fix(announces): type peers/relays by aspect, stop relays leaking into Peers

[torlando-tech]

Problem

In the Network announces tab, the Peers filter (the default) also showed relays.

Two stacked bugs:

1. The filter keyed on the wrong field. .peers filtered on aspect == "lxmf.delivery" || aspect == nil while .relays filtered on the independent isRelay flag. A contact that was isRelay == true and carried a delivery/nil aspect satisfied both predicates, so it appeared under the default Peers view.
2. The classifier mis-flagged peers as relays (root cause). Contact.init(from:) ran PropagationNodeInfo.parse(app_data) before the aspect branches, and that parser accepts any ≥3-element msgpack array without verifying the destination is a propagation node. So a lxmf.delivery (or audio/site) announce carrying such app_data got tagged isRelay = true while keeping its real aspect. (Columba's own delivery announces are 2-element arrays, which is why only some relays leaked — it fired on other clients / newer LXMF formats.)

Fix

Make the aspect the sole signal for announce typing, matching the reference clients:

• Sideband types each announce by which RNS aspect handler received it (aspect_filter == "lxmf.propagation" vs "lxmf.delivery").
• Android does the same via NodeType.fromAspect.

Changes:

• Contact.init: removed the PropagationNodeInfo.parse(app_data) relay heuristic. Relays are now strictly aspect == "lxmf.propagation" (via entry.isLXMFPropagationNode); an unrecognized/empty aspect defaults to .peer (matches NodeType.fromAspect's default).
• Filters: .peers / .relays now key on the mutually-exclusive badgeType (.peer / .relay), symmetric with .sites, so the four node types can't overlap.
• handlePythonEvent: added a warning if an announce ever arrives with an empty/unrecognized aspect — since there's no app_data fallback anymore, this surfaces it loudly instead of silently classifying as a peer.
• app/rns_bridge.py: corrected the stale "lxmf.delivery only" scope docstring so the per-aspect propagation handler can't be narrowed away by accident.

Why nothing regresses

Both announce backends guarantee a genuine propagation node arrives tagged "lxmf.propagation" or is dropped before reaching Contact.init, so the removed heuristic only ever caught false positives:

• Python bridge registers one RNS announce handler per aspect (aspect_filter), and RNS enforces the filter by cryptographic destination-hash match — a real propagation node can only emit "lxmf.propagation". Worst case is omission (no event), which the fallback couldn't have rescued.
• Native Swift/NE derive detectedAspect cryptographically; a propagation node yields "lxmf.propagation" or nil (dropped before emit). A mistag would require a SHA-256 collision.

Tests

Adds AnnounceClassificationTests (the filter had no prior coverage) — 7 cases pinning the classification, including the exact leak (delivery + ≥3-element app_data → .peer), the new strict contract (untagged propagation-shaped app_data → .peer, not relay), and unknown aspect → .peer. Full ColumbaAppTests suite passes (98 tests, 0 failures) on the iPhone 17 simulator.

🤖 Generated with Claude Code

[greptile-apps]

Greptile Summary

Fixes a double bug where relays leaked into the default Peers view: the filter keyed on aspect (which didn't exclude isRelay), and the classifier mis-flagged non-relay announces as relays by accepting any ≥3-element msgpack array as propagation-shaped app_data. The fix makes badgeType (set exclusively from entry.isLXMFPropagationNode) the single discriminator for all four filters, removing the now-redundant PropagationNodeInfo.parse heuristic.

• Contact.init: drops the PropagationNodeInfo.parse app_data fallback; all announce typing now flows through isLXMFPropagationNode / isLXSTTelephony / detectedAspect, matching Sideband and Android's aspect-first contract.
• applyAnnounceFilters: all four filter cases (.peers, .audio, .relays, .sites) now key on the mutually-exclusive badgeType, so overlap is impossible by construction.
• AnnounceClassificationTests: adds 7 new tests including a direct regression pin for the delivery-peer-with-3-element-app_data leak case.

Confidence Score: 5/5

Safe to merge — the fix is surgical, well-tested, and aligns Columba with the reference clients' aspect-first classification contract.

The change removes a single mis-scoped heuristic and replaces it with a single, mutually-exclusive discriminator that matches how both Sideband and Android already type announces. All four filter predicates now key on the same field, so the overlap that caused the leak is structurally impossible. Seven new tests pin the regression case and the full test suite passes.

No files require special attention.

Important Files Changed

Filename	Overview
Sources/ColumbaApp/ViewModels/ContactsViewModel.swift	Core fix: removes PropagationNodeInfo.parse fallback from Contact.init and rewrites all four filter predicates to key on the mutually-exclusive badgeType. Logic is sound and symmetric.
Tests/ColumbaAppTests/AnnounceClassificationTests.swift	New test file with 7 cases covering the regression (delivery peer with ≥3-element app_data), relay classification, unknown/nil aspect defaulting to peer, and audio/site no longer mis-flagged.
Sources/ColumbaApp/Services/AppServices.swift	Adds a diagnostic warning log when an announce arrives with an unrecognized aspect. The knownAspects set is correct and the warning fires before the PathEntry is inserted.
app/rns_bridge.py	Docstring correction only: broadens the 'lxmf.delivery only' scope note to reflect all four tracked aspects and warns against narrowing, preventing future regression by accident.
Columba.xcodeproj/project.pbxproj	Registers the new AnnounceClassificationTests.swift source file in the Xcode project with consistent ACT01/ACT02 UUID references.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[PathEntry arrives] --> B{entry.isLXMFPropagationNode?}
    B -- yes --> C[badgeType = .relay\nisRelay = true]
    B -- no --> D{entry.isLXSTTelephony?}
    D -- yes --> E[badgeType = .audio\nisRelay = false]
    D -- no --> F{detectedAspect ==\n'nomadnetwork.node'?}
    F -- yes --> G[badgeType = .node\nisRelay = false]
    F -- no --> H[badgeType = .peer\nisRelay = false\n'lxmf.delivery' or unknown]

    subgraph applyAnnounceFilters
        I[.peers] -->|badgeType == .peer| J[✔ peers only]
        K[.relays] -->|badgeType == .relay| L[✔ relays only]
        M[.audio] -->|badgeType == .audio| N[✔ audio only]
        O[.sites] -->|badgeType == .node| P[✔ sites only]
    end

    C --> K
    E --> M
    G --> O
    H --> I

Loading

%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
flowchart TD
    A[PathEntry arrives] --> B{entry.isLXMFPropagationNode?}
    B -- yes --> C[badgeType = .relay\nisRelay = true]
    B -- no --> D{entry.isLXSTTelephony?}
    D -- yes --> E[badgeType = .audio\nisRelay = false]
    D -- no --> F{detectedAspect ==\n'nomadnetwork.node'?}
    F -- yes --> G[badgeType = .node\nisRelay = false]
    F -- no --> H[badgeType = .peer\nisRelay = false\n'lxmf.delivery' or unknown]

    subgraph applyAnnounceFilters
        I[.peers] -->|badgeType == .peer| J[✔ peers only]
        K[.relays] -->|badgeType == .relay| L[✔ relays only]
        M[.audio] -->|badgeType == .audio| N[✔ audio only]
        O[.sites] -->|badgeType == .node| P[✔ sites only]
    end

    C --> K
    E --> M
    G --> O
    H --> I

Loading

_{Reviews (2): Last reviewed commit: "fix(announces): key .audio filter on bad..." | Re-trigger Greptile}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!