filters: supporting alternative characters for quote

The regexp syntax can have some nasty characters to handle quoting
correctly, in practice. In certain scenarios a double quote works well,
but it can affect readability for certain regexps. This introduces a
quoting mode that treats `/` and `|` as a double quote to make the
quoting in regular expressions more familiar and readable.

This change is introduced in a backwards compatible manner, so existing
regexp quoting is not affected.

Signed-off-by: Stephen J Day <stephen.day@docker.com>

filters/filter_test.go

@@ -135,6 +135,11 @@ func TestFilters(t *testing.T) {
 				corpus[8],
 			},
 		},
+		{
+			name:      "LabelValueNoAltQuoting",
+			input:     "labels.|foo|==omg_asdf.asdf-qwer",
+			errString: "filters: parse error: [labels. >|||< foo|==omg_asdf.asdf-qwer]: invalid quote encountered",
+		},
 		{
 			name:  "Name",
 			input: "name==bar",
@@ -178,6 +183,27 @@ func TestFilters(t *testing.T) {
 				corpus[7],
 			},
 		},
+		{
+			name:  "RegexpQuotedValue",
+			input: "other~=/[abc]+/,name!=foo",
+			expected: []interface{}{
+				corpus[6],
+				corpus[7],
+			},
+		},
+		{
+			name:  "RegexpQuotedValue",
+			input: "other~=/[abc]{1,2}/,name!=foo",
+			expected: []interface{}{
+				corpus[6],
+				corpus[7],
+			},
+		},
+		{
+			name:  "RegexpQuotedValueGarbage",
+			input: "other~=/[abc]{0,1}\"\\//,name!=foo",
+			// valid syntax, but doesn't match anything
+		},
 		{
 			name:  "NameAndLabelValue",
 			input: "name==bar,labels.bar==true",

filters/parser.go

@@ -3,7 +3,6 @@ package filters
 import (
 	"fmt"
 	"io"
-	"strconv"
 
 	"github.com/containerd/containerd/errdefs"
 	"github.com/pkg/errors"
@@ -134,7 +133,12 @@ func (p *parser) selector() (selector, error) {
 		return selector{}, err
 	}
 
-	value, err := p.value()
+	var allowAltQuotes bool
+	if op == operatorMatches {
+		allowAltQuotes = true
+	}
+
+	value, err := p.value(allowAltQuotes)
 	if err != nil {
 		if err == io.EOF {
 			return selector{}, io.ErrUnexpectedEOF
@@ -188,7 +192,7 @@ func (p *parser) field() (string, error) {
 	case tokenField:
 		return s, nil
 	case tokenQuoted:
-		return p.unquote(pos, s)
+		return p.unquote(pos, s, false)
 	}
 
 	return "", p.mkerr(pos, "expected field or quoted")
@@ -213,21 +217,25 @@ func (p *parser) operator() (operator, error) {
 	return 0, p.mkerr(pos, `expected an operator ("=="|"!="|"~=")`)
 }
 
-func (p *parser) value() (string, error) {
+func (p *parser) value(allowAltQuotes bool) (string, error) {
 	pos, tok, s := p.scanner.scan()
 
 	switch tok {
 	case tokenValue, tokenField:
 		return s, nil
 	case tokenQuoted:
-		return p.unquote(pos, s)
+		return p.unquote(pos, s, allowAltQuotes)
 	}
 
 	return "", p.mkerr(pos, "expected value or quoted")
 }
 
-func (p *parser) unquote(pos int, s string) (string, error) {
+func (p *parser) unquote(pos int, s string, allowAlts bool) (string, error) {
-	uq, err := strconv.Unquote(s)
+	if !allowAlts && s[0] != '\'' && s[0] != '"' {
+		return "", p.mkerr(pos, "invalid quote encountered")
+	}
+
+	uq, err := unquote(s)
 	if err != nil {
 		return "", p.mkerr(pos, "unquoting failed: %v", err)
 	}

