The themed Button forwards its ref straight to SButton, a class
component, so the ref resolves to the class instance instead of the DOM
node. Semantic UI's Modal/Popup Portal calls node.contains(...) on the
trigger's ref during handleDocumentClick, which throws on a non-DOM
value. This crashed the Control page whenever a Service Logs (or QR /
download-error) modal was open and the document was clicked.
Wrap SButton in Ref (findDOMNode) so a forwarded ref always resolves to
the underlying <button> element, fixing every Modal/Popup that uses a
themed Button as its trigger.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The dashboard nudge is useless on Safari and other browsers that cannot
run the WROLPi extension. Share detectBrowser and gate the banner with
isExtensionSupportedBrowser.
Drop .pgpass handling, Postgres ownership repairs, and the docker db
container viewability rules. Update comments and pi-gen map docs for
SQLite/PMTiles. Help submodule documents the new database layout.
Catch useForm fetcher errors and surface them so DomainEditPage can show a not-found message instead of hanging on the loader (and so the webpack overlay no longer flakes Cypress).
- Require `line === 'auto'` in the auto-caption signature so a cue that
is deliberately placed (e.g. `line:0` for top-edge) is left alone even
if it is also `align:start position:0%`. YouTube auto-captions never
set `line`.
- For lazily-loaded language tracks, only detach the cuechange listener
after the cue list is actually populated, in case a cuechange fires
before cues load.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
YouTube auto-generated captions (downloaded via yt-dlp) stamp every cue
with `align:start position:0%`, which pins the caption text to the
bottom-left of the player and makes it hard to read. Properly-authored
VTT files have no cue settings and already render centered.
Detect that auto-caption signature and reset only those cues to the
browser's centered default (align:center / position:auto / line:auto).
This mutates the in-memory VTTCue objects the browser parsed; the .vtt
file on disk is never modified, and deliberately-positioned cues are
left untouched.
Wired via each <track>'s onLoad. Non-default language tracks are
`disabled` at load (no cues yet), so they are normalized lazily on the
first cuechange when the user selects them.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The React /map page rendered blank (only pins) on fresh builds while the
standalone :8084 viewer worked. maplibre-gl's Web Worker (inlined into the
production main.js) threw `ReferenceError: a is not defined` in its setLayers
handler, so the worker's style-layer index never built and every vector
source's tiles failed to parse ("Cannot read properties of undefined
(reading '<source>')"). The unminified dev build was unaffected — it is a
production minification bug: Terser mangles the inlined worker into broken code.
Root cause: the "npm audit fix" in 62660529 (intended for shell-quote/
form-data/ws) regenerated the lockfile and dragged @babel/core 7.20.12 ->
7.29.7 (whole @babel core pipeline). Babel 7.29 transpiles the maplibre worker
differently; CRA's (unchanged) Terser mangle pass then breaks it. Proven on
10.0.0.8: rebuilding the minified prod bundle with @babel 7.20 renders the map
(0 worker errors); rebuilding back to 7.29 breaks it again with the identical
bundle hash. The transform plugins were never bumped, so pinning @babel/core
alone is insufficient — the core AST pipeline (traverse/generator/types/
helpers) must be pinned too.
Fix: pin the 16 @babel packages that differ from the last-good tree
(6b7bd079) via package.json overrides, keeping the shell-quote/form-data/ws
security fixes. Frontend jest suite passes (810/810).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Verify ARRL wire-antenna formulas (936/468/234/585), fix feet/inches
rollover ("N ft 12 in"), make wavelength edits update lengths, and cover
the math with 50 unit tests.
Applies non-breaking transitive lockfile bumps for three still-present
advisories on the app dependency tree:
* shell-quote 1.8.0 -> 1.10.0 (critical: unescaped newlines in quote())
* form-data 3.0.4 -> 3.0.5 / 4.0.4 -> 4.0.6 (high: CRLF injection via
unescaped multipart field names/filenames)
* ws 7.5.10 -> 7.5.12 / 8.18.3 -> 8.21.1 (high: memory-exhaustion DoS)
package-lock.json only; no package.json / direct-dependency changes. Runs
`npm audit fix` without --force, so no breaking dev-chain (react-scripts)
upgrades are pulled in. Verified on trixie/Node 20: jest 760/760 and the
production build both green.
