Merge pull request #1708 from ahoppen/ahoppen/memory-corruption

Fix two memory corruption bugs when modifying SwiftSyntax nodes

Author: ahoppen

CodeGeneration/Sources/generate-swiftsyntax/templates/swiftparser/ParserEntryFile.swift

@@ -63,9 +63,16 @@ let parserEntryFile = SourceFileSyntax(leadingTrivia: copyrightHeader) {
         """
         extension \(node.kind.syntaxType): SyntaxParseable {
           public static func parse(from parser: inout Parser) -> Self {
+            // Keep the parser alive so that the arena in which `raw` is allocated
+            // doesn’t get deallocated before we have a chance to create a syntax node
+            // from it. We can’t use `parser.arena` as the parameter to
+            // `Syntax(raw:arena:)` because the node might have been re-used during an
+            // incremental parse and would then live in a different arena than
+            // `parser.arena`.
+            defer { withExtendedLifetime(parser) {} }
             let node = parser.\(raw: parserFunction)()
             let raw = RawSyntax(parser.parseRemainder(into: node))
-            return Syntax(raw: raw).cast(Self.self)
+            return Syntax(raw: raw, rawNodeArena: raw.arena).cast(Self.self)
           }
         }
         """

CodeGeneration/Sources/generate-swiftsyntax/templates/swiftsyntax/SyntaxNodesFile.swift

@@ -129,7 +129,7 @@ func syntaxNode(emitKind: SyntaxNodeKind) -> SourceFileSyntax {
                   """
                 )
               }
-              StmtSyntax("return SyntaxData.forRoot(raw)")
+              StmtSyntax("return SyntaxData.forRoot(raw, rawNodeArena: arena)")
             }
           )
 
@@ -173,7 +173,7 @@ func syntaxNode(emitKind: SyntaxNodeKind) -> SourceFileSyntax {
             AccessorDeclSyntax(
               """
               set(value) {
-                self = \(node.kind.syntaxType)(data.replacingChild(at: \(raw: index), with: value\(raw: child.isOptional ? "?" : "").raw, arena: SyntaxArena()))
+                self = \(node.kind.syntaxType)(data.replacingChild(at: \(raw: index), with: value\(raw: child.isOptional ? "?" : "").data, arena: SyntaxArena()))
               }
               """
             )
@@ -207,7 +207,7 @@ func syntaxNode(emitKind: SyntaxNodeKind) -> SourceFileSyntax {
                   collection = RawSyntax.makeLayout(kind: SyntaxKind.\(raw: childNode.varOrCaseName),
                                                     from: [element.raw], arena: arena)
                 }
-                let newData = data.replacingChild(at: \(raw: index), with: collection, arena: arena)
+                let newData = data.replacingChild(at: \(raw: index), with: collection, rawNodeArena: arena, allocationArena: arena)
                 return \(node.kind.syntaxType)(newData)
               }
               """

CodeGeneration/Sources/generate-swiftsyntax/templates/swiftsyntax/SyntaxRewriterFile.swift

@@ -325,8 +325,8 @@ let syntaxRewriterFile = SourceFileSyntax(leadingTrivia: copyrightHeader) {
           let arena = SyntaxArena()
           let newRaw = node.raw.layoutView!.replacingLayout(with: Array(newLayout), arena: arena)
           // 'withExtendedLifetime' to keep 'SyntaxArena's of them alive until here.
-          return withExtendedLifetime((arena, rewrittens)) {
-            Syntax(raw: newRaw).cast(SyntaxType.self)
+          return withExtendedLifetime(rewrittens) {
+            Syntax(raw: newRaw, rawNodeArena: arena).cast(SyntaxType.self)
           }
         } else {
           // No child node was rewritten. So no need to change this node as well.
@@ -342,8 +342,9 @@ let syntaxRewriterFile = SourceFileSyntax(leadingTrivia: copyrightHeader) {
       /// a parent if `node` had one.
       public func rewrite(_ node: Syntax) -> Syntax {
         let rewritten = self.visit(node)
-        let arena = SyntaxArena()
-        return Syntax(node.data.replacingSelf(rewritten.raw, arena: arena))
+        return withExtendedLifetime(rewritten) {
+          return Syntax(node.data.replacingSelf(rewritten.raw, rawNodeArena: rewritten.raw.arena, allocationArena: SyntaxArena()))
+        }
       }
       """
     )

