app.py

from fastapi import FastAPI, UploadFile, File
from fastapi.middleware.cors import CORSMiddleware
import fitz
import tempfile
import os
import base64
import re
import json

app = FastAPI()
app.add_middleware(CORSMiddleware, allow_origins=["*"], allow_methods=["*"], allow_headers=["*"])


@app.post("/debug")
async def debug_document(file: UploadFile = File(...)):
    """Show what PyMuPDF sees — helps diagnose image ordering."""
    suffix = os.path.splitext(file.filename)[1].lower()
    with tempfile.NamedTemporaryFile(delete=False, suffix=suffix) as tmp:
        content = await file.read()
        tmp.write(content)
        tmp_path = tmp.name

    try:
        doc = fitz.open(tmp_path)
        result = {"pymupdf_version": fitz.VersionBind, "pages": []}

        for page_num, page in enumerate(doc):
            pg = {"page": page_num + 1, "text_blocks": [], "images": []}

            # Text blocks
            for b in page.get_text("dict", flags=fitz.TEXTFLAGS_TEXT)["blocks"]:
                if b["type"] == 0:
                    text = ""
                    for line in b.get("lines", []):
                        for span in line.get("spans", []):
                            text += span.get("text", "")
                    text = text.strip()
                    if text:
                        pg["text_blocks"].append({
                            "y": round(b["bbox"][1], 1),
                            "text": text[:60],
                        })

            # Images
            for img in page.get_images(full=True):
                xref = img[0]
                info = {"xref": xref}
                try:
                    rects = page.get_image_rects(xref)
                    if rects:
                        r = rects[0]
                        info["y"] = round(r.y0, 1)
                        info["w"] = round(r.width, 1)
                        info["h"] = round(r.height, 1)
                    else:
                        info["y"] = "no_rects"
                except Exception as e:
                    info["y"] = f"error: {e}"

                # Also try get_image_info
                try:
                    for ii in page.get_image_info(xrefs=True):
                        if ii.get("xref") == xref:
                            info["info_bbox"] = [round(x, 1) for x in ii.get("bbox", [])]
                            break
                except Exception:
                    pass

                pg["images"].append(info)

            result["pages"].append(pg)

        doc.close()
        return result
    except Exception as e:
        return {"error": str(e), "traceback": repr(e)}
    finally:
        os.unlink(tmp_path)


def extract_pdf(tmp_path: str) -> dict:
    doc = fitz.open(tmp_path)
    page_count = len(doc)
    raw_elements = []

    for page_num, page in enumerate(doc):
        page_height = page.rect.height
        page_width = page.rect.width
        header_zone = page_height * 0.08
        footer_zone = page_height * 0.92

        # Tables
        table_rects = []
        try:
            found_tables = page.find_tables()
            for tab in found_tables:
                rect = fitz.Rect(tab.bbox)
                table_rects.append(rect)
                html = "<table>"
                for row in tab.extract():
                    html += "<tr>"
                    for cell in row:
                        html += f"<td>{cell or ''}</td>"
                    html += "</tr>"
                html += "</table>"
                raw_elements.append((page_num, rect.y0, {
                    "content": html, "type": "table", "page": page_num + 1,
                }))
        except Exception:
            pass

        # Images — try get_image_rects first, fall back to get_image_info
        seen_xrefs = set()
        for img in page.get_images(full=True):
            xref = img[0]
            if xref in seen_xrefs:
                continue
            seen_xrefs.add(xref)

            # Try to find y-position
            img_y = None
            img_w = 0
            img_h = 0

            # Method 1: get_image_rects (most reliable)
            try:
                rects = page.get_image_rects(xref)
                if rects:
                    r = rects[0]
                    img_y = r.y0
                    img_w = r.width
                    img_h = r.height
            except Exception:
                pass

            # Method 2: get_image_info (fallback)
            if img_y is None:
                try:
                    for ii in page.get_image_info(xrefs=True):
                        if ii.get("xref") == xref:
                            bbox = ii.get("bbox", (0, 0, 0, 0))
                            img_y = bbox[1]
                            img_w = bbox[2] - bbox[0]
                            img_h = bbox[3] - bbox[1]
                            break
                except Exception:
                    pass

            # Method 3: just put it after all text on this page
            if img_y is None or img_y == 0:
                img_y = page_height * 0.5  # middle of page as fallback

            # Filters
            if img_y < header_zone or (img_y + img_h) > footer_zone:
                continue
            if img_w < 50 or img_h < 50:
                continue
            if img_w > page_width * 0.8 and img_h < 30:
                continue

            try:
                base_img = doc.extract_image(xref)
                if not base_img or not base_img.get("image"):
                    continue
                w = base_img.get("width", 0)
                h = base_img.get("height", 0)
                if w < 50 or h < 50:
                    continue

                b64 = base64.b64encode(base_img["image"]).decode("utf-8")
                ext = base_img.get("ext", "png")

                raw_elements.append((page_num, img_y, {
                    "content": f"data:image/{ext};base64,{b64}",
                    "type": "image", "page": page_num + 1,
                    "width": w, "height": h,
                }))
            except Exception:
                continue

