Nuxt/scripts/crawl.mjs

121 lines
3.3 KiB
JavaScript
Raw Normal View History

import Crawler from 'crawler'
import consola from 'consola'
import { parseURL, withoutTrailingSlash } from 'ufo'
import chalk from 'chalk'
import * as actions from '@actions/core'
import { isCI } from 'std-env'
const logger = consola.withTag('crawler')
const baseURL = withoutTrailingSlash(process.env.BASE_URL || 'https://nuxt.com')
const startingURL = baseURL + '/'
const excludedExtensions = ['svg', 'png', 'jpg', 'sketch', 'ico', 'gif']
const urlsToOmit = ['http://localhost:3000']
// TODO: remove when migrating to Nuxt 3/Docus
const errorsToIgnore = [
'/guide/directory-structure/nuxt.config',
'/guide/directory-structure',
'/guide/directory-structure/app.config',
2022-10-07 20:39:38 +00:00
'/api/configuration/nuxt-config',
'/guide/deploy',
'/guide/features/app-config'
]
// GLOBALS
const urls = new Set([startingURL])
const erroredUrls = new Set()
/**
* @param {string} path Path to check
* @param {string | undefined} referrer The referring page
*/
function queue (path, referrer) {
if (!path) {
const message = chalk.red(`${chalk.bold('✗')} ${referrer} linked to empty href`)
if (isCI) { actions.error(message) }
logger.log(message)
return
}
if (urlsToOmit.some(url => path.startsWith(url))) { return }
const { pathname, origin } = new URL(path, referrer)
// Don't crawl the same page more than once
const url = `${origin}${pathname}`
if (!url || urls.has(url) || !crawler) { return }
// Don't try to visit linked assets (e.g. SVGs)
const extension = url.split('.').pop()
if (extension && excludedExtensions.includes(extension)) { return }
// Don't crawl external URLs
if (origin !== baseURL) { return }
urls.add(url)
crawler.queue(url)
}
const crawler = new Crawler({
maxConnections: 100,
callback (error, res, done) {
const { $ } = res
const { uri } = res.options
// @ts-ignore
const { statusCode } = res.request.response
if (error || ![200, 301, 302].includes(statusCode) || !$) {
if (errorsToIgnore.includes(parseURL(uri).pathname)) {
const message = chalk.gray(`${chalk.bold('✗')} ${uri} (${statusCode}) (ignored)`)
logger.log(message)
return done()
}
const message = chalk.red(`${chalk.bold('✗')} ${uri} (${statusCode})`)
if (isCI) { actions.error(message) }
logger.log(message)
erroredUrls.add(uri)
return done()
}
if (!$) {
const message = `Could not parse HTML for ${uri}`
logger.error(message)
if (isCI) { actions.warning(message) }
return done()
}
$('a:not([href*=mailto]):not([href*=tel])').each((_, el) => {
if ('attribs' in el && 'href' in el.attribs) {
queue(el.attribs.href, uri)
}
})
logger.success(chalk.green(uri))
logger.debug(uri, `[${crawler.queueSize} / ${urls.size}]`)
if (!isCI && crawler.queueSize === 1) {
logger.log('')
logger.info(`Checked \`${urls.size}\` pages.`)
// Tasks to run at the end.
if (erroredUrls.size) {
const message = `${chalk.bold(erroredUrls.size)} errors found on ${chalk.bold(baseURL)}.`
const error = new Error(`\n\n${message}\n`)
error.message = message
error.stack = ''
throw error
}
}
done()
}
})
logger.log('')
logger.info(`Checking \`${baseURL}\`.`)
logger.info(`Ignoring file extensions: \`${excludedExtensions.join(', ')}.\`\n`)
crawler.queue(startingURL)