gecko-dev/mobile/android/modules/WebsiteMetadata.jsm

494 строки
18 KiB
JavaScript

/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
"use strict";
const { classes: Cc, interfaces: Ci, utils: Cu } = Components;
this.EXPORTED_SYMBOLS = ["WebsiteMetadata"];
Cu.import("resource://gre/modules/XPCOMUtils.jsm");
XPCOMUtils.defineLazyModuleGetter(this, "EventDispatcher", "resource://gre/modules/Messaging.jsm");
XPCOMUtils.defineLazyModuleGetter(this, "Task", "resource://gre/modules/Task.jsm");
var WebsiteMetadata = {
/**
* Asynchronously parse the document extract metadata. A 'Website:Metadata' event with the metadata
* will be sent.
*/
parseAsynchronously: function(doc) {
Task.spawn(function() {
let metadata = getMetadata(doc, doc.location.href, {
image_url: metadataRules.image_url,
provider: metadataRules.provider,
description_length: metadataRules.description_length
});
// No metadata was extracted, so don't bother sending it.
if (Object.keys(metadata).length === 0) {
return;
}
let msg = {
type: "Website:Metadata",
location: doc.location.href,
hasImage: metadata.image_url && metadata.image_url !== "",
metadata: JSON.stringify(metadata),
};
EventDispatcher.instance.sendRequest(msg);
});
}
};
// #################################################################################################
// # Modified version of makeUrlAbsolute() to not import url parser library (and dependencies)
// #################################################################################################
function makeUrlAbsolute(context, relative) {
var a = context.doc.createElement("a");
a.href = relative;
return a.href;
}
// #################################################################################################
// # page-metadata-parser
// # https://github.com/mozilla/page-metadata-parser/
// # 61c58cbd0f0bf2153df832a388a79c66b288b98c
// #################################################################################################
function buildRuleset(name, rules, processors) {
const reversedRules = Array.from(rules).reverse();
const builtRuleset = ruleset(...reversedRules.map(([query, handler], order) => rule(
dom(query),
node => [{
score: order,
flavor: name,
notes: handler(node),
}]
)));
return (doc, context) => {
const kb = builtRuleset.score(doc);
const maxNode = kb.max(name);
if (maxNode) {
let value = maxNode.flavors.get(name);
if (processors) {
processors.forEach(processor => {
value = processor(value, context);
});
}
if (value) {
if (value.trim) {
return value.trim();
}
return value;
}
}
};
}
const descriptionRules = [
['meta[property="og:description"]', node => node.element.getAttribute("content")],
['meta[name="description"]', node => node.element.getAttribute("content")],
];
const metadataRules = {
description: {
rules: descriptionRules
},
description_length: {
rules: descriptionRules,
processors: [
(description) => description.length
]
},
icon_url: {
rules: [
['link[rel="apple-touch-icon"]', node => node.element.getAttribute("href")],
['link[rel="apple-touch-icon-precomposed"]', node => node.element.getAttribute("href")],
['link[rel="icon"]', node => node.element.getAttribute("href")],
['link[rel="fluid-icon"]', node => node.element.getAttribute("href")],
['link[rel="shortcut icon"]', node => node.element.getAttribute("href")],
['link[rel="Shortcut Icon"]', node => node.element.getAttribute("href")],
['link[rel="mask-icon"]', node => node.element.getAttribute("href")],
],
processors: [
(icon_url, context) => makeUrlAbsolute(context, icon_url)
]
},
image_url: {
rules: [
['meta[property="og:image:secure_url"]', node => node.element.getAttribute("content")],
['meta[property="og:image:url"]', node => node.element.getAttribute("content")],
['meta[property="og:image"]', node => node.element.getAttribute("content")],
['meta[property="twitter:image"]', node => node.element.getAttribute("content")],
['meta[name="thumbnail"]', node => node.element.getAttribute("content")],
],
processors: [
(image_url, context) => makeUrlAbsolute(context, image_url)
],
},
keywords: {
rules: [
['meta[name="keywords"]', node => node.element.getAttribute("content")],
],
processors: [
(keywords) => keywords.split(",").map((keyword) => keyword.trim()),
]
},
title: {
rules: [
['meta[property="og:title"]', node => node.element.getAttribute("content")],
['meta[property="twitter:title"]', node => node.element.getAttribute("content")],
['meta[name="hdl"]', node => node.element.getAttribute("content")],
["title", node => node.element.text],
],
},
type: {
rules: [
['meta[property="og:type"]', node => node.element.getAttribute("content")],
],
},
url: {
rules: [
['meta[property="og:url"]', node => node.element.getAttribute("content")],
['link[rel="canonical"]', node => node.element.getAttribute("href")],
],
},
provider: {
rules: [
['meta[property="og:site_name"]', node => node.element.getAttribute("content")]
]
},
};
function getMetadata(doc, url, rules) {
const metadata = {};
const context = {url, doc};
const ruleSet = rules || metadataRules;
Object.keys(ruleSet).map(metadataKey => {
const metadataRule = ruleSet[metadataKey];
if (Array.isArray(metadataRule.rules)) {
