neo-scraper/src/engines/Shimmie2.ts

55 lines
2.0 KiB
TypeScript

import { ScrapeEngineBase, ScrapeResult, ScrapedPost, ScrapedTag, ScrapeEngineFeature } from "../ScrapeEngine.js";
import { guessContentType } from "../Utility.js";
export default class Shimmie2 extends ScrapeEngineBase {
name = "shimmie2";
features: ScrapeEngineFeature[] = ["content", "tags", "source"];
notes = ["Rating is assumed to be unsafe."];
supportedHosts = ["rule34.paheal.net", "rule34hentai.net"];
scrapeDocument(document: Document): ScrapeResult {
const result = new ScrapeResult(this.name);
const post = new ScrapedPost();
post.pageUrl = document.location.href;
// Set image url
const originalImageElements = Array.from(document.querySelectorAll(".image_info > tbody > tr > td > a"))
.map((x) => x as HTMLAnchorElement)
.filter((x) => x.innerText == "Image Only" || x.innerText == "File Only");
if (originalImageElements.length > 0) {
post.contentUrl = originalImageElements[0].href;
} else {
const dowloadBtnEl = document.querySelector("a[download]") as HTMLAnchorElement;
if (dowloadBtnEl) {
post.contentUrl = dowloadBtnEl.href;
}
}
// Set content type
post.contentType = guessContentType(post.contentUrl);
// Set safety
post.rating = "unsafe"; // Usually it is unsafe
// Set tags
// Shimmie2 doesn't support categories so we don't need to worry about those.
// They also use uppercase tags, so we toLowerCase() those.
// Tags might not match with other booru-likes, but that is for the user to figure out.
// For example by deleting 'wrong' tags or setting up some aliases in szurubooru.
post.tags = Array.from(document.querySelectorAll("a.tag_name"))
.map((x) => (x as HTMLAnchorElement).innerText)
.map((x) => new ScrapedTag(x));
// Set source
const sourceEl = document.querySelector(".image_info > tbody > tr > td > div > a") as HTMLAnchorElement;
if (sourceEl) {
post.sources = [sourceEl.href];
}
result.tryAddPost(post);
return result;
}
}