url: speed up WHATWG URL parsing

lib/internal/url.js

@@ -167,41 +167,62 @@ function lazyCryptoRandom() {
   return cryptoRandom;
 }
 
+/**
+ * Copy href and the latest `urlComponents` snapshot into a URLContext.
+ * Property assignment order matches the historical URLContext fields so
+ * `util.inspect(..., { showHidden: true })` stays stable.
+ * @param {object} ctx
+ * @param {string} href
+ */
+function setURLContextFromBinding(ctx, href) {
+  const c = bindingUrl.urlComponents;
+  ctx.href = href;
+  ctx.protocol_end = c[0];
+  ctx.username_end = c[1];
+  ctx.host_start = c[2];
+  ctx.host_end = c[3];
+  ctx.pathname_start = c[5];
+  ctx.search_start = c[6];
+  ctx.hash_start = c[7];
+  ctx.port = c[4];
+  ctx.scheme_type = c[8];
+}
+
 // This class provides the internal state of a URL object. An instance of this
 // class is stored in every URL object and is accessed internally by setters
 // and getters. It roughly corresponds to the concept of a URL record in the
 // URL Standard, with a few differences. It is also the object transported to
 // the C++ binding.
 // Refs: https://url.spec.whatwg.org/#concept-url
+//
+// scheme_type refers to ada::scheme::type:
+//   HTTP = 0, NOT_SPECIAL = 1, HTTPS = 2, WS = 3, FTP = 4, WSS = 5, FILE = 6
 class URLContext {
   // This is the maximum value uint32_t can get.
   // Ada uses uint32_t(-1) for declaring omitted values.
   static #omitted = 4294967295;
 
-  href = '';
-  protocol_end = 0;
-  username_end = 0;
-  host_start = 0;
-  host_end = 0;
-  pathname_start = 0;
-  search_start = 0;
-  hash_start = 0;
-  port = 0;
   /**
-   * Refers to `ada::scheme::type`
-   *
-   * enum type : uint8_t {
-   *   HTTP = 0,
-   *   NOT_SPECIAL = 1,
-   *   HTTPS = 2,
-   *   WS = 3,
-   *   FTP = 4,
-   *   WSS = 5,
-   *   FILE = 6
-   * };
-   * @type {number}
+   * @param {string} [href] Parsed href. When omitted, create an empty context
+   *   (used by `URL.parse` on invalid input). When provided, `bindingUrl.parse`
+   *   / `update` has just written `urlComponents`.
    */
-  scheme_type = 1;
+  constructor(href) {
+    if (href === undefined) {
+      this.href = '';
+      this.protocol_end = 0;
+      this.username_end = 0;
+      this.host_start = 0;
+      this.host_end = 0;
+      this.pathname_start = 0;
+      this.search_start = 0;
+      this.hash_start = 0;
+      this.port = 0;
+      this.scheme_type = 1;
+      return;
+    }
+    setURLContextFromBinding(this, href);
+  }
 
   get hasPort() {
     return this.port !== URLContext.#omitted;
@@ -819,7 +840,7 @@ const kCreateURLFromPosixPathSymbol = Symbol('kCreateURLFromPosixPath');
 const kCreateURLFromWindowsPathSymbol = Symbol('kCreateURLFromWindowsPath');
 
 class URL {
-  #context = new URLContext();
+  #context;
   #searchParams;
   #searchParamsModified;
 
@@ -844,16 +865,16 @@ class URL {
   }
 
   constructor(input, base = undefined, parseSymbol = undefined) {
-    markTransferMode(this, false, false);
-
     if (arguments.length === 0) {
       throw new ERR_MISSING_ARGS('url');
     }
 
     // StringPrototypeToWellFormed is not needed.
-    input = `${input}`;
+    if (typeof input !== 'string') {
+      input = `${input}`;
+    }
 
-    if (base !== undefined) {
+    if (base !== undefined && typeof base !== 'string') {
       base = `${base}`;
     }
 
@@ -868,9 +889,12 @@ class URL {
         bindingUrl.pathToFileURL(input, interpretAsWindowsPath, base) :
         bindingUrl.parse(input, base, raiseException);
     }
-    if (href) {
-      this.#updateContext(href);
-    }
+
+    // Delay context allocation until parse finishes so invalid URLs that
+    // throw do not pay for an unused URLContext. Initialize in one shot
+    // from the binding snapshot instead of writing an empty context first.
+    this.#context = href ? new URLContext(href) : new URLContext();
+    markTransferMode(this, false, false);
   }
 
   static parse(input, base = undefined) {
@@ -939,29 +963,7 @@ class URL {
     const previousSearch = shouldUpdateSearchParams && this.#searchParams &&
       (this.#searchParamsModified ? this.#getSearchFromParams() : this.#getSearchFromContext());
 
