Round 2 (166e9f9e) fixed the stash-key mismatch, but @mkoreen's
2026-06-23 bundle showed BS's MQTT project_file arrived 85 ms past the
2.0 s wait timeout (FTP done 00:42:02.509, "No slicer options cached"
00:42:04.509, MQTT 00:42:04.594). Queue item was committed with
settings defaults; nozzle_mapping never made it onto the wire.
Three pieces:
1. _SLICER_OPTIONS_WAIT_TIMEOUT module constant, 2.0 -> 5.0 s. Covers
wireless / loaded-Pi jitter; one-time +3 s cost only for legacy
slicers that never send MQTT.
2. _RECENT_QUEUE_ITEM_TTL fallback: on_print_command retroactively
UPDATEs slicer-driven fields on a recently-committed queue item
when the event wait already gave up. Tracked via
_recent_queue_items dict (30 s TTL, evicted on every queue-add).
Gated on status='pending' so we never race the dispatcher.
Multi-plate covered via WHERE id IN (...).
3. Post-commit last-chance pop. Audit caught a race in (2): MQTT could
arrive during any await inside _add_to_print_queue (wait_for,
archive_print, db.flush, db.commit), and on_print_command would
stash data with no event consumer AND no _recent_queue_items entry
yet. After populating _recent_queue_items, _add_to_print_queue now
pops _slicer_print_options[file_path.name] one last time and
routes any hit through _restamp inline.
New setting "Auto-add unknown RFID spools" under Settings -> Filament -> Filament Tracking,
default ON for back-compat. When turned off, the backend stops auto-creating an inventory
record for an unknown RFID tag and instead broadcasts an unknown_tag WS event that pops
a global confirmation modal in the Bambuddy UI showing the printer / AMS-X label / slot /
material / colour. Add or Cancel; no nag on every MQTT push.
Backend
- Module-level _unknown_tag_last_broadcast dict dedupes per (printer, slot, tag). Set is
committed AFTER ws_manager.broadcast() returns so a crashed broadcast doesn't poison
the dedup and permanently silence the slot.
- Empty-slot MQTT push clears that slot's entry, so remove+reinsert reliably re-prompts.
- Successful matches via get_spool_by_tag / find_matching_untagged_spool / create_spool
also clear the entry so a future tag swap re-prompts.
- Tray data (tray_type, tray_color, tray_sub_brands, tray_count) shipped in the WS payload
directly so the modal renders the real material / colour instead of relying on the
React Query cache that lags the WS event by several seconds.
- Two new endpoints back the modal's confirm action:
POST /api/v1/inventory/spools/from-slot (INVENTORY_UPDATE)
POST /api/v1/spoolman/spools/from-slot (FILAMENTS_UPDATE)
Both look up the slot's tray data server-side and create + auto-assign atomically.
- Spoolman /from-slot now raises HTTP 500 when the slot-assignment INSERT fails instead
of returning success while the DB rolled back the binding.
- sync_ams_tray gained an optional auto_add_unknown_rfid kwarg (default True so existing
callers are unaffected); auto-sync and both manual sync routes thread the setting.
Frontend
- useUnknownTagPrompt hook listens for the unknown-tag CustomEvent, reads the tray fields
out of the event detail, and feeds a single-modal queue. No long-lived dismissed set;
the backend dedup handles spam suppression.
- UnknownSpoolModal wraps the existing ConfirmModal with a material + colour-swatch
preview block.
- Mounted in Layout.tsx alongside useSponsorPrompt so SpoolBuddy kiosk / login / setup
routes are excluded.
- getAmsLabel moved to utils/amsHelpers.ts; ConfigureAmsSlotModal.tsx and PrintersPage.tsx
both import the shared version (canonical AMS-A / HT-A / External labels).
- AppSettings TS interface gained spoolman_enabled, auto_add_unknown_rfid, spoolman_url
so the runtime cast in the hook is no longer needed.
- SpoolmanSettings.tsx gets a new toggle row in the Filament Tracking card, visible in
both built-in and Spoolman branches; auto-save + toast already wired.
When the printer reports ams_filament_backup=True,
compute_deficit_for_queue_item pools remaining_grams across spools
matching (preset, colour) on the same printer (scoped per extruder on
dual-nozzle) before declaring a per-slot shortfall. Identity is strict:
same slicer_filament preset AND same colour (alpha-normalised). Two
PETG HF spools in different colours are NOT pooled — the firmware would
swap correctly but the print would change colour mid-run. Spoolman side
mirrors the rule via filament.id + color_hex. Backup OFF falls back to
the pre-PR per-slot accounting line-for-line.
8 new test cases in TestFilamentDeficitBackupAware pin pool covers,
pool insufficient, different presets, backup-OFF regression, dual-
extruder side scoping, no-preset never pairs, colour-strict, and
alpha-hex normalisation. The 8 pre-existing test_filament_deficit.py
cases stay green.
feat(printers): AMS Filament Backup modal with BS-style ring per pair
Badge click on the Filaments section header (#1766) now opens a
modal: filament-colour ring per backup pair, material name + rotation
count in the centre, slot labels distributed around the colour band on
contrast-aware pills. Closely modelled on Bambu Studio's Auto Refill
widget. Lone slots are intentionally not listed. R / L badges per ring
when the extruder map carries two distinct values; collapses to no-
badge rendering for single-nozzle printers misflagged as dual.
Esc keypress closes the modal. Theme-aware via CSS variables matching
AMSHistoryModal. computeBackupGroups helper in utils/amsHelpers
defensively dedupes duplicate ams.id entries observed on switch-VP
aggregations.
10 modal render cases pin: Esc closes / unmount nulls the listener /
ring renders for pairs and omits lone slots / R-L badges only when
extruder map has distinct values / empty state / toggle gating.
13 frontend cases pin computeBackupGroups identity rules.
feat(printers): active-print P-N pill on AMS slot tiles during RUNNING
While the printer is mid-print, each AMS slot tile referenced by
status.ams_mapping carries a small "P1 / P2 / P3" pill in the top-
right corner, naming which print-slot is mapped to that AMS slot.
Catches the #1762 comment-2 scenario: a queue job set for "any X1C"
staged to a printer with mismatched filament, no way to verify mid-
print. Same wire data (status.ams_mapping is already on the wire) —
the addition is purely surface.
The existing ring-bambu-green highlight for effectiveTrayNow keeps its
meaning (currently extruding RIGHT NOW); the pill is the per-slot
static assignment for the active print.
chore(scheduler): log Print Anyway short-circuit at INFO
_block_on_filament_deficit logs at INFO when it honours
item.skip_filament_check, so a future "Print Anyway didn't work" report
(third commenter on #1762 hit this shape) has actionable evidence in
the standard support bundle without DEBUG. Bundled because the deficit
fix makes the original symptom disappear for users with backup ON.
