TokenStreamCreator.swift

//===----------------------------------------------------------------------===//
//
// This source file is part of the Swift.org open source project
//
// Copyright (c) 2014 - 2019 Apple Inc. and the Swift project authors
// Licensed under Apache License v2.0 with Runtime Library Exception
//
// See https://swift.org/LICENSE.txt for license information
// See https://swift.org/CONTRIBUTORS.txt for the list of Swift project authors
//
//===----------------------------------------------------------------------===//

import Foundation
import SwiftFormatConfiguration
import SwiftFormatCore
import SwiftOperators
import SwiftSyntax

/// Visits the nodes of a syntax tree and constructs a linear stream of formatting tokens that
/// tell the pretty printer how the source text should be laid out.
fileprivate final class TokenStreamCreator: SyntaxVisitor {
  private var tokens = [Token]()
  private var beforeMap = [TokenSyntax: [Token]]()
  private var afterMap = [TokenSyntax: [[Token]]]()
  private let config: Configuration
  private let operatorTable: OperatorTable
  private let maxlinelength: Int

  /// The index of the most recently appended break, or nil when no break has been appended.
  private var lastBreakIndex: Int? = nil

  /// Whether newlines can be merged into the most recent break, based on which tokens have been
  /// appended since that break.
  private var canMergeNewlinesIntoLastBreak = false

  /// Keeps track of the kind of break that should be used inside a multiline string. This differs
  /// depending on surrounding context due to some tricky special cases, so this lets us pass that
  /// information down to the strings that need it.
  private var pendingMultilineStringBreakKinds = [StringLiteralExprSyntax: BreakKind]()

  /// Lists tokens that shouldn't be appended to the token stream as `syntax` tokens. They will be
  /// printed conditionally using a different type of token.
  private var ignoredTokens = Set<TokenSyntax>()

  /// Lists the expressions that have been visited, from the outermost expression, where contextual
  /// breaks and start/end contextual breaking tokens have been inserted.
  private var preVisitedExprs = Set<SyntaxIdentifier>()

  /// Tracks the "root" exprs where previsiting for contextual breaks started so that
  /// `preVisitedExprs` can be emptied after exiting an expr tree.
  private var rootExprs = Set<SyntaxIdentifier>()

  /// Lists the tokens that are the closing or final delimiter of a node that shouldn't be split
  /// from the preceding token. When breaks are inserted around compound expressions, the breaks are
  /// moved past these tokens.
  private var closingDelimiterTokens = Set<TokenSyntax>()

  /// Tracks closures that are never allowed to be laid out entirely on one line (e.g., closures
  /// in a function call containing multiple trailing closures).
  private var forcedBreakingClosures = Set<SyntaxIdentifier>()

  init(configuration: Configuration, operatorTable: OperatorTable) {
    self.config = configuration
    self.operatorTable = operatorTable
    self.maxlinelength = config.lineLength
    super.init(viewMode: .all)
  }

  func makeStream(from node: Syntax) -> [Token] {
    // Because `walk` takes an `inout` argument, and we're a class, we have to do the following
    // dance to pass ourselves in.
    self.walk(node)
    defer { tokens = [] }
    return tokens
  }

  var openings = 0

  /// If the syntax token is non-nil, enqueue the given list of formatting tokens before it in the
  /// token stream.
  func before(_ token: TokenSyntax?, tokens: Token...) {
    before(token, tokens: tokens)
  }

  /// If the syntax token is non-nil, enqueue the given list of formatting tokens before it in the
  /// token stream.
  func before(_ token: TokenSyntax?, tokens: [Token]) {
    guard let tok = token else { return }
    beforeMap[tok, default: []] += tokens
  }

  /// If the syntax token is non-nil, enqueue the given list of formatting tokens after it in the
  /// token stream.
  func after(_ token: TokenSyntax?, tokens: Token...) {
    after(token, tokens: tokens)
  }

  /// If the syntax token is non-nil, enqueue the given list of formatting tokens after it in the
  /// token stream.
  func after(_ token: TokenSyntax?, tokens: [Token]) {
    guard let tok = token else { return }
    afterMap[tok, default: []].append(tokens)
  }

  /// Enqueues the given list of formatting tokens between each element of the given syntax
  /// collection (but not before the first one nor after the last one).
  private func insertTokens<Node: SyntaxCollection>(
    _ tokens: Token...,
    betweenElementsOf collectionNode: Node
  ) where Node.Element == Syntax {
    for element in collectionNode.dropLast() {
      after(element.lastToken, tokens: tokens)
    }
  }

  /// Enqueues the given list of formatting tokens between each element of the given syntax
  /// collection (but not before the first one nor after the last one).
  private func insertTokens<Node: SyntaxCollection>(
    _ tokens: Token...,
    betweenElementsOf collectionNode: Node
  ) where Node.Element: SyntaxProtocol {
    for element in collectionNode.dropLast() {
      after(element.lastToken, tokens: tokens)
    }
  }

  /// Enqueues the given list of formatting tokens between each element of the given syntax
  /// collection (but not before the first one nor after the last one).
  private func insertTokens<Node: SyntaxCollection>(
    _ tokens: Token...,
    betweenElementsOf collectionNode: Node
  ) where Node.Element == DeclSyntax {
    for element in collectionNode.dropLast() {
      after(element.lastToken, tokens: tokens)
    }
  }

  private func verbatimToken(_ node: Syntax, indentingBehavior: IndentingBehavior = .allLines) {
    if let firstToken = node.firstToken {
      appendBeforeTokens(firstToken)
    }

    appendToken(.verbatim(Verbatim(text: node.description, indentingBehavior: indentingBehavior)))

    if let lastToken = node.lastToken {
      // Extract any comments that trail the verbatim block since they belong to the next syntax
      // token. Leading comments don't need special handling since they belong to the current node,
      // and will get printed.
      appendAfterTokensAndTrailingComments(lastToken)
    }
  }

  // MARK: - Type declaration nodes

  override func visit(_ node: ClassDeclSyntax) -> SyntaxVisitorContinueKind {
    arrangeTypeDeclBlock(
      Syntax(node),
      attributes: node.attributes,
      modifiers: node.modifiers,
      typeKeyword: node.classKeyword,
      identifier: node.identifier,
      genericParameterOrPrimaryAssociatedTypeClause: node.genericParameterClause.map(Syntax.init),
      inheritanceClause: node.inheritanceClause,
      genericWhereClause: node.genericWhereClause,
      members: node.members)
    return .visitChildren
  }

  override func visit(_ node: ActorDeclSyntax) -> SyntaxVisitorContinueKind {
    arrangeTypeDeclBlock(
      Syntax(node),
      attributes: node.attributes,
      modifiers: node.modifiers,
      typeKeyword: node.actorKeyword,
      identifier: node.identifier,
      genericParameterOrPrimaryAssociatedTypeClause: node.genericParameterClause.map(Syntax.init),
      inheritanceClause: node.inheritanceClause,
      genericWhereClause: node.genericWhereClause,
      members: node.members)
    return .visitChildren
  }

  override func visit(_ node: StructDeclSyntax) -> SyntaxVisitorContinueKind {
    arrangeTypeDeclBlock(
      Syntax(node),
      attributes: node.attributes,
      modifiers: node.modifiers,
      typeKeyword: node.structKeyword,
      identifier: node.identifier,
      genericParameterOrPrimaryAssociatedTypeClause: node.genericParameterClause.map(Syntax.init),
      inheritanceClause: node.inheritanceClause,
      genericWhereClause: node.genericWhereClause,
      members: node.members)
    return .visitChildren
  }

  override func visit(_ node: EnumDeclSyntax) -> SyntaxVisitorContinueKind {
    arrangeTypeDeclBlock(
      Syntax(node),
      attributes: node.attributes,
      modifiers: node.modifiers,
      typeKeyword: node.enumKeyword,
      identifier: node.identifier,
      genericParameterOrPrimaryAssociatedTypeClause: node.genericParameters.map(Syntax.init),
      inheritanceClause: node.inheritanceClause,
      genericWhereClause: node.genericWhereClause,
      members: node.members)
    return .visitChildren
  }

  override func visit(_ node: ProtocolDeclSyntax) -> SyntaxVisitorContinueKind {
    arrangeTypeDeclBlock(
      Syntax(node),
      attributes: node.attributes,
      modifiers: node.modifiers,
      typeKeyword: node.protocolKeyword,
      identifier: node.identifier,
      genericParameterOrPrimaryAssociatedTypeClause:
        node.primaryAssociatedTypeClause.map(Syntax.init),
      inheritanceClause: node.inheritanceClause,
      genericWhereClause: node.genericWhereClause,
      members: node.members)
    return .visitChildren
  }

  override func visit(_ node: ExtensionDeclSyntax) -> SyntaxVisitorContinueKind {
    guard let lastTokenOfExtendedType = node.extendedType.lastToken else {
      fatalError("ExtensionDeclSyntax.extendedType must have at least one token")
    }
    arrangeTypeDeclBlock(
      Syntax(node),
      attributes: node.attributes,
      modifiers: node.modifiers,
      typeKeyword: node.extensionKeyword,
      identifier: lastTokenOfExtendedType,
      genericParameterOrPrimaryAssociatedTypeClause: nil,
      inheritanceClause: node.inheritanceClause,
      genericWhereClause: node.genericWhereClause,
      members: node.members)
    return .visitChildren
  }

  /// Applies formatting tokens to the tokens in the given type declaration node (i.e., a class,
  /// struct, enum, protocol, or extension).
  private func arrangeTypeDeclBlock(
    _ node: Syntax,
    attributes: AttributeListSyntax?,
    modifiers: ModifierListSyntax?,
    typeKeyword: TokenSyntax,
    identifier: TokenSyntax,
    genericParameterOrPrimaryAssociatedTypeClause: Syntax?,
    inheritanceClause: TypeInheritanceClauseSyntax?,
    genericWhereClause: GenericWhereClauseSyntax?,
    members: MemberDeclBlockSyntax
  ) {
    before(node.firstToken, tokens: .open)

    arrangeAttributeList(attributes)

    // Prioritize keeping "<modifiers> <keyword> <name>:" together (corresponding group close is
    // below at `lastTokenBeforeBrace`).
    let firstTokenAfterAttributes = modifiers?.firstToken ?? typeKeyword
    before(firstTokenAfterAttributes, tokens: .open)
    after(typeKeyword, tokens: .break)

    arrangeBracesAndContents(of: members, contentsKeyPath: \.members)

    if let genericWhereClause = genericWhereClause {
      before(genericWhereClause.firstToken, tokens: .break(.same), .open)
      after(members.leftBrace, tokens: .close)
    }

    let lastTokenBeforeBrace = inheritanceClause?.colon
      ?? genericParameterOrPrimaryAssociatedTypeClause?.lastToken
      ?? identifier
    after(lastTokenBeforeBrace, tokens: .close)

    after(node.lastToken, tokens: .close)
  }

  // MARK: - Function and function-like declaration nodes (initializers, deinitializers, subscripts)

  override func visit(_ node: FunctionDeclSyntax) -> SyntaxVisitorContinueKind {
    let hasArguments = !node.signature.input.parameterList.isEmpty

    // Prioritize keeping ") throws -> <return_type>" together. We can only do this if the function
    // has arguments.
    if hasArguments && config.prioritizeKeepingFunctionOutputTogether {
      // Due to visitation order, the matching .open break is added in ParameterClauseSyntax.
      after(node.signature.lastToken, tokens: .close)
    }

    let mustBreak = node.body != nil || node.signature.output != nil
    arrangeParameterClause(node.signature.input, forcesBreakBeforeRightParen: mustBreak)

    // Prioritize keeping "<modifiers> func <name>(" together. Also include the ")" if the parameter
    // list is empty.
    let firstTokenAfterAttributes = node.modifiers?.firstToken ?? node.funcKeyword
    before(firstTokenAfterAttributes, tokens: .open)
    after(node.funcKeyword, tokens: .break)
    if hasArguments || node.genericParameterClause != nil {
      after(node.signature.input.leftParen, tokens: .close)
    } else {
      after(node.signature.input.rightParen, tokens: .close)
    }

    // Add a non-breaking space after the identifier if it's an operator, to separate it visually
    // from the following parenthesis or generic argument list. Note that even if the function is
    // defining a prefix or postfix operator, the token kind always comes through as
    // `binaryOperator`.
    if case .binaryOperator = node.identifier.tokenKind {
      after(node.identifier.lastToken, tokens: .space)
    }

    arrangeFunctionLikeDecl(
      Syntax(node),
      attributes: node.attributes,
      genericWhereClause: node.genericWhereClause,
      body: node.body,
      bodyContentsKeyPath: \.statements)

    return .visitChildren
  }

  override func visit(_ node: InitializerDeclSyntax) -> SyntaxVisitorContinueKind {
    let hasArguments = !node.signature.input.parameterList.isEmpty

    // Prioritize keeping ") throws" together. We can only do this if the function
    // has arguments.
    if hasArguments && config.prioritizeKeepingFunctionOutputTogether {
      // Due to visitation order, the matching .open break is added in ParameterClauseSyntax.
      after(node.signature.lastToken, tokens: .close)
    }

    arrangeParameterClause(node.signature.input, forcesBreakBeforeRightParen: node.body != nil)

    // Prioritize keeping "<modifiers> init<punctuation>" together.
    let firstTokenAfterAttributes = node.modifiers?.firstToken ?? node.initKeyword
    before(firstTokenAfterAttributes, tokens: .open)

    if hasArguments || node.genericParameterClause != nil {
      after(node.signature.input.leftParen, tokens: .close)
    } else {
      after(node.signature.input.rightParen, tokens: .close)
    }

    arrangeFunctionLikeDecl(
      Syntax(node),
      attributes: node.attributes,
      genericWhereClause: node.genericWhereClause,
      body: node.body,
      bodyContentsKeyPath: \.statements)

    return .visitChildren
  }

  override func visit(_ node: DeinitializerDeclSyntax) -> SyntaxVisitorContinueKind {
    arrangeFunctionLikeDecl(
      Syntax(node),
      attributes: node.attributes,
      genericWhereClause: nil,
      body: node.body,
      bodyContentsKeyPath: \.statements)
    return .visitChildren
  }

  override func visit(_ node: SubscriptDeclSyntax) -> SyntaxVisitorContinueKind {
    let hasArguments = !node.indices.parameterList.isEmpty

    before(node.firstToken, tokens: .open)

    // Prioritize keeping "<modifiers> subscript" together.
    if let firstModifierToken = node.modifiers?.firstToken {
      before(firstModifierToken, tokens: .open)

      if hasArguments || node.genericParameterClause != nil {
        after(node.indices.leftParen, tokens: .close)
      } else {
        after(node.indices.rightParen, tokens: .close)
      }
    }

    // Prioritize keeping ") -> <return_type>" together. We can only do this if the subscript has
    // arguments.
    if hasArguments && config.prioritizeKeepingFunctionOutputTogether {
      // Due to visitation order, the matching .open break is added in ParameterClauseSyntax.
      after(node.result.lastToken, tokens: .close)
    }

    arrangeAttributeList(node.attributes)

    if let genericWhereClause = node.genericWhereClause {
      before(genericWhereClause.firstToken, tokens: .break(.same), .open)
      after(genericWhereClause.lastToken, tokens: .close)
    }

    before(node.result.firstToken, tokens: .break)

    if let accessorOrCodeBlock = node.accessor {
      switch accessorOrCodeBlock {
      case .accessors(let accessorBlock):
        arrangeBracesAndContents(of: accessorBlock)
      case .getter(let codeBlock):
        arrangeBracesAndContents(of: codeBlock, contentsKeyPath: \.statements)
      }
    }

    after(node.lastToken, tokens: .close)

    arrangeParameterClause(node.indices, forcesBreakBeforeRightParen: true)

    return .visitChildren
  }

  /// Applies formatting tokens to the tokens in the given function or function-like declaration
  /// node (e.g., initializers, deinitiailizers, and subscripts).
  private func arrangeFunctionLikeDecl<Node: BracedSyntax, BodyContents: SyntaxCollection>(
    _ node: Syntax,
    attributes: AttributeListSyntax?,
    genericWhereClause: GenericWhereClauseSyntax?,
    body: Node?,
    bodyContentsKeyPath: KeyPath<Node, BodyContents>?
  ) where BodyContents.Element: SyntaxProtocol {
    before(node.firstToken, tokens: .open)

    arrangeAttributeList(attributes)
    arrangeBracesAndContents(of: body, contentsKeyPath: bodyContentsKeyPath)

    if let genericWhereClause = genericWhereClause {
      before(genericWhereClause.firstToken, tokens: .break(.same), .open)
      after(body?.leftBrace ?? genericWhereClause.lastToken, tokens: .close)
    }

    after(node.lastToken, tokens: .close)
  }

  // MARK: - Property and subscript accessor block nodes

  override func visit(_ node: AccessorListSyntax) -> SyntaxVisitorContinueKind {
    for child in node.dropLast() {
      // If the child doesn't have a body (it's just the `get`/`set` keyword), then we're in a
      // protocol and we want to let them be placed on the same line if possible. Otherwise, we
      // place a newline between each accessor.
      let newlines: NewlineBehavior = child.body == nil ? .elective : .soft
      after(child.lastToken, tokens: .break(.same, size: 1, newlines: newlines))
    }
    return .visitChildren
  }

  override func visit(_ node: AccessorDeclSyntax) -> SyntaxVisitorContinueKind {
    arrangeAttributeList(node.attributes)

    if let asyncKeyword = node.effectSpecifiers?.asyncSpecifier {
      if node.effectSpecifiers?.throwsSpecifier != nil {
        before(asyncKeyword, tokens: .break, .open)
      } else {
        before(asyncKeyword, tokens: .break)
      }
    }

    if let throwsKeyword = node.effectSpecifiers?.throwsSpecifier {
      before(node.effectSpecifiers?.throwsSpecifier, tokens: .break)
      if node.effectSpecifiers?.asyncSpecifier != nil {
        after(throwsKeyword, tokens: .close)
      }
    }

    arrangeBracesAndContents(of: node.body, contentsKeyPath: \.statements)
    return .visitChildren
  }

  override func visit(_ node: AccessorParameterSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  // MARK: - Control flow statement nodes

  override func visit(_ node: LabeledStmtSyntax) -> SyntaxVisitorContinueKind {
    after(node.labelColon, tokens: .space)
    return .visitChildren
  }

  override func visit(_ node: IfExprSyntax) -> SyntaxVisitorContinueKind {
    // There may be a consistent breaking group around this node, see `CodeBlockItemSyntax`. This
    // group is necessary so that breaks around and inside of the conditions aren't forced to break
    // when the if-stmt spans multiple lines.
    before(node.conditions.firstToken, tokens: .open)
    after(node.conditions.lastToken, tokens: .close)

    after(node.ifKeyword, tokens: .space)

