ZZZ Digital Lab

Back to PDF tables to sheets

tests/run.py

"""Offline regression checks against PDFs generated in a project-local temp folder."""
import importlib.util,json,sys,tempfile,unittest,subprocess,csv
from decimal import Decimal
from pathlib import Path
from openpyxl import load_workbook
from pypdf import PdfReader,PdfWriter
from reportlab.pdfgen import canvas
ROOT=Path(__file__).resolve().parents[1]; sys.path.insert(0,str(ROOT))
from generate_sources import generate
from extract import extract_document,run,decimal_cell,safe_text,LayoutError
spec=importlib.util.spec_from_file_location('bad_extract',ROOT/'before/extract.py'); bad=importlib.util.module_from_spec(spec); spec.loader.exec_module(bad)

class ExtractionTests(unittest.TestCase):
 @classmethod
 def setUpClass(cls):
  (ROOT/'tmp').mkdir(exist_ok=True); cls.temp=tempfile.TemporaryDirectory(dir=ROOT/'tmp'); cls.base=Path(cls.temp.name)
  cls.addClassCleanup(cls.temp.cleanup)
  cls.source=generate(cls.base/'sources'); cls.out=cls.base/'out'; cls.qa=run(cls.source,cls.out)
  cls.invoice,cls.invoice_qa=extract_document(cls.source/'invoice.pdf'); cls.statement,cls.statement_qa=extract_document(cls.source/'statement.pdf')
 def test_01_before_silently_omits_second_page(self):
  rows=bad.extract_bad(self.source/'invoice.pdf',self.base/'bad.csv'); self.assertEqual(len(rows),3); self.assertNotEqual(len(rows),6)
 def test_02_before_splits_description_into_wrong_columns(self):
  rows=bad.extract_bad(self.source/'invoice.pdf',self.base/'bad.csv'); self.assertGreater(len(rows[0]),5)
 def test_03_before_guesses_ambiguous_amount(self):
  rows=bad.extract_bad(self.source/'statement.pdf',self.base/'bad-st.csv'); self.assertIn('05.00',rows[3])
 def test_04_all_pages_sequence_and_declared_row_counts(self):
  self.assertEqual(len(self.invoice),6); self.assertEqual(len(self.statement),8)
  for qa in (self.invoice_qa,self.statement_qa): self.assertEqual(qa['pages'],2); self.assertEqual(qa['row_count']['status'],'PASS'); self.assertEqual(qa['page_sequence'],'PASS')
  reader=PdfReader(self.source/'invoice.pdf')
  for name,indices in [('missing',[0]),('duplicate',[0,0]),('reordered',[1,0])]:
   writer=PdfWriter()
   for index in indices: writer.add_page(reader.pages[index])
   target=self.base/(name+'.pdf'); writer.write(target)
   with self.assertRaisesRegex(LayoutError,'PAGE_SEQUENCE'): extract_document(target)
 def test_05_exact_multiword_text_and_amounts(self):
  self.assertEqual(self.invoice[2]['Description'],'Synthetic fixture pack'); self.assertEqual(self.invoice[2]['Amount'],'45.50')
  self.assertEqual(self.invoice[5]['Description'],'Handoff walkthrough')
 def test_06_invoice_totals_reconcile_without_binary_floats(self):
  self.assertEqual(self.invoice_qa['totals']['Amount'],{'status':'PASS','actual':'550.50','expected':'550.50','difference':'0.00'})
  self.assertEqual(self.invoice_qa['totals']['Qty']['actual'],'13'); self.assertEqual(self.invoice_qa['quality'],'PASS')
 def test_07_all_invoice_line_products_checked(self):
  self.assertEqual(len(self.invoice_qa['row_checks']),6); self.assertTrue(all(r['status']=='PASS' for r in self.invoice_qa['row_checks']))
 def test_08_low_confidence_value_is_blank_not_guessed(self):
  row=self.statement[3]; self.assertEqual(row['Debit'],''); self.assertEqual(row['review_status'],'REVIEW')
  flag=self.statement_qa['low_confidence_cells'][0]; self.assertEqual(flag['raw_value'],'O5.00'); self.assertEqual(flag['column'],'Debit'); self.assertEqual(flag['confidence'],'LOW')
 def test_09_missing_amount_prevents_false_total_pass(self):
  total=self.statement_qa['totals']['Debit']; self.assertEqual(total['actual'],'274.75'); self.assertEqual(total['expected'],'279.75'); self.assertEqual(total['status'],'INCOMPLETE'); self.assertIsNone(total['difference'])
  self.assertEqual(self.statement_qa['totals']['Net movement to closing']['status'],'INCOMPLETE')
 def test_10_other_statement_controls_are_checked(self):
  self.assertEqual(self.statement_qa['totals']['Credit']['status'],'PASS'); self.assertEqual(self.statement_qa['totals']['Closing balance']['status'],'PASS')
  statuses=[r['status'] for r in self.statement_qa['row_checks']]; self.assertEqual(statuses.count('PASS'),7); self.assertEqual(statuses.count('INCOMPLETE'),1)
 def test_11_each_row_has_page_and_table_row_provenance(self):
  self.assertEqual(self.invoice[3]['provenance'],'invoice.pdf:p2:r1'); self.assertEqual(self.statement[-1]['provenance'],'statement.pdf:p2:r4')
  self.assertEqual(len({r['provenance'] for r in self.invoice+self.statement}),14)
 def test_12_formula_like_text_stays_text_in_csv_and_xlsx(self):
  self.assertTrue(self.invoice[4]['Description'].startswith("'="))
  wb=load_workbook(self.out/'extracted.xlsx'); self.assertEqual(wb['invoice']['B6'].data_type,'s'); self.assertTrue(wb['invoice']['B6'].value.startswith("'=")); wb.close()
 def test_13_xlsx_matches_csv_and_contains_review_queue(self):
  wb=load_workbook(self.out/'extracted.xlsx'); self.assertEqual(wb.sheetnames,['invoice','statement','review_queue']); self.assertEqual(wb['invoice'].max_row,7); self.assertEqual(wb['statement'].max_row,9); self.assertEqual(wb['review_queue'].max_row,2)
  with (self.out/'invoice.csv').open(newline='',encoding='utf-8') as f: rows=list(csv.DictReader(f))
  self.assertEqual(Decimal(rows[0]['Amount']),Decimal(str(wb['invoice']['E2'].value))); self.assertEqual(wb['invoice']['E2'].data_type,'n'); self.assertEqual(wb['statement']['A2'].number_format,'yyyy-mm-dd'); wb.close()
 def test_14_csv_preserves_empty_debit_and_report_is_explicit(self):
  with (self.out/'statement.csv').open(newline='',encoding='utf-8') as f: rows=list(csv.DictReader(f))
  self.assertEqual(rows[3]['Debit'],''); text=(self.out/'QA_REPORT.md').read_text(encoding='utf-8'); self.assertIn('REVIEW_REQUIRED',text); self.assertIn('statement.pdf:p1:r4',text); self.assertIn('INCOMPLETE',text)
 def test_15_numeric_parser_rejects_ambiguous_and_malformed_values(self):
  for value in ['O5.00','1,2.00','NaN','12 USD','1e3','-5','1.001','=5']:
   self.assertIsNone(decimal_cell(value),value)
  self.assertEqual(decimal_cell('1,200.00'),Decimal('1200.00')); self.assertEqual(decimal_cell('0.00'),Decimal('0.00'))
 def test_16_wrong_source_total_produces_failed_qa(self):
  source=generate(self.base/'wrong-total',invoice_total='999.00'); rows,qa=extract_document(source/'invoice.pdf')
  self.assertEqual(qa['quality'],'FAIL'); self.assertEqual(qa['totals']['Amount']['status'],'MISMATCH'); self.assertEqual(qa['totals']['Amount']['difference'],'-448.50')
 def test_17_cli_returns_nonzero_for_failed_control(self):
  source=generate(self.base/'bad-cli',invoice_total='999.00')
  result=subprocess.run([sys.executable,'-B',str(ROOT/'extract.py'),'--source',str(source),'--output',str(self.base/'bad-cli-output')],capture_output=True,text=True)
  self.assertEqual(result.returncode,2); self.assertIn('FAIL',result.stdout)
 def test_18_unknown_layout_fails_closed(self):
  pdf=self.base/'unknown.pdf'; c=canvas.Canvas(str(pdf)); c.drawString(40,700,'SYNTHETIC unsupported layout'); c.save()
  with self.assertRaisesRegex(LayoutError,'UNSUPPORTED'): extract_document(pdf)
 def test_19_known_title_without_expected_table_fails_closed(self):
  pdf=self.base/'missing-table.pdf'; c=canvas.Canvas(str(pdf)); c.drawString(40,700,'Invoice No: SYN-OTHER'); c.save()
  with self.assertRaisesRegex(LayoutError,'TABLE_SCHEMA'): extract_document(pdf)
 def test_20_repeated_runs_have_identical_structured_qa(self):
  second=run(self.source,self.base/'rerun'); self.assertEqual(second,self.qa); self.assertEqual(self.qa['total_rows'],14); self.assertEqual(self.qa['low_confidence_cell_count'],1)

