250 lines
7.2 KiB
Python
250 lines
7.2 KiB
Python
"""Testes do pipeline de extração de texto de PDF.
|
|
|
|
Cobre a detecção de camada textual corrompida (o PDF real do projeto), o fallback
|
|
para OCR, o cache por página e o isolamento de falhas de OCR.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import sys
|
|
from pathlib import Path
|
|
|
|
import pytest
|
|
|
|
sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
|
|
|
|
from app.extraction import OcrError, PdfExtractor, legible_ratio # noqa: E402
|
|
|
|
|
|
class FakePage:
|
|
"""Página de PDF simulada."""
|
|
|
|
def __init__(self, text: str = "", pixmap=None):
|
|
self._text = text
|
|
self._pixmap = pixmap or FakePixmap()
|
|
|
|
def get_text(self, kind: str) -> str:
|
|
assert kind == "text"
|
|
return self._text
|
|
|
|
def get_pixmap(self, dpi: int = 200):
|
|
return self._pixmap
|
|
|
|
|
|
class FakeDocument:
|
|
def __init__(self, pages):
|
|
self._pages = pages
|
|
self.closed = False
|
|
|
|
def __len__(self):
|
|
return len(self._pages)
|
|
|
|
def __getitem__(self, index):
|
|
return self._pages[index]
|
|
|
|
def __iter__(self):
|
|
return iter(self._pages)
|
|
|
|
def close(self):
|
|
self.closed = True
|
|
|
|
|
|
class FakePixmap:
|
|
def __init__(self, width=10, height=10, blank=False):
|
|
self.width = width
|
|
self.height = height
|
|
self.n = 3
|
|
self.blank = blank
|
|
self.saved_to: list[Path] = []
|
|
|
|
@property
|
|
def samples(self) -> bytes:
|
|
if self.blank:
|
|
return bytes([255]) * (self.width * self.height * self.n)
|
|
return bytes(range(self.width * self.height * self.n))[: self.width * self.height * self.n]
|
|
|
|
def save(self, path):
|
|
self.saved_to.append(Path(path))
|
|
Path(path).write_bytes(b"PNG")
|
|
|
|
|
|
class FakePyMuPDF:
|
|
"""Dublê do módulo pymupdf para testes."""
|
|
|
|
def __init__(self, document):
|
|
self._document = document
|
|
self.opened: list[Path] = []
|
|
|
|
def open(self, path):
|
|
self.opened.append(Path(path))
|
|
return self._document
|
|
|
|
|
|
@pytest.fixture
|
|
def tmp_cache(tmp_path):
|
|
return tmp_path / "ocr_cache"
|
|
|
|
|
|
def test_legible_ratio_is_high_for_normal_text():
|
|
text = "No princípio criou Deus os céus e a terra. E a terra era sem forma e vazia."
|
|
assert legible_ratio(text) > 0.9
|
|
|
|
|
|
def test_legible_ratio_is_low_for_control_characters():
|
|
text = "".join(chr(i) for i in range(1, 32))
|
|
assert legible_ratio(text) < 0.35
|
|
|
|
|
|
def test_legible_ratio_of_empty_text_is_zero():
|
|
assert legible_ratio("") == 0.0
|
|
assert legible_ratio(" \n\t ") == 0.0
|
|
|
|
|
|
def test_legible_text_does_not_trigger_ocr(tmp_cache, tmp_path, monkeypatch):
|
|
pdf = tmp_path / "biblia.pdf"
|
|
pdf.write_bytes(b"%PDF-1.3")
|
|
doc = FakeDocument([FakePage(text="1 No princípio criou Deus os céus e a terra.")])
|
|
extractor = PdfExtractor(cache_dir=tmp_cache, pymupdf_module=FakePyMuPDF(doc))
|
|
|
|
def fail_ocr(*args, **kwargs):
|
|
raise AssertionError("OCR não deveria ser chamado para texto legível")
|
|
|
|
monkeypatch.setattr(extractor, "_run_tesseract", fail_ocr)
|
|
|
|
pages = list(extractor.extract_document(pdf))
|
|
|
|
assert len(pages) == 1
|
|
assert "No princípio" in pages[0].text
|
|
assert pages[0].page == 1
|
|
|
|
|
|
def test_corrupted_text_layer_triggers_ocr(tmp_cache, tmp_path, monkeypatch):
|
|
corrupted = "".join(chr(i) for i in range(1, 32))
|
|
page = FakePage(text=corrupted, pixmap=FakePixmap())
|
|
doc = FakeDocument([page])
|
|
extractor = PdfExtractor(cache_dir=tmp_cache, pymupdf_module=FakePyMuPDF(doc))
|
|
|
|
calls: list[int] = []
|
|
|
|
def fake_tesseract(pixmap, language, timeout):
|
|
calls.append(pixmap.width)
|
|
return "1 No princípio criou Deus os céus e a terra."