On the FINISH-state fallback path bambu_mqtt.py:3258 dispatches
on_finish_photo_moment and on_print_complete back-to-back, so the
moment-producer's RTSP grab and the print-complete consumer's cache
read race — consumer wins the empty pop, then its own RTSP fallback
times out against the producer's in-flight grab (Bambu printers allow
one RTSP client). 394 KB frame captured, notification went text-only.
Add a per-printer asyncio.Event in _stage22_finish_in_flight: producer
registers before first await, sets it in finally on every exit;
consumer awaits with a 20s timeout (15s producer grab + headroom)
before the cache pop. Closes the race AND the concurrent-RTSP timeout
in one change. Timelapse path skips the event, so its branch is
unchanged.
First-attempt fix (d196cfc5) was wrong about the cause. Real root,
traced via @mkoreen's BAMBUDDY_VP_DUMP_WIRE capture + 2026-06-21
support bundle:
mqtt_server.py:1296 was passing the slicer's bare subtask_name
(e.g. "Model_Name") into on_print_command, which stashed under
that key. _add_to_print_queue looked up under file_path.name
(the FTP filename WITH extension, "Model_Name.gcode.3mf"). The
two strings never matched. pop returned None, the 2s wait fired
against a key the stash side never signaled, every captured
slicer field silently fell back to settings defaults.
Affected EVERY Bambu Studio "Send" upload across EVERY model —
not just H2C nozzle_mapping. bed_leveling / flow_cali /
vibration_cali / layer_inspect / timelapse from the original
#1403 capture have been silently ignored since BambuStudio
started splitting subtask_name (bare) from file (with extension).
Unit tests passed because fixtures called on_print_command with
file_path.name directly, bypassing the broken caller.
Fix in manager.py::on_print_command: derive
stash_key = data.get("file") or filename and use it for both
_slicer_print_options and the event lookup. filename
(subtask_name) still flows unchanged to _schedule_finish_release
— push_status echoes it back as gcode_file / subtask_name and
the slicer matches against its own subtask_name there, so
re-routing that path was a separate regression I caught and
reverted mid-audit.
Also: nozzles_info field was a wrong guess in d196cfc5 —
BambuStudio never sends it (confirmed via wire capture). Drop
the capture, dispatch, schema, kwarg, and route paths. DB
column stays nullable so old rows still load; nothing reads
or writes it.
Diagnostic: DEBUG log when _add_to_print_queue finds no slicer
options after the 2s wait, including the looked-up key and the
actual cache keys present. Future stash/lookup mismatches will
be obvious from a log line instead of needing a wire capture.
Behaviour change worth flagging: users on Bambu Studio whose
slicer-side bed-leveling / flow-cali / vibration-cali /
layer-inspect / timelapse differ from Bambuddy's
default-workflow settings will see their slicer choices
honored now instead of silently overridden. Restores #1403's
original intent.
Batches were stuck where they were created. The parent row had no drag
handle and wasn't registered with dnd-kit's SortableContext, so the
only way to move a grouped item was to ungroup, drag, and re-group.
Collapsed batches also acted as unmovable obstacles for adjacent items.
The batch parent now registers under a synthetic "batch-<id>" string
id and carries a GripVertical handle in the header (gated by
queue:reorder). handleDragEnd resolves both endpoints: dragging a
batch moves all its children as one block, dropping onto a batch
anchors at the batch's first child so the group lands immediately
before it. Direction-aware insert uses the first moving id's index
instead of the previously single dragged id, so multi-row drags and
batch drags share the same insert math. Within-batch child reorder
is unchanged. DragOverlay gained a batch ghost showing
"<name> (<N> copies)".
Reporter @thenewguy runs an engineering farm with one AMS per material
(PLA, ASA, Nylon, PVB, HIPS) — Bambuddy's single global ams_humidity_fair
threshold (default 60%) was driving both the queue / ambient auto-drying
trigger AND the hourly humidity alarm uniformly, which is wrong for
multi-material setups where Nylon wants <10% and PLA is fine at 60%.
Drying RUN parameters were already per-filament via drying_presets;
this commit adds the missing per-filament TRIGGER.
New setting ams_humidity_thresholds — JSON map of filament-type to
threshold percent with a "default" key for unknown / unmapped types.
Empty / unset → both consumers fall back to ams_humidity_fair so the
upgrade is silent.
Resolver lives in PrintScheduler.resolve_humidity_threshold(trays,
thresholds, fallback) — picks the lowest (most-restrictive) threshold
across all loaded tray types, matching the conservative-params strategy
_get_conservative_drying_params already uses for temp / hours. Empty
tray slots contribute no constraint; all-empty AMS falls through to the
"default" key. Filament names normalized to uppercase base (so
"PLA Basic" / "pla basic" both map to PLA).
Two consumer sites rewired through the same resolver so the scheduler
and the alarm path can never disagree about whether an AMS is "too
humid":
- print_scheduler.py::_check_auto_drying — per-AMS humidity comparison
for start / stop / skip decisions.
- main.py AMS sensor / alarm worker — hourly humidity alarm notifier.
UI: new table in Settings → Workflow → Auto-Drying, below the existing
Drying Presets table. Default row + 8 default filament types
(PLA / PETG / TPU / ABS / ASA / PA / PC / PVA) pre-filled from the
current ams_humidity_fair value so the editor starts sensibly.
Input pattern: draft-on-edit / commit-on-blur (transient humidityDrafts
state per row). onChange only updates the draft; onBlur (and Enter)
parses + clamps to [5, 95] + commits. Empty value on blur clears the
override and falls back to default. Caught mid-PR via a typing test:
the naive per-keystroke clamp snapped "3" → 5 before the user could
type the second digit of "30".
Setting is in the public _UI_PREFERENCE_FIELDS allowlist (same rationale
as drying_presets and ams_humidity_fair — non-sensitive integer map,
no SETTINGS_READ permission required for badge-color rendering).
Operator-flipped out-of-service state per printer for three scenarios:
parallel Bambuddy installs (dev + prod where the printer rejects all
but one MQTT client), printers under repair, and temporary suspension.
The backend Printer.is_active gate has shipped since day one and is
already honoured by every consumer — MQTT (printer_manager), queue
dispatch (print_scheduler, print_queue), metrics, scheduler, picker,
backup, maintenance dashboard. The missing piece was UI exposure.
