Repository navigation
GroveHealthKitFHIR rework - #221
Merged
Merged
Conversation
Spezi was renamed to Grove in 0.3.0 (SchmiedmayerLab/Grove, PR #43): every product `Spezi*` -> `Grove*` and `XCTSpezi*` -> `XCTGrove*`, the core type `Spezi` -> `Grove`, and no source-compatibility typealiases. This moves the app onto 0.3.0-beta.6. Package graph: - Both manifests and the Xcode project point at SchmiedmayerLab/Grove.git. The old URL still resolves through GitHub's rename redirect, but SwiftPM derives identity from the last path component, so a graph mixing the two URLs resolves one repository under two identities -- they have to move together. - MyHeartCounts-StudyDefinitions is pinned to a revision until its own Grove release is tagged (SchmiedmayerLab/MyHeartCounts-StudyDefinitions#46); every 0.2.x of it still requires the `spezi` identity. - The watch target's GroveHealthKit product dependency pointed at a package reference for the retired StanfordSpezi/SpeziHealthKit that was not in the project's packageReferences at all. It now resolves against the Grove reference, and the orphan is gone. Renames the compiler cannot do for you, because the replacement is not the `Spezi` token: `.spezi(_:)` -> `.grove(_:)`, `@Application(\.spezi)` -> `\.grove`, `SpeziAppDelegate.spezi` -> `GroveAppDelegate.grove`, and `UTType.speziStudyBundle` -> `UTType.studyBundle` (the vendor prefix is dropped, not swapped). Identifiers that name stored or transmitted data: - BGTaskSchedulerPermittedIdentifiers now lists `$(PRODUCT_BUNDLE_IDENTIFIER).scheduler.refresh`. Grove honours the old string for one release and then stops; registering an unpermitted identifier fails silently, so this cannot be left to drift. - The scheduler notification userInfo is read through `SchedulerNotifications.taskId(fromUserInfo:)`, which accepts the current key and the pre-Grove one. A direct subscript still compiles and would silently stop matching notifications already pending on a device. - MHC's own coding system moved off the framework's name to an authority we own: `https://myheartcounts.stanford.edu/fhir/CodeSystem/sampleType`. Reads accept the superseded `https://spezi.stanford.edu` spelling, so historical bloodLipids / nicotineExposure / dietMEPAScore / mentalWellbeingScore observations keep parsing. - MHC's hand-rolled sourceRevision writer derives its urls from Grove's `FHIRExtensionURL.sourceRevision` instead of hardcoding bdh.stanford.edu, so resources this app writes agree with the ones Grove writes. The backend still emits the old spellings: MyHeartCounts-Firebase#213. Left alone deliberately: the `mhc.spezi.stanford.edu` config host, the "based on the Stanford Spezi Template Application" provenance headers, and the App Store description copy. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0182tQC5aCA5KdUYixuB7tvP
The `.bundledWithApp` fallback only caught `unableToDecode`, so a fetch failure -- an offline launch, or a bucket that does not yet carry the object under its current name -- threw `unableToFetchFromServer` straight out of `_update(using:)`. That is not a soft landing: `update()`'s `returnCachedBundleOnError` can only return a bundle already loaded in this process, which on a cold launch is nothing, so the study bundle simply fails to load. Both failures now take the same path, still only for the firebase selector: an explicit `--studyBundle <url>` that cannot be fetched keeps erroring rather than quietly loading something else. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0182tQC5aCA5KdUYixuB7tvP
Grove deprecates `RWLock` (pthread_rwlock is slower than the alternatives in nearly every access pattern, and it cannot own the state it guards), which left the app with two deprecation warnings. `MyHeartCountsStandard.currentEnrollmentInfo` becomes a `Mutex<CurrentEnrollmentInfo?>`, so the lock owns the value instead of guarding a `nonisolated(unsafe) static var` beside it, and the enrollment snapshot is built before taking the lock rather than reading MainActor state inside it. `Lifecycle` guards a keypath write rather than a value, so it takes a `Mutex<Void>` -- the same swap Grove made internally for `LocalPreference`. Behaviourally equivalent: no reader relied on read-lock concurrency, and both values were only ever read and written whole. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0182tQC5aCA5KdUYixuB7tvP
# Conflicts: # MyHeartCounts.xcodeproj/project.pbxproj # MyHeartCounts.xcodeproj/project.xcworkspace/xcshareddata/swiftpm/Package.resolved # MyHeartCounts/Heart Health Dashboard/Health Dashboard/HealthDashboard+BloodPressure.swift # MyHeartCounts/Heart Health Dashboard/Health Dashboard/HealthDashboard+SleepAnalysis.swift # MyHeartCounts/Heart Health Dashboard/Health Dashboard/HealthDashboard.swift # MyHeartCounts/Heart Health Dashboard/Health Dashboard/HealthDashboardQuantityLabel.swift # MyHeartCounts/Heart Health Dashboard/Health Dashboard/HealthDashboardTile.swift # MyHeartCounts/Heart Health Dashboard/Health Dashboard/HealthStatsChart.swift # MyHeartCounts/Heart Health Dashboard/QuantitySample+FHIR.swift # MyHeartCounts/Heart Health Dashboard/QuantitySamplesQuery.swift # MyHeartCounts/Heart Health Dashboard/SleepSessionsQuery.swift # MyHeartCounts/Modules/Lifecycle.swift # MyHeartCounts/Modules/MHCBackgroundTasks.swift # MyHeartCounts/Modules/ManagedFileUpload.swift # MyHeartCounts/Modules/SetupTestEnvironment.swift # MyHeartCounts/MyHeartCountsDelegate.swift # MyHeartCounts/Onboarding/OnboardingDisclaimerStep.swift # MyHeartCounts/Root View/RootView.swift # MyHeartCounts/Supporting Files/Info.plist # MyHeartCounts/Utils/HealthKit.swift # MyHeartCountsShared/Package.swift # MyHeartCountsTests/ManagedFileUploadTests.swift
|
Important
This repository does not receive automatic reviews because it has fewer than 10 stars.