Supersedes the stale dependabot PRs #487 (ws), #486 (form-data), and
equal-or-newer patched versions.
Co-authored-by: Roland <roland.henrie@o2x.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Clearing the temperature input stored parseFloat('') = NaN, rendering
every Kelvin cell as NaN and persisting the invalid value in localStorage
across reloads. Clearing now falls back to room temperature, non-numeric
input is ignored, and a bad stored value decodes to the default.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Semantic UI paints a definition table's empty corner cell white to fake a
"cut-out" corner against a white page. Our dark theme has a dark page, so
the corner rendered as a white square. Override the background and border
box-shadows to transparent for inverted tables only.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Shows the expected millivolt drop across a single wire at 1A, 5A, and 10A
so a wire can be verified with a known current and sense leads. Shares the
length, units, and wire type with the Power Loss calculator (state lifted
to the parent, same localStorage keys). Length is not doubled because the
sense leads measure across one conductor.
Includes a wire temperature input (default room temperature, °F/°C follows
the SAE toggle, stored canonically in Celsius) which corrects resistance
using the material's temperature coefficient.
Adds Solid/Stranded Aluminum wire types with published resistance tables
for both unit systems, and aluminum's 0.403%/°C coefficient.
Fixes erroneous copper resistance values: solid 6 AWG was 0.3134 instead
of the published 0.3951 Ohm/kft, and stranded 0/2/4 AWG were decimal typos
(0.022/0.019/0.038 instead of 0.122/0.194/0.308 - the table is solid x
1.24 throughout). New tests enforce monotonicity and ratio consistency.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
yt-dlp 2026.7.4 (via upstream PR #16965) again extracts view_count from
YouTube's lockupViewModel channel/playlist listings, so the 'views'
download order deprecated in fcb64fed can come back — verified live
against a real channel listing.
Unlike the old implementation, the sort degrades gracefully:
- Entries missing view_count sort last instead of raising KeyError.
- If NO entry has a view_count (yt-dlp older than ~2026.07), the
download fails cleanly with UnrecoverableDownloadError telling the
user to update yt-dlp or switch to Newest/Oldest — no retry loop.
Listing counts are approximate ("3K" -> 3000), which is fine for
ordering. No requirements.txt change needed; the >=2026.03.17 range
picks up 2026.7.4 automatically.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The Disks page treated /media/wrolpi as fully protected, so a
manually-mounted primary drive could not be persisted from the UI — and
the Persist toggle would have been wrong anyway: it wrote fstab.yaml,
which the wrolpi-mounts reconciler refuses to apply for the primary
mount (RESERVED_MOUNT_POINTS). Route Persist for the primary mount to
the host's /etc/fstab (as onboarding does), report its entry in the
fstab list endpoint, and enable the toggle in both UIs while keeping
Unmount blocked. Mounts outside /media (e.g. /boot/efi, which showed
an Unmount button) are now entirely untouchable System mounts.
Also:
- Run `ntfsfix -d` before mounting ntfs/ntfs3 volumes so the Windows
dirty flag (Fast Startup, unclean unplug) cannot block mounting;
non-fatal and a no-op on clean volumes.
- Add uid/gid fstab options for ntfs/ntfs3, matching mount_drive().
- Log reconcile diagnostics (desired/already-mounted/skipped with
reasons) so an idle wrolpi-mounts run explains itself in the journal.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Semantic UI may activate a Dropdown.Item from the keyboard without
following its href, which would save the slicer preference without
launching the slicer. Prevent the default anchor navigation and call
window.location.assign() so every activation path behaves the same.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Adds a split button to the STL and 3MF preview toolbars that opens the
model directly in a desktop slicer via its URL scheme
(scheme://open?file=<absolute media URL>). The slicer downloads the
file itself over the network.
