mirror of
https://github.com/iptv-org/database.git
synced 2024-11-21 10:32:32 -05:00
257 lines
7.7 KiB
TypeScript
257 lines
7.7 KiB
TypeScript
import { Collection, Storage, File, Dictionary, Logger } from '@freearhey/core'
|
|
import { DATA_DIR } from '../constants'
|
|
import { program } from 'commander'
|
|
import Joi from 'joi'
|
|
import { CSVParser, IDCreator } from '../core'
|
|
import chalk from 'chalk'
|
|
|
|
program.argument('[filepath]', 'Path to file to validate').parse(process.argv)
|
|
|
|
const logger = new Logger()
|
|
const buffer = new Dictionary()
|
|
const files = new Dictionary()
|
|
const schemes: { [key: string]: object } = require('../schemes')
|
|
|
|
async function main() {
|
|
const dataStorage = new Storage(DATA_DIR)
|
|
const _files = await dataStorage.list('*.csv')
|
|
let globalErrors = new Collection()
|
|
const parser = new CSVParser()
|
|
|
|
for (const filepath of _files) {
|
|
const file = new File(filepath)
|
|
if (file.extension() !== 'csv') continue
|
|
|
|
const csv = await dataStorage.load(file.basename())
|
|
|
|
const rows = csv.split(/\r\n/)
|
|
const headers = rows[0].split(',')
|
|
for (const [i, line] of rows.entries()) {
|
|
if (!line.trim()) continue
|
|
if (line.indexOf('\n') > -1)
|
|
return handleError(
|
|
`Error: row ${i + 1} has the wrong line ending character, should be CRLF (${filepath})`
|
|
)
|
|
if (line.split(/,(?=(?:(?:[^"]*"){2})*[^"]*$)/).length !== headers.length)
|
|
return handleError(`Error: row ${i + 1} has the wrong number of columns (${filepath})`)
|
|
}
|
|
|
|
const data = await parser.parse(csv)
|
|
const filename = file.name()
|
|
|
|
let grouped
|
|
switch (filename) {
|
|
case 'blocklist':
|
|
grouped = data.keyBy(item => item.channel)
|
|
break
|
|
case 'categories':
|
|
case 'channels':
|
|
grouped = data.keyBy(item => item.id)
|
|
break
|
|
default:
|
|
grouped = data.keyBy(item => item.code)
|
|
break
|
|
}
|
|
|
|
buffer.set(filename, grouped)
|
|
files.set(filename, data)
|
|
}
|
|
|
|
const filesToCheck = program.args.length ? program.args : _files
|
|
for (const filepath of filesToCheck) {
|
|
const file = new File(filepath)
|
|
const filename = file.name()
|
|
if (!schemes[filename]) return handleError(`Error: "${filename}" scheme is missing`)
|
|
|
|
const rows: Collection = files.get(filename)
|
|
const rowsCopy = JSON.parse(JSON.stringify(rows.all()))
|
|
|
|
let fileErrors = new Collection()
|
|
switch (filename) {
|
|
case 'channels':
|
|
fileErrors = fileErrors.concat(findDuplicatesBy(rowsCopy, 'id'))
|
|
for (const [i, row] of rowsCopy.entries()) {
|
|
fileErrors = fileErrors.concat(validateChannelId(row, i))
|
|
fileErrors = fileErrors.concat(validateChannelBroadcastArea(row, i))
|
|
fileErrors = fileErrors.concat(
|
|
checkValue(i, row, 'id', 'subdivision', buffer.get('subdivisions'))
|
|
)
|
|
fileErrors = fileErrors.concat(
|
|
checkValue(i, row, 'id', 'categories', buffer.get('categories'))
|
|
)
|
|
fileErrors = fileErrors.concat(
|
|
checkValue(i, row, 'id', 'replaced_by', buffer.get('channels'))
|
|
)
|
|
fileErrors = fileErrors.concat(
|
|
checkValue(i, row, 'id', 'languages', buffer.get('languages'))
|
|
)
|
|
fileErrors = fileErrors.concat(
|
|
checkValue(i, row, 'id', 'country', buffer.get('countries'))
|
|
)
|
|
}
|
|
break
|
|
case 'blocklist':
|
|
for (const [i, row] of rowsCopy.entries()) {
|
|
fileErrors = fileErrors.concat(validateChannel(row.channel, i))
|
|
}
|
|
break
|
|
case 'countries':
|
|
fileErrors = fileErrors.concat(findDuplicatesBy(rowsCopy, 'code'))
|
|
for (const [i, row] of rowsCopy.entries()) {
|
|
fileErrors = fileErrors.concat(
|
|
checkValue(i, row, 'code', 'languages', buffer.get('languages'))
|
|
)
|
|
}
|
|
break
|
|
case 'subdivisions':
|
|
fileErrors = fileErrors.concat(findDuplicatesBy(rowsCopy, 'code'))
|
|
for (const [i, row] of rowsCopy.entries()) {
|
|
fileErrors = fileErrors.concat(
|
|
checkValue(i, row, 'code', 'country', buffer.get('countries'))