const builtRule = buildRuleset(metadataKey, metadataRule.rules, metadataRule.processors);
metadata[metadataKey] = builtRule(doc, context);
} else {
metadata[metadataKey] = getMetadata(doc, url, metadataRule);
}
});
return metadata;
}
// #################################################################################################
// # Fathom dependencies resolved
// #################################################################################################
// const {forEach} = require('wu');
function forEach(fn, obj) {
for (let x of obj) {
fn(x);
}
}
function best(iterable, by, isBetter) {
let bestSoFar, bestKeySoFar;
let isFirst = true;
forEach(
function(item) {
const key = by(item);
if (isBetter(key, bestKeySoFar) || isFirst) {
bestSoFar = item;
bestKeySoFar = key;
isFirst = false;
}
},
iterable);
if (isFirst) {
throw new Error("Tried to call best() on empty iterable");
}
return bestSoFar;
}
// const {max} = require('./utils');
function max(iterable, by = identity) {
return best(iterable, by, (a, b) => a > b);
}
// #################################################################################################
// # Fathom
// # https://github.com/mozilla/fathom
// # cac59e470816f17fc1efd4a34437b585e3e451cd
// #################################################################################################
// Get a key of a map, first setting it to a default value if it's missing.
function getDefault(map, key, defaultMaker) {
if (map.has(key)) {
return map.get(key);
}
const defaultValue = defaultMaker();
map.set(key, defaultValue);
return defaultValue;
}
// Construct a filtration network of rules.
function ruleset(...rules) {
const rulesByInputFlavor = new Map(); // [someInputFlavor: [rule, ...]]
// File each rule under its input flavor:
forEach(rule => getDefault(rulesByInputFlavor, rule.source.inputFlavor, () => []).push(rule),
rules);
return {
// Iterate over a DOM tree or subtree, building up a knowledgebase, a
// data structure holding scores and annotations for interesting
// elements. Return the knowledgebase.
//
// This is the "rank" portion of the rank-and-yank algorithm.
score: function(tree) {
const kb = knowledgebase();
// Introduce the whole DOM into the KB as flavor 'dom' to get
// things started:
const nonterminals = [[{tree}, "dom"]]; // [[node, flavor], [node, flavor], ...]
// While there are new facts, run the applicable rules over them to
// generate even newer facts. Repeat until everything's fully
// digested. Rules run in no particular guaranteed order.
while (nonterminals.length) {
const [inNode, inFlavor] = nonterminals.pop();
for (let rule of getDefault(rulesByInputFlavor, inFlavor, () => [])) {
const outFacts = resultsOf(rule, inNode, inFlavor, kb);
for (let fact of outFacts) {
const outNode = kb.nodeForElement(fact.element);
// No matter whether or not this flavor has been
// emitted before for this node, we multiply the score.
// We want to be able to add rules that refine the
// scoring of a node, without having to rewire the path
// of flavors that winds through the ruleset.
//
// 1 score per Node is plenty. That simplifies our
// data, our rankers, our flavor system (since we don't
// need to represent score axes), and our engine. If
// somebody wants more score axes, they can fake it
// themselves with notes, thus paying only for what
// they eat. (We can even provide functions that help
// with that.) Most rulesets will probably be concerned
// with scoring only 1 thing at a time anyway. So,
// rankers return a score multiplier + 0 or more new
// flavors with optional notes. Facts can never be
// deleted from the KB by rankers (or order would start
// to matter); after all, they're *facts*.
outNode.score *= fact.score;
// Add a new annotation to a node--but only if there
// wasn't already one of the given flavor already
// there; otherwise there's no point.
//
// You might argue that we might want to modify an
// existing note here, but that would be a bad
// idea. Notes of a given flavor should be
// considered immutable once laid down. Otherwise, the
// order of execution of same-flavored rules could
// matter, hurting pluggability. Emit a new flavor and
// a new note if you want to do that.
//
// Also, choosing not to add a new fact to nonterminals
// when we're not adding a new flavor saves the work of
// running the rules against it, which would be
// entirely redundant and perform no new work (unless
// the rankers were nondeterministic, but don't do
// that).
if (!outNode.flavors.has(fact.flavor)) {
outNode.flavors.set(fact.flavor, fact.notes);
kb.indexNodeByFlavor(outNode, fact.flavor); // TODO: better encapsulation rather than indexing explicitly
nonterminals.push([outNode, fact.flavor]);
}
}
}
}
return kb;
}
};
}
// Construct a container for storing and querying facts, where a fact has a
// flavor (used to dispatch further rules upon), a corresponding DOM element, a
// score, and some other arbitrary notes opaque to fathom.
function knowledgebase() {
const nodesByFlavor = new Map(); // Map{'texty' -> [NodeA],
// 'spiffy' -> [NodeA, NodeB]}
// NodeA = {element: <someElement>,
//
// // Global nodewide score. Add
// // custom ones with notes if
// // you want.