    // Add break groups, using open continuation breaks, around any conditions after the first so
    // that continuations inside of the conditions can stack in addition to continuations between
    // the conditions. There are no breaks around the first condition because if-statements look
    // better without a break between the "if" and the first condition.
    for condition in node.conditions.dropFirst() {
      before(condition.firstToken, tokens: .break(.open(kind: .continuation), size: 0))
      after(condition.lastToken, tokens: .break(.close(mustBreak: false), size: 0))
    }

    arrangeBracesAndContents(of: node.body, contentsKeyPath: \.statements)

    if let elseKeyword = node.elseKeyword {
      // Add a token before the else keyword. Breaking before `else` is explicitly allowed when
      // there's a comment.
      if config.lineBreakBeforeControlFlowKeywords {
        before(elseKeyword, tokens: .break(.same, newlines: .soft))
      } else if elseKeyword.leadingTrivia.hasLineComment {
        before(elseKeyword, tokens: .break(.same, size: 1))
      } else {
        before(elseKeyword, tokens: .space)
      }

      // Breaks are only allowed after `else` when there's a comment; otherwise there shouldn't be
      // any newlines between `else` and the open brace or a following `if`.
      if let tokenAfterElse = elseKeyword.nextToken(viewMode: .all), tokenAfterElse.leadingTrivia.hasLineComment {
        after(node.elseKeyword, tokens: .break(.same, size: 1))
      } else if let elseBody = node.elseBody, elseBody.is(IfExprSyntax.self) {
        after(node.elseKeyword, tokens: .space)
      }
    }

    arrangeBracesAndContents(of: node.elseBody?.as(CodeBlockSyntax.self), contentsKeyPath: \.statements)

    return .visitChildren
  }

  override func visit(_ node: GuardStmtSyntax) -> SyntaxVisitorContinueKind {
    after(node.guardKeyword, tokens: .space)

    // Add break groups, using open continuation breaks, around all conditions so that continuations
    // inside of the conditions can stack in addition to continuations between the conditions.
    for condition in node.conditions {
      before(condition.firstToken, tokens: .break(.open(kind: .continuation), size: 0))
      after(condition.lastToken, tokens: .break(.close(mustBreak: false), size: 0))
    }

    before(node.elseKeyword, tokens: .break(.reset), .open)
    after(node.elseKeyword, tokens: .space)
    before(node.body.leftBrace, tokens: .close)

    arrangeBracesAndContents(
      of: node.body, contentsKeyPath: \.statements, shouldResetBeforeLeftBrace: false)

    return .visitChildren
  }

  override func visit(_ node: ForInStmtSyntax) -> SyntaxVisitorContinueKind {
    // If we have a `(try) await` clause, allow breaking after the `for` so that the `(try) await`
    // can fall onto the next line if needed, and if both `try await` are present, keep them
    // together. Otherwise, keep `for` glued to the token after it so that we break somewhere later
    // on the line.
    if let awaitKeyword = node.awaitKeyword {
      after(node.forKeyword, tokens: .break)
      if let tryKeyword = node.tryKeyword {
        before(tryKeyword, tokens: .open)
        after(tryKeyword, tokens: .break)
        after(awaitKeyword, tokens: .close, .break)
      } else {
        after(awaitKeyword, tokens: .break)
      }
    } else {
      after(node.forKeyword, tokens: .space)
    }

    after(node.caseKeyword, tokens: .space)
    before(node.inKeyword, tokens: .break)
    after(node.inKeyword, tokens: .space)

    if let typeAnnotation = node.typeAnnotation {
      after(
        typeAnnotation.colon,
        tokens: .break(.open(kind: .continuation), newlines: .elective(ignoresDiscretionary: true)))
      after(typeAnnotation.lastToken, tokens: .break(.close(mustBreak: false), size: 0))
    }

    arrangeBracesAndContents(of: node.body, contentsKeyPath: \.statements)

    return .visitChildren
  }

  override func visit(_ node: WhileStmtSyntax) -> SyntaxVisitorContinueKind {
    after(node.whileKeyword, tokens: .space)

    // Add break groups, using open continuation breaks, around any conditions after the first so
    // that continuations inside of the conditions can stack in addition to continuations between
    // the conditions. There are no breaks around the first condition because there was historically
    // not break after the while token and adding such a break would cause excessive changes to
    // previously formatted code.
    // This has the side effect that the label + `while` + tokens up to the first break in the first
    // condition could be longer than the column limit since there are no breaks between the label
    // or while token.
    for condition in node.conditions.dropFirst() {
      before(condition.firstToken, tokens: .break(.open(kind: .continuation), size: 0))
      after(condition.lastToken, tokens: .break(.close(mustBreak: false), size: 0))
    }

    arrangeBracesAndContents(of: node.body, contentsKeyPath: \.statements)

    return .visitChildren
  }

  override func visit(_ node: RepeatWhileStmtSyntax) -> SyntaxVisitorContinueKind {
    arrangeBracesAndContents(of: node.body, contentsKeyPath: \.statements)

    if config.lineBreakBeforeControlFlowKeywords {
      before(node.whileKeyword, tokens: .break(.same), .open)
      after(node.condition.lastToken, tokens: .close)
    } else {
      // The length of the condition needs to force the breaks around the braces of the repeat
      // stmt's body, so that there's always a break before the right brace when the while &
      // condition is too long to be on one line.
      before(node.whileKeyword, tokens: .space)
      // The `open` token occurs after the ending tokens for the braced `body` node.
      before(node.body.rightBrace, tokens: .open)
      after(node.condition.lastToken, tokens: .close)
    }
    after(node.whileKeyword, tokens: .space)
    return .visitChildren
  }

  override func visit(_ node: DoStmtSyntax) -> SyntaxVisitorContinueKind {
    arrangeBracesAndContents(of: node.body, contentsKeyPath: \.statements)
    return .visitChildren
  }

  override func visit(_ node: CatchClauseSyntax) -> SyntaxVisitorContinueKind {
    let catchPrecedingBreak = config.lineBreakBeforeControlFlowKeywords
      ? Token.break(.same, newlines: .soft) : Token.space
    before(node.catchKeyword, tokens: catchPrecedingBreak)

    if let catchItems = node.catchItems {
      // If there are multiple items in the `catch` clause, wrap each in open/close breaks so that
      // their internal breaks stack correctly. Otherwise, if there is only a single clause, use the
      // old (pre-SE-0276) behavior (a fixed space after the `catch` keyword).
      if catchItems.count > 1 {
        for catchItem in catchItems {
          before(catchItem.firstToken, tokens: .break(.open(kind: .continuation)))
          after(catchItem.lastToken, tokens: .break(.close(mustBreak: false), size: 0))
        }
      } else {
        before(node.catchItems?.firstToken, tokens: .space)
      }
    }

    arrangeBracesAndContents(of: node.body, contentsKeyPath: \.statements)

    return .visitChildren
  }

  override func visit(_ node: DeferStmtSyntax) -> SyntaxVisitorContinueKind {
    arrangeBracesAndContents(of: node.body, contentsKeyPath: \.statements)
    return .visitChildren
  }

  override func visit(_ node: BreakStmtSyntax) -> SyntaxVisitorContinueKind {
    before(node.label, tokens: .break)
    return .visitChildren
  }

  override func visit(_ node: ReturnStmtSyntax) -> SyntaxVisitorContinueKind {
    if let expression = node.expression {
      if leftmostMultilineStringLiteral(of: expression) != nil {
        before(expression.firstToken, tokens: .break(.open))
        after(expression.lastToken, tokens: .break(.close(mustBreak: false)))
      } else {
        before(expression.firstToken, tokens: .break)
      }
    }
    return .visitChildren
  }

  override func visit(_ node: ThrowStmtSyntax) -> SyntaxVisitorContinueKind {
    before(node.expression.firstToken, tokens: .break)
    return .visitChildren
  }

  override func visit(_ node: ContinueStmtSyntax) -> SyntaxVisitorContinueKind {
    before(node.label, tokens: .break)
    return .visitChildren
  }

  override func visit(_ node: SwitchExprSyntax) -> SyntaxVisitorContinueKind {
    before(node.switchKeyword, tokens: .open)
    after(node.switchKeyword, tokens: .space)
    before(node.leftBrace, tokens: .break(.reset))
    after(node.leftBrace, tokens: .close)

    // An if-configuration clause around a switch-case encloses the case's node, so an
    // if-configuration clause requires a break here in order to be allowed on a new line.
    for ifConfigDecl in node.cases.filter({ $0.is(IfConfigDeclSyntax.self) }) {
      if config.indentSwitchCaseLabels {
        before(ifConfigDecl.firstToken, tokens: .break(.open))
        after(ifConfigDecl.lastToken, tokens: .break(.close, size: 0))
      } else {
        before(ifConfigDecl.firstToken, tokens: .break(.same))
      }
    }

    let newlines: NewlineBehavior =
      areBracesCompletelyEmpty(node, contentsKeyPath: \.cases) ? .elective : .soft
    before(node.rightBrace, tokens: .break(.same, size: 0, newlines: newlines))

    return .visitChildren
  }

  override func visit(_ node: SwitchCaseSyntax) -> SyntaxVisitorContinueKind {
    // If switch/case labels were configured to be indented, use an `open` break; otherwise, use
    // the default `same` break.
    let openBreak: Token
    if config.indentSwitchCaseLabels {
      openBreak = .break(.open, newlines: .elective)
    } else {
      openBreak = .break(.same, newlines: .soft)
    }
    before(node.firstToken, tokens: openBreak)

    after(node.unknownAttr?.lastToken, tokens: .space)
    after(node.label.lastToken, tokens: .break(.reset, size: 0), .break(.open), .open)

    // If switch/case labels were configured to be indented, insert an extra `close` break after the
    // case body to match the `open` break above
    var afterLastTokenTokens: [Token] = [.break(.close, size: 0), .close]
    if config.indentSwitchCaseLabels {
      afterLastTokenTokens.append(.break(.close, size: 0))
    }
    after(node.lastToken, tokens: afterLastTokenTokens)

    return .visitChildren
  }

  override func visit(_ node: SwitchCaseLabelSyntax) -> SyntaxVisitorContinueKind {
    before(node.caseKeyword, tokens: .open)
    after(node.caseKeyword, tokens: .space)

    // If an item with a `where` clause follows an item without a `where` clause, the compiler emits
    // a warning telling the user that they should insert a newline between them to disambiguate
    // their appearance. We enforce that "requirement" here to avoid spurious warnings, especially
    // following a `NoCasesWithOnlyFallthrough` transformation that might merge cases.
    let caseItems = Array(node.caseItems)
    for (index, item) in caseItems.enumerated() {
      before(item.firstToken, tokens: .open)
      if let trailingComma = item.trailingComma {
        // Insert a newline before the next item if it has a where clause and this item doesn't.
        let nextItemHasWhereClause =
          index + 1 < caseItems.endIndex && caseItems[index + 1].whereClause != nil
        let requiresNewline = item.whereClause == nil && nextItemHasWhereClause
        let newlines: NewlineBehavior = requiresNewline ? .soft : .elective
        after(trailingComma, tokens: .close, .break(.continue, size: 1, newlines: newlines))
      } else {
        after(item.lastToken, tokens: .close)
      }
    }

    after(node.colon, tokens: .close)
    closingDelimiterTokens.insert(node.colon)
    return .visitChildren
  }

  override func visit(_ node: YieldStmtSyntax) -> SyntaxVisitorContinueKind {
    // As of https://github.com/apple/swift-syntax/pull/895, the token following a `yield` keyword
    // *must* be on the same line, so we cannot break here.
    after(node.yieldKeyword, tokens: .space)
    return .visitChildren
  }

  // TODO: - Other nodes (yet to be organized)

  override func visit(_ node: DeclNameArgumentsSyntax) -> SyntaxVisitorContinueKind {
    after(node.leftParen, tokens: .break(.open, size: 0), .open(argumentListConsistency()))
    before(node.rightParen, tokens: .break(.close(mustBreak: false), size: 0), .close)
    insertTokens(.break(.same, size: 0), betweenElementsOf: node.arguments)
    return .visitChildren
  }

  override func visit(_ node: TupleExprSyntax) -> SyntaxVisitorContinueKind {
    // We'll do nothing if it's a zero-element tuple, because we just want to keep the empty `()`
    // together.
    let elementCount = node.elementList.count

    if elementCount == 1 {
      // A tuple with one element is a parenthesized expression; add a group around it to keep it
      // together when possible, but breaks are handled elsewhere (see calls to
      // `stackedIndentationBehavior`).
      after(node.leftParen, tokens: .open)
      before(node.rightParen, tokens: .close)
      closingDelimiterTokens.insert(node.rightParen)

      // When there's a comment inside of a parenthesized expression, we want to allow the comment
      // to exist at the EOL with the left paren or on its own line. The contents are always
      // indented on the following lines, since parens always create a scope. An open/close break
      // pair isn't used here to avoid forcing the closing paren down onto a new line.
      if node.leftParen.nextToken(viewMode: .all)?.leadingTrivia.hasLineComment ?? false {
        after(node.leftParen, tokens: .break(.continue, size: 0))
      }
    } else if elementCount > 1 {
      // Tuples with more than one element are "true" tuples, and should indent as block structures.
      after(node.leftParen, tokens: .break(.open, size: 0), .open)
      before(node.rightParen, tokens: .break(.close, size: 0), .close)

      insertTokens(.break(.same), betweenElementsOf: node.elementList)

      for element in node.elementList {
        arrangeAsTupleExprElement(element)
      }
    }

    return .visitChildren
  }

  override func visit(_ node: TupleExprElementListSyntax) -> SyntaxVisitorContinueKind {
    // Intentionally do nothing here. Since `TupleExprElement`s are used both in tuple expressions
    // and function argument lists, which need to be formatted, differently, those nodes manually
    // loop over the nodes and arrange them in those contexts.
    return .visitChildren
  }

  override func visit(_ node: TupleExprElementSyntax) -> SyntaxVisitorContinueKind {
    // Intentionally do nothing here. Since `TupleExprElement`s are used both in tuple expressions
    // and function argument lists, which need to be formatted, differently, those nodes manually
    // loop over the nodes and arrange them in those contexts.
    return .visitChildren
  }

  /// Arranges the given tuple expression element as a tuple element (rather than a function call
  /// argument).
  ///
  /// - Parameter node: The tuple expression element to be arranged.
  private func arrangeAsTupleExprElement(_ node: TupleExprElementSyntax) {
    before(node.firstToken, tokens: .open)
    after(node.colon, tokens: .break)
    after(node.lastToken, tokens: .close)
    if let trailingComma = node.trailingComma {
      closingDelimiterTokens.insert(trailingComma)
    }
  }

  override func visit(_ node: ArrayExprSyntax) -> SyntaxVisitorContinueKind {
    if !node.elements.isEmpty || node.rightSquare.leadingTrivia.numberOfComments > 0 {
      after(node.leftSquare, tokens: .break(.open, size: 0), .open)
      before(node.rightSquare, tokens: .break(.close, size: 0), .close)
    }
    return .visitChildren
  }

  override func visit(_ node: ArrayElementListSyntax) -> SyntaxVisitorContinueKind {
    insertTokens(.break(.same), betweenElementsOf: node)

    for element in node {
      before(element.firstToken, tokens: .open)
      after(element.lastToken, tokens: .close)
      if let trailingComma = element.trailingComma {
        closingDelimiterTokens.insert(trailingComma)
      }
    }

    if let lastElement = node.last {
      if let trailingComma = lastElement.trailingComma {
        ignoredTokens.insert(trailingComma)
      }
      before(node.first?.firstToken, tokens: .commaDelimitedRegionStart)
      let endToken =
        Token.commaDelimitedRegionEnd(
          hasTrailingComma: lastElement.trailingComma != nil,
          isSingleElement: node.first == lastElement)
      after(lastElement.expression.lastToken, tokens: [endToken])
    }
    return .visitChildren
  }

  override func visit(_ node: ArrayElementSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: DictionaryExprSyntax) -> SyntaxVisitorContinueKind {
    // The node's content is either a `DictionaryElementListSyntax` or a `TokenSyntax` for a colon
    // token (for an empty dictionary).
    if !(node.content.as(DictionaryElementListSyntax.self)?.isEmpty ?? true)
      || node.content.leadingTrivia?.numberOfComments ?? 0 > 0
      || node.rightSquare.leadingTrivia.numberOfComments > 0
    {
      after(node.leftSquare, tokens: .break(.open, size: 0), .open)
      before(node.rightSquare, tokens: .break(.close, size: 0), .close)
    }
    return .visitChildren
  }

  override func visit(_ node: DictionaryElementListSyntax) -> SyntaxVisitorContinueKind {
    insertTokens(.break(.same), betweenElementsOf: node)

    for element in node {
      before(element.firstToken, tokens: .open)
      after(element.colon, tokens: .break)
      after(element.lastToken, tokens: .close)
      if let trailingComma = element.trailingComma {
        closingDelimiterTokens.insert(trailingComma)
      }
    }

    if let lastElement = node.last {
      if let trailingComma = lastElement.trailingComma {
        ignoredTokens.insert(trailingComma)
      }
      before(node.first?.firstToken, tokens: .commaDelimitedRegionStart)
      let endToken =
        Token.commaDelimitedRegionEnd(
          hasTrailingComma: lastElement.trailingComma != nil,
          isSingleElement: node.first == node.last)
      after(lastElement.lastToken, tokens: endToken)
    }
    return .visitChildren
  }

  override func visit(_ node: DictionaryTypeSyntax) -> SyntaxVisitorContinueKind {
    after(node.colon, tokens: .break)
    return .visitChildren
  }

  override func visit(_ node: DictionaryElementSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: MemberAccessExprSyntax) -> SyntaxVisitorContinueKind {
    preVisitInsertingContextualBreaks(node)

    return .visitChildren
  }

  override func visitPost(_ node: MemberAccessExprSyntax) {
    clearContextualBreakState(node)
  }

  override func visit(_ node: FunctionCallExprSyntax) -> SyntaxVisitorContinueKind {
    preVisitInsertingContextualBreaks(node)

