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.
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.
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.
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.
The two "0.0.0.0" comparisons in test_support_helpers verify the
support-bundle net.info[*].ip redaction sentinel (mirrors the
support.py:1193 annotation), not a socket bind. Annotate inline.
The "0.0.0.0" written into the support bundle is a JSON sentinel that
scrubs the printer's local IP plus the gateway/peers it sees — not a
socket bind address. Annotate inline so bandit stops flagging it.
The "0.0.0.0" written into the support bundle is a JSON sentinel that
scrubs the printer's local IP plus the gateway/peers it sees — not a
socket bind address. Annotate inline so bandit stops flagging it.
VP queue-mode multi-plate Send All
==========================================
BambuStudio / OrcaSlicer "Send All" of a multi-plate project uploads ONE
3MF containing every plate (one FTP STOR, single filename) — slice_info.config
inside the file lists N <plate> blocks with their own index metadata and
their own Metadata/plate_N.gcode payload. Pre-#1733 the VP queue path
called _extract_plate_id which returned only the FIRST plate index, and
_add_to_print_queue built exactly one PrintQueueItem from it. Plates 2..N
silently dropped on the floor. From the user's perspective: Send All of a
3-plate project produced 1 queue item, indistinguishable from a regular
single-plate Send, with no log line to explain the discrepancy.
The wire was confirmed against the live H2D-1 Proxy VP: the same file
ships whether the user clicked Send or Send All; the only intent signal
is the count of <plate> blocks inside slice_info.config.
Fix: replaced _extract_plate_id (-> int | None) with _extract_plate_ids
(-> list[int]). The list contains every <plate> block's index in order;
falls back to [1] when slice_info.config is missing / unparseable so the
single-plate case is preserved. _add_to_print_queue now loops over the
list and creates one PrintQueueItem per plate, with:
- plate-specific position = MAX(position) + iteration_number, so the
items inherit consecutive positions and the slicer's plate order
becomes the queue execution order.
- per-plate required_filament_types / filament_overrides via
extract_filament_requirements(file_path, plate_id) — the plate-aware
filter shipped with #1697 — so the scheduler's per-printer "Any X"
matching dispatches each plate onto a printer with the right
colours loaded for THAT plate, not for plate 1's filament set.
- shared archive_id across all plates (one upload = one archive row).
- the VP's auto_dispatch + manual_start posture inherited unchanged.
Net behaviour: single-plate Send hits the loop once → exactly today's
result (one queue item, plate_id from the slicer, one archive). Multi-
plate Send All of a 3-plate file → 3 queue items, plate_id 1/2/3,
consecutive positions, all referencing the same backing archive.
Archive delete cascades to queue rows
=============================================
Previously the soft-delete path (the default the trash-can button uses)
called _cancel_pending_queue_items which only flipped queue rows with
status='pending' to status='cancelled' while leaving every other status
alone AND leaving every row in the DB. The Send All multi-plate work
above made this much more visible: deleting an archive backed by N
queue items now had to clean up N rows, and what users saw instead was
N "cancelled" rows lingering in the queue history.
Backend:
- Replaced _cancel_pending_queue_items with _delete_related_queue_items
(db, archive_id) -> int. DELETEs every queue row where
archive_id = X regardless of status. Matches what the hard-delete
path already did via the ON DELETE CASCADE FK on
print_queue.archive_id — both paths now produce the same end state.
- Print history lives in PrintLogEntry (FK ON DELETE SET NULL) and is
untouched; Quick Stats / accuracy bands are preserved across both
delete paths.
- 409 guard on archives.py::delete_archive when any related queue
item is currently status='printing'. Both soft and hard delete are
gated; deleting the archive while a print is live would strip the
dispatcher's metadata trail (filament / plate / ams_mapping) out
from under the running print.
- New GET /archives/{id}/delete-impact endpoint returns
{related_queue_items: N, currently_printing: M}. Cheap, single
endpoint, deliberately NOT folded into the archive list response
so the much larger list endpoint isn't forced to run the same
query per row.
