Bug 1506842 - Always restyle / repaint when a visited query finishes. r=dholbert

Differential Revision: https://phabricator.services.mozilla.com/D50810

--HG--
extra : moz-landing-system : lando

dom/base/Link.cpp

@@ -10,6 +10,8 @@
 #include "mozilla/MemoryReporting.h"
 #include "mozilla/dom/Element.h"
 #include "mozilla/IHistory.h"
+#include "mozilla/StaticPrefs_layout.h"
+#include "nsLayoutUtils.h"
 #include "nsIURL.h"
 #include "nsIURIMutator.h"
 #include "nsISizeOf.h"
@@ -109,7 +111,12 @@ void Link::VisitedQueryFinished(bool aVisited) {
   // Tell the element to update its visited state
   mElement->UpdateState(true);
 
-  // FIXME(emilio, bug 1506842): Repaint here unconditionally.
+  if (StaticPrefs::layout_css_always_repaint_on_unvisited()) {
+    // Even if the state didn't actually change, we need to repaint in order for
+    // the visited state not to be observable.
+    nsLayoutUtils::PostRestyleEvent(GetElement(), RestyleHint::RestyleSubtree(),
+                                    nsChangeHint_RepaintFrame);
+  }
 }
 
 EventStates Link::LinkState() const {

layout/base/RestyleManager.cpp

@@ -3205,13 +3205,25 @@ void RestyleManager::ContentStateChanged(nsIContent* aContent,
 
   const EventStates kVisitedAndUnvisited =
       NS_EVENT_STATE_VISITED | NS_EVENT_STATE_UNVISITED;
-  // NOTE: We want to return ASAP for visitedness changes, but we don't want to
+
-  // mess up the situation where the element became a link or stopped being one.
+  // When visited links are disabled, they cannot influence style for obvious
-  if (aChangedBits.HasAllStates(kVisitedAndUnvisited) &&
+  // reasons.
-      !Gecko_VisitedStylesEnabled(element.OwnerDoc())) {
+  //
-    aChangedBits &= ~kVisitedAndUnvisited;
+  // When layout.css.always-repaint-on-unvisited is true, we'll restyle when the
-    if (aChangedBits.IsEmpty()) {
+  // relevant visited query finishes, regardless of the style (see
-      return;
+  // Link::VisitedQueryFinished). So there's no need to do anything as a result
+  // of this state change just yet.
+  //
+  // Note that this check checks for _both_ bits: This is only true when visited
+  // changes to unvisited or vice-versa, but not when we start or stop being a
+  // link itself.
+  if (aChangedBits.HasAllStates(kVisitedAndUnvisited)) {
+    if (!Gecko_VisitedStylesEnabled(element.OwnerDoc()) ||
+        StaticPrefs::layout_css_always_repaint_on_unvisited()) {
+      aChangedBits &= ~kVisitedAndUnvisited;
+      if (aChangedBits.IsEmpty()) {
+        return;
+      }
     }
   }
 

modules/libpref/init/StaticPrefList.yaml

@@ -4720,6 +4720,12 @@
   value: @IS_NIGHTLY_BUILD@
   mirror: always
 
+# Whether we always restyle / repaint as a result of a visited query
+- name: layout.css.always-repaint-on-unvisited
+  type: RelaxedAtomicBool
+  value: false
+  mirror: always
+
 # Make `zoom` a `transform` + `transform-origin` alias.
 - name: layout.css.zoom-transform-hack.enabled
   type: RelaxedAtomicBool

servo/components/style/invalidation/element/state_and_attributes.rs

@@ -158,6 +158,14 @@ where
         // If we the visited state changed, we force a restyle here. Matching
         // doesn't depend on the actual visited state at all, so we can't look
         // at matching results to decide what to do for this case.
+        //
+        // TODO(emilio): This should be contains(), to avoid doing subtree
+        // restyles when adding or removing an href attribute, but invalidation
+        // for that case is broken right now (bug 1591987).
+        //
+        // This piece of code should be removed when
+        // layout.css.always-repaint-on-unvisited is true, since we cannot get
+        // into this situation in that case.
         if state_changes.intersects(ElementState::IN_VISITED_OR_UNVISITED_STATE) {
             trace!(" > visitedness change, force subtree restyle");
             // We can't just return here because there may also be attribute