Add per-device truth for multi-computer dives (PRO-67, PRO-34)

[shaheinm]

Summary

Implements the multi-computer dive design: one dive record, per-device truth for samples, gases, and deco settings, with the device picker as the source of truth in the UI.

Phase A — import correctness (PRO-67):

• Extract shared GasMixMergeHelper (dedupe by composition, gasmixIndex remap, per-device device_id) used by both BLE and Shearwater Cloud import paths — Cloud merge previously never remapped sample gas indices
• Filter imported gases to slots actually referenced by that device's samples (no more phantom programmed-slot gases)
• Gate Cloud time-proximity merging on device ownership — buddy computers (ownership: other) no longer merge into owned dives
• Dive-level GF/deco metadata comes from the primary device instead of silent first-wins

Phase B — per-device settings + UI (PRO-34):

• Migration 018: dive_device_settings table (GF low/high, deco model, salinity, surface pressure, is_primary per computer); populated on all import paths, backfilled on re-import (extends the PRO-62 pattern)
• DiveDetailView: device picker drives gas list, deco/GF display, and Replay prefill; defaults to primary device on multi-computer dives; subtle footnote when computers disagree ("Varies by computer — showing Petrel")
• ReplayProfileSheet: GF/deco model prefill from the selected device's settings
• splitDive moves the device's settings row to the new dive and promotes a new primary on the original
• DiveService.hasConflictingDeviceSettings() + getDeviceSettings() helpers

Repair path: re-importing the Shearwater Cloud .db fixes existing merged dives (gas attribution, per-device settings rows) — same flow as PRO-62.

Closes PRO-67, PRO-34.

Test plan

• 413 Swift tests pass (swift test), including new GasMixMergeHelperTests, Cloud ownership-gate/settings tests, and split settings tests
• make lint clean (SwiftLint 0 violations)
• macOS build (xcodebuild -scheme Profundum -destination 'platform=macOS')
• Manual QA: re-import Shearwater .db on a Petrel+Perdix dive; confirm gas list and GF follow the device picker; verify conflict footnote when GFs differ

Made with Cursor

[shaheinm]

Self-review (project checklist)

Validation run locally:

• make lint — SwiftLint 0 violations
• swift test — 413 tests, 0 failures (Rust core untouched, so rust-test/security/mutants not applicable)
• macOS build via xcodebuild -scheme Profundum — succeeds, only pre-existing ImportSession Swift-6 warnings
• Migration 018 is append-only; composite PK + ON DELETE CASCADE on dive_id; GRDB decodes is_primary INTEGER → Bool correctly (covered by split/import tests)

Correctness spot-checks:

• gasmixIndex remap is nil-safe: a sample referencing an index with no persisted mix stores nil (previously the Cloud path stored dangling raw indices)
• Cloud new-dive path re-fetches existing mixes per device inside one write transaction — SQLite sees its own writes, so cross-device dedupe and nextMixIndex sequencing are correct
• Backfill only parses the binary log when a dive-level field or the per-device settings row is missing — re-import of fully populated DBs stays cheap (PRO-62 invariant preserved)
• Legacy dives (no settings rows yet): deco section falls back to dive-level fields in both the visibility gate and the section body

Known gaps, deliberately deferred:

1. ExportService does not export dive_device_settings — an export/import round-trip loses per-device settings (recoverable via Shearwater re-import). Suggest a small follow-up ticket.
2. divesBackfilled counter reports dive-level field backfills only; settings-row-only backfills aren't counted (cosmetic).
3. Ownership gate uses chain grouping: an ownership: other row between two owned rows within the 2-min window prevents the owned rows from merging with each other. Unlikely in practice (requires a buddy device starting between your two computers).
4. Devices with settings rows can't be hard-deleted (FK, no cascade) — consistent with existing samples.device_id / gas_mixes.device_id FKs; devices are archived, not deleted.

Manual QA still required (device-dependent): re-import Shearwater .db on a Petrel+Perdix dive; confirm gas list + GF follow the device picker and the conflict footnote appears when GFs differ.

[shaheinm]

Second-opinion review (GPT-5.3 Codex) — findings addressed in f9b9996

The independent review found two major issues, both real and both fixed:

1. Interleaved buddy row broke owned-computer merging. The ownership gate compared each row only to the previous one, so mine A → other B → mine C within the 2-minute window split A and C into separate dives. Fixed: buddy-owned rows now form singleton groups without terminating the current group, so owned computers chain directly. Regression test: testOwnedComputersMergeAcrossInterleavedBuddyRow.

2. mergeDevices didn't reassign dive_device_settings. Device merge (Cloud/BLE duplicate consolidation) moved dives, samples, gas mixes, and fingerprints to the winner but left settings rows pointing at the archived loser. Fixed: settings rows are reassigned, and when a dive already has a winner row the loser's duplicate is dropped instead of violating the composite PK. Regression tests: testMergeDevicesReassignsDeviceSettings, testMergeDevicesDropsDuplicateSettingsRow.

3 (minor). Test coverage gaps — covered by the three tests above.

Full suite: 416 tests, 0 failures. Reviewer verdict was "request changes"; both changes are now in.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!