https://arxiv.org/abs/2606.28344. The end of web parsing. The beginning of scalable pixel-native search. link: https://pixelrag.ai/
"""Selected PDF pages retain the same zero-based IDs as a full render."""
import json
import sys
from types import ModuleType
import pytest
from PIL import Image
from pixelrag_render import render_pdf
@pytest.mark.parametrize("pages", [None, [1], [3], [4], [2, 3], [1, 3], [2, 4], [4, 2]])
def test_pdf_page_indices_match_full_render(tmp_path, monkeypatch, pages):
source = tmp_path / "document.pdf"
source.touch()
colors = ["red", "green", "blue", "yellow"]
converter = ModuleType("pdf2image")
def convert_from_path(**kwargs):
first = kwargs.get("first_page", 1)
last = kwargs.get("last_page", 4)
return [
Image.new("RGB", (16, 16), colors[i - 1]) for i in range(first, last + 1)
]
converter.convert_from_path = convert_from_path
monkeypatch.setitem(sys.modules, "pdf2image", converter)
full_dir = render_pdf(source, tmp_path / "full")[0]
selected_dir = render_pdf(source, tmp_path / "selected", pages=pages)[0]
expected_ids = [0, 1, 2, 3] if pages is None else sorted(p - 1 for p in pages)
expected_files = [f"tile_{i:04d}.jpg" for i in expected_ids]
manifest = json.loads((selected_dir / "tiles.json").read_text())
chunks = json.loads((selected_dir / "chunks.json").read_text())["chunks"]
assert manifest["tiles"] == expected_files
assert [c["tile_index"] for c in chunks] == expected_ids
assert [c["file"] for c in chunks] == expected_files
assert [c["tile"] for c in chunks] == expected_files
assert sorted(p.name for p in selected_dir.glob("tile_*.jpg")) == expected_files
for filename in expected_files:
assert (selected_dir / filename).read_bytes() == (
full_dir / filename
).read_bytes()