Add functions searching a word boundary without VisualPosition to HTMLTextFormControlElement.

Source/core/html/HTMLTextFormControlElement.cpp

Index: Source/core/html/HTMLTextFormControlElement.cpp
diff --git a/Source/core/html/HTMLTextFormControlElement.cpp b/Source/core/html/HTMLTextFormControlElement.cpp
index 0434728f24a557dcb6aa4a94af75ce08ceaccc60..34d175ee5f0b302ff158a3dd9fdc08a3daa90022 100644
--- a/Source/core/html/HTMLTextFormControlElement.cpp
+++ b/Source/core/html/HTMLTextFormControlElement.cpp
@@ -30,6 +30,7 @@
 #include "core/HTMLNames.h"
 #include "core/accessibility/AXObjectCache.h"
 #include "core/dom/Document.h"
+#include "core/dom/NodeList.h"
 #include "core/dom/NodeTraversal.h"
 #include "core/dom/Text.h"
 #include "core/dom/shadow/ShadowRoot.h"
@@ -43,6 +44,7 @@
 #include "core/rendering/RenderBlock.h"
 #include "core/rendering/RenderTheme.h"
 #include "platform/heap/Handle.h"
+#include "platform/text/TextBoundaries.h"
 #include "wtf/text/StringBuilder.h"
 
 namespace WebCore {
@@ -659,4 +661,246 @@ HTMLElement* HTMLTextFormControlElement::innerEditorElement() const
     return toHTMLElement(userAgentShadowRoot()->getElementById(ShadowElementNames::innerEditor()));
 }
 
+static Position innerNodePosition(const Position& innerPosition)
+{
+    HTMLElement* element = toHTMLElement(innerPosition.anchorNode());

[Yuta Kitamura, 2014/07/03 08:39:27]

What if the anchor node isn't an HTMLElement?

[yoichio, 2014/07/04 01:29:44]

This function is called with only innerEditorElement.

[Yuta Kitamura, 2014/07/04 06:12:34]

Then that should be ASSERTed.

+    RefPtrWillBeRawPtr<NodeList> childNodes = element->childNodes();
+    if (!childNodes->length())
+        return Position(element, 0, Position::PositionIsOffsetInAnchor);
+
+    unsigned offset = 0;
+
+    switch (innerPosition.anchorType()) {
+    case Position::PositionIsOffsetInAnchor:
+        offset = std::max(0, std::min(innerPosition.computeOffsetInContainerNode(), (int)childNodes->length()));

[Yuta Kitamura, 2014/07/03 08:39:27]

1. Can the second argument of max() become negative?

2. nit: Prefer static_cast<> over C-style cast.

[yoichio, 2014/07/04 01:29:43]

Sorry, This must be offsetInContainerNode(), which returns just m_offset.
Done

+        break;
+    case Position::PositionIsAfterChildren:
+        offset = childNodes->length();
+        break;
+    default:

[Yuta Kitamura, 2014/07/03 08:39:27]

What about other Position types?

[yoichio, 2014/07/04 01:29:43]

PositionIsBeforeChildren means offset is 0.

In our use cases, the anchorType can not be PositionIsBeforeAnchor and
PositionIsAfterAnchor.

[Yuta Kitamura, 2014/07/04 05:23:34]

Then you should ASSERT that. It's not okay to have
such an implicit precondition.

[yoichio, 2014/07/04 05:44:41]

Done.

+        break;
+    }
+
+    if (offset == childNodes->length())
+        return Position(childNodes->item(childNodes->length() - 1), Position::PositionIsAfterAnchor);

[Yuta Kitamura, 2014/07/03 08:39:27]

item(...) can be lastChild().

[yoichio, 2014/07/04 01:29:44]

Done.

+
+    Node* node = childNodes->item(offset);
+    if (node->isTextNode())
+        return Position(toText(node), 0);
+
+    return Position(node, Position::PositionIsBeforeAnchor);
+}
+
+static Position findWordBoundary(const HTMLElement* innerEditor, const Position& startPosition, const Position endPosition, bool findStart)

[Yuta Kitamura, 2014/07/03 08:39:27]

The type of the third argument should be "const Position&".

[yoichio, 2014/07/04 01:29:43]

Done.

+{
+    Vector<UChar> characters;
+    Vector<unsigned> lengthList;
+    Vector<Text*> textList;
+    // Trace text nodes.
+    for (Node* node = startPosition.anchorNode(); node; node = NodeTraversal::next(*node, innerEditor)) {

[Yuta Kitamura, 2014/07/03 08:39:27]

If startPosition is of type AfterAnchor or AfterChildren, this
does not work correctly. 

[yoichio, 2014/07/04 01:29:44]

A goal of this loop is to collect all strings in text nodes.
The start/end position must be of type PositionIsOffsetInAnchor so I add
assertions.