Three entry points to flip is_active:
- Three-dot overflow menu (Enter / Exit maintenance mode, wrench icon)
- Exit button inside the in-card amber panel
- Checkbox in EditPrinterModal
Card UI: expanded mode shows an amber panel (Wrench + "In Maintenance"
+ subtitle + Exit) where the cover/progress container would normally
render — same height, no layout shift. Compact mode shows an amber
pill in place of the progress bar. Header pill swaps Connected/Offline
for "Maintenance" and the diagnostic CTA is suppressed (deliberate
state, not involuntary offline). HMS / Queue / Firmware pills fall
away naturally via the existing status?.connected gates.
Mid-print entry (RUNNING / PAUSE) triggers a confirmation dialog —
disconnecting MQTT mid-print stops progress tracking and completion
notifications for the in-flight job. Idle / FINISH / FAILED skip the
dialog and toggle directly.
Scope: no backend change, no new permission, no behaviour change for
any other consumer. PrinterCreate.is_active?: boolean added to the
TypeScript surface so the field flows through api.updatePrinter.
The Printers page rendered three fan widgets (part / aux / chamber)
for every printer unconditionally. Open-frame models (A1, A1 Mini,
A2L, P1P) have no chamber fan — the firmware reports big_fan2_speed
as 0, so the badge always rendered greyed-out and let users "set" a
fan speed on hardware that doesn't exist.
Adds MODELS_WITH_CHAMBER_FAN allowlist (X1C / X1 / X1E / X2D / P1S /
P2S / H2D / H2D Pro / H2C / H2S) near mapModelCode, and the chamber
entry is spread into fanItems only when the printer's model is in the
set. Open-frame printers now show two badges (part + aux), which
matches their actual hardware.
Allowlist not denylist: mirrors the file's existing classification
pattern (the enclosure-door badge gate uses the same shape), and the
failure mode is preferable — a missing badge on a real chambered
printer is obvious; a phantom badge on a future open-frame model
would look correct and silently lie.
In-app "Install Update" on Windows installer installs failed with "Could
not find git executable" because (1) _find_executable's fallback paths
are Unix-only, and (2) the installer stages backend/ via shutil.copytree
so there is no .git directory — even with Git for Windows installed, the
fetch would die on "not a git repository". Adding Windows paths would
only have changed which error users saw.
Switches the Windows installer path to a fourth update_method
("windows_installer") that mirrors the existing docker / ha_addon
branches — surface a link to the release .exe and let the user re-run
the installer, matching the Discord / Spotify Windows update model.
Backend:
- New _is_windows_installer_install() — true iff sys.platform == "win32"
AND no .git in app_dir, so Windows devs with a real git clone keep
the git path.
- New _find_windows_installer_asset() picks the matching release asset
(prefers versioned bambuddy-<ver>-windows-x64-setup.exe, falls back
to the unversioned alias on non-daily tags).
- /updates/check now returns is_windows_installer / update_method /
installer_download_url.
- /updates/apply short-circuits with a friendly message after the
existing HA / Docker guards — defense in depth, the frontend swaps
the button so the POST should not fire on Windows.
Frontend:
- UpdateCheckResult extended with the new fields and 'windows_installer'
in the update_method union.
- SettingsPage renders a Bambu-green styled <a target="_blank"
rel="noopener"> between the Docker snippet and the in-app Update
button, with installer_download_url falling back to release_url then
the tag page so the link is never broken.
- applyUpdateMutation onSuccess toast guard extended to treat
is_windows_installer the same as HA / Docker.
ghcr.io pull baseline (~10k/day rising → ~8-12k active installs) puts
sponsor conversion at 0.08% — roughly an order of magnitude under
industry-benchmark for OSS with visible CTA. The Settings banner from
0d4b9d4e gives passive every-visit visibility on one page; this adds
opt-out-able active visibility at moments where the user has just
earned something with Bambuddy.
Five trigger families with a 14-day cross-family cooldown: prints
(100/500/1000/2500/5000), cost (100/500/1000 tracked filament +
energy), archives (50/250/1000), anniversary (1 year), version-update
(re-armable on each major bump). New sponsor_toast_state table with
nullable user_id so auth-disabled installs get the same trigger logic
through one code path (NULL-keyed install-default row).
Matomo shows only 1.18% of website visitors reach /sponsors despite
29% hitting /installation. The ask was discoverable on the marketing
site but invisible in-app where users actually live.
Full-width gradient banner sits above the three-column layout on the
default landing tab and links to bambuddy.cool/sponsors.html with a
?from=app-settings tracking param so conversion lift is measurable
in Matomo. Three new sponsors.* keys translated to all 11 locales.
New PrinterSensorHistory table + 60s recorder + GET/DELETE /printer-sensor-history
route gated behind a new PRINTER_SENSOR_HISTORY_READ scope (separate from AMS). UI
adds a 10x10 LineChart icon on each heater tile - click body opens the existing
target-temp popover unchanged, click icon opens a HeaterHistoryModal mirroring the
AMSHistoryModal shape (kind toggle + 6h/24h/48h/7d range + current/avg/min/max +
recharts line for value + dashed target). Read-only X1C/P2S chamber tile finally
gets an interaction. Retention configurable via printer_sensor_history_retention_days
(default 30, sibling of ams_history_retention_days). 8 new i18n keys translated in
all 11 locales, parity green. 4 backend + 6 frontend tests added; full pytest -n 30
6226/6226, vitest 2176/2176, ruff/eslint/build all clean.
Reporter forcefully started a print needing ~260 g with 180 g on the
first spool and a backup spool in the AMS. Printer correctly consumed
spool 1, AMS Backup switched, spool 2 finished the print. Bambuddy
attributed all 260 g to spool 2 -- spool 1 untouched in inventory.
Two stacking bugs produced the exact "all to second spool" symptom for
prints without per-layer 3MF gcode data:
1. bambu_mqtt.py:2135 wrote state.total_layers = int(data["total_layer_num"])
unconditionally. P1S firmware pushes total_layer_num=0 at print end
(same reset pattern other models do for layer_num / progress). The
unconditional write clobbered the slicer's actual total to 0 before
the usage tracker read it.
2. usage_tracker.py:1129-1137 linear-fallback dumped EVERYTHING onto the
last segment when total_layers was 0:
if total_layers > 0:
segment_grams = total_weight * (seg_end_layer - seg_start_layer) / total_layers
else:
segment_grams = 0.0 # <- entire print weight ends up on last segment
Path 2 (AMS remain% delta) couldn't recover because (a) the emptied
spool reported remain=-1 and (b) Bug-A had already added the second
spool's key to handled_trays, suppressing the Path 2 lookup.