class RecordedResult(unittest.TextTestResult):
 def __init__(self,*args,**kwargs): super().__init__(*args,**kwargs); self.records=[]
 def addSuccess(self,test): super().addSuccess(test); self.records.append({'name':test._testMethodName,'status':'PASS'})
 def addFailure(self,test,err): super().addFailure(test,err); self.records.append({'name':test._testMethodName,'status':'FAIL'})
 def addError(self,test,err): super().addError(test,err); self.records.append({'name':str(test),'status':'ERROR'})

def summarize_result(result):
 # Class setup/cleanup errors do not count as executed test methods.
 # Count completed successes directly instead of subtracting fixture errors.
 return {'synthetic':True,'scope':'Generated text PDFs, real local pdfplumber extraction and openpyxl workbook round-trip',
         'tests_run':result.testsRun,'passed':sum(r['status']=='PASS' for r in result.records),
         'failed':len(result.failures),'errors':len(result.errors),'skipped':len(result.skipped),
         'successful':result.wasSuccessful(),'results':result.records}

if __name__=='__main__':
 result=unittest.TextTestRunner(verbosity=2,resultclass=RecordedResult).run(unittest.defaultTestLoader.loadTestsFromTestCase(ExtractionTests))
 report=summarize_result(result)
 (ROOT/'evidence').mkdir(exist_ok=True); (ROOT/'evidence/tests.json').write_text(json.dumps(report,indent=2)+'\n',encoding='utf-8'); sys.exit(0 if result.wasSuccessful() else 1)