    // If there are multiple trailing closures, force all the closures in the call to break.
    if let additionalTrailingClosures = node.additionalTrailingClosures {
      if let closure = node.trailingClosure {
        forcedBreakingClosures.insert(closure.id)
      }
      for additionalTrailingClosure in additionalTrailingClosures {
        forcedBreakingClosures.insert(additionalTrailingClosure.closure.id)
      }
    }

    if let calledMemberAccessExpr = node.calledExpression.as(MemberAccessExprSyntax.self) {
      if let base = calledMemberAccessExpr.base, base.is(IdentifierExprSyntax.self) {
        // When this function call is wrapped by a try-expr or await-expr, the group applied when
        // visiting that wrapping expression is sufficient. Adding another group here in that case
        // can result in unnecessarily breaking after the try/await keyword.
        if !(base.firstToken?.previousToken(viewMode: .all)?.parent?.is(TryExprSyntax.self) ?? false
          || base.firstToken?.previousToken(viewMode: .all)?.parent?.is(AwaitExprSyntax.self) ?? false) {
          before(base.firstToken, tokens: .open)
          after(calledMemberAccessExpr.name.lastToken, tokens: .close)
        }
      }
    }

    let arguments = node.argumentList

    // If there is a trailing closure, force the right parenthesis down to the next line so it
    // stays with the open curly brace.
    let breakBeforeRightParen =
      (node.trailingClosure != nil && !isCompactSingleFunctionCallArgument(arguments))
      || mustBreakBeforeClosingDelimiter(of: node, argumentListPath: \.argumentList)

    before(
      node.trailingClosure?.leftBrace,
      tokens: .break(.same, newlines: .elective(ignoresDiscretionary: true)))

    arrangeFunctionCallArgumentList(
      arguments,
      leftDelimiter: node.leftParen,
      rightDelimiter: node.rightParen,
      forcesBreakBeforeRightDelimiter: breakBeforeRightParen)

    return .visitChildren
  }

  override func visitPost(_ node: FunctionCallExprSyntax) {
    clearContextualBreakState(node)
  }

  override func visit(_ node: MultipleTrailingClosureElementSyntax)
    -> SyntaxVisitorContinueKind
  {
    before(node.label, tokens: .space)
    after(node.colon, tokens: .space)
    return .visitChildren
  }

  /// Arrange the given argument list (or equivalently, tuple expression list) as a list of function
  /// arguments.
  ///
  /// - Parameters:
  ///   - arguments: The argument list/tuple expression list to arrange.
  ///   - leftDelimiter: The left parenthesis or bracket surrounding the arguments, if any.
  ///   - rightDelimiter: The right parenthesis or bracket surrounding the arguments, if any.
  ///   - forcesBreakBeforeRightDelimiter: True if a line break should be forced before the right
  ///     right delimiter if a line break occurred after the left delimiter, or false if the right
  ///     delimiter is allowed to hang on the same line as the final argument.
  private func arrangeFunctionCallArgumentList(
    _ arguments: TupleExprElementListSyntax,
    leftDelimiter: TokenSyntax?,
    rightDelimiter: TokenSyntax?,
    forcesBreakBeforeRightDelimiter: Bool
  ) {
    if !arguments.isEmpty {
      var afterLeftDelimiter: [Token] = [.break(.open, size: 0)]
      var beforeRightDelimiter: [Token] = [
        .break(.close(mustBreak: forcesBreakBeforeRightDelimiter), size: 0),
      ]

      if shouldGroupAroundArgumentList(arguments) {
        afterLeftDelimiter.append(.open(argumentListConsistency()))
        beforeRightDelimiter.append(.close)
      }

      after(leftDelimiter, tokens: afterLeftDelimiter)
      before(rightDelimiter, tokens: beforeRightDelimiter)
    }

    let shouldGroupAroundArgument = !isCompactSingleFunctionCallArgument(arguments)
    for argument in arguments {
      if let trailingComma = argument.trailingComma {
        closingDelimiterTokens.insert(trailingComma)
      }
      arrangeAsFunctionCallArgument(argument, shouldGroup: shouldGroupAroundArgument)
    }
  }

  /// Arranges the given tuple expression element as a function call argument.
  ///
  /// - Parameters:
  ///   - node: The tuple expression element.
  ///   - shouldGroup: If true, group around the argument to prefer keeping it together if possible.
  private func arrangeAsFunctionCallArgument(
    _ node: TupleExprElementSyntax,
    shouldGroup: Bool
  ) {
    if shouldGroup {
      before(node.firstToken, tokens: .open)
    }

    var additionalEndTokens = [Token]()
    if let colon = node.colon {
      // If we have an open delimiter following the colon, use a space instead of a continuation
      // break so that we don't awkwardly shift the delimiter down and indent it further if it
      // wraps.
      var tokensAfterColon: [Token] = [
        startsWithOpenDelimiter(Syntax(node.expression)) ? .space : .break
      ]

      if leftmostMultilineStringLiteral(of: node.expression) != nil {
        tokensAfterColon.append(.break(.open(kind: .block), size: 0))
        additionalEndTokens = [.break(.close(mustBreak: false), size: 0)]
      }

      after(colon, tokens: tokensAfterColon)
    }

    if let trailingComma = node.trailingComma {
      before(trailingComma, tokens: additionalEndTokens)
      var afterTrailingComma: [Token] = [.break(.same)]
      if shouldGroup {
        afterTrailingComma.insert(.close, at: 0)
      }
      after(trailingComma, tokens: afterTrailingComma)
    } else if shouldGroup {
      after(node.lastToken, tokens: additionalEndTokens + [.close])
    }
  }

  override func visit(_ node: ClosureExprSyntax) -> SyntaxVisitorContinueKind {
    let newlineBehavior: NewlineBehavior
    if forcedBreakingClosures.remove(node.id) != nil {
      newlineBehavior = .soft
    } else {
      newlineBehavior = .elective
    }

    if let signature = node.signature {
      after(node.leftBrace, tokens: .break(.open))
      if node.statements.count > 0 {
        after(signature.inTok, tokens: .break(.same, newlines: newlineBehavior))
      } else {
        after(signature.inTok, tokens: .break(.same, size: 0, newlines: newlineBehavior))
      }
      before(node.rightBrace, tokens: .break(.close))
    } else {
      // Closures without signatures can have their contents laid out identically to any other
      // braced structure. The leading reset is skipped because the layout depends on whether it is
      // a trailing closure of a function call (in which case that function call supplies the reset)
      // or part of some other expression (where we want that expression's same/continue behavior to
      // apply).
      arrangeBracesAndContents(
        of: node,
        contentsKeyPath: \.statements,
        shouldResetBeforeLeftBrace: false,
        openBraceNewlineBehavior: newlineBehavior)
    }
    return .visitChildren
  }

  override func visit(_ node: ClosureParamSyntax) -> SyntaxVisitorContinueKind {
    after(node.trailingComma, tokens: .break(.same))
    return .visitChildren
  }

  override func visit(_ node: ClosureSignatureSyntax) -> SyntaxVisitorContinueKind {
    before(node.firstToken, tokens: .open)

    arrangeAttributeList(
      node.attributes, suppressFinalBreak: node.input == nil && node.capture == nil)

    if let input = node.input {
      // We unconditionally put a break before the `in` keyword below, so we should only put a break
      // after the capture list's right bracket if there are arguments following it or we'll end up
      // with an extra space if the line doesn't wrap.
      after(node.capture?.rightSquare, tokens: .break(.same))

      // When it's parenthesized, the input is a `ParameterClauseSyntax`. Otherwise, it's a
      // `ClosureParamListSyntax`. The parenthesized version is wrapped in open/close breaks so that
      // the parens create an extra level of indentation.
      if let parameterClause = input.as(ParameterClauseSyntax.self) {
        // Whether we should prioritize keeping ") throws -> <return_type>" together. We can only do
        // this if the closure has arguments.
        let keepOutputTogether =
          !parameterClause.parameterList.isEmpty && config.prioritizeKeepingFunctionOutputTogether

        // Keep the output together by grouping from the right paren to the end of the output.
        if keepOutputTogether {
          // Due to visitation order, the matching .open break is added in ParameterClauseSyntax.
          // Since the output clause is optional but the in-token is required, placing the .close
          // before `inTok` ensures the close gets into the token stream.
          before(node.inTok, tokens: .close)
        } else  {
          // Group outside of the parens, so that the argument list together, preferring to break
          // between the argument list and the output.
          before(input.firstToken, tokens: .open)
          after(input.lastToken, tokens: .close)
        }

        arrangeParameterClause(parameterClause, forcesBreakBeforeRightParen: true)
      } else {
        // Group around the arguments, but don't use open/close breaks because there are no parens
        // to create a new scope.
        before(input.firstToken, tokens: .open(argumentListConsistency()))
        after(input.lastToken, tokens: .close)
      }
    }

    before(node.effectSpecifiers?.asyncSpecifier, tokens: .break)
    before(node.effectSpecifiers?.throwsSpecifier, tokens: .break)
    if let asyncKeyword = node.effectSpecifiers?.asyncSpecifier, let throwsTok = node.effectSpecifiers?.throwsSpecifier {
      before(asyncKeyword, tokens: .open)
      after(throwsTok, tokens: .close)
    }

    before(node.output?.arrow, tokens: .break)
    after(node.lastToken, tokens: .close)
    before(node.inTok, tokens: .break(.same))
    return .visitChildren
  }

  override func visit(_ node: ClosureCaptureSignatureSyntax) -> SyntaxVisitorContinueKind {
    after(node.leftSquare, tokens: .break(.open, size: 0), .open)
    before(node.rightSquare, tokens: .break(.close, size: 0), .close)
    return .visitChildren
  }

  override func visit(_ node: ClosureCaptureItemSyntax) -> SyntaxVisitorContinueKind {
    before(node.firstToken, tokens: .open)
    after(node.specifier?.lastToken, tokens: .break)
    before(node.assignToken, tokens: .break)
    after(node.assignToken, tokens: .break)
    if let trailingComma = node.trailingComma {
      before(trailingComma, tokens: .close)
      after(trailingComma, tokens: .break(.same))
    } else {
      after(node.lastToken, tokens: .close)
    }
    return .visitChildren
  }

  override func visit(_ node: SubscriptExprSyntax) -> SyntaxVisitorContinueKind {
    preVisitInsertingContextualBreaks(node)

    if let calledMemberAccessExpr = node.calledExpression.as(MemberAccessExprSyntax.self) {
      if let base = calledMemberAccessExpr.base, base.is(IdentifierExprSyntax.self) {
        before(base.firstToken, tokens: .open)
        after(calledMemberAccessExpr.name.lastToken, tokens: .close)
      }
    }

    let arguments = node.argumentList

    // If there is a trailing closure, force the right bracket down to the next line so it stays
    // with the open curly brace.
    let breakBeforeRightBracket =
      node.trailingClosure != nil
      || mustBreakBeforeClosingDelimiter(of: node, argumentListPath: \.argumentList)

    before(
      node.trailingClosure?.leftBrace,
      tokens: .break(.same, newlines: .elective(ignoresDiscretionary: true)))

    arrangeFunctionCallArgumentList(
      arguments,
      leftDelimiter: node.leftBracket,
      rightDelimiter: node.rightBracket,
      forcesBreakBeforeRightDelimiter: breakBeforeRightBracket)

    return .visitChildren
  }

  override func visitPost(_ node: SubscriptExprSyntax) {
    clearContextualBreakState(node)
  }

  override func visit(_ node: ExpressionSegmentSyntax) -> SyntaxVisitorContinueKind {
    // TODO: For now, just use the raw text of the node and don't try to format it deeper. In the
    // future, we should find a way to format the expression but without wrapping so that at least
    // internal whitespace is fixed.
    appendToken(.syntax(node.description))
    // Visiting children is not needed here.
    return .skipChildren
  }

  override func visit(_ node: MacroExpansionDeclSyntax) -> SyntaxVisitorContinueKind {
    arrangeFunctionCallArgumentList(
      node.argumentList,
      leftDelimiter: node.leftParen,
      rightDelimiter: node.rightParen,
      forcesBreakBeforeRightDelimiter: false)
    return .visitChildren
  }

  override func visit(_ node: MacroExpansionExprSyntax) -> SyntaxVisitorContinueKind {
    arrangeFunctionCallArgumentList(
      node.argumentList,
      leftDelimiter: node.leftParen,
      rightDelimiter: node.rightParen,
      forcesBreakBeforeRightDelimiter: false)
    return .visitChildren
  }

  override func visit(_ node: ParameterClauseSyntax) -> SyntaxVisitorContinueKind {
    // Prioritize keeping ") throws -> <return_type>" together. We can only do this if the function
    // has arguments.
    if !node.parameterList.isEmpty && config.prioritizeKeepingFunctionOutputTogether {
      // Due to visitation order, this .open corresponds to a .close added in FunctionDeclSyntax
      // or SubscriptDeclSyntax.
      before(node.rightParen, tokens: .open)
    }

    return .visitChildren
  }

  override func visit(_ node: FunctionParameterSyntax) -> SyntaxVisitorContinueKind {
    before(node.firstToken, tokens: .open)
    arrangeAttributeList(node.attributes)
    before(
      node.secondName,
      tokens: .break(.continue, newlines: .elective(ignoresDiscretionary: true)))
    after(node.colon, tokens: .break)

    if let trailingComma = node.trailingComma {
      after(trailingComma, tokens: .close, .break(.same))
    } else {
      after(node.lastToken, tokens: .close)
    }
    return .visitChildren
  }

  override func visit(_ node: ReturnClauseSyntax) -> SyntaxVisitorContinueKind {
    if node.parent?.is(FunctionTypeSyntax.self) ?? false {
      // `FunctionTypeSyntax` used to not use `ReturnClauseSyntax` and had 
      // slightly different formatting behavior than the normal 
      // `ReturnClauseSyntax`. To maintain the previous formatting behavior, 
      // add a special case.
      before(node.arrow, tokens: .break)
      before(node.returnType.firstToken, tokens: .break)
    } else {
      after(node.arrow, tokens: .space)
    }

    // Member type identifier is used when the return type is a member of another type. Add a group
    // here so that the base, dot, and member type are kept together when they fit.
    if node.returnType.is(MemberTypeIdentifierSyntax.self) {
      before(node.returnType.firstToken, tokens: .open)
      after(node.returnType.lastToken, tokens: .close)
    }
    return .visitChildren
  }

  override func visit(_ node: IfConfigDeclSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: IfConfigClauseSyntax) -> SyntaxVisitorContinueKind {
    switch node.poundKeyword.tokenKind {
    case .poundIfKeyword, .poundElseifKeyword:
      after(node.poundKeyword, tokens: .space)
    case .poundElseKeyword:
      break
    default:
      preconditionFailure()
    }

    let breakKindOpen: BreakKind
    let breakKindClose: BreakKind
    if config.indentConditionalCompilationBlocks {
      breakKindOpen = .open
      breakKindClose = .close
    } else {
      breakKindOpen = .same
      breakKindClose = .same
    }

    let tokenToOpenWith = node.condition?.lastToken ?? node.poundKeyword
    after(tokenToOpenWith, tokens: .break(breakKindOpen), .open)

    // Unlike other code blocks, where we may want a single statement to be laid out on the same
    // line as a parent construct, the content of an `#if` block must always be on its own line;
    // the newline token inserted at the end enforces this.
    if let lastElemTok = node.elements?.lastToken {
      after(lastElemTok, tokens: .break(breakKindClose, newlines: .soft), .close)
    } else {
      before(tokenToOpenWith.nextToken(viewMode: .all), tokens: .break(breakKindClose, newlines: .soft), .close)
    }

    if isNestedInPostfixIfConfig(node: Syntax(node)) {
      before(
        node.firstToken,
        tokens: [
          .printerControl(kind: .enableBreaking),
          .break(.reset),
        ]
      )
    } else if let condition = node.condition {
      before(condition.firstToken, tokens: .printerControl(kind: .disableBreaking))
      after(
        condition.lastToken,
        tokens: .printerControl(kind: .enableBreaking), .break(.reset, size: 0))
    }

    return .visitChildren
  }

  override func visit(_ node: MemberDeclBlockSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: MemberDeclListSyntax) -> SyntaxVisitorContinueKind {
    // Skip ignored items, because the tokens after `item.lastToken` would be ignored and leave
    // unclosed open tokens.
    for item in node where !shouldFormatterIgnore(node: Syntax(item)) {
      before(item.firstToken, tokens: .open)
      let newlines: NewlineBehavior =
        item != node.last && shouldInsertNewline(basedOn: item.semicolon) ? .soft : .elective
      let resetSize = item.semicolon != nil ? 1 : 0
      after(item.lastToken, tokens: .close, .break(.reset, size: resetSize, newlines: newlines))
    }
    return .visitChildren
  }

  override func visit(_ node: MemberDeclListItemSyntax) -> SyntaxVisitorContinueKind {
    if shouldFormatterIgnore(node: Syntax(node)) {
      appendFormatterIgnored(node: Syntax(node))
      return .skipChildren
    }
    return .visitChildren
  }

  override func visit(_ node: SourceFileSyntax) -> SyntaxVisitorContinueKind {
    if shouldFormatterIgnore(file: node) {
      appendFormatterIgnored(node: Syntax(node))
      return .skipChildren
    }
    after(node.eofToken, tokens: .break(.same, newlines: .soft))
    return .visitChildren
  }

  override func visit(_ node: EnumCaseDeclSyntax) -> SyntaxVisitorContinueKind {
    before(node.firstToken, tokens: .open)

    arrangeAttributeList(node.attributes)

    after(node.caseKeyword, tokens: .break)
    after(node.lastToken, tokens: .close)
    return .visitChildren
  }

  override func visit(_ node: OperatorDeclSyntax) -> SyntaxVisitorContinueKind {
    after(node.operatorKeyword, tokens: .break)
    return .visitChildren
  }

  override func visit(_ node: OperatorPrecedenceAndTypesSyntax) -> SyntaxVisitorContinueKind {
    before(node.colon, tokens: .space)
    after(node.colon, tokens: .break(.open), .open)
    after(node.designatedTypes.lastToken ?? node.lastToken, tokens: .break(.close, size: 0), .close)
    return .visitChildren
  }

  override func visit(_ node: DesignatedTypeElementSyntax) -> SyntaxVisitorContinueKind {
    after(node.leadingComma, tokens: .break(.same))
    return .visitChildren
  }

  override func visit(_ node: EnumCaseElementSyntax) -> SyntaxVisitorContinueKind {
    after(node.trailingComma, tokens: .break)

    if let associatedValue = node.associatedValue {
      arrangeParameterClause(associatedValue, forcesBreakBeforeRightParen: false)
    }

    return .visitChildren
  }

  override func visit(_ node: ObjCSelectorSyntax) -> SyntaxVisitorContinueKind {
    insertTokens(.break(.same, size: 0), betweenElementsOf: node)
    return .visitChildren
  }

  override func visit(_ node: PrecedenceGroupDeclSyntax) -> SyntaxVisitorContinueKind {
    after(node.precedencegroupKeyword, tokens: .break)
    after(node.identifier, tokens: .break(.reset))
    after(node.leftBrace, tokens: .break(.open, newlines: .soft))
    before(node.rightBrace, tokens: .break(.close))
    return .visitChildren
  }

  override func visit(_ node: PrecedenceGroupRelationSyntax) -> SyntaxVisitorContinueKind {
    after(node.colon, tokens: .break(.open))
    after(node.lastToken, tokens: .break(.close, newlines: .soft))
    return .visitChildren
  }

  override func visit(_ node: PrecedenceGroupAssignmentSyntax) -> SyntaxVisitorContinueKind {
    after(node.colon, tokens: .break(.open))
    after(node.lastToken, tokens: .break(.close, newlines: .soft))
    return .visitChildren
  }

  override func visit(_ node: PrecedenceGroupNameElementSyntax) -> SyntaxVisitorContinueKind {
    after(node.trailingComma, tokens: .break(.same))
    return .visitChildren
  }

  override func visit(_ node: PrecedenceGroupAssociativitySyntax) -> SyntaxVisitorContinueKind {
    after(node.colon, tokens: .break(.open))
    after(node.lastToken, tokens: .break(.close, newlines: .soft))
    return .visitChildren
  }

  override func visit(_ node: CodeBlockSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: CodeBlockItemListSyntax) -> SyntaxVisitorContinueKind {
    // Skip ignored items, because the tokens after `item.lastToken` would be ignored and leave
    // unclosed open tokens.
    for item in node where !shouldFormatterIgnore(node: Syntax(item)) {
      before(item.firstToken, tokens: .open)
      let newlines: NewlineBehavior =
        item != node.last && shouldInsertNewline(basedOn: item.semicolon) ? .soft : .elective
      let resetSize = item.semicolon != nil ? 1 : 0
      after(item.lastToken, tokens: .close, .break(.reset, size: resetSize, newlines: newlines))
    }
    return .visitChildren
  }

  override func visit(_ node: CodeBlockItemSyntax) -> SyntaxVisitorContinueKind {
    if shouldFormatterIgnore(node: Syntax(node)) {
      appendFormatterIgnored(node: Syntax(node))
      return .skipChildren
    }