+        bool isStartNode = node == startPosition.anchorNode();
+        bool isEndNode = node == endPosition.anchorNode();

[Yuta Kitamura, 2014/07/03 08:39:27]

These booleans are smelly, too, if they are of type BeforeAnchor
or AfterAnchor (these positions are basically anchored by
anchorNode()->parentNode()).

[yoichio, 2014/07/04 01:29:44]

ditto

+        if (node->isTextNode()) {
+            Text* text = toText(node);
+            const unsigned start = isStartNode ? startPosition.computeOffsetInContainerNode() : 0;
+            const unsigned end = isEndNode ? endPosition.computeOffsetInContainerNode() : text->data().length();
+            for (unsigned offset = start; offset < end; ++offset)
+                characters.append(text->data()[offset]);

[Yuta Kitamura, 2014/07/03 08:39:27]

This loop looks error-prone.

If you use StringBuilder instead of Vector<UChar>. you can
make use of StringBuilder::append() which supports adding
a substring of a String.

[yoichio, 2014/07/04 01:29:44]

Since findWordBoundary takes UChar*, we need to convert strings if it is 8bit.
text->data()[offset] also do that conversion.

[Yuta Kitamura, 2014/07/04 05:23:34]

It's silly to write this kind of loop by hand. You should make
use of libraries as much as possible.

In this case, you can use StringBuilder to build a string and
then use String::append() to extract the content into Vector<UChar>.

(I think it's pretty obvious to have an overloaded version of
findWordBoundary taking LChar*, though.)

[yoichio, 2014/07/04 05:44:41]

Done.
Thanks! I have not gotten the way :)

+            const unsigned length = end - start;
+
+            lengthList.append(length);
+            textList.append(text);
+        }
+
+        if (isEndNode)
+            break;
+    }
+
+    if (!characters.size())
+        return startPosition;
+
+    int start, end;
+    if (!findStart && characters.data()[0] == '\n') {
+        // findWordBoundary("\ntext", 0, &start, &end) assigns 1 to |end| but we expect 0 at the case.
+        start = 0;
+        end = 0;
+    } else {
+        findWordBoundary(characters.data(), characters.size(), findStart ? characters.size() : 0, &start, &end);
+    }
+    ASSERT(start >= 0);
+    ASSERT(end >= 0);
+    unsigned restOffset = findStart ? start : end;

[Yuta Kitamura, 2014/07/03 08:39:27]

nit: "rest" isn't an adjective. It's a noun. You probably
meant "remaining".

[yoichio, 2014/07/04 01:29:44]

Done.

