Restock & Price detection - Fix decimal-comma prices ("39,99") being read as 3999

[jaime-king]

Fixes #4493

Root cause

Restock.parse_currency() (processors/restock_diff/__init__.py:19) already normalises both separator styles: "39,99" → 39.99, "1.299,00" → 1299.0, "1,299.00" → 1299.0. But the extractors never let it see the original value. Each one strips everything except [\d.] first and calls float() directly:

• processors/restock_diff/processor.py:50 and :54, in _deduplicate_prices() (extruct path, microdata and JSON-LD)
• processors/restock_diff/pure_python_extractor.py:244, the JSON-LD string fallback in query_price_availability()
• processors/restock_diff/pure_python_extractor.py:138, the microdata text-node path

So a decimal comma is deleted before parsing: "39,99" → "3999" → 3999.0. Detection still fires because every price is off by the same factor. But the stored price, the watch overview, {{restock.price}} / {{restock.previous_price}} in notifications, and the absolute price_change_min / price_change_max limits are all 100× too high for any shop that writes prices this way. That's normal for EUR, e.g. every EUR-region Steam store page (example in #4493).

The fix

Route all four sites through Restock().parse_currency() instead of float(re.sub(...)). Values it can't parse are skipped rather than added as None.

Routing through parse_currency() would change one case that the old strip-first code handled correctly by accident: a value with only thousands commas, e.g. "1,299". parse_currency() reads a lone comma as a decimal separator, which would give 1.299. So parse_currency() now treats commas followed by exact three-digit groups (\d{1,3}(,\d{3})+) as thousands separators, and "1,299" stays 1299.0.

Source change: +14 / −8 across the three files; the rest is the test.

How I verified

Run inside the published dgtlmoon/changedetection.io:0.60.7 image, with this branch (based on d789fe3) copied in:

• New tests/unit/test_restock_logic.py::TestDiffBuilder::test_price_decimal_comma covers 39,99, 12.56, 1.299,00, 1,299.00, 1,299, € 39,99, $159 and 159. Each goes through parse_currency(), extruct microdata, extruct JSON-LD, pure-Python JSON-LD, and pure-Python microdata text.
• Differential: with the source files reverted to master, the new test fails (AssertionError: microdata '39,99'). With the fix it passes.
• Regression: tests/test_restock_itemprop.py tests/test_automatic_follow_ldjson_price.py tests/test_notification_restock_token.py tests/unit/test_restock_logic.py tests/unit/test_restock_deal_filter.py: 22 passed, no failures.
• End to end, before the fix: a live EUR Steam watch records 3999.0 for a content="39,99" price. A bol.com watch (JSON-LD "price":"12.56") records 12.56 and is unaffected.

What I did not run: the full run_basic_tests.sh / browser-fetcher suites. The change is confined to price parsing in restock_diff, and the CI run on this PR covers the rest.

Known limitations (unchanged by this PR)

These are genuinely ambiguous without knowing the page's locale; both behave exactly as on master today:

• A dot as the only thousands separator with no decimals ("1.299") is read as 1.299.
• A comma followed by exactly three digits ("0,299", "12,500") is read as thousands (299, 12500).

---

Disclosure: the change, its test and this description were prepared with an AI agent (Claude Code).

🤖 Generated with Claude Code