    // This group applies to a top-level if-stmt so that all of the bodies will have the same
    // breaking behavior.
    if let exprStmt = node.item.as(ExpressionStmtSyntax.self),
       let ifStmt = exprStmt.expression.as(IfExprSyntax.self) {
      before(ifStmt.conditions.firstToken, tokens: .open(.consistent))
      after(ifStmt.lastToken, tokens: .close)
    }
    return .visitChildren
  }

  override func visit(_ node: GenericParameterClauseSyntax) -> SyntaxVisitorContinueKind {
    after(node.leftAngleBracket, tokens: .break(.open, size: 0), .open(argumentListConsistency()))
    before(node.rightAngleBracket, tokens: .break(.close, size: 0), .close)
    return .visitChildren
  }

  override func visit(_ node: PrimaryAssociatedTypeClauseSyntax) -> SyntaxVisitorContinueKind {
    after(node.leftAngleBracket, tokens: .break(.open, size: 0), .open(argumentListConsistency()))
    before(node.rightAngleBracket, tokens: .break(.close, size: 0), .close)
    return .visitChildren
  }

  override func visit(_ node: ArrayTypeSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: TupleTypeSyntax) -> SyntaxVisitorContinueKind {
    after(node.leftParen, tokens: .break(.open, size: 0), .open)
    before(node.rightParen, tokens: .break(.close, size: 0), .close)
    return .visitChildren
  }

  override func visit(_ node: TupleTypeElementSyntax) -> SyntaxVisitorContinueKind {
    before(node.firstToken, tokens: .open)
    before(
      node.secondName,
      tokens: .break(.continue, newlines: .elective(ignoresDiscretionary: true)))
    after(node.colon, tokens: .break)

    if let trailingComma = node.trailingComma {
      after(trailingComma, tokens: .close, .break(.same))
    } else {
      after(node.lastToken, tokens: .close)
    }
    return .visitChildren
  }

  override func visit(_ node: FunctionTypeSyntax) -> SyntaxVisitorContinueKind {
    after(node.leftParen, tokens: .break(.open, size: 0), .open)
    before(node.rightParen, tokens: .break(.close, size: 0), .close)
    before(node.effectSpecifiers?.asyncSpecifier, tokens: .break)
    before(node.effectSpecifiers?.throwsSpecifier, tokens: .break)
    return .visitChildren
  }

  override func visit(_ node: GenericArgumentClauseSyntax) -> SyntaxVisitorContinueKind {
    after(node.leftAngleBracket, tokens: .break(.open, size: 0), .open)
    before(node.rightAngleBracket, tokens: .break(.close, size: 0), .close)
    return .visitChildren
  }

  override func visit(_ node: TuplePatternSyntax) -> SyntaxVisitorContinueKind {
    after(node.leftParen, tokens: .break(.open, size: 0), .open)
    before(node.rightParen, tokens: .break(.close, size: 0), .close)
    return .visitChildren
  }

  override func visit(_ node: TryExprSyntax) -> SyntaxVisitorContinueKind {
    before(
      node.expression.firstToken,
      tokens: .break(.continue, newlines: .elective(ignoresDiscretionary: true)))

    // Check for an anchor token inside of the expression to group with the try keyword.
    if let anchorToken = findTryAwaitExprConnectingToken(inExpr: node.expression) {
      before(node.tryKeyword, tokens: .open)
      after(anchorToken, tokens: .close)
    }

    return .visitChildren
  }

  override func visit(_ node: AwaitExprSyntax) -> SyntaxVisitorContinueKind {
    before(
      node.expression.firstToken,
      tokens: .break(.continue, newlines: .elective(ignoresDiscretionary: true)))

    // Check for an anchor token inside of the expression to group with the await keyword.
    if !(node.parent?.is(TryExprSyntax.self) ?? false),
      let anchorToken = findTryAwaitExprConnectingToken(inExpr: node.expression)
    {
      before(node.awaitKeyword, tokens: .open)
      after(anchorToken, tokens: .close)
    }

    return .visitChildren
  }

  /// Searches the AST from `expr` to find a token that should be grouped with an enclosing
  /// try-expr or await-expr. Returns that token, or nil when no such token is found.
  ///
  /// - Parameter expr: An expression that is wrapped by a try-expr or await-expr.
  /// - Returns: A token that should be grouped with the try-expr or await-expr, or nil.
  func findTryAwaitExprConnectingToken(inExpr expr: ExprSyntax) -> TokenSyntax? {
    if let awaitExpr = expr.as(AwaitExprSyntax.self) {
      // If we were called from the `try` of a `try await <expr>`, drill into the child expression.
      return findTryAwaitExprConnectingToken(inExpr: awaitExpr.expression)
    }
    if let callingExpr = expr.asProtocol(CallingExprSyntaxProtocol.self) {
      return findTryAwaitExprConnectingToken(inExpr: callingExpr.calledExpression)
    }
    if let memberAccessExpr = expr.as(MemberAccessExprSyntax.self), let base = memberAccessExpr.base
    {
      // When there's a simple base (i.e. identifier), group the entire `try/await <base>.<name>`
      // sequence. This check has to happen here so that the `MemberAccessExprSyntax.name` is
      // available.
      if base.is(IdentifierExprSyntax.self) {
        return memberAccessExpr.name.lastToken
      }
      return findTryAwaitExprConnectingToken(inExpr: base)
    }
    if expr.is(IdentifierExprSyntax.self) {
      return expr.lastToken
    }
    return nil
  }

  override func visit(_ node: TypeExprSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: AttributeSyntax) -> SyntaxVisitorContinueKind {
    before(node.firstToken, tokens: .open)
    switch node.argument {
    case .argumentList(let argumentList)?:
      if let leftParen = node.leftParen, let rightParen = node.rightParen {
        arrangeFunctionCallArgumentList(
          argumentList,
          leftDelimiter: leftParen,
          rightDelimiter: rightParen,
          forcesBreakBeforeRightDelimiter: false)
      }
    case .some:
      // Wrap the attribute's arguments in their own group, so arguments stay together with a higher
      // affinity than the overall attribute (e.g. allows a break after the opening "(" and then
      // having the entire argument list on 1 line). Necessary spaces and breaks are added inside of
      // the argument, using type specific visitor methods.
      after(node.leftParen, tokens: .break(.open, size: 0), .open(argumentListConsistency()))
      before(node.rightParen, tokens: .break(.close, size: 0), .close)
    case nil:
      break
    }
    after(node.lastToken, tokens: .close)
    return .visitChildren
  }

  override func visit(_ node: AvailabilitySpecListSyntax) -> SyntaxVisitorContinueKind {
    insertTokens(.break(.same, size: 1), betweenElementsOf: node)
    return .visitChildren
  }

  override func visit(_ node: AvailabilityLabeledArgumentSyntax) -> SyntaxVisitorContinueKind {
    before(node.label, tokens: .open)
    after(node.colon, tokens: .break(.continue, newlines: .elective(ignoresDiscretionary: true)))
    after(node.value.lastToken, tokens: .close)
    return .visitChildren
  }

  override func visit(_ node: AvailabilityVersionRestrictionSyntax) -> SyntaxVisitorContinueKind {
    before(node.firstToken, tokens: .open)
    after(node.platform, tokens: .break(.continue, size: 1))
    after(node.lastToken, tokens: .close)
    return .visitChildren
  }

  override func visit(_ node: ConditionElementSyntax) -> SyntaxVisitorContinueKind {
    before(node.firstToken, tokens: .open)
    if let comma = node.trailingComma {
      after(comma, tokens: .close, .break(.same))
      closingDelimiterTokens.insert(comma)
    } else {
      after(node.lastToken, tokens: .close)
    }
    return .visitChildren
  }

  override func visit(_ node: InOutExprSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: ImportDeclSyntax) -> SyntaxVisitorContinueKind {
    after(node.attributes?.lastToken, tokens: .space)
    after(node.importTok, tokens: .space)
    after(node.importKind, tokens: .space)
    return .visitChildren
  }

  override func visit(_ node: KeyPathExprSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: TernaryExprSyntax) -> SyntaxVisitorContinueKind {
    // The order of the .open/.close tokens here is intentional. They are normally paired with the
    // corresponding breaks, but in this case, we want to prioritize keeping the entire `? a : b`
    // part together if some part of the ternary wraps, instead of keeping `c ? a` together and
    // wrapping after that.
    before(node.questionMark, tokens: .break(.open(kind: .continuation)), .open)
    after(node.questionMark, tokens: .space)
    before(
      node.colonMark,
      tokens: .break(.close(mustBreak: false), size: 0), .break(.open(kind: .continuation)), .open)
    after(node.colonMark, tokens: .space)

    // When the ternary is wrapped in parens, absorb the closing paren into the ternary's group so
    // that it is glued to the last token of the ternary.
    let closeScopeToken: TokenSyntax?
    if let parenExpr = outermostEnclosingNode(from: Syntax(node.secondChoice)) {
      closeScopeToken = parenExpr.lastToken
    } else {
      closeScopeToken = node.secondChoice.lastToken
    }
    after(closeScopeToken, tokens: .break(.close(mustBreak: false), size: 0), .close, .close)
    return .visitChildren
  }

  override func visit(_ node: WhereClauseSyntax) -> SyntaxVisitorContinueKind {
    // We need to special case `where`-clauses associated with `catch` blocks when
    // `lineBreakBeforeControlFlowKeywords == false`, because that's the one situation where we
    // want the `where` keyword to be treated as a continuation; that way, we get this:
    //
    //     } catch LongExceptionName
    //       where longCondition
    //     {
    //       ...
    //     }
    //
    // instead of this:
    //
    //     } catch LongExceptionName
    //     where longCondition {
    //       ...
    //     }
    //
    let wherePrecedingBreak: Token
    if !config.lineBreakBeforeControlFlowKeywords,
      let parent = node.parent, parent.is(CatchItemSyntax.self)
    {
      wherePrecedingBreak = .break(.continue)
    } else {
      wherePrecedingBreak = .break(.same)
    }
    before(node.whereKeyword, tokens: wherePrecedingBreak, .open)
    after(node.whereKeyword, tokens: .break)
    after(node.lastToken, tokens: .close)
    return .visitChildren
  }

  override func visit(_ node: DeclModifierSyntax) -> SyntaxVisitorContinueKind {
    // Due to the way we currently use spaces after let/var keywords in variable bindings, we need
    // this special exception for `async let` statements to avoid breaking prematurely between the
    // `async` and `let` keywords.
    let breakOrSpace: Token
    if node.name.tokenKind == .keyword(.async) {
      breakOrSpace = .space
    } else {
      breakOrSpace = .break
    }
    after(node.lastToken, tokens: breakOrSpace)
    return .visitChildren
  }

  override func visit(_ node: FunctionSignatureSyntax) -> SyntaxVisitorContinueKind {
    before(node.effectSpecifiers?.asyncSpecifier, tokens: .break)
    before(node.effectSpecifiers?.throwsSpecifier, tokens: .break)
    if let asyncOrReasyncKeyword = node.effectSpecifiers?.asyncSpecifier,
      let throwsOrRethrowsKeyword = node.effectSpecifiers?.throwsSpecifier
    {
      before(asyncOrReasyncKeyword, tokens: .open)
      after(throwsOrRethrowsKeyword, tokens: .close)
    }
    before(node.output?.firstToken, tokens: .break)
    return .visitChildren
  }

  override func visit(_ node: MetatypeTypeSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: OptionalTypeSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: InfixOperatorExprSyntax) -> SyntaxVisitorContinueKind {
    // FIXME: This is a workaround/hack for https://github.com/apple/swift-syntax/issues/928. For
    // keypaths like `\.?.foo`, they get represented (after folding) as an infix operator expression
    // with an empty keypath, followed by the "binary operator" `.?.`, followed by other
    // expressions. We can detect this and treat the whole thing as a verbatim node, which mimics
    // what we do today for keypaths (i.e., nothing).
    if let keyPathExpr = node.leftOperand.as(KeyPathExprSyntax.self),
      keyPathExpr.components.isEmpty
    {
      // If there were spaces in the trailing trivia of the previous token, they would have been
      // ignored (since this expression would be expected to insert its own preceding breaks).
      // Preserve that whitespace verbatim for now.
      if let previousToken = node.firstToken?.previousToken {
        appendTrailingTrivia(previousToken, forced: true)
      }
      verbatimToken(Syntax(node), indentingBehavior: .none)
      return .skipChildren
    }

    let binOp = node.operatorOperand
    let rhs = node.rightOperand
    maybeGroupAroundSubexpression(rhs, combiningOperator: binOp)

    let wrapsBeforeOperator = !isAssigningOperator(binOp)

    if shouldRequireWhitespace(around: binOp) {
      if isAssigningOperator(binOp) {
        var beforeTokens: [Token]

        // If the rhs starts with a parenthesized expression, stack indentation around it.
        // Otherwise, use regular continuation breaks.
        if let (unindentingNode, _, breakKind) = stackedIndentationBehavior(after: binOp, rhs: rhs)
        {
          beforeTokens = [.break(.open(kind: breakKind))]
          after(unindentingNode.lastToken, tokens: [.break(.close(mustBreak: false), size: 0)])
        } else {
          beforeTokens = [.break(.continue)]
        }