Offers Orca Slicer, Bambu Studio, and Cura; the last-used slicer is
remembered in localStorage and becomes the split button's default,
since browsers cannot detect which protocol handlers are installed.
PrusaSlicer is deliberately omitted: it only downloads from domains
whitelisted by Prusa (printables.com, thingiverse.com, cults3d.com),
so it cannot fetch from a WROLPi.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* POST /api/hotspot/settings checks is_docker_mode() directly instead of
string-matching the error message (still 500, matching the sibling
hotspot/bluetooth/throttle endpoints).
* Escape \ ; , " in the WiFi QR code per the WPA QR format.
* Skip the controller.yaml write when a settings update changes nothing.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Hotspot was reported as "not supported" on x86 machines (Ubuntu/Debian PCs)
because the Controller only watched the hardcoded default wlan0 device, while
predictable interface naming produces names like wlp2s0. The Settings page
saved hotspot_device to wrolpi.yaml, which the Controller never read.
The Controller now owns hotspot settings in controller.yaml:
* New endpoints: GET/POST /api/hotspot/settings, GET /api/hotspot/devices
(WiFi interfaces discovered via nmcli).
* Legacy hotspot_* values in wrolpi.yaml are migrated into controller.yaml
once on startup, unless controller.yaml already sets them.
* The Hotspot SSID/password/device fields moved from the Settings page to the
Control page next to the Hotspot toggle; the device is now a dropdown of
the WiFi devices found on the machine.
* The Controller fallback UI gets the same hotspot settings form for parity.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Add a client-side filter above the File Browser table. Typing reduces
the visible rows to files/directories whose name matches, plus their
parent folders. A folder whose own name matches keeps its entire
subtree, and opening a matching folder shows its contents unfiltered.
* 'f' focuses the filter input, like the search inputs on other pages.
* An attached X button clears the filter; disabled while empty.
* A collapse-all (angle double up) button closes all open folders,
clearing the ?folders= query param and deselecting hidden children.
* The filter row is right-aligned and double-width on desktop, and
fills a single row on mobile.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The DeepSearchHint only appeared on zero results, so a fast search with
matches gave no indication that captions/contents were not searched.
Module searches have no deep estimate, so show a generic reminder
whenever a fast search is active: "More results may be available in
deep search." with results, "Not finding what you need?" without.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The file_group.textsearch tsvector indexes all four text columns;
d_text (captions, document text, zip contents) is ~90% of the data,
making ts_rank searches take seconds on large libraries. Add a
textsearch_abc generated column (title/author/description only, ~8x
smaller) and default all searches to it, with deep=true opting into the
full column.
Database:
- Migration adds the generated column and a GIN index (rewrites the
table; a few minutes on large libraries).
Backend:
- FilesSearchRequest, VideoSearchRequest, ArchiveSearchRequest, and
DocSearchRequest accept `deep`; each search function picks the column.
- /api/search_file_estimates returns both counts in one index-bound
query (COUNT FILTER on the abc subset): `file_groups` (fast) and
`file_groups_deep`.
UI (one coherent story across the count surfaces):
- The typing dropdown always suggests the deep count (what is findable).
- The tab badge shows the active mode's count so it matches the result
list.
- The gap is reconciled by a hint offering deep search: count-aware on
the global search ("No results in fast search. N results available in
deep search." / "N more results available..."), generic on the module
pages (no deep estimate there). The hint resets pagination.
- "Search Depth" toggle in the search filter modals, persisted as
?deep=true.
Tests that relied on d_text matches now pass deep=True, with fast-mode
assertions alongside. Also skip flaky test_download_channel on
CircleCI.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
A failed "Filter files by detecting device" only surfaced the download-mode
reminder down in the Connect segment, off-screen. Extract the message into a
shared renderer and show it both places, so the error appears right where the
user clicked on the "Choose from your WROLPi" tab.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- ESP_CHIP_ID_NAMES disagreed with the backend ESP_CHIP_IDS: it mapped 18 to
ESP32-C5 (18 is ESP32-P4) and omitted 17 entirely, so C5 firmware showed as
"chip id 17" and P4 firmware was mislabeled "ESP32-C5". Now matches the
backend (17=ESP32-C5, 18=ESP32-P4), with a test guarding it.
