Sent-entry deduplication was hardcoded to a 24h TTL (SEEN_TTL_SECS), so an entry still present in a feed after a day would be re-posted to the MUC. Add an optional [xmpp] cache_ttl_secs key that controls how long keys are kept in the SQLite dedup cache, defaulting to 86400 so existing deployments keep the current behavior. purge_old() now takes the TTL from the config, and --verbose diagnostics render it in readable units through a new format_ttl() helper. Tests cover config parsing with and without the key, format_ttl cases, and test_purge_old now exercises purge_old() directly instead of duplicating its SQL. Document the option in README.md, config.example.toml, and AGENTS.md; AGENTS.md also gains a Repository and Tickets section describing the fossil checkout and its ticket workflow, which was previously undocumented. Version bumped 0.7.0 to 0.8.0 (new config option).

Ticket da9cde75f6e23dbf95451bc45028243c11e438b2. leaf check-in: da2a3798e0 user: andrew tags: master, trunk

Feed URLs that return HTML pages instead of RSS/Atom XML now produce a clear error message instead of cryptic XML parser errors. A fallback entity repair step also fixes bare ampersands and invalid entity references in feeds that have minor XML well-formedness issues. check-in: 427a4acede user: sourcehut@lewman.us tags: master, trunk

roxmltree 0.21 rejects documents with <!DOCTYPE> by default, which caused NOAA NDBC feeds (and other DTD-bearing RSS feeds) to fail with "XML with DTD detected". Parse with allow_dtd: true and keep external entity resolution disabled.

Extract feed parsing into parse_feed_entries for testability and add a unit test covering a DTD-bearing NOAA RSS document so this does not regress.

Bump patch version to 0.4.1. check-in: 95d197adde user: sourcehut@lewman.us tags: master, trunk

NOAA station RSS feeds keep the same

and <guid> while only the observation data in <description> changes. The previous ID-based cache key caused every observation update to be skipped as "already seen". Compute the cache key from a StableHasher hash of the formatted MUC message instead, so any change in the observation content produces a new key and gets posted.

Bump version to 0.4.0.

The text extraction helper was collecting each XML text string twice — once from the element node and once from the text node itself. This caused NOAA station observation values (and cache keys) to appear duplicated in MUC output. Filter descendants to text nodes only, add regression tests, and bump version to 0.3.1. check-in: 8edc3e8ddb user: sourcehut@lewman.us tags: master, trunk

Previously only the feed title and link were posted. The new "noaa-rss" format strips the HTML <description>, decodes entities, and sends a multi-line message with timestamp, wind, pressure, water temp, and station link. Also corrected the README cache_file default. check-in: 6e99afd760 user: sourcehut@lewman.us tags: master, trunk

The tokio runtime hangs indefinitely during shutdown because the XMPP crate's background TCP read tasks never terminate, leaving processes stuck after disconnect. Call std::process::exit(0) at every exit path following an XMPP connection to bypass the hung runtime cleanup. check-in: 4b403663c7 user: sourcehut@lewman.us tags: master, trunk

Previously, resp.text() only handled UTF-8, causing garbled text or parse failures for feeds in legacy encodings like GB2312 (Chinese), Big5 (Taiwanese), or Shift_JIS (Japanese). Fetch raw bytes instead, detect encoding from the XML prolog or Content-Type charset header, and decode using encoding_rs with UTF-8 fallback. Bump version to 0.2.5. check-in: ca8cf4ad25 user: sourcehut@lewman.us tags: master, trunk

Byte-slicing at a fixed index could split a multi-byte character (e.g. curly quotes), causing a panic. Use char-boundary-safe truncation instead. Bump version to 0.2.4. check-in: 2057bd6679 user: sourcehut@lewman.us tags: master, trunk

- .gitignore: add config.toml, seen_alerts.db, and seen_alerts.db.tmp to prevent committing live credentials and runtime data - Replace unwrap() on reqwest client builder with graceful error handling - Add HTTP status code check before reading feed body responses - Log body read and XML parse errors instead of silently swallowing them - Error on --config flag without a path argument instead of silently falling through to default config search - Fix RSS messages using empty link text when guid/title is the fallback ID - Bump version to 0.2.3 check-in: 121f508ca3 user: sourcehut@lewman.us tags: master, trunk

Document versioning policy in AGENTS.md and fix stale dedup flow description to reflect that entries are cached before sending. check-in: d39e66e759 user: sourcehut@lewman.us tags: master, trunk

0.2.0: SQLite dedup cache, --verbose, --cache-info flags 0.2.1: Fix duplicate posts, disconnect hang, no-alert MUC message check-in: 2fd809a402 user: sourcehut@lewman.us tags: master, trunk

Mark feed entries as seen in SQLite BEFORE sending to the MUC instead of after, so a failed XMPP delivery no longer causes every entry to be reposted on the next run.