Frontend:
- ArchivesPage delete-confirm modal queries the new endpoint when the
modal opens (useQuery with enabled: showDeleteConfirm) and renders
an amber "N queue items linked to this archive will also be removed"
line when total > 0 AND printing = 0, OR a red "Cannot delete —
M queue items are currently printing" line when printing > 0
(confirm button disabled in that case so the user can't bonk the
409 on submit).
- ConfirmModal gained an optional confirmDisabled?: boolean prop —
isLoading was the only disable knob before; this adds the external-
precondition path.
- 2 new i18n keys (deleteQueueItemsWarning, deleteBlockedByPrinting)
translated across all 11 locales per feedback_translate_dont_fallback —
no English fallbacks.
No DB migration — the CASCADE FK was already in place; only the helper's
semantics changed.
capture_finish_photo (default-on) was forcing the timelapse MQTT field to
true on every print, even when the user explicitly unchecked Timelapse in
the slicer send dialog. On profiles with Timelapse Type = Smooth, that
flipped the printer's timelapse_record_flag and un-gated the per-layer
M622 J1 wipe blocks the slicer had baked in — toolhead parked off the
part every layer, on prints the user opted out of recording.
Root cause: #1397 implemented the finish-photo feature as a side channel
of "force the printer into timelapse-recording mode at dispatch" so the
last-frame extractor had a video to pull from. That conflated recording a
timelapse with snapping a finish photo, and the per-layer side effects
were decided at slice time by the user's timelapse_type, which Bambuddy
has no visibility into post-slice.
Fix: replace the force-on with a clean MQTT-state-driven trigger.
bambu_mqtt.py fires a new on_finish_photo_moment callback when
stg_cur transitions INTO 22 ("Filament unloading") while
_was_running AND end-of-print gate matches (progress >= 99 OR
layer_num >= total_layers OR remaining_time <= 0). The gate
disambiguates from mid-print color swaps (which also transit
stage 22 but at progress < 99). FINISH-state fallback in the same
handler fires the callback at the existing transition if stage 22
never arrived (cancel, external-spool-only, HMS halt, firmware
variants).
main.py registers on_finish_photo_moment as a top-level handler.
It pre-captures one camera frame at the trigger edge (external cam
→ buffered RTSP → fresh RTSP via capture_camera_frame_bytes) and
caches the JPEG bytes in _stage22_finish_frames[printer_id].
_background_finish_photo consumes the cached bytes before its
existing live-grab chain, so the saved photo has the better
framing (toolhead parked, before bed drop) without restructuring
the archive-resolution / fallback / notification wiring.
When a timelapse IS actively recording (user explicitly opted in),
pre-capture is skipped — _capture_finish_photo_from_timelapse
still extracts the last frame, which is still the best framing
and now has no force-on side effects because the user wanted the
video.
Removed: resolve_effective_timelapse, _resolve_effective_timelapse
wrapper, both background_dispatch call sites, the print_scheduler call
site, the archive.bambuddy_forced_timelapse write, _cleanup_forced_timelapse
(~75 lines including the FTP-DELE walk across /timelapse, /timelapse/video,
/record, /recording) and its call site. All paths now read
bool(item.timelapse) / bool(job.options.get("timelapse", False)) directly.
The archive.bambuddy_forced_timelapse DB column stays defined (default
False) for back-compat with existing rows — no consumer reads it anymore.
VP bridges bound to a target printer (Proxy mode, Queue mode with
specific target) forwarded the printer's raw AMS push_status to the
slicer untouched. bambu_mqtt.py::_handle_ams_data applies a
tray_exist_bits-driven cleanup to Bambuddy's internal state
(promote empty slots to state=9, wipe stale tray_type / tray_color /
tray_info_idx / tag_uid / tray_uuid / remain) so the AMS card renders
empty slots as Empty, but the VP bridge cache never ran the same
cleanup. Net result on real hardware: a printer with 3 loaded
filaments and several previously-loaded-now-empty slots had Bambuddy's
AMS card render those slots correctly as Empty, but BambuStudio after
Sync painted them as phantom loaded filaments with stale color and
material from before the slot went empty.