Sources/SwiftParser/generated/Parser+Entry.swift

@@ -43,129 +43,289 @@ public protocol SyntaxParseable: SyntaxProtocol {
 
 extension AccessorDeclSyntax: SyntaxParseable {
   public static func parse(from parser: inout Parser) -> Self {
+    // Keep the parser alive so that the arena in which `raw` is allocated
+    // doesn’t get deallocated before we have a chance to create a syntax node
+    // from it. We can’t use `parser.arena` as the parameter to
+    // `Syntax(raw:arena:)` because the node might have been re-used during an
+    // incremental parse and would then live in a different arena than
+    // `parser.arena`.
+    defer {
+      withExtendedLifetime(parser) {
+      }
+    }
     let node = parser.parseAccessorDecl()
     let raw = RawSyntax(parser.parseRemainder(into: node))
-    return Syntax(raw: raw).cast(Self.self)
+    return Syntax(raw: raw, rawNodeArena: raw.arena).cast(Self.self)
   }
 }
 
 extension AttributeSyntax: SyntaxParseable {
   public static func parse(from parser: inout Parser) -> Self {
+    // Keep the parser alive so that the arena in which `raw` is allocated
+    // doesn’t get deallocated before we have a chance to create a syntax node
+    // from it. We can’t use `parser.arena` as the parameter to
+    // `Syntax(raw:arena:)` because the node might have been re-used during an
+    // incremental parse and would then live in a different arena than
+    // `parser.arena`.
+    defer {
+      withExtendedLifetime(parser) {
+      }
+    }
     let node = parser.parseAttribute()
     let raw = RawSyntax(parser.parseRemainder(into: node))
-    return Syntax(raw: raw).cast(Self.self)
+    return Syntax(raw: raw, rawNodeArena: raw.arena).cast(Self.self)
   }
 }
 
 extension CatchClauseSyntax: SyntaxParseable {
   public static func parse(from parser: inout Parser) -> Self {
+    // Keep the parser alive so that the arena in which `raw` is allocated
+    // doesn’t get deallocated before we have a chance to create a syntax node
+    // from it. We can’t use `parser.arena` as the parameter to
+    // `Syntax(raw:arena:)` because the node might have been re-used during an
+    // incremental parse and would then live in a different arena than
+    // `parser.arena`.
+    defer {
+      withExtendedLifetime(parser) {
+      }
+    }
     let node = parser.parseCatchClause()
     let raw = RawSyntax(parser.parseRemainder(into: node))
-    return Syntax(raw: raw).cast(Self.self)
+    return Syntax(raw: raw, rawNodeArena: raw.arena).cast(Self.self)
   }
 }
 
 extension ClosureParameterSyntax: SyntaxParseable {
   public static func parse(from parser: inout Parser) -> Self {
+    // Keep the parser alive so that the arena in which `raw` is allocated
+    // doesn’t get deallocated before we have a chance to create a syntax node
+    // from it. We can’t use `parser.arena` as the parameter to
+    // `Syntax(raw:arena:)` because the node might have been re-used during an
+    // incremental parse and would then live in a different arena than
+    // `parser.arena`.
+    defer {
+      withExtendedLifetime(parser) {
+      }
+    }
     let node = parser.parseClosureParameter()
     let raw = RawSyntax(parser.parseRemainder(into: node))
-    return Syntax(raw: raw).cast(Self.self)
+    return Syntax(raw: raw, rawNodeArena: raw.arena).cast(Self.self)
   }
 }
 
 extension CodeBlockItemSyntax: SyntaxParseable {
   public static func parse(from parser: inout Parser) -> Self {
+    // Keep the parser alive so that the arena in which `raw` is allocated
+    // doesn’t get deallocated before we have a chance to create a syntax node
+    // from it. We can’t use `parser.arena` as the parameter to
+    // `Syntax(raw:arena:)` because the node might have been re-used during an
+    // incremental parse and would then live in a different arena than
+    // `parser.arena`.
+    defer {
+      withExtendedLifetime(parser) {
+      }
+    }
     let node = parser.parseNonOptionalCodeBlockItem()
     let raw = RawSyntax(parser.parseRemainder(into: node))
-    return Syntax(raw: raw).cast(Self.self)
+    return Syntax(raw: raw, rawNodeArena: raw.arena).cast(Self.self)
   }
 }
 
 extension DeclSyntax: SyntaxParseable {
   public static func parse(from parser: inout Parser) -> Self {
+    // Keep the parser alive so that the arena in which `raw` is allocated
+    // doesn’t get deallocated before we have a chance to create a syntax node
+    // from it. We can’t use `parser.arena` as the parameter to
+    // `Syntax(raw:arena:)` because the node might have been re-used during an
+    // incremental parse and would then live in a different arena than
+    // `parser.arena`.
+    defer {
+      withExtendedLifetime(parser) {
+      }
+    }
     let node = parser.parseDeclaration()
     let raw = RawSyntax(parser.parseRemainder(into: node))
-    return Syntax(raw: raw).cast(Self.self)
+    return Syntax(raw: raw, rawNodeArena: raw.arena).cast(Self.self)
   }
 }
 
