Also two new §3 traps paid for during the gate re-run: daemon-surviving
desktop JVMs after a client kill (use pkill -9 before re-running against
the same data dir), and never parallelizing two --data-dir runs (leg B
against leg A's half-written library produces bogus FAILs).
Closes milestone M3 (DESKTOP_PHASE_PLAN §5).
Code (one addition): --key-export / --key-import CLI legs in Main.kt —
the SecurityActions menu's exact codec calls, headless. This makes the
pending physical-phone leg a one-liner once a Waydroid container is
live (recipe recorded in the feasibility doc).
Docs: DESKTOP_FEASIBILITY.md gains the M3.1–M3.4 status section with
the actual run output (leg A/B, KEYENV logs, 0600 ls lines); plan §1
lists the two crypto legs among the must-stay-green commands; §5 marked
DONE; first-actions now point a fresh session at M4 task 1.
Exit criteria, per the plan's wording:
- encrypted desktop library passes EncryptionVerifier's trial-decrypt
sweep: MET (leg A verified=1 failed=0, wrong key -> AEAD failure
count, leg B relaunch all-ok).
- phone-exported recovery key decrypts on desktop: MET at format level
(shared codec source, 84B INCKEY1 envelope, AEADBadTagException on
wrong passphrase, byte-identical key into a fresh 0600 file); the
physical phone export is FLAGGED pending (Waydroid container FROZEN,
start is sudo-gated) with the exact re-run recipe — not assumed done.
Verified: ./gradlew assembleDebug, :core:data-jvm:smoke,
:spike:data-jvm:run, :desktopApp:run --self-test, and both crypto legs
— all green in one final pass (one self-test teardown hang under the
compositor, rc=124, clean on immediate rerun — recorded, environmental).
Manual Waydroid pass (phone exports encrypted -> re-imports the same
file through the real UI) caught it: the confirm dialog closed but no
import ran and no snackbar appeared. PassphraseDialog.onConfirm's
shared preamble nulls all URI slots before routing on mode, and the
IMPORT branch never re-armed pendingImportUri -- so the preview
confirm dialog's Import button read null and skipped viewModel.import
entirely. Plain-JSON imports bypass the passphrase dialog, which is
why every scripted desktop leg (M2.1-2.5) stayed green.
Fix: re-arm pendingImportUri in the IMPORT branch before preview;
wrong-pw/dismiss paths still null it and zero the retained chars.
Verified on-device after rebuild + `waydroid app install`: export
snackbar, magic-sniffed "Decrypt backup" sheet, 4-note preview, and
the dedupe snackbar "Imported 0 note(s), skipped 4 already imported."
Also lands the Waydroid test rig used for the pass (tools/WAYDROID_
TESTING.md + tools/inject2.c: uinput tap/move/key/text/swipe injector)
and the full M2.4b write-up in DESKTOP_FEASIBILITY.md. M2 exit
criteria met: bidirectional lossless round-trip plain+encrypted.
Phase 1 M2 task 5 (tools/DESKTOP_PHASE_PLAN.md §4): measure & record,
no redesign. New :core:data-jvm:backupMemory probe (2g heap) samples
heap-used at 1 ms during BackupBridge.exportPlain — the phone's own
export path — across four fresh libraries:
24 MB data (3 att) → peak 230 MB | 2 MB text x2000 notes → 140 MB
80 MB data (10 att) → peak 777 MB | 320 MB data (40 att) → OOM at -Xmx2g
Attachments drive memory (~9–10x allocation pressure: store load →
base64 String → UTF-8 bytes → JsonElement tree); note count alone is
trivial. The plan's 250 MB gate trips at ~25–30 MB of attachments, and
the phone (no largeHeap) shares the ceiling — a pre-existing Android
limitation, unchanged by the bridge extraction. The real library
exports at 366 KB ≈ 1000x headroom. Bonus: a ~1.6 GB library dies in
objectbox-java's DbFullException before the bridge ever sees it.
Verdict: record, don't redesign. Streaming formatVersion=2 stays parked
for Phase 2, required only if attachment libraries approach ~25 MB.
Phase 1 M2 task 4, desktop half (tools/DESKTOP_PHASE_PLAN.md §4). New
--backup-export-enc <path> + --backup-passphrase <pw> flags drive the
exact calls the RequirePassphrase/Passphrase dialogs make, headlessly.
Verified across two launches on fresh libraries:
run A: import plain(27) → reimport skipped=27 → exported-enc=27 magic ok
run B: preview encrypted=true → imported=27 → wrong-pw preview rejected
→ reimport imported=0 skipped=27
Gotcha found (and honoured): EncryptedBackupCodec zeroes the passphrase
CharArray in BOTH encrypt and decrypt — the array is single-use fuel.