Root cause: two consumers of the same payload, only one wired to the
cleanup. _handle_ams_data ran it on every push; mqtt_bridge.py::
_on_printer_raw merged the ams blob via _merge_ams_dict but copied
tray_exist_bits through as an opaque scalar without acting on it.
Fix: factored the bit-clear logic out of _handle_ams_data into a
module-level helper apply_tray_exist_bits(units, tray_exist_bits_str,
*, power_on_flag, log_label). Internal path replaced with a single
call. Bridge calls it after _merge_ams_dict on the merged ams dict,
before the merged state is stored as the 1 Hz cached-as-base source.
Shared shutdown guard kept on both sides: all-zero bits +
power_on_flag=False is the printer-off pattern (#765, would
propagate phantom empties on every reconnect); nonzero bits +
power-off is valid idle-printer state (#1365, X1C between prints)
and still applies. AMS-HT units (id >= 128) skipped on both sides.
Tests: new TestApplyTrayExistBitsHelper (10 cases) pins the helper
contract directly. 3 new bridge regression tests reproduce the
#1726 wire shape, the shutdown guard, and the AMS-HT skip on the
cached slicer-facing state. Existing internal-state tests for the
bit-clear logic (covers state=9 promotion, loaded-slot preserve,
genuine-removal-with-power-on) continue to pass against the
refactored path.
One pre-existing bridge fixture had an inconsistent tray_exist_bits
('3' for 2 AMS units each with slot 0 loaded — bit 4 missing). The
shared cleanup exposed it; corrected to '11' (bits 0 + 4) to match
real-printer wire shape.
Reported by @needo37 with full code-level analysis including the
suggested fix shape and the BAMBUDDY_VP_DUMP_WIRE diagnostic to
verify on a live system.
The Windows installer's embedded Python doesn't carry an IANA tz
database, and the stdlib zoneinfo has no system DB to read on Windows.
ZoneInfo("UTC") raises ZoneInfoNotFoundError on those installs, and
the new /api/local-backup/status endpoint 500s on the resulting
uncaught exception. Surfaced via a Windows traceback from a user's log:
File "...\backend\app\services\local_backup.py", line 32, in _local_zone
return ZoneInfo("UTC")
zoneinfo._common.ZoneInfoNotFoundError: 'No time zone found with key UTC'
_local_zone()'s try/except only covered the TZ-env branch — both
fallbacks unconditionally called ZoneInfo("UTC") and re-raised.
Fix (two parts):
1. services/local_backup.py — return type widened from ZoneInfo to
tzinfo, the UTC fallback is wrapped in its own try, and the
last-resort fallback returns datetime.timezone.utc (stdlib, no
IANA DB needed). str(timezone.utc) == "UTC" so the response shape
on /api/local-backup/status is unchanged. The astimezone call in
_calculate_next_run accepts any tzinfo — no other call sites
affected.
2. requirements.txt — pin tzdata>=2024.1; sys_platform == "win32" so
the next Windows installer build ships the IANA DB, and any non-
UTC TZ value (e.g. Europe/Berlin) resolves correctly. The stdlib
fallback can only ever give UTC. Linux/macOS unaffected by the
platform marker — they already have the system tz database.
The Re-print and Schedule modal toggles for Flow Calibration and Nozzle
Offset Calibration accepted "off" correctly and flowed it through to the
project_file MQTT publish — Bambuddy sent extrude_cali_flag: 2 and
nozzle_offset_cali: 2 per the "1 = run, 2 = skip" reading inherited from
the #1478 / #1682 work. Live test on H2D 01.x: with both toggles off,
the stg queue still scheduled stage 8 ("Calibrating dynamic flow") and
stage 39 ("Nozzle offset calibration"), and the printer ran both at
print start.
