Allow unbounded quoted sequences \Q...

PCRE and ICU both support quoted sequences that
don't have a terminating `\E`. Update the parsing
to allow this.

Author: hamishknight

Sources/_RegexParser/Regex/Parse/LexicalAnalysis.swift

@@ -565,7 +565,7 @@ extension Source {
 
   /// Try to consume quoted content
   ///
-  ///     Quote -> '\Q' (!'\E' .)* '\E'
+  ///     Quote -> '\Q' (!'\E' .)* '\E'?
   ///
   /// With `SyntaxOptions.experimentalQuotes`, also accepts
   ///
@@ -578,9 +578,21 @@ extension Source {
   mutating func lexQuote(context: ParsingContext) throws -> AST.Quote? {
     let str = try recordLoc { src -> String? in
       if src.tryEat(sequence: #"\Q"#) {
-        return try src.expectQuoted(endingWith: #"\E"#).value
+        let contents = src.lexUntil { src in
+          src.isEmpty || src.starts(with: #"\E"#)
+        }.value
+
+        // If we have an ending, the sequence shouldn't be empty. This isn't a
+        // particulary useful thing to express, and [\Q\E] should definitely be
+        // illegal. The unbounded case may however be empty.
+        let ending = src.tryEat(sequence: #"\E"#)
+        if ending && contents.isEmpty {
+          throw ParseError.expectedNonEmptyContents
+        }
+        return contents
       }
       if context.experimentalQuotes, src.tryEat("\"") {
+        // TODO: Can experimental quotes be empty?
         return try src.expectQuoted(endingWith: "\"", ignoreEscaped: true).value
       }
       return nil

Tests/RegexTests/ParseTests.swift

@@ -742,6 +742,11 @@ extension RegexTests {
     // This follows the PCRE behavior.
     parseTest(#"\Q\\E"#, quote("\\"))
 
+    // Quotes may be unterminated.
+    parseTest(#"\Qab"#, quote("ab"))
+    parseTest(#"\Q"#, quote(""))
+    parseTest("\\Qab\\", quote("ab\\"))
+
     parseTest(#"a" ."b"#, concat("a", quote(" ."), "b"),
               syntax: .experimental)
     parseTest(#"a" .""b""#, concat("a", quote(" ."), quote("b")),
@@ -2486,8 +2491,6 @@ extension RegexTests {
     diagnosticTest(#"(?P"#, .expected(")"))
     diagnosticTest(#"(?R"#, .expected(")"))
 
-    diagnosticTest(#"\Qab"#, .expected("\\E"))
-    diagnosticTest("\\Qab\\", .expected("\\E"))
     diagnosticTest(#""ab"#, .expected("\""), syntax: .experimental)
     diagnosticTest(#""ab\""#, .expected("\""), syntax: .experimental)
     diagnosticTest("\"ab\\", .expectedEscape, syntax: .experimental)
@@ -2560,6 +2563,10 @@ extension RegexTests {
     // TODO: Custom diagnostic for missing '\Q'
     diagnosticTest(#"\E"#, .invalidEscape("E"))
 
+    diagnosticTest(#"\Q\E"#, .expectedNonEmptyContents)
+    diagnosticTest(#"[\Q\E]"#, .expectedNonEmptyContents)
+    diagnosticTest(#"[\Q]"#, .expected("]"))
+
     // PCRE treats these as octal, but we require a `0` prefix.
     diagnosticTest(#"[\1]"#, .invalidEscape("1"))
     diagnosticTest(#"[\123]"#, .invalidEscape("1"))