+    // Find position.
+    for (unsigned i = 0; i < lengthList.size(); ++i) {
+        if (restOffset <= lengthList[i])
+            return Position(textList[i], (textList[i] == startPosition.anchorNode()) ? restOffset + startPosition.computeOffsetInContainerNode() : restOffset);
+        restOffset -= lengthList[i];
+    }
+
+    ASSERT_NOT_REACHED();
+    return Position();
+}
+
+Position HTMLTextFormControlElement::startOfWord(const Position& position)
+{
+    const HTMLTextFormControlElement* textFormControl = enclosingTextFormControl(position);
+    ASSERT(textFormControl);
+    HTMLElement* innerEditor = textFormControl->innerEditorElement();
+
+    const Position startPosition = startOfSentence(position);
+    if (startPosition == position)
+        return position;
+    const Position endPosition = (position.anchorNode() == innerEditor) ? innerNodePosition(position) : position;
+
+    return findWordBoundary(innerEditor, startPosition, endPosition, true);
+}
+
+Position HTMLTextFormControlElement::endOfWord(const Position& position)
+{
+    const HTMLTextFormControlElement* textFormControl = enclosingTextFormControl(position);
+    ASSERT(textFormControl);
+    HTMLElement* innerEditor = textFormControl->innerEditorElement();
+
+
+    const Position endPosition = endOfSentence(position);
+    if (endPosition == position)
+        return position;
+    const Position startPosition = (position.anchorNode() == innerEditor) ? innerNodePosition(position) : position;
+
+    return findWordBoundary(innerEditor, startPosition, endPosition, false);
+}
+
+static int findLastLineBreak(const Text& text, int offset)
+{
+    for (; offset >= 0; --offset) {
+        if (text.data()[offset] == '\n')
+            return offset;
+    }
+    return -1;
+}
+
+static Position endOfPrevious(const Node& node, HTMLElement* innerEditor)
+{
+    Node* prev = NodeTraversal::previous(node, innerEditor);
+    if (!prev)
+        return Position();
+
+    if (prev->hasTagName(brTag))
+        return Position(prev, Position::PositionIsAfterAnchor);
+
+    if (prev->isTextNode()) {
+        Text* textNode = toText(prev);
+        return Position(textNode, textNode->length());
+    }
+
+    return Position();
+}
+
+static Position previousIfPositionIsAfterLineBreak(const Position& position, HTMLElement* innerEditor)
+{
+    if (position.isNull())
+        return Position();
+
+    // Move back if position is just after line break.
+    if (isHTMLBRElement(*position.anchorNode())) {
+        switch (position.anchorType()) {
+        case Position::PositionIsAfterAnchor:
+            return Position(position.anchorNode(), Position::PositionIsBeforeAnchor);
+        case Position::PositionIsBeforeAnchor:
+            return previousIfPositionIsAfterLineBreak(endOfPrevious(*position.anchorNode(), innerEditor), innerEditor);
+        default:
+            ASSERT_NOT_REACHED();
+        }
+    } else if (position.anchorNode()->isTextNode()) {
+        Text* textNode = toText(position.anchorNode());
+        unsigned offset = position.offsetInContainerNode();
+        if (textNode->length() == 0 || offset <= 0) {
+            return previousIfPositionIsAfterLineBreak(endOfPrevious(*position.anchorNode(), innerEditor), innerEditor);
+        }
+
+        if (offset <= textNode->length() && textNode->data()[offset - 1] == '\n') {
+            return Position(textNode, offset - 1);
+        }
+    }
+
+    return position;
+}
+
+static inline Position startOfInnerText(const HTMLTextFormControlElement* textFormControl)
+{
+    return Position(textFormControl->innerEditorElement(), 0, Position::PositionIsOffsetInAnchor);
+}
+
+Position HTMLTextFormControlElement::startOfSentence(const Position& position)
+{
+    HTMLTextFormControlElement* textFormControl = enclosingTextFormControl(position);
+    ASSERT(textFormControl);
+
+    HTMLElement* innerEditor = textFormControl->innerEditorElement();
+    if (!innerEditor->childNodes()->length())
+        return startOfInnerText(textFormControl);
+
+    const Position innerPosition = position.anchorNode() == innerEditor ? innerNodePosition(position) : position;
+    const Position pivotPosition = previousIfPositionIsAfterLineBreak(innerPosition, innerEditor);
+    if (pivotPosition.isNull())
+        return startOfInnerText(textFormControl);
+
+    for (Node* node = pivotPosition.anchorNode(); node; node = NodeTraversal::previous(*node, innerEditor)) {
+        bool isPivotNode = (node == pivotPosition.anchorNode());
+
+        if (node->isTextNode()) {
+            Text* textNode = toText(node);
+            int lastLineBreak = findLastLineBreak(*textNode, isPivotNode ? pivotPosition.offsetInContainerNode() - 1 : (int)textNode->length() - 1);
+            if (lastLineBreak >= 0)
+                return Position(textNode, lastLineBreak + 1);
+        } else if (isHTMLBRElement(node) && (!isPivotNode || pivotPosition.anchorType() == Position::PositionIsAfterAnchor)) {
+            return Position(node, Position::PositionIsAfterAnchor);
+        }
+    }
+    return startOfInnerText(textFormControl);
+}
+
+static int findFirstLineBreak(const Text& text, int offset)
+{
+    for (; (unsigned)offset < text.length(); ++offset) {
+        if (text.data()[offset] == '\n')
+            return offset;
+    }
+    return -1;
+}
+
+static Position endOfInnerText(const HTMLTextFormControlElement* textFormControl)
+{
+    HTMLElement* innerEditor = textFormControl->innerEditorElement();
+    return Position(innerEditor, innerEditor->childNodes()->length(), Position::PositionIsOffsetInAnchor);
+}
+
+Position HTMLTextFormControlElement::endOfSentence(const Position& position)
+{
+    HTMLTextFormControlElement* textFormControl = enclosingTextFormControl(position);
+    ASSERT(textFormControl);
+
+    HTMLElement* innerEditor = textFormControl->innerEditorElement();
+    if (!innerEditor->childNodes()->length())
+        return startOfInnerText(textFormControl);
+
+    const Position pivotPosition = position.anchorNode() == innerEditor ? innerNodePosition(position) : position;
+    if (pivotPosition.isNull())
+        return startOfInnerText(textFormControl);
+
+    for (Node* node = pivotPosition.anchorNode(); node; node = NodeTraversal::next(*node, innerEditor)) {
+        bool isPivotNode = (node == pivotPosition.anchorNode());
+
+        if (node->isTextNode()) {
+            Text* textNode = toText(node);
+            int firstLineBreak = findFirstLineBreak(*textNode, isPivotNode ? pivotPosition.offsetInContainerNode() : 0);
+            if (firstLineBreak >= 0)
+                return Position(textNode, firstLineBreak + 1);
+        } else if (isHTMLBRElement(node)) {
+            return Position(node, Position::PositionIsAfterAnchor);
+        }
+    }
+    return endOfInnerText(textFormControl);
+}
+
 } // namespace Webcore