Root cause: 2 means "skip the explicit pass but still apply / verify
stored PA via the calibration stage" — close to a no-op K-factor wise
but the per-print physical sequence still runs. 0 is the encoding that
actually drops the stage from stg. A BambuStudio Send-dialog capture
on the same firmware showed 0 for both fields when the user unchecked
the calibrations — contradicting the #1478 commit's read of "BambuStudio
never sends 0."
Fix:
- extrude_cali_flag = 1 if flow_cali else 0 (was: else 2)
- nozzle_offset_cali = 1 if (nozzle_offset_cali and is_dual_nozzle) else 0
(was: else 2)
Dual-nozzle gate stays; single-nozzle prints continue to force-skip the
nozzle-offset cali their head doesn't support (#1682). The 1 (run)
branch is unchanged.
Verified live on the same H2D after the patch: stg dropped to
[29, 13, 4, 14, 3] (cooling, homing, filament change, nozzle cleaning,
vibration comp). Stages 8 and 39 gone.
Vibration compensation is NOT fixed by this commit: vibration_cali is a
bool in our and BambuStudio's wire format, and the H2D firmware queues
stage 3 regardless of the false value. Firmware-side, not solvable at
the dispatch layer with the current field. Filed as a follow-up.
Two warnings polluting every A1 support bundle on healthy prints, both
unrelated to the timelapse-default behaviour the issue actually reports.
1. mqtt_bridge.py's post-bind nudge calls request_status_update on the
real printer's MQTT client to populate the bridge cache without
waiting for the next periodic pushall. The bind frequently races the
TLS handshake, especially on A1 firmware. Skip the nudge when
state.connected is False — the periodic pushall fills the cache
anyway. The WARNING in bambu_mqtt.py stays for the genuinely-
actionable callers (refresh-status API, bug reporter).
2. Post-finish SD-card cleanup (and the symmetric forced-timelapse dir
walk) used delete_file_async's bool return to drive a WARNING when
all candidates failed. A1 firmware self-cleans the SD card before
our cleanup runs — every candidate FTP-DELE returns 550, we burn
the retry budget, then WARN on a successful print. Introduce
DeleteResult.{DELETED,NOT_FOUND,FAILED} so the helpers only WARN
on real network/auth/transient failures. NOT_FOUND advances to the
next candidate without consuming the 2s backoff. User-facing delete
endpoint returns 404 on NOT_FOUND.
The support bundle shipped support-info.json + bambuddy.log, but the raw
shape of the printer's MQTT push_status — the field that blocks per-model
work like AMS Backup detection (deferred in 85fbd7fc) and every vt_tray /
vir_slot / mapping shape regression — was never captured.
Each connected printer now contributes push-status/printer-{i}.json with
{model, firmware_version, captured_at, raw_data}, indexed against
support-info.json["printers"]. Two-pass redaction: a structural walk
drops user-private keys (subtask_name, gcode_file, subtask_id, task_id,
project_id, design_id, profile_id, model_id, gcode_state,
gcode_file_prepare_percent) and rewrites net.info[*].ip to 0.0.0.0
(matches the #1429 VP bridge fix); then the JSON runs through the same
DB-derived sensitive_strings sanitizer the log path uses, catching any
printer name / serial / access code / cloud email that leaked into a
nested string field.
print.cfg, print.option, ams, vt_tray, vir_slot, mapping,
ams_extruder_map, and hardware fields are all preserved — those are the
fields per-model work needs.
Always-on inside the existing debug-logging-required gate; no opt-in
toggle (the bundle is already user-initiated and downloads locally
before the user chooses to send).
Right after the slicer picks a filament for the external spool (vt_tray, ams_id=255),
Bambu firmware pushes a partial vt_tray carrying just {tray_info_idx, tray_color} -
~18 fields shorter than the pushall shape the slicer expects. The #1622 round-4
per-field accumulate (da799447) only carried over prev keys NOT in new, so the
cached vt_tray was replaced wholesale with the 2-field partial. The next 1 Hz
cached-as-base push delivered the stripped dict and BambuStudio rendered the
external slot as invalid (color only, no tray_type / state / k / n / cali_idx /
nozzle_temp_*). Reload restored it because the reconnect-triggered pushall
re-seeded vt_tray, then the cycle repeated. AMS slots didn't suffer because
_merge_ams_dict deep-merged them.
