The reproducible build OOM-killed the Gradle daemon ('daemon disappeared
unexpectedly') mid flutter-build-apk on the shared runner. The app's
gradle.properties reserves -Xmx8G+4G metaspace (~12G), and the fdroid
build also compiles tesseract4android from source first (its daemon
lingers), so combined peak RAM trips the OS OOM-killer under contention
(the same build succeeded on a quieter host earlier). Scope the fix to
the fdroid build (don't touch the committed 8G the play/local builds
rely on): run tesseract's gradle with --no-daemon so it frees memory
before the app build, and append a lower -Xmx4g + org.gradle.daemon=false
override to gradle.properties in prebuild (last-key-wins; F-Droid runs
the same steps and memory settings don't change output bytes, so it stays
reproducible).
The plugin's android/build.gradle pins its own AGP 7.1.2 and resolves it
via jcenter() (dead since 2021). On a cold Gradle cache the AGP-7.1.2
transitives (apkzlib -> fastutil/json-simple/javawriter) only resolve
from jcenter and fail hard, breaking the reproducible fdroid build.
Warm-cache runs masked it (the one green armeabi-v7a run never hit
jcenter). Strip that buildscript block in prebuild so the plugin
inherits the app's AGP 8.11.1 (settings.gradle.kts) and never touches
jcenter or old AGP at all.
Mirror the es/ pattern: docs/legal/pt/ + docs/o-que-e-tane.md as the
Portuguese source of truth, build-legal.sh extended to generate the
Hugo pt.* pages from them, config.toml gets a [languages.pt] entry,
i18n/pt.json translates the UI chrome. og-pt.png is a placeholder copy
of og-en.png until a proper Portuguese social card is made by hand
(same as the other og-*.png, per DEPLOY.md). Verified locally with the
Docker build: /pt/, /pt/about/, all four /pt/legal/ pages and the
lang-switch link all resolve correctly.
Device compatibility (regression fix):
- The CAMERA permission (from zxing_barcode_scanner / image_picker) implicitly
required android.hardware.camera, excluding camera-less devices — Play showed
Automotive -96%, Chromebook -86%, TV -25%. Declare camera & location
uses-feature required="false" to keep those devices supported.
- Detect a real camera at runtime (PackageManager.FEATURE_CAMERA_ANY via the
existing MethodChannel), cache it at bootstrap, and hide the QR scan button and
the camera photo-source option when absent, so no broken actions are offered.
Play "app optimization" recommendations:
- Enable R8 (isMinifyEnabled + isShrinkResources) with keep rules for the
OCR (tesseract4android), SQLCipher and notifications JNI/reflection code.
- Bitmap downscaling: cacheWidth/cacheHeight (and ResizeImage for avatars) on
list thumbnails and avatars so photos decode to on-screen size, not full res.
- Edge-to-edge: opt in with transparent system bars in main().
Release flow:
- Tagged builds now publish to the production track at 100% (was internal);
add a manual deploy_internal lane as a QA safety net.
- Document country/region availability as a Play Console setting (not in-repo).
Closes the physical↔digital loop the labels opened: the QR on a printed
envelope (tane://seed, already encoded by seed_label_codec) can now be
scanned from the inventory bar. A known label opens its variety; an
unknown one asks before adding anything, then creates the variety (species
linked by exact scientific name when bundled) with a lot carrying the
label's year and origin. Scanner is pure ZXing (zxing_barcode_scanner,
MIT) — no ML Kit, no Play Services, F-Droid-safe, the stack Ğ1nkgo ships.
Mobile-only; other platforms don't show the button (OCR precedent).
The post-scan flow is a plain function (handleScannedPayload) so it's
widget-tested without a camera.
Bumps to 0.1.1+3 (versionCode 3). Adds CHANGELOG [0.1.1], fastlane
changelogs (en-US, es-ES) for versionCode 3, and a 0.1.1 build entry in
the F-Droid recipe (commit v0.1.1, CurrentVersion/Code 3).
- chown /opt/flutter to vagrant (build runs as non-root; fixes cache write)
- AutoUpdateMode: Version (drop invalid 'v%v' per schema)
- UpdateCheckMode: Tags v[\d.]+ + UpdateCheckData reading version from
apps/app_seeds/pubspec.yaml (Flutter versionCode isn't a literal in gradle)
Validated locally: schema, lint, checkupdates (exit 0), and flutter build as vagrant.
