// ==UserScript==
// @name No Yandex Ads
// @namespace lainverse_no_yandex_ads
// @description Removes ads in Yandex search results.
// @author lainverse
// @contributor James Lewis
// @license CC-BY-SA
// @version 3.6.1
// @grant none
// @match http://*/yandsearch?*
// @match https://*/yandsearch?*
// @match http://yandex.ru/search/*
// @match http://*.yandex.ru/search/*
// @match https://yandex.ru/search/*
// @match https://*.yandex.ru/search/*
// @run-at document-end
// ==/UserScript==
/*!
* Shim for MutationObserver interface
* Author: Graeme Yeates (github.com/megawac)
* Repository: https://github.com/megawac/MutationObserver.js
* License: WTFPL V2, 2004 (wtfpl.net).
* Though credit and staring the repo will make me feel pretty, you can modify and redistribute as you please.
* Attempts to follow spec (http:// www.w3.org/TR/dom/#mutation-observers) as closely as possible for native javascript
* See https://github.com/WebKit/webkit/blob/master/Source/WebCore/dom/MutationObserver.cpp for current webkit source c++ implementation
*/
/**
* prefix bugs:
- https://bugs.webkit.org/show_bug.cgi?id=85161
- https://bugzilla.mozilla.org/show_bug.cgi?id=749920
* Don't use WebKitMutationObserver as Safari (6.0.5-6.1) use a buggy implementation
*/
window.MutationObserver = window.MutationObserver || window.MozMutationObserver || (function (undefined) {
'use strict';
/**
* @param {function(Array.<MutationRecord>, MutationObserver)} listener
* @constructor
*/
function MutationObserver(listener) {
/**
* @type {Array.<Object>}
* @private
*/
this._watched = [];
/** @private */
this._listener = listener;
}
/**
* Start a recursive timeout function to check all items being observed for mutations
* @type {MutationObserver} observer
* @private
*/
function startMutationChecker(observer) {
(function check() {
var mutations = observer.takeRecords();
if (mutations.length) // fire away
// calling the listener with context is not spec but currently consistent with FF and WebKit
observer._listener(mutations, observer);
/** @private */
observer._timeout = setTimeout(check, MutationObserver._period);
})();
}
/**
* Period to check for mutations (~32 times/sec)
* @type {number}
* @expose
*/
MutationObserver._period = 30; /*ms+runtime*/
/**
* Exposed API
* @expose
* @final
*/
MutationObserver.prototype = {
/**
* see http:// dom.spec.whatwg.org/#dom-mutationobserver-observe
* not going to throw here but going to follow the current spec config sets
* @param {Node|null} $target
* @param {Object|null} config : MutationObserverInit configuration dictionary
* @expose
* @return undefined
*/
observe: function($target, config) {
/**
* Using slightly different names so closure can go ham
* @type {!Object} : A custom mutation config
*/
var settings = {
attr: !!(config.attributes || config.attributeFilter || config.attributeOldValue),
// some browsers are strict in their implementation that config.subtree and childList must be set together. We don't care - spec doesn't specify
kids: !!config.childList, descendents: !!config.subtree,
charData: !!(config.characterData || config.characterDataOldValue)
}, watched = this._watched;
// remove already observed target element from pool
for (var i = 0; i < watched.length; i++)
if (watched[i].tar === $target) watched.splice(i, 1);
if (config.attributeFilter)
/**
* converts to a {key: true} dict for faster lookup
* @type {Object.<String,Boolean>}
*/
settings.afilter = reduce(config.attributeFilter, function(a, b) {
a[b] = true;
return a;
}, {});
watched.push({
tar: $target,
fn: createMutationSearcher($target, settings)
});
// reconnect if not connected
if (!this._timeout) startMutationChecker(this);
},
/**
* Finds mutations since last check and empties the "record queue" i.e. mutations will only be found once
* @expose
* @return {Array.<MutationRecord>}
*/
takeRecords: function() {
var mutations = [], watched = this._watched, wl = watched.length;
for (var i = 0; i < wl; i++) watched[i].fn(mutations);
return mutations;
},
/**
* @expose