Fix: for every top-level push_status key whose prev AND new are both dicts,
overlay incoming keys onto prev rather than replace. ams is excluded (already
deep-merged). The same shape protects device / online / upgrade_state / ipcam /
upload / net against future firmware partials. net.info IP rewrite is unaffected -
_rewrite_net_info_ips runs before caching and overlay lets the freshly-rewritten
list win over prev when present.
GitGuardian still flagged the file after the previous round even though
every call site used a constant — the constant itself was a static
string built by concatenation, which the generic-password detector still
matched on. Generate the test credential per process via secrets.token_urlsafe
so no password literal lives in the source, and mark the single line where
the variable is bound with the standard `pragma: allowlist secret` marker
ggshield / detect-secrets honour.
GitGuardian flagged the seven hard-coded passwords used by the
privilege-escalation regression suite as potential secrets. They are
test-only credentials whose value is irrelevant — the suite asserts
the admin authorization gate, not password handling — but the pattern
matches the high-confidence detector.
Replace each call-site literal with a single _FIXTURE_PW module
constant, built from string concatenation so it doesn't hash to a
recognisable token, with a comment explaining the purpose and the
complexity rule it satisfies. No behavioural change; all 11 tests
still pass.
slice_and_persist writes a .gcode.3mf ZIP container but persisted the row
with file_type="gcode". The G-code preview endpoint short-circuits on
file_type == "gcode" and returns the bytes as text/plain, so the embedded
viewer received the raw ZIP body instead of the embedded toolpath.
- Persist file_type="gcode.3mf" on sliced rows (matches _classify_file_type
and external-scan rows).
- get_gcode also routes to the unzip branch when the filename ends with
.gcode.3mf, so rows already written under the bug self-heal on first
preview without a DB migration.
- Extend FileManagerPage badge + viewer-eye gate and ProjectDetailPage badge
to accept "gcode.3mf"; isSlicedFilename / isSliceableFilename already do.
- Add test_library_get_gcode_recovers_legacy_gcode_type_for_3mf: legacy
row preview must be text/plain, contain G28, and NOT start with PK.
Bundle import never delivered what it implied: BambuStudio's .bbscfg export
strips system processes/filaments, so importing a bundle left users without
process presets and slicing fell back to embedded settings on STL. Bundle
mode also hid the standard tier behind a constrained dropdown, the actual
trap reported here.
Removed end-to-end:
- backend: POST/GET/DELETE /slicer/bundles*, SliceRequest.bundle,
SliceBundleSpec, dispatch fork in library.py, bundle-context params on
the filament-requirements endpoints, bundle-fingerprint cache key in
slice_preview.py, SlicerApiService.{import,list,get,delete}_bundle and
slice_with_bundle, BundleSummary / BundleNotFoundError.
- frontend: BundlePicker + BundleStringDropdown, isBundleMode + every
branch, bundle state/queries/dispatch in SliceModal.tsx, SlicerBundle /
SliceBundleSpec types, three bundle API methods. buildCompatibilityIndex
loses its bundle path; presetCompatibility keeps compatible_printers
plus the @BBL fallback.
- SlicerBundlesPanel turns into a permanent static notice explaining the
removal, alternative import paths, and the new slice-time lookup order
(Imported > Orca Cloud > Bambu Cloud > Standard sidecar fallback).
- i18n: slicerBundlesRemoved.{title,description,alternatives,lookupOrder}
translated across all 11 locales; slice.bundle*, slicerBundles.* keys
removed.
Fixed (surfaced by removing bundle mode):
- _resolve_cloud and _resolve_orca_cloud now force type per slot and pin
from: "system" on the payload before json.dumps. Bambu Cloud ships
type as "printer"/"print" and routinely empty `from`; the BS CLI's
--load-settings parser rejects both with return -5 / "input preset
file invalid". Standard tier already did this; cloud paths now match.
