[Cocoa] Strip DataDetectors links when copying content to the pasteboard

• Component: HTML Editing
• Status: RESOLVED FIXED

[Aditya Keerthi]

...

[Aditya Keerthi]

<rdar://problem/71045590>

[Aditya Keerthi]

Created attachment 417417 [details]
Patch

[Ryosuke Niwa]

Comment on attachment 417417 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=417417&action=review

> LayoutTests/editing/pasteboard/copy-paste-data-detected-links.html:6
> +    <span>Meeting <a href="x-apple-data-detectors://0" dir="ltr" x-apple-data-detectors="true" x-apple-data-detectors-type="calendar-event" x-apple-data-detectors-result="0" style="color: currentcolor; text-decoration-color: rgba(128, 128, 128, 0.38);">on Friday 11/6 at 4pm</a> at <a href="x-apple-data-detectors://1" dir="ltr" x-apple-data-detectors="true" x-apple-data-detectors-type="address" x-apple-data-detectors-result="1" style="color: currentcolor; text-decoration-color: rgba(128, 128, 128, 0.38);">1 Apple Park Way, Cupertino CA</a>

If this anchor element gets, say, font-weight: bold, do we preserve that using a span?
Otherwise, we need a span conversion instead of simply removing the anchor.
Also, please add a test for this case.

[Aditya Keerthi]

Created attachment 417459 [details]
Patch

[Aditya Keerthi]

Added tests for the WebArchive and RTF cases.

[Aditya Keerthi]

(In reply to Ryosuke Niwa from comment #3)
> Comment on attachment 417417 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=417417&action=review
> 
> > LayoutTests/editing/pasteboard/copy-paste-data-detected-links.html:6
> > +    <span>Meeting <a href="x-apple-data-detectors://0" dir="ltr" x-apple-data-detectors="true" x-apple-data-detectors-type="calendar-event" x-apple-data-detectors-result="0" style="color: currentcolor; text-decoration-color: rgba(128, 128, 128, 0.38);">on Friday 11/6 at 4pm</a> at <a href="x-apple-data-detectors://1" dir="ltr" x-apple-data-detectors="true" x-apple-data-detectors-type="address" x-apple-data-detectors-result="1" style="color: currentcolor; text-decoration-color: rgba(128, 128, 128, 0.38);">1 Apple Park Way, Cupertino CA</a>
> 
> If this anchor element gets, say, font-weight: bold, do we preserve that
> using a span?
> Otherwise, we need a span conversion instead of simply removing the anchor.
> Also, please add a test for this case.

I'm not sure span conversion is necessary, since DD links are behind SPI, and are only used in non-editable contexts.

If we did perform span conversion, we would also preserve the "color" and "currentcolor" styles added by us when creating a DD link, which doesn't make sense to preserve.

[Aditya Keerthi]

(In reply to Aditya Keerthi from comment #6)
> (In reply to Ryosuke Niwa from comment #3)
> > Comment on attachment 417417 [details]
> > Patch
> > 
> > View in context:
> > https://bugs.webkit.org/attachment.cgi?id=417417&action=review
> > 
> > > LayoutTests/editing/pasteboard/copy-paste-data-detected-links.html:6
> > > +    <span>Meeting <a href="x-apple-data-detectors://0" dir="ltr" x-apple-data-detectors="true" x-apple-data-detectors-type="calendar-event" x-apple-data-detectors-result="0" style="color: currentcolor; text-decoration-color: rgba(128, 128, 128, 0.38);">on Friday 11/6 at 4pm</a> at <a href="x-apple-data-detectors://1" dir="ltr" x-apple-data-detectors="true" x-apple-data-detectors-type="address" x-apple-data-detectors-result="1" style="color: currentcolor; text-decoration-color: rgba(128, 128, 128, 0.38);">1 Apple Park Way, Cupertino CA</a>
> > 
> > If this anchor element gets, say, font-weight: bold, do we preserve that
> > using a span?
> > Otherwise, we need a span conversion instead of simply removing the anchor.
> > Also, please add a test for this case.
> 
> I'm not sure span conversion is necessary, since DD links are behind SPI,
> and are only used in non-editable contexts.
> 
> If we did perform span conversion, we would also preserve the "color" and
> "currentcolor" styles added by us when creating a DD link, which doesn't
> make sense to preserve.

I was incorrect about DD links being SPI only (https://developer.apple.com/documentation/webkit/wkwebviewconfiguration/1641937-datadetectortypes?language=objc).

