agent: lumotia — Phase A.7 fix startup-order race that silently orphaned legacy data

Critical bug surfaced by the dogfood drill: every upgrading Magnotia user
would silently keep a fresh empty Lumotia install while their Magnotia
data sat orphaned next to it. Drill caught it on the first real run
under sandboxed HOME.

ROOT CAUSE

src-tauri/src/lib.rs::run() previously called the migrations from inside
the Tauri setup hook (post `tauri::Builder::default()`). But three
sequential actions BEFORE the setup hook had already created the
destination directories:

  1. init_tracing() -> logs_dir() -> create_dir_all(app_data_dir/logs)
     creates the lumotia/ root.
  2. install_panic_hook() -> crashes_dir() -> create_dir_all() ditto.
  3. Tauri's WebKitGTK runtime / plugin chain creates the bundle-id-keyed
     consulting.corbel.lumotia/ dir eagerly when the WebContext spins up
     (mediakeys, storage, WebKitCache subdirs appeared even without our
     hook explicitly creating them).

By the time the setup-hook migrations fired, every legacy candidate
returned `TargetAlreadyExists` (paths.rs) or `BothExistLegacyPreserved`
(tauri_app_data_migration.rs) — both silent no-op codepaths. Legacy
data was left untouched, fresh Lumotia install gained no transcripts,
settings, or window state.

FIX

Migrate BEFORE any other code touches app_data_dir().

src-tauri/src/tauri_app_data_migration.rs:
  - NEW_BUNDLE_ID const ("consulting.corbel.lumotia"). MUST agree with
    tauri.conf.json#identifier; reviewer-enforced invariant.
  - Renamed private `legacy_tauri_app_data_dir_for` -> public
    `tauri_app_data_dir_for(identifier)`. Function is parameterised by
    bundle id; the "legacy" name was misleading after this change.
  - New `current_tauri_app_data_dir()` resolves the NEW bundle path
    from platform env vars (same convention Tauri 2 uses), so the
    pre-runtime migration can address its destination without
    needing an AppHandle.

src-tauri/src/lib.rs:
  - New `migrate_user_data_pre_runtime()` orchestrates the two
    migrations + ambiguity guard. Uses `eprintln!` for surface events
    (tracing not yet initialised at this stage; stderr lands in
    journald / foreground terminal which is the right transport for
    boot-phase output). FATAL errors call process::exit(1) — the
    setup-hook version returned Err from the closure, equivalent
    effect.
  - run() now calls migrate_user_data_pre_runtime() as its first line,
    BEFORE init_tracing(), install_panic_hook(), and the Tauri
    builder.
  - Setup-hook migration blocks deleted (~90 lines). Setup hook now
    starts with a one-line comment pointing at the pre-runtime fn.

VERIFICATION

Re-ran the dogfood drill (scripts/dogfood-rebrand-drill.sh) — 8/8 probes
pass after the fix (was 4/8). Both stderr lines fire:

  [lumotia-startup] migrated legacy magnotia data dir to lumotia:
      .../magnotia -> .../lumotia (renamed_db=true, elapsed_ms=0)
  [lumotia-startup] migrated Tauri app_data_dir from legacy bundle
      identifier: .../uk.co.corbel.magnotia ->
      .../consulting.corbel.lumotia (elapsed_ms=0)

On-disk post-state confirms: magnotia/ gone, lumotia/ has migrated db
+ recordings, uk.co.corbel.magnotia/ preserved as backup,
consulting.corbel.lumotia/localStorage/leveldb/ has migrated data.

- cargo fmt --check: clean
- cargo clippy --workspace --all-targets -- -D warnings: clean
- cargo test --workspace: 409/0 (no regression)
This commit is contained in:
2026-05-14 13:59:08 +01:00
parent 2aac366f32
commit ff8dda06d0
2 changed files with 168 additions and 116 deletions