 extension EnumCaseParameterSyntax: SyntaxParseable {
   public static func parse(from parser: inout Parser) -> Self {
+    // Keep the parser alive so that the arena in which `raw` is allocated
+    // doesn’t get deallocated before we have a chance to create a syntax node
+    // from it. We can’t use `parser.arena` as the parameter to
+    // `Syntax(raw:arena:)` because the node might have been re-used during an
+    // incremental parse and would then live in a different arena than
+    // `parser.arena`.
+    defer {
+      withExtendedLifetime(parser) {
+      }
+    }
     let node = parser.parseEnumCaseParameter()
     let raw = RawSyntax(parser.parseRemainder(into: node))
-    return Syntax(raw: raw).cast(Self.self)
+    return Syntax(raw: raw, rawNodeArena: raw.arena).cast(Self.self)
   }
 }
 
 extension ExprSyntax: SyntaxParseable {
   public static func parse(from parser: inout Parser) -> Self {
+    // Keep the parser alive so that the arena in which `raw` is allocated
+    // doesn’t get deallocated before we have a chance to create a syntax node
+    // from it. We can’t use `parser.arena` as the parameter to
+    // `Syntax(raw:arena:)` because the node might have been re-used during an
+    // incremental parse and would then live in a different arena than
+    // `parser.arena`.
+    defer {
+      withExtendedLifetime(parser) {
+      }
+    }
     let node = parser.parseExpression()
     let raw = RawSyntax(parser.parseRemainder(into: node))
-    return Syntax(raw: raw).cast(Self.self)
+    return Syntax(raw: raw, rawNodeArena: raw.arena).cast(Self.self)
   }
 }
 
 extension FunctionParameterSyntax: SyntaxParseable {
   public static func parse(from parser: inout Parser) -> Self {
+    // Keep the parser alive so that the arena in which `raw` is allocated
+    // doesn’t get deallocated before we have a chance to create a syntax node
+    // from it. We can’t use `parser.arena` as the parameter to
+    // `Syntax(raw:arena:)` because the node might have been re-used during an
+    // incremental parse and would then live in a different arena than
+    // `parser.arena`.
+    defer {
+      withExtendedLifetime(parser) {
+      }
+    }
     let node = parser.parseFunctionParameter()
     let raw = RawSyntax(parser.parseRemainder(into: node))
-    return Syntax(raw: raw).cast(Self.self)
+    return Syntax(raw: raw, rawNodeArena: raw.arena).cast(Self.self)
   }
 }
 
 extension GenericParameterClauseSyntax: SyntaxParseable {
   public static func parse(from parser: inout Parser) -> Self {
+    // Keep the parser alive so that the arena in which `raw` is allocated
+    // doesn’t get deallocated before we have a chance to create a syntax node
+    // from it. We can’t use `parser.arena` as the parameter to
+    // `Syntax(raw:arena:)` because the node might have been re-used during an
+    // incremental parse and would then live in a different arena than
+    // `parser.arena`.
+    defer {
+      withExtendedLifetime(parser) {
+      }
+    }
     let node = parser.parseGenericParameters()
     let raw = RawSyntax(parser.parseRemainder(into: node))
-    return Syntax(raw: raw).cast(Self.self)
+    return Syntax(raw: raw, rawNodeArena: raw.arena).cast(Self.self)
   }
 }
 
 extension MemberDeclBlockSyntax: SyntaxParseable {
   public static func parse(from parser: inout Parser) -> Self {
+    // Keep the parser alive so that the arena in which `raw` is allocated
+    // doesn’t get deallocated before we have a chance to create a syntax node
+    // from it. We can’t use `parser.arena` as the parameter to
+    // `Syntax(raw:arena:)` because the node might have been re-used during an
+    // incremental parse and would then live in a different arena than
+    // `parser.arena`.
+    defer {
+      withExtendedLifetime(parser) {
+      }
+    }
     let node = parser.parseMemberDeclList()
     let raw = RawSyntax(parser.parseRemainder(into: node))
-    return Syntax(raw: raw).cast(Self.self)
+    return Syntax(raw: raw, rawNodeArena: raw.arena).cast(Self.self)
   }
 }
 