Will update the patch to perform span conversion and add new tests.

[Aditya Keerthi]

Created attachment 417494 [details]
Patch

[Aditya Keerthi]

The new patch replaces the nodes with <span> elements, rather than simply removing them.

The tests account for style preservation with the addition of "font-weight: bold" to the style attribute.

[Ryosuke Niwa]

Comment on attachment 417494 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=417494&action=review

> Source/WebCore/editing/cocoa/DataDetection.mm:650
> +    for (auto* child = ElementTraversal::firstChild(element); child; child = ElementTraversal::nextSibling(*child))
> +        convertDataDetectedLinkToSpan(*child);

Why are we recursing here? We've already found all anchor elements in replaceDataDetectedLinksInContainer.

> Source/WebCore/editing/cocoa/DataDetection.mm:662
> +    spanElement->removeAttribute(x_apple_data_detectorsAttr);

We should probably also remove color and background-color properties if they match the one we add by default.

> Source/WebCore/editing/markup.cpp:639
> +        if (is<HTMLAnchorElement>(node) && equalIgnoringASCIICase(downcast<HTMLAnchorElement>(node).attributeWithoutSynchronization(x_apple_data_detectorsAttr), "true"))
> +            m_containsDataDetectedLink = true;

Instead of this, we should just modify StyledMarkupAccumulator::appendStartTag to emit a span instead of <a.
Namely, look at what we do in StyledMarkupAccumulator::appendStartTag for HTMLSlotElement.
We just need to do something similar and then after the line where we merge inline style:
        if (is<StyledElement>(element) && downcast<StyledElement>(element).inlineStyle())
            newInlineStyle->overrideWithStyle(*downcast<StyledElement>(element).inlineStyle());
we need some code to delete the those inline styles inserted by WebKit for data detecter anchor elements.

> Tools/TestWebKitAPI/Tests/WebKitCocoa/CopyRTF.mm:156
> +            ASSERT_NOT_REACHED();

Please use FAIL() instead.

[Darin Adler]

Comment on attachment 417494 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=417494&action=review

>> Source/WebCore/editing/cocoa/DataDetection.mm:650
>> +        convertDataDetectedLinkToSpan(*child);
> 
> Why are we recursing here? We've already found all anchor elements in replaceDataDetectedLinksInContainer.

An error like this makes me think we need to stress test with an absurdly deep hierarchy so we can quickly detect if we accidentally make algorithms have unacceptable complexity.

[Aditya Keerthi]

Created attachment 417548 [details]
Patch

[Aditya Keerthi]

(In reply to Ryosuke Niwa from comment #10)
> Comment on attachment 417494 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=417494&action=review
> 
> > Source/WebCore/editing/cocoa/DataDetection.mm:650
> > +    for (auto* child = ElementTraversal::firstChild(element); child; child = ElementTraversal::nextSibling(*child))
> > +        convertDataDetectedLinkToSpan(*child);
> 
> Why are we recursing here? We've already found all anchor elements in
> replaceDataDetectedLinksInContainer.

Right, this was unnecessary. Note that this code has been removed entirely from the latest patch, due to the new approach to <span> conversion.
 
> > Source/WebCore/editing/cocoa/DataDetection.mm:662
> > +    spanElement->removeAttribute(x_apple_data_detectorsAttr);
> 
> We should probably also remove color and background-color properties if they
> match the one we add by default.

We don't add a "background-color" property, but we do add "color: currentcolor", which is already removed when creating the inline style.
 
> > Source/WebCore/editing/markup.cpp:639
> > +        if (is<HTMLAnchorElement>(node) && equalIgnoringASCIICase(downcast<HTMLAnchorElement>(node).attributeWithoutSynchronization(x_apple_data_detectorsAttr), "true"))
> > +            m_containsDataDetectedLink = true;
> 
> Instead of this, we should just modify
> StyledMarkupAccumulator::appendStartTag to emit a span instead of <a.
> Namely, look at what we do in StyledMarkupAccumulator::appendStartTag for
> HTMLSlotElement.
> We just need to do something similar and then after the line where we merge
> inline style:
>         if (is<StyledElement>(element) &&
> downcast<StyledElement>(element).inlineStyle())
>            
> newInlineStyle->overrideWithStyle(*downcast<StyledElement>(element).
> inlineStyle());
> we need some code to delete the those inline styles inserted by WebKit for
> data detecter anchor elements.

