Ban ] as literal first character of custom character class

PCRE, Oniguruma, and ICU allow `]` to appear as
the first member of a custom character class, and
treat it as literal, due to empty character classes
being invalid.

However this behavior isn't particularly intuitive,
and makes lexing heuristics harder to implement
properly. Instead, reject such character classes
as being empty, and require escaping if `]` is
meant as the first character.

Author: hamishknight

Sources/_RegexParser/Regex/Parse/Parse.swift

@@ -465,16 +465,6 @@ extension Parser {
     var members: Array<Member> = []
     try parseCCCMembers(into: &members)
 
-    // If we didn't parse any semantic members, we can eat a ']' character, as
-    // PCRE, Oniguruma, and ICU forbid empty character classes, and assume an
-    // initial ']' is literal.
-    if members.none(\.isSemantic) {
-      if let loc = source.tryEatWithLoc("]") {
-        members.append(.atom(.init(.char("]"), loc)))
-        try parseCCCMembers(into: &members)
-      }
-    }
-
     // If we have a binary set operator, parse it and the next members. Note
     // that this means we left associate for a chain of operators.
     // TODO: We may want to diagnose and require users to disambiguate, at least

Tests/RegexTests/MatchTests.swift

@@ -1282,7 +1282,6 @@ extension RegexTests {
     firstMatchTest(#"(?xx)[ \t]+"#, input: " \t \t", match: "\t")
 
     firstMatchTest("(?xx)[ a && ab ]+", input: " aaba ", match: "aa")
-    firstMatchTest("(?xx)[ ] a ]+", input: " a]]a ] ", match: "a]]a")
   }
 
   func testASCIIClasses() {

Tests/RegexTests/ParseTests.swift

@@ -528,23 +528,7 @@ extension RegexTests {
     ))
 
     parseTest("[-]", charClass("-"))
-
-    // Empty character classes are forbidden, therefore these are character
-    // classes containing literal ']'.
-    parseTest("[]]", charClass("]"))
-    parseTest("[]a]", charClass("]", "a"))
-    parseTest("(?x)[ ]]", concat(
-      changeMatchingOptions(matchingOptions(adding: .extended)),
-      charClass("]")
-    ))
-    parseTest("(?x)[ ]  ]", concat(
-      changeMatchingOptions(matchingOptions(adding: .extended)),
-      charClass("]")
-    ))
-    parseTest("(?x)[ ] a ]", concat(
-      changeMatchingOptions(matchingOptions(adding: .extended)),
-      charClass("]", "a")
-    ))
+    parseTest(#"[\]]"#, charClass("]"))
 
     // These are metacharacters in certain contexts, but normal characters
     // otherwise.
@@ -2497,10 +2481,15 @@ extension RegexTests {
 
     diagnosticTest("[a", .expected("]"))
 
-    // The first ']' of a custom character class is literal, so these are
-    // missing the closing bracket.
-    diagnosticTest("[]", .expected("]"))
-    diagnosticTest("(?x)[  ]", .expected("]"))
+    // Character classes may not be empty.
+    diagnosticTest("[]", .expectedCustomCharacterClassMembers)
+    diagnosticTest("[]]", .expectedCustomCharacterClassMembers)
+    diagnosticTest("[]a]", .expectedCustomCharacterClassMembers)
+    diagnosticTest("(?x)[  ]", .expectedCustomCharacterClassMembers)
+    diagnosticTest("(?x)[ ]  ]", .expectedCustomCharacterClassMembers)
+    diagnosticTest("(?x)[ ] a ]", .expectedCustomCharacterClassMembers)
+    diagnosticTest("(?xx)[ ] a ]+", .expectedCustomCharacterClassMembers)
+    diagnosticTest("(?x)[ ]]", .expectedCustomCharacterClassMembers)
 
     diagnosticTest("[&&]", .expectedCustomCharacterClassMembers)
     diagnosticTest("[a&&]", .expectedCustomCharacterClassMembers)