mirror of
https://github.com/NikolaiT/se-scraper.git
synced 2024-11-22 23:53:09 +01:00
198 lines
7.7 KiB
JavaScript
198 lines
7.7 KiB
JavaScript
const se_scraper = require('./../index.js');
|
|
var assert = require('chai').assert;
|
|
|
|
/*
|
|
* Use chai and mocha for tests.
|
|
* https://mochajs.org/#installation
|
|
*/
|
|
|
|
const normal_search_keywords = ['apple tree', 'weather tomorrow'];
|
|
|
|
async function normal_search_test() {
|
|
let config = {
|
|
compress: false,
|
|
debug_level: 1,
|
|
keyword_file: '',
|
|
headless: true,
|
|
output_file: '',
|
|
block_assets: true,
|
|
user_agent: 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/70.0.3538.110 Safari/537.36',
|
|
random_user_agent: false,
|
|
};
|
|
|
|
let scrape_config = {
|
|
search_engine: 'google',
|
|
keywords: normal_search_keywords,
|
|
num_pages: 3,
|
|
};
|
|
|
|
console.log('normal_search_test()');
|
|
normal_search_test_case( await se_scraper.scrape(config, scrape_config) );
|
|
}
|
|
|
|
// we test with a callback function to our handler
|
|
function normal_search_test_case(response) {
|
|
assert.equal(response.headers['Content-Type'], 'text/json', 'content type is not text/json');
|
|
assert.equal(response.statusCode, 200, 'status code must be 200');
|
|
assert.equal(response.metadata.num_requests, 6);
|
|
|
|
for (let query in response.results) {
|
|
let total_rank = 1;
|
|
|
|
assert.containsAllKeys(response.results, normal_search_keywords, 'not all keywords were scraped.');
|
|
|
|
for (let page_number in response.results[query]) {
|
|
|
|
assert.isNumber(parseInt(page_number), 'page_number must be numeric');
|
|
|
|
let obj = response.results[query][page_number];
|
|
|
|
assert.containsAllKeys(obj, ['results', 'time', 'no_results', 'num_results', 'effective_query'], 'not all keys are in the object');
|
|
|
|
assert.isAtLeast(obj.results.length, 7, 'results must have at least 8 SERP objects');
|
|
assert.equal(obj.no_results, false, 'no results should be false');
|
|
assert.typeOf(obj.num_results, 'string', 'num_results must be a string');
|
|
assert.isAtLeast(obj.num_results.length, 5, 'num_results should be a string of at least 5 chars');
|
|
assert.typeOf(Date.parse(obj.time), 'number', 'time should be a valid date');
|
|
|
|
for (let res of obj.results) {
|
|
|
|
assert.containsAllKeys(res, ['link', 'title', 'rank', 'visible_link'], 'not all keys are in the SERP object');
|
|
|
|
assert.isOk(res.link, 'link must be ok');
|
|
assert.typeOf(res.link, 'string', 'link must be string');
|
|
assert.isAtLeast(res.link.length, 5, 'link must have at least 5 chars');
|
|
|
|
assert.isOk(res.visible_link, 'visible_link must be ok');
|
|
assert.typeOf(res.visible_link, 'string', 'visible_link must be string');
|
|
assert.isAtLeast(res.visible_link.length, 5, 'visible_link must have at least 5 chars');
|
|
|
|
assert.isOk(res.title, 'title must be ok');
|
|
assert.typeOf(res.title, 'string', 'title must be string');
|
|
assert.isAtLeast(res.title.length, 10, 'title must have at least 10 chars');
|
|
|
|
assert.isOk(res.snippet, 'snippet must be ok');
|
|
assert.typeOf(res.snippet, 'string', 'snippet must be string');
|
|
assert.isAtLeast(res.snippet.length, 10, 'snippet must have at least 10 chars');
|
|
|
|
assert.isNumber(res.rank, 'rank must be integer');
|
|
assert.equal(res.rank, total_rank++, 'rank ist wrong');
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
const keywords_no_results = ['fgskl34440abJAksafkl34a44dsflkjaQQuBBdfk',];
|
|
|
|
async function no_results_test() {
|
|
let config = {
|
|
compress: false,
|
|
debug_level: 1,
|
|
keyword_file: '',
|
|
headless: true,
|
|
output_file: '',
|
|
block_assets: true,
|
|
user_agent: 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/70.0.3538.110 Safari/537.36',
|
|
random_user_agent: false,