Fix:
- bambu_mqtt.py: only overwrite state.total_layers when the incoming
value is positive (mirror of the existing _last_valid_layer_num
pattern at line 2127). Explicit reset on new print start at
_handle_print_start so the previous print's total can't bleed in.
- usage_tracker.py: cascade the linear-fallback denominator -
state.total_layers, then last_layer_num (already threaded in for
the last_progress fallback), then equal-split as a bounded fence.
Equal-split is still wrong but never dumps the whole print on the
last segment, which was strictly worse.
Two tightly-coupled deliverables in one drop -- a new AMS Filament Backup
status/control surface, and the #1766 fix that depends on it.
Added -- AMS Filament Backup status + control
- Parse bit 18 of top-level print.cfg into PrinterState.ams_filament_backup
on every push_status. Verified against OrcaSlicer source
(DeviceManager.cpp:4961) and a live H2D ON/OFF capture. Tri-state
(None = A1 family / pre-cfg push) preserves today's behaviour.
- Hold-timer guard (3 s) prevents stale frames from flickering the badge
back to the printer's old cfg after a user-initiated toggle.
- POST /printers/{id}/ams-backup toggle, set_ams_filament_backup() client
method calling _set_print_option("auto_switch_filament", enabled).
- GET /printers/{id}/inventory-remain endpoint exposes the same map the
dispatcher uses (internal and Spoolman modes both work uniformly).
- Small icon badge in the printer card's "Filaments" section header
(placement reads as printer-wide because the cfg bit is printer-wide,
not per-AMS). Click to toggle, success toast.
- 5 i18n keys x 11 locales for the badge UI.
Fixed -- #1766: prefer_lowest didn't pick lowest, ignored backup state
- Backend gate in _compute_ams_mapping_for_printer: coerce prefer_lowest
to False when status.ams_filament_backup is False; log the skip.
- New effectivePreferLowest(setting, backup) helper applied at every
frontend sort entry point: single-printer PrintModal, multi-printer
hook per-printer, PrinterSelector InlineMappingEditor, FilamentMapping
standalone editor (the last had NO preferLowest awareness at all
before this change).
- New preferLowestSortKey(f, inventoryByTrayId) mirrors backend's two-tier
key exactly, including the banding tie-break (regular AMS < AMS-HT <
external) so the client-side pre-compute matches the dispatch-time pick.
An earlier draft used a flat `amsId * 4 + trayId` priority which gave
external slots (ams_id = -1) a NEGATIVE priority -- caught in code
review before commit.
- Settings -> Filament -> "Prefer lowest remaining filament" gets an
explanatory note about the printer-side AMS Backup dependency, with
i18n key in all 11 locales.
The AMS humidity/temperature stats modal hardcoded color literals
gated on a light/dark boolean, so it ignored the active background
variant (neutral / warm / cool / oled / slate / forest) and the
light-bg variants. Switched modal chrome, stat cards, and the
recharts grid / axes / tooltip to read --bg-secondary, --bg-primary,
--border-color, --text-primary, --text-secondary, --text-muted from
the active theme so the modal follows mode AND background variant.
The 3MF parser sums prediction + weight across every plate (#1593) so the
archive card can headline the whole project — correct for the card, wrong
for the completion notification of a single plate. The queue UI already
re-reads the 3MF per-plate at print_queue.py:272-285; mirror that for the
notification path so Discord / Pushover / email show the plate's actual
duration and grams instead of the project sum. Helper fails open on every
error path so a missing or corrupt 3MF can't block the notification.
BambuStudio's project_file MQTT command for O1C2 (the H2C dual-
nozzle-rack variant) carries nozzle_mapping (per-filament physical
nozzle position IDs) and nozzles_info (per-extruder rack metadata).
The VP intake was dropping both, so the H2C firmware fell back to
"last matching nozzle type" auto-pick and ignored the user's
slicer choice — every HF print landed on R2, every standard print
landed on R4.
Carry both fields through the VP intake → queue item → MQTT
dispatch path. New nullable TEXT columns on print_queue, non-
branched ALTER (matches ams_mapping / filament_overrides
precedent). Dual-nozzle gate at start_print() keeps the fields
off single-nozzle dispatches. Fail-open on malformed JSON —
firmware auto-picks, never worse than pre-fix.
Stamps both fields on every plate in the multi-plate Send All
loop (#1697 / #1188 precedent).
ams_mapping2 still handles H2D/X2D dual-extruder routing
unchanged; this fix is scoped to the O1C2 rack-swap mechanism.
install/docker-install.sh::create_install_dir ran `mkdir -p
"$INSTALL_PATH"` without sudo while DEFAULT_INSTALL_PATH was
/opt/bambuddy, root-owned on every Linux distro. set -e then
aborted the whole script before docker compose could pull the
image — anyone running the documented `curl ... | bash` flow as
a normal user hit this on first install.
Fix: try the unprivileged `mkdir -p ... 2>/dev/null` first so
--path ~/bambuddy, /srv/bambuddy and other writable targets don't
trigger a needless password prompt, then fall back to
`sudo mkdir -p` + `sudo chown -R "$USER:$USER"` only when the
first attempt failed. The chown is load-bearing: without it the
script would later try to write docker-compose.yml + .env into a
root-owned dir as the invoking user and cascade further EACCES
failures.
Not changing the default path: install/update.sh and
install/update_macos.sh both default INSTALL_DIR to /opt/bambuddy,
and install/README.md's update flow documents the same — flipping
the install default to ~/bambuddy without coordinating those
would silently break self-service updates for anyone following
the docs verbatim. The default stays /opt/bambuddy; only the
escalation gap closes.
set -e survives the redirected stderr because the `if !` form is
the documented escape hatch for an expected-failure check.
Smoke-tested writable-target, idempotent-rerun, and the
failing-mkdir-then-sudo-fallback branches.
Bambuddy's archive cards were blank for every print sliced through the
BS or Orca docker sidecars. The "Some recent prints couldn't be archived
with thumbnails" banner pointed at install step 4 which is unrelated —
that flag only fires on FTP-fetch failures, not on missing-thumb in the
sliced 3MF.
Root cause is upstream of Bambuddy: neither slicer CLI renders
Metadata/plate_N.png when invoked headlessly with --slice --export-3mf.