View File

@@ -298,8 +298,136 @@ fn init_tracing() {
}); });
} }
/// One-shot data migration that runs BEFORE anything else in `run()`.
///
/// CRITICAL ORDERING: this must come before `init_tracing`,
/// `install_panic_hook`, AND `tauri::Builder::default()`. Each of those
/// either calls `create_dir_all` on a child of `app_data_dir()` or causes
/// Tauri/WebKitGTK to create its own bundle-identifier-keyed dir for
/// plugin state. Either way, by the time the migration would otherwise
/// run from inside the Tauri setup hook, the destination already exists
/// and the migration short-circuits via `TargetAlreadyExists` /
/// `BothExistLegacyPreserved` — silently leaving the user's legacy
/// Magnotia data orphaned next to a fresh empty Lumotia install. The
/// `scripts/dogfood-rebrand-drill.sh` integration probe is what surfaced
/// this race; see commit history for the original buggy ordering.
///
/// Tracing isn't initialised yet, so migration outcomes are written to
/// stderr via `eprintln!`. The format mirrors the structured fields the
/// setup-hook tracing layer would have emitted — same content, different
/// transport. systemd-journald + a foreground terminal both capture
/// stderr, which is the audit surface that matters at boot.
///
/// Fatal failures (data-dir migration error, ambiguous lumotia paths on
/// disk) call `process::exit(1)` rather than panicking. We refuse to
/// start with the wrong path resolved — silently continuing would
/// orphan user data, which is the failure mode this fix is closing.
fn migrate_user_data_pre_runtime() {
// 1. Hand-rolled data-dir migration: ~/.local/share/magnotia (and
// macOS / Windows equivalents) -> ~/.local/share/lumotia. Drives
// every legacy candidate independently so multi-legacy Linux
// users (`~/.magnotia` AND `~/.local/share/magnotia` from
// different historical builds) get all of them migrated, not
// just the first one probed.
let t = std::time::Instant::now();
match migrate_legacy_data_dir() {
Ok(statuses) => {
for status in &statuses {
match status {
MigrationStatus::Migrated {
from,
to,
renamed_db,
} => {
eprintln!(
"[lumotia-startup] migrated legacy magnotia data dir to lumotia: \
{from} -> {to} (renamed_db={renamed_db}, elapsed_ms={ms})",
from = from.display(),
to = to.display(),
renamed_db = *renamed_db,
ms = t.elapsed().as_millis(),
);
}
MigrationStatus::TargetAlreadyExists { .. }
| MigrationStatus::NoLegacyFound => {
// Steady state on the second-or-later boot. Chatty
// logging here would dwarf the genuine first-boot
// event, so we stay silent.
}
}
}
}
Err(e) => {
eprintln!(
"[lumotia-startup] FATAL: legacy data dir migration failed — refusing \
to start (would orphan user data): {e}"
);
std::process::exit(1);
}
}
// 2. Tauri app_data_dir migration: ~/.local/share/uk.co.corbel.magnotia
// -> ~/.local/share/consulting.corbel.lumotia. Copy-via-staging so
// a half-written destination cannot appear on disk. Legacy is
// preserved as a backup.
use crate::tauri_app_data_migration::{
current_tauri_app_data_dir, legacy_tauri_app_data_dir,
migrate_tauri_app_data_dir_with_paths, AppDataMigrationStatus,
};
let t = std::time::Instant::now();
if let (Some(legacy), Some(current)) =
(legacy_tauri_app_data_dir(), current_tauri_app_data_dir())
{
match migrate_tauri_app_data_dir_with_paths(&legacy, &current) {
AppDataMigrationStatus::Migrated { old, new } => {
eprintln!(
"[lumotia-startup] migrated Tauri app_data_dir from legacy bundle \
identifier: {old} -> {new} (elapsed_ms={ms})",
old = old.display(),
new = new.display(),
ms = t.elapsed().as_millis(),
);
}
AppDataMigrationStatus::BothExistLegacyPreserved { old, new } => {
// Reachable on the second-or-later boot OR if the user
// ran a side-by-side install. Either is benign — we
// preserve legacy, use new — so log at INFO not WARN.
eprintln!(
"[lumotia-startup] Tauri app_data_dir present at new path; legacy \
preserved: old={old} new={new}",
old = old.display(),
new = new.display(),
);
}
AppDataMigrationStatus::NoLegacyFound => {}
}
}
// 3. Ambiguity guard. After migrations, if more than one lumotia
// target candidate exists on disk (`~/.lumotia` AND
// `~/.local/share/lumotia` from split runs), refuse to start
// rather than silently pick one. The user must consolidate
// manually.
if let Err(amb) = check_target_ambiguity() {
eprintln!(
"[lumotia-startup] FATAL: ambiguous lumotia data directory — refusing to \
start: {amb}"
);
std::process::exit(1);
}
}
#[cfg_attr(mobile, tauri::mobile_entry_point)] #[cfg_attr(mobile, tauri::mobile_entry_point)]
pub fn run() { pub fn run() {
// Step 1: migrate legacy magnotia data BEFORE anything else touches
// the lumotia data dir. init_tracing, install_panic_hook, and
// tauri::Builder::default() all lazily create directories under
// app_data_dir on first use; if any of those run before migration,
