The changes field was required, so a single record without it failed the
whole response and emptied the updates list. A live instance returns five
update records of which one has no changes key, so this is reachable rather
than theoretical.
This affects every servarr, since Update is shared.
Add ReadarrEvent::GetDiskSpace, mapped to GET /diskspace, and
get_readarr_diskspace, which stores the result on ReadarrData for the
root folder and system views.
Surface it as `managarr readarr list disk-space`, matching the other
three servarrs.
Readarr returns a label field on some records that DiskSpace does not
model; serde drops it, so the CLI prints only the path and the free and
total space, as it does for the other three servarrs.
Add ReadarrEvent::GetStatus, mapped to GET /system/status, and
get_readarr_status in the new readarr_network/system module. It stores
the version and start time on ReadarrData so the system view and the
tick refresh can read them.
Surface it as `managarr readarr get system-status`, matching the
Radarr, Sonarr, and Lidarr spelling of the same call. A `list status`
spelling would have made the same command a different verb on the
fourth servarr, which is a difference the user would have to remember
for no benefit.
Readarr instances answer /system/status with thirty keys; SystemStatus
models the two that are used and serde ignores the rest, as it does for
the other three servarrs.
This is the first ReadarrCommand variant, so ReadarrCliHandler now
reads its app and network fields.
Inhabit ReadarrEvent with its first variant, HealthCheck, mapped to
GET /health, and add get_readarr_healthcheck alongside the equivalent
Lidarr, Radarr, and Sonarr fetchers. The call is internal only and has
no CLI surface, matching the other three servarrs; it exists so that an
unreachable instance surfaces an error banner on tick.
The healthcheck fetcher lives in readarr_network/mod.rs rather than a
submodule, so its tests live in readarr_network_tests.rs next to the
resource() and NetworkEvent conversion tests, mirroring Lidarr.
MockServarrApi::build_for matched NetworkEvent by value with no Radarr
fallback arm. It compiled only while ReadarrEvent was uninhabited and
no value could reach it; the first variant makes the match non
exhaustive, so it now assigns the config to the Readarr server tab.
Promote Route::Readarr from a unit stub to a tuple variant and wire Readarr
through every layer, mirroring the Lidarr tree. Readarr appears as a fourth
servarr tab and its library view renders the Authors table.
Adds ReadarrData and its modals, the tick and context clue modules, the
top-level ReadarrHandler, ReadarrUi with the Authors library view, and an
empty ReadarrCommand behind ReadarrCliHandler. The default port is 8787.
Promoting the route variant compiles cleanly without any of the arms that
give it behaviour, so all six of those sites are wired explicitly here:
App::on_tick, handlers::handle_events, handle_prompt_toggle,
ServarrContextClueProvider, ui(), and verify_config_present_for_cli. Each was
checked by removing the arm and confirming its covering assertion then fails.
ReadarrEvent and ReadarrCommand are both declared empty. They gain variants
alongside the network calls and CLI commands that use them, which keeps each
later change to a single function.
Adding the tab changes the rendered tab bar, so 13 existing snapshots
regenerate. The only delta in each is that single tab bar line.
This scaffold adds 24 dead_code warnings, all inside the Readarr tree and
none outside it. They clear as the handlers, UI, network events, and CLI
commands that consume these items land: block arrays awaiting their accepts
implementations and BlockSelectionState seeding, modal and state fields
awaiting their handlers, the CLI handler fields awaiting an inhabited command
enum, and the top-level handler fields awaiting sub-handlers. The count must
strictly decrease from here, so a later increase is a defect rather than
transient noise. The 12 pre-existing collapsible_match warnings are unrelated
and untouched.
SystemStatus was declared three times, byte for byte identically, in
radarr_models, sonarr_models, and lidarr_models. MetadataProfile and the
Lidarr release download body were each declared in a single servarr module
despite having nothing servarr specific about them: MetadataProfile is an
id and name pair structurally identical to QualityProfile, which already
lives in servarr_models, and a download body of a guid plus an indexer id
is the generic form of that request.
Move all three into servarr_models, preserving that file's alphabetical
ordering, and drop the prefix from LidarrReleaseDownloadBody now that it is
no longer owned by one servarr.
RadarrReleaseDownloadBody and SonarrReleaseDownloadBody deliberately stay
where they are. Radarr's carries a movie id and Sonarr's carries optional
series, episode, and season fields, so the three types are not
interchangeable and only the Lidarr shape is the common one.
Behavior is unchanged. Every remaining edit is an import path update or the
rename at a call site.
- Rename `ui_scroll_tick_count: u64` → `should_text_scroll: bool` across all 33
call sites; `on_ui_scroll_tick` now assigns true/false instead of 0/1
- Remove two now-redundant comments from `test_on_ui_scroll_tick`
- Move `WHITESPACE_RE` static declaration below module imports in network/mod.rs
- Remove `MANAGARR_TICK_RATE_MS` env var and its parsing helpers; `recv()`-based
blocking already brings idle CPU to near-0, and tuning tick rate silently
stretches the poll cadence — keep `DEFAULT_TICK_RATE_MS` const only
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01C3FpbniZme9mJqgFkMVowz
- Revert src/main.rs to draw-at-top-of-loop (original structure) so
auto-scrolling is no longer coupled to event rate. Lock is released
before blocking on recv() so the network task is never starved.
- Replace ticks_until_scroll counter in on_ui_scroll_tick() with an
Instant + Duration (default 100ms). Scroll now advances at a fixed
wall-clock rate (10 chars/sec) regardless of draw rate or tick rate,
which is the correct long-term fix for the animation regression.
- Update tests: assert scroll_interval == 100ms in default/new tests;
rewrite test_on_ui_scroll_tick to cover zero-interval (always fires)
and long-interval (suppressed) cases.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01C3FpbniZme9mJqgFkMVowz
- Compile whitespace regex once via OnceLock instead of on every
non-success response; makes format_error_body infallible (String
return, no Result) eliminating the only practical failure mode
- Revert default tick rate from 100ms back to 50ms to preserve the
existing 20-second Servarr poll cadence (tick_until_poll=400 × 50ms);
the recv() change is the real CPU fix — the tick rate change was
unnecessary and silently halved the data refresh rate
- Keep MANAGARR_TICK_RATE_MS as opt-in override; note that tuning it
proportionally stretches the poll interval alongside the tick rate
- Replace non-blocking try_recv() event loop with blocking recv() to
eliminate idle spin; the sender tick timeout still drives minimum
wakeup rate so the UI stays responsive
- Lower default tick rate from 50 ms to 100 ms (10 ticks/sec vs 20);
add MANAGARR_TICK_RATE_MS env var for runtime tuning
- Move terminal draws to event boundaries (key + tick) instead of every
loop pass, reducing unnecessary redraws
- In network error handling, prefer concise JSON fields (message /
errorMessage / error) over the raw HTTP body, so Radarr ASP.NET
stack traces don't flood the TUI banner; full body still logged
- Add unit tests for tick-rate parsing and Radarr-style JSON error body
Observed: ~0.6% total CPU at idle vs measurably higher with prod defaults
Dark-Alex-17