Bridge cache replaced prev state wholesale on each incremental, re-merging
only a 14-key allowlist. Capability/lifecycle fields (cali_version,
print_type, mc_print_stage, device, ...) drained out within one 1Hz tick,
greying out BambuStudio's Device-tab UIs (manage-calibration, AMS-slot
dropdown) once the cache thinned. Most P1S users miss it by timing — they
click Device tab while the cache is still fat from the connect pushall.
Switch to per-field accumulate matching bambu_mqtt.py's internal state
handler: prev keys carry over verbatim when not present in the incoming
push, new values overwrite when present. _merge_ams_dict for partial AMS
blobs unchanged (#1387 / #1371 regression guards stay green).
_SLICER_VISIBLE_STICKY_KEYS removed — new logic is a strict superset.
Round-2 cmd.jsonl from shaddowlink proves the bridge forwards both commands and
responses correctly: ams_filament_setting round-trips with result=success on P1S,
the cached push_status carries tray_info_idx=GFA11/tray_type=PLA-AERO/K-n/cali_idx
intact, and the visible "unload" symptom comes from the slicer's choice of
extrusion_cali_set (push K direct, P1S firmware rejects) vs extrusion_cali_sel
(select by id, both H2D and P1S accept). The open question is what makes the
slicer pick _set vs _sel — likely the info.get_version response Bambuddy
synthesises or the first cached pushall reply the slicer reads at connect.
Round 2 captured neither; the JSONL had slicer_to_bridge and printer_to_slicer
but no direction for the bridge's own synthesised replies.
Same BAMBUDDY_VP_DUMP_WIRE=1 flag now also appends a bridge_to_slicer line for
every bridge-synthesised reply (info.get_version answer, project_file ack,
on-demand pushall response). Capture is in _publish_to_report — the single
chokepoint — gated on a new log_event param; the 1Hz periodic push threads
log_event=False so the JSONL isn't flooded (~60 lines/min/VP) because
dump_wire already covers cache shape per tick.
Diagnostic-only, no data-path change. Default param preserves every existing
call site's behaviour.
Round 2 fixed the model-mode FilamentOverride: tray_info_idx →
sub-brand, plus a material-disambiguated colour name from a new
/inventory/colors/by-material endpoint. The printer-mode panel that
renders the same 3MF (FilamentMapping) was reading the same raw
fields — item.type for the required label, getColorName(item.color)
for the swatch tooltip — and was not touched, so picking "Specific
Printer" still showed "Required: PLA - Black" for a slice the
"Any H2D" branch already labelled "Bambu PLA Matte - Charcoal".
Extract the three-query resolution machinery from FilamentOverride
into a shared hook useFilamentLabels (returns positional
{resolvedName, colorLabel} per slot). Both panels call it; both
read the same labels. The hook also owns extractMaterialHint so the
"strip leading brand token" rule has one source of truth.
FilamentMapping required-side now reads {resolvedName} instead of
{item.type}; swatch tooltip reads `Required: {resolvedName} -
{colorLabel}` instead of `Required: {item.type} -
getColorName(item.color)`.
The shape-of-payload dump shipped earlier rules out cache wipes —
shaddowlink's round-1 captures show AMS data reaches the slicer
byte-identical to what the printer sent. The remaining symptom
(picking a generic filament in archive mode "unloads" the slot) lives
on the command path, which the snapshot dump doesn't see: it writes
only the cached _latest_print_state and the periodic 1Hz push.
Add append_event() in _debug.py — same env flag, separate file at
<log_dir>/vp_wire/<vp>_cmd.jsonl. One JSONL line per event with UTC
iso timestamp, direction (slicer_to_bridge / printer_to_slicer), MQTT
topic, <channel>.<command> grep handle, and parsed payload. Wired at
two points: mqtt_server._handle_publish for slicer publishes (after
JSON decode so the trace matches what the bridge actually parsed) and
mqtt_bridge._on_printer_raw "everything else" branch for printer
responses (after serial rewrite so the trace matches what the slicer
sees on the wire). Pushall / get_version stay out — both are handled
locally and never round-trip through the bridge.
