/**
 * Layout prompt and decoder contract shared with okraPDF's parser-gemini
 * implementation. Kept in this standalone package so a public checkout can
 * install and build without monorepo workspace dependencies.
 *
 * Derived from ParseBench and @okrapdf/parser-gemini (MIT).
 */

export type DecodedLayoutBlock = {
  bbox: [number, number, number, number];
  label: string;
  text: string;
  page?: number;
};

const DIV_BLOCK = /<div\b([^>]*)>([\s\S]*?)<\/div>/gi;
const DATA_BBOX = /\bdata-bbox=["']([^"']+)["']/i;
const DATA_LABEL = /\bdata-label=["']([^"']+)["']/i;
const CLASS_LABEL = /\bclass=["']([^"']+)["']/i;
const DATA_PAGE = /data-page=["'](\d+)["']/i;

function tryParseBbox(raw: string): [number, number, number, number] | null {
  const value = raw.trim();
  try {
    const parsed: unknown = JSON.parse(value.startsWith('[') ? value : `[${value}]`);
    if (
      Array.isArray(parsed) &&
      parsed.length === 4 &&
      parsed.every((number) => typeof number === 'number' && Number.isFinite(number))
    ) {
      return parsed as [number, number, number, number];
    }
  } catch {
    // Ignore malformed blocks while continuing to decode later valid blocks.
  }
  return null;
}

export function parseLayoutBlocks(content: string): DecodedLayoutBlock[] {
  const blocks: DecodedLayoutBlock[] = [];
  for (const match of content.matchAll(DIV_BLOCK)) {
    const attributes = match[1];
    const bboxMatch = DATA_BBOX.exec(attributes);
    const labelMatch = DATA_LABEL.exec(attributes) ?? CLASS_LABEL.exec(attributes);
    if (!bboxMatch || !labelMatch) continue;
    const bbox = tryParseBbox(bboxMatch[1]);
    if (!bbox) continue;
    const pageMatch = DATA_PAGE.exec(attributes);
    const page = pageMatch ? Number.parseInt(pageMatch[1], 10) : undefined;
    blocks.push({
      bbox,
      label: labelMatch[1].trim().split(/\s+/)[0],
      text: match[2].trim(),
      ...(page && page > 0 ? { page } : {}),
    });
  }
  return blocks;
}

export function swapGeminiBbox(blocks: DecodedLayoutBlock[]): DecodedLayoutBlock[] {
  return blocks.map((block) => {
    const [yMin, xMin, yMax, xMax] = block.bbox;
    return { ...block, bbox: [xMin, yMin, xMax, yMax] };
  });
}

export function itemsToMarkdown(blocks: DecodedLayoutBlock[]): string {
  const parts: string[] = [];
  for (const block of blocks) {
    const label = block.label.toLowerCase();
    if (!block.text) continue;
    if (label === 'title') parts.push(`# ${block.text}`);
    else if (label === 'section-header' || label === 'section_header') {
      parts.push(`## ${block.text}`);
    } else if (label === 'formula') {
      parts.push(`$$\n${block.text}\n$$`);
    } else {
      parts.push(block.text);
    }
  }
  return parts.join('\n\n');
}

const LABEL_MAP: Record<string, string> = {
  caption: 'Caption',
  footnote: 'Footnote',
  formula: 'Formula',
  'list-item': 'List-item',
  list_item: 'List-item',
  'page-footer': 'Page-footer',
  page_footer: 'Page-footer',
  'page-header': 'Page-header',
  page_header: 'Page-header',
  picture: 'Picture',
  figure: 'Picture',
  'section-header': 'Section-header',
  section_header: 'Section-header',
  table: 'Table',
  text: 'Text',
  title: 'Title',
};

export function canonicalLabel(raw: string): string {
  return LABEL_MAP[raw.toLowerCase()] ?? raw;
}

export const SYSTEM_PROMPT_LAYOUT =
  'You are a document parser. Your task is to convert ' +
  'document images to clean, well-structured markdown.' +
  '\n\nGuidelines:\n' +
  '- Preserve the document structure ' +
  '(headings, paragraphs, lists, tables)\n' +
  '- Convert tables to HTML format ' +
  '(<table>, <tr>, <th>, <td>)\n' +
  '- For existing tables in the document: use colspan ' +
  'and rowspan attributes to preserve merged cells ' +
  'and hierarchical headers\n' +
  '- For charts/graphs being converted to tables: use ' +
  'flat combined column headers (e.g., ' +
  '"Primary 2015" not separate rows) so each data ' +
  "cell's row contains all its labels\n" +
  '- Describe images/figures briefly in square brackets ' +
  'like [Figure: description]\n' +
  '- Preserve any code blocks with appropriate syntax ' +
  'highlighting\n' +
  '- Maintain reading order (left-to-right, ' +
  'top-to-bottom for Western documents)\n' +
  '- Do not add commentary or explanations ' +
  '- only output the parsed content' +
  '\n\n' +
  'Additionally, wrap each layout element in a <div> tag with:\n' +
  '- data-bbox="[x1, y1, x2, y2]" — bounding box in normalized 0-1000 ' +
  'coordinates where x is horizontal (left edge = 0, right edge = 1000) ' +
  'and y is vertical (top = 0, bottom = 1000). ' +
  'x1,y1 is the top-left corner and x2,y2 is the bottom-right corner.\n' +
  '- data-label="<category>" — one of: Caption, Footnote, Formula, ' +
  'List-item, Page-footer, Page-header, Picture, Section-header, ' +
  'Table, Text, Title\n\n' +
  'Place elements in reading order. Every piece of content must be ' +
  'inside exactly one <div> wrapper.';

export const USER_PROMPT_LAYOUT =
  'Parse this document page and output its content as ' +
  'clean markdown, with each layout element wrapped in a ' +
  '<div data-bbox="[x1,y1,x2,y2]" data-label="Category"> tag. ' +
  'Use HTML tables for any tabular data. ' +
  'For charts/graphs, use flat combined column headers. ' +
  'Output ONLY the parsed content with div wrappers, ' +
  'no explanations.';

export const SYSTEM_PROMPT_LAYOUT_GEMINI = SYSTEM_PROMPT_LAYOUT.replace(
  '"[x1, y1, x2, y2]" — bounding box in normalized 0-1000 ' +
    'coordinates where x is horizontal (left edge = 0, right edge = 1000) ' +
    'and y is vertical (top = 0, bottom = 1000). ' +
    'x1,y1 is the top-left corner and x2,y2 is the bottom-right corner.',
  '"[y_min, x_min, y_max, x_max]" — bounding box in normalized 0-1000 ' +
    'coordinates where x is horizontal (left edge = 0, right edge = 1000) ' +
    'and y is vertical (top = 0, bottom = 1000). ' +
    'The order is [y_min, x_min, y_max, x_max].',
);

export const USER_PROMPT_LAYOUT_GEMINI = USER_PROMPT_LAYOUT.replace(
  '[x1,y1,x2,y2]',
  '[y_min,x_min,y_max,x_max]',
);