Japscan not loading images dirty fix (#2315) (#2404)

Japscan not loading images dirty fix (#2315)
This commit is contained in:
morcefaster 2020-03-11 17:04:28 +02:00 committed by GitHub
parent 1285139a67
commit 98604a69cb
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 70 additions and 4 deletions

View File

@ -5,7 +5,7 @@ ext {
appName = 'Tachiyomi: Japscan' appName = 'Tachiyomi: Japscan'
pkgNameSuffix = 'fr.japscan' pkgNameSuffix = 'fr.japscan'
extClass = '.Japscan' extClass = '.Japscan'
extVersionCode = 16 extVersionCode = 18
libVersion = '1.2' libVersion = '1.2'
} }

View File

@ -30,6 +30,24 @@ class Japscan : ParsedHttpSource() {
override val supportsLatest = true override val supportsLatest = true
val keysheetChapterUrl1 = "https://www.japscan.co/lecture-en-ligne/bloody-kiss/volume-1/"
val keysheetChapterUrl2= "https://www.japscan.co/lecture-en-ligne/1-nen-a-gumi-no-monster/1/"
var keysheet = "0123456789abcdefghijklmnopqrstuvwxyz"
val realpage_bk_2 = "221518b2f9a20842e4c5379223a5881255d2e522c32277923242520284a542f28085b952e422b1e5a97216b262e5e6621535a22/552c5c112f3e2c4f2e122f025517294925042f322d54513e25822c7e2ev0274751116a4266386a0d56w2506a578b562f6exd2e6/65f435e996fs95fw1160125x715062eve104e33va38454bu632x1315b4fxa10244e2c438932ta38v1158d4e323cx237x933u435/1038xa1"
val realpage_bk_4 = "a336f98370e3d9130516d873045639a35633468374c3389353f36313759623633176406345c3e2e60033e713c37687239616b3f6732612a3d4437/5738213e1d642b3d5a3f113d4e3b666c423e973a8d3bja3a50662772597a4270196akc6b74609a613277l1337d6c0776ge60k3220735l85410640/33a25325d2a927b446al56al36fi8672e4dl120113dj7205543ja455354i847lc436955lf26315739529643h744j6249c5e4e40l34el14ci74822/49l12"
val realpage_bk_14 = "4033163017801660626355d0118326c0c3202380719055b0e070e020a2e3d0c04883e750d2a029f397704470f0e3a4208303f0b370a37930816022/f0097038e3a9a0e2c0781021d0f303a120f6d075b05nd0e27379d48204b164f8a36od374a3c6f3e0b4fpa0841357f44kc3co798740c8c02p924853/e7b0398052d996e491534p331p331m8319616p6958d01n8922c1cn2122827m915pb1e372ep29a0e2d0c24681el51dna926f291714p71fpe1am2189/01fpb9"
val realpage_bk_29 = "e952450996d945591172e4b9e0528549828912293089e469a939a939c172c9d997b276f991c9a842566973f96922e359c222f9b21/932984980e931e908d98742f819f179a7a9d0c972e2a07975794419az6931c248032173809377523a028322b51279a3fbf903e2d6/03dw72bae8a6399bd1f7f26bf8d7196za85140fz00d121ey504b0092d14b581971e9b1e5508x408z8815213080eb60ab30ay30c87/0cbd8"
val realpage_monster_1="d7b023a0f83391903313f630b090311054d04910118397b0c140003368b3e3d343e0a390a76032c0d990f16387104953d110405378d030/4388f0d7a052f05820e853203377a04760e06034a0c5b019c05p805223198472d41174f863aqe394f3e6c32014ar30642387f41m13dqe9/f8a0f7700rf95890era9a9e26211a1a29062dqb12ra1fp319p993r11fq31bp9936115pd1421214819pb9c8410p396qb1e2d14qe1ap99a8/603rc9"
val realPageUrls = listOf(
realpage_bk_2,
realpage_bk_4,
realpage_bk_14,
realpage_bk_29,
realpage_monster_1
)
override val client: OkHttpClient = network.cloudflareClient.newBuilder().addInterceptor { chain -> override val client: OkHttpClient = network.cloudflareClient.newBuilder().addInterceptor { chain ->
val indicator = "&decodeImage" val indicator = "&decodeImage"
@ -57,6 +75,17 @@ class Japscan : ParsedHttpSource() {
} }
} }
fun loadKeysheetChapters() {