That render is a separate code path triggered by --export-png, which is
mutually exclusive with --export-3mf and additionally needs a working
display backend (BS 02.07.x's bundled GLFW is hard-locked to Wayland —
even XDG_SESSION_TYPE=x11 + GDK_BACKEND=x11 + QT_QPA_PLATFORM=xcb don't
switch it back). An Xvfb display in the sidecar wouldn't help even if we
wired the second-pass call. The Orca sidecar has been silently shipping
thumbnail-less 3MFs from STL inputs since launch; nobody noticed.
Fill the gap on the Bambuddy side: new plate_thumbnail.py renders the
missing thumbnails after the slice returns. inject_plate_thumbnails_if_missing
parses the sliced zip, finds every Metadata/plate_N.gcode entry that
doesn't have a matching plate_N.png, loads 3D/3dmodel.model via trimesh,
renders an isometric Bambu-green-on-dark view at 512x512 + 128x128 via
the same matplotlib Agg pipeline as stl_thumbnail.py, and re-packs the
zip with the PNGs injected. Visual style matches Bambuddy's existing
library thumbnails — archive cards stay consistent inside Bambuddy rather
than chasing parity with desktop Studio's plate render. Best-effort:
input bytes are returned unchanged on any failure so the slice flow itself
can never fail because of a missing thumbnail. Idempotent: re-running on
an already-injected 3MF returns the input verbatim.
Wired into both library.py slice paths via result._replace; covers the
cross-class merged-multi-plate path automatically (merged bytes flow into
the same write site). No sidecar Dockerfile change required — an earlier
attempt to install Xvfb in Dockerfile.bambu-studio was a false start and
is not part of this drop.
Dependencies: trimesh's 3MF loader uses networkx (scene-graph traversal)
and lxml (model.xml parse) lazily inside the 3MF code path — both added
to requirements.txt because they aren't strict trimesh transitives.
Follow-up to the temperature & fan-speed presets feature — the
TestUiPreferencesEndpoint.test_returns_expected_field_set test pins
the exact set of fields the endpoint exposes (so adding a sensitive
field by accident fails the assert). The 4 preset fields were added
to _UI_PREFERENCE_FIELDS without updating the pin, breaking the full
backend test run.
- The Speed and AMS load/unload tests broke after the printer-card
refactor in #1661 (icon-only Gauge button replaced the "100%" badge,
hover-card actions replaced the kebab "Slot options" button). Add
data-testid="speed-control" + data-testid="filament-slot" as stable
test hooks, rewrite both files around them. Parametrize the four-mode
speed-selection test. Update the "RUNNING hides menu" assertion to
"Load/Unload buttons exist but are disabled" — the new UX shows
actions on hover and disables them rather than hiding the trigger.
- Drop `animate-in fade-in-0 zoom-in-95 duration-150` from
FilamentHoverCard and EmptySlotHoverCard. The card briefly painted
at the offscreen (-9999, -9999) coords during the zoom-in transition,
reading as a fly-in from the upper-left corner. With the animation
gone it just appears in place at its computed position.
PR #1661 swapped the visible speed badge ("100%") for an icon-only Gauge
button and replaced the kebab "Slot options" button with hover-card
actions inside FilamentHoverCard. The two existing test files weren't
updated alongside the refactor and stayed broken on dev.
- Add data-testid="speed-control" to the Gauge button and rewrite the
Speed tests around it; assertions on the percentage text are gone
because that text no longer renders. Parametrize the four-mode API
call test.
- Add data-testid="filament-slot" to FilamentHoverCard's trigger
wrapper and rewrite the AMS load/unload tests around
fireEvent.mouseEnter → portaled actions. Replace the "hides menu
while RUNNING" assertion with "Load/Unload buttons exist but are
disabled" — matches the new UX shape.
The matcher's tray_info_idx vs color vs type_only bucket choice was
debug-only, so a bug report's bundle never showed which path won. Emit
the sorted candidate trays and the picked bucket per filament req when
prefer_lowest=True. Behaviour-neutral; existing 89 matcher tests pass.
GET /api/v1/printers/ and /api/v1/printers/{id} return access_code
only when the caller holds PRINTERS_UPDATE. Adds PrinterResponseWithSecret
as the elevated response shape; PrinterResponse no longer carries the
field. Auth-disabled single-trust mode preserved.
The Slicer -> Local Profiles page kept showing a just-deleted row for
the ~hundreds of ms it took invalidateQueries to refetch. A quick
re-click on the same row opened a second delete-confirm modal that
resolved to a 404 from the backend.
Add an optimistic queryClient.setQueryData filter in deleteMutation's
onSuccess so the row disappears the instant the DELETE returns 200.
Existing invalidateQueries calls stay in place to reconcile any drift.
Found while reproducing #1713 (verifying maziggy's setup against the
reporter's). Unrelated to that investigation but caught here.
Major version bump for the frontend build:
- vite ^7.3.2 -> ^8.0.16
- @vitejs/plugin-react ^5.1.1 -> ^5.2.0
Vite 8 swaps Rollup for Rolldown as the default bundler
(Rust-backed, same plugin contract). The bump also lifts the
transitive esbuild floor to 0.28.1, closing the last open
advisory in the audit chain.
vite.config.ts surface audited and unchanged:
- defineConfig, Connect type
- serveGcodeViewer configureServer middleware
- server.proxy with WebSocket upgrade for /api/v1/ws
- build.outDir / emptyOutDir / chunkSizeWarningLimit
- resolve.alias for @
- base: '/' regression guard from #1221
vitest@4.1.8 already accepts vite 8 in its peer range
(^6 || ^7 || ^8); no test-runner bump required.
Node floor for vite 8 is ^20.19.0 || >=22.12.0; CI Node 20.x
line satisfies this.
Not taken: plugin-react v6 — it requires
babel-plugin-react-compiler and @rolldown/plugin-babel as
peers and is a separate scope.
The SpoolBuddy inventory page reimplemented its filter inline and only
matched material/subtype/brand/color_name/note, while Bambuddy's main
inventory uses the shared filterSpoolsByQuery helper which also matches
spool ID, slicer_filament_name, and storage_location. Delegate to the
shared helper so both pages stay in lockstep.
- frontend/src/pages/spoolbuddy/SpoolBuddyInventoryPage.tsx: replace
inline filter with filterSpoolsByQuery
- frontend/src/__tests__/pages/SpoolBuddyInventorySearch.test.ts: lock
in ID / partial ID / pre-fix fields / parity-gain fields
navPermissions in Layout.tsx gated three resources on the LEGACY *:read flag.
Default Operators group is seeded with *_own only (and the migration map flips
legacy → _own on existing groups), so non-admin users never held the legacy
permission and the sidebar hid Archives / Queue / Files even though the
underlying API accepted their requests. Reporter only spotted Files; same bug
shape applied to Archives and Queue.