-    this.#context.href = href;
-
-    const {
-      0: protocol_end,
-      1: username_end,
-      2: host_start,
-      3: host_end,
-      4: port,
-      5: pathname_start,
-      6: search_start,
-      7: hash_start,
-      8: scheme_type,
-    } = bindingUrl.urlComponents;
-
-    this.#context.protocol_end = protocol_end;
-    this.#context.username_end = username_end;
-    this.#context.host_start = host_start;
-    this.#context.host_end = host_end;
-    this.#context.port = port;
-    this.#context.pathname_start = pathname_start;
-    this.#context.search_start = search_start;
-    this.#context.hash_start = hash_start;
-    this.#context.scheme_type = scheme_type;
+    setURLContextFromBinding(this.#context, href);
 
     if (this.#searchParams) {
       // If the search string has updated, URL becomes the source of truth, and we update URLSearchParams.
@@ -1186,10 +1188,12 @@ class URL {
       throw new ERR_MISSING_ARGS('url');
     }
 
-    url = `${url}`;
+    if (typeof url !== 'string') {
+      url = `${url}`;
+    }
 
     if (base !== undefined) {
-      return bindingUrl.canParse(url, `${base}`);
+      return bindingUrl.canParse(url, typeof base === 'string' ? base : `${base}`);
     }
 
     // It is important to differentiate the canParse call statements

src/node_url.cc

@@ -8,6 +8,7 @@
 #include "node_metadata.h"
 #include "node_process-inl.h"
 #include "path.h"
+#include "simdutf.h"
 #include "util-inl.h"
 #include "v8-fast-api-calls.h"
 #include "v8-local-handle.h"
@@ -33,6 +34,38 @@ using v8::SnapshotCreator;
 using v8::String;
 using v8::Value;
 
+namespace {
+
+// Parse a V8 string as a URL. One-byte ASCII inputs are parsed in place
+// without allocating a UTF-8 copy. `reuse_input` is set when the serialized
+// href is identical to that ASCII input so the caller can return the original
+// V8 string. Non-ASCII inputs are never reused: UTF-8 conversion may replace
+// unpaired surrogates, so the original string may not match href.
+ada::result<ada::url_aggregator> ParseUrlFromV8String(
+    Isolate* isolate,
+    Local<String> input,
+    const ada::url_aggregator* base_url,
+    bool* reuse_input) {
+  {
+    String::ValueView view(isolate, input);
+    if (view.is_one_byte()) {
+      const char* data = reinterpret_cast<const char*>(view.data8());
+      const size_t length = static_cast<size_t>(view.length());
+      if (simdutf::validate_ascii(data, length)) [[likely]] {
+        const std::string_view input_view(data, length);
+        auto out = ada::parse<ada::url_aggregator>(input_view, base_url);
+        *reuse_input = out.has_value() && out->get_href() == input_view;
+        return out;
+      }
+    }
+  }
+  *reuse_input = false;
+  Utf8Value utf8(isolate, input);
+  return ada::parse<ada::url_aggregator>(utf8.ToStringView(), base_url);
+}
+
+}  // namespace
+
 void BindingData::MemoryInfo(MemoryTracker* tracker) const {
   tracker->TrackField("url_components_buffer", url_components_buffer_);
 }
@@ -392,32 +425,51 @@ void BindingData::Parse(const FunctionCallbackInfo<Value>& args) {
   Realm* realm = Realm::GetCurrent(args);
   BindingData* binding_data = realm->GetBindingData<BindingData>();
   Isolate* isolate = realm->isolate();
-  std::optional<std::string> base_{};
+  Local<String> input_string = args[0].As<String>();
 
-  Utf8Value input(isolate, args[0]);
   ada::result<ada::url_aggregator> base;
   ada::url_aggregator* base_pointer = nullptr;
   if (args[1]->IsString()) {
-    base_ = Utf8Value(isolate, args[1]).ToString();
-    base = ada::parse<ada::url_aggregator>(*base_);
-    if (!base && raise_exception) {
-      return ThrowInvalidURL(realm->env(), input.ToStringView(), base_);
-    } else if (!base) {
+    bool unused_reuse = false;
+    base = ParseUrlFromV8String(
+        isolate, args[1].As<String>(), nullptr, &unused_reuse);
+    if (!base) {
+      if (raise_exception) {
+        Utf8Value input(isolate, input_string);
+        Utf8Value base_utf8(isolate, args[1]);
+        return ThrowInvalidURL(
+            realm->env(), input.ToStringView(), base_utf8.ToString());
+      }
       return;
     }
     base_pointer = &base.value();
   }
-  auto out =
-      ada::parse<ada::url_aggregator>(input.ToStringView(), base_pointer);
 
-  if (!out && raise_exception) {
-    return ThrowInvalidURL(realm->env(), input.ToStringView(), base_);
-  } else if (!out) {
+  bool reuse_input = false;
+  auto out =
+      ParseUrlFromV8String(isolate, input_string, base_pointer, &reuse_input);
+  if (!out) {
+    if (raise_exception) {
+      Utf8Value input(isolate, input_string);
+      std::optional<std::string> base_error;
+      if (args[1]->IsString()) {
+        base_error = Utf8Value(isolate, args[1]).ToString();
+      }
+      return ThrowInvalidURL(
+          realm->env(), input.ToStringView(), std::move(base_error));
+    }
     return;
   }
 
   binding_data->UpdateComponents(out->get_components(), out->type);
 
