, err := cmd.Output()
		if err != nil {
			iter <- nil
		}
		for _, line := range strings.Split(string(b), "\n") {
			if len(line) > 0 {
				iter <- NewDidMatch(line, m.enableSep, nil)
			}
		}
		iter <- nil
	}()
MATCH:
	for {
		select {
		case <-quit:
			go func() {
				defer func() { recover() }()
				close(iter)
			}()
			break MATCH
		case match := <-iter:
			if match == nil {
				break MATCH
			}
			results = append(results, match)
		}
	}

	return results
}