// every migrate_one() probe returns TargetAlreadyExists / both-exist
// and the legacy data is silently orphaned.
migrate_user_data_pre_runtime();
// Step 2: structured logging on the (now correctly migrated) logs dir.
init_tracing(); init_tracing();
#[cfg(target_os = "linux")] #[cfg(target_os = "linux")]
@@ -345,117 +473,14 @@ pub fn run() {
builder builder
.setup(|app| { .setup(|app| {
// Tauri 2 keys both `app_data_dir` and the webview's data // Data migrations + ambiguity guard ran in
// store (localStorage, IndexedDB, cookies, service worker // `migrate_user_data_pre_runtime()` BEFORE tauri::Builder
// storage, cache) plus all Tauri plugin state files // was constructed — see the doc comment on that function
// (window-state geometry, autostart enable flag) by the // for why setup-hook timing is too late. By the time we get
// bundle identifier. Commit `14313cf` renamed the bundle // here every app_data_dir path is the post-rebrand one and
// from `uk.co.corbel.magnotia` to `consulting.corbel.lumotia`, // any legacy magnotia state has either been migrated or
// which would silently orphan every existing user's // preserved as a backup.
// webview state under the new identifier. Migrate the let _ = app;
// legacy identifier-keyed app_data_dir first thing in
// setup, before the webview loads its first page and
// touches the on-disk store. Failure is logged but does
// NOT block startup: the hand-rolled `paths.rs` migration
// below covers the user's transcripts and models, so a
// total failure here only loses webview-keyed state.
//
// This is a separate concern from the
// `~/.local/share/magnotia` → `~/.local/share/lumotia`
// migration just below — that one lives under a
// hand-rolled path the app picks itself; this one lives
// under Tauri's identifier-keyed convention.
let t_app_data = Instant::now();
match app.path().app_data_dir() {
Ok(new_app_data) => {
use crate::tauri_app_data_migration::{
legacy_tauri_app_data_dir, migrate_tauri_app_data_dir_with_paths,
AppDataMigrationStatus,
};
if let Some(legacy) = legacy_tauri_app_data_dir() {
match migrate_tauri_app_data_dir_with_paths(&legacy, &new_app_data) {
AppDataMigrationStatus::Migrated { old, new } => {
tracing::info!(
target: "lumotia_startup",
elapsed_ms = t_app_data.elapsed().as_millis(),
old = %old.display(),
new = %new.display(),
"migrated Tauri app_data_dir from legacy bundle identifier; legacy preserved as backup"
);
}
AppDataMigrationStatus::BothExistLegacyPreserved { old, new } => {
tracing::warn!(
target: "lumotia_startup",
old = %old.display(),
new = %new.display(),
"both legacy and new Tauri app_data_dir exist; using new, legacy preserved"
);
}
AppDataMigrationStatus::NoLegacyFound => {}
}
}
}
Err(e) => {
tracing::warn!(
target: "lumotia_startup",
error = %e,
"could not resolve Tauri app_data_dir; skipping bundle-identifier migration"
);
}
}
// One-shot legacy data-dir migration: rename ~/.local/share/magnotia
// (and macOS/Windows equivalents) to the convention-preserving
// lumotia path on first launch after the rebrand. Idempotent —
// safe to call on every boot. A migration error is fatal: silently
// continuing past it would orphan the user's transcripts and
// settings behind a fresh empty lumotia dir.
let t_migrate = Instant::now();
match migrate_legacy_data_dir() {
Ok(statuses) => {
// Drive every legacy candidate independently: on Linux a
// user may have both `~/.magnotia` and
// `~/.local/share/magnotia`, and migrating only one
// would orphan the other forever.
for status in &statuses {
match status {
MigrationStatus::Migrated { from, to, renamed_db } => {
tracing::info!(
target: "lumotia_startup",
elapsed_ms = t_migrate.elapsed().as_millis(),
from = %from.display(),
to = %to.display(),
renamed_db = *renamed_db,
"migrated legacy magnotia data dir to lumotia"
);
}
MigrationStatus::TargetAlreadyExists { .. } => {}
MigrationStatus::NoLegacyFound => {}
}
}
}
Err(e) => {
tracing::error!(
target: "lumotia_startup",
error = %e,
"legacy data dir migration failed — refusing to start (would orphan user data)"
);
return Err(Box::new(e) as Box<dyn std::error::Error>);
}
}
// After migration, refuse to start if more than one lumotia
// target candidate exists on disk (e.g. both `~/.lumotia` AND
// `~/.local/share/lumotia`). Silently picking one would point
// the app at the wrong half of a split data directory.
if let Err(amb) = check_target_ambiguity() {
tracing::error!(
target: "lumotia_startup",
error = %amb,
"ambiguous lumotia data directory — refusing to start"
);
return Err(Box::new(amb) as Box<dyn std::error::Error>);
}
// Initialise database and startup settings in one runtime entry. // Initialise database and startup settings in one runtime entry.
let db_path = database_path(); let db_path = database_path();