filters/quote.go

@@ -0,0 +1,237 @@
+package filters
+
+import (
+	"unicode/utf8"
+
+	"github.com/pkg/errors"
+)
+
+// NOTE(stevvooe): Most of this code in this file is copied from the stdlib
+// strconv package and modified to be able to handle quoting with `/` and `|`
+// as delimiters.  The copyright is held by the Go authors.
+
+var errQuoteSyntax = errors.New("quote syntax error")
+
+// UnquoteChar decodes the first character or byte in the escaped string
+// or character literal represented by the string s.
+// It returns four values:
+//
+//	1) value, the decoded Unicode code point or byte value;
+//	2) multibyte, a boolean indicating whether the decoded character requires a multibyte UTF-8 representation;
+//	3) tail, the remainder of the string after the character; and
+//	4) an error that will be nil if the character is syntactically valid.
+//
+// The second argument, quote, specifies the type of literal being parsed
+// and therefore which escaped quote character is permitted.
+// If set to a single quote, it permits the sequence \' and disallows unescaped '.
+// If set to a double quote, it permits \" and disallows unescaped ".
+// If set to zero, it does not permit either escape and allows both quote characters to appear unescaped.
+//
+// This is from Go strconv package, modified to support `|` and `/` as double
+// quotes for use with regular expressions.
+func unquoteChar(s string, quote byte) (value rune, multibyte bool, tail string, err error) {
+	// easy cases
+	switch c := s[0]; {
+	case c == quote && (quote == '\'' || quote == '"' || quote == '/' || quote == '|'):
+		err = errQuoteSyntax
+		return
+	case c >= utf8.RuneSelf:
+		r, size := utf8.DecodeRuneInString(s)
+		return r, true, s[size:], nil
+	case c != '\\':
+		return rune(s[0]), false, s[1:], nil
+	}
+
+	// hard case: c is backslash
+	if len(s) <= 1 {
+		err = errQuoteSyntax
+		return
+	}
+	c := s[1]
+	s = s[2:]
+
+	switch c {
+	case 'a':
+		value = '\a'
+	case 'b':
+		value = '\b'
+	case 'f':
+		value = '\f'
+	case 'n':
+		value = '\n'
+	case 'r':
+		value = '\r'
+	case 't':
+		value = '\t'
+	case 'v':
+		value = '\v'
+	case 'x', 'u', 'U':
+		n := 0
+		switch c {
+		case 'x':
+			n = 2
+		case 'u':
+			n = 4
+		case 'U':
+			n = 8
+		}
+		var v rune
+		if len(s) < n {
+			err = errQuoteSyntax
+			return
+		}
+		for j := 0; j < n; j++ {
+			x, ok := unhex(s[j])
+			if !ok {
+				err = errQuoteSyntax
+				return
+			}
+			v = v<<4 | x
+		}
+		s = s[n:]
+		if c == 'x' {
+			// single-byte string, possibly not UTF-8
+			value = v
+			break
+		}
+		if v > utf8.MaxRune {
+			err = errQuoteSyntax
+			return
+		}
+		value = v
+		multibyte = true
+	case '0', '1', '2', '3', '4', '5', '6', '7':
+		v := rune(c) - '0'
+		if len(s) < 2 {
+			err = errQuoteSyntax
+			return
+		}
+		for j := 0; j < 2; j++ { // one digit already; two more