        // When the RHS is a simple expression, even if is requires multiple lines, we don't add a
        // group so that as much of the expression as possible can stay on the same line as the
        // operator token.
        if isCompoundExpression(rhs) {
          beforeTokens.append(.open)
          after(rhs.lastToken, tokens: .close)
        }

        after(binOp.lastToken, tokens: beforeTokens)
      } else if let (unindentingNode, shouldReset, breakKind) =
        stackedIndentationBehavior(after: binOp, rhs: rhs)
      {
        // For parenthesized expressions and for unparenthesized usages of `&&` and `||`, we don't
        // want to treat all continue breaks the same. If we did, then all operators would line up
        // at the same alignment regardless of whether they were, for example, `&&` or something
        // between a pair of `&&`. To make long expressions/conditionals format more cleanly, we
        // use open-continuation/close pairs around such operators and their right-hand sides so
        // that the continuation breaks inside those scopes "stack", instead of receiving the
        // usual single-level "continuation line or not" behavior.
        let openBreakTokens: [Token] = [.break(.open(kind: breakKind)), .open]
        if wrapsBeforeOperator {
          before(binOp.firstToken, tokens: openBreakTokens)
        } else {
          after(binOp.lastToken, tokens: openBreakTokens)
        }

        let closeBreakTokens: [Token] =
          (shouldReset ? [.break(.reset, size: 0)] : [])
          + [.break(.close(mustBreak: false), size: 0), .close]
        after(unindentingNode.lastToken, tokens: closeBreakTokens)
      } else {
        if wrapsBeforeOperator {
          before(binOp.firstToken, tokens: .break(.continue))
        } else {
          after(binOp.lastToken, tokens: .break(.continue))
        }
      }

      if wrapsBeforeOperator {
        after(binOp.lastToken, tokens: .space)
      } else {
        before(binOp.firstToken, tokens: .space)
      }
    }

    return .visitChildren
  }

  override func visit(_ node: PrefixOperatorExprSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: PostfixUnaryExprSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: AsExprSyntax) -> SyntaxVisitorContinueKind {
    before(node.asTok, tokens: .break(.continue), .open)
    before(node.typeName.firstToken, tokens: .space)
    after(node.lastToken, tokens: .close)
    return .visitChildren
  }

  override func visit(_ node: IsExprSyntax) -> SyntaxVisitorContinueKind {
    before(node.isTok, tokens: .break(.continue), .open)
    before(node.typeName.firstToken, tokens: .space)
    after(node.lastToken, tokens: .close)
    return .visitChildren
  }

  override func visit(_ node: SequenceExprSyntax) -> SyntaxVisitorContinueKind {
    preconditionFailure("SequenceExpr should have already been folded.")
  }

  override func visit(_ node: AssignmentExprSyntax) -> SyntaxVisitorContinueKind {
    // Breaks and spaces are inserted at the `InfixOperatorExpr` level.
    return .visitChildren
  }

  override func visit(_ node: BinaryOperatorExprSyntax) -> SyntaxVisitorContinueKind {
    // Breaks and spaces are inserted at the `InfixOperatorExpr` level.
    return .visitChildren
  }

  override func visit(_ node: ArrowExprSyntax) -> SyntaxVisitorContinueKind {
    // The break before the `throws` keyword is inserted at the `InfixOperatorExpr` level so that it
    // is placed in the correct relative position to the group surrounding the "operator".
    after(node.effectSpecifiers?.throwsSpecifier, tokens: .break)
    return .visitChildren
  }

  override func visit(_ node: SuperRefExprSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: VariableDeclSyntax) -> SyntaxVisitorContinueKind {
    arrangeAttributeList(node.attributes)

    if node.bindings.count == 1 {
      // If there is only a single binding, don't allow a break between the `let/var` keyword and
      // the identifier; there are better places to break later on.
      after(node.bindingKeyword, tokens: .space)
    } else {
      // If there is more than one binding, we permit an open-break after `let/var` so that each of
      // the comma-delimited items will potentially receive indentation. We also add a group around
      // the individual bindings to bind them together better. (This is done here, not in
      // `visit(_: PatternBindingSyntax)`, because we only want that behavior when there are
      // multiple bindings.)
      after(node.bindingKeyword, tokens: .break(.open))

      for binding in node.bindings {
        before(binding.firstToken, tokens: .open)
        after(binding.trailingComma, tokens: .break(.same))
        after(binding.lastToken, tokens: .close)
      }

      after(node.lastToken, tokens: .break(.close, size: 0))
    }

    return .visitChildren
  }

  override func visit(_ node: PatternBindingSyntax) -> SyntaxVisitorContinueKind {
    // If the type annotation and/or the initializer clause need to wrap, we want those
    // continuations to stack to improve readability. So, we need to keep track of how many open
    // breaks we create (so we can close them at the end of the binding) and also keep track of the
    // right-most token that will anchor the close breaks.
    var closesNeeded: Int = 0
    var closeAfterToken: TokenSyntax? = nil

    if let typeAnnotation = node.typeAnnotation, !typeAnnotation.type.is(MissingTypeSyntax.self) {
      after(
        typeAnnotation.colon,
        tokens: .break(.open(kind: .continuation), newlines: .elective(ignoresDiscretionary: true)))
      closesNeeded += 1
      closeAfterToken = typeAnnotation.lastToken
    }
    if let initializer = node.initializer {
      let expr = initializer.value

      if let (unindentingNode, _, breakKind) = stackedIndentationBehavior(rhs: expr) {
        after(initializer.equal, tokens: .break(.open(kind: breakKind)))
        after(unindentingNode.lastToken, tokens: .break(.close(mustBreak: false), size: 0))
      } else {
        after(initializer.equal, tokens: .break(.continue))
      }
      closeAfterToken = initializer.lastToken

      // When the RHS is a simple expression, even if is requires multiple lines, we don't add a
      // group so that as much of the expression as possible can stay on the same line as the
      // operator token.
      if isCompoundExpression(expr) {
        before(expr.firstToken, tokens: .open)
        after(expr.lastToken, tokens: .close)
      }
    }

    if let accessorOrCodeBlock = node.accessor {
      switch accessorOrCodeBlock {
      case .accessors(let accessorBlock):
        arrangeBracesAndContents(of: accessorBlock)
      case .getter(let codeBlock):
        arrangeBracesAndContents(of: codeBlock, contentsKeyPath: \.statements)
      }
    } else if let trailingComma = node.trailingComma {
      // If this is one of multiple comma-delimited bindings, move any pending close breaks to
      // follow the comma so that it doesn't get separated from the tokens before it.
      closeAfterToken = trailingComma
      closingDelimiterTokens.insert(trailingComma)
    }

    if closeAfterToken != nil && closesNeeded > 0 {
      let closeTokens = [Token](repeatElement(.break(.close, size: 0), count: closesNeeded))
      after(closeAfterToken, tokens: closeTokens)
    }

    return .visitChildren
  }

  override func visit(_ node: InheritedTypeSyntax) -> SyntaxVisitorContinueKind {
    after(node.trailingComma, tokens: .break(.same))
    return .visitChildren
  }

  override func visit(_ node: IsTypePatternSyntax) -> SyntaxVisitorContinueKind {
    after(node.isKeyword, tokens: .space)
    return .visitChildren
  }

  override func visit(_ node: TypealiasDeclSyntax) -> SyntaxVisitorContinueKind {
    arrangeAttributeList(node.attributes)

    after(node.typealiasKeyword, tokens: .break)

    if let genericWhereClause = node.genericWhereClause {
      before(genericWhereClause.firstToken, tokens: .break(.same), .open)
      after(node.lastToken, tokens: .close)
    }
    return .visitChildren
  }

  override func visit(_ node: TypeInitializerClauseSyntax) -> SyntaxVisitorContinueKind {
    before(node.equal, tokens: .space)
    after(node.equal, tokens: .break)
    return .visitChildren
  }

  override func visit(_ node: AttributedTypeSyntax) -> SyntaxVisitorContinueKind {
    arrangeAttributeList(node.attributes)
    after(
      node.specifier,
      tokens: .break(.continue, newlines: .elective(ignoresDiscretionary: true)))
    return .visitChildren
  }

  override func visit(_ node: IdentifierExprSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: NilLiteralExprSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: SpecializeExprSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: TypeAnnotationSyntax) -> SyntaxVisitorContinueKind {
    before(node.type.firstToken, tokens: .open)
    after(node.type.lastToken, tokens: .close)
    return .visitChildren
  }

  override func visit(_ node: ConstrainedSugarTypeSyntax) -> SyntaxVisitorContinueKind {
    after(node.someOrAnySpecifier, tokens: .space)
    return .visitChildren
  }

  override func visit(_ node: CompositionTypeSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: FallthroughStmtSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: ForcedValueExprSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: GenericArgumentSyntax) -> SyntaxVisitorContinueKind {
    before(node.firstToken, tokens: .open)
    if let trailingComma = node.trailingComma {
      after(trailingComma, tokens: .close, .break(.same))
    } else {
      after(node.lastToken, tokens: .close)
    }
    return .visitChildren
  }

  override func visit(_ node: WildcardPatternSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: DeclNameArgumentSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: FloatLiteralExprSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: GenericParameterSyntax) -> SyntaxVisitorContinueKind {
    before(node.firstToken, tokens: .open)
    after(node.colon, tokens: .break)
    if let trailingComma = node.trailingComma {
      after(trailingComma, tokens: .close, .break(.same))
    } else {
      after(node.lastToken, tokens: .close)
    }
    return .visitChildren
  }

  override func visit(_ node: PrimaryAssociatedTypeSyntax) -> SyntaxVisitorContinueKind {
    before(node.firstToken, tokens: .open)
    if let trailingComma = node.trailingComma {
      after(trailingComma, tokens: .close, .break(.same))
    } else {
      after(node.lastToken, tokens: .close)
    }
    return .visitChildren
  }

  override func visit(_ node: ExpressionPatternSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: ValueBindingPatternSyntax) -> SyntaxVisitorContinueKind {
    after(node.bindingKeyword, tokens: .break)
    return .visitChildren
  }

  override func visit(_ node: IdentifierPatternSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: InitializerClauseSyntax) -> SyntaxVisitorContinueKind {
    before(node.equal, tokens: .space)

    // InitializerClauses that are children of a PatternBindingSyntax are already handled in the
    // latter node, to ensure that continuations stack appropriately.
    if node.parent == nil || !node.parent!.is(PatternBindingSyntax.self) {
      after(node.equal, tokens: .break)
    }
    return .visitChildren
  }

  override func visit(_ node: StringLiteralExprSyntax) -> SyntaxVisitorContinueKind {
    if node.openQuote.tokenKind == .multilineStringQuote {
      // Looks up the correct break kind based on prior context.
      let breakKind = pendingMultilineStringBreakKinds[node, default: .same]
      after(node.openQuote, tokens: .break(breakKind, size: 0, newlines: .hard(count: 1)))
      before(node.closeQuote, tokens: .break(breakKind, newlines: .hard(count: 1)))
    }
    return .visitChildren
  }

  override func visit(_ node: StringSegmentSyntax) -> SyntaxVisitorContinueKind {
    // Looks up the correct break kind based on prior context.
    func breakKind() -> BreakKind {
      if let stringLiteralSegments = node.parent?.as(StringLiteralSegmentsSyntax.self),
        let stringLiteralExpr = stringLiteralSegments.parent?.as(StringLiteralExprSyntax.self)
      {
        return pendingMultilineStringBreakKinds[stringLiteralExpr, default: .same]
      } else {
        return .same
      }
    }

    let segmentText = node.content.text
    if segmentText.hasSuffix("\n") {
      // If this is a multiline string segment, it will end in a newline. Remove the newline and
      // append the rest of the string, followed by a break if it's not the last line before the
      // closing quotes. (The `StringLiteralExpr` above does the closing break.)
      let remainder = node.content.text.dropLast()
      if !remainder.isEmpty {
        appendToken(.syntax(String(remainder)))
      }
      appendToken(.break(breakKind(), newlines: .hard(count: 1)))
    } else {
      appendToken(.syntax(segmentText))
    }

    if node.trailingTrivia?.containsBackslashes == true {
      // Segments with trailing backslashes won't end with a literal newline; the backslash is
      // considered trivia. To preserve the original text and wrapping, we need to manually render
      // the backslash and a break into the token stream.
      appendToken(.syntax("\\"))
      appendToken(.break(breakKind(), newlines: .hard(count: 1)))
    }
    return .skipChildren
  }

  override func visit(_ node: AssociatedtypeDeclSyntax) -> SyntaxVisitorContinueKind {
    arrangeAttributeList(node.attributes)

    after(node.associatedtypeKeyword, tokens: .break)

    if let genericWhereClause = node.genericWhereClause {
      before(genericWhereClause.firstToken, tokens: .break(.same), .open)
      after(node.lastToken, tokens: .close)
    }
    return .visitChildren
  }

  override func visit(_ node: BooleanLiteralExprSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: GenericWhereClauseSyntax) -> SyntaxVisitorContinueKind {
    guard node.whereKeyword != node.lastToken else {
      verbatimToken(Syntax(node))
      return .skipChildren
    }

    after(node.whereKeyword, tokens: .break(.open))
    after(node.lastToken, tokens: .break(.close, size: 0))

    before(node.requirementList.firstToken, tokens: .open(genericRequirementListConsistency()))
    after(node.requirementList.lastToken, tokens: .close)

    return .visitChildren
  }

  override func visit(_ node: IntegerLiteralExprSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: AccessPathComponentSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: GenericRequirementSyntax) -> SyntaxVisitorContinueKind {
    before(node.firstToken, tokens: .open)
    if let trailingComma = node.trailingComma {
      after(trailingComma, tokens: .close, .break(.same))
    } else {
      after(node.lastToken, tokens: .close)
    }
    return .visitChildren
  }

  override func visit(_ node: SameTypeRequirementSyntax) -> SyntaxVisitorContinueKind {
    before(node.equalityToken, tokens: .break)
    after(node.equalityToken, tokens: .space)

    return .visitChildren
  }

  override func visit(_ node: ConformanceRequirementSyntax) -> SyntaxVisitorContinueKind {
    after(node.colon, tokens: .break)

    return .visitChildren
  }

  override func visit(_ node: TuplePatternElementSyntax) -> SyntaxVisitorContinueKind {
    after(node.trailingComma, tokens: .break(.same))
    return .visitChildren
  }

  override func visit(_ node: MemberTypeIdentifierSyntax) -> SyntaxVisitorContinueKind {
    before(node.period, tokens: .break(.continue, size: 0))
    return .visitChildren
  }

  override func visit(_ node: OptionalChainingExprSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: SimpleTypeIdentifierSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: AvailabilityConditionSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: DiscardAssignmentExprSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: EditorPlaceholderExprSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: TypeInheritanceClauseSyntax) -> SyntaxVisitorContinueKind {
    // Normally, the open-break is placed before the open token. In this case, it's intentionally
    // ordered differently so that the inheritance list can start on the current line and only
    // breaks if the first item in the list would overflow the column limit.
    before(node.inheritedTypeCollection.firstToken, tokens: .open, .break(.open, size: 1))
    after(node.inheritedTypeCollection.lastToken, tokens: .break(.close, size: 0), .close)
    return .visitChildren
  }

  override func visit(_ node: UnresolvedPatternExprSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: CompositionTypeElementSyntax) -> SyntaxVisitorContinueKind {
    before(node.ampersand, tokens: .break)
    after(node.ampersand, tokens: .space)
    return .visitChildren
  }

  override func visit(_ node: MatchingPatternConditionSyntax) -> SyntaxVisitorContinueKind {
    before(node.firstToken, tokens: .open)
    after(node.caseKeyword, tokens: .break)
    after(node.lastToken, tokens: .close)
    return .visitChildren
  }

  override func visit(_ node: OptionalBindingConditionSyntax) -> SyntaxVisitorContinueKind {
    after(node.bindingKeyword, tokens: .break)

    if let typeAnnotation = node.typeAnnotation {
      after(
        typeAnnotation.colon,
        tokens: .break(.open(kind: .continuation), newlines: .elective(ignoresDiscretionary: true)))
      after(typeAnnotation.lastToken, tokens: .break(.close(mustBreak: false), size: 0))
    }

    return .visitChildren
  }

  override func visit(_ node: ImplicitlyUnwrappedOptionalTypeSyntax) -> SyntaxVisitorContinueKind {
    return .visitChildren
  }

  override func visit(_ node: DifferentiableAttributeArgumentsSyntax) -> SyntaxVisitorContinueKind {
    // This node encapsulates the entire list of arguments in a `@differentiable(...)` attribute.
    var needsBreakBeforeWhereClause = false

    if let diffParamsComma = node.diffParamsComma {
      after(diffParamsComma, tokens: .break(.same))
    } else if node.diffParams != nil {
      // If there were diff params but no comma following them, then we have "wrt: foo where ..."
      // and we need a break before the where clause.
      needsBreakBeforeWhereClause = true
    }

    if let whereClause = node.whereClause {
      if needsBreakBeforeWhereClause {
        before(whereClause.firstToken, tokens: .break(.same))
      }
      before(whereClause.firstToken, tokens: .open)
      after(whereClause.lastToken, tokens: .close)
    }
    return .visitChildren
  }

  override func visit(_ node: DifferentiabilityParamsSyntax) -> SyntaxVisitorContinueKind {
    after(node.leftParen, tokens: .break(.open, size: 0), .open)
    before(node.rightParen, tokens: .break(.close, size: 0), .close)
    return .visitChildren
  }

  override func visit(_ node: DifferentiabilityParamSyntax) -> SyntaxVisitorContinueKind {
    after(node.trailingComma, tokens: .break(.same))
    return .visitChildren
  }

  // `DerivativeRegistrationAttributeArguments` was added after the Swift 5.2 release was cut.
  #if HAS_DERIVATIVE_REGISTRATION_ATTRIBUTE
    override func visit(_ node: DerivativeRegistrationAttributeArgumentsSyntax)
      -> SyntaxVisitorContinueKind
    {
      // This node encapsulates the entire list of arguments in a `@derivative(...)` or
      // `@transpose(...)` attribute.
      before(node.ofLabel, tokens: .open)
      after(node.colon, tokens: .break(.continue, newlines: .elective(ignoresDiscretionary: true)))
      // The comma after originalDeclName is optional and is only present if there are diffParams.
      after(node.comma ?? node.originalDeclName.lastToken, tokens: .close)

      if let diffParams = node.diffParams {
        before(diffParams.firstToken, tokens: .break(.same), .open)
        after(diffParams.lastToken, tokens: .close)
      }

      return .visitChildren
    }
  #endif

  override func visit(_ node: DifferentiabilityParamsClauseSyntax) -> SyntaxVisitorContinueKind {
    // This node encapsulates the `wrt:` label and value/variable in a `@differentiable`,
    // `@derivative`, or `@transpose` attribute.
    after(node.colon, tokens: .break(.continue, newlines: .elective(ignoresDiscretionary: true)))
    return .visitChildren
  }

  // MARK: - Nodes representing unexpected or malformed syntax

  override func visit(_ node: UnexpectedNodesSyntax) -> SyntaxVisitorContinueKind {
    verbatimToken(Syntax(node))
    return .skipChildren
  }

  // MARK: - Token handling

  override func visit(_ token: TokenSyntax) -> SyntaxVisitorContinueKind {
    // Arrange the tokens and trivia such that before tokens that start a new "scope" (which may
    // increase indentation) are inserted in the stream *before* the leading trivia, but tokens
    // that end an existing "scope" (which may reduce indentation) are inserted *after* the leading
    // trivia. In general, comments before a token should included in the same scope as the token.
    let (openScopeTokens, closeScopeTokens) = splitScopingBeforeTokens(of: token)
    openScopeTokens.forEach(appendToken)
    extractLeadingTrivia(token)
    closeScopeTokens.forEach(appendToken)

    if !ignoredTokens.contains(token) {
      // Otherwise, it's just a regular token, so add the text as-is.
      appendToken(.syntax(token.presence == .present ? token.text : ""))
    }

    appendTrailingTrivia(token)
    appendAfterTokensAndTrailingComments(token)

    // It doesn't matter what we return here, tokens do not have children.
    return .skipChildren
  }

  /// Appends the before-tokens of the given syntax token to the token stream.
  private func appendBeforeTokens(_ token: TokenSyntax) {
    if let before = beforeMap.removeValue(forKey: token) {
      before.forEach(appendToken)
    }
  }