 extension PatternSyntax: SyntaxParseable {
   public static func parse(from parser: inout Parser) -> Self {
+    // Keep the parser alive so that the arena in which `raw` is allocated
+    // doesn’t get deallocated before we have a chance to create a syntax node
+    // from it. We can’t use `parser.arena` as the parameter to
+    // `Syntax(raw:arena:)` because the node might have been re-used during an
+    // incremental parse and would then live in a different arena than
+    // `parser.arena`.
+    defer {
+      withExtendedLifetime(parser) {
+      }
+    }
     let node = parser.parsePattern()
     let raw = RawSyntax(parser.parseRemainder(into: node))
-    return Syntax(raw: raw).cast(Self.self)
+    return Syntax(raw: raw, rawNodeArena: raw.arena).cast(Self.self)
   }
 }
 
 extension SourceFileSyntax: SyntaxParseable {
   public static func parse(from parser: inout Parser) -> Self {
+    // Keep the parser alive so that the arena in which `raw` is allocated
+    // doesn’t get deallocated before we have a chance to create a syntax node
+    // from it. We can’t use `parser.arena` as the parameter to
+    // `Syntax(raw:arena:)` because the node might have been re-used during an
+    // incremental parse and would then live in a different arena than
+    // `parser.arena`.
+    defer {
+      withExtendedLifetime(parser) {
+      }
+    }
     let node = parser.parseSourceFile()
     let raw = RawSyntax(parser.parseRemainder(into: node))
-    return Syntax(raw: raw).cast(Self.self)
+    return Syntax(raw: raw, rawNodeArena: raw.arena).cast(Self.self)
   }
 }
 
 extension StmtSyntax: SyntaxParseable {
   public static func parse(from parser: inout Parser) -> Self {
+    // Keep the parser alive so that the arena in which `raw` is allocated
+    // doesn’t get deallocated before we have a chance to create a syntax node
+    // from it. We can’t use `parser.arena` as the parameter to
+    // `Syntax(raw:arena:)` because the node might have been re-used during an
+    // incremental parse and would then live in a different arena than
+    // `parser.arena`.
+    defer {
+      withExtendedLifetime(parser) {
+      }
+    }
     let node = parser.parseStatement()
     let raw = RawSyntax(parser.parseRemainder(into: node))
-    return Syntax(raw: raw).cast(Self.self)
+    return Syntax(raw: raw, rawNodeArena: raw.arena).cast(Self.self)
   }
 }
 
 extension SwitchCaseSyntax: SyntaxParseable {
   public static func parse(from parser: inout Parser) -> Self {
+    // Keep the parser alive so that the arena in which `raw` is allocated
+    // doesn’t get deallocated before we have a chance to create a syntax node
+    // from it. We can’t use `parser.arena` as the parameter to
+    // `Syntax(raw:arena:)` because the node might have been re-used during an
+    // incremental parse and would then live in a different arena than
+    // `parser.arena`.
+    defer {
+      withExtendedLifetime(parser) {
+      }
+    }
     let node = parser.parseSwitchCase()
     let raw = RawSyntax(parser.parseRemainder(into: node))
-    return Syntax(raw: raw).cast(Self.self)
+    return Syntax(raw: raw, rawNodeArena: raw.arena).cast(Self.self)
   }
 }
 
 extension TypeSyntax: SyntaxParseable {
   public static func parse(from parser: inout Parser) -> Self {
+    // Keep the parser alive so that the arena in which `raw` is allocated
+    // doesn’t get deallocated before we have a chance to create a syntax node
+    // from it. We can’t use `parser.arena` as the parameter to
+    // `Syntax(raw:arena:)` because the node might have been re-used during an
+    // incremental parse and would then live in a different arena than
+    // `parser.arena`.
+    defer {
+      withExtendedLifetime(parser) {
+      }
+    }
     let node = parser.parseType()
     let raw = RawSyntax(parser.parseRemainder(into: node))
-    return Syntax(raw: raw).cast(Self.self)
+    return Syntax(raw: raw, rawNodeArena: raw.arena).cast(Self.self)
   }
 }
 

Sources/SwiftSyntax/Raw/RawSyntax.swift

@@ -209,7 +209,8 @@ public struct RawSyntax {
     rawData.arenaReference
   }
 
-  internal var arena: SyntaxArena {
+  @_spi(RawSyntax)
+  public var arena: SyntaxArena {
     rawData.arenaReference.value
   }
 

Sources/SwiftSyntax/Raw/RawSyntaxLayoutView.swift

@@ -54,6 +54,8 @@ public struct RawSyntaxLayoutView {
   }
 
   /// Creates a new node of the same kind but with children replaced by `elements`.
+  ///
+  /// The newly created syntax node is allocated in `arena`.
   @_spi(RawSyntax)
   public func replacingLayout(
     with elements: some Collection<RawSyntax?>,