fix: deduplicate BlueBubbles webhook messages

[yu-xin-c]

Summary

• reuse the shared MessageDeduplicator in the BlueBubbles adapter
• fingerprint webhook event shape so exact replays are suppressed while same-GUID text and attachment lifecycle updates remain deliverable
• reject duplicate events before attachment download/cache work, preventing replayed side effects as well as duplicate dispatch
• cover exact replay, attachment replay, distinct GUIDs, text updates, and attachment completion

Fixes #33327

Validation

• uv run --extra dev --extra messaging python -m pytest tests/gateway/test_bluebubbles.py -q -o addopts= (61 passed)
• uv run --extra dev --extra messaging ruff check gateway/platforms/bluebubbles.py tests/gateway/test_bluebubbles.py scripts/ci/classify_changes.py
• git diff --check

[alt-glitch]

Competing with #30996 which also wires MessageDeduplicator into BlueBubbles adapter for GUID-based dedup. Both fix #30708. Also overlaps with #18395, #19976, #22116, #27985 (all open BlueBubbles dedup PRs). This PR additionally touches scripts/release.py for contributor email mapping.

[yu-xin-c]

Updated this PR to address the scope concern:

• Removed the unrelated scripts/release.py change and restored its executable mode.
• Rebased/merged onto latest main so the branch is no longer conflicting.
• Current diff is limited to gateway/platforms/bluebubbles.py and tests/gateway/test_bluebubbles.py.
• Local syntax check on the rebased files: python3 -m py_compile /tmp/hermes33337/gateway/platforms/bluebubbles.py /tmp/hermes33337/tests/gateway/test_bluebubbles.py.

The new head is 917511c2ba19702d26691ab6070c3ab746e4724b; CI is running again now.

[Mushy-Snugglebites-badonkadonk]

Thanks for working on this — deduping repeated BlueBubbles webhook deliveries is definitely useful.

One caution from testing a similar path: deduping solely by message GUID can be a little too broad unless the key also accounts for the webhook lifecycle state. BlueBubbles may deliver multiple events for the same GUID as a message transitions or gets enriched, and some of those same-GUID updates can still be meaningful.

Cases worth guarding with regression tests before this lands:

• exact replay of the same new-message payload should be suppressed;
• media/attachment completion for the same GUID should not be dropped if the first event did not yet contain usable attachment metadata;
• edits/retractions should not be suppressed only because they share the original GUID;
• reactions/Tapbacks should not be suppressed if they arrive as a later same-GUID event shape.

A possible safer shape is to dedupe on a small event fingerprint rather than GUID alone, e.g. message GUID plus event/update type plus the relevant attachment/reaction/edit marker. That still suppresses true webhook replays while preserving legitimate same-message lifecycle updates.

Happy to help adapt tests around those lifecycle cases if useful.

Disclosure: This comment was prepared with AI assistance under human direction and reviewed before posting.

[yu-xin-c]

Follow-up on the same-GUID lifecycle concern: updated the branch so webhook dedup no longer keys solely on the BlueBubbles message GUID. It now fingerprints the event shape (event type, text, associated/reply fields, item type, and attachment metadata), so exact replay payloads are still suppressed while same-GUID text updates and attachment completion events are allowed through. Added regression coverage for both same-GUID text updates and same-GUID attachment completion.\n\nVerification:\n- uv run --with ruff ruff check gateway/platforms/bluebubbles.py tests/gateway/test_bluebubbles.py\n- uv run --with pytest --with pytest-timeout --with pytest-asyncio --with aiohttp pytest tests/gateway/test_bluebubbles.py -q (60 passed)

[alt-glitch]

This was generated by AI during triage.

Duplicate of #18395 — the earliest still-open PR wiring MessageDeduplicator into the BlueBubbles adapter for GUID-based inbound dedup (both fix #30708). The previously-cross-referenced #30996 has since been closed, so this PR is now linked to the canonical open one. Saturated cluster also includes #19976, #22116, #27985, #45717 — a maintainer should pick one. (Re-pointed from closed #30996 during re-triage.)

[teknium1]

Thanks for addressing the verified replay-dispatch gap and for updating the key after the same-GUID lifecycle feedback.

Problems

• gateway/platforms/bluebubbles.py:1056 checks dedup only after the attachment loop at lines 962-996. An exact replay with an attachment still performs the second download/cache side effect before being suppressed. Move the check before that loop and add a replay-with-attachment assertion that _download_attachment runs once.
• scripts/ci/classify_changes.py:89 is unrelated to BlueBubbles and reverses current main's explicit MCP-catalog skip at line 90. Please remove this hunk from the focused fix.

Suggested changes

• Preserve the raw event fingerprint before attachment processing, then gate all downstream attachment and dispatch work on it.

Automated hermes-sweeper review.

[teknium1]

This guard is reached only after the attachment loop at lines 962-996, so an exact replay still downloads and caches every attachment a second time. Build/check the fingerprint after raw text extraction and before attachment processing; add a replay-with-attachment test asserting one download and one dispatch.

[teknium1]

This unrelated CI behavior change contradicts the current-main comment that empty/.github diffs should skip the MCP catalog lane unless catalog files changed. Please drop it from this focused BlueBubbles fix.