OLD | NEW |
---|---|
(Empty) | |
1 // Copyright 2015 The Chromium Authors. All rights reserved. | |
2 // Use of this source code is governed by a BSD-style license that can be | |
3 // found in the LICENSE file. | |
4 | |
5 #include "config.h" | |
6 #include "DocumentStatisticsCollector.h" | |
7 | |
8 #include "core/HTMLNames.h" | |
9 #include "core/InputTypeNames.h" | |
10 #include "core/css/CSSComputedStyleDeclaration.h" | |
11 #include "core/dom/ElementTraversal.h" | |
12 #include "core/dom/NodeComputedStyle.h" | |
13 #include "core/editing/iterators/TextIterator.h" | |
14 #include "core/frame/FrameHost.h" | |
15 #include "core/html/HTMLHeadElement.h" | |
16 #include "core/html/HTMLInputElement.h" | |
17 #include "core/html/HTMLMetaElement.h" | |
18 #include "public/platform/WebDistillability.h" | |
19 #include "wtf/text/StringBuilder.h" | |
20 #include "wtf/text/StringImpl.h" | |
21 | |
22 namespace blink { | |
23 | |
24 using namespace HTMLNames; | |
25 | |
26 namespace { | |
27 | |
28 // Saturate the length of a paragraph to save time. | |
29 const int kTextContentLengthSaturation = 1000; | |
30 | |
31 unsigned textContentLengthSaturated(Element& root) | |
32 { | |
33 unsigned length = 0; | |
34 // This skips shadow dom intentionally. | |
dglazkov
2015/11/03 04:45:37
Please explain why in the comment.
wychen
2015/11/03 07:13:41
Done.
| |
35 for (Node& node : NodeTraversal::inclusiveDescendantsOf(root)) { | |
36 if (!node.isTextNode()) { | |
37 continue; | |
38 } | |
39 length += toText(node).length(); | |
40 if (length > kTextContentLengthSaturation) { | |
41 return kTextContentLengthSaturation; | |
42 } | |
43 } | |
44 return length; | |
45 } | |
46 | |
47 bool isVisible(Element& element) | |
48 { | |
49 const ComputedStyle* style = element.computedStyle(); | |
50 if (!style) { | |
dglazkov
2015/11/03 04:45:37
Don't need braces here.
wychen
2015/11/03 07:13:41
Done.
| |
51 return false; | |
52 } | |
53 ASSERT(style->display() != NONE); | |
54 return ( | |
55 style->visibility() != HIDDEN | |
56 && style->opacity() != 0 | |
57 ); | |
58 } | |
59 | |
60 bool matchAttributes(Element& element, const Vector<String>& words) | |
61 { | |
62 const String& classes = element.getClassAttribute(); | |
63 const String& id = element.getIdAttribute(); | |
64 for (const String& word : words) { | |
65 if (classes.findIgnoringCase(word) != WTF::kNotFound | |
66 || id.findIgnoringCase(word) != WTF::kNotFound) { | |
67 return true; | |
68 } | |
69 } | |
70 return false; | |
71 } | |
72 | |
73 // underListItem denotes that at least one of the ancesters is <li> element. | |
74 void collectFeatures(Element& root, WebDistillabilityFeatures& features, bool un derListItem = false) | |
75 { | |
76 DEFINE_STATIC_LOCAL(Vector<String>, unlikelyCandidates, ()); | |
77 if (unlikelyCandidates.isEmpty()) { | |
78 for (auto word : {"banner", "combx", "comment", "community", "disqus", " extra", "foot", "header", "menu", "related", "remark", "rss", "share", "shoutbox ", "sidebar", "skyscraper", "sponsor", "ad-break", "agegate", "pagination", "pag er", "popup"}) { | |
79 unlikelyCandidates.append(word); | |
80 } | |
81 } | |
82 DEFINE_STATIC_LOCAL(Vector<String>, highlyLikelyCandidates, ()); | |
83 if (highlyLikelyCandidates.isEmpty()) { | |
84 for (auto word : {"and", "article", "body", "column", "main", "shadow"}) { | |
85 highlyLikelyCandidates.append(word); | |
86 } | |
87 } | |
88 // Filter out short P elements. The threshold is set to around 2 English sen tences. | |
89 const unsigned kParagraphLengthThreshold = 140; | |
90 | |
91 // Saturate the scores to save time. The max is the score of 6 long paragrap hs. | |
92 const double kMozScoreSaturation = 6 * sqrt(kTextContentLengthSaturation - k ParagraphLengthThreshold); | |
93 const double kMozScoreAllSqrtSaturation = 6 * sqrt(kTextContentLengthSaturat ion); | |
94 const double kMozScoreAllLinearSaturation = 6 * kTextContentLengthSaturation ; | |
95 | |
96 for (Node& node : NodeTraversal::childrenOf(root)) { | |
97 if (node.isTextNode()) { | |
98 features.textContentLength += toText(node).length(); | |
