Forbid empty character classes

As per PCRE, Oniguruma, and ICU, a first character
of `]` is treated as literal.

Author: hamishknight

Sources/_MatchingEngine/Regex/Parse/Parse.swift

@@ -434,6 +434,12 @@ extension Parser {
     try source.expectNonEmpty()
 
     var members: Array<Member> = []
+
+    // We can eat an initial ']', as PCRE, Oniguruma, and ICU forbid empty
+    // character classes, and assume an initial ']' is literal.
+    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

Tests/RegexTests/ParseTests.swift

@@ -428,6 +428,10 @@ extension RegexTests {
 
     parseTest("[-]", charClass("-"))
 
+    // Empty character classes are forbidden, therefore this is a character
+    // class of literal ']'.
+    parseTest("[]]", charClass("]"))
+
     // These are metacharacters in certain contexts, but normal characters
     // otherwise.
     parseTest(
@@ -1901,6 +1905,10 @@ extension RegexTests {
     diagnosticTest("(?<a-b", .expected(">"))
     diagnosticTest("(?<a-b>", .expected(")"))
 
+    // The first ']' of a custom character class is literal, so this is missing
+    // the closing bracket.
+    diagnosticTest("[]", .expected("]"))
+
     // MARK: Bad escapes
 
     diagnosticTest("\\", .expectedEscape)