- Removed the stale filesSearch import/mock/beforeEach from the test; the
flasher no longer calls it.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Prevents the wrong-chip flash that bricks a board (e.g. an ESP32-S2 bootloader
in an ESP32 set): after connecting, each selected file's ESP image header is
checked against the device's chip_id and the offending row is flagged (a
warning, not a block). Changing the firmware selection now disconnects so the
check is always re-run against a freshly detected chip.
- "Choose from your WROLPi" is now a table (File / Path / Size / Chip / Kind
/ add), served by a single /flasher/search call
- search_esp_firmware returns every .bin annotated when unfiltered (non-ESP
parts like partition tables, boot_app0, and littlefs stay pickable); a chip
filter still narrows to that chip's ESP images
- Detect-device filter now searches by the canonical chip name (CHIP_NAME,
e.g. "ESP32") instead of the variant description ("ESP32-D0WD-V3"), which
matched nothing
- Chip column uses deterministic, colorblind-safe colors; same-width chip
names always get distinct colors, and the palette has headroom to grow
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Browser-based ESP32/ESP8266 flasher (esptool-js + Web Serial); nothing is
uploaded to WROLPi, the flash is written straight to the device over USB.
Firmware sources (themed, tabbed): a saved set, the WROLPi media directory,
or a local file. Local files can be dropped anywhere on the page, which jumps
to the "Add from computer" dropzone. The media picker searches .bin files by
path and can detect the connected chip to filter firmware to matching images.
- Indexed file_group.suffix + path filter for fast .bin search (alembic
migration backfills via split_path_stem_and_suffix)
- modules/flasher: chip detection from ESP image headers, /search endpoint,
and flasher.yaml saved configurations (/configs CRUD)
- Flash progress modal with ETA, connect timeout, boot-mode hint, auto-
scrolling logs
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
YouTube's new listing layout means yt-dlp no longer returns `view_count`
in flat channel/playlist entries, so sorting by views raised KeyError and
the download failed and retried forever.
Sorting by views now requires per-video extraction (hundreds of requests
per channel), so the feature is deprecated rather than degraded:
- downloader: the 'views' branch raises UnrecoverableDownloadError so the
download fails cleanly (no retry loop) with a message to switch to
Newest/Oldest. May be revisited via the YouTube Data API.
- schema: download_order validation drops 'views' (newest/oldest/null only).
- UI: removed "Most Views" from the download-order dropdown.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
getSettings() in api.js remaps the /settings response through a fixed field
whitelist, which dropped download_daily_limit_global/per_domain before they
reached the SettingsContext. The values saved correctly but always rendered
empty. Add both fields to the mapping.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- Settings.js: send 0 (not null) when a daily-limit field is cleared, so the
settings API actually clears a previously-saved limit instead of dropping the
null and keeping the old value.
- downloader.py: document that the daily limit is a best-effort soft cap; minor
bounded overshoot is possible when create_downloads' background dispatch overlaps
the perpetual cycle, and self-corrects the next cycle. A strict cross-process cap
is intentionally not implemented.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds two optional DownloadManager settings:
- download_daily_limit_per_domain: max non-recurring downloads per domain per day
- download_daily_limit_global: max non-recurring downloads across all domains per day
Recurring Channel/RSS catalog updates are never gated and never counted; only the
videos/archives they spawn and one-off downloads count. Both successful and failed
attempts consume a slot (counted once per download per day). Limits reset at local
midnight (configured timezone). Over-limit downloads stay "new" and resume the next day.
Implementation:
- New Download.last_download_attempt column (+ index, migration) set at dispatch time so
the count survives restarts and is correct across workers.
