Blazing fast HTML escaping.

Sources/TemplateKit/Utilities/HTMLEscape.swift

@@ -1,13 +1,98 @@
 import Foundation
 
 extension String {
+    /// Lookup table mapping the characters that need escaping to their escaped representation.
+    private static let htmlEscapeMap: [UInt8: String] = [
+        UInt8(ascii: "&"): "&",
+        UInt8(ascii: "\""): """,
+        UInt8(ascii: "'"): "'",
+        UInt8(ascii: "<"): "&lt;",
+        UInt8(ascii: ">"): "&gt;"]
+
+    /// Stores an inline byte array to avoid the memory overhead of using `[UInt8]`.
+    private struct InlineByteArray {
+        private(set) var eightBytes: Int64 = 0
+        private(set) var count: Int
+
+        init(bytes: [UInt8]) {
+            assert(bytes.count <= 8)
+            self.count = bytes.count
+            let selfPointer = UnsafeMutableRawPointer(&self)
+            bytes.withUnsafeBytes { bytesPointer in
+                selfPointer.copyMemory(from: bytesPointer.baseAddress!, byteCount: bytes.count)
+            }
+        }
+    }
+
+    private struct SixteenBytes {
+        let firstEight: Int64 = 0
+        let secondEight: Int64 = 0
+
+        init() { }
+    }
+
+    /// Same as `htmlEscapeMap`, but stored as an array indexed from 0 to 255 to avoid dictionary lookups.
+    /// In addition, we store `InlineByteArray`s instead of `String`s in order to avoid memory management overhead.
+    /// If no escaping is required for a character, the character itself is stored.
+    /// Using an array-typed lookup table is much faster than a dictionary-typed one or `if`-based branching.
+    private static let htmlEscapeMapASCIIByteArray: [InlineByteArray] = (UInt8(0)...UInt8(255)).map { byte in
+        if let escaped = String.htmlEscapeMap[byte] {
+            return InlineByteArray(bytes: Array(escaped.utf8))
+        } else {
+            return InlineByteArray(bytes: [byte])
+        }
+    }
+
     /// Escapes HTML entities in a `String`.
-    internal func htmlEscaped() -> String {
-        /// FIXME: performance
-        return replacingOccurrences(of: "&", with: "&amp;")
-            .replacingOccurrences(of: "\"", with: "&quot;")
-            .replacingOccurrences(of: "'", with: "&#39;")
-            .replacingOccurrences(of: "<", with: "&lt;")
-            .replacingOccurrences(of: ">", with: "&gt;")
+    public func htmlEscaped() -> String {
+        var expectedLength = 0
+        // Using `withUnsafeBufferPointer` is minimally faster than calling `String.htmlEscapeMapASCIILengths[Int(character)]` for each character.
+        String.htmlEscapeMapASCIIByteArray.withUnsafeBufferPointer { lengths in
+            for character in self.utf8 {
+                expectedLength += lengths[Int(character)].count
+            }
+        }
+
+        guard expectedLength != self.utf8.count else {
+            // Shortcut: no replacements necessary; skip them altogether.
+            return self
+        }
+
+        func writeEscapedString(_ resultBytes: UnsafeMutableRawPointer) -> Void {
+            var raw = resultBytes
+            let end = raw + expectedLength
+            for character in self.utf8 {
+                var escaped = String.htmlEscapeMapASCIIByteArray[Int(character)]
+                assert(raw + escaped.count <= end)
+                raw.copyMemory(from: &escaped, byteCount: escaped.count)
+                raw += escaped.count
+            }
+        }
+
+        if expectedLength <= 15 {
+            // Avoid the `Array<UInt8>` heap allocation for strings consisting
+            // of at most 15 UTF-8 code units, where `String`'s small string
+            // optimization avoids a memory allocation.
+            // This provides another ~5x speedup compared to the "slow" path below.
+            // Note: This might be slightly less efficient (but still correct,
+            // and still faster than the slow path) for non-ASCII Strings on Swift 4.2.
+            var resultData = SixteenBytes()
+            let resultBytes = UnsafeMutableRawPointer(&resultData)
+            writeEscapedString(resultBytes)
+
+            // Note: Byte 16 should always be zero to make sure the string is null-terminated.
+            // This is ensured by `raw + escaped.count <= end = expectedLength <= 15` above.
+            return String(cString: resultBytes.assumingMemoryBound(to: UInt8.self))
+        } else {
+            var resultData = Array<UInt8>(repeating: 0, count: expectedLength)
+            resultData.withUnsafeMutableBytes {
+                writeEscapedString($0.baseAddress!)
+            }
+
+            // TODO: It might be possible to gain further improvements
+            // by re-using the byte array allocated by `resultData`
+            // to avoid copying the string's bytes here.
+            return String(bytes: resultData, encoding: .utf8)!  // Guaranteed to succeed.
+        }
     }
 }