+			x := rune(s[j]) - '0'
+			if x < 0 || x > 7 {
+				err = errQuoteSyntax
+				return
+			}
+			v = (v << 3) | x
+		}
+		s = s[2:]
+		if v > 255 {
+			err = errQuoteSyntax
+			return
+		}
+		value = v
+	case '\\':
+		value = '\\'
+	case '\'', '"', '|', '/':
+		if c != quote {
+			err = errQuoteSyntax
+			return
+		}
+		value = rune(c)
+	default:
+		err = errQuoteSyntax
+		return
+	}
+	tail = s
+	return
+}
+
+// unquote interprets s as a single-quoted, double-quoted,
+// or backquoted Go string literal, returning the string value
+// that s quotes.  (If s is single-quoted, it would be a Go
+// character literal; Unquote returns the corresponding
+// one-character string.)
+//
+// This is modified from the standard library to support `|` and `/` as quote
+// characters for use with regular expressions.
+func unquote(s string) (string, error) {
+	n := len(s)
+	if n < 2 {
+		return "", errQuoteSyntax
+	}
+	quote := s[0]
+	if quote != s[n-1] {
+		return "", errQuoteSyntax
+	}
+	s = s[1 : n-1]
+
+	if quote == '`' {
+		if contains(s, '`') {
+			return "", errQuoteSyntax
+		}
+		if contains(s, '\r') {
+			// -1 because we know there is at least one \r to remove.
+			buf := make([]byte, 0, len(s)-1)
+			for i := 0; i < len(s); i++ {
+				if s[i] != '\r' {
+					buf = append(buf, s[i])
+				}
+			}
+			return string(buf), nil
+		}
+		return s, nil
+	}
+	if quote != '"' && quote != '\'' && quote != '|' && quote != '/' {
+		return "", errQuoteSyntax
+	}
+	if contains(s, '\n') {
+		return "", errQuoteSyntax
+	}
+
+	// Is it trivial?  Avoid allocation.
+	if !contains(s, '\\') && !contains(s, quote) {
+		switch quote {
+		case '"', '/', '|': // pipe and slash are treated like double quote
+			return s, nil
+		case '\'':
+			r, size := utf8.DecodeRuneInString(s)
+			if size == len(s) && (r != utf8.RuneError || size != 1) {
+				return s, nil
+			}
+		}
+	}
+
+	var runeTmp [utf8.UTFMax]byte
+	buf := make([]byte, 0, 3*len(s)/2) // Try to avoid more allocations.
+	for len(s) > 0 {
+		c, multibyte, ss, err := unquoteChar(s, quote)
+		if err != nil {
+			return "", err
+		}
+		s = ss
+		if c < utf8.RuneSelf || !multibyte {
+			buf = append(buf, byte(c))
+		} else {
+			n := utf8.EncodeRune(runeTmp[:], c)
+			buf = append(buf, runeTmp[:n]...)
+		}
+		if quote == '\'' && len(s) != 0 {
+			// single-quoted must be single character
+			return "", errQuoteSyntax
+		}
+	}
+	return string(buf), nil
+}
+
+// contains reports whether the string contains the byte c.
+func contains(s string, c byte) bool {
+	for i := 0; i < len(s); i++ {
+		if s[i] == c {
+			return true
+		}
+	}
+	return false
+}
+
+func unhex(b byte) (v rune, ok bool) {
+	c := rune(b)
+	switch {
+	case '0' <= c && c <= '9':
+		return c - '0', true
+	case 'a' <= c && c <= 'f':
+		return c - 'a' + 10, true
+	case 'A' <= c && c <= 'F':
+		return c - 'A' + 10, true
+	}
+	return
+}

