RSS - RSS pubDate is wrong on any non-UTC container: local time is relabelled as UTC - #4309 (#4314)

changedetectionio/blueprint/rss/_util.py

@@ -150,7 +150,8 @@ def populate_feed_entry(fe, watch, content, guid, timestamp, link=None, title_su
     fe.guid(guid, permalink=False)
 
     # Set pubDate using the timestamp of this specific change
-    dt = datetime.datetime.fromtimestamp(int(timestamp))
-    dt = dt.replace(tzinfo=pytz.UTC)
+    # Note: tz= must be passed to fromtimestamp(), otherwise we get a naive datetime in the
+    # container's local timezone and relabelling it as UTC shifts every pubDate by the local offset
+    dt = datetime.datetime.fromtimestamp(int(timestamp), tz=pytz.UTC)
     fe.pubDate(dt)
 

changedetectionio/run_basic_tests.sh

@@ -91,6 +91,10 @@ export HIDE_REFERER=True
 REMOVE_REQUESTS_OLD_SCREENSHOTS=false pytest -vv -s --maxfail=1 tests/test_notification.py tests/test_access_control.py
 
 
+# Re #4309 - RSS pubDate/timestamps must be correct on containers that don't run UTC
+# (Europe/Athens is UTC+2/+3, so any naive local->UTC relabelling shows up as a shifted date)
+TZ=Europe/Athens pytest -vv -s --maxfail=1 tests/test_rss.py
+
 # Re-run a few tests that will trigger brotli based storage
 # And again with brotli+screenshot attachment
 SNAPSHOT_BROTLI_COMPRESSION_THRESHOLD=5 REMOVE_REQUESTS_OLD_SCREENSHOTS=false pytest -vv -s --maxfail=1 --dist=load tests/test_backend.py tests/test_rss.py tests/test_unique_lines.py tests/test_notification.py  tests/test_access_control.py

changedetectionio/tests/test_rss.py

@@ -1,6 +1,8 @@
 #!/usr/bin/env python3
+import email.utils
 import os
 import time
+import pytest
 from flask import url_for
 from .util import set_original_response, set_modified_response, live_server_setup, wait_for_all_checks, extract_rss_token_from_UI, \
     extract_UUID_from_client, delete_all_watches
@@ -354,3 +356,54 @@ def test_rss_single_watch_feed(client, live_server, measure_memory_usage, datast
     assert (">3<" in descriptions[2] or "Version 3" in descriptions[2]) and "content" in descriptions[2], \
         f"Third item should show Version 3, but got: {descriptions[2][:500]}"
 
+    # #4309 - pubDate must be the real UTC time of the change, whatever the server's local timezone is.
+    # The GUID carries the snapshot timestamp, so the two must agree to the second.
+    for item in items:
+        guid_timestamp = int(item.findtext('guid').rsplit('/', 1)[1])
+        pub_date = email.utils.parsedate_to_datetime(item.findtext('pubDate'))
+        assert pub_date.timestamp() == guid_timestamp, \
+            f"pubDate {item.findtext('pubDate')} does not match snapshot timestamp {guid_timestamp}"
+
+
+@pytest.mark.skipif(not hasattr(time, 'tzset'), reason="Changing TZ at runtime needs a POSIX platform")
+@pytest.mark.parametrize("tz_name", ['UTC', 'Europe/Athens', 'America/New_York', 'Australia/Sydney'])
+def test_rss_pubdate_is_utc_regardless_of_local_timezone(tz_name):
+    """#4309 - datetime.fromtimestamp() without tz= returns local wall-clock time, and relabelling
+    that as UTC shifts every pubDate by the local offset (items appear in the future on TZ=Europe/Athens).
+
+    No live server needed - this drives populate_feed_entry() directly under several timezones."""
+    import xml.etree.ElementTree as ET
+    from feedgen.feed import FeedGenerator
+    from ..blueprint.rss._util import populate_feed_entry
+
+    timestamp = 1700000000  # Tue, 14 Nov 2023 22:13:20 UTC
+
+    original_tz = os.environ.get('TZ')
+    try:
+        os.environ['TZ'] = tz_name
+        time.tzset()
+
+        fg = FeedGenerator()
+        fg.title('test')
+        fg.link(href='https://example.com', rel='self')
+        fg.description('test')
+        fe = fg.add_entry()
+
+        populate_feed_entry(fe=fe,
+                            watch={'uuid': 'fake-uuid', 'url': 'https://example.com'},
+                            content='some content',
+                            guid=f'fake-uuid/{timestamp}',
+                            timestamp=timestamp)
+
+        pub_date = ET.fromstring(fg.rss_str()).findtext('.//item/pubDate')
+    finally:
+        if original_tz is None:
+            os.environ.pop('TZ', None)
+        else:
+            os.environ['TZ'] = original_tz
+        time.tzset()
+
+    # parsedate_to_datetime() honours the offset in the header, so this compares real instants
+    assert email.utils.parsedate_to_datetime(pub_date).timestamp() == timestamp, \
+        f"With TZ={tz_name} the feed said {pub_date}, expected the instant {timestamp}"
+