Tests/LinuxMain.swift

@@ -1,6 +1,8 @@
 import XCTest
-@testable import TemplateKitTests
 
-XCTMain([
-    testCase(TemplateDataEncoderTests.allTests),
-])
+import TemplateKitTests
+
+var tests = [XCTestCaseEntry]()
+tests += TemplateKitTests.__allTests()
+
+XCTMain(tests)

Tests/TemplateKitTests/HTMLEscapeTests.swift

@@ -0,0 +1,91 @@
+@testable import TemplateKit
+import XCTest
+
+class HTMLEscapeTests: XCTestCase {
+    func testCorrectness() {
+        XCTAssertEqual("".htmlEscaped(), "")
+        XCTAssertEqual("abcdef".htmlEscaped(), "abcdef")
+        XCTAssertEqual("abc&<>\"'".htmlEscaped(), "abc&amp;&lt;&gt;&quot;&#39;")
+        XCTAssertEqual("abc&".htmlEscaped(), "abc&amp;")
+    }
+
+    func testShortStringNoReplacements() {
+        let string = "abcde12345"
+        measure {
+            for _ in 0..<10_000_000 {
+                _ = string.htmlEscaped()
+            }
+        }
+    }
+
+    func testShortStringWithReplacements() {
+        // The result should still fit into 15 bytes to hit the in-place String storage optimization.
+        let string = "<abcdef>"
+        measure {
+            for _ in 0..<1_000_000 {
+                _ = string.htmlEscaped()
+            }
+        }
+    }
+
+    static let mediumStringNoReplacements: String = {
+        let lowercase = Array(UInt8(ascii: "a")...UInt8(ascii: "z"))
+        let digits = Array(UInt8(ascii: "0")...UInt8(ascii: "9"))
+        let uppercase = Array(UInt8(ascii: "A")...UInt8(ascii: "Z"))
+
+        return String(bytes: lowercase + digits + uppercase, encoding: .utf8)!
+    }()
+
+    func testMediumStringNoReplacements() {
+        measure {
+            for _ in 0..<2_000_000 {
+                _ = HTMLEscapeTests.mediumStringNoReplacements.htmlEscaped()
+            }
+        }
+    }
+
+    static let mediumStringWithReplacements: String = {
+        let lowercase = Array(UInt8(ascii: "a")...UInt8(ascii: "z"))
+        let digits = Array(UInt8(ascii: "0")...UInt8(ascii: "9"))
+        let uppercase = Array(UInt8(ascii: "A")...UInt8(ascii: "Z"))
+        let allCharacters = [[UInt8(ascii: "&")], lowercase, [UInt8(ascii: "\"")], digits, [UInt8(ascii: "'")], uppercase, [UInt8(ascii: "<")], [UInt8(ascii: ">")]]
+            .flatMap { $0 }
+
+        return String(bytes: allCharacters, encoding: .utf8)!
+    }()
+
+    func testMediumStringWithReplacements() {
+        measure {
+            for _ in 0..<500_000 {
+                _ = HTMLEscapeTests.mediumStringWithReplacements.htmlEscaped()
+            }
+        }
+    }
+
+    func testMediumStringWithOnlyReplacements() {
+        let string = Array(repeating: "&<>\"'", count: 10).joined(separator: "")
+        measure {
+            for _ in 0..<500_000 {
+                _ = string.htmlEscaped()
+            }
+        }
+    }
+
+    func testLongStringNoReplacements() {
+        let longString = Array(repeating: HTMLEscapeTests.mediumStringNoReplacements, count: 20).joined(separator: "")
+        measure {
+            for _ in 0..<200_000 {
+                _ = longString.htmlEscaped()
+            }
+        }
+    }
+
+    func testLongStringWithReplacements() {
+        let longString = Array(repeating: HTMLEscapeTests.mediumStringWithReplacements, count: 20).joined(separator: "")
+        measure {
+            for _ in 0..<50_000 {
+                _ = longString.htmlEscaped()
+            }
+        }
+    }
+}