filters/scanner.go

@@ -87,7 +87,7 @@ func (s *scanner) peek() rune {
 	return ch
 }
 
-func (s *scanner) scan() (int, token, string) {
+func (s *scanner) scan() (nextp int, tk token, text string) {
 	var (
 		ch  = s.next()
 		pos = s.pos
@@ -101,6 +101,7 @@ chomp:
 		s.scanQuoted(ch)
 		return pos, tokenQuoted, s.input[pos:s.ppos]
 	case isSeparatorRune(ch):
+		s.value = false
 		return pos, tokenSeparator, s.input[pos:s.ppos]
 	case isOperatorRune(ch):
 		s.scanOperator()
@@ -241,7 +242,7 @@ func isOperatorRune(r rune) bool {
 
 func isQuoteRune(r rune) bool {
 	switch r {
-	case '"': // maybe add single quoting?
+	case '/', '|', '"': // maybe add single quoting?
 		return true
 	}
 

filters/scanner_test.go

@@ -95,6 +95,20 @@ func TestScanner(t *testing.T) {
 				{pos: 22, token: tokenEOF},
 			},
 		},
+		{
+			name:  "RegexpQuotedValue",
+			input: `name~=/[abc]{0,2}/,foo=test`,
+			expected: []tokenResult{
+				{pos: 0, token: tokenField, text: "name"},
+				{pos: 4, token: tokenOperator, text: "~="},
+				{pos: 6, token: tokenQuoted, text: "/[abc]{0,2}/"},
+				{pos: 18, token: tokenSeparator, text: ","},
+				{pos: 19, token: tokenField, text: "foo"},
+				{pos: 22, token: tokenOperator, text: "="},
+				{pos: 23, token: tokenValue, text: "test"},
+				{pos: 27, token: tokenEOF},
+			},
+		},
 		{
 			name:  "Cowsay",
 			input: "name~=牛,labels.moo=true",
@@ -111,6 +125,22 @@ func TestScanner(t *testing.T) {
 				{pos: 25, token: tokenEOF},
 			},
 		},
+		{
+			name:  "CowsayRegexpQuoted",
+			input: "name~=|牛|,labels.moo=true",
+			expected: []tokenResult{
+				{pos: 0, token: tokenField, text: "name"},
+				{pos: 4, token: tokenOperator, text: "~="},
+				{pos: 6, token: tokenQuoted, text: "|牛|"},
+				{pos: 11, token: tokenSeparator, text: ","},
+				{pos: 12, token: tokenField, text: "labels"},
+				{pos: 18, token: tokenSeparator, text: "."},
+				{pos: 19, token: tokenField, text: "moo"},
+				{pos: 22, token: tokenOperator, text: "="},
+				{pos: 23, token: tokenValue, text: "true"},
+				{pos: 27, token: tokenEOF},
+			},
+		},
 		{
 			name:  "Escapes",
 			input: `name~="asdf\n\tfooo"`,
@@ -187,15 +217,45 @@ func TestScanner(t *testing.T) {
 		},
 		{
 			name:  "MissingValue",
-			input: "input==,id?=ff",
+			input: "input==,id!=ff",
 			expected: []tokenResult{
 				{pos: 0, token: tokenField, text: "input"},
 				{pos: 5, token: tokenOperator, text: "=="},
 				{pos: 7, token: tokenSeparator, text: ","},
-				{pos: 8, token: tokenValue, text: "id?=ff"},
+				{pos: 8, token: tokenField, text: "id"},
+				{pos: 10, token: tokenOperator, text: "!="},
+				{pos: 12, token: tokenValue, text: "ff"},
 				{pos: 14, token: tokenEOF},
 			},
 		},
+		{
+			name:  "QuotedRegexp",
+			input: "input~=/foo\\/bar/,id!=ff",
+			expected: []tokenResult{
+				{pos: 0, token: tokenField, text: "input"},
+				{pos: 5, token: tokenOperator, text: "~="},
+				{pos: 7, token: tokenQuoted, text: "/foo\\/bar/"},
+				{pos: 17, token: tokenSeparator, text: ","},
+				{pos: 18, token: tokenField, text: "id"},
+				{pos: 20, token: tokenOperator, text: "!="},
+				{pos: 22, token: tokenValue, text: "ff"},
+				{pos: 24, token: tokenEOF},
+			},
+		},
+		{
+			name:  "QuotedRegexpAlt",
+			input: "input~=|foo/bar|,id!=ff",
+			expected: []tokenResult{
+				{pos: 0, token: tokenField, text: "input"},
+				{pos: 5, token: tokenOperator, text: "~="},
+				{pos: 7, token: tokenQuoted, text: "|foo/bar|"},
+				{pos: 16, token: tokenSeparator, text: ","},
+				{pos: 17, token: tokenField, text: "id"},
+				{pos: 19, token: tokenOperator, text: "!="},
+				{pos: 21, token: tokenValue, text: "ff"},
+				{pos: 23, token: tokenEOF},
+			},
+		},
 	} {
 		t.Run(testcase.name, func(t *testing.T) {
 			var sc scanner