  /// Handle trailing trivia that might contain garbage text that we don't want to indiscriminantly
  /// discard.
  ///
  /// In syntactically valid code, trailing trivia will only contain spaces or tabs, so we can
  /// usually ignore it entirely. If there is garbage text after a token, however, then we preserve
  /// it (and any whitespace immediately before it) and "glue" it to the end of the preceding token
  /// using a `verbatim` formatting token. Any whitespace following the last garbage text in the
  /// trailing trivia will be discarded, with the assumption that the formatter will have inserted
  /// some kind of break there that would be more appropriate (and we want to avoid inserting
  /// trailing whitespace on a line).
  ///
  /// The choices above are admittedly somewhat arbitrary, but given that garbage text in trailing
  /// trivia represents a malformed input (as opposed to garbage text in leading trivia, which has
  /// some legitimate uses), this is a reasonable compromise to keep the garbage text roughly in the
  /// same place but still let surrounding formatting occur somewhat as expected.
  private func appendTrailingTrivia(_ token: TokenSyntax, forced: Bool = false) {
    let trailingTrivia = Array(token.trailingTrivia)
    let lastIndex: Array<TriviaPiece>.Index
    if forced {
      lastIndex = trailingTrivia.index(before: trailingTrivia.endIndex)
    } else {
      guard
        let lastUnexpectedIndex = trailingTrivia.lastIndex(where: { $0.isUnexpectedText })
      else {
        return
      }
      lastIndex = lastUnexpectedIndex
    }

    var verbatimText = ""
    for piece in trailingTrivia[...lastIndex] {
      switch piece {
      case .shebang, .unexpectedText, .spaces, .tabs, .formfeeds, .verticalTabs:
        piece.write(to: &verbatimText)
      default:
        // The implementation of the lexer today ensures that newlines, carriage returns, and
        // comments will not be present in trailing trivia. Ignore them for now (rather than assert,
        // in case that changes in a future version).
        break
      }
    }

    appendToken(.verbatim(Verbatim(text: verbatimText, indentingBehavior: .none)))
  }

  /// Appends the after-tokens and trailing comments (if present) of the given syntax token
  /// to the token stream.
  ///
  /// After-tokens require special care because the location of trailing comments (being in the
  /// trivia of the *next* token) sometimes can interfere with the ordering of formatting tokens
  /// being enqueued during visitation. Specifically:
  ///
  /// * If the trailing comment is a block comment, we append it first to the stream before any
  ///   other formatting tokens. This keeps the comment closely bound to the syntax token
  ///   preceding it; for example, if the comment occurs after the last token in a group, it
  ///   will stay inside the group.
  ///
  /// * If the trailing comment is a line comment, we first append any enqueued after-tokens
  ///   that are *not* breaks or newlines, then we append the comment, and then the remaining
  ///   after-tokens. Due to visitation ordering, this ensures that a trailing line comment is
  ///   not incorrectly inserted into the token stream *after* a break or newline.
  private func appendAfterTokensAndTrailingComments(_ token: TokenSyntax) {
    let (wasLineComment, trailingCommentTokens) = afterTokensForTrailingComment(token)
    let afterGroups = afterMap.removeValue(forKey: token) ?? []
    var hasAppendedTrailingComment = false

    if !wasLineComment {
      trailingCommentTokens.forEach(appendToken)
    }

    for after in afterGroups.reversed() {
      after.forEach { afterToken in
        var shouldExtractTrailingComment = false
        if wasLineComment && !hasAppendedTrailingComment {
          switch afterToken {
          case .break: shouldExtractTrailingComment = true
          default: break
          }
        }
        if shouldExtractTrailingComment {
          trailingCommentTokens.forEach(appendToken)
          hasAppendedTrailingComment = true
        }
        appendToken(afterToken)
      }
    }

    if wasLineComment && !hasAppendedTrailingComment {
      trailingCommentTokens.forEach(appendToken)
    }
  }

  // MARK: - Various other helper methods

  /// Applies formatting tokens around and between the attributes in an attribute list.
  private func arrangeAttributeList(
    _ attributes: AttributeListSyntax?,
    suppressFinalBreak: Bool = false
  ) {
    if let attributes = attributes {
      before(attributes.firstToken, tokens: .open)
      insertTokens(.break(.same), betweenElementsOf: attributes)
      var afterAttributeTokens = [Token.close]
      if !suppressFinalBreak {
        afterAttributeTokens.append(.break(.same))
      }
      after(attributes.lastToken, tokens: afterAttributeTokens)
    }
  }

  /// Returns a value indicating whether or not the given braced syntax node is completely empty;
  /// that is, it contains neither child syntax nodes (aside from the braces) *nor* any comments.
  ///
  /// Checking for comments separately is vitally important, because a code block that appears to be
  /// "empty" because it doesn't contain any statements might still contain comments, and if those
  /// are line comments, we need to make sure to insert the same breaks that we would if there were
  /// other statements there to get the same layout.
  ///
  /// Note the slightly different generic constraints on this and the other overloads. All are
  /// required because protocols in Swift do not conform to themselves, so if the element type of
  /// the collection is *precisely* `Syntax`, the constraint `BodyContents.Element: Syntax` is not
  /// satisfied and we must constrain it by `BodyContents.Element == Syntax` instead.
  ///
  /// - Parameters:
  ///   - node: A node that conforms to `BracedSyntax`.
  ///   - contentsKeyPath: A keypath describing how to get from `node` to the contents of the node
  ///     (a `Collection` whose elements are of a type that conforms to `Syntax`).
  /// - Returns: True if the collection at the node's keypath is empty and there are no comments.
  private func areBracesCompletelyEmpty<Node: BracedSyntax, BodyContents: SyntaxCollection>(
    _ node: Node,
    contentsKeyPath: KeyPath<Node, BodyContents>
  ) -> Bool where BodyContents.Element: SyntaxProtocol {
    // If the collection is empty, then any comments that might be present in the block must be
    // leading trivia of the right brace.
    let commentPrecedesRightBrace = node.rightBrace.leadingTrivia.numberOfComments > 0
    // We can't use `count` here because it also includes missing children. Instead, we get an
    // iterator and check if it returns `nil` immediately.
    var contentsIterator = node[keyPath: contentsKeyPath].makeIterator()
    return contentsIterator.next() == nil && !commentPrecedesRightBrace
  }

  /// Returns a value indicating whether or not the given braced syntax node is completely empty;
  /// that is, it contains neither child syntax nodes (aside from the braces) *nor* any comments.
  ///
  /// - Parameters:
  ///   - node: A node that conforms to `BracedSyntax`.
  ///   - contentsKeyPath: A keypath describing how to get from `node` to the contents of the node
  ///     (a `Collection` whose elements are of type `Syntax`).
  /// - Returns: True if the collection at the node's keypath is empty and there are no comments.
  private func areBracesCompletelyEmpty<Node: BracedSyntax, BodyContents: SyntaxCollection>(
    _ node: Node,
    contentsKeyPath: KeyPath<Node, BodyContents>
  ) -> Bool where BodyContents.Element == Syntax {
    // If the collection is empty, then any comments that might be present in the block must be
    // leading trivia of the right brace.
    let commentPrecedesRightBrace = node.rightBrace.leadingTrivia.numberOfComments > 0
    // We can't use `count` here because it also includes missing children. Instead, we get an
    // iterator and check if it returns `nil` immediately.
    var contentsIterator = node[keyPath: contentsKeyPath].makeIterator()
    return contentsIterator.next() == nil && !commentPrecedesRightBrace
  }

  /// Returns a value indicating whether or not the given braced syntax node is completely empty;
  /// that is, it contains neither child syntax nodes (aside from the braces) *nor* any comments.
  ///
  /// - Parameters:
  ///   - node: A node that conforms to `BracedSyntax`.
  ///   - contentsKeyPath: A keypath describing how to get from `node` to the contents of the node
  ///     (a `Collection` whose elements are of type `DeclSyntax`).
  /// - Returns: True if the collection at the node's keypath is empty and there are no comments.
  private func areBracesCompletelyEmpty<Node: BracedSyntax, BodyContents: SyntaxCollection>(
    _ node: Node,
    contentsKeyPath: KeyPath<Node, BodyContents>
  ) -> Bool where BodyContents.Element == DeclSyntax {
    // If the collection is empty, then any comments that might be present in the block must be
    // leading trivia of the right brace.
    let commentPrecedesRightBrace = node.rightBrace.leadingTrivia.numberOfComments > 0
    // We can't use `count` here because it also includes missing children. Instead, we get an
    // iterator and check if it returns `nil` immediately.
    var contentsIterator = node[keyPath: contentsKeyPath].makeIterator()
    return contentsIterator.next() == nil && !commentPrecedesRightBrace
  }

  /// Applies formatting to a collection of parameters for a decl.
  ///
  /// - Parameters:
  ///    - parameters: A node that contains the parameters that can be passed to a decl when its
  ///      called.
  ///    - forcesBreakBeforeRightParen: Whether a break should be required before the right paren
  ///      when the right paren is on a different line than the corresponding left paren.
  private func arrangeParameterClause(
    _ parameters: ParameterClauseSyntax, forcesBreakBeforeRightParen: Bool
  ) {
    guard !parameters.parameterList.isEmpty else { return }

    after(parameters.leftParen, tokens: .break(.open, size: 0), .open(argumentListConsistency()))
    before(
      parameters.rightParen,
      tokens: .break(.close(mustBreak: forcesBreakBeforeRightParen), size: 0), .close)
  }

  /// Applies consistent formatting to the braces and contents of the given node.
  ///
  /// - Parameters:
  ///   - node: A node that conforms to `BracedSyntax`.
  ///   - contentsKeyPath: A keypath describing how to get from `node` to the contents of the node
  ///     (a `Collection` whose elements are of a type that conforms to `Syntax`).
  ///   - shouldResetBeforeLeftBrace: If true, a `reset` break will be inserted before the node's
  ///     left brace (the default behavior). Passing false will suppress this break, which is useful
  ///     if you have already placed a `reset` elsewhere (for example, in a `guard` statement, the
  ///     `reset` is inserted before the `else` keyword to force both it and the brace down to the
  ///     next line).
  ///   - openBraceNewlineBehavior: The newline behavior to apply to the break following the open
  ///     brace; defaults to `.elective`.
  private func arrangeBracesAndContents<Node: BracedSyntax, BodyContents: SyntaxCollection>(
    of node: Node?,
    contentsKeyPath: KeyPath<Node, BodyContents>?,
    shouldResetBeforeLeftBrace: Bool = true,
    openBraceNewlineBehavior: NewlineBehavior = .elective
  ) where BodyContents.Element: SyntaxProtocol {
    guard let node = node, let contentsKeyPath = contentsKeyPath else { return }

    if shouldResetBeforeLeftBrace {
      before(
        node.leftBrace,
        tokens: .break(.reset, size: 1, newlines: .elective(ignoresDiscretionary: true)))
    }
    if !areBracesCompletelyEmpty(node, contentsKeyPath: contentsKeyPath) {
      after(
        node.leftBrace, tokens: .break(.open, size: 1, newlines: openBraceNewlineBehavior), .open)
      before(node.rightBrace, tokens: .break(.close, size: 1), .close)
    } else {
      after(node.leftBrace, tokens: .break(.open, size: 0, newlines: openBraceNewlineBehavior))
      before(node.rightBrace, tokens: .break(.close, size: 0))
    }
  }

  /// Applies consistent formatting to the braces and contents of the given node.
  ///
  /// - Parameters:
  ///   - node: A node that conforms to `BracedSyntax`.
  ///   - contentsKeyPath: A keypath describing how to get from `node` to the contents of the node
  ///     (a `Collection` whose elements are of type `Syntax`).
  ///   - shouldResetBeforeLeftBrace: If true, a `reset` break will be inserted before the node's
  ///     left brace (the default behavior). Passing false will suppress this break, which is useful
  ///     if you have already placed a `reset` elsewhere (for example, in a `guard` statement, the
  ///     `reset` is inserted before the `else` keyword to force both it and the brace down to the
  ///     next line).
  private func arrangeBracesAndContents<Node: BracedSyntax, BodyContents: SyntaxCollection>(
    of node: Node?,
    contentsKeyPath: KeyPath<Node, BodyContents>?,
    shouldResetBeforeLeftBrace: Bool = true
  ) where BodyContents.Element == Syntax {
    guard let node = node, let contentsKeyPath = contentsKeyPath else { return }

    if shouldResetBeforeLeftBrace {
      before(node.leftBrace, tokens: .break(.reset, size: 1))
    }
    if !areBracesCompletelyEmpty(node, contentsKeyPath: contentsKeyPath) {
      after(node.leftBrace, tokens: .break(.open, size: 1), .open)
      before(node.rightBrace, tokens: .break(.close, size: 1), .close)
    } else {
      after(node.leftBrace, tokens: .break(.open, size: 0))
      before(node.rightBrace, tokens: .break(.close, size: 0))
    }
  }

  /// Applies consistent formatting to the braces and contents of the given node.
  ///
  /// - Parameters:
  ///   - node: A node that conforms to `BracedSyntax`.
  ///   - contentsKeyPath: A keypath describing how to get from `node` to the contents of the node
  ///     (a `Collection` whose elements are of type `DeclSyntax`).
  ///   - shouldResetBeforeLeftBrace: If true, a `reset` break will be inserted before the node's
  ///     left brace (the default behavior). Passing false will suppress this break, which is useful
  ///     if you have already placed a `reset` elsewhere (for example, in a `guard` statement, the
  ///     `reset` is inserted before the `else` keyword to force both it and the brace down to the
  ///     next line).
  private func arrangeBracesAndContents<Node: BracedSyntax, BodyContents: SyntaxCollection>(
    of node: Node?,
    contentsKeyPath: KeyPath<Node, BodyContents>?,
    shouldResetBeforeLeftBrace: Bool = true
  ) where BodyContents.Element == DeclSyntax {
    guard let node = node, let contentsKeyPath = contentsKeyPath else { return }

    if shouldResetBeforeLeftBrace {
      before(node.leftBrace, tokens: .break(.reset, size: 1))
    }
    if !areBracesCompletelyEmpty(node, contentsKeyPath: contentsKeyPath) {
      after(node.leftBrace, tokens: .break(.open, size: 1), .open)
      before(node.rightBrace, tokens: .break(.close, size: 1), .close)
    } else {
      after(node.leftBrace, tokens: .break(.open, size: 0))
      before(node.rightBrace, tokens: .break(.close, size: 0))
    }
  }

  /// Applies consistent formatting to the braces and contents of the given node.
  ///
  /// - Parameter node: An `AccessorBlockSyntax` node.
  private func arrangeBracesAndContents(of node: AccessorBlockSyntax) {
    // If the collection is empty, then any comments that might be present in the block must be
    // leading trivia of the right brace.
    let commentPrecedesRightBrace = node.rightBrace.leadingTrivia.numberOfComments > 0
    // We can't use `count` here because it also includes missing children. Instead, we get an
    // iterator and check if it returns `nil` immediately.
    var accessorsIterator = node.accessors.makeIterator()
    let areAccessorsEmpty = accessorsIterator.next() == nil
    let bracesAreCompletelyEmpty = areAccessorsEmpty && !commentPrecedesRightBrace

    before(node.leftBrace, tokens: .break(.reset, size: 1))

    if !bracesAreCompletelyEmpty {
      after(node.leftBrace, tokens: .break(.open, size: 1), .open)
      before(node.rightBrace, tokens: .break(.close, size: 1), .close)
    } else {
      after(node.leftBrace, tokens: .break(.open, size: 0))
      before(node.rightBrace, tokens: .break(.close, size: 0))
    }
  }

  /// Returns the group consistency that should be used for argument lists based on the user's
  /// current configuration.
  private func argumentListConsistency() -> GroupBreakStyle {
    return config.lineBreakBeforeEachArgument ? .consistent : .inconsistent
  }

  /// Returns the group consistency that should be used for generic requirement lists based on
  /// the user's current configuration.
  private func genericRequirementListConsistency() -> GroupBreakStyle {
    return config.lineBreakBeforeEachGenericRequirement ? .consistent : .inconsistent
  }

  private func afterTokensForTrailingComment(_ token: TokenSyntax)
    -> (isLineComment: Bool, tokens: [Token])
  {
    let nextToken = token.nextToken
    guard let trivia = nextToken?.leadingTrivia,
      let firstPiece = trivia.first
    else {
      return (false, [])
    }

    switch firstPiece {
    case .lineComment(let text):
      return (
        true,
        [
          .space(size: 2, flexible: true),
          .comment(Comment(kind: .line, text: text), wasEndOfLine: true),
          // There must be a break with a soft newline after the comment, but it's impossible to
          // know which kind of break must be used. Adding this newline is deferred until the
          // comment is added to the token stream.
      ])

    case .blockComment(let text):
      return (
        false,
        [
          .space(size: 1, flexible: true),
          .comment(Comment(kind: .block, text: text), wasEndOfLine: false),
          // We place a size-0 break after the comment to allow a discretionary newline after
          // the comment if the user places one here but the comment is otherwise adjacent to a
          // text token.
          .break(.same, size: 0),
        ]
      )

    default:
      return (false, [])
    }
  }

  /// Splits the before tokens for the given token into an opening-scope collection and a
  /// closing-scope collection. The opening-scope collection contains `.open` and `.break` tokens
  /// that start a "scope" before the token. The closing-scope collection contains `.close` and
  /// `.break` tokens that end a "scope" after the token.
  private func splitScopingBeforeTokens(of token: TokenSyntax) -> (
    openingScope: [Token], closingScope: [Token]
  ) {
    guard let beforeTokens = beforeMap[token] else {
      return ([], [])
    }

    // Find the first index of a non-opening-scope token, and split into the two sections.
    for (index, beforeToken) in beforeTokens.enumerated() {
      switch beforeToken {
      case .break(.open, _, _), .break(.continue, _, _), .break(.same, _, _),
        .break(.contextual, _, _), .open:
        break
      default:
        if index > 0 {
          return (Array(beforeTokens[0...(index - 1)]), Array(beforeTokens[index...]))
        } else {
          return ([], beforeTokens)
        }
      }
    }
    // Never found a closing-scope token, so assume they're all opening-scope.
    return (beforeTokens, [])
  }

  private func extractLeadingTrivia(_ token: TokenSyntax) {
    var isStartOfFile = token.previousToken(viewMode: .all) == nil
    let trivia = token.leadingTrivia

    // If we're at the end of the file, determine at which index to stop checking trivia pieces to
    // prevent trailing newlines.
    var cutoffIndex: Int? = nil
    if token.tokenKind == TokenKind.eof {
      cutoffIndex = 0
      for (index, piece) in trivia.enumerated() {
        switch piece {
        case .newlines(_), .carriageReturns(_), .carriageReturnLineFeeds(_):
          continue
        default:
          cutoffIndex = index + 1
        }
      }
    }