View File

@@ -30,6 +30,14 @@ use lumotia_core::paths::copy_dir_recursive;
/// (commit `14313cf`). /// (commit `14313cf`).
pub const OLD_BUNDLE_ID: &str = "uk.co.corbel.magnotia"; pub const OLD_BUNDLE_ID: &str = "uk.co.corbel.magnotia";
/// Bundle identifier in use post-rebrand. MUST match the `identifier`
/// field in `src-tauri/tauri.conf.json` — if those two ever drift,
/// the pre-runtime migration would copy data into a path Tauri never
/// looks at. There is no compile-time check for this invariant;
/// reviewer's job to catch a tauri.conf.json edit that doesn't update
/// this const.
pub const NEW_BUNDLE_ID: &str = "consulting.corbel.lumotia";
/// Outcome of an `app_data_dir` migration attempt. Mostly informational /// Outcome of an `app_data_dir` migration attempt. Mostly informational
/// for the setup-hook tracing layer; the function never aborts startup. /// for the setup-hook tracing layer; the function never aborts startup.
#[derive(Debug, Clone, PartialEq, Eq)] #[derive(Debug, Clone, PartialEq, Eq)]
@@ -47,9 +55,10 @@ pub enum AppDataMigrationStatus {
Migrated { old: PathBuf, new: PathBuf }, Migrated { old: PathBuf, new: PathBuf },
} }
/// Resolve the OLD Tauri `app_data_dir` from platform conventions. The /// Resolve the OLD Tauri `app_data_dir` from platform conventions.
/// AppHandle no longer knows the legacy identifier, so this is /// Used by the pre-runtime migration; the AppHandle isn't available
/// hand-rolled per platform. /// yet at that stage AND the AppHandle is keyed by the NEW identifier
/// anyway, so it couldn't surface the legacy path even if we had it.
/// ///
/// Returns `None` when the environment variable required to anchor the /// Returns `None` when the environment variable required to anchor the
/// path is missing or empty — `HOME` on Unix, `APPDATA` on Windows. In /// path is missing or empty — `HOME` on Unix, `APPDATA` on Windows. In
@@ -57,10 +66,28 @@ pub enum AppDataMigrationStatus {
/// fallback (the new path also depends on the same env vars and would /// fallback (the new path also depends on the same env vars and would
/// be equally unrooted). /// be equally unrooted).
pub fn legacy_tauri_app_data_dir() -> Option<PathBuf> { pub fn legacy_tauri_app_data_dir() -> Option<PathBuf> {
legacy_tauri_app_data_dir_for(OLD_BUNDLE_ID) tauri_app_data_dir_for(OLD_BUNDLE_ID)
} }
fn legacy_tauri_app_data_dir_for(identifier: &str) -> Option<PathBuf> { /// Resolve the CURRENT (post-rebrand) Tauri `app_data_dir` from
/// platform conventions. Used by the pre-runtime migration as the
/// destination for the copy — pre-runtime means we run before
/// `tauri::Builder::default()` so we can't call `app.path().app_data_dir()`.
///
/// CRITICAL: this resolver MUST agree with Tauri 2's own resolution at
/// runtime, otherwise the migration copies data into a path Tauri never
/// reads. Both sides use the same XDG / Library / APPDATA conventions
/// keyed by the bundle identifier, so they agree by construction as
/// long as `NEW_BUNDLE_ID` matches `tauri.conf.json#identifier`.
pub fn current_tauri_app_data_dir() -> Option<PathBuf> {
tauri_app_data_dir_for(NEW_BUNDLE_ID)
}
/// Platform-aware resolver shared by [`legacy_tauri_app_data_dir`] and
/// [`current_tauri_app_data_dir`]. Public so integration tests in
/// sibling crates can substitute an arbitrary identifier — production
/// callers should prefer the two wrappers above.
pub fn tauri_app_data_dir_for(identifier: &str) -> Option<PathBuf> {
// Exactly one of the four cfg blocks below is present per target // Exactly one of the four cfg blocks below is present per target
// compile. Each is a tail expression that becomes the function's // compile. Each is a tail expression that becomes the function's
// return value. Avoiding explicit `return` keeps clippy's // return value. Avoiding explicit `return` keeps clippy's