Bytes payloads get the same \x00-tolerance fix from #927 so
OrcaSlicer's C-string-null publishes parse cleanly; un-parseable
bytes fall back to {"raw": "..."} so every line stays valid JSON.
Native installs that follow the systemd template
WorkingDirectory=/opt/bambuddy
Environment="DATA_DIR=/srv/bambuddy/data"
(or any layout where DATA_DIR is not a subdirectory of the install)
could not apply in-app updates. Every git subprocess in _perform_update
used cwd=settings.base_dir and safe.directory={base_dir}. On standard
installs (DATA_DIR=INSTALL_PATH/data) this happened to work by accident
because git walks up from a subdirectory of the repo to find .git; on
separate-mount layouts the walk has nowhere to go and every call
returns "fatal: not a git repository." safe.directory was also wrong
even on the standard install -- it must equal the repo root git
discovers, not the data dir.
Resolve app_dir = settings.app_dir at the top of _perform_update and
route all four git subprocesses (remote get-url, remote set-url, fetch,
reset --hard) and the embedded safe.directory through it. Rename the
base_dir parameter on _origin_points_at_repo to app_dir so the
signature documents the contract.
Four #1712 issues from the 2026-06-04 Orca Cloud integration:
(1) Tier order put Orca Cloud above everything across SliceModal,
auto-pick scoring, dropdown groups, the AMS slot picker, and the
backend precedence. Bambu-Cloud-only users saw their profiles
deprioritised behind an empty Orca tier.
(2) Cross-tier dedup hid a same-named preset in all but the highest-
priority tier. A user with both a local-imported and an Orca-synced
"Bambu PLA Basic" couldn't see the Orca copy as a picker option.
(3) CloudStatusBanner nagged signed-out users with a permanent
"Sign in to Orca Cloud" line at the top of every slice -- even after
explicit logout. Bambu Cloud had the symmetric problem.
(4) ConfigureAmsSlotModal source badges were inconsistent: Orca rows
showed only "Custom" (no source identity), Bambu Cloud built-in rows
had no badge at all, and the orthogonal isUser-driven "Custom" badge
collided with the source badge for cloud user presets.
Order is local > orca_cloud > cloud > standard everywhere it lives
(SliceModal SLICE_MODAL_TIER_ORDER + TIER_BONUS + dropdown tier list,
ConfigureAmsSlotModal sourceOrder, and backend precedence). The order
drives auto-pick + visual group rendering; it does NOT hide profiles.
_dedupe_by_name replaced with _enrich_cloud_metadata: every tier
returns its full list across all three slots (printer / process /
filament). The function still backfills Bambu Cloud filament metadata
from same-named local / orca_cloud / standard entries so cloud
filaments score in pickFilamentForSlot.
CloudStatusBanner silently no-ops on not_authenticated for both clouds;
expired / unreachable still surface. The not_authenticated i18n keys
stay in the locale files dormant.
ConfigureAmsSlotModal: one source badge per row, one colour per source
(green Local / purple Orca Cloud / bambu-blue Bambu Cloud / amber
Built-in). The legacy isUser-driven "Custom" badge is gone; every row
identifies its tier consistently.
Add a BAMBUDDY_VP_DUMP_WIRE=1 escape hatch that writes the bridge's
cached push_status (in) and the 1Hz slicer-facing copy (out) to
<log_dir>/vp_wire/<vp_name>_<direction>.json, overwritten each tick.
#1622's symptom — empty filament dropdown in slicer's AMS slot details
for P1S/A1 but not H2D in non-proxy VP modes — needs visibility into
the actual wire bytes flowing through the bridge to bisect between
"cache is missing fields" and "_send_status_report strips them on copy."
The existing logs prove the bridge is bound and pushing at 1Hz, but
not what's in the payload.
