Isolate File Behavior

Selection tests have no lasting effects. A JSON writer does, so a useful test must choose where those bytes may exist. A fixed selected.json in the project would make tests depend on old state and leave artifacts behind.

Pytest supplies a fresh Path named tmp_path to each test that requests it. Add this file-behavior test:

import json

from catalog import write_catalog_json


def test_json_file_keeps_visible_unicode(tmp_path):
    # Arrange
    output_path = tmp_path / "catalog-output.json"
    item = {
        "sku": "CAFE-7",
        "name": "Café Cards",
        "quantity": 7,
        "unit_price": 2.5,
        "in_stock": True,
    }

    # Act
    write_catalog_json([item], output_path)
    raw = output_path.read_text(encoding="utf-8")

    # Assert with an independent JSON oracle and serialized text contract
    parsed = json.loads(raw)
    assert parsed == {"schema_version": 1, "items": [item]}
    assert type(parsed["schema_version"]) is int
    saved = parsed["items"][0]
    assert type(saved["quantity"]) is int
    assert type(saved["unit_price"]) is float
    assert type(saved["in_stock"]) is bool
    assert "Café Cards" in raw
    assert "\\u00e9" not in raw
    assert raw.endswith("\n")
    assert not raw.endswith("\n\n")

The expected document belongs to the test. It does not come from the writer’s return value or the learner’s JSON reader. The explicit type checks matter because Python otherwise considers values such as True and 1 equal. If two production helpers shared the same bug, asking one to certify the other would be false confidence.

Run the complete suite:

python -m pytest -q

This lesson must end red. Collection remains healthy and the selection tests pass, but the new test fails because the seeded writer escapes é as \\u00e9. Read the failing node, the assertion, and pytest’s actual value. Do not edit catalog.py yet; Lesson 5 owns that repair.

Check that you are red for the right reason. You want status 1, with the selection tests still passing and your new test failing on its Unicode assertion. That is a healthy red: the suite ran, found what it was looking for, and disagreed with production.

Status 2 or 5 would be a different story. Those mean pytest could not collect or import your tests, so nothing was learned about the Unicode contract at all. A red suite is only evidence when the failure is the one you were asking about.

One rule while you are here: no test may write into the project folder, even if it tidies up afterwards. Cleanup does not run when a test fails partway, which is exactly when you least want yesterday’s file confusing tomorrow’s run. That is what tmp_path is for.