Bug 128538 – ContentData equals() methods are not inline-able

• Component: CSS
• Status: RESOLVED FIXED

[David Kilzer (:ddkilzer)]

Per Darin Adler in Bug 128510 Comment #2:

(From update of attachment 223660 [details])
View in context: https://bugs.webkit.org/attachment.cgi?id=223660&action=review

> Source/WebCore/rendering/style/ContentData.h:105
> +inline bool ImageContentData::equals(const ContentData& data) const
> +{
> +    if (!data.isImage())
> +        return false;
> +    return *toImageContentData(data).image() == *image();
> +}

I think we could move this to the .cpp file, because I think everyone calls it polymorphically and so nobody really inlines it.

I also think this could read nicely with && rather than early return.

> Source/WebCore/rendering/style/ContentData.h:135
> +inline bool TextContentData::equals(const ContentData& data) const
> +{
> +    if (!data.isText())
> +        return false;
> +    return toTextContentData(data).text() == text();
> +}

Ditto.

> Source/WebCore/rendering/style/ContentData.h:169
> +inline bool CounterContentData::equals(const ContentData& data) const
> +{
> +    if (!data.isCounter())
> +        return false;
> +    return *toCounterContentData(data).counter() == *counter();
> +}

Again.

> Source/WebCore/rendering/style/ContentData.h:199
> +inline bool QuoteContentData::equals(const ContentData& data) const
> +{
> +    if (!data.isQuote())
> +        return false;
> +    return toQuoteContentData(data).quote() == quote();
> +}

Ditto.

[David Kilzer (:ddkilzer)]

(In reply to comment #0)
> Per Darin Adler in Bug 128510 Comment #2:
> 
> (From update of attachment 223660 [details] [details])
> View in context: https://bugs.webkit.org/attachment.cgi?id=223660&action=review
> 
> > Source/WebCore/rendering/style/ContentData.h:105
> > +inline bool ImageContentData::equals(const ContentData& data) const
> > +{
> > +    if (!data.isImage())
> > +        return false;
> > +    return *toImageContentData(data).image() == *image();
> > +}
> 
> I think we could move this to the .cpp file, because I think everyone calls it polymorphically and so nobody really inlines it.
> 
> I also think this could read nicely with && rather than early return.

The only place the polymorphism (and the operator==() method) is used is in StyleRareNonInheritedData.cpp:

bool StyleRareNonInheritedData::contentDataEquivalent(const StyleRareNonInheritedData& o) const
{
    ContentData* a = m_content.get();
    ContentData* b = o.m_content.get();

    while (a && b && *a == *b) { // operator==() used here.
        a = a->next();
        b = b->next();
    }

    return !a && !b;
}

Is it worth leaving ContentData as an abstract base class for this one use?

An alternative approach would be to introduce a type enum for each subclass (a la CachedResource), which would let us devirtualize the type methods (isCounter, isImage, isQuote, isText), and implement the equals() methods as inline operator==() overloads instead of as pure virtual methods.

This is what the current operator==() method would look like with that change (plus individual operator==() methods for each specific type):

inline bool operator==(const ContentData& a, const ContentData& b)
{
    if (a.type() != b.type())
        return false;

    switch (a.type()) {
    case (ContentData::CounterType):
        return toCounterContentData(a) == toCounterContentData(b);
    case (ContentData::ImageType):
        return toImageContentData(a) == toImageContentData(b);
    case (ContentData::QuoteType):
        return toQuoteContentData(a) == toQuoteContentData(b);
    case (ContentData::TextType):
        return toTextContentData(a) == toTextContentData(b);
    }

    ASSERT_NOT_REACHED();
    return false;
}

Thoughts?

[David Kilzer (:ddkilzer)]

Created attachment 223726 [details]
Patch v1

[Darin Adler]

Comment on attachment 223726 [details]
Patch v1

View in context: https://bugs.webkit.org/attachment.cgi?id=223726&action=review

> Source/WebCore/rendering/style/ContentData.h:68
> +    ContentData(Type type)

explicit

> Source/WebCore/rendering/style/ContentData.h:160
>      const CounterContent* counter() const { return m_counter.get(); }

Seems like this should return a reference rather than a pointer, unless it can be null.

[David Kilzer (:ddkilzer)]

Committed r163936: <http://trac.webkit.org/changeset/163936>

[David Kilzer (:ddkilzer)]

(In reply to comment #3)
> (From update of attachment 223726 [details])
> > Source/WebCore/rendering/style/ContentData.h:160
> >      const CounterContent* counter() const { return m_counter.get(); }
> 
> Seems like this should return a reference rather than a pointer, unless it can be null.

Filed Bug 128671.