Updated the patch to modify appendStartTag/appendEndTag rather than manipulating the nodes in a DocumentFragment.

Note that the only style removed explicitly is CSSPropertyTextDecorationColor, which we add when creating the link.
 
> > Tools/TestWebKitAPI/Tests/WebKitCocoa/CopyRTF.mm:156
> > +            ASSERT_NOT_REACHED();
> 
> Please use FAIL() instead.

Done.

[Ryosuke Niwa]

Comment on attachment 417548 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=417548&action=review

> Source/WebCore/editing/cocoa/HTMLConverter.mm:1966
> -    if (range.length > 0 && element.hasTagName(aTag)) {
> +    if (range.length > 0 && element.hasTagName(aTag) && ![@"true" isEqualToString:element.getAttribute(x_apple_data_detectorsAttr)]) {

Can we just call DataDetection::isDataDetectorElement(element)?

> Source/WebCore/editing/markup.cpp:532
> +#if ENABLE(DATA_DETECTION)
> +    const bool isDataDetectorElement = DataDetection::isDataDetectorElement(element);
> +#endif

Rather than checking this condition twice, why don't we make shouldReplaceElementWithSpan return an enum class.
e.g.
enum class SpanReplacementType {
    None,
    Slot,
    DataDetector
};
auto replacementType = spanReplacementForElement(element)

> Source/WebCore/editing/markup.cpp:554
> +            if (isDataDetectorElement) {
> +                if (attributeName == x_apple_data_detectorsAttr)

Can we extract this as a helper function in DataDetector like this:
DataDetector::isDataDetectorAttribute(attribute.name())
attribute.name() is super cheap (just loading from offset) so there is no need to avoid calling that twice here really.

> Source/WebCore/editing/markup.cpp:586
> +        if (isDataDetectorElement && newInlineStyle->style())
> +            newInlineStyle->style()->removeProperty(CSSPropertyTextDecorationColor);

How about color: currentcolor?

[Ryosuke Niwa]

(In reply to Aditya Keerthi from comment #13)
> (In reply to Ryosuke Niwa from comment #10)
> > Comment on attachment 417494 [details]
> > Patch
> > 
> > View in context:
> > https://bugs.webkit.org/attachment.cgi?id=417494&action=review
> >
> > > Source/WebCore/editing/cocoa/DataDetection.mm:662
> > > +    spanElement->removeAttribute(x_apple_data_detectorsAttr);
> > 
> > We should probably also remove color and background-color properties if they
> > match the one we add by default.
> 
> We don't add a "background-color" property, but we do add "color:
> currentcolor", which is already removed when creating the inline style.

Ah, missed this comment. nvm.

[Aditya Keerthi]

Created attachment 417559 [details]
Patch

[Aditya Keerthi]

(In reply to Ryosuke Niwa from comment #14)
> Comment on attachment 417548 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=417548&action=review
> 
> > Source/WebCore/editing/cocoa/HTMLConverter.mm:1966
> > -    if (range.length > 0 && element.hasTagName(aTag)) {
> > +    if (range.length > 0 && element.hasTagName(aTag) && ![@"true" isEqualToString:element.getAttribute(x_apple_data_detectorsAttr)]) {
> 
> Can we just call DataDetection::isDataDetectorElement(element)?

Done. I also moved the link creation logic into its own method.
 
> > Source/WebCore/editing/markup.cpp:532
> > +#if ENABLE(DATA_DETECTION)
> > +    const bool isDataDetectorElement = DataDetection::isDataDetectorElement(element);
> > +#endif
> 
> Rather than checking this condition twice, why don't we make
> shouldReplaceElementWithSpan return an enum class.
> e.g.
> enum class SpanReplacementType {
>     None,
>     Slot,
>     DataDetector
> };
> auto replacementType = spanReplacementForElement(element)

Updated to use an enum.
 
> > Source/WebCore/editing/markup.cpp:554
> > +            if (isDataDetectorElement) {
> > +                if (attributeName == x_apple_data_detectorsAttr)
> 
> Can we extract this as a helper function in DataDetector like this:
> DataDetector::isDataDetectorAttribute(attribute.name())
> attribute.name() is super cheap (just loading from offset) so there is no
> need to avoid calling that twice here really.

Done.

[EWS]

Committed r271507: <https://trac.webkit.org/changeset/271507>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 417559 [details].