Fix: navPermissions accepts Permission | Permission[]; the three affected
resources list all three tiers. isHidden checks .some(hasPermission) for
arrays. Permission type extended with the matching *_own / *_all variants —
backend already shipped them, the TS type just didn't declare them.
The provider toggle, schema, template, and NotificationService.on_printer_offline
all shipped, but no caller invoked the dispatcher — the offline event was an
orphan toggle. Edge detection in on_printer_status_change now schedules a
debounced (60s) background task on the connected→disconnected transition;
reconnect before the window elapses cancels it. Covers both upstream paths
(smart-plug power-off via mark_printer_offline, and MQTT staleness via
check_staleness), both of which already route through the status callback.
The "back online" channel is the existing print-failure notification on
firmware FAILED report — no symmetric on_printer_online needed.
ProtectedRoute and PermissionRoute now pass the requested location as
router state when redirecting to /login. LoginPage stashes it in
sessionStorage before the OIDC provider redirect (since window.location
kills React state) and consumes it on all three post-login navigations
(credentials, 2FA, OIDC token exchange). Targets are sanitized to
same-origin internal paths only — protocol-relative and /login itself
are rejected to prevent open-redirect.
QR labels (https://host/inventory?spool=N) now land on the scanned
spool instead of the printer page after authentik / any OIDC SSO login.
Older print_archives rows (and rows that landed via the SQLite ↔ Postgres
cross-DB restore path) can have created_at = NULL because the column was
originally created without a DEFAULT clause — server_default=func.now()
only fires at table creation, not for existing rows or raw cross-DB
inserts. The list_archives response model required a datetime, so a
single NULL row 500'd the whole endpoint via Pydantic ResponseValidationError.
- Boot-time backfill: COALESCE(completed_at, started_at, now()) for
any row where created_at IS NULL. Dialect-branched (SQLite datetime('now')
vs Postgres NOW()).
- Schema: created_at is now Optional on ArchiveDuplicate, ArchiveResponse,
and ArchiveSlim so a future NULL-leaking path doesn't break the list
endpoint again.
Round 2 (166e9f9e) fixed the stash-key mismatch, but @mkoreen's
2026-06-23 bundle showed BS's MQTT project_file arrived 85 ms past the
2.0 s wait timeout (FTP done 00:42:02.509, "No slicer options cached"
00:42:04.509, MQTT 00:42:04.594). Queue item was committed with
settings defaults; nozzle_mapping never made it onto the wire.
Three pieces:
1. _SLICER_OPTIONS_WAIT_TIMEOUT module constant, 2.0 -> 5.0 s. Covers
wireless / loaded-Pi jitter; one-time +3 s cost only for legacy
slicers that never send MQTT.
2. _RECENT_QUEUE_ITEM_TTL fallback: on_print_command retroactively
UPDATEs slicer-driven fields on a recently-committed queue item
when the event wait already gave up. Tracked via
_recent_queue_items dict (30 s TTL, evicted on every queue-add).
Gated on status='pending' so we never race the dispatcher.
Multi-plate covered via WHERE id IN (...).
3. Post-commit last-chance pop. Audit caught a race in (2): MQTT could
arrive during any await inside _add_to_print_queue (wait_for,
archive_print, db.flush, db.commit), and on_print_command would
stash data with no event consumer AND no _recent_queue_items entry
yet. After populating _recent_queue_items, _add_to_print_queue now
pops _slicer_print_options[file_path.name] one last time and
routes any hit through _restamp inline.
Bulk operations on the Inventory page in both built-in and Spoolman modes.
Reporter wanted ten-of-the-same-spool edits without ten round-trips through
the per-spool editor.
Frontend
- New checkbox column on the inventory table (header / row / group). Sticky
toolbar appears when at least one row is selected with Edit / Print labels /
Reset usage / Archive (or Restore in the Archived tab) / Delete / Clear.
Selection clears on any filter / tab / search change so the count can't
drift from what is on screen.
- BulkEditSpoolsModal is a three-state-per-field form. The user opts in per
field by ticking its checkbox or just typing into it; only ticked + non-
empty fields are sent. Clearing fields in bulk is intentionally NOT
supported per the issue discussion.
- A new SearchableSelect renders all categorical fields (material, sub-type,
brand, category, slicer preset name, slicer filament, storage location)
with the same dropdown pattern the per-spool editor uses - text input +
chevron + filtered button list, click-outside / Escape closes. No native
select anywhere in the modal. Options merge the canonical constants from
spool-form/constants.ts with whatever already exists in the user's
inventory. Slicer-preset dropdowns fetch the same sources as the per-spool
form (Bambu Cloud + Orca Cloud + local + built-in) through buildFilament
Options() and three useQuery calls gated on isOpen.
- onSuccess handlers surface three outcomes: all-succeeded (green toast),
partial-success (yellow toast with ok / failed counts), all-failed (red
toast that keeps the selection and modal open so the user can retry).
The first cut silently dropped errors / not_found arrays - audited and
fixed before merge.
- Invalid rgba hex is flagged inline with a red border + helper text and
the Apply button is gated on a hasDroppedTickedField guard, so silently
dropping a ticked field is no longer possible.
- bulkResetConsumedCounterMutation.onSuccess now closes the confirm modal +
clears selection, matching the other three bulk mutations.
Backend
- Four new endpoints per inventory mode (eight total):
POST /api/v1/inventory/spools/bulk-update INVENTORY_UPDATE
POST /api/v1/inventory/spools/bulk-delete INVENTORY_UPDATE
POST /api/v1/inventory/spools/bulk-archive INVENTORY_UPDATE
POST /api/v1/inventory/spools/bulk-restore INVENTORY_UPDATE
POST /api/v1/spoolman/inventory/spools/bulk-* FILAMENTS_UPDATE
- Built-in update runs the same prepare_internal_spool_payload(...) +
weight_used / weight_locked auto-stamp as the per-spool PATCH.
- Spoolman update loops the per-spool update_spool route function so the
filament re-linking / extra-dict / extra-lock / shared-filament rules
stay byte-identical to single-spool edits.
- Per-spool failures inside the batch are collected. Spoolman bulk-delete /
archive / restore now catch non-HTTPException too (matches bulk-update) -
a mid-batch httpx.ConnectError or TimeoutError no longer aborts the route
with a 500 and skips the WS broadcast.
- Both modes broadcast a single inventory_changed WS event at the end of
the batch.