Add a 10-second timeout on disconnect to prevent the process from hanging indefinitely when the XMPP server doesn't respond to the shutdown handshake.

Send "Nothing new to report." to the MUC when there are no new alerts, so the bot's presence is visible even on idle runs. check-in: b2d47d801f user: sourcehut@lewman.us tags: master, trunk

rusqlite 0.40 pulls libsqlite3-sys 0.38 which uses the unstable cfg_select! macro, breaking builds on platforms with older Rust toolchains like OpenBSD. rusqlite 0.39 (libsqlite3-sys 0.37) is the latest version that works on stable Rust. check-in: 5b9eb4c57b user: sourcehut@lewman.us tags: master, trunk

The TOML cache silently reset to empty on parse failures or write errors, causing every feed entry to be reposted on each run. SQLite provides atomic transactions, efficient key lookups, and SQL-based TTL purging. Uses rusqlite with the bundled feature so SQLite is compiled from source with no system library dependency.

Dedup flow: poll feeds, check each entry against the SQLite DB, send new alerts to MUC, insert keys only on successful delivery, purge entries older than 24 hours. Failed sends leave keys uninserted so they retry on the next run.

Adds --verbose for per-feed dedup stats and --cache-info to dump the SQLite cache contents. Includes 4 unit tests covering dedup, multi- feed, TTL purge, and idempotency. check-in: 9ff8280e96 user: sourcehut@lewman.us tags: master, trunk

Atom entries were never matched because the code looked for <id> as an XML attribute instead of a child element, so every Atom entry was silently skipped. RSS dedup now prefers <guid> over <link> or

, which is the element explicitly designed for this purpose. Cache keys are now namespaced by feed URL to prevent cross-feed ID collisions when two feeds share the same link URL or title.

Each feed now supports an optional emoji field to customize the prefix shown in posted messages. Defaults to the warning emoji when not specified. check-in: 2a244bd46f user: sourcehut@lewman.us tags: master, trunk

Messages now read "⚠️ Feed: Title URL" instead of just the feed name and title, for both RSS (link text) and Atom (link href attribute). check-in: 3f17ba745e user: sourcehut@lewman.us tags: master, trunk

Previously each alert opened its own XMPP connection, joined the MUC, sent one message, and disconnected. Now all alerts are collected and sent through one connection with a 500ms gap between messages. Also fixes the disconnect panic by removing the timeout wrapper that could drop the shutdown future mid-teardown. check-in: 0115effbe0 user: sourcehut@lewman.us tags: master, trunk

TLS ALPN negotiation could select HTTP/2, causing hyper-util to panic on platforms where reqwest's http2 feature isn't enabled. check-in: 44607352f8 user: sourcehut@lewman.us tags: master, trunk

Replace the append-only plaintext cache with a TOML-serialized HashMap that records when each entry was first seen. Entries older than 24 hours are pruned on load, keeping the cache file bounded. The cache is now written only after successful XMPP delivery — failed sends remove the ID so it retries on the next run. check-in: a7e1309884 user: sourcehut@lewman.us tags: master, trunk

Add a "Strict Build Requirements" section instructing agents to verify all changes with `RUSTFLAGS="-D warnings"` for both cargo build and clippy, treating warnings as hard errors and prohibiting #[allow(...)] suppression. Also correct outdated dependency versions (roxmltree 0.10 → 0.21, toml 0.8 → 1) and the tokio features description to match Cargo.toml. check-in: a565521e23 user: sourcehut@lewman.us tags: master, trunk

Use tokio::task::JoinSet to fetch all feeds in parallel, so a slow feed no longer blocks the rest. Deduplication against the seen-cache is now done centrally after all tasks complete. check-in: e8d73d1088 user: sourcehut@lewman.us tags: master, trunk

- Switch TLS from rustls/aws-lc-rs to native-tls (system OpenSSL), removing the rustls dependency and crypto provider initialization - Update reqwest 0.12→0.13, roxmltree 0.10→0.21, toml 0.8→1 - Trim tokio features from "full" to only rt-multi-thread, time, macros - Rename config.toml to config.example.toml and gitignore config.toml to prevent credential leaks - Update README.md and AGENTS.md to reflect all changes check-in: 3d8fbb6a3c user: sourcehut@lewman.us tags: master, trunk

A CLI tool that polls RSS/Atom feeds for new entries and posts them to an XMPP MUC room. Designed to run periodically via cron.

Includes config template with USGS earthquake and NWS weather feeds. check-in: 68f534d32d user: sourcehut@lewman.us tags: master, trunk