test_image_pdf.py

"""
Image-only / scanned-PDF path: detect no-text PDFs, render pages to images, and
build a deck from them. The render/detect plumbing is testable without a GPU;
the model call is exercised only in stub mode here (real vision inference is
covered by the deploy, not locally). Skips when the real-model image deps
(PyMuPDF + Pillow) aren't installed — they live in requirements-model.txt.

    RECALL_STUB=1 python3 -m pytest test_image_pdf.py
"""
import io
import os
import tempfile

import pytest

os.environ.setdefault("RECALL_STUB", "1")

fitz = pytest.importorskip("fitz", reason="needs PyMuPDF (real-model dep)")
PIL_Image = pytest.importorskip("PIL.Image", reason="needs Pillow (real-model dep)")
from PIL import Image, ImageDraw  # noqa: E402

import content_pipeline as cp  # noqa: E402


def _make_scanned_pdf(pages: int = 2) -> str:
    """A PDF with no text layer — each page is a rasterized image."""
    fd, path = tempfile.mkstemp(suffix=".pdf")
    os.close(fd)
    doc = fitz.open()
    for i in range(pages):
        img = Image.new("RGB", (600, 800), "white")
        ImageDraw.Draw(img).text((40, 40), f"Scanned page {i + 1}", fill="black")
        buf = io.BytesIO()
        img.save(buf, format="PNG")
        page = doc.new_page(width=600, height=800)
        page.insert_image(fitz.Rect(0, 0, 600, 800), stream=buf.getvalue())
    doc.save(path)
    doc.close()
    return path


def _make_text_pdf() -> str:
    fd, path = tempfile.mkstemp(suffix=".pdf")
    os.close(fd)
    doc = fitz.open()
    doc.new_page().insert_text((72, 72), "Selectable text about photosynthesis.")
    doc.save(path)
    doc.close()
    return path


def test_is_image_only_pdf_detects_scanned():
    path = _make_scanned_pdf()
    try:
        assert cp.is_image_only_pdf(path) is True
    finally:
        os.unlink(path)


def test_text_pdf_is_not_image_only():
    path = _make_text_pdf()
    try:
        assert cp.is_image_only_pdf(path) is False
    finally:
        os.unlink(path)


def test_non_pdf_is_not_image_only_and_renders_nothing():
    assert cp.is_image_only_pdf("notes.txt") is False
    assert cp.render_pdf_images("notes.txt") == []


def test_render_pdf_images_returns_pages():
    path = _make_scanned_pdf(pages=3)
    try:
        images = cp.render_pdf_images(path)
        assert len(images) == 3
        assert all(isinstance(im, Image.Image) for im in images)
    finally:
        os.unlink(path)


def test_render_pdf_images_respects_max_pages():
    path = _make_scanned_pdf(pages=5)
    try:
        assert len(cp.render_pdf_images(path, max_pages=2)) == 2
    finally:
        os.unlink(path)


def test_generate_deck_from_images_stub():
    # STUB returns the canned demo deck regardless of the images passed. Pin
    # llm.STUB explicitly (and restore) so this is robust to other test files'
    # import-order / STUB mutations rather than relying on ambient state.
    import llm
    prev = llm.STUB
    llm.STUB = True
    try:
        deck = cp.generate_deck_from_images([object(), object()])
    finally:
        llm.STUB = prev
    assert isinstance(deck, list) and len(deck) >= 1
    assert all(c.get("question") and c.get("answer") for c in deck)


if __name__ == "__main__":
    test_is_image_only_pdf_detects_scanned()
    test_text_pdf_is_not_image_only()
    test_non_pdf_is_not_image_only_and_renders_nothing()
    test_render_pdf_images_returns_pages()
    test_render_pdf_images_respects_max_pages()
    test_generate_deck_from_images_stub()
    print("All image-PDF tests passed.")