263 lines
8.9 KiB
JavaScript
263 lines
8.9 KiB
JavaScript
//https://github.com/iptv-org/epg/blob/master/sites/telenet.tv/telenet.tv.config.js
|
|
|
|
const axios = require('axios')
|
|
const dayjs = require('dayjs')
|
|
const doFetch = require('@ntlab/sfetch')
|
|
|
|
const API_STATIC_ENDPOINT = 'https://staticqbr-prod-be.gnp.cloud.telenet.tv/eng/web/epg-service-lite/be'
|
|
const API_PROD_ENDPOINT = 'https://spark-prod-be.gnp.cloud.telenet.tv/eng/web/linear-service/v2'
|
|
const API_IMAGE_ENDPOINT = 'https://staticqbr-prod-be.gnp.cloud.telenet.tv/image-service'
|
|
|
|
// a segment holds 6 hours of guide for every channel at once, so it is kept around
|
|
// and shared between all the channels of the same day
|
|
const segments = {}
|
|
|
|
// which day emitted an event, per channel, so a broadcast running past midnight is not repeated
|
|
// by the day it ends in
|
|
const claimed = {}
|
|
|
|
module.exports = {
|
|
site: 'telenet.tv',
|
|
days: 2,
|
|
request: {
|
|
cache: {
|
|
ttl: 60 * 60 * 1000 // 1 hour
|
|
}
|
|
},
|
|
url: function ({ date, channel, segment = 0 }) {
|
|
return `${API_STATIC_ENDPOINT}/${channel.lang}/events/segments/${date.format(
|
|
'YYYYMMDD'
|
|
)}${segment.toString().padStart(2, '0')}0000`
|
|
},
|
|
async parser({ content, channel, date }) {
|
|
const items = await loadItems({ content, channel, date })
|
|
if (!items.length) return []
|
|
|
|
const details = await loadProgramDetails(items, channel)
|
|
|
|
return items.map((item, index) => {
|
|
const detail = details[index] || {}
|
|
|
|
return {
|
|
title: item.title,
|
|
subTitle: detail.episodeName || item.seriesName,
|
|
icon: parseIcon(item),
|
|
image: parseImage(item),
|
|
description: detail.longDescription || detail.shortDescription,
|
|
category: detail.genres,
|
|
actors: detail.actors,
|
|
directors: detail.directors,
|
|
producers: detail.producers,
|
|
season: parseSeason(detail),
|
|
episode: parseEpisode(detail),
|
|
date: parseYear(detail),
|
|
country: detail.countryOfOrigin,
|
|
rating: parseRating(detail, item),
|
|
language: parseLanguages(detail, item),
|
|
subtitles: parseSubtitles(detail, item),
|
|
new: Boolean(detail.premiere || item.premiere),
|
|
start: parseStart(item),
|
|
stop: parseStop(item)
|
|
}
|
|
})
|
|
},
|
|
async channels() {
|
|
const data = await axios
|
|
.get(`${API_PROD_ENDPOINT}/channels?cityId=28001&language=en&productClass=Orion-DASH`)
|
|
.then(r => r.data)
|
|
.catch(console.log)
|
|
|
|
return data.map(item => {
|
|
return {
|
|
lang: 'nl',
|
|
site_id: item.id,
|
|
name: item.name
|
|
}
|
|
})
|
|
}
|
|
}
|
|
|
|
async function loadItems({ content, channel, date }) {
|
|
const urls = [0, 6, 12, 18].map(segment => module.exports.url({ date, channel, segment }))
|
|
|
|
// the first segment has already been downloaded by the grabber itself
|
|
cacheSegment(urls[0], content)
|
|
|
|
const missing = urls.filter(url => segments[url] === undefined)
|
|
if (missing.length) {
|
|
await doFetch(missing, (url, res) => cacheSegment(url, res))
|
|
}
|
|
|
|
const items = urls.flatMap(url => findEvents(segments[url], channel)).filter(isBroadcast)
|
|
|
|
// an event that spans a boundary is listed twice: by both segments, or by both days
|
|
return uniqueItems(items, channel, date)
|
|
}
|
|
|
|
// `res` is left out when a request fails, but only when sfetch is checking results, which is a
|
|
// setting shared with the other sites using it, so every program is built from its event first
|
|
async function loadProgramDetails(items, channel) {
|
|
const details = []
|
|
const queues = items
|
|
.map((item, index) => ({ item, index }))
|
|
.filter(({ item }) => item.id)
|
|
.map(({ item, index }) => ({
|
|
url: `${API_PROD_ENDPOINT}/replayEvent/${item.id}?returnLinearContent=true&language=${channel.lang}`,
|
|
index
|
|
}))
|
|
|
|
if (queues.length) {
|
|
await doFetch(queues, (queue, res) => {
|
|
if (res) details[queue.index] = res
|
|
})
|
|
}
|
|
|
|
return details
|
|
}
|
|
|
|
function cacheSegment(url, content) {
|
|
if (segments[url] !== undefined) return
|
|
const entries = parseEntries(content)
|
|
// a failed download must not be cached, the next channel gets to retry it
|
|
if (entries.length) segments[url] = entries
|
|
}
|
|
|
|
// the grabber hands over the raw response, while axios has already parsed the ones fetched here
|
|
function parseEntries(content) {
|
|
if (!content) return []
|
|
try {
|
|
const data = Array.isArray(content.entries) ? content : JSON.parse(content)
|
|
|
|
return Array.isArray(data.entries) ? data.entries : []
|
|
} catch (err) {
|
|
console.error(`Unable to parse guide: ${err.message}!`)
|
|
|
|
return []
|
|
}
|
|
}
|
|
|
|
function parseStart(item) {
|
|
return dayjs.unix(item.startTime)
|
|
}
|
|
|
|
function parseStop(item) {
|
|
return dayjs.unix(item.endTime)
|
|
}
|
|
|
|
// telenet fills the hours a channel is off air with "Geen uitzending", often several in a row.
