fix(tray): persist module start/stop to [autostart] modules

[ErikBjare]

Addresses a user report with two halves:

1. Turning watchers off in the tray doesn't stick across restarts.
2. "after removing aw-watcher-window and aw-watcher-afk from the config files, they're still automatically enabled anytime activitywatch starts".

1. Tray toggles now persist (fixed here)

handle_system_click only changed the running state, so a tray toggle lasted for the session and the module came back on the next launch. Now, in both the GUI and mini mode, a tray start or stop adds the module to, or removes it from, [autostart] modules in config.toml (new src-tauri/src/config_edit.rs).

• The file is re-read from disk and edited in place with toml_edit. Comments, layout and hand edits made while the app runs are kept. Comments on and around entries stay exactly once when adding or removing, with or without a trailing comma.
• Both forms the config accepts are edited: inline modules = [...] (names and { name, args }) and [[autostart.modules]] tables.
• A file that doesn't parse is left untouched (the failure is only logged).
• Removing an entry with inline args moves them to [module_args], replacing any existing value there, since the inline args were the ones in effect. A later manual start still uses them.
• Writes are atomic (temp file in the same dir, then rename; permissions kept, symlinks written through). This also applies to the autostart toggle and write_formatted_config. Writes share autostart::PERSIST_LOCK.
• Tray clicks on names that aren't discovered modules are ignored.

aw-notify

aw-notify's autostart is governed by the shared settings.aw-notify.enabled opt-in flag (ActivityWatch/activitywatch#1435: single source of truth, managers start it if and only if it is set). So a Modules-menu click on aw-notify now goes through the same path as the "Enable notifications" toggle: write the flag, then start or stop it. It is never written to [autostart]. At launch aw-notify now follows the flag alone: it is started when the flag is on, even if unlisted, and skipped when it is off, even if listed. Previously it also had to be listed, so the dedicated toggle didn't survive a restart either.

2. Hand-edited removals reverting to defaults (root cause; not changed here)

This isn't caused by the config path or the per-profile path. Default and testing profiles still use the legacy activitywatch/aw-tauri/config.toml root, and the defaults are only written when the file is missing. The cause is the parse-failure fallback in get_config() (src-tauri/src/lib.rs:603-615 on master):

• Any deserialize error makes it silently use UserConfig::default() for the whole session, which on macOS/Windows (and on Linux without aw-awatcher) is ["aw-watcher-afk", "aw-watcher-window"] (lib.rs:535, 568).
• It never writes that back, so the file still looks edited.
• In GUI mode it shows a one-off "Malformed config file" dialog that's easy to miss on a login autostart. In daemon or mini mode it only logs a warning.

AutostartConfig.modules has no #[serde(default)] (lib.rs:482), so the most natural ways to "remove" the watchers trigger that fallback. I checked with toml::from_str::<UserConfig>:

Edit	Result
modules = [\n]	OK, nothing autostarts
modules = [\n "aw-awatcher",\n] (trailing comma)	OK
modules line deleted or commented out	missing field modules, so defaults are used
a comma left behind ([ , ])	parse error, so defaults are used
a missing comma between entries	parse error, so defaults are used
[autostart] section deleted	missing field autostart, so defaults are used

Follow-up fix (separate PR): add #[serde(default)] on AutostartConfig.modules (and on autostart itself) so a missing list means "start nothing". On a real syntax error, don't autostart the default watchers: keep the defaults for port and paths but start no modules, and say so in the tray or notification instead of a one-off dialog.

With this PR, users no longer need to hand-edit the file to disable watchers. Also, the tray never writes over a malformed file.

Tests

• config_edit::tests: one-per-line, single-line and array-of-tables edits; comments before, inside and after entries (with and without a trailing comma); inline args winning over [module_args]; a malformed config left alone; write_atomic (replace, create, no temp leftovers, symlink write-through). All of these use scratch dirs under the system temp dir and never the real config dir.
• manager::tests: autostart_names_follow_notify_opt_in and tray_click_on_unknown_module_is_ignored.
• cargo fmt --check, cargo clippy --all-targets -D warnings and cargo test --lib (60 tests) pass locally.

[ErikBjare]

@greptileai review

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-10T12:45:05.411850Z	c9af666	New commits

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 623b849bbd

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[greptile-apps]

Confidence Score: 5/5

[High risk] Adds file-write library and refactors config persistence.

The PR appears safe to merge; all previous findings are addressed and no new actionable issues remain.