|
|
)
|
|
}
|
|
break
|
|
case 'regions':
|
|
fileErrors = fileErrors.concat(findDuplicatesBy(rowsCopy, 'code'))
|
|
for (const [i, row] of rowsCopy.entries()) {
|
|
fileErrors = fileErrors.concat(
|
|
checkValue(i, row, 'code', 'countries', buffer.get('countries'))
|
|
)
|
|
}
|
|
break
|
|
case 'categories':
|
|
fileErrors = fileErrors.concat(findDuplicatesBy(rowsCopy, 'id'))
|
|
break
|
|
case 'languages':
|
|
fileErrors = fileErrors.concat(findDuplicatesBy(rowsCopy, 'code'))
|
|
break
|
|
}
|
|
|
|
const schema = Joi.object(schemes[filename])
|
|
rows.all().forEach((row: { [key: string]: string }, i: number) => {
|
|
const { error } = schema.validate(row, { abortEarly: false })
|
|
if (error) {
|
|
error.details.forEach(detail => {
|
|
fileErrors.push({ line: i + 2, row, message: detail.message })
|
|
})
|
|
}
|
|
})
|
|
|
|
if (fileErrors.count()) {
|
|
logger.info(`\n${chalk.underline(filepath)}`)
|
|
fileErrors.forEach(err => {
|
|
const position = err.line.toString().padEnd(6, ' ')
|
|
const id = err.row && err.row.id ? ` ${err.row.id}:` : ''
|
|
logger.info(` ${chalk.gray(position)}${id} ${err.message}`)
|
|
})
|
|
globalErrors = globalErrors.concat(fileErrors)
|
|
}
|
|
}
|
|
|
|
if (globalErrors.count()) return handleError(`${globalErrors.count()} error(s)`)
|
|
}
|
|
|
|
main()
|
|
|
|
function checkValue(
|
|
i: number,
|
|
row: { [key: string]: string[] | string | boolean },
|
|
key: string,
|
|
field: string,
|
|
collection: Collection
|
|
) {
|
|
const errors = new Collection()
|
|
let values: string[] = []
|
|
if (Array.isArray(row[field])) {
|
|
values = row[field] as string[]
|
|
} else if (typeof row[field] === 'string') {
|
|
values = new Array(row[field]) as string[]
|
|
}
|
|
|
|
values.forEach((value: string) => {
|
|
if (collection.missing(value)) {
|
|
errors.push({
|
|
line: i + 2,
|
|
message: `"${row[key]}" has an invalid ${field} "${value}"`
|
|
})
|
|
}
|
|
})
|
|
|
|
return errors
|
|
}
|
|
|
|
function validateChannel(channelId: string, i: number) {
|
|
const errors = new Collection()
|
|
const channels = buffer.get('channels')
|
|
|
|
if (channels.missing(channelId)) {
|
|
errors.push({
|
|
line: i + 2,
|
|
message: `"${channelId}" is missing in the channels.csv`
|
|
})
|
|
}
|
|
|
|
return errors
|
|
}
|
|
|
|
function findDuplicatesBy(rows: { [key: string]: string }[], key: string) {
|
|
const errors = new Collection()
|
|
const buffer = new Dictionary()
|
|
|
|
rows.forEach((row, i) => {
|
|
const normId = row[key].toLowerCase()
|
|
if (buffer.has(normId)) {
|
|
errors.push({
|
|
line: i + 2,
|
|
message: `entry with the ${key} "${row[key]}" already exists`
|
|
})
|
|
}
|
|
|
|
buffer.set(normId, true)
|
|
})
|
|
|
|
return errors
|
|
}
|
|
|
|
function validateChannelId(row: { [key: string]: string }, i: number) {
|
|
const errors = new Collection()
|
|
|
|
const expectedId = new IDCreator().create(row.name, row.country)
|
|
|
|
if (expectedId !== row.id) {
|
|
errors.push({
|
|
line: i + 2,
|
|
message: `"${row.id}" must be derived from the channel name "${row.name}" and the country code "${row.country}"`
|
|
})
|
|
}
|
|
|
|
return errors
|
|
}
|
|
|
|
function validateChannelBroadcastArea(row: { [key: string]: string[] }, i: number) {
|
|
const errors = new Collection()
|
|
const regions = buffer.get('regions')
|
|
const countries = buffer.get('countries')
|
|
const subdivisions = buffer.get('subdivisions')
|
|
|
|
row.broadcast_area.forEach((areaCode: string) => {
|
|
const [type, code] = areaCode.split('/')
|
|
if (
|
|
(type === 'r' && regions.missing(code)) ||
|
|
(type === 'c' && countries.missing(code)) ||
|
|
(type === 's' && subdivisions.missing(code))
|
|
) {
|
|
errors.push({
|
|
line: i + 2,
|
|
message: `"${row.id}" has the wrong broadcast_area "${areaCode}"`
|
|
})
|
|
}
|
|
})
|
|
|
|
return errors
|
|
}
|
|
|
|
function handleError(message: string) {
|
|
logger.error(chalk.red(message))
|
|
process.exit(1)
|
|
}
|