The dialogs already comply (fresh toCharArray() per call); the scripted
block re-derives it via newPw() for the same reason.
Device half (4b: phone re-imports a desktop-made envelope + sees the
dedupe snackbar) follows on Waydroid with the M2 wrap-up.
Phase 1 M2 task 3 (tools/DESKTOP_PHASE_PLAN.md §4). Logic lives in
BackupBridge so both platforms get it through the one bridge:
- BackupBridge.import gains skipExisting (default true) and returns
ImportSummary(imported, skipped). A note is skipped when the library
already holds the key (createdAt, updatedAt, sha256(content)) — the
triple the export preserves verbatim, trustworthy only since the M2
import-timestamp fix. Keys are snapshotted BEFORE importing, so two
genuinely identical notes inside one file still both import into a
fresh library; only a second import of the same file is a no-op.
Snapshotted keys include trashed notes. skipExisting=false restores
the old additive behaviour (the documented override).
- Android: BackupManager.import passes the flag and returns the summary
(signature-identical call for SettingsViewModel); its snackbar now
reports the skip count. Screen layouts untouched — exit criterion
"Android backup screens unchanged" holds.
- Desktop: the import confirm dialog swaps its "duplicates persist"
warning for a working "Skip notes that are already here" checkbox
(default ON); scripted path prints the re-import outcome.
- Format unchanged: formatVersion stays 1 (streaming/dedupe-at-format
level stays a Phase-2 formatVersion=2 idea).
Verification passed: :core:data-jvm:backupSmoke (new step — re-import
skipped=27, override added 27, store=54; RESULT M2 round-trip PASS on
the real export) + :desktopApp:run --args="--data-dir /tmp/incido-m2t3
--backup-import <export> --backup-export … --exit-after 8" (imported=27
skipped=0 → exported=27 → reimport imported=0 skipped=27, clean exit) +
assembleDebug + :core:data-jvm:smoke + :desktopApp:run --self-test —
all green; output recorded in DESKTOP_FEASIBILITY.md.
Known limitation (by design): attachments are not in the key.
Phase 1 M2 task 2 (tools/DESKTOP_PHASE_PLAN.md §4): :desktopApp can now
speak the phone's backup format through the extracted bridge.
- New BackupActions.kt: top-bar overflow menu (Import from file… /
Export plain / Export encrypted). Import = AWT FileDialog ->
BackupBridge.preview -> passphrase dialog on INCDBK1 (wrong passphrase
stays open for retry) -> BackupPreview confirm dialog -> import.
Export = save-picker -> exportPlain / passphrase + exportEncrypted.
All heavy bridge calls run on Dispatchers.IO; results via Snackbar.
- Same package-level BackupBridge calls Android's BackupManager shim
makes — format shared by construction, not reimplementation. Android's
PassphraseDialog.kt deliberately NOT ported (M4 moves the real one to
common code).
- :desktopApp gains implementation(:core:backup).
- Scripted headless path for CI: --backup-import/--backup-export drive
the identical bridge calls without dialog clicking.
Verification passed: :desktopApp:run --args="--data-dir /tmp/...
--backup-import <real phone export> --backup-export /tmp/m2t4-exp.json
--exit-after 8" (preview 27 / imported 27 / exported 27, clean exit) +
:core:data-jvm:backupSmoke on the DESKTOP-written export (all PASS) +
python cross-check: notes (createdAt,updatedAt,content) and attachment
Base64 identical phone->desktop->phone + assembleDebug +
:core:data-jvm:smoke + :desktopApp:run --self-test — all green.
Output recorded in DESKTOP_FEASIBILITY.md (incl. one unreproduced
first-run exit hang to watch in M5).
Phase 1 M2 verification probe, driven by a real phone export (27 notes,
1 attachment, 14 orphan tags; the file is user data and stays out of the
repo):
- :core:data-jvm:backupSmoke --args="<export.incido.json> [libDir]" —
preview, import into a fresh JVM library, re-export + field-by-field
compare, encrypted envelope export (wrong passphrase rejected),
encrypted import into a 2nd library with attachment-byte equality.
The probe caught a PRE-EXISTING Android bug the phone also has: upsert /
saveNoteWithHashtags stamped createdAt+updatedAt to now on every insert,
so backup import rewrote every note's history — and would defeat the
(createdAt, updatedAt, sha256(content)) dedupe key M2 task 3 depends on.
Justified behavior change (invariant 1: not silent): non-zero caller
timestamps are now honoured, 0 still means now. The Android editor
passes its own stamps (verified at both Note(...) sites), so only the
import path changes. addAttachment gains createdAt: Long = 0L;
importSnapshot passes the phone value through.