New setting "Auto-add unknown RFID spools" under Settings -> Filament -> Filament Tracking,
default ON for back-compat. When turned off, the backend stops auto-creating an inventory
record for an unknown RFID tag and instead broadcasts an unknown_tag WS event that pops
a global confirmation modal in the Bambuddy UI showing the printer / AMS-X label / slot /
material / colour. Add or Cancel; no nag on every MQTT push.
Backend
- Module-level _unknown_tag_last_broadcast dict dedupes per (printer, slot, tag). Set is
committed AFTER ws_manager.broadcast() returns so a crashed broadcast doesn't poison
the dedup and permanently silence the slot.
- Empty-slot MQTT push clears that slot's entry, so remove+reinsert reliably re-prompts.
- Successful matches via get_spool_by_tag / find_matching_untagged_spool / create_spool
also clear the entry so a future tag swap re-prompts.
- Tray data (tray_type, tray_color, tray_sub_brands, tray_count) shipped in the WS payload
directly so the modal renders the real material / colour instead of relying on the
React Query cache that lags the WS event by several seconds.
- Two new endpoints back the modal's confirm action:
POST /api/v1/inventory/spools/from-slot (INVENTORY_UPDATE)
POST /api/v1/spoolman/spools/from-slot (FILAMENTS_UPDATE)
Both look up the slot's tray data server-side and create + auto-assign atomically.
- Spoolman /from-slot now raises HTTP 500 when the slot-assignment INSERT fails instead
of returning success while the DB rolled back the binding.
- sync_ams_tray gained an optional auto_add_unknown_rfid kwarg (default True so existing
callers are unaffected); auto-sync and both manual sync routes thread the setting.
Frontend
- useUnknownTagPrompt hook listens for the unknown-tag CustomEvent, reads the tray fields
out of the event detail, and feeds a single-modal queue. No long-lived dismissed set;
the backend dedup handles spam suppression.
- UnknownSpoolModal wraps the existing ConfirmModal with a material + colour-swatch
preview block.
- Mounted in Layout.tsx alongside useSponsorPrompt so SpoolBuddy kiosk / login / setup
routes are excluded.
- getAmsLabel moved to utils/amsHelpers.ts; ConfigureAmsSlotModal.tsx and PrintersPage.tsx
both import the shared version (canonical AMS-A / HT-A / External labels).
- AppSettings TS interface gained spoolman_enabled, auto_add_unknown_rfid, spoolman_url
so the runtime cast in the hook is no longer needed.
- SpoolmanSettings.tsx gets a new toggle row in the Filament Tracking card, visible in
both built-in and Spoolman branches; auto-save + toast already wired.
Extends the appliance endpoint that landed in the previous commit with a
time_synced field, sourced from /run/bambuddy/time-synced (the appliance's
ntp-gate.sh writes this once chronyd reports sync, or with a "warning"
marker after the 3-minute timeout). The RPi 5 has no battery-backed RTC,
so on a fresh boot the system clock is wrong until NTP catches up -- JWT
expiries and TLS certificate validity windows depend on this being right.
Exposing the gate lets the SPA render a "time not synced" indicator while
that's still true and clear it once "ok" comes through.
backend/app/core/local_config.py
New read_ntp_gate(path) function alongside read_local_toml. Three states:
"ok" chrony reported sync within the 3-minute window
"warning" 3-minute timeout elapsed without sync; user already waited
and the wizard proceeded with a degraded clock
None file absent (non-appliance install), OSError, empty content,
unknown marker, or binary garbage -- "unknown / don't gate"
Defensive read mode (errors="replace") survives non-utf8 content without
crashing. Module docstring broadened from "local.toml reader" to "small
readers for appliance-set state files".
backend/app/api/routes/system.py
/system/appliance now returns:
{hostname, timezone, locale, time_synced}
with the same no-auth posture: bootstrap surfaces (i18n init, time-sync
banner) read this before auth might be set up, and the contents are
non-secret (user-set defaults + a public sync flag). The endpoint
docstring expands to explain the RTC motivation -- otherwise the
time_synced field reads like a leftover.
Closes the cross-repo contract started in bambuddy-appliance: the firstboot
wizard writes /etc/bambuddy/local.toml with the user's hostname / timezone /
locale, but nothing on the main app side read it. Hostname + timezone are
already applied by the appliance's firstboot.sh via hostnamectl /
timedatectl. This PR closes the loop for the third field — locale — so the
language the user picked in the wizard actually shows up on first SPA load.
backend/app/core/local_config.py
New module. read_local_toml(path) returns a LocalConfig TypedDict
({hostname?, timezone?, locale?}) parsed from /etc/bambuddy/local.toml.
Defensive on every failure mode -- missing file returns {}, invalid TOML
returns {} + log warning, non-string values dropped with warning. The
reader never raises; a malformed config never blocks startup.
backend/app/api/routes/system.py
New endpoint GET /system/appliance. Returns {hostname, timezone, locale}
with null for any field not present in the TOML. No auth required: the
frontend i18n bootstrap reads this before auth might be set up, and the
contents are user-set defaults, not secrets. The function calls
read_local_toml() with no args (default path) so tests can monkeypatch
the module's read_local_toml reference to inject fixtures.
frontend/src/i18n/index.ts
One-shot applyApplianceLocale() runs after i18n.init(). Gated by a
bambuddy_appliance_locale_consumed localStorage flag so it runs at most
once per appliance. Fetches /api/v1/system/appliance, validates the
returned locale against supportedLngs, calls i18n.changeLanguage if
valid. Silent .catch() because the endpoint absent / unreachable means
non-appliance install or dev environment -- we leave the LanguageDetector's
choice in place. The consumed flag is set on success; future loads skip
the fetch entirely. Won't override a user's explicit language pick (the
language picker writes to a separate localStorage key, bambutrack_language).
Third and final piece of #1268, alongside the recursive-search +
README-panel commit that landed earlier in 0.2.5b1. Folders express
hierarchy (one home per file); tags are orthogonal labels — "toy",
"kid-safe", "petg-only" — and a single file can carry as many as the
user wants. Reporter wanted to find "every toy regardless of which
folder it lives in"; folders alone can't do that without forcing the
file into one bucket.
Design decisions locked with maziggy before code:
- file-only (folders already express hierarchy)
- multi-tag filter = AND
- tag filter IGNORES the selected folder (cross-cutting by design)
- bulk-tagging from multi-select toolbar in v1
- no auto-tags from 3MF metadata (user-authored only)
- label-only chips, no color/icon
Backend
- LibraryTag (id, name, name_key UNIQUE = LOWER(TRIM(name)))
in backend/app/models/library.py. Case-insensitive UNIQUE
collapses "Toys"/"toys"/"TOYS " into one row, so the route
returns 409 instead of silently fragmenting the catalog.