|
|
|
|
monkeypatch.setattr(extractor, "_run_tesseract", fake_tesseract)
|
|
|
|
pdf = tmp_path / "biblia.pdf"
|
|
pdf.write_bytes(b"%PDF-1.3")
|
|
pages = list(extractor.extract_document(pdf))
|
|
|
|
assert calls == [10], "OCR deveria ter sido chamado uma vez"
|
|
assert pages[0].text.startswith("1 No princípio")
|
|
|
|
|
|
def test_ocr_result_is_cached_and_reused(tmp_cache, tmp_path, monkeypatch):
|
|
corrupted = "".join(chr(i) for i in range(1, 32))
|
|
page = FakePage(text=corrupted, pixmap=FakePixmap())
|
|
extractor = PdfExtractor(
|
|
cache_dir=tmp_cache,
|
|
pymupdf_module=FakePyMuPDF(FakeDocument([page])),
|
|
)
|
|
|
|
calls = 0
|
|
|
|
def fake_tesseract(pixmap, language, timeout):
|
|
nonlocal calls
|
|
calls += 1
|
|
return "texto ocr"
|
|
|
|
monkeypatch.setattr(extractor, "_run_tesseract", fake_tesseract)
|
|
|
|
pdf = tmp_path / "biblia.pdf"
|
|
pdf.write_bytes(b"%PDF-1.3")
|
|
first = list(extractor.extract_document(pdf, file_hash="abc123"))
|
|
second = list(extractor.extract_document(pdf, file_hash="abc123"))
|
|
|
|
assert calls == 1, "segunda extração deveria usar o cache"
|
|
assert first[0].text == second[0].text == "texto ocr"
|
|
|
|
|
|
def test_cache_is_invalidated_by_different_file_hash(tmp_cache, tmp_path, monkeypatch):
|
|
corrupted = "".join(chr(i) for i in range(1, 32))
|
|
page = FakePage(text=corrupted, pixmap=FakePixmap())
|
|
extractor = PdfExtractor(
|
|
cache_dir=tmp_cache,
|
|
pymupdf_module=FakePyMuPDF(FakeDocument([page])),
|
|
)
|
|
|
|
calls = 0
|
|
|
|
def fake_tesseract(pixmap, language, timeout):
|
|
nonlocal calls
|
|
calls += 1
|
|
return "texto ocr"
|
|
|
|
monkeypatch.setattr(extractor, "_run_tesseract", fake_tesseract)
|
|
|
|
pdf = tmp_path / "biblia.pdf"
|
|
pdf.write_bytes(b"%PDF-1.3")
|
|
list(extractor.extract_document(pdf, file_hash="hash1"))
|
|
list(extractor.extract_document(pdf, file_hash="hash2"))
|
|
|
|
assert calls == 2, "hash diferente deve invalidar o cache"
|
|
|
|
|
|
def test_ocr_failure_is_isolated(tmp_cache, tmp_path, monkeypatch):
|
|
corrupted = "".join(chr(i) for i in range(1, 32))
|
|
page = FakePage(text=corrupted, pixmap=FakePixmap())
|
|
extractor = PdfExtractor(
|
|
cache_dir=tmp_cache,
|
|
pymupdf_module=FakePyMuPDF(FakeDocument([page])),
|
|
)
|
|
|
|
def failing_tesseract(pixmap, language, timeout):
|
|
raise OcrError("tesseract falhou")
|
|
|
|
monkeypatch.setattr(extractor, "_run_tesseract", failing_tesseract)
|
|
|
|
pdf = tmp_path / "biblia.pdf"
|
|
pdf.write_bytes(b"%PDF-1.3")
|
|
pages = list(extractor.extract_document(pdf))
|
|
|
|
assert len(pages) == 1
|
|
assert pages[0].text == ""
|
|
assert pages[0].failed is True
|
|
|
|
|
|
def test_extract_page_returns_empty_text_for_blank_page(tmp_cache, monkeypatch):
|
|
page = FakePage(text="", pixmap=FakePixmap(blank=True))
|
|
extractor = PdfExtractor(
|
|
cache_dir=tmp_cache,
|
|
pymupdf_module=FakePyMuPDF(FakeDocument([page])),
|
|
)
|
|
|
|
def fail_ocr(*args, **kwargs):
|
|
raise AssertionError("OCR não deveria ser chamado em página em branco")
|
|
|
|
monkeypatch.setattr(extractor, "_run_tesseract", fail_ocr)
|
|
|
|
result = extractor.extract_page(page, "abc", 1)
|
|
|
|
assert result.text == ""
|
|
assert result.failed is False
|
|
|
|
|
|
def test_real_pdf_page_is_detected_as_corrupted(tmp_path):
|
|
"""O PDF do projeto tem camada textual corrompida: precisa cair no OCR."""
|
|
import pymupdf
|
|
|
|
real_pdf = Path(__file__).resolve().parents[2] / "PDF" / "biblia.pdf"
|
|
if not real_pdf.exists():
|
|
pytest.skip("PDF real não disponível")
|
|
|
|
doc = pymupdf.open(real_pdf)
|
|
try:
|
|
page = doc[5]
|
|
text = page.get_text("text")
|
|
finally:
|
|
doc.close()
|
|
|
|
assert legible_ratio(text) < 0.35, "camada textual do PDF real deveria ser ilegível"
|