// score: 8,
//
// // Flavors is a map of flavor names to notes:
// flavors: Map{'texty' -> {ownText: 'blah',
// someOtherNote: 'foo',
// someCustomScore: 10},
// // This is an empty note:
// 'fluffy' -> undefined}}
const nodesByElement = new Map();
return {
// Return the "node" (our own data structure that we control) that
// corresponds to a given DOM element, creating one if necessary.
nodeForElement: function(element) {
return getDefault(nodesByElement,
element,
() => ({element,
score: 1,
flavors: new Map()}));
},
// Return the highest-scored node of the given flavor, undefined if
// there is none.
max: function(flavor) {
const nodes = nodesByFlavor.get(flavor);
return nodes === undefined ? undefined : max(nodes, node => node.score);
},
// Let the KB know that a new flavor has been added to an element.
indexNodeByFlavor: function(node, flavor) {
getDefault(nodesByFlavor, flavor, () => []).push(node);
},
nodesOfFlavor: function(flavor) {
return getDefault(nodesByFlavor, flavor, () => []);
}
};
}
// Apply a rule (as returned by a call to rule()) to a fact, and return the
// new facts that result.
function resultsOf(rule, node, flavor, kb) {
// If more types of rule pop up someday, do fancier dispatching here.
return rule.source.flavor === "flavor" ? resultsOfFlavorRule(rule, node, flavor) : resultsOfDomRule(rule, node, kb);
}
// Pull the DOM tree off the special property of the root "dom" fact, and query
// against it.
function* resultsOfDomRule(rule, specialDomNode, kb) {
// Use the special "tree" property of the special starting node:
const matches = specialDomNode.tree.querySelectorAll(rule.source.selector);
for (let i = 0; i < matches.length; i++) { // matches is a NodeList, which doesn't conform to iterator protocol
const element = matches[i];
const newFacts = explicitFacts(rule.ranker(kb.nodeForElement(element)));
for (let fact of newFacts) {
if (fact.element === undefined) {
fact.element = element;
}
if (fact.flavor === undefined) {
throw new Error("Rankers of dom() rules must return a flavor in each fact. Otherwise, there is no way for that fact to be used later.");
}
yield fact;
}
}
}
function* resultsOfFlavorRule(rule, node, flavor) {
const newFacts = explicitFacts(rule.ranker(node));
for (let fact of newFacts) {
// If the ranker didn't specify a different element, assume it's
// talking about the one we passed in:
if (fact.element === undefined) {
fact.element = node.element;
}
if (fact.flavor === undefined) {
fact.flavor = flavor;
}
yield fact;
}
}
// Take the possibly abbreviated output of a ranker function, and make it
// explicitly an iterable with a defined score.
//
// Rankers can return undefined, which means "no facts", a single fact, or an
// array of facts.
function* explicitFacts(rankerResult) {
const array = (rankerResult === undefined) ? [] : (Array.isArray(rankerResult) ? rankerResult : [rankerResult]);
for (let fact of array) {
if (fact.score === undefined) {
fact.score = 1;
}
yield fact;
}
}
// TODO: For the moment, a lot of responsibility is on the rankers to return a
// pretty big data structure of up to 4 properties. This is a bit verbose for
// an arrow function (as I hope we can use most of the time) and the usual case
// will probably be returning just a score multiplier. Make that case more
// concise.
// TODO: It is likely that rankers should receive the notes of their input type
// as a 2nd arg, for brevity.
// Return a condition that uses a DOM selector to find its matches from the
// original DOM tree.
//
// For consistency, Nodes will still be delivered to the transformers, but
// they'll have empty flavors and score = 1.
//
// Condition constructors like dom() and flavor() build stupid, introspectable
// objects that the query engine can read. They don't actually do the query
// themselves. That way, the query planner can be smarter than them, figuring
// out which indices to use based on all of them. (We'll probably keep a heap
// by each dimension's score and a hash by flavor name, for starters.) Someday,
// fancy things like this may be possible: rule(and(tag('p'), klass('snork')),
// ...)
function dom(selector) {
return {
flavor: "dom",
inputFlavor: "dom",
selector
};
}
// Return a condition that discriminates on nodes of the knowledgebase by flavor.
function flavor(inputFlavor) {
return {
flavor: "flavor",
inputFlavor
};
}
function rule(source, ranker) {
return {
source,
ranker
};
}