        # Text blocks
        blocks = page.get_text("dict", flags=fitz.TEXTFLAGS_TEXT)["blocks"]

        for block in blocks:
            if block["type"] != 0:
                continue
            block_rect = fitz.Rect(block["bbox"])
            if any(block_rect.intersects(tr) for tr in table_rects):
                continue

            y_pos = block_rect.y0

            lines_data = []
            for line in block.get("lines", []):
                lt = ""
                ms = 0
                bold = False
                for span in line.get("spans", []):
                    lt += span.get("text", "")
                    if span.get("size", 0) > ms:
                        ms = span["size"]
                    if "bold" in span.get("font", "").lower():
                        bold = True
                lt = lt.strip()
                if lt:
                    lines_data.append({"text": lt, "font_size": ms, "is_bold": bold})

            if not lines_data:
                continue

            paragraphs = []
            cp = []
            cit = False
            for ld in lines_data:
                it = (
                    ld["font_size"] > 13 or
                    (ld["is_bold"] and len(ld["text"]) < 120) or
                    (ld["text"].isupper() and 2 < len(ld["text"]) < 100)
                )
                if it != cit and cp:
                    pt = " ".join(cp).strip()
                    if pt:
                        paragraphs.append({"text": pt, "is_title": cit})
                    cp = []
                cit = it
                cp.append(ld["text"])
            if cp:
                pt = " ".join(cp).strip()
                if pt:
                    paragraphs.append({"text": pt, "is_title": cit})

            for para in paragraphs:
                el_type = "title" if para["is_title"] else "text"
                text = para["text"]

                if el_type == "text" and len(text) > 500:
                    sents = re.split(r'(?<=[.!?])\s+', text)
                    chunk = ""
                    offset = 0
                    for s in sents:
                        if len(chunk) + len(s) > 400 and chunk:
                            raw_elements.append((page_num, y_pos + offset * 0.01, {
                                "content": chunk.strip(),
                                "type": "text", "page": page_num + 1,
                            }))
                            offset += 1
                            chunk = s
                        else:
                            chunk += (" " if chunk else "") + s
                    if chunk.strip():
                        raw_elements.append((page_num, y_pos + offset * 0.01, {
                            "content": chunk.strip(),
                            "type": "text", "page": page_num + 1,
                        }))
                else:
                    raw_elements.append((page_num, y_pos, {
                        "content": text, "type": el_type, "page": page_num + 1,
                    }))

    doc.close()

    raw_elements.sort(key=lambda x: (x[0], x[1]))

    all_elements = []
    for idx, (pg, yp, el) in enumerate(raw_elements):
        el["index"] = idx
        all_elements.append(el)

    text_sections = [e for e in all_elements if e["type"] == "text"]
    tables = [e for e in all_elements if e["type"] == "table"]
    images = [e for e in all_elements if e["type"] == "image"]
    titles = [e for e in all_elements if e["type"] == "title"]

    return {
        "text_sections": text_sections,
        "tables": tables,
        "images": images,
        "titles": titles,
        "page_count": page_count,
        "counts": {
            "text_sections": len(text_sections),
            "tables": len(tables),
            "images": len(images),
            "titles": len(titles),
        }
    }


def extract_docx(tmp_path: str) -> dict:
    from docx import Document as DocxDocument
    doc = DocxDocument(tmp_path)
    elements = []
    idx = 0
    for para in doc.paragraphs:
        text = para.text.strip()
        if not text:
            continue
        if para.style and para.style.name and "Heading" in para.style.name:
            elements.append({"index": idx, "content": text, "type": "title"})
        else:
            elements.append({"index": idx, "content": text, "type": "text"})
        idx += 1
    for table in doc.tables:
        html = "<table>"
        for row in table.rows:
            html += "<tr>"
            for cell in row.cells:
                html += f"<td>{cell.text}</td>"
            html += "</tr>"
        html += "</table>"
        elements.append({"index": idx, "content": html, "type": "table"})
        idx += 1
    text_sections = [e for e in elements if e["type"] == "text"]
    tables_list = [e for e in elements if e["type"] == "table"]
    titles = [e for e in elements if e["type"] == "title"]
    return {
        "text_sections": text_sections, "tables": tables_list, "images": [],
        "titles": titles,
        "counts": {"text_sections": len(text_sections), "tables": len(tables_list), "images": 0, "titles": len(titles)}
    }


@app.post("/partition")
async def partition_document(file: UploadFile = File(...)):
    suffix = os.path.splitext(file.filename)[1].lower()
    with tempfile.NamedTemporaryFile(delete=False, suffix=suffix) as tmp:
        content = await file.read()
        tmp.write(content)
        tmp_path = tmp.name
    try:
        if suffix == ".pdf":
            return extract_pdf(tmp_path)
        elif suffix == ".docx":
            return extract_docx(tmp_path)
        else:
            with open(tmp_path, "r", encoding="utf-8", errors="ignore") as f:
                text = f.read()
            paragraphs = [p.strip() for p in text.split("\n\n") if p.strip()]
            return {
                "text_sections": [{"index": i, "content": p, "type": "text"} for i, p in enumerate(paragraphs)],
                "tables": [], "images": [], "titles": [],
                "counts": {"text_sections": len(paragraphs), "tables": 0, "images": 0, "titles": 0},
            }
    except Exception as e:
        return {"error": str(e), "type": type(e).__name__}
    finally:
        os.unlink(tmp_path)


@app.get("/health")
def health():
    return {"status": "ok"}