Categories now Inventory + Market & Price + Science & Education (F-Droid has no
'Tools'; these match the app). Build entry and CurrentVersionCode bumped to 2 to
match the shipped 0.1.0 release. Description/screenshots come from the fastlane
metadata tree, not this recipe.
aaron runs forgejo + jenkins but no act_runner, so workflows queue until a runner
is registered (or CI moves to Jenkins). Record that the keystore alias is tane-upload.
git.comunes.org runs Forgejo Actions, not GitLab CI, so the tag-triggered
signed-AAB build + fastlane Play upload now lives in .forgejo/workflows/release.yml
(single job, no cross-job artifact passing). Removed the equivalent GitLab jobs;
.gitlab-ci.yml keeps only the analyze/test gate. Play service account is reused
from Ğ1nkgo. docs/release.md updated (Forgejo secrets).
- fastlane: Appfile/Fastfile/Gemfile with deploy_play lane (AAB -> Play internal track)
- CI: tag-gated build:android (signed AAB/APK from CI secrets) + deploy:play jobs
- F-Droid: fdroiddata build recipe at docs/fdroid/org.comunes.tane.yml
- Play compliance: Data Safety / content-rating answer sheet (docs/legal/internal)
- docs/release.md: automated tag-triggered flow, dedicated tane-upload keystore
- pubspec: description now mentions the market, not just the inventory
Beta-tester feedback triage. The data model already had LotType.seedling as
first class; the gaps were copy and UX.
- Seedlings/plantones: broaden landing hero (EN/ES), in-app legal notice, and
all legal texts (site + docs masters) from seeds-only to "seeds and
seedlings", with a live-plant phytosanitary/transport caveat. Seeds stay the
hero — targeted, not a blanket rename.
- Backups: settings line now states the 7-day cadence explicitly; cadence lives
in AutoBackupService.backupInterval as the single source for text and schedule.
- Community servers: replace the manual wss:// text box with a checklist of the
known servers (defaults visible/toggleable) plus an "add server" affordance
with basic validation. No jargon in the UI.
i18n across en/es/pt/fr/de/ast (ja falls back). Tests updated + a new
server-picker widget test.
- Hugo site (site/): EN/ES intro + legal (generated from docs/legal), an
ES/EN About page from docs/{que-es-tane,what-is-tane}.md, seed-green theme,
responsive header, store-badge placeholders, multi-stage Hugo->nginx image.
- Golden screenshot harness (test/screenshots/): en,es,fr,de,pt,ja + RTL demo,
real fonts via DejaVu Sans + FontManifest; skip-by-default tag so CI stays
green. collect_screenshots.sh feeds site/ and fastlane phoneScreenshots.
- Drop editorial notes from the public explainer.
Flip plantare-bilateral.md from "design, not yet built" to "v1 shipped",
summarize what landed (transport / schema v12 / service / UI), resolve
the open questions inline with the decisions made, and list the honest
follow-ups (durable outbox, accept-after-restart, rating anchor,
provenance DAG, paper/group/forgiveness sync).
Bundle Noto Sans Arabic + Noto Sans JP (SIL OFL 1.1) as per-glyph fallbacks
for both the UI text theme and every generated PDF, so RTL (Arabic) and CJK
render on all platforms — including desktop, where the system font may lack
them — instead of tofu. A shared pdf_fonts.dart centralizes the PDF theme and
the three label/catalog/recovery services reuse it.
Add Japanese (ja) as the CJK reference locale — fittingly, the app's own name
is Japanese (種, tane, 'seed'). Core strings translated; the rest fall back to
English via slang. Wired into supportedLocales (automatic) and the settings
picker.
Fix three physical Alignment.centerLeft -> AlignmentDirectional.centerStart in
variety_detail_screen, and make month-abbreviation and avatar-initial
truncation grapheme-safe (.characters) so CJK text is never cut mid-character.
Tests: CJK/Arabic PDF glyph render, ja locale resolution + English fallback.
The chat-bubble mirroring flagged by the audit was verified correct (already
respects Directionality) and left unchanged.
The Plantaré screen stored madeOn/dueBy/varietyId but showed none of them,
so it felt half-built. Surface what's already there (no schema change):
- watchPlantareViews() joins each commitment with its variety label
- tile shows the made-on date, a "Return by {date}" line (amber + "overdue"
when an open promise is past due, via new seedWarning colour), the variety
name, and taps through to /variety/:id
- add sheet gains an optional return-by date picker (createPlantare already
took dueBy)
- variety detail shows a read-only commitments section (hidden when empty,
no add button — honours the single hand-over door)
- i18n en/es/pt/ast; tests for the join, dueBy round-trip, tile, and section
Also specs the deferred two-sided record in docs/design/plantare-bilateral.md
(Nostr pubkey counterparty, PlantareTransport propose->accept->counter-sign,
deferred key/signature/movement columns, provenance DAG, reputation anchor),
with a faithful transcription of the paper Plantaré v0.4 (BAH-Semillero 2009,
CC-BY-SA). Cross-linked from data-model §2.7 and sharing-model §6.