- LibraryFileTag(file_id, tag_id) association, composite PK,
ON DELETE CASCADE both directions. Deleting a tag drops every
chip; files survive. Deleting a file drops its tag links; the
catalog row survives.
- Both tables auto-create via Base.metadata.create_all — no
explicit run_migrations step needed for new tables.
- New router at backend/app/api/routes/library_tags.py with:
GET /library/tags (list + per-tag file_count)
POST /library/tags (create, 409 on case-insensitive dup)
PATCH /library/tags/{id} (rename, 409 on collision, self-rename OK)
DELETE /library/tags/{id} (cascade)
POST /library/tags/bulk-assign (add | remove | replace)
- Bulk-assign add is idempotent; replace with empty tag_ids clears
the file's tag set. Per-file ownership enforced — *_OWN callers
can only modify their own files; unknown file_ids quietly
skipped (matches library_trash bulk shape).
- list_files gains tag_ids: list[int] query param. AND semantics
via JOIN + GROUP BY + HAVING COUNT(DISTINCT) — portable across
SQLite and Postgres. When tag_ids is non-empty, folder_id /
project_id / include_root / recursive are all bypassed so the
result is cross-cutting.
- FileListResponse gains tags: list[{id, name}] via
selectinload(LibraryFile.tags) — N+1-free chip render.
- Permissions reuse existing constants: LIBRARY_UPDATE_ALL for
catalog mutations (global catalog, ownership-aware update isn't
meaningful), LIBRARY_UPDATE_ALL/OWN pair for bulk-assign,
LIBRARY_READ_ALL/OWN for list — file_count projection narrows
for *_OWN callers so chip counts match what they actually see.
Frontend
- LibraryTagsModal (catalog CRUD) opens from the toolbar's new
Tags button. max-w-4xl so multi-language subtitles don't wrap.
Delete-with-warning when file_count > 0 ("removes the chip from
all of them; files themselves are untouched").
- BulkTagsPickerModal opens from the multi-select toolbar (new
Tag button between Move and Delete). Add/Remove radio,
checkbox list, inline "create new tag" disabled on dup.
Apply disabled until at least one tag is selected. The replace
action is exposed in the API but deliberately NOT in this UI —
arbitrary multi-file replace is destructive and confusing.
- FileManagerPage integration:
* selectedTagIds state, sorted into the useQuery key so the
cache hits are stable regardless of toggle order
* filter rail above the file list lists EVERY catalog tag as
a togglable chip — inactive outlined, active filled green
with an X. Clear all when 1+ active. Bar hidden entirely
when catalog is empty.
* useEffect prunes selectedTagIds when a tag is deleted from
the catalog so the filter never strands on a phantom id
* dedicated Tags column in list view at minmax(0,200px)
between Prints and Actions
* grid view chips render below the metadata block
* chip clicks stop propagation so they don't toggle file
selection
- libraryTagsQueryKey extracted to frontend/src/utils/
libraryTagsQuery.ts so component files export only components
(Vite react-refresh rule).
- LibraryFileListItem.tags is OPTIONAL even though the backend
always emits an empty array — legacy msw mocks in pre-existing
tests construct partial file shapes without the field. Without
the ? the FileCard renderer crashed on .length and broke 49
unrelated tests across FileManagerPage + FileManagerExternalFolder.
Read sites use file.tags ?? [].
Reporter (@zumik3-del, seconded by @unLieb) asked for three File Manager
improvements: recursive search, tags, and a markdown preview side panel.
This commit ships the two scoped ones; tags is held back gated on the
"give the issue a thumbs up" interest check Martin posted on the issue
because it's a much larger surface (M2M schema, CRUD endpoints, tag UI +
filter + autocomplete + i18n for the management surface) and isn't the
right call without a real demand signal.
1) Recursive search inside the selected folder.
Until now, selecting "Toys" and typing "robot" only found files
directly in Toys/ — anything under Toys/Cars/Race/ stayed invisible.
The page's client-side filter ran over a server-narrowed listing
(/library/files?folder_id=X is strict equality on folder_id), so the
client filter couldn't see what the listing never loaded.
list_files (backend/app/api/routes/library.py:1729+) gains a
recursive=true query param. When combined with folder_id, the route
walks library_folders.parent_id via a recursive CTE rooted at the
requested folder and returns every descendant folder's files in one
query. Recursive CTEs work on both SQLite >=3.8.3 (2014, well below
Bambuddy's runtime floor) and Postgres without dialect branching.
Default off so the existing folder-browsing call sites (Project /
Archive detail, the FE's no-search case) keep their narrow scope.
FE opts in only when both a folder is selected AND searchQuery is
non-empty (FileManagerPage.tsx — derived as searchExpandsSubfolders,
threaded through the useQuery key so the cache invalidates on
toggle). Small "Including subfolders" caption renders under the
search input when active so the user understands why a file from two
levels deep showed up.
2) Per-folder markdown description panel.
New endpoint GET /library/folders/{folder_id}/readme returns the
first .md file in the folder as {filename, content, truncated}.
Selection prefers README.md / readme.md / description.md
(case-insensitive via func.lower(filename) LIKE '%.md' + an
in-Python stem-preference sort), falls back to the
alphabetically-first *.md otherwise. 404 when no markdown is present
so the FE can hide the side panel — non-users pay no UI cost.
Bytes are clipped at 512 KiB (_README_BYTES_CAP) with a truncated
flag so the panel can warn the reader. UTF-8 decode uses
errors="replace" so one bad byte never blanks the panel.
New FolderReadmePanel.tsx fetches on folder-select and renders via
react-markdown@9 + remark-gfm@4 (tables, strikethrough, task lists).
Collapsible (default expanded), max-height 24rem with internal
scroll. react-markdown 9 doesn't render raw HTML by default — no
dompurify needed. Links open in a new tab with rel=noopener
noreferrer. Tailwind has no typography plugin in this project so
per-element components map h1/h2/h3/p/ul/ol/code/blockquote/table
to explicit utility classes that match the rest of the app.
Scope and permissions.
Both endpoints reuse the existing LIBRARY_READ_ALL / LIBRARY_READ_OWN
ownership-aware pair, so a viewer-tier user with read_own only sees
their own files in recursive listings and can only fetch the README of
folders containing their own files. No new permission, no DB migration.
The recursive CTE is a single SQL query — no N+1, no per-folder
round-trip, scales to deeply-nested model libraries.
maziggy