99 continue; | |
100 } | |
101 if (!node.isElementNode()) { | |
102 continue; | |
103 } | |
104 | |
105 features.elementCount++; | |
106 Element& element = toElement(node); | |
107 if (element.hasTagName(aTag)) { | |
108 features.anchorCount++; | |
109 } else if (element.hasTagName(formTag)) { | |
110 features.formCount++; | |
111 } else if (element.hasTagName(inputTag)) { | |
112 const HTMLInputElement& input = toHTMLInputElement(element); | |
113 if (equalIgnoringCase(input.type(), InputTypeNames::text)) { | |
114 features.textInputCount++; | |
115 } else if (equalIgnoringCase(input.type(), InputTypeNames::password) ) { | |
116 features.passwordInputCount++; | |
117 } | |
118 } else if (element.hasTagName(pTag) || element.hasTagName(preTag)) { | |
119 if (element.hasTagName(pTag)) { | |
120 features.pCount++; | |
121 } else { | |
122 features.preCount++; | |
123 } | |
124 if (!underListItem | |
125 && (features.mozScore < kMozScoreSaturation | |
126 || features.mozScoreAllSqrt < kMozScoreAllSqrtSaturation | |
127 || features.mozScoreAllLinear < kMozScoreAllLinearSaturation ) | |
128 && isVisible(element) | |
129 && (!matchAttributes(element, unlikelyCandidates) || matchAttrib utes(element, highlyLikelyCandidates)) | |
130 ) { | |
131 unsigned length = textContentLengthSaturated(element); | |
dglazkov
2015/11/03 04:45:37
Is this an O(NxM) built in here?
wychen
2015/11/03 07:13:41
I'm not quite sure about what you meant here.
tex
| |
132 if (length >= kParagraphLengthThreshold) { | |
133 features.mozScore += sqrt(length - kParagraphLengthThreshold ); | |
134 features.mozScore = std::min(features.mozScore, kMozScoreSat uration); | |
135 } | |
136 features.mozScoreAllSqrt += sqrt(length); | |
137 features.mozScoreAllSqrt = std::min(features.mozScoreAllSqrt, kM ozScoreAllSqrtSaturation); | |
138 | |
139 features.mozScoreAllLinear += length; | |
140 features.mozScoreAllLinear = std::min(features.mozScoreAllLinear , kMozScoreAllLinearSaturation); | |
141 } | |
142 } else if (element.hasTagName(liTag)) { | |
143 underListItem = true; | |
144 } | |
145 collectFeatures(element, features, underListItem); | |
146 } | |
147 } | |
148 | |
149 bool hasOpenGraphArticle(const Element& head) | |
150 { | |
151 DEFINE_STATIC_LOCAL(AtomicString, ogType, ("og:type")); | |
152 DEFINE_STATIC_LOCAL(AtomicString, propertyAttr, ("property")); | |
153 for (const Element* child = ElementTraversal::firstChild(head); child; child = ElementTraversal::nextSibling(*child)) { | |
154 if (!isHTMLMetaElement(*child)) | |
155 continue; | |
156 const HTMLMetaElement& meta = toHTMLMetaElement(*child); | |
157 | |
158 if (meta.name() == ogType || meta.getAttribute(propertyAttr) == ogType) { | |
159 if (equalIgnoringCase(meta.content(), "article")) { | |
160 return true; | |
161 } | |
162 } | |
163 } | |
164 return false; | |
165 } | |
166 | |
167 bool isMobileFriendly(Document& document) | |
168 { | |
169 FrameHost* frameHost = document.frameHost(); | |
170 if (!frameHost) | |
171 return false; | |
172 return frameHost->visualViewport().shouldDisableDesktopWorkarounds(); | |
173 } | |
174 | |
175 } // namespace | |
176 | |
177 WebDistillabilityFeatures DocumentStatisticsCollector::collectStatistics(Documen t& document) | |
178 { | |
179 WebDistillabilityFeatures features = WebDistillabilityFeatures(); | |
180 | |
181 if (!document.frame() || !document.frame()->isMainFrame()) | |
182 return features; | |
183 | |
184 ASSERT(document.hasFinishedParsing()); | |
185 | |
186 if (!document.body() || !document.head()) | |
dglazkov
2015/11/03 04:45:37
Both of these are traversals, so might be good to
wychen
2015/11/03 07:13:41
Done.
| |
187 return features; | |
188 | |
189 if (isMobileFriendly(document)) { | |
190 features.isMobileFriendly = true; | |
191 return features; | |
192 } | |
193 | |
194 // Traverse the DOM tree and collect statistics. | |
195 collectFeatures(*document.body(), features); | |
196 features.openGraph = hasOpenGraphArticle(*document.head()); | |
197 | |
198 return features; | |
199 } | |
200 | |
201 } | |
OLD | NEW |