var response = client.newCall(GET(keysheetChapterUrl1, headers)).execute()
var doc = response.asJsoup()
response = client.newCall(GET(keysheetChapterUrl2, headers)).execute()
var doc2 = response.asJsoup()
// 「MULTI-DOCUMENT DRIFTING!!」
createKeysheet(doc, doc2)
}
//Popular //Popular
override fun popularMangaRequest(page: Int): Request { override fun popularMangaRequest(page: Int): Request {
return GET("$baseUrl/mangas/", headers) return GET("$baseUrl/mangas/", headers)
@ -195,19 +224,56 @@ class Japscan : ParsedHttpSource() {
} }
} }
fun createKeysheet(doc: Document, doc2: Document) {
val pageUrls = mutableListOf<String>()
var pages = doc.select("select#pages").first()?.select("option")!!; // if this is null we're done
for (i in listOf(1,3,13,28)) {
pageUrls.add(pages[i].attr("data-img").substring("https://c.japscan.co/".length, pages[i].attr("data-img").length - 4))
}
pages = doc2.select("select#pages").first()?.select("option")!!; // if this is null we're done
for (i in listOf(0)) {
pageUrls.add(pages[i].attr("data-img").substring("https://c.japscan.co/".length, pages[i].attr("data-img").length - 4))
}
var az = "0123456789abcdefghijklmnopqrstuvwxyz".toCharArray()
var ks = "0123456789abcdefghijklmnopqrstuvwxyz".toCharArray()
for (i in 0 until realPageUrls.count())
for (j in 0 until realPageUrls[i].length) {
if(realPageUrls[i][j] != pageUrls[i][j]) {
ks[az.indexOf(pageUrls[i][j])] = realPageUrls[i][j]
}
}
keysheet = ks.joinToString("")
}
override fun pageListParse(document: Document): List<Page> { override fun pageListParse(document: Document): List<Page> {
loadKeysheetChapters()
val pages = mutableListOf<Page>() val pages = mutableListOf<Page>()
var imagePath = "(.*\\/).*".toRegex().find(document.select("#image").attr("data-src"))?.groupValues?.get(1) var imagePath = "(.*\\/).*".toRegex().find(document.select("#image").attr("data-src"))?.groupValues?.get(1)
val imageScrambled = if (!document.select("script[src^='/js/iYFbYi_U']").isNullOrEmpty()) "&decodeImage" else "" val imageScrambled = if (!document.select("script[src^='/js/iYFbYi_U']").isNullOrEmpty()) "&decodeImage" else ""
document.select("select#pages").first()?.select("option")?.forEach { document.select("select#pages").first()?.select("option")?.forEach {
if (it.attr("data-img").startsWith("http")) imagePath = "" if (it.attr("data-img").startsWith("http")) imagePath = ""
pages.add(Page(pages.size, "", "$imagePath${it.attr("data-img")}$imageScrambled")) pages.add(Page(pages.size, "", decodeImageUrl("$imagePath${it.attr("data-img")}")+"$imageScrambled"))
} }
return pages return pages
} }
private fun decodeImageUrl(url: String): String {
val az = "0123456789abcdefghijklmnopqrstuvwxyz"
// skip https://, cut after next slash and before extension
var urlBase = url.substring(0, url.indexOf('/', 10)+1)
var extension = url.substring(url.length - 4, url.length)
var encodedPart = url.substring(url.indexOf('/', 10)+1, url.length-4)
return urlBase+encodedPart.map { if (az.indexOf(it) < 0) it else keysheet[az.indexOf(it)]}.joinToString("")+extension
}
override fun imageUrlParse(document: Document): String = "" override fun imageUrlParse(document: Document): String = ""
private fun decodeImage(img: InputStream): ByteArray { private fun decodeImage(img: InputStream): ByteArray {