    // Updated throughout the loop to indicate whether the next newline *must* be honored (for
    // example, even if discretionary newlines are discarded). This is the case when the preceding
    // trivia was a line comment or garbage text.
    var requiresNextNewline = false

    for (index, piece) in trivia.enumerated() {
      if let cutoff = cutoffIndex, index == cutoff { break }
      switch piece {
      case .lineComment(let text):
        if index > 0 || isStartOfFile {
          appendToken(.comment(Comment(kind: .line, text: text), wasEndOfLine: false))
          appendNewlines(.soft)
          isStartOfFile = false
        }
        requiresNextNewline = true

      case .blockComment(let text):
        if index > 0 || isStartOfFile {
          appendToken(.comment(Comment(kind: .block, text: text), wasEndOfLine: false))
          // There is always a break after the comment to allow a discretionary newline after it.
          var breakSize = 0
          if index + 1 < trivia.endIndex {
            let nextPiece = trivia[index + 1]
            // The original number of spaces is intentionally discarded, but 1 space is allowed in
            // case the comment is followed by another token instead of a newline.
            if case .spaces = nextPiece { breakSize = 1 }
          }
          appendToken(.break(.same, size: breakSize))
          isStartOfFile = false
        }
        requiresNextNewline = false

      case .docLineComment(let text):
        appendToken(.comment(Comment(kind: .docLine, text: text), wasEndOfLine: false))
        appendNewlines(.soft)
        isStartOfFile = false
        requiresNextNewline = true

      case .docBlockComment(let text):
        appendToken(.comment(Comment(kind: .docBlock, text: text), wasEndOfLine: false))
        appendNewlines(.soft)
        isStartOfFile = false
        requiresNextNewline = false

      case .newlines(let count), .carriageReturns(let count), .carriageReturnLineFeeds(let count):
        guard !isStartOfFile else { break }

        if requiresNextNewline ||
          (config.respectsExistingLineBreaks && isDiscretionaryNewlineAllowed(before: token))
        {
          appendNewlines(.soft(count: count, discretionary: true))
        } else {
          // Even if discretionary line breaks are not being respected, we still respect multiple
          // line breaks in order to keep blank separator lines that the user might want.
          // TODO: It would be nice to restrict this to only allow multiple lines between statements
          // and declarations; as currently implemented, multiple newlines will locally ignore the
          // configuration setting.
          if count > 1 {
            appendNewlines(.soft(count: count, discretionary: true))
          }
        }

      case .shebang(let text), .unexpectedText(let text):
        // Garbage text in leading trivia might be something meaningful that would be disruptive to
        // throw away when formatting the file, like a hashbang line or Unicode byte-order marker at
        // the beginning of a file, or source control conflict markers. Keep it as verbatim text so
        // that it is printed exactly as we got it.
        appendToken(.verbatim(Verbatim(text: text, indentingBehavior: .none)))

        // Unicode byte-order markers shouldn't allow leading newlines to otherwise appear in the
        // file, nor should they modify our detection of the beginning of the file.
        let isBOM = text == "\u{feff}"
        requiresNextNewline = !isBOM
        isStartOfFile = isStartOfFile && isBOM

      default:
        break
      }
    }
  }

  /// Returns a value indicating whether or not discretionary newlines are permitted before the
  /// given syntax token.
  ///
  /// Discretionary newlines are allowed before any token (ignoring open/close group tokens, which
  /// do not contribute to this) that is preceded by an existing newline or that is preceded by a
  /// break whose `ignoresDiscretionary` property is false. In other words, this means that users
  /// may insert their own breaks in places where the pretty printer allows them, even if those
  /// breaks wouldn't cause wrapping based on the column limit, but they may not place them in
  /// places where the pretty printer would not break (for example, at a space token that is
  /// intended to keep two tokens glued together).
  ///
  /// Furthermore, breaks with `ignoresDiscretionary` equal to `true` are in effect "last resort"
  /// breaks; a user's newline will be discarded unless the algorithm *must* break there. For
  /// example, an open curly brace on a non-continuation line should always be kept on the same line
  /// as the tokens before it unless the tokens before it are exactly the length of the line and a
  /// break must be inserted there to prevent the brace from going over the limit.
  private func isDiscretionaryNewlineAllowed(before token: TokenSyntax) -> Bool {
    func isBreakMoreRecentThanNonbreakingContent(_ tokens: [Token]) -> Bool? {
      for token in tokens.reversed() as ReversedCollection {
        switch token {
        case .break(_, _, .elective(ignoresDiscretionary: true)): return false
        case .break: return true
        case .comment, .space, .syntax, .verbatim: return false
        default: break
        }
      }
      return nil
    }

    // First, check the pretty printer tokens that will be added before the text token. If we find
    // a break or newline before we find some other text, we allow a discretionary newline. If we
    // find some other content, we don't allow it.
    //
    // If there were no before tokens, then we do the same check the token stream created thus far,
    // returning true if there were no tokens at all in the stream (which would mean there was a
    // discretionary newline at the beginning of the file).
    if let beforeTokens = beforeMap[token],
      let foundBreakFirst = isBreakMoreRecentThanNonbreakingContent(beforeTokens)
    {
      return foundBreakFirst
    }
    return isBreakMoreRecentThanNonbreakingContent(tokens) ?? true
  }

  /// Appends the newlines to the token stream.
  ///
  /// The newlines will be inserted using one of the following approaches:
  /// - As a new break, whose kind is compatible with the most recent break.
  /// - Overwriting the newlines of the most recent break.
  /// - Appending to the newlines of the most recent break.
  private func appendNewlines(_ newlines: NewlineBehavior) {
    guard let lastBreakIndex = lastBreakIndex else {
      // When there haven't been any breaks yet, there can't be any indentation to maintain so a
      // same break is safe here.
      appendToken(.break(.same, size: 0, newlines: newlines))
      return
    }

    let lastBreak = tokens[lastBreakIndex]
    guard case .break(let kind, let size, let existingNewlines) = lastBreak else {
      fatalError("Found non-break token at lastBreakIndex. TokenStreamCreator is invalid.")
    }

    guard !canMergeNewlinesIntoLastBreak else {
      tokens[lastBreakIndex] = .break(kind, size: size, newlines: existingNewlines + newlines)
      return
    }

    // Otherwise, create and insert a new break whose `kind` is compatible with last break.
    let compatibleKind: BreakKind
    switch kind {
    case .open, .close, .reset, .same:
      compatibleKind = .same
    case .continue, .contextual:
      compatibleKind = kind
    }
    appendToken(.break(compatibleKind, size: 0, newlines: newlines))
  }

  /// Appends a formatting token to the token stream.
  ///
  /// This function also handles collapsing neighboring tokens in situations where that is
  /// desired, like merging adjacent comments and newlines.
  private func appendToken(_ token: Token) {
    if let last = tokens.last {
      switch (last, token) {
      case (.comment(let c1, _), .comment(let c2, _))
      where c1.kind == .docLine && c2.kind == .docLine:
        var newComment = c1
        newComment.addText(c2.text)
        tokens[tokens.count - 1] = .comment(newComment, wasEndOfLine: false)
        return

      // If we see a pair of spaces where one or both are flexible, combine them into a new token
      // with the maximum of their counts.
      case (.space(let first, let firstFlexible), .space(let second, let secondFlexible))
      where firstFlexible || secondFlexible:
        tokens[tokens.count - 1] = .space(size: max(first, second), flexible: true)
        return

      default:
        break
      }
    }

    switch token {
    case .break:
      lastBreakIndex = tokens.endIndex
      canMergeNewlinesIntoLastBreak = true
    case .open, .printerControl, .contextualBreakingStart:
      break
    default:
      canMergeNewlinesIntoLastBreak = false
    }
    tokens.append(token)
  }

  /// Returns true if the first token of the given node is an open delimiter that may desire
  /// special breaking behavior in some cases.
  private func startsWithOpenDelimiter(_ node: Syntax) -> Bool {
    guard let token = node.firstToken else { return false }
    switch token.tokenKind {
    case .leftBrace, .leftParen, .leftSquareBracket: return true
    default: return false
    }
  }

  /// Returns true if open/close breaks should be inserted around the entire function call argument
  /// list.
  private func shouldGroupAroundArgumentList(_ arguments: TupleExprElementListSyntax) -> Bool {
    let argumentCount = arguments.count

    // If there are no arguments, there's no reason to break.
    if argumentCount == 0 { return false }

    // If there is more than one argument, we must open/close break around the whole list.
    if argumentCount > 1 { return true }

    return !isCompactSingleFunctionCallArgument(arguments)
  }

  /// Returns whether the `reset` break before an expression's closing delimiter must break when
  /// it's on a different line than the opening delimiter.
  /// - Parameters:
  ///   - expr: An expression that includes opening and closing delimiters and arguments.
  ///   - argumentListPath: A key path for accessing the expression's function call argument list.
  private func mustBreakBeforeClosingDelimiter<T: ExprSyntaxProtocol>(
    of expr: T, argumentListPath: KeyPath<T, TupleExprElementListSyntax>
  ) -> Bool {
    guard let parent = expr.parent, parent.is(MemberAccessExprSyntax.self) else { return false }
    let argumentList = expr[keyPath: argumentListPath]

    // When there's a single compact argument, there is no extra indentation for the argument and
    // the argument's own internal reset break will reset indentation.
    return !isCompactSingleFunctionCallArgument(argumentList)
  }

  /// Returns true if the argument list can be compacted, even if it spans multiple lines (where
  /// compact means that it can start immediately after the open parenthesis).
  ///
  /// This is true for any argument list that contains a single argument (labeled or unlabeled) that
  /// is an array, dictionary, or closure literal.
  func isCompactSingleFunctionCallArgument(_ argumentList: TupleExprElementListSyntax) -> Bool {
    guard argumentList.count == 1 else { return false }

    let expression = argumentList.first!.expression
    return expression.is(ArrayExprSyntax.self) || expression.is(DictionaryExprSyntax.self)
      || expression.is(ClosureExprSyntax.self)
  }

  /// Returns a value indicating whether a statement or member declaration should have a newline
  /// inserted after it, based on the presence of a semicolon and whether or not the formatter is
  /// respecting existing newlines.
  private func shouldInsertNewline(basedOn semicolon: TokenSyntax?) -> Bool {
    if config.respectsExistingLineBreaks {
      // If we are respecting existing newlines, then we only want to force a newline at the end of
      // statements and declarations that don't have a semicolon (i.e., where they are required).
      return semicolon == nil
    } else {
      // If we are not respecting existing newlines, then we always force a newline (this forces
      // even semicolon-delimited statements onto separate lines).
      return true
    }
  }

  /// Adds a grouping around certain subexpressions during `InfixOperatorExpr` visitation.
  ///
  /// Adding groups around these expressions allows them to prefer breaking onto a newline before
  /// the expression, keeping the entire expression together when possible, before breaking inside
  /// the expression. This is a hand-crafted list of expressions that generally look better when the
  /// break(s) before the expression fire before breaks inside of the expression.
  private func maybeGroupAroundSubexpression(
    _ expr: ExprSyntax, combiningOperator operatorExpr: ExprSyntax? = nil
  ) {
    switch Syntax(expr).as(SyntaxEnum.self) {
    case .memberAccessExpr, .subscriptExpr:
      before(expr.firstToken, tokens: .open)
      after(expr.lastToken, tokens: .close)
    default:
      break
    }

    // When a function call expression is assigned to an lvalue, we omit the group around the
    // function call so that the callee and open parenthesis can remain on the same line, if they
    // fit. This is a frequent enough case that the outcome looks better with the exception in
    // place.
    if expr.is(FunctionCallExprSyntax.self),
      let operatorExpr = operatorExpr, !isAssigningOperator(operatorExpr)
    {
      before(expr.firstToken, tokens: .open)
      after(expr.lastToken, tokens: .close)
    }
  }

  /// Returns whether the given expression consists of multiple subexpressions. Certain expressions
  /// that are known to wrap an expression, e.g. try expressions, are handled by checking the
  /// expression that they contain.
  private func isCompoundExpression(_ expr: ExprSyntax) -> Bool {
    switch Syntax(expr).as(SyntaxEnum.self) {
    case .awaitExpr(let awaitExpr):
      return isCompoundExpression(awaitExpr.expression)
    case .infixOperatorExpr, .ternaryExpr, .isExpr, .asExpr:
      return true
    case .tryExpr(let tryExpr):
      return isCompoundExpression(tryExpr.expression)
    case .tupleExpr(let tupleExpr) where tupleExpr.elementList.count == 1:
      return isCompoundExpression(tupleExpr.elementList.first!.expression)
    default:
      return false
    }
  }

  /// Returns whether the given operator behaves as an assignment, to assign a right-hand-side to a
  /// left-hand-side in a `InfixOperatorExpr`.
  ///
  /// Assignment is defined as either being an assignment operator (i.e. `=`) or any operator that
  /// uses "assignment" precedence.
  private func isAssigningOperator(_ operatorExpr: ExprSyntax) -> Bool {
    if operatorExpr.is(AssignmentExprSyntax.self) {
      return true
    }
    if let binOpExpr = operatorExpr.as(BinaryOperatorExprSyntax.self) {
      if let binOp = operatorTable.infixOperator(named: binOpExpr.operatorToken.text),
        let precedenceGroup = binOp.precedenceGroup, precedenceGroup == "AssignmentPrecedence"
      {
        return true
      }
    }
    return false
  }

  /// Walks the expression and returns the leftmost subexpression if it is parenthesized (which
  /// might be the expression itself).
  ///
  /// - Parameter expr: The expression whose parenthesized leftmost subexpression should be
  ///   returned.
  /// - Returns: The parenthesized leftmost subexpression, or nil if the leftmost subexpression was
  ///   not parenthesized.
  private func parenthesizedLeftmostExpr(of expr: ExprSyntax) -> TupleExprSyntax? {
    switch Syntax(expr).as(SyntaxEnum.self) {
    case .tupleExpr(let tupleExpr) where tupleExpr.elementList.count == 1:
      return tupleExpr
    case .infixOperatorExpr(let infixOperatorExpr):
      return parenthesizedLeftmostExpr(of: infixOperatorExpr.leftOperand)
    case .ternaryExpr(let ternaryExpr):
      return parenthesizedLeftmostExpr(of: ternaryExpr.conditionExpression)
    default:
      return nil
    }
  }

  /// Walks the expression and returns the leftmost multiline string literal (which might be the
  /// expression itself) if the leftmost child is a multiline string literal.
  ///
  /// - Parameter expr: The expression whose leftmost multiline string literal should be returned.
  /// - Returns: The leftmost multiline string literal, or nil if the leftmost subexpression was
  ///   not a multiline string literal.
  private func leftmostMultilineStringLiteral(of expr: ExprSyntax) -> StringLiteralExprSyntax? {
    switch Syntax(expr).as(SyntaxEnum.self) {
    case .stringLiteralExpr(let stringLiteralExpr)
      where stringLiteralExpr.openQuote.tokenKind == .multilineStringQuote:
      return stringLiteralExpr
    case .infixOperatorExpr(let infixOperatorExpr):
      return leftmostMultilineStringLiteral(of: infixOperatorExpr.leftOperand)
    case .ternaryExpr(let ternaryExpr):
      return leftmostMultilineStringLiteral(of: ternaryExpr.conditionExpression)
    default:
      return nil
    }
  }

  /// Returns the outermost node enclosing the given node whose closing delimiter(s) must be kept
  /// alongside the last token of the given node. Any tokens between `node.lastToken` and the
  /// returned node's `lastToken` are delimiter tokens that shouldn't be preceded by a break.
  private func outermostEnclosingNode(from node: Syntax) -> Syntax? {
    guard let afterToken = node.lastToken?.nextToken(viewMode: .all), closingDelimiterTokens.contains(afterToken)
    else {
      return nil
    }
    var parenthesizedExpr = afterToken.parent
    while let nextToken = parenthesizedExpr?.lastToken?.nextToken(viewMode: .all),
      closingDelimiterTokens.contains(nextToken),
      let nextExpr = nextToken.parent
    {
      parenthesizedExpr = nextExpr
    }
    return parenthesizedExpr
  }

  /// Determines if indentation should be stacked around a subexpression to the right of the given
  /// operator, and, if so, returns the node after which indentation stacking should be closed and
  /// whether or not the continuation state should be reset as well.
  ///
  /// Stacking is applied around parenthesized expressions, but also for low-precedence operators
  /// that frequently occur in long chains, such as logical AND (`&&`) and OR (`||`) in conditional
  /// statements. In this case, the extra level of indentation helps to improve readability with the
  /// operators inside those conditions even when parentheses are not used.
  private func stackedIndentationBehavior(
    after operatorExpr: ExprSyntax? = nil,
    rhs: ExprSyntax
  ) -> (unindentingNode: Syntax, shouldReset: Bool, breakKind: OpenBreakKind)? {
    // Check for logical operators first, and if it's that kind of operator, stack indentation
    // around the entire right-hand-side. We have to do this check before checking the RHS for
    // parentheses because if the user writes something like `... && (foo) > bar || ...`, we don't
    // want the indentation stacking that starts before the `&&` to stop after the closing
    // parenthesis in `(foo)`.
    //
    // We also want to reset after undoing the stacked indentation so that we have a visual
    // indication that the subexpression has ended.
    if let binOpExpr = operatorExpr?.as(BinaryOperatorExprSyntax.self) {
      if let binOp = operatorTable.infixOperator(named: binOpExpr.operatorToken.text),
        let precedenceGroup = binOp.precedenceGroup,
        precedenceGroup == "LogicalConjunctionPrecedence"
          || precedenceGroup == "LogicalDisjunctionPrecedence"
      {
        // When `rhs` side is the last sequence in an enclosing parenthesized expression, absorb the
        // paren into the right hand side by unindenting after the final closing paren. This glues
        // the paren to the last token of `rhs`.
        if let unindentingParenExpr = outermostEnclosingNode(from: Syntax(rhs)) {
          return (
            unindentingNode: unindentingParenExpr, shouldReset: true, breakKind: .continuation)
        }
        return (unindentingNode: Syntax(rhs), shouldReset: true, breakKind: .continuation)
      }
    }

    // If the right-hand-side is a ternary expression, stack indentation around the condition so
    // that it is indented relative to the `?` and `:` tokens.
    if let ternaryExpr = rhs.as(TernaryExprSyntax.self) {
      // We don't try to absorb any parens in this case, because the condition of a ternary cannot
      // be grouped with any exprs outside of the condition.
      return (
        unindentingNode: Syntax(ternaryExpr.conditionExpression), shouldReset: false,
        breakKind: .continuation)
    }