Tests/TemplateKitTests/XCTestManifests.swift

@@ -0,0 +1,52 @@
+#if !canImport(ObjectiveC)
+import XCTest
+
+extension HTMLEscapeTests {
+    // DO NOT MODIFY: This is autogenerated, use:
+    //   `swift test --generate-linuxmain`
+    // to regenerate.
+    static let __allTests__HTMLEscapeTests = [
+        ("testCorrectness", testCorrectness),
+        ("testLongStringNoReplacements", testLongStringNoReplacements),
+        ("testLongStringWithReplacements", testLongStringWithReplacements),
+        ("testMediumStringNoReplacements", testMediumStringNoReplacements),
+        ("testMediumStringWithOnlyReplacements", testMediumStringWithOnlyReplacements),
+        ("testMediumStringWithReplacements", testMediumStringWithReplacements),
+        ("testShortStringNoReplacements", testShortStringNoReplacements),
+        ("testShortStringWithReplacements", testShortStringWithReplacements),
+    ]
+}
+
+extension TemplateDataEncoderTests {
+    // DO NOT MODIFY: This is autogenerated, use:
+    //   `swift test --generate-linuxmain`
+    // to regenerate.
+    static let __allTests__TemplateDataEncoderTests = [
+        ("testArray", testArray),
+        ("testComplexEncodable", testComplexEncodable),
+        ("testDictionary", testDictionary),
+        ("testDouble", testDouble),
+        ("testEncodable", testEncodable),
+        ("testEncodeSuperCustomImplementation", testEncodeSuperCustomImplementation),
+        ("testEncodeSuperCustomImplementationWithSuperEncoder1", testEncodeSuperCustomImplementationWithSuperEncoder1),
+        ("testEncodeSuperCustomImplementationWithSuperEncoder2", testEncodeSuperCustomImplementationWithSuperEncoder2),
+        ("testEncodeSuperDefaultImplementation", testEncodeSuperDefaultImplementation),
+        ("testEncodingPerformanceExampleModel", testEncodingPerformanceExampleModel),
+        ("testEncodingPerformanceExampleModelJSONBaseline", testEncodingPerformanceExampleModelJSONBaseline),
+        ("testGH10", testGH10),
+        ("testGH20", testGH20),
+        ("testNestedArray", testNestedArray),
+        ("testNestedDictionary", testNestedDictionary),
+        ("testNestedEncodable", testNestedEncodable),
+        ("testString", testString),
+        ("testTemplabeByteScannerPeak", testTemplabeByteScannerPeak),
+    ]
+}
+
+public func __allTests() -> [XCTestCaseEntry] {
+    return [
+        testCase(HTMLEscapeTests.__allTests__HTMLEscapeTests),
+        testCase(TemplateDataEncoderTests.__allTests__TemplateDataEncoderTests),
+    ]
+}
+#endif