+  // Already-serialized ASCII URLs are the common case. Reuse the input
+  // string instead of allocating an identical V8 string from href.
+  if (reuse_input) {
+    args.GetReturnValue().Set(args[0]);
+    return;
+  }
+
   Local<Value> ret;
   if (ToV8Value(realm->context(), out->get_href(), isolate).ToLocal(&ret))
       [[likely]] {
@@ -439,13 +491,15 @@ void BindingData::Update(const FunctionCallbackInfo<Value>& args) {
     return;
   }
   enum url_update_action action = static_cast<enum url_update_action>(val);
-  Utf8Value input(isolate, args[0].As<String>());
   Utf8Value new_value(isolate, args[2].As<String>());
 
   std::string_view new_value_view = new_value.ToStringView();
   // A serialized URL is not always reparsable: the IDNA encoder can emit a
   // host label that the decoder rejects. Fail the update instead of crashing.
-  auto out = ada::parse<ada::url_aggregator>(input.ToStringView());
+  // Existing hrefs are typically already-serialized ASCII, so parse in place.
+  bool unused_reuse = false;
+  auto out = ParseUrlFromV8String(
+      isolate, args[0].As<String>(), nullptr, &unused_reuse);
   if (!out) {
     return args.GetReturnValue().Set(false);
   }

test/parallel/test-whatwg-url-parse-fast-path.js

@@ -0,0 +1,88 @@
+'use strict';
+
+// Covers the URL constructor parse paths that avoid a UTF-8 copy and/or
+// reuse the input string when it is already a serialized ASCII href.
+
+const { hasIntl } = require('../common');
+const assert = require('assert');
+
+const alreadySerialized = [
+  'https://nodejs.org/en/blog/',
+  'http://nodejs.org:89/docs/latest/api/foo/bar/qua/13949281/0f28b/' +
+    '/5d49/b3020/url.html#test?payload1=true&payload2=false&test=1' +
+    '&benchmark=3&foo=38.38.011.293&bar=1234834910480&test=19299&3992&' +
+    'key=f5c65e1e98fe07e648249ad41e1cfdb0',
+  'https://user:pass@example.com/path?search=1',
+  'file:///foo/bar/test/node.js',
+  'ws://localhost:9229/f46db715-70df-43ad-a359-7f9949f39868',
+];
+
+for (const href of alreadySerialized) {
+  const url = new URL(href);
+  assert.strictEqual(url.href, href);
+  assert.strictEqual(URL.parse(href).href, href);
+  assert.strictEqual(URL.canParse(href), true);
+}
+
+// Special-scheme URLs with an empty path gain a trailing slash.
+{
+  const url = new URL('https://example.com');
+  assert.strictEqual(url.href, 'https://example.com/');
+  assert.strictEqual(url.pathname, '/');
+}
+
+// Dot-segment normalization must still rewrite the path.
+{
+  const url = new URL('https://example.org/./a/../b/./c');
+  assert.strictEqual(url.href, 'https://example.org/b/c');
+  assert.strictEqual(url.pathname, '/b/c');
+}
+
+// Relative resolution against a base URL.
+{
+  const url = new URL('/path?x=1#h', 'https://example.com:8443/base');
+  assert.strictEqual(url.href, 'https://example.com:8443/path?x=1#h');
+  assert.strictEqual(url.host, 'example.com:8443');
+}
+
+// Non-string input is still stringified.
+{
+  const url = new URL({ toString: () => 'https://example.com/from-object' });
+  assert.strictEqual(url.href, 'https://example.com/from-object');
+}
+
+// Invalid input still throws from the constructor and is null from parse().
+{
+  assert.throws(() => new URL('not a url'), {
+    code: 'ERR_INVALID_URL',
+    name: 'TypeError',
+  });
+  assert.strictEqual(URL.parse('not a url'), null);
+  assert.strictEqual(URL.canParse('not a url'), false);
+}
+
+// Unpaired surrogates must not be returned as-is from href.
+{
+  const input = 'https://example.com/\uD800';
+  const url = new URL(input);
+  assert.notStrictEqual(url.href, input);
+  assert.ok(url.href.startsWith('https://example.com/'));
+}
+
+if (hasIntl) {
+  const url = new URL('http://你好你好.在线');
+  assert.ok(url.hostname.startsWith('xn--'));
+  assert.ok(url.href.startsWith('http://xn--'));
+}
+
+// Setters re-parse the existing href; keep component updates correct.
+{
+  const url = new URL('https://example.com/old');
+  url.pathname = '/new';
+  url.search = 'q=1';
+  url.hash = 'frag';
+  assert.strictEqual(url.href, 'https://example.com/new?q=1#frag');
+  assert.strictEqual(url.pathname, '/new');
+  assert.strictEqual(url.search, '?q=1');
+  assert.strictEqual(url.hash, '#frag');
+}