bmp-project/packages/extractor/index.ts

import { join } from "node:path";
import { dataPath } from "../../deployPaths";
import { jsonError, normalizeStem, securityHeaders, withAuth } from "../../security";

const UPLOAD_PAGE_PATH = join(import.meta.dir, "upload.html");
import { mkdir } from "node:fs/promises";
import { flattenTemplate, extractFields, buildOutput, getNested } from "./extractor";

const TEMPLATE_PATH =
  process.env.BMP_TEMPLATE_PATH ??
  dataPath(join(import.meta.dir, "../templatebuilder/template.json"), "templatebuilder", "template.json");
const OUTPUTS_DIR =
  process.env.BMP_EXTRACTOR_OUTPUTS_DIR ??
  dataPath(join(import.meta.dir, "outputs"), "extractor", "outputs");
const MAX_UPLOAD_BYTES = Number(process.env.BMP_MAX_UPLOAD_BYTES ?? 15 * 1024 * 1024);
const MAX_PDF_PAGES = Number(process.env.BMP_MAX_PDF_PAGES ?? 35);

await mkdir(OUTPUTS_DIR, { recursive: true });

async function loadTemplate(): Promise<Record<string, unknown>> {
  const file = Bun.file(TEMPLATE_PATH);
  if (!(await file.exists())) {
    throw new Error(
      `template.json not found at ${TEMPLATE_PATH}. Run the templatebuilder first.`,
    );
  }
  return file.json();
}

function normalizeFilenamePart(value: string): string {
  return normalizeStem(value);
}

function toSafeFilename(...parts: unknown[]): string {
  const normalizedParts: string[] = [];

  for (const part of parts) {
    if (typeof part !== "string") continue;
    const trimmed = part.trim();
    if (!trimmed) continue;

    const normalized = normalizeFilenamePart(trimmed);
    const previous = normalizedParts[normalizedParts.length - 1];

    if (normalized && normalized !== previous) {
      normalizedParts.push(normalized);
    }
  }

  return normalizedParts.join("_");
}

function extractScalarText(value: unknown): string | undefined {
  if (typeof value === "string") return value.trim() || undefined;
  if (typeof value === "object" && value !== null) {
    const antwort = (value as Record<string, unknown>).antwort;
    if (typeof antwort === "string") return antwort.trim() || undefined;
  }
  return undefined;
}

export const routes = {
  "/extractor": {
    GET: withAuth(async () => new Response(Bun.file(UPLOAD_PAGE_PATH), {
      headers: { ...securityHeaders, "Content-Type": "text/html" },
    }),
    ),
  },

  "/extract": {
    POST: withAuth(async (req: Request) => {
      const contentLength = Number(req.headers.get("content-length") ?? 0);
      if (contentLength > MAX_UPLOAD_BYTES + 4096) {
        return jsonError(`PDF is too large. Maximum size is ${Math.floor(MAX_UPLOAD_BYTES / 1024 / 1024)} MB.`, 413);
      }

      let form;
      try {
        form = await req.formData();
      } catch {
        return Response.json({ error: "Expected multipart/form-data" }, { status: 400 });
      }

      const entry = form.get("pdf");
      if (!(entry instanceof File)) {
        return Response.json({ error: 'Missing "pdf" file field' }, { status: 400 });
      }
      if (!entry.name.toLowerCase().endsWith(".pdf") || entry.size > MAX_UPLOAD_BYTES) {
        return jsonError(`Only PDF uploads up to ${Math.floor(MAX_UPLOAD_BYTES / 1024 / 1024)} MB are allowed.`, 400);
      }

      let template: Record<string, unknown>;
      try {
        template = await loadTemplate();
      } catch (err) {
        return Response.json({ error: String(err) }, { status: 500 });
      }

      const index  = flattenTemplate(template);
      const data   = await entry.arrayBuffer();
      if (!isPdfMagic(data)) {
        return jsonError("Uploaded file is not a valid PDF.", 400);
      }
      // pdfjs transfers the ArrayBuffer to its Worker, detaching it (byteLength → 0).
      // Slice a copy first so we still have the original bytes to write to disk.
      const pdfBytes = data.slice(0);
      let values;
      try {
        values = await extractFields(data, { maxPages: MAX_PDF_PAGES });
      } catch (error) {
        return jsonError(`PDF could not be processed: ${String(error).replace(/^Error:\s*/, "")}`, 400);
      }
      const output = buildOutput(values, index);

      const name = extractScalarText(getNested(output, "kontakt.unternehmen.name"));
      const rechtsform = extractScalarText(getNested(output, "kontakt.unternehmen.rechtsform"));
      const stem =
        name && rechtsform && normalizeFilenamePart(name).endsWith(normalizeFilenamePart(rechtsform))
          ? toSafeFilename(name)
          : toSafeFilename(name, rechtsform) || "output";

      const outDir  = join(OUTPUTS_DIR, stem);
      await mkdir(outDir, { recursive: true });

      const jsonPath = join(outDir, `${stem}.json`);
      const pdfPath  = join(outDir, `${stem}.pdf`);

      await Promise.all([
        Bun.write(jsonPath, JSON.stringify(output, null, 2)),
        Bun.write(pdfPath, pdfBytes),
      ]);

      return Response.json({ ok: true, stem, files: { json: `${stem}.json`, pdf: `${stem}.pdf` } });
    }, {
      csrf: true,
      limit: { key: "extract", max: 12, windowMs: 60_000 },
    }),
  },

  "/template": {
    GET: withAuth(async () => {
      try {
        return Response.json(await loadTemplate());
      } catch (err) {
        return Response.json({ error: String(err) }, { status: 500 });
      }
    }),
  },
} as const;

function isPdfMagic(data: ArrayBuffer): boolean {
  const header = new TextDecoder().decode(new Uint8Array(data.slice(0, 5)));
  return header === "%PDF-";
}