 |
HealthObservationBatch is gone. Converting is a static requirement, HealthObservation.export(_:at:using:exporter:receive:): HealthKit samples go through the Grove exporter, observations the app models itself become their own resources and never ask for an exporter. uploadHealthObservations and HealthUploadStaging are generic over the observation type and dispatch statically, so nothing is cast per element. Self-modelled observations build their issued-at FHIR instant once per batch, since creating FHIR date/time primitives from a Date was measured to be expensive. HealthUploadBatch keeps only the upload file format: the convert-and-write overload is gone and the file-name rule moved in as HealthUploadBatch.filename. Grove refusals of HealthKit samples log as errors. Tests follow the new API. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
FHIRExchangeLedger holds the app's one HealthKit exporter, fenced to its account generation. Getting it is a lock and a comparison; it throws healthKitExporterUnavailable while none is built and staleAccountGeneration for a batch captured before a logout. MyHeartCountsStandard rebuilds it in refreshHealthKitExporter() at launch, sign-in and after enrollment changes, and clears it on logout, so the identity secret is read and the producer set up only when the account or the enrollment changes, not per batch. Live staging, retractions and the storage upload path look the exporter up; the conversion closure is synchronous, and self-modelled observations still never ask for one. Retractions run on the exporter (healthKitRetractions(of:recordedAt:)). The historical uploader still builds its exporter per batch. Tests follow the new API; tests keep building their exporter from the staging overrides. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…rter With the Historical Upload HealthKitFHIRExporter.RecordProvider replaces HealthObservationRecords: one parameter instead of a HealthKit and a records source. .healthKit(HealthKit) fetches an ECG's voltages and correlated symptoms and states its voltage count for the batch split; _custom exists only for the batch-export tests. HealthObservation.export takes the provider, and the exporter as an autoclosure, so self-modelled observations never evaluate it. HealthKitSamplesFHIRUploader holds its HealthKit and is created from the Standard asynchronously. It now looks up the app's one HealthKit exporter, the same instance live ingestion uses, instead of building one per batch; a batch that runs before the launch refresh has built it refreshes once and retries. This replaces the per-batch build that e2ffd4b still left for the historical upload. Tests follow. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The ledger used a rollback journal (journal_mode = TRUNCATE) with synchronous FULL and fullfsync, which Xcode flags as excessive I/O for a database not in WAL mode. It now sets journal_mode = WAL itself in prepareDatabase, keeping synchronous = FULL, fullfsync and adding checkpoint_fullfsync: every commit still reaches storage before it returns, with fewer syncs per commit. GRDB's journalMode = .wal is not used because it lowers synchronous to NORMAL after prepareDatabase ran (seen in GRDB 7.11.1's Database.setUpWALMode(), not documented), and a lost reservation could mint an event sequence again. An existing database switches to WAL when next opened; the process-exclusive lock guarantees no other connection holds it. The log files live in the ledger's directory, which is already excluded from backup and deleted as a whole on a discard. The durability test now pins wal and checkpoint_fullfsync. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With the --measureExporterThroughput launch argument, the app's HealthKit exporter measures its throughput (Grove's Options.measuresThroughput); every export batch then logs the cumulative report, and the debug form's "HealthKit Exporter Throughput" screen shows it live: totals and rates, each phase's share, and each HealthKit type, with a button that copies the report. Live ingestion and the historical upload share the one exporter, so the numbers cover both. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Grove's HealthKit, SensorKit and Questionnaire exporters now export and retract asynchronously and build a call's graphs on a task group, on every core by default. Their entry points are nonisolated(nonsending), so a call runs on the caller's executor and the receive closures keep capturing local state unchanged; MyHeartCounts keeps the default concurrency. The batch export's flush, the Questionnaire observation extraction, the SensorKit publication and the HealthKit retraction helper await the calls, and their callers await them in turn. The test support's withTemporaryDirectory takes an async body, and the concurrent ledger contract test runs its nine workers on a task group, each returning its failure count, instead of on DispatchQueue.concurrentPerform. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The two tests that corrupt the ledger's database file expected the open connection to read the overwritten pages back, which bumping the file change counter forced under the rollback journal. Since the ledger keeps a write-ahead log, an open connection never reads that counter: it serves the pages it cached, and the newest pages live in the log, so neither test saw an error. The helper now checkpoints the log into the database file, closes the ledger's connection and then overwrites the ledgerEntry pages, so the transaction after the reopen reads them and fails as corrupt. Both tests keep their assertions; corruptPageDiscardsTheDatabase now states that the page is corrupted while the ledger is closed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…codings Each SensorKit upload strategy names its sidecar's file extension again where it writes the bytes, as on main: PPG recordings are .mhcPPG2, tabular recordings .csv, ECG and structured payloads .json. SensorKitUploadSidecar carries that extension instead of a recording format whose computed extension nothing tied to the bytes (Grove dropped it); PPG sidecars were .bin. decode-ppg reads both PPG encodings through Grove: .mhcPPG files with Grove's legacy decoder and .mhcPPG2 files with the registered one. The bytes carry no version, so the extension decides; a .zstd or .zlib suffix on an uncompressed earlier file is looked through. On 14 real .mhcPPG files the new command yields the same samples as the previous one; on a .mhcPPG2 file the previous command read every value byte-swapped. Dates are now ISO 8601 with the stored instant's exact fractional seconds; .iso8601 dropped them, shifting every sample's offset by up to a second. MHCDataSpec describes both encodings. Building SensorKitCLI from MyHeartCountsShared alone needs its Grove pin moved to a revision with the legacy decoder. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Moves the Grove dependency from e1cbf2fd to b20cc933 (the head of Grove's lukas/grove-fhir-api-changes) and MyHeartCounts-StudyDefinitions from 49289ddd to aaa19dcf (its lukas/grove-fhir-api-changes, which pins the same Grove revision), in the app project and in MyHeartCountsShared. In Package.resolved only these two pins and the origin hash change; every other package stays at its committed version. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Grove reads legacy .mhcPPG files (SensorKitPPGRecording(legacyMyHeartCountsData:)) and writes the current format, and the SensorKit CLI decodes both through it, so MyHeartCountsShared's BinaryEncoder/BinaryDecoder and PPGSample's binary conformance had no caller left. They are removed with their tests, and so is the swift-nio dependency only they used. Grove keeps a test that decodes bytes this encoder wrote for the same three samples. PPGSample's JSON round trip, which the CLI's output relies on, stays as PPGSampleCodingTests. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Moves Grove from b20cc933 to eeab2ef5, the head of Grove's lukas/grove-fhir-api-changes, and the study definitions from aaa19dc to 9346b78, which pins the same Grove revision. Grove now reports a workout activity only as a shared activity it is a kind of, starts PPG recordings with a version byte, keeps concurrent exports from waiting on each other, and projects category, percent and rate observations back to HealthKit, which questionnaire extraction uses. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Launched with --recordHealthKitExportLog, the app appends one JSON line per HealthKit batch export to Library/Application Support/ HealthKitExportLog/export-log.jsonl: the batch's origin (live or historical), samples by type, duration and outcome, with Grove's refusals and warnings counted by kind or the error it failed with, then the exporter's running throughput totals by phase and source type. Every line names the app launch, device, OS and build it was written on, so a test run's log can be copied off a connected device with devicectl and analysed per run. The flag also turns on the exporter's throughput measurement; without it nothing is written. A historical batch counts as exported once its upload file is staged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A category's `Progress` and its pending files' size change without notifying SwiftUI, so the debug view showed the queue as it was when opened. It now rereads the queue every two seconds and when pulled down, shows when it last did, and states each category's completed and total upload counts. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Moves Grove from eeab2ef5 to 46f5ef5f, the head of Grove's lukas/grove-fhir-api-changes, and the study definitions from 9346b78 to 3a40a9c, which pins the same Grove revision. Grove now hashes every attachment through one shared SHA-1, so its FHIR modules build on Linux, and drains each built record's temporaries inside its concurrent build. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
lukaskollmer
marked this pull request as ready for review
October 16, 2026 20:17
Moves Grove from 46f5ef5f to 6e40cfbf and the study definitions from 3a40a9c to fa1b3ee, which pins the same Grove revision. Both histories were re-created with every commit signed; the code is identical, only the commit identifiers changed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
lukaskollmer
force-pushed
the
lukas/grove-fhir-api-changes
branch
from
October 16, 2026 20:32
14e0166 to
3239341
Compare
github-project-automation
Bot
moved this from In Progress
to Done
in MyHeart Counts Project Planning
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
♻️ Current situation & Problem
reworked and updated MHC to use the new Grove FHIR APIs introduced in SchmiedmayerLab/Grove#118
⚙️ Release Notes
HealthKitFHIRExporterAPI (see also [WIP] rework GroveHealthKitFHIR Grove#118)📚 Documentation
yes
✅ Testing
yes
Code of Conduct & Contributing Guidelines
By creating and submitting this pull request, you agree to follow our Code of Conduct and Contributing Guidelines: