[WIP] Add support for Android SuggestionSpans when editing text

third_party/WebKit/Source/core/editing/commands/CompositeEditCommand.cpp

Index: third_party/WebKit/Source/core/editing/commands/CompositeEditCommand.cpp
diff --git a/third_party/WebKit/Source/core/editing/commands/CompositeEditCommand.cpp b/third_party/WebKit/Source/core/editing/commands/CompositeEditCommand.cpp
index 0c7fb5795e80adc76d9cb1cd2c79949553efcf3d..f238e95ebd8aa2edbb24d7dbd9d1ad2d74d0e01d 100644
--- a/third_party/WebKit/Source/core/editing/commands/CompositeEditCommand.cpp
+++ b/third_party/WebKit/Source/core/editing/commands/CompositeEditCommand.cpp
@@ -532,13 +532,16 @@ void CompositeEditCommand::replaceTextInNode(Text* node,
                                              const String& replacementText) {
   // DeleteFromTextNodeCommand and InsertIntoTextNodeCommand are never
   // aborted.
-  applyCommandToComposite(
-      DeleteFromTextNodeCommand::create(node, offset, count),
-      ASSERT_NO_EDITING_ABORT);
+
+  // Insert must happen before delete to preserve markers correctly
+  // See algorithm in https://dom.spec.whatwg.org/#concept-cd-replace
   if (!replacementText.isEmpty())
     applyCommandToComposite(
         InsertIntoTextNodeCommand::create(node, offset, replacementText),
         ASSERT_NO_EDITING_ABORT);
+  applyCommandToComposite(DeleteFromTextNodeCommand::create(

[rlanday, 2017/01/31 19:50:21]

Again, hope this doesn't break anything...

[rlanday, 2017/01/31 22:33:45]

I have a test case that requires this fix to work properly, but this change
breaks some other test cases, I will look into this.

Also, one of the CLs I'm depending on to fix a few bugs
(https://codereview.chromium.org/2370663002/ ) currently breaks a couple of
InputMethodController tests as well...

[esprehn, 2017/01/31 22:41:34]

What did the code do before and what's different? This seems like it's a web
exposed change in the way editing works. We probably need tests for that.

[rlanday, 2017/01/31 22:58:43]

Actually this doesn't break any of the webkit_unit_tests, I'd just messed up my
local copy of this file when I was double-checking that all these changes were
necessary :)

The reason I'm making this change is that there's a method
DeleteSelectionCommand:fixupWhitespace():
https://cs.chromium.org/chromium/src/third_party/WebKit/Source/core/editing/c...

I'd written a test (not currently in this CL but I will add it) that does the
following:

- Create an editable div containing the text "Initial text."
- Underline "text"
- Set composition to "Initial"
- Delete the composition (leaving us with " text.")
- Verify that the underline is still attached to "text"

Without this fix, fixupWhitespace() would come along and do some operation (I
think replacing a space with a non-breaking space?) and it would first delete
the space, causing the marker to be shifted left by a character (so it's still
on "text"), and then it would insert the nbsp or whatever and it would get
included at the beginning of the range, which seems clearly broken in this case.
If you always do the insert before the delete, the information about whether or
not the replaced character should be included in the new text gets preserved.

This is a fairly low-level editing operation, if it breaks anything, I suspect
it would be the way DOM ranges get updated, but I think I'm actually updating
the DocumentMarkers now in the same way DOM ranges are updated, and I'm pretty
sure there are tests for DOM ranges (although I think I've only tried running
webkit_unit_tests so far).

+                              node, offset + replacementText.length(), count),
+                          ASSERT_NO_EDITING_ABORT);
 }
 
 Position CompositeEditCommand::replaceSelectedTextInNode(const String& text) {