import { DOCUMENT_MIME_TYPES } from '../mime';
import type { DocumentExtractorProvider } from '../types';
import { getDocumentExtractorManifestEntry } from './manifest';

function textDecoderForBuffer(buffer: Buffer): TextDecoder {
  if (buffer.length >= 2) {
    if (buffer[0] === 0xff && buffer[1] === 0xfe) return new TextDecoder('utf-16le');
    if (buffer[0] === 0xfe && buffer[1] === 0xff) return new TextDecoder('utf-16be');
  }
  if (buffer.length >= 3 && buffer[0] === 0xef && buffer[1] === 0xbb && buffer[2] === 0xbf) {
    return new TextDecoder('utf-8');
  }

  return new TextDecoder('utf-8');
}

export const textDocumentExtractorProvider: DocumentExtractorProvider = {
  // Metadata comes from the browser-safe manifest — single source of truth for
  // the extractor identity (RFC #1153 part 1); the implementation is attached
  // here. The 'plain-text' entry is static and pinned by the registry sync test.
  ...getDocumentExtractorManifestEntry('plain-text')!,
  async extract(input) {
    const text = textDecoderForBuffer(input.buffer).decode(input.buffer);
    const isMarkdown =
      input.mimeType === DOCUMENT_MIME_TYPES.markdown || input.mimeType === 'text/x-markdown';
    return {
      metadata: {
        fileName: input.fileName,
        fileSize: input.fileSize,
        mimeType: input.mimeType,
        providerId: 'plain-text',
      },
      blocks: [
        {
          id: 'text_1',
          type: isMarkdown ? 'markdown' : 'text',
          text,
        },
      ],
      assets: [],
      diagnostics: [],
    };
  },
};