migrate web engine implementation to null-safe Dart #19172

yjbanov · 2020-06-19T21:28:05Z

Description

This upgrades the dart:_engine code to null-safe Dart.

Notable changes:

Apply the new syntax (?, !, etc)
Implicit downcasts become explicit (required by the new language)
Disable dart2js inlining for tests for better stack traces
CSS attributes are no longer nullable => changed to assigning empty string instead of null
Changed html.window.navigator.clipboard?.writeText != null and html.window.navigator.clipboard?.readText != null to just html.window.navigator.clipboard != null because the tear-offs are never null.
Unimplemented non-null API that used to return null now throws UnsupportedError.
PathMetrics iterators throw RangeError instead of returning null (matches mobile behavior).
Refactor compositor/rasterizer.dart and compositor/surface.dart to initialize HtmlViewEmbedder upon construction so that it can be final and non-null.
Throw errors when CanvasKit fails to construct a GL surface, instead of just printing to window.console (this removes many null situations).
Replace List<Conic> with _ConicPair as temporary storage of results (NNBD does not support fixed-capacity lists).
Converted some constructors to factories so that final fields can be immediately initialized to non-null values as opposed to lazy initializing them (e.g. WriteBuffer, SurfaceCanvas)
EngineLineMetrics.withText uses bogus values for unsupported fields
Factories no longer return null (e.g. factory AutofillInfo.fromFrameworkMessage). This is a new language requirement.
Fixes an issue with getBoxesForRange in RTL mode (see paragraph_test.dart)

How to review this PR

This PR touches every file in the dart:_engine. It is not practical for one person to carefully review all of it. Instead, each reviewer is assigned a subset of files (although do feel free to comment on anything). The assignments are tracked in this spreadsheet: https://docs.google.com/spreadsheets/d/1Efar1J6eDARoeDlDI4qXdoldJuxUcMUeLpXpu3gQDMA/edit?pli=1#gid=180643056
The PR does not fix all nullability issues. It is not practical to address them all in a PR of this size. If you see places where something should be non-null but isn't, please file an issue so we remember to fix this in a smaller follow-up PR. The reverse is not true, however. If something has been make non-null where it should be nullable could lead to a bug, so please comment on that.

Related Issues

Fixes flutter/flutter#58811

ferhatb

First batch of comments.

lib/web_ui/build.html.yaml

lib/web_ui/build.canvaskit.yaml

lib/web_ui/lib/src/engine/bitmap_canvas.dart

lib/web_ui/lib/src/engine/canvas_pool.dart

ferhatb

Comments Batch2

lib/web_ui/lib/src/engine/conic.dart

lib/web_ui/lib/src/engine/dom_canvas.dart

lib/web_ui/lib/src/engine/dom_renderer.dart

ferhatb · 2020-06-21T00:11:26Z

lib/web_ui/lib/src/engine/houdini_canvas.dart

@@ -22,7 +22,7 @@ class HoudiniCanvas extends EngineCanvas with SaveElementStackTracking {
  /// where this canvas paints.
  ///
  /// Painting outside the bounds of this rectangle is cropped.
-  final ui.Rect bounds;
+  final ui.Rect? bounds;


why nullable?

Should we kill HoudiniCanvas until we're ready for a proper investigation? Not sure if it's worth maintaining it with no tests and users.

Please go ahead.

lib/web_ui/lib/src/engine/mouse_cursor.dart

lib/web_ui/lib/src/engine/picture.dart

lib/web_ui/lib/src/engine/rrect_renderer.dart

lib/web_ui/lib/src/engine/shadow.dart

mdebbar

This is a great effort! Thank you so much for working on it.

I'm still not done with the review, but wanted to submit the comments I have so far. All my comments are things we can iterate on later, not blockers.

mdebbar · 2020-06-22T03:23:42Z

lib/web_ui/lib/src/engine/browser_detection.dart


 /// Returns the [OperatingSystem] the current browsers works on.
 ///
 /// This is used to implement operating system specific behavior such as
 /// soft keyboards.
-OperatingSystem get operatingSystem {
+OperatingSystem? get operatingSystem {


IIUC, this should never return null. debugOperatingSystemOverride is returned only if it's non-null. And _detectOperatingSystem() returns a non-nullable.

Maybe we need the same trick you did in line 45?

mdebbar · 2020-06-22T03:27:25Z

lib/web_ui/lib/src/engine/browser_detection.dart

-    return debugBrowserEngineOverride;
-  }
-  return _browserEngine ??= _detectBrowserEngine();
+  return debugBrowserEngineOverride ?? (_browserEngine ??= _detectBrowserEngine());


Is this required to make the analyzer happy? I think it's possible to infer from the old code that the result is always non-nullable. Is this a known issue or should we file one?