- dispatch_downloads gates non-recurring downloads: per-domain skips that domain, global
stops dispatching counted downloads for the cycle. daily_download_counts derives the
day's totals from the DB.
- Domain grouping uses normalized URLs via a new Downloader.normalize_domain method
(base strips www/m/mobile + port); VideoDownloader/ChannelDownloader override it to
collapse youtu.be, /shorts/, and tracking params to youtube.com. ArchiveDownloader is
the seam for future archive URL normalization.
- Settings wired through WROLPiConfig, the /settings API (GET/PATCH, non-negative
validation, 0 = unlimited), and the React Settings page.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The dark-theme rule hardcoded the inverted Accordion's title/content
background to #171616, which is the Modal content color. Inside a Modal
this matched, but on the /admin/status page (and any Segment, which
renders at #1B1C1D) the Accordion showed up as a darker band.
Make the inverted Accordion transparent so it inherits whatever
container it sits in, matching how the light theme already renders
accordions.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* Restore the censored-video search filter
Censored videos (no longer publicly available — detected when a channel
refresh drops them from the channel list, or when comment downloads fail
with "video unavailable/private/removed") are already flagged on
FileGroup.censored, but there was no way to find them since the search
rework dropped the old filter.
Add a "Censored" toggle to the reworked search filter modal, shown only on
the Videos page (showCensored). When enabled the video search returns only
videos with FileGroup.censored = TRUE, tracked via a ?censored=true query
param (useSearchCensored).
* Move Availability filter below the Tags/Results row
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* Surface coarse SMART health for USB-bridge drives
USB enclosures that reject ATA pass-through (e.g. Seagate Expansion) are
invisible to pySMART/DeviceList, so the drive shows no SMART at all on the
Controller page. Fall back to smartctl over the SCSI transport, which many
such bridges still answer, to report a coarse PASS/FAIL health plus drive
identity (model/serial/capacity).
The health (-d scsi -H) and identity (-d scsi -i) reads are issued as
separate calls: combining them makes smartctl suppress the health line when
the identity read reports "SMART support is: Unavailable" (which these
bridges do). The entry is marked smart_limited (labelled "(limited)" in both
UIs) so a green PASS is not mistaken for full monitoring — these drives
expose no attributes or temperature, and some bridges report "OK" statically.
* Harden SCSI health fallback (Greptile review)
- _scsi_health: treat any unrecognised status (e.g. "UNKNOWN", "N/A") as
unreadable (None) instead of a false FAIL. Only "OK" -> PASS and "FAIL*"
-> FAIL; anything else omits the drive rather than showing a red FAIL that
contradicts the smart_limited disclaimer.
- get_all_smart_status: isolate the SCSI supplement loop in its own
try/except so an error there cannot discard the pySMART results already
collected.
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* Consolidate search options into a single Filter modal
Reworks the file-search UI on the Dashboard, Videos, Archives, and Docs
pages. The crowded row of controls (sort, tags, limit, file-type filter,
date selector) is replaced with a single "Filter" button that opens one
comprehensive modal.
- Add SearchFilterModal + SearchFilterButton (violet + count badge when
active). Sections (Sort, File Type, Tags, Results Per Page, Published
Date) are shown per-page via props; layout uses a Semantic Grid.
- Selections are drafted locally and applied in a single updateQuery only
on close/Done, so the search isn't re-run on every field change.
- Results Per Page and Sort use buttons (fewer clicks, no clipped dropdown).
- Extract shared SearchControlBar (view toggle + input + Filter) for the
three results pages; move the global-search view toggle onto the tab row.
- Rework the Dashboard search line into one adaptive flex row; clear button
only shows when there's an active search.
- Remove now-dead FilesView limit/tag returns and SearchLimitDropdown.
* Address review: reset limit on Clear All; use replace over replaceAll
- handleClearAll now resets Results Per Page to the default page size
(DEFAULT_SEARCH_LIMIT = 24) so "Clear All" actually clears every setting.
