fix bug by truncating corrupted jsonl lines.

test/test_resume.py

@@ -458,7 +458,10 @@ def test_resume_corrupted_jsonl_last_line():
     """Test that JSONL resume correctly handles corrupted/incomplete last line.
 
     When the previous run was interrupted mid-write leaving an incomplete JSON
-    line, the resume should detect and remove the corrupted line before appending.
+    line, the resume should:
+    1. Find the resume point from the last valid line (no checkpoint file needed)
+    2. Truncate the corrupted trailing data
+    3. Append new data, resulting in valid JSONL
     """
     tester_full = WikiqTester(SAILORMOON, "resume_corrupt_full", in_compression="7z", out_format="jsonl")
 
@@ -481,19 +484,20 @@ def test_resume_corrupted_jsonl_last_line():
         # Write incomplete JSON (simulates crash mid-write)
         f.write('{"revid": 999, "articleid": 123, "incomplet')
 
-    # Write checkpoint pointing to a valid revision (last complete row)
-    checkpoint_path = get_checkpoint_path(corrupt_output_path)
-    with open(checkpoint_path, 'w') as f:
-        json.dump({"pageid": full_rows[resume_idx - 1]["articleid"],
-                   "revid": full_rows[resume_idx - 1]["revid"]}, f)
+    # Record file size before resume
+    size_before = os.path.getsize(corrupt_output_path)
 
-    # Resume should detect and remove the corrupted line, then append new data
+    # NO checkpoint file - JSONL resume works from last valid line in the file
+    checkpoint_path = get_checkpoint_path(corrupt_output_path)
+    assert not os.path.exists(checkpoint_path), "Test setup error: checkpoint should not exist"
+
+    # Resume should detect corrupted line, truncate it, then append new data
     try:
         tester_corrupt.call_wikiq("--fandom-2020", "--resume")
     except subprocess.CalledProcessError as exc:
         pytest.fail(f"Resume failed unexpectedly: {exc.stderr.decode('utf8')}")
 
-    # Verify the file is valid JSONL and readable
+    # Verify the file is valid JSONL and readable (no corrupted lines)
     resumed_rows = read_jsonl(corrupt_output_path)
 
     # Full data equivalence check