|
|
};
|
|
|
|
let scrape_config = {
|
|
search_engine: 'google',
|
|
keywords: keywords_no_results,
|
|
num_pages: 1,
|
|
};
|
|
|
|
console.log('no_results_test()');
|
|
test_case_no_results( await se_scraper.scrape(config, scrape_config) );
|
|
}
|
|
|
|
// we test with a callback function to our handler
|
|
function test_case_no_results(response) {
|
|
assert.equal(response.headers['Content-Type'], 'text/json', 'content type is not text/json');
|
|
assert.equal(response.statusCode, 200, 'status code must be 200');
|
|
assert.equal(response.metadata.num_requests, 1);
|
|
|
|
results = response.results;
|
|
for (let query in response.results) {
|
|
|
|
assert.containsAllKeys(response.results, keywords_no_results, 'not all keywords were scraped.');
|
|
|
|
for (let page_number in response.results[query]) {
|
|
|
|
assert.isNumber(parseInt(page_number), 'page_number must be numeric');
|
|
|
|
let obj = response.results[query][page_number];
|
|
|
|
assert.containsAllKeys(obj, ['results', 'time', 'no_results', 'num_results', 'effective_query'], 'not all keys are in the object');
|
|
|
|
assert(obj.results.length === 0, 'results must have 0 SERP objects');
|
|
assert.equal(obj.no_results, true, 'no results should be true');
|
|
assert.isEmpty(obj.num_results, 'no results should be a empty string');
|
|
assert.typeOf(Date.parse(obj.time), 'number', 'time should be a valid date');
|
|
}
|
|
}
|
|
}
|
|
|
|
const effective_query_keywords = ['mount evverrest'];
|
|
|
|
async function effective_query_test() {
|
|
let config = {
|
|
compress: false,
|
|
debug_level: 1,
|
|
keyword_file: '',
|
|
headless: true,
|
|
output_file: '',
|
|
block_assets: true,
|
|
user_agent: 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/70.0.3538.110 Safari/537.36',
|
|
random_user_agent: false,
|
|
};
|
|
|
|
let scrape_config = {
|
|
search_engine: 'google',
|
|
keywords: effective_query_keywords,
|
|
num_pages: 1,
|
|
};
|
|
|
|
console.log('effective_query_test()');
|
|
test_case_effective_query( await se_scraper.scrape(config, scrape_config) );
|
|
}
|
|
|
|
// we test with a callback function to our handler
|
|
function test_case_effective_query(response) {
|
|
assert.equal(response.headers['Content-Type'], 'text/json', 'content type is not text/json');
|
|
assert.equal(response.statusCode, 200, 'status code must be 200');
|
|
assert.equal(response.metadata.num_requests, 1);
|
|
|
|
results = response.results;
|
|
for (let query in response.results) {
|
|
|
|
assert.containsAllKeys(response.results, effective_query_keywords, 'not all keywords were scraped.');
|
|
|
|
for (let page_number in response.results[query]) {
|
|
|
|
assert.isNumber(parseInt(page_number), 'page_number must be numeric');
|
|
|
|
let obj = response.results[query][page_number];
|
|
|
|
assert.containsAllKeys(obj, ['results', 'time', 'no_results', 'num_results', 'effective_query'], 'not all keys are in the object');
|
|
|
|
// effective query must be different to the original keyword
|
|
assert.isOk(obj.effective_query, 'effective query must be ok');
|
|
assert.isNotEmpty(obj.effective_query, 'effective query must be valid');
|
|
assert(obj.effective_query !== query, 'effective query must be different from keyword');
|
|
|
|
assert.isAtLeast(obj.results.length, 7, 'results must have at least 8 SERP objects');
|
|
assert.equal(obj.no_results, false, 'no results should be false');
|
|
assert.typeOf(obj.num_results, 'string', 'num_results must be a string');
|
|
assert.isAtLeast(obj.num_results.length, 5, 'num_results should be a string of at least 5 chars');
|
|
assert.typeOf(Date.parse(obj.time), 'number', 'time should be a valid date');
|
|
}
|
|
}
|
|
}
|
|
|
|
(async () => {
|
|
await normal_search_test();
|
|
await no_results_test();
|
|
await effective_query_test();
|
|
})();
|