What we checked:

• Config locks do not deadlock: Module clicks take the manager lock before PERSIST_LOCK. The other config-writing path does not take the manager lock, so there is no reverse waiting path.

Summary

Tray module choices now survive restarts in GUI and mini mode. Config edits preserve layout and comments, support both list forms, and retain module arguments.

• aw-notify uses its shared notification flag instead of the autostart list.
• Config writes replace files atomically and follow symlinks.
• The latest change keeps toggles and saves in the same order and adds a duplicate-arguments regression test.
• All five previous findings are addressed. No new actionable issues were found.
• ErikBjare explicitly defers changes to the malformed-config fallback to a separate PR.
• Tests were inspected, not run during this review.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  A[Tray module click] --> B{aw-notify?}
  B -->|Yes| C[Save notification flag]
  C -->|Saved| D[Start or stop aw-notify]
  B -->|No| E[Start or stop discovered module]
  E --> F[Edit autostart modules while holding manager lock]
  F --> G[Write config atomically]

Loading

_{Reviews (4) · Last reviewed commit: "fix(tray): save module toggles under the..." · Reviewed by Greptile}

[TimeToBuildBob]

🤖 AI code review

This PR adds a new config_edit module that edits [autostart] modules in the user's config.toml in place using toml_edit, so tray toggles of modules persist across restarts. It also routes aw-notify clicks through the settings flag rather than the modules list, makes all config writes atomic via a new write_atomic function (temp file + rename with symlink resolution), and refactors handle_system_click to return an Option<bool> so unknown modules are ignored and module changes are persisted. GUI and mini mode now share a single handle_module_menu_click entry point.

Safe to merge — no P0/P1 findings

Confidence 5/5

⚠️ 1 generated file excluded and NOT reviewed: src-tauri/Cargo.lock (+2/-0 lines, 0KB). Machine-generated output is removed from the diff before review — from what the model sees and from the size gate alike — so nothing below says anything about it.

✅ No thread-worthy findings. Advisory notes follow; they are retained without opening review threads.

1 advisory finding (summary-only, not scored)

These P2 guard, heuristic, trade-off, or documentation claims are retained for judgment without opening review threads.

⚠️ P2 medium — src-tauri/src/config_edit.rs

This is a fix(...) PR but no test files are included in the diff. Erik's feedback: 'where is the repro & fixes they are supposed to catch' (gptme#3441), 'that measurement should come with a regression test' (gptme#3446). Add a test that would have caught this bug. (Advisory: Erik merged all such PRs but consistently requested tests.)

Add a test file that reproduces the bug before the fix and passes after it.

_{How this was verified: static preflight: fix-commit + touched-files scan (rule 7)}

Files changed (6) — the diff as I read it

• src-tauri/Cargo.toml — Adds the toml_edit dependency (0.25).
• src-tauri/src/autostart.rs — Makes PERSIST_LOCK pub(crate), updates its doc comment, and switches persist_enabled to use config_edit::write_atomic.
• src-tauri/src/config_edit.rs — New module: write_atomic (atomic replace with symlink resolution), persist_module_autostart/click, and set_module_autostart for inline arrays and arrays of tables, plus extensive tests.
• src-tauri/src/lib.rs — Registers config_edit module, uses write_atomic in write_formatted_config, and routes tray module clicks to manager::handle_module_menu_click.
• src-tauri/src/manager.rs — handle_system_click now returns Option<bool> and rejects unknown modules; adds NOTIFY_MODULE, autostart_names, and handle_module_menu_click; set_notify_enabled drops the unused AppHandle parameter.
• src-tauri/src/mini.rs — Routes mini-mode module menu clicks through manager::handle_module_menu_click instead of calling handle_system_click directly.

Previous review passes

commit	score	findings	engine	when
623b849bbd7b	4/5	1	llm	2026-10-07 10:19 UTC
6a39182251c5	3/5	2	llm	2026-10-07 12:52 UTC
d39dd4b20c87	5/5	0	llm	2026-10-07 13:04 UTC

_{Reviewed d39dd4b20c87 · openrouter/deepseek/deepseek-v4-flash-0731 · llm engine · 591s · about this reviewer}

_{Maintainer commands}

_{@TimeToBuildBob review (own line) — fresh review · @TimeToBuildBob fix — a worker acts on the findings. Once per comment; 👀 = received.}

[ErikBjare]

@greptileai review

@TimeToBuildBob review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: df4b4ba6fb

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[ErikBjare]

@greptileai review

@TimeToBuildBob review

[ErikBjare]

@greptileai review

@TimeToBuildBob review