    // If the right-hand-side of the operator is or starts with a parenthesized expression, stack
    // indentation around the operator and those parentheses. We don't need to reset here because
    // the parentheses are sufficient to provide a visual indication of the nesting relationship.
    if let parenthesizedExpr = parenthesizedLeftmostExpr(of: rhs) {
      // When `rhs` side is the last sequence in an enclosing parenthesized expression, absorb the
      // paren into the right hand side by unindenting after the final closing paren. This glues the
      // paren to the last token of `rhs`.
      if let unindentingParenExpr = outermostEnclosingNode(from: Syntax(rhs)) {
        return (unindentingNode: unindentingParenExpr, shouldReset: true, breakKind: .continuation)
      }

      if let innerExpr = parenthesizedExpr.elementList.first?.expression,
        let stringLiteralExpr = innerExpr.as(StringLiteralExprSyntax.self),
        stringLiteralExpr.openQuote.tokenKind == .multilineStringQuote
      {
        pendingMultilineStringBreakKinds[stringLiteralExpr] = .continue
        return nil
      }

      return (
        unindentingNode: Syntax(parenthesizedExpr), shouldReset: false, breakKind: .continuation)
    }

    // If the expression is a multiline string that is unparenthesized, create a block-based
    // indentation scope and have the segments aligned inside it.
    if let stringLiteralExpr = leftmostMultilineStringLiteral(of: rhs) {
      pendingMultilineStringBreakKinds[stringLiteralExpr] = .same
      return (unindentingNode: Syntax(stringLiteralExpr), shouldReset: false, breakKind: .block)
    }

    // Otherwise, don't stack--use regular continuation breaks instead.
    return nil
  }

  /// Returns a value indicating whether whitespace should be required around the given operator,
  /// for the given configuration.
  ///
  /// If spaces are not required (for example, range operators), then the formatter will also forbid
  /// breaks around the operator. This is to prevent situations where a break could occur before an
  /// unspaced operator (e.g., turning `0...10` into `0<newline>...10`), which would be a breaking
  /// change because it would treat it as a prefix operator `...10` instead of an infix operator.
  private func shouldRequireWhitespace(around operatorExpr: ExprSyntax) -> Bool {
    // Note that we look at the operator itself to make this determination, not the token kind.
    // The token kind (spaced or unspaced operator) represents how the *user* wrote it, and we want
    // to ignore that and apply our own rules.
    if let binaryOperator = operatorExpr.as(BinaryOperatorExprSyntax.self) {
      let token = binaryOperator.operatorToken
      if !config.spacesAroundRangeFormationOperators,
         let binOp = operatorTable.infixOperator(named: token.text),
         let precedenceGroup = binOp.precedenceGroup, precedenceGroup == "RangeFormationPrecedence"
      {
        // We want to omit whitespace around range formation operators if possible. We can't do this
        // if the token is either preceded by a postfix operator, followed by a prefix operator, or
        // followed by a dot (for example, in an implicit member reference)---removing the spaces in
        // those situations would cause the parser to greedily treat the combined sequence of
        // operator characters as a single operator.
        if case .postfixOperator? = token.previousToken(viewMode: .all)?.tokenKind { return true }

        switch token.nextToken(viewMode: .all)?.tokenKind {
        case .prefixOperator?, .period?: return true
        default: return false
        }
      }
    }

    // For all other operators, we want to require whitespace on each side. That's always safe, so
    // we don't need to be concerned about neighboring operator tokens. For example, we don't need
    // to be concerned about the user writing "4+-5" when they meant "4 + -5", because Swift would
    // always parse the former as "4 +- 5".
    return true
  }

  /// Appends the given node to the token stream without applying any formatting or printing tokens.
  ///
  /// - Parameter node: A node that is ignored by the formatter.
  private func appendFormatterIgnored(node: Syntax) {
    // The first line of text in the `verbatim` token is printed with correct indentation, based on
    // the previous tokens. The leading trivia of the first token needs to be excluded from the
    // `verbatim` token in order for the first token to be printed with correct indentation. All
    // following lines in the ignored node are printed as-is with no changes to indentation.
    var nodeText = node.description
    if let firstToken = node.firstToken {
      extractLeadingTrivia(firstToken)
      let leadingTriviaText = firstToken.leadingTrivia.reduce(into: "") { $1.write(to: &$0) }
      nodeText = String(nodeText.dropFirst(leadingTriviaText.count))
    }

    // The leading trivia of the next token, after the ignored node, may contain content that
    // belongs with the ignored node. The trivia extraction that is performed for `lastToken` later
    // excludes that content so it needs to be extracted and added to the token stream here.
    if let next = node.lastToken?.nextToken(viewMode: .all), let trivia = next.leadingTrivia.first {
      switch trivia {
      case .lineComment, .blockComment:
        trivia.write(to: &nodeText)
        break
      default:
        // All other kinds of trivia are inserted into the token stream by `extractLeadingTrivia`
        // when the relevant token is visited.
        break
      }
    }

    appendToken(.verbatim(Verbatim(text: nodeText, indentingBehavior: .firstLine)))

    // Add this break so that trivia parsing will allow discretionary newlines after the node.
    appendToken(.break(.same, size: 0))
  }

  /// Cleans up state related to inserting contextual breaks throughout expressions during
  /// `visitPost` for an expression that is the root of an expression tree.
  private func clearContextualBreakState<T: ExprSyntaxProtocol>(_ expr: T) {
    let exprID = expr.id
    if rootExprs.remove(exprID) != nil {
      preVisitedExprs.removeAll()
    }
  }

  /// Visits the given expression node and all of the nested expression nodes, inserting tokens
  /// necessary for contextual breaking throughout the expression. Records the nodes that were
  /// visited so that they can be skipped later.
  private func preVisitInsertingContextualBreaks<T: ExprSyntaxProtocol & Equatable>(_ expr: T) {
    let exprID = expr.id
    if !preVisitedExprs.contains(exprID) {
      rootExprs.insert(exprID)
      insertContextualBreaks(ExprSyntax(expr), isTopLevel: true)
    }
  }

  /// Recursively visits nested expressions from the given expression inserting contextual breaking
  /// tokens. When visiting an expression node, `preVisitInsertingContextualBreaks(_:)` should be
  /// called instead of this helper.
  @discardableResult
  private func insertContextualBreaks(_ expr: ExprSyntax, isTopLevel: Bool) -> (
    hasCompoundExpression: Bool, hasMemberAccess: Bool
  ) {
    preVisitedExprs.insert(expr.id)
    if let memberAccessExpr = expr.as(MemberAccessExprSyntax.self) {
      // When the member access is part of a calling expression, the break before the dot is
      // inserted when visiting the parent node instead so that the break is inserted before any
      // scoping tokens (e.g. `contextualBreakingStart`, `open`).
      if memberAccessExpr.base != nil &&
          expr.parent?.isProtocol(CallingExprSyntaxProtocol.self) != true {
        before(memberAccessExpr.dot, tokens: .break(.contextual, size: 0))
      }
      var hasCompoundExpression = false
      if let base = memberAccessExpr.base {
        (hasCompoundExpression, _) = insertContextualBreaks(base, isTopLevel: false)
      }
      if isTopLevel {
        before(expr.firstToken, tokens: .contextualBreakingStart)
        after(expr.lastToken, tokens: .contextualBreakingEnd)
      }
      return (hasCompoundExpression, true)
    } else if let callingExpr = expr.asProtocol(CallingExprSyntaxProtocol.self) {
      let calledExpression = callingExpr.calledExpression
      let (hasCompoundExpression, hasMemberAccess) =
        insertContextualBreaks(calledExpression, isTopLevel: false)

      let shouldGroup =
        hasMemberAccess && (hasCompoundExpression || !isTopLevel)
        && config.lineBreakAroundMultilineExpressionChainComponents
      let beforeTokens: [Token] =
        shouldGroup ? [.contextualBreakingStart, .open] : [.contextualBreakingStart]
      let afterTokens: [Token] =
        shouldGroup ? [.contextualBreakingEnd, .close] : [.contextualBreakingEnd]

      if let calledMemberAccessExpr = calledExpression.as(MemberAccessExprSyntax.self) {
        if calledMemberAccessExpr.base != nil {
          if isNestedInPostfixIfConfig(node: Syntax(calledMemberAccessExpr)) {
            before(calledMemberAccessExpr.dot, tokens: [.break(.same, size: 0)])
          } else {
            before(calledMemberAccessExpr.dot, tokens: [.break(.contextual, size: 0)])
          }
        }
        before(calledMemberAccessExpr.dot, tokens: beforeTokens)
        after(expr.lastToken, tokens: afterTokens)
        if isTopLevel {
          before(expr.firstToken, tokens: .contextualBreakingStart)
          after(expr.lastToken, tokens: .contextualBreakingEnd)
        }
      } else {
        before(expr.firstToken, tokens: beforeTokens)
        after(expr.lastToken, tokens: afterTokens)
      }
      return (true, hasMemberAccess)
    }

    // Otherwise, it's an expression that isn't calling another expression (e.g. array or
    // dictionary, identifier, etc.). Wrap it in a breaking context but don't try to pre-visit
    // children nodes.
    before(expr.firstToken, tokens: .contextualBreakingStart)
    after(expr.lastToken, tokens: .contextualBreakingEnd)
    let hasCompoundExpression = !expr.is(IdentifierExprSyntax.self)
    return (hasCompoundExpression, false)
  }
}

private func isNestedInPostfixIfConfig(node: Syntax) -> Bool {
    var this: Syntax? = node

    while this?.parent != nil {
      if this?.parent?.is(PostfixIfConfigExprSyntax.self) == true {
        return true
      }

      this = this?.parent
    }

    return false
}

extension Syntax {
  /// Creates a pretty-printable token stream for the provided Syntax node.
  func makeTokenStream(configuration: Configuration, operatorTable: OperatorTable) -> [Token] {
    let commentsMoved = CommentMovingRewriter().visit(self)
    return TokenStreamCreator(configuration: configuration, operatorTable: operatorTable)
      .makeStream(from: commentsMoved)
  }
}

/// Rewriter that relocates comment trivia around nodes where comments are known to be better
/// formatted when placed before or after the node.
///
/// For example, comments after binary operators are relocated to be before the operator, which
/// results in fewer line breaks with the comment closer to the relevant tokens.
class CommentMovingRewriter: SyntaxRewriter {
  /// Map of tokens to alternate trivia to use as the token's leading trivia.
  var rewriteTokenTriviaMap: [TokenSyntax: Trivia] = [:]

  override func visit(_ node: SourceFileSyntax) -> SourceFileSyntax {
    if shouldFormatterIgnore(file: node) {
      return node
    }
    return super.visit(node)
  }

  override func visit(_ node: CodeBlockItemSyntax) -> CodeBlockItemSyntax {
    if shouldFormatterIgnore(node: Syntax(node)) {
      return node
    }
    return super.visit(node)
  }

  override func visit(_ node: MemberDeclListItemSyntax) -> MemberDeclListItemSyntax {
    if shouldFormatterIgnore(node: Syntax(node)) {
      return node
    }
    return super.visit(node)
  }

  override func visit(_ token: TokenSyntax) -> TokenSyntax {
    if let rewrittenTrivia = rewriteTokenTriviaMap[token] {
      return token.with(\.leadingTrivia, rewrittenTrivia)
    }
    return token
  }

  override func visit(_ node: InfixOperatorExprSyntax) -> ExprSyntax {
    if let binaryOperatorExpr = node.operatorOperand.as(BinaryOperatorExprSyntax.self),
      let followingToken = binaryOperatorExpr.operatorToken.nextToken(viewMode: .all),
      followingToken.leadingTrivia.hasLineComment
    {
      // Rewrite the trivia so that the comment is in the operator token's leading trivia.
      let (remainingTrivia, extractedTrivia) = extractLineCommentTrivia(from: followingToken)
      let combinedTrivia = binaryOperatorExpr.operatorToken.leadingTrivia + extractedTrivia
      rewriteTokenTriviaMap[binaryOperatorExpr.operatorToken] = combinedTrivia
      rewriteTokenTriviaMap[followingToken] = remainingTrivia
    }
    return super.visit(node)
  }

  /// Extracts trivia containing and related to line comments from `token`'s leading trivia. Returns
  /// 2 trivia collections: the trivia that wasn't extracted and should remain in `token`'s leading
  /// trivia and the trivia that meets the criteria for extraction.
  /// - Parameter token: A token whose leading trivia should be split to extract line comments.
  private func extractLineCommentTrivia(from token: TokenSyntax) -> (
    remainingTrivia: Trivia, extractedTrivia: Trivia
  ) {
    var pendingPieces = [TriviaPiece]()
    var keepWithTokenPieces = [TriviaPiece]()
    var extractingPieces = [TriviaPiece]()

    // Line comments and adjacent newlines are extracted so they can be moved to a different token's
    // leading trivia, while all other kinds of tokens are left as-is.
    var lastPiece: TriviaPiece?
    for piece in token.leadingTrivia {
      defer { lastPiece = piece }
      switch piece {
      case .lineComment:
        extractingPieces.append(contentsOf: pendingPieces)
        pendingPieces.removeAll()
        extractingPieces.append(piece)
      case .blockComment, .docLineComment, .docBlockComment:
        keepWithTokenPieces.append(contentsOf: pendingPieces)
        pendingPieces.removeAll()
        keepWithTokenPieces.append(piece)
      case .newlines, .carriageReturns, .carriageReturnLineFeeds:
        if case .lineComment = lastPiece {
          extractingPieces.append(piece)
        } else {
          pendingPieces.append(piece)
        }
      default:
        pendingPieces.append(piece)
      }
    }
    keepWithTokenPieces.append(contentsOf: pendingPieces)
    return (Trivia(pieces: keepWithTokenPieces), Trivia(pieces: extractingPieces))
  }
}

extension TriviaPiece {
  /// True if the trivia piece is unexpected text.
  fileprivate var isUnexpectedText: Bool {
    switch self {
    case .unexpectedText: return true
    default: return false
    }
  }
}

/// Returns whether the given trivia includes a directive to ignore formatting for the next node.
///
/// - Parameters:
///   - trivia: Leading trivia for a node that the formatter supports ignoring.
///   - isWholeFile: Whether to search for a whole-file ignore directive or per node ignore.
/// - Returns: Whether the trivia contains the specified type of ignore directive.
fileprivate func isFormatterIgnorePresent(inTrivia trivia: Trivia, isWholeFile: Bool) -> Bool {
  func isFormatterIgnore(in commentText: String, prefix: String, suffix: String) -> Bool {
    let trimmed =
      commentText.dropFirst(prefix.count)
        .dropLast(suffix.count)
        .trimmingCharacters(in: .whitespaces)
    let pattern = isWholeFile ? "swift-format-ignore-file" : "swift-format-ignore"
    return trimmed == pattern
  }

  for piece in trivia {
    switch piece {
    case .lineComment(let text):
      if isFormatterIgnore(in: text, prefix: "//", suffix: "") { return true }
      break
    case .blockComment(let text):
      if isFormatterIgnore(in: text, prefix: "/*", suffix: "*/") { return true }
      break
    default:
      break
    }
  }
  return false
}

/// Returns whether the formatter should ignore the given node by printing it without changing the
/// node's internal text representation (i.e. print all text inside of the node as it was in the
/// original source).
///
/// - Note: The caller is responsible for ensuring that the given node is a type of node that can
/// be safely ignored.
///
/// - Parameter node: A node that can be safely ignored.
fileprivate func shouldFormatterIgnore(node: Syntax) -> Bool {
  // Regardless of the level of nesting, if the ignore directive is present on the first token
  // contained within the node then the entire node is eligible for ignoring.
  if let firstTrivia = node.firstToken?.leadingTrivia {
    return isFormatterIgnorePresent(inTrivia: firstTrivia, isWholeFile: false)
  }
  return false
}

/// Returns whether the formatter should ignore the given file by printing it without changing the
/// any if its nodes' internal text representation (i.e. print all text inside of the file as it was
/// in the original source).
///
/// - Parameter file: The root syntax node for a source file.
fileprivate func shouldFormatterIgnore(file: SourceFileSyntax) -> Bool {
  if let firstTrivia = file.firstToken?.leadingTrivia {
    return isFormatterIgnorePresent(inTrivia: firstTrivia, isWholeFile: true)
  }
  return false
}

extension NewlineBehavior {
  static func +(lhs: NewlineBehavior, rhs: NewlineBehavior) -> NewlineBehavior {
    switch (lhs, rhs) {
    case (.elective, _):
      // `rhs` is either also elective or a required newline, which overwrites elective.
      return rhs
    case (_, .elective):
      // `lhs` is either also elective or a required newline, which overwrites elective.
      return lhs

    case (.soft(let lhsCount, let lhsDiscretionary), .soft(let rhsCount, let rhsDiscretionary)):
      let mergedCount: Int
      if lhsDiscretionary && rhsDiscretionary {
        mergedCount = lhsCount + rhsCount
      } else if lhsDiscretionary {
        mergedCount = lhsCount
      } else if rhsDiscretionary {
        mergedCount = rhsCount
      } else {
        mergedCount = max(lhsCount, rhsCount)
      }
      return .soft(count: mergedCount, discretionary: lhsDiscretionary || rhsDiscretionary)

    case (.soft(let softCount, _), .hard(let hardCount)),
      (.hard(let hardCount), .soft(let softCount, _)):
      return .hard(count: max(softCount, hardCount))

    case (.hard(let lhsCount), .hard(let rhsCount)):
      return .hard(count: lhsCount + rhsCount)
    }
  }
}

/// Common protocol implemented by expression syntax types that support calling another expression.
protocol CallingExprSyntaxProtocol: ExprSyntaxProtocol {
  var calledExpression: ExprSyntax { get }
}

extension FunctionCallExprSyntax: CallingExprSyntaxProtocol { }
extension SubscriptExprSyntax: CallingExprSyntaxProtocol { }

extension Syntax {
  func asProtocol(_: CallingExprSyntaxProtocol.Protocol) -> CallingExprSyntaxProtocol? {
    return self.asProtocol(SyntaxProtocol.self) as? CallingExprSyntaxProtocol
  }
  func isProtocol(_: CallingExprSyntaxProtocol.Protocol) -> Bool {
    return self.asProtocol(CallingExprSyntaxProtocol.self) != nil
  }
}

extension ExprSyntax {
  func asProtocol(_: CallingExprSyntaxProtocol.Protocol) -> CallingExprSyntaxProtocol? {
    return Syntax(self).asProtocol(SyntaxProtocol.self) as? CallingExprSyntaxProtocol
  }
  func isProtocol(_: CallingExprSyntaxProtocol.Protocol) -> Bool {
    return self.asProtocol(CallingExprSyntaxProtocol.self) != nil
  }
}