|
|
// Nothing is behind them, so they are dropped and the guide simply has a hole there.
|
|
function isBroadcast(item) {
|
|
return !/^geen uitzending\b/i.test(item.title || '')
|
|
}
|
|
|
|
function findEvents(entries, channel) {
|
|
if (!Array.isArray(entries)) return []
|
|
const channelData = entries.find(e => e.channelId === channel.site_id)
|
|
if (!channelData) return []
|
|
|
|
return Array.isArray(channelData.events) ? channelData.events : []
|
|
}
|
|
|
|
// Both guards are needed: the local set drops an event listed by two segments of this day, while
|
|
// the claim drops one the previous day already emitted. Re-parsing a day it claimed itself stays
|
|
// idempotent, so parsing the same day twice keeps returning the same programs.
|
|
function uniqueItems(items, channel, date) {
|
|
const day = date.format('YYYYMMDD')
|
|
const claimedByChannel = claimed[channel.site_id] || (claimed[channel.site_id] = new Map())
|
|
const seen = new Set()
|
|
|
|
return items.filter(item => {
|
|
if (seen.has(item.id)) return false
|
|
|
|
const owner = claimedByChannel.get(item.id)
|
|
if (owner !== undefined && owner !== day) return false
|
|
|
|
seen.add(item.id)
|
|
claimedByChannel.set(item.id, day)
|
|
|
|
return true
|
|
})
|
|
}
|
|
|
|
function parseSeason(detail) {
|
|
if (!detail.seasonNumber) return null
|
|
if (String(detail.seasonNumber).length > 2) return null
|
|
return detail.seasonNumber
|
|
}
|
|
|
|
function parseEpisode(detail) {
|
|
if (!detail.episodeNumber) return null
|
|
if (String(detail.episodeNumber).length > 3) return null
|
|
return detail.episodeNumber
|
|
}
|
|
|
|
function parseYear(detail) {
|
|
if (!detail.productionDate) return null
|
|
if (!/^\d{4}$/.test(String(detail.productionDate))) return null
|
|
return detail.productionDate
|
|
}
|
|
|
|
// the minimum age follows the Kijkwijzer (NICAM) classification
|
|
function parseRating(detail, item) {
|
|
const minimumAge = detail.minimumAge != null ? detail.minimumAge : item.minimumAge
|
|
if (minimumAge == null || minimumAge === '') {
|
|
if (detail.isAdult || item.isAdult) return { system: 'Kijkwijzer', value: '18' }
|
|
return null
|
|
}
|
|
return { system: 'Kijkwijzer', value: String(minimumAge) }
|
|
}
|
|
|
|
function parseLanguages(detail, item) {
|
|
return parseLangCodes(detail.audioLanguages || item.audioLanguages)
|
|
}
|
|
|
|
function parseSubtitles(detail, item) {
|
|
const captions = parseLangCodes(detail.captionLanguages || item.captionLanguages)
|
|
const signed = parseLangCodes(detail.signLanguages || item.signLanguages)
|
|
|
|
return [
|
|
...captions.map(language => ({ language })),
|
|
...signed.map(language => ({ type: 'deaf-signed', language }))
|
|
]
|
|
}
|
|
|
|
// the same language can be listed more than once, once per purpose (e.g. audio description)
|
|
function parseLangCodes(languages) {
|
|
if (!Array.isArray(languages)) return []
|
|
return [...new Set(languages.map(language => language.lang).filter(Boolean))]
|
|
}
|
|
|
|
// Three assets hang off one event id. Only the poster is always there: over a day of the guide,
|
|
// episodeStill answered for 91% of programmes and titleTreatment for 66%, and nothing in the event
|
|
// says which, so all three are emitted and a reader has to expect a 404.
|
|
//
|
|
// Poster first because the DTD asks for the most authoritative image first -- though Telenet's own
|
|
// player prefers the still, its bundle carrying the literal list ['episodeStill','posterTile'].
|
|
const IMAGE_INTENTS = [
|
|
// Key art cropped to 2:3, always from the season or the movie, never per episode.
|
|
{ intent: 'posterTile', type: 'poster', orient: 'P' },
|
|
// 16:9; a real production still for two thirds, season or show artwork for the rest. The app
|
|
// picks the landscape image on `type` alone, so dropping this one hides it there silently.
|
|
{ intent: 'episodeStill', type: 'still', orient: 'L' },
|
|
// A transparent wordmark, which is none of the DTD's five types, so `system` is all it has.
|
|
{ intent: 'titleTreatment', orient: 'L' },
|
|
]
|
|
|
|
function parseIcon(item) {
|
|
return imageUrl(item, 'posterTile')
|
|
}
|
|
|
|
function imageUrl(item, intent) {
|
|
return `${API_IMAGE_ENDPOINT}/intent/${item.id}/${intent}`
|
|
}
|
|
|
|
// <icon> repeats the poster because the two elements are read by different clients: <icon> has only
|
|
// a src, while <image> is the one that can say what the picture is.
|
|
function parseImage(item) {
|
|
return IMAGE_INTENTS.map(({ intent, type, orient }) => ({
|
|
...(type ? { type } : {}),
|
|
orient,
|
|
system: intent,
|
|
value: imageUrl(item, intent),
|
|
}))
|
|
} |