* @return undefined
*/
disconnect: function() {
this._watched = []; // clear the stuff being observed
clearTimeout(this._timeout); // ready for garbage collection
/** @private */
this._timeout = null;
}
};
/**
* Simple MutationRecord pseudoclass. No longer exposing as its not fully compliant
* @param {Object} data
* @return {Object} a MutationRecord
*/
function MutationRecord(data) {
var settings = { // technically these should be on proto so hasOwnProperty will return false for non explicitly set props
type: null, target: null, addedNodes: [], removedNodes: [], previousSibling: null,
nextSibling: null, attributeName: null, attributeNamespace: null, oldValue: null
};
for (var prop in data)
if (has(settings, prop) && data[prop] !== undefined) settings[prop] = data[prop];
return settings;
}
/**
* Creates a func to find all the mutations
*
* @param {Node} $target
* @param {!Object} config : A custom mutation config
*/
function createMutationSearcher($target, config) {
/** type {Elestuct} */
var $oldstate = clone($target, config); // create the cloned datastructure
/**
* consumes array of mutations we can push to
*
* @param {Array.<MutationRecord>} mutations
*/
return function(mutations) {
var olen = mutations.length, dirty;
// Alright we check base level changes in attributes... easy
if (config.attr && $oldstate.attr)
findAttributeMutations(mutations, $target, $oldstate.attr, config.afilter);
// check childlist or subtree for mutations
if (config.kids || config.descendents)
dirty = searchSubtree(mutations, $target, $oldstate, config);
// reclone data structure if theres changes
if (dirty || mutations.length !== olen)
/** type {Elestuct} */
$oldstate = clone($target, config);
};
}
/* attributes + attributeFilter helpers */
/**
* fast helper to check to see if attributes object of an element has changed
* doesnt handle the textnode case
*
* @param {Array.<MutationRecord>} mutations
* @param {Node} $target
* @param {Object.<string, string>} $oldstate : Custom attribute clone data structure from clone
* @param {Object} filter
*/
function findAttributeMutations(mutations, $target, $oldstate, filter) {
var checked = {}, attributes = $target.attributes, i = attributes.length, attr, name;
while (i--) {
attr = attributes[i];
name = attr.name;
if (!filter || has(filter, name)) {
if (attr.value !== $oldstate[name])
// The pushing is redundant but gzips very nicely
mutations.push(MutationRecord({
type: 'attributes', target: $target, attributeName: name, oldValue: $oldstate[name],
attributeNamespace: attr.namespaceURI // in ie<8 it incorrectly will return undefined
}));
checked[name] = true;
}
}
for (name in $oldstate)
if (!(checked[name]) && $oldstate.hasOwnProperty(name))
mutations.push(MutationRecord({target: $target, type: 'attributes', attributeName: name, oldValue: $oldstate[name]}));
}
/**
* searchSubtree: array of mutations so far, element, element clone, bool
* synchronous dfs comparision of two nodes
* This function is applied to any observed element with childList or subtree specified
* Sorry this is kind of confusing as shit, tried to comment it a bit...
* codereview.stackexchange.com/questions/38351 discussion of an earlier version of this func
*
* @param {Array} mutations
* @param {Node} $target
* @param {!Object} $oldstate : A custom cloned node from clone()
* @param {!Object} config : A custom mutation config
*/
function searchSubtree(mutations, $target, $oldstate, config) {
// Track if the tree is dirty and has to be recomputed (#14).
var dirty;
/*
* Helper to identify node rearrangment and stuff...
* There is no gaurentee that the same node will be identified for both added and removed nodes
* if the positions have been shuffled.
* conflicts array will be emptied by end of operation
*/
function resolveConflicts(conflicts, node, $kids, $oldkids, numAddedNodes) {
// the distance between the first conflicting node and the last
var distance = conflicts.length - 1,
// prevents same conflict being resolved twice consider when two nodes switch places.