Fastlane title.txt + full_description.txt (en-US, es-ES) now lead with
Tane and keep tanemaki (種まき) as etymology; CI comment and the Material
mockups updated too.
Make Tane primary in README, VISION, CLAUDE, the intros and the
plain-language explainer (renamed que-es-tanemaki.md → que-es-tane.md,
dropping its "provisional name" caveats); keep Tanemaki (種まき) only as
the etymology.
PLAN.md is the origin record: its naming section §5-ter is preserved
verbatim as history under a dated "superseded" note, and the Nota de
vigencia now records Tane + the tane.comunes.org domain.
A short, all-audiences write-up derived from VISION.md: names concepts
like free software / copyleft / decentralised and explains each in plain
words rather than assuming them. States the name is tentative.
- Anchor the message list to the bottom (`reverse: true`) so new messages
stay in view instead of landing below the fold.
- Move chat history from the OS keystore (O(n²) JSON blob, silent 200-msg
cap) to a separate encrypted Drift/SQLCipher DB (`ChatDatabase`):
indexed append, uncapped history, dedup as a unique-key invariant. It's
an ephemeral per-device cache, isolated from the inventory schema, its
migrations, and its sync. No data migration (pre-release).
- Add day separators (Today/Yesterday/locale date) and a per-bubble time,
all via ICU (12/24h per locale; Localizations locale maps Asturian →
Spanish for intl date symbols).
- Add peer avatars (deterministic colour from the pubkey + name initial),
surface send failures that were previously silent, and make bubble text
selectable (addresses, links).
- New i18n keys in en/es/pt/ast; tests for grouping, formatting, avatars,
scroll anchoring, storage and send errors.
Docs: docs/design/chat-storage.md + open-decisions.md.
The most foundational Block-2+ gap: the data model was built CRDT-ready
(HLC, tombstones, LWW, append-only Movement) and the import reconciler
proves the merge — but nothing replicated a device's inventory to your
OTHER devices. Only manual file backup/restore moved data between devices.
This lands the transport foundation in commons_core (seed-agnostic, on the
shared connection — the 'one connection, N interfaces' shape):
- SyncTransport: moves an opaque, encrypted snapshot blob between ONE
identity's own devices. The core carries bytes + does the crypto; the app
decides what the bytes are and how to merge them.
- NostrSyncTransport: NIP-78 app-data (kind 30078, addressable/replaceable
per device via the d-tag), content NIP-44-encrypted TO SELF, discovery
filtered to your own author key. Each device keeps one replaceable record.
Tests (MiniRelay, pure Dart): a snapshot replicates to the identity's other
device; the relay only ever sees ciphertext; another identity neither
receives nor could decrypt; re-push replaces; two devices keep their own.
Confirms the open-decisions §D.4 guarantee (sync leaks neither the key nor
plaintext). NEXT slice (app): a SyncService that serializes the inventory
(reusing ExportImportService), pushes on mutation, and runs the existing
idempotent importInventory on receive; a stable per-install device id;
wire sync into SocialSession.
Slice 4 of Block 2. The pure rule was already parameterised; this adds the
policy, cold-start and UI around it.
- commons_core: WotParams (sigQty/stepMax/sigValidity, Duniter defaults) +
WebOfTrust.membersWith; npubToHex helper (NIP-19 decode) for adding roots.
- WotSettings (keystore): the active parameters, configurable — a young
network loosens them and tightens as it grows, as Ğ1 did. Defaults Duniter.
- TrustReferents: the bootstrap 'seeds' membership is measured from — a
bundled asset (empty until real founders are curated, no invented keys)
unioned with referents the user adds by npub/QR. Honest cold-start.
- TrustCubit: computes the full membership verdict against referents+params
alongside the personal circle, and exposes a TrustTier (networkMember >
inYourCircle > vouched > unknown). Certifications issued with the active
validity (they expire and renew, Duniter rule).
- UI: chat trust badge by tier; a 'Network of trust' screen (manage roots +
advanced params) reached from the profile. i18n en/es/pt/ast.
Tests: WotParams/membersWith, npubToHex, TrustReferents, WotSettings, and
TrustCubit tiers/membership. Resolves the WoT-parameters decision
(open-decisions §B). Trust net stays empty/undetermined until seeded — by
design; users bootstrap their own roots.
Adds pseudonymous, switchable social identities derived from the SAME root
seed via an account index (NostrKeyDerivation.deriveFromSeed(seed, account)).
