Use a lazy-quoting @sh encoder (#1548)

* Use a lazy-quoting @sh encoder

* Add internal quoting style switch to @sh

* Add test for stray empty quotes in @sh

pkg/yqlib/doc/operators/encode-decode.md

@@ -478,7 +478,7 @@ yq '.coolData | @sh' sample.yml
 ```
 will output
 ```yaml
-'strings with spaces and a \'quote\''
+strings' with spaces and a '\'quote\'
 ```
 
 ## Decode a base64 encoded string

pkg/yqlib/encoder_sh.go

@@ -9,13 +9,14 @@ import (
 	yaml "gopkg.in/yaml.v3"
 )
 
-var pattern = regexp.MustCompile(`[^\w@%+=:,./-]`)
+var unsafeChars = regexp.MustCompile(`[^\w@%+=:,./-]`)
 
 type shEncoder struct {
+	quoteAll bool
 }
 
 func NewShEncoder() Encoder {
-	return &shEncoder{}
+	return &shEncoder{false}
 }
 
 func (e *shEncoder) CanHandleAliases() bool {
@@ -36,9 +37,43 @@ func (e *shEncoder) Encode(writer io.Writer, originalNode *yaml.Node) error {
 		return fmt.Errorf("cannot encode %v as URI, can only operate on strings. Please first pipe through another encoding operator to convert the value to a string", node.Tag)
 	}
 
-	value := originalNode.Value
-	if pattern.MatchString(value) {
-		value = "'" + strings.ReplaceAll(value, "'", "\\'") + "'"
-	}
-	return writeString(writer, value)
+	return writeString(writer, e.encode(originalNode.Value))
+}
+
+// put any (shell-unsafe) characters into a single-quoted block, close the block lazily
+func (e *shEncoder) encode(input string) string {
+	const quote = '\''
+	var inQuoteBlock = false
+	var encoded strings.Builder
+	encoded.Grow(len(input))
+
+	for _, ir := range input {
+		// open or close a single-quote block
+		if ir == quote {
+			if inQuoteBlock {
+				// get out of a quote block for an input quote
+				encoded.WriteRune(quote)
+				inQuoteBlock = !inQuoteBlock
+			}
+			// escape the quote with a backslash
+			encoded.WriteRune('\\')
+		} else {
+			if e.shouldQuote(ir) && !inQuoteBlock {
+				// start a quote block for any (unsafe) characters
+				encoded.WriteRune(quote)
+				inQuoteBlock = !inQuoteBlock
+			}
+		}
+		// pass on the input character
+		encoded.WriteRune(ir)
+	}
+	// close any pending quote block
+	if inQuoteBlock {
+		encoded.WriteRune(quote)
+	}
+	return encoded.String()
+}
+
+func (e *shEncoder) shouldQuote(ir rune) bool {
+	return e.quoteAll || unsafeChars.MatchString(string(ir))
 }

pkg/yqlib/operator_encoder_decoder_test.go

@@ -263,9 +263,19 @@ var encoderDecoderOperatorScenarios = []expressionScenario{
 		document:       "coolData: strings with spaces and a 'quote'",
 		expression:     ".coolData | @sh",
 		expected: []string{
-			"D0, P[coolData], (!!str)::'strings with spaces and a \\'quote\\''\n",
+			"D0, P[coolData], (!!str)::strings' with spaces and a '\\'quote\\'\n",
 		},
 	},
+	{
+		description:    "Encode a string to sh",
+		subdescription: "Watch out for stray '' (empty strings)",
+		document:       "coolData: \"'starts, contains more '' and ends with a quote'\"",
+		expression:     ".coolData | @sh",
+		expected: []string{
+			"D0, P[coolData], (!!str)::\\'starts,' contains more '\\'\\'' and ends with a quote'\\'\n",
+		},
+		skipDoc: true,
+	},
 	{
 		description:    "Decode a base64 encoded string",
 		subdescription: "Decoded data is assumed to be a string.",