// only one should be given a mutation event (note -~ is used as a math.ceil shorthand)
counter = -~((distance - numAddedNodes) / 2), $cur, oldstruct, conflict;
while (conflict = conflicts.pop()) {
$cur = $kids[conflict.i];
oldstruct = $oldkids[conflict.j];
// attempt to determine if there was node rearrangement... won't gaurentee all matches
// also handles case where added/removed nodes cause nodes to be identified as conflicts
if (config.kids && counter && Math.abs(conflict.i - conflict.j) >= distance) {
mutations.push(MutationRecord({
type: 'childList', target: node, addedNodes: [$cur], removedNodes: [$cur],
// haha don't rely on this please
nextSibling: $cur.nextSibling, previousSibling: $cur.previousSibling
}));
counter--; // found conflict
}
// Alright we found the resorted nodes now check for other types of mutations
if (config.attr && oldstruct.attr) findAttributeMutations(mutations, $cur, oldstruct.attr, config.afilter);
if (config.charData && $cur.nodeType === 3 && $cur.nodeValue !== oldstruct.charData)
mutations.push(MutationRecord({type: 'characterData', target: $cur}));
// now look @ subtree
if (config.descendents) findMutations($cur, oldstruct);
}
}
/**
* Main worker. Finds and adds mutations if there are any
* @param {Node} node
* @param {!Object} old : A cloned data structure using internal clone
*/
function findMutations(node, old) {
var $kids = node.childNodes, $oldkids = old.kids, klen = $kids.length,
// $oldkids will be undefined for text and comment nodes
olen = $oldkids ? $oldkids.length : 0;
// if (!olen && !klen) return; // both empty; clearly no changes
// we delay the intialization of these for marginal performance in the expected case (actually quite signficant on large subtrees when these would be otherwise unused)
// map of checked element of ids to prevent registering the same conflict twice
var map,
// array of potential conflicts (ie nodes that may have been re arranged)
conflicts, id, // element id from getElementId helper
idx, // index of a moved or inserted element
oldstruct,
// current and old nodes
$cur, $old,
// track the number of added nodes so we can resolve conflicts more accurately
numAddedNodes = 0,
// iterate over both old and current child nodes at the same time
i = 0, j = 0;
// while there is still anything left in $kids or $oldkids (same as i < $kids.length || j < $oldkids.length;)
while (i < klen || j < olen) {
// current and old nodes at the indexs
$cur = $kids[i];
oldstruct = $oldkids[j];
$old = oldstruct && oldstruct.node;
if ($cur === $old) { // expected case - optimized for this case
// check attributes as specified by config
if (config.attr && oldstruct.attr) /* oldstruct.attr instead of textnode check */ findAttributeMutations(mutations, $cur, oldstruct.attr, config.afilter);
// check character data if node is a comment or textNode and it's being observed
if (config.charData && oldstruct.charData !== undefined && $cur.nodeValue !== oldstruct.charData)
mutations.push(MutationRecord({type: 'characterData', target: $cur}));
// resolve conflicts; it will be undefined if there are no conflicts - otherwise an array
if (conflicts) resolveConflicts(conflicts, node, $kids, $oldkids, numAddedNodes);
// recurse on next level of children. Avoids the recursive call when there are no children left to iterate
if (config.descendents && ($cur.childNodes.length || oldstruct.kids && oldstruct.kids.length)) findMutations($cur, oldstruct);
i++;
j++;
} else { // (uncommon case) lookahead until they are the same again or the end of children
dirty = true;
if (!map) { // delayed initalization (big perf benefit)
map = {};
conflicts = [];
}
if ($cur) {
// check id is in the location map otherwise do a indexOf search
if (!(map[id = getElementId($cur)])) { // to prevent double checking
// mark id as found
map[id] = true;
// custom indexOf using comparitor checking oldkids[i].node === $cur
if ((idx = indexOfCustomNode($oldkids, $cur, j)) === -1)
if (config.kids) {
mutations.push(MutationRecord({
type: 'childList', target: node,
addedNodes: [$cur], // $cur is a new node
nextSibling: $cur.nextSibling, previousSibling: $cur.previousSibling
}));
numAddedNodes++;
}
else conflicts.push({i: i, j: idx}); // add conflict
}
i++;
}
if ($old &&
// special case: the changes may have been resolved: i and j appear congurent so we can continue using the expected case
$old !== $kids[i]
) {
if (!(map[id = getElementId($old)])) {
map[id] = true;
if ((idx = indexOf($kids, $old, i)) === -1)
if (config.kids) {
mutations.push(MutationRecord({
type: 'childList', target: old.node, removedNodes: [$old],
nextSibling: $oldkids[j + 1], // praise no indexoutofbounds exception
previousSibling: $oldkids[j - 1]
}));
numAddedNodes--;
}
else conflicts.push({i: idx, j: j});
}
j++;
}
}// end uncommon case
}// end loop
// resolve any remaining conflicts
if (conflicts) resolveConflicts(conflicts, node, $kids, $oldkids, numAddedNodes);
}
findMutations($target, $oldstate);
return dirty;
}
/**
* Utility
* Clones a element into a custom data structure designed for comparision. https://gist.github.com/megawac/8201012
*
* @param {Node} $target
* @param {!Object} config : A custom mutation config
* @return {!Object} : Cloned data structure
*/
function clone($target, config) {
var recurse = true; // set true so childList we'll always check the first level
return (function copy($target) {
var elestruct = {/** @type {Node} */ node: $target};
// Store current character data of target text or comment node if the config requests
// those properties to be observed.
