Bug 1307853 - Set inline-resize flag when the content-box size is changing (in addition to border-box). r=mats

MozReview-Commit-ID: FjUwuQB8g5j

layout/generic/ReflowInput.cpp

@@ -522,9 +522,34 @@ void
 ReflowInput::InitResizeFlags(nsPresContext* aPresContext, nsIAtom* aFrameType)
 {
   const WritingMode wm = mWritingMode; // just a shorthand
+  // We should report that we have a resize in the inline dimension if
+  // *either* the border-box size or the content-box size in that
+  // dimension has changed.  It might not actually be necessary to do
+  // this if the border-box size has changed and the content-box size
+  // has not changed, but since we've historically used the flag to mean
+  // border-box size change, continue to do that.  (It's possible for
+  // the content-box size to change without a border-box size change or
+  // a style change given (1) a fixed width (possibly fixed by max-width
+  // or min-width), (2) box-sizing:border-box or padding-box, and
+  // (3) percentage padding.)
+  //
+  // However, we don't actually have the information at this point to
+  // tell whether the content-box size has changed, since both style
+  // data and the UsedPaddingProperty() have already been updated.  So,
+  // instead, we explicitly check for the case where it's possible for
+  // the content-box size to have changed without either (a) a change in
+  // the border-box size or (b) an nsChangeHint_NeedDirtyReflow change
+  // hint due to change in border or padding.  Thus we test using the
+  // conditions from the previous paragraph, except without testing (1)
+  // since it's complicated to test properly and less likely to help
+  // with optimizing cases away.
   bool isIResize =
+    // is the border-box resizing?
     mFrame->ISize(wm) !=
-      ComputedISize() + ComputedLogicalBorderPadding().IStartEnd(wm);
+      ComputedISize() + ComputedLogicalBorderPadding().IStartEnd(wm) ||
+    // or is the content-box resizing?  (see comment above)
+    (mStylePosition->mBoxSizing != StyleBoxSizing::Content &&
+     mStylePadding->IsWidthDependent());
 
   if ((mFrame->GetStateBits() & NS_FRAME_FONT_INFLATION_FLOW_ROOT) &&
       nsLayoutUtils::FontSizeInflationEnabled(aPresContext)) {

layout/style/nsStyleStruct.cpp

@@ -336,6 +336,10 @@ nsStylePadding::CalcDifference(const nsStylePadding& aNewData) const
   // to force children to reflow so that we can reposition them, since their
   // offsets are from our frame bounds but our content rect's position within
   // those bounds is moving.
+  // FIXME: It would be good to return a weaker hint here that doesn't
+  // force reflow of all descendants, but the hint would need to force
+  // reflow of the frame's children (see how
+  // ReflowInput::InitResizeFlags initializes the inline-resize flag).
   return NS_STYLE_HINT_REFLOW & ~nsChangeHint_ClearDescendantIntrinsics;
 }
 
@@ -470,8 +474,13 @@ nsStyleBorder::Destroy(nsPresContext* aContext) {
 nsChangeHint
 nsStyleBorder::CalcDifference(const nsStyleBorder& aNewData) const
 {
-  // XXXbz we should be able to return a more specific change hint for
-  // at least GetComputedBorder() differences...
+  // FIXME: XXXbz: As in nsStylePadding::CalcDifference, many of these
+  // differences should not need to clear descendant intrinsics.
+  // FIXME: It would be good to return a weaker hint for the
+  // GetComputedBorder() differences (and perhaps others) that doesn't
+  // force reflow of all descendants, but the hint would need to force
+  // reflow of the frame's children (see how
+  // ReflowInput::InitResizeFlags initializes the inline-resize flag).
   if (mTwipsPerPixel != aNewData.mTwipsPerPixel ||
       GetComputedBorder() != aNewData.GetComputedBorder() ||
       mFloatEdge != aNewData.mFloatEdge ||