- Replace start-anchored replaceAll(/^-/g) with replace(/^-/) for the sort key.
- Add a test covering the Clear All limit reset.
* Remove now-unused Media import from Files.js
The FilesSearchView refactor dropped the last <Media> usage in this file,
leaving an unused import (a build warning). Remove it.
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Inventories are a config-only MultiFileConfig (config/inventory/<slug>.yaml) and
were not reachable by the admin config Restore UI, so there was no way to re-import
a hand-edited/copied-in file or restore an inventory from its daily backup.
Backend:
- MultiFileConfig gains generic helpers: get_backup_dates(slug)/_get_backup_file
(surface the daily config/backup/<subdir>/<slug>-DATE.yaml files) and reimport()
(rebuild the in-memory store from disk; picks up hand-edits and copied-in files,
drops files deleted on disk). backup_date is constrained to YYYYMMDD so it cannot
traverse out of the backup directory.
- InventoriesConfig.preview_restore/apply_restore: overwrite replaces the whole
inventory; merge unions items by id (keeps current fields and existing items).
The pre-restore state is backed up automatically, so a restore is undoable.
- API: POST /inventory/reimport, GET /inventory/<slug>/backups,
POST /inventory/<slug>/restore/preview, POST /inventory/<slug>/restore.
Frontend:
- New InventoryImportModal (opened from a Restore button in the Inventory toolbar):
Re-import from disk, plus a dated-backup table with per-row Merge/Overwrite that
previews the change before applying.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A user can type a calculation into any numeric inventory field (number,
quantity magnitude, calories) and have it evaluated in place — e.g. weigh a
400 g box of nails, enter '400 - 20', and Tab/Enter swaps in 380. The result
feeds dependent logic like count-by-weight automatically.
- evaluateExpression() (units.js) reuses the scoped mathjs instance, guarded by
a strict regex so only bare arithmetic (digits, + - * / ( ) and spaces) is
evaluated; anything with a letter, a plain number, or unparseable input is
left untouched. Float noise is rounded off.
- FieldCell evaluates on blur (Tab/click-away) and on Enter: in a numeric field
the first Enter resolves a pending expression, a second submits the row.
- NUMERIC_TYPES is exported once from units.js and shared by the table and cell
editors.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Counting bulk hardware (screws, nails) by hand is slow. This adds a workflow
to estimate the count from weight instead: weigh one item, weigh the lot, and
the count is filled in automatically.
- The Fields modal gains a "Count by Weight" button that adds two quantity
fields, Unit Weight and Total Weight (default grams), and tags the Count
field as computed. An existing `count` field is linked in place rather than
duplicated, so hand-counted items keep their values.
- The link is plain metadata on the Count field
(`compute: {kind: 'count_by_weight', total, unit}`), so it round-trips
through the config/API with no backend change (covered by a backend test).
- During item entry, Count = round(Total Weight / Unit Weight), converting
mismatched units first (g/kg/oz) and refusing to divide when only one side
has a unit. It rounds to a whole number (no 403.7 nails).
- When a weight changes, the count always reflects the new reality: it is set
to the computed value (0 for a sub-unit total) when both weights are present,
or cleared when they can't produce one (a weight blanked, incompatible
units) — so a stale count can never persist against edited weights. A
hand-typed count is only touched when a weight actually changes.
countByWeight lives in units.js; the field preset, config detection, and the
entry-time apply hook live in a new computeFields.js. Covered by unit tests
(countByWeight, computeFields), InventoryTable auto-fill / clear / sub-unit on
draft and inline edit, the FieldSchemaEditor preset, and a backend
field-metadata round-trip.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Clicking a cell in the Items table now focuses that field's input and selects
its contents, so the user can type a replacement value immediately. FieldCell
gains an `autoFocus` prop (focus + select on mount via a merged ref that still
supports the draft row's first-field ref); InventoryTable tracks which cell was
clicked and flags the matching FieldCell.
Also add a FieldSchemaEditor test covering the select-options comma fix.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