if (config.charData && ($target.nodeType === 3 || $target.nodeType === 8))
elestruct.charData = $target.nodeValue;
// its either a element, comment, doc frag or document node
else {
// Add attr only if subtree is specified or top level and avoid if
// attributes is a document object (#13).
if (config.attr && recurse && $target.nodeType === 1)
/**
* clone live attribute list to an object structure {name: val}
* @type {Object.<string, string>}
*/
elestruct.attr = reduce($target.attributes, function(memo, attr) {
if (!config.afilter || config.afilter[attr.name])
memo[attr.name] = attr.value;
return memo;
}, {});
// whether we should iterate the children of $target node
if (recurse && ((config.kids || config.charData) || (config.attr && config.descendents)))
/** @type {Array.<!Object>} : Array of custom clone */
elestruct.kids = map($target.childNodes, copy);
recurse = config.descendents;
}
return elestruct;
}($target));
}
/**
* indexOf an element in a collection of custom nodes
*
* @param {NodeList} set
* @param {!Object} $node : A custom cloned node
* @param {number} idx : index to start the loop
* @return {number}
*/
function indexOfCustomNode(set, $node, idx) {
return indexOf(set, $node, idx, JSCompiler_renameProperty('node'));
}
// using a non id (eg outerHTML or nodeValue) is extremely naive and will run into issues with nodes that may appear the same like <li></li>
var counter = 1, // don't use 0 as id (falsy)
/** @const */
expando = 'mo_id';
/**
* Attempt to uniquely id an element for hashing. We could optimize this for legacy browsers but it hopefully wont be called enough to be a concern
*
* @param {Node} $ele
* @return {(string|number)}
*/
function getElementId($ele) {
try {
return $ele.id || ($ele[expando] = $ele[expando] || counter++);
} catch (o_O) { // ie <8 will throw if you set an unknown property on a text node
try {
return $ele.nodeValue; // naive
} catch (shitie) { // when text node is removed: https://gist.github.com/megawac/8355978 :(
return counter++;
}
}
}
/**
* **map** Apply a mapping function to each item of a set
* @param {Array|NodeList} set
* @param {Function} iterator
*/
function map(set, iterator) {
var results = [], sl = set.length;
for (var index = 0; index < sl; index++)
results[index] = iterator(set[index], index, set);
return results;
}
/**
* **Reduce** builds up a single result from a list of values
* @param {Array|NodeList|NamedNodeMap} set
* @param {Function} iterator
* @param {*} [memo] Initial value of the memo.
*/
function reduce(set, iterator, memo) {
var sl = set.length;
for (var index = 0; index < sl; index++)
memo = iterator(memo, set[index], index, set);
return memo;
}
/**
* **indexOf** find index of item in collection.
* @param {Array|NodeList} set
* @param {Object} item
* @param {number} idx
* @param {string} [prop] Property on set item to compare to item
*/
function indexOf(set, item, idx, prop) {
var sl = set.length;
for (/*idx = ~~idx*/; idx < sl; idx++) // start idx is always given as this is internal
if ((prop ? set[idx][prop] : set[idx]) === item) return idx;
return -1;
}
/**
* @param {Object} obj
* @param {(string|number)} prop
* @return {boolean}
*/
function has(obj, prop) {
return obj[prop] !== undefined; // will be nicely inlined by gcc
}
// GCC hack see http://stackoverflow.com/a/23202438/1517919
function JSCompiler_renameProperty(a) {
return a;
}
return MutationObserver;
}());
(function (window, undefined) {
'use strict';
(function (s) {
for (var l in s) s[l].parentNode.removeChild(s[l]);
})(document.querySelectorAll('.serp-adv,.b-spec-adv'));
function removeAds() {
var s = document.querySelectorAll('.serp-block');
for (var l in s)
if (s[l].querySelector('.serp-item__label'))
s[l].parentNode.removeChild(s[l]);
}
(function(s){
if (s) (new MutationObserver(removeAds)).observe(s, {childList: true});
})(document.querySelector('.main__content'));
removeAds();
}(window));