HKDF is one-way so accounts are unlinkable to the Ğ1 key; account 0 is the
original identity, byte-for-byte unchanged (no rotation for current users),
and every account regenerates from the single seed — so switching adds
nothing to back up.
- SocialAccountStore: keystore-backed active account + max created.
- Per-identity stores (chats, profile, name cache) namespaced by account
scope (account 0 = legacy keys, no migration) so identities never mix.
- switchSocialAccount() re-derives the identity, re-scopes the stores and
restarts the inbox listener; RestartWidget rebuilds the tree to pick up
the new social singletons. DB/inventory untouched.
- Profile 'Your identities' switcher: list, create, switch (with a note
that messages/contacts are kept separate per identity). i18n en/es/pt/ast.
Tests: account-indexed derivation (legacy 0 unchanged, accounts distinct
yet deterministic, negatives rejected); SocialAccountStore; per-identity
store scope isolation. Resolves the flagged 'change identity' decision
(open-decisions §B).
Known: switching resets navigation to home (full tree rebuild).
User-flagged (on-device feedback): no way to change the social identity today
(derived once from the root seed at startup). Records the two options —
pseudonymous social key (recommended) vs full root-seed reset — and the runtime
re-derivation implications, for a future round.
The de-risking spike validated the whole social happy path, so Block 2 moves
from "do NOT start" to STARTED. Records the build order (transport foundation in
commons_core first, on the vetted pure-Dart nostr lib) and marks open-decisions
D.3/D.7 as resolved by the spike.
Adds TrustTransport as the third interface on the shared NostrConnection,
completing the social-layer happy path in the spike:
- WebOfTrust: pure, Flutter-free Duniter membership rule (N certs from members +
within distance D of bootstrap referents), iterated to a fixpoint. The
commons_core-worthy piece; TDD'd (threshold, distance cutoff, transitive
growth, expired/revoked/self exclusion).
- NostrTrustTransport: certifications as a custom addressable kind (30777,
keyed by issuer+subject) — certify renews, revoke replaces, certs expire.
- Trust filters spam (network-trust §2): seeds certify Alice; Eve stays
uncertified; Bob discovers both offers and annotates authors — Alice known,
Eve unknown — all over ONE shared connection.
Findings updated: WoT risk High -> Medium (policy/bootstrap, not feasibility);
overall scope Medium-High -> Medium — all three transport contracts now proven.
30 tests green, analyzer clean, Block 1 untouched.
Extends the Block 2 spike to attack the highest-risk unbuilt piece (private
messaging) and to prove the Q2 architecture recommendation with running code:
- NostrConnection: one shared socket+key+sign+REQ/EOSE lifecycle. Refactored
NostrOfferTransport onto it; added NostrMessageTransport — two thin contracts
on ONE connection (the 'one connection, three interfaces' shape).
- NIP-44 v2 encryption (secp256k1 ECDH -> ChaCha20 + HMAC, length-hiding pad).
- NIP-17/NIP-59 gift-wrap: rumor(14) -> seal(13) -> wrap(1059, ephemeral key).
Alice->Bob DM round-trips through the relay: encrypted, sender authenticated,
and metadata-private (relay/eavesdropper sees only ciphertext, an ephemeral
author and the recipient p-tag — Alice stays hidden).
Findings updated: NIP-17 risk drops Medium-High -> Medium (hardening, not
feasibility); next de-risking target is the web of trust. NIP-44 not yet
vector-verified (noted). 20 tests green, analyzer clean, Block 1 untouched.
Throwaway research spike (spike/ branch, outside pub workspace, no prod deps
touched). Validates the open Block 2 decisions before committing a funded round:
1. Deterministic one-way secp256k1 (Nostr) key derived from the Ğ1 root seed
via HKDF — user still backs up ONE thing. Reproducible + signs (BIP340).
2. OfferTransport abstraction over Nostr NIP-99 (kind 30402); Offer stays
inventory/location-agnostic, geohash coarsened on the wire (tested).
3. publish->discover-by-geohash proven end-to-end against an in-process
hermetic mini-relay (~34ms round-trip).
Findings + risks + recommendation in docs/design/spike-block2-findings.md.
Block 1 suite untouched. 14 tests green, analyzer clean.
- Android release signing reads a gitignored key.properties, falling back
to debug keys so contributors can still build --release.
- CHANGELOG.md (human, newest-first) + Fastlane/F-Droid metadata (en/es)
with descriptions derived from docs/intro.md.
- docs/release.md: the by-hand build + manual smoke checklist (no CI yet).
- inventory_scale_test guards the list query against N+1 at 3000 varieties.
