[NFC] Reduce SwiftParser stack usage in debug mode

I’ve had trouble running SwiftParser’s tests—particularly the torture tests in the swiftlang/swift repo—locally in debug mode because some common parser functions would use many kilobytes of stack for each call, consuming the stack too quickly for the `maximumNestingLevel` to control. Break up a number of problematic functions to help with this, and also insert a missing stack overflow check in `parseClosureExpression()`.

Author: beccadax

CodeGeneration/Sources/generate-swift-syntax/templates/swiftsyntax/KeywordFile.swift

@@ -40,20 +40,27 @@ let keywordFile = SourceFileSyntax(leadingTrivia: copyrightHeader) {
       try! SwitchExprSyntax("switch text.count") {
         for (length, keywords) in keywordsByLength() {
           SwitchCaseSyntax("case \(raw: length):") {
-            try! SwitchExprSyntax("switch text") {
-              for keyword in keywords {
-                SwitchCaseSyntax("case \(literal: keyword.name):") {
-                  ExprSyntax("self = .\(keyword.enumCaseCallName)")
-                }
-              }
-              SwitchCaseSyntax("default: return nil")
-            }
+            ExprSyntax("self.init(_length\(raw: length): text)")
           }
         }
         SwitchCaseSyntax("default: return nil")
       }
     }
 
+    // Split into individual initializers by length to reduce stack use
+    for (length, keywords) in keywordsByLength() {
+      try! InitializerDeclSyntax("private init?(_length\(raw: length) text: SyntaxText)") {
+        try! SwitchExprSyntax("switch text") {
+          for keyword in keywords {
+            SwitchCaseSyntax("case \(literal: keyword.name):") {
+              ExprSyntax("self = .\(keyword.enumCaseCallName)")
+            }
+          }
+          SwitchCaseSyntax("default: return nil")
+        }
+      }
+    }
+
     DeclSyntax(
       """
       /// This is really unfortunate. Really, we should have a `switch` in

Sources/SwiftParser/Declarations.swift

@@ -63,37 +63,7 @@ extension TokenConsumer {
 
     var subparser = self.lookahead()
 
-    var hasAttribute = false
-    var attributeProgress = LoopProgressCondition()
-    while subparser.hasProgressed(&attributeProgress) {
-      if subparser.at(.atSign) {
-        _ = subparser.consumeAttributeList()
-        hasAttribute = true
-      } else if subparser.at(.poundIf) && subparser.consumeIfConfigOfAttributes() {
-        subparser.skipSingle()
-        hasAttribute = true
-      } else {
-        break
-      }
-    }
-
-    var hasModifier = false
-    if subparser.currentToken.isLexerClassifiedKeyword || subparser.currentToken.rawTokenKind == .identifier {
-      var modifierProgress = LoopProgressCondition()
-      while let (modifierKind, handle) = subparser.at(anyIn: DeclarationModifier.self),
-        modifierKind != .class,
-        subparser.hasProgressed(&modifierProgress)
-      {
-        hasModifier = true
-        subparser.eat(handle)
-        if modifierKind != .open && subparser.at(.leftParen) && modifierKind.canHaveParenthesizedArgument {
-          // When determining whether we are at a declaration, don't consume anything in parentheses after 'open'
-          // so we don't consider a function call to open as a decl modifier. This matches the C++ parser.
-          subparser.consumeAnyToken()
-          subparser.consume(to: .rightParen)
-        }
-      }
-    }
+    let (hasAttribute, hasModifier) = subparser.skipAttributesAndModifiers()
 
     if hasAttribute {
       if subparser.at(.rightBrace) || subparser.at(.endOfFile) || subparser.at(.poundEndif) {
@@ -118,17 +88,7 @@ extension TokenConsumer {
     switch declStartKeyword {
     case .lhs(.actor):
       // actor Foo {}
-      if subparser.peek().rawTokenKind == .identifier {
-        return true
-      }
-      // actor may be somewhere in the modifier list. Eat the tokens until we get
-      // to something that isn't the start of a decl. If that is an identifier,
-      // it's an actor declaration, otherwise, it isn't.
-      var lookahead = subparser.lookahead()
-      repeat {
-        lookahead.consumeAnyToken()
-      } while lookahead.atStartOfDeclaration(isAtTopLevel: isAtTopLevel, allowInitDecl: allowInitDecl)
-      return lookahead.at(.identifier)
+      return subparser.atStartOfActor(isAtTopLevel: isAtTopLevel, allowInitDecl: allowInitDecl)
     case .lhs(.case):
       // When 'case' appears inside a function, it's probably a switch
       // case, not an enum case declaration.
@@ -152,23 +112,7 @@ extension TokenConsumer {
         return false
       }
 
-      var lookahead = subparser.lookahead()
-
-      // Consume 'using'
-      lookahead.consumeAnyToken()
-
-      // Allow parsing 'using' as declaration only if
-      // it's immediately followed by either `@` or
-      // an identifier.
-      if lookahead.atStartOfLine {
-        return false
-      }
-
-      guard lookahead.at(.atSign) || lookahead.at(.identifier) else {
-        return false
-      }
-
-      return true
+      return subparser.atStartOfUsing()
     case .some(_):
       // All other decl start keywords unconditionally start a decl.
       return true
@@ -190,6 +134,81 @@ extension TokenConsumer {
   }
 }
 
+extension Parser.Lookahead {
+  fileprivate mutating func skipAttributesAndModifiers() -> (hasAttribute: Bool, hasModifier: Bool) {
+    var hasAttribute = false
+    var attributeProgress = LoopProgressCondition()
+    while self.hasProgressed(&attributeProgress) {
+      if self.at(.atSign) {
+        _ = self.consumeAttributeList()
+        hasAttribute = true
+      } else if self.at(.poundIf) && self.consumeIfConfigOfAttributes() {
+        self.skipSingle()
+        hasAttribute = true
+      } else {
+        break
+      }
+    }
+
+    var hasModifier = false
+    if self.currentToken.isLexerClassifiedKeyword || self.currentToken.rawTokenKind == .identifier {
+      var modifierProgress = LoopProgressCondition()
+      while let (modifierKind, handle) = self.at(anyIn: DeclarationModifier.self),
+        modifierKind != .class,
+        self.hasProgressed(&modifierProgress)
+      {
+        hasModifier = true
+        self.eat(handle)
+        if modifierKind != .open && self.at(.leftParen) && modifierKind.canHaveParenthesizedArgument {
+          // When determining whether we are at a declaration, don't consume anything in parentheses after 'open'
+          // so we don't consider a function call to open as a decl modifier. This matches the C++ parser.
+          self.consumeAnyToken()
+          self.consume(to: .rightParen)
+        }
+      }
+    }
+
+    return (hasAttribute, hasModifier)
+  }
+
+  fileprivate mutating func atStartOfActor(
+    isAtTopLevel: Bool,
+    allowInitDecl: Bool
+  ) -> Bool {
+    if self.peek().rawTokenKind == .identifier {
+      return true
+    }
+    // actor may be somewhere in the modifier list. Eat the tokens until we get
+    // to something that isn't the start of a decl. If that is an identifier,
+    // it's an actor declaration, otherwise, it isn't.
+    var lookahead = self.lookahead()
+    repeat {
+      lookahead.consumeAnyToken()
+    } while lookahead.atStartOfDeclaration(isAtTopLevel: isAtTopLevel, allowInitDecl: allowInitDecl)
+    return lookahead.at(.identifier)
+  }
+
+  fileprivate mutating func atStartOfUsing() -> Bool {
+    var lookahead = self.lookahead()
+
+    // Consume 'using'
+    lookahead.consumeAnyToken()
+
+    // Allow parsing 'using' as declaration only if
+    // it's immediately followed by either `@` or
+    // an identifier.
+    if lookahead.atStartOfLine {
+      return false
+    }
+
+    guard lookahead.at(.atSign) || lookahead.at(.identifier) else {
+      return false
+    }
+
+    return true
+  }
+}
+
 extension Parser {
   struct DeclAttributes {
     var attributes: RawAttributeListSyntax