Yes, this is to satisfy the type inference. It's a known issue, but not easily solvable. Maybe sealed classes will help one day.

mdebbar · 2020-06-22T03:34:06Z

lib/web_ui/lib/src/engine/browser_location.dart

@@ -23,17 +23,17 @@ abstract class LocationStrategy {
  ui.VoidCallback onPopState(html.EventListener fn);

  /// The active path in the browser history.
-  String get path;
+  String? get path;


Is there a reason this is nullable? In the HashLocationStrategy it always returns a string.

mdebbar · 2020-06-22T03:40:56Z

lib/web_ui/lib/src/engine/test_embedding.dart

@@ -60,7 +60,7 @@ class TestLocationStrategy extends LocationStrategy {
  bool get withinAppHistory => _currentEntryIndex >= 0;

  @override
-  void pushState(dynamic state, String title, String url) {
+  void pushState(dynamic state, String title, String? url) {


The nullable url here is probably what's causing multiple other things to be nullable (e.g. String? get path, etc). Let's make it non-nullable?

mdebbar · 2020-06-22T03:41:08Z

lib/web_ui/lib/src/engine/test_embedding.dart

@@ -77,7 +77,7 @@ class TestLocationStrategy extends LocationStrategy {
  }

  @override
-  void replaceState(dynamic state, String title, String url) {
+  void replaceState(dynamic state, String title, String? url) {


mdebbar · 2020-06-22T05:01:00Z

lib/web_ui/lib/src/engine/pointer_binding.dart


-  static void initInstance(html.Element glassPaneElement) {
+  static void initInstance(html.Element? glassPaneElement) {


glassPaneElement should never be null.

mdebbar · 2020-06-22T05:03:55Z

lib/web_ui/lib/src/engine/pointer_binding.dart

  _PointerDataCallback _callback;
-  PointerDataConverter _pointerDataConverter;
+  PointerDataConverter? _pointerDataConverter;


I couldn't find a situation where this can be null. It's always initialized in the constructor.

mdebbar · 2020-06-22T05:07:13Z

lib/web_ui/lib/src/engine/profiler.dart

@@ -71,7 +71,7 @@ class Profiler {
    return Profiler._instance ??= Profiler._();
  }

-  static Profiler get instance {
+  static Profiler? get instance {


This never returns null. If _instance is null, it throws.

mdebbar · 2020-06-22T05:11:57Z

lib/web_ui/lib/src/engine/text/measurement.dart

@@ -147,8 +147,8 @@ class RulerManager {
  /// The returned ruler is marked as hit so there's no need to do that
  /// elsewhere.
  @visibleForTesting
-  ParagraphRuler findOrCreateRuler(ParagraphGeometricStyle style) {
-    ParagraphRuler ruler = _rulers[style];
+  ParagraphRuler? findOrCreateRuler(ParagraphGeometricStyle style) {


In all cases, this function returns a ruler.

mdebbar · 2020-06-22T05:21:41Z

lib/web_ui/lib/src/engine/text/paragraph.dart

+  final String? _ellipsis;
+  final ui.Locale? _locale;
+
+  String? get _effectiveFontFamily {


This is never null.

nturgut · 2020-06-22T17:13:33Z

lib/web_ui/lib/src/engine/text_editing/input_type.dart

@@ -66,7 +66,7 @@ abstract class EngineInputType {
  html.HtmlElement createDomElement() => html.InputElement();

  /// Given a [domElement], set attributes that are specific to this input type.
-  void configureInputMode(html.HtmlElement domElement) {
+  void configureInputMode(html.HtmlElement? domElement) {


When calling this method, dom_element should not be null.

nturgut · 2020-06-22T17:14:17Z

lib/web_ui/lib/src/engine/text_editing/input_type.dart

@@ -75,7 +75,7 @@ abstract class EngineInputType {
    // keyboard shows up.
    if (operatingSystem == OperatingSystem.iOs ||
        operatingSystem == OperatingSystem.android) {
-      domElement.setAttribute('inputmode', inputmodeAttribute);
+      domElement!.setAttribute('inputmode', inputmodeAttribute!);


dom_element should not be null

nturgut · 2020-06-22T17:22:04Z

lib/web_ui/lib/src/engine/text_editing/input_type.dart

@@ -14,7 +14,7 @@ part of engine;
 abstract class EngineInputType {
  const EngineInputType();

-  static EngineInputType fromName(String name) {
+  static EngineInputType fromName(String? name) {


name shouldn't be null. It always get defaulted to TextInputType.text on the framework side.

lib/web_ui/lib/src/engine/surface/platform_view.dart

yjbanov · 2020-06-23T22:35:10Z

I think responded to all the comments (at least those I was able to find). PTAL! This is no longer a draft. While I'm still testing manually, this is good to go.

nturgut · 2020-06-24T00:24:20Z

lib/web_ui/lib/src/engine/semantics/semantics_helper.dart

@@ -49,7 +49,7 @@ class SemanticsHelper {
    return _semanticsEnabler.shouldEnableSemantics(event);
  }

-  html.Element prepareAccesibilityPlaceholder() {
+  html.Element? prepareAccesibilityPlaceholder() {


prepareAccesibilityPlaceholder always returns a value so this is never null.

nturgut · 2020-06-24T00:25:37Z

lib/web_ui/lib/src/engine/semantics/semantics_helper.dart

@@ -189,12 +189,12 @@ class DesktopSemanticsEnabler extends SemanticsEnabler {
  }

  @override
-  html.Element prepareAccesibilityPlaceholder() {
-    _semanticsPlaceholder = html.Element.tag('flt-semantics-placeholder');
+  html.Element? prepareAccesibilityPlaceholder() {


this is never null (related to the previous comment)

nturgut · 2020-06-24T00:31:39Z