Verification passed: :core:data-jvm:backupSmoke (RESULT: M2 round-trip
PASS on the real export) + assembleDebug + :core:data-jvm:smoke (after a
forced codegen regen: "ID model file unchanged") + :desktopApp:run
--self-test + :spike:data-jvm:run — all green; output recorded in
DESKTOP_FEASIBILITY.md.
Known pre-existing gap left open (documented): importSnapshot never
recreates orphaned hashtag rows.
Phase 1 M2 task 1 (tools/DESKTOP_PHASE_PLAN.md §4): extract the backup
bridge from :composeApp into a plain kotlin("jvm") module so the desktop
build can import/export the phone file format.
- git mv BackupCodec / EncryptedBackupCodec / EncryptedKeyCodec into
core/backup, keeping the org.incido.compose.data.backup package so the
Android diff is file moves + one dependency line (no import churn).
- New BackupBridge: the four BackupManager operations over
InputStream/OutputStream instead of Uri (incl. the v0.3.3.3 preview
hotfix catches); BackupPreview moves with it, same package.
- BackupManager stays in :composeApp as a thin Uri -> stream SAF shim,
public API unchanged. Dedupe = task 3, desktop picker UI = task 2.
Verification passed: ./gradlew :core:backup:build :core:data-jvm:smoke
:spike:data-jvm:run + assembleDebug + :desktopApp:run --args="--data-dir
/tmp/incido-desktop-smoke --self-test --exit-after 6" — all green; run
output recorded in tools/DESKTOP_FEASIBILITY.md (Phase 1 status).
Adds tools/DESKTOP_PHASE_PLAN.md: everything a fresh session needs with no
prior context — current module shape, the exact commands that must stay
green, pinned versions, the invariants (tracked ObjectBox model file, the
objectbox-android / objectbox-java classpath rule, kapt and Compose
Desktop plugin quirks), the traps already paid for, and the M2/M3/M4/M5
milestones with exit criteria.
Every claim is checked against the working tree rather than remembered:
the coupling inventory (per-file android.* import counts), the 8
DataStore keys, AesGcmFieldCipher's single getOrCreateKey() call, the
portability of BackupCodec/EncryptedBackupCodec, and the pinned versions
(Kotlin 2.0.21, AGP 8.7.2, Compose Multiplatform 1.7.3, ObjectBox 4.0.3).
Open questions are marked as such rather than guessed — the JVM EXIF
decoder and OS-keyring libraries are unverified, and the AGP/KMP
friction around the composeApp conversion is called out as the least-known
part of M4, with a fallback shape and a time-boxed decision rule.
Phase 1 milestone M1b: a Compose Multiplatform window driven by the same
NoteRepositoryImpl the Android app uses, over its own ObjectBox library.
DesktopContainer mirrors AppContainer's construction order (store,
CipherHolder, repository). Owns its library at
$XDG_DATA_HOME/incido — a separate library per machine,
since ObjectBox is single-writer; the backup file (M2)
is the only bridge to the phone. --data-dir opens
another library, e.g. one pulled off a phone.
Main.kt two panes, save-as-you-type with a 400ms debounce,
favourite toggle. Material 3 defaults on purpose: the
app's theme and screens arrive with the M4 composeApp
conversion, which deletes this stand-in.
Verified headlessly (--self-test drives the same repository calls the "+"
button and the star make, --exit-after auto-closes), against a library
previously written by :core:data-jvm:smoke — i.e. a different module's
generated MyObjectBox, same schema:
RENDER notes=0 selected=null first frame (collectAsState initial value)
RENDER notes=1 selected=1 ObjectBox flow delivered
SELFTEST wrote note id=2 write path through the UI's own calls
RENDER notes=2 selected=2 and it came back through the flow
reopen -> RENDER notes=2 persisted
Gotcha recorded in the findings doc: the first composition always sees the
empty initial value of collectAsState, so anything asserting on note counts
must watch the flow rather than the first frame.
Still synthetic: theme + screens. Not wired yet: settings, encryption (M3),
backup bridge (M2), packaging (M5). assembleDebug verified green.
Adds the throwaway Phase-0 spike (Development Plan §6.3) that answers
whether Incido can become a Compose Multiplatform desktop app without
abandoning ObjectBox, its crypto layer, or its UI layer. All six
questions came back PASS; findings live in tools/DESKTOP_FEASIBILITY.md.
:spike:data-jvm ObjectBox kapt codegen + native runtime on a plain
JVM, DataStore Preferences without a Context, and
the app's own EncryptedKeyCodec round-tripping.
:spike:compose-jvm Compose Desktop rendering the app's real theme
layer + core:domain, plus jpackage packaging.
Additive only: nothing outside spike/ depends on these, and the Android
build is untouched (assembleDebug verified green).