Off by default, single env flag, single file per VP per direction
(bounded disk footprint), failures swallowed at debug so a broken
dump can never break the 1Hz loop. 21 tests pin the helper contract:
disabled-by-default, atomic writes, sanitized vp_name (no path
escape), per-call env check so toggling without restart works.
Reprints reuse the source archive row via the expected-print promotion
branch, but that branch never reset archive.timelapse_path. The stale
path made _scan_for_timelapse_with_retries early-return (so the new
run's MP4 was never downloaded), and _capture_finish_photo_from_timelapse
then extracted the *original* run's last frame and shipped it to Telegram
as the new run's finish photo. Surface was specific to the
timelapse-prefer path (data.timelapse_was_active=true and no external
camera) — fallback live-camera paths were unaffected, which is why this
took a P2S reporter to surface.
Clear archive.timelapse_path at promotion and unlink the stale on-disk
MP4 (best-effort; missing files are logged and skipped). The orphan
unlink also kills a long-standing file-leak: every reprint used to leave
its predecessor's timelapse on disk forever. archive.photos is
deliberately left alone — accumulating one finish photo per run is
correct.
Brings the Windows installer work from dev to main without merging
the rest of the 0.2.5b1 release content. Squashes 12 commits from
dev (8711c54e..7bb11df2) into a single net-effect commit on main.
Includes:
- installers/windows/ — Inno Setup .iss script, build.py, vendored
NSSM 2.24, bambuddy.ico (multi-resolution app icon), service
install/uninstall .bat files, build pipeline README
- backend/app/services/network_utils.py — Windows psutil branch so
the VP bind-IP dropdown enumerates interfaces; Linux/macOS path
unchanged
- .github/workflows/windows-installer.yml — reconciles main's
kludge-pushed copy with dev's accumulated changes (NSSM
vendoring, version-from-tag, unversioned alias step, etc.)
CHANGELOG and README entries for the Windows installer stay on
dev — they reference unreleased 0.2.5b1 release notes that aren't
on main yet.
get_network_interfaces() used fcntl ioctls and get_all_interface_ips()
shelled out to `ip -j addr show` — both Linux-only. On Windows the
fcntl path raised ImportError and returned [], so the VP "bind IP"
dropdown showed no interfaces.
Add a Windows branch using psutil.net_if_addrs() + net_if_stats()
(psutil is already a dep). Same dict shape as the Linux path, filters
loopback / link-local / down interfaces by address class instead of
by name prefix. No name-based exclusion — users may legitimately
bind a VP to a Hyper-V / WSL / Tailscale virtual adapter.
The Linux fcntl path is untouched.
Internal code N9 (from BambuStudio resources/profiles/BBL/machine/Bambu Lab A2L.json),
serial prefix 26A19 (5-char, same shape as H2C's late 31B8B). Capabilities from
Bambu's official A2L specs page: linear rail, single FDM extruder + integrated
cutter/plotter, no Ethernet (2.4 GHz Wi-Fi only), low-rate chamber camera on
port 6000.
The BambuStudio profile's use_double_extruder_default_texture: true flag
describes two TOOL HEADS (FDM + cutter), not dual filament extrusion — A2L
must NOT be classified as dual-nozzle or AMS routing will target the deputy
slot and firmware rejects with 07FF_8012.
Registry updates: printer_models.py, firmware_check.py, virtual_printer/manager.py,
virtual_printer/mqtt_server.py, PrintersPage.tsx, SpoolBuddyAmsPage.tsx.
12 new test cases in TestA2LModel pin every dimension.
A1 and A1 Mini ship without a MicroSD slot at all - there is no
firmware-side "Store sent files on external storage" toggle and the
slicers don't surface a slicer-side equivalent either. The connection
diagnostic was reading state.store_to_sdcard (home_flag bit 11), which
is never set on these models, so the check fell through to fail for
every A1-series user. Combined with the absent slicer UI it left users
thinking Bambuddy was wrong about a setting their hardware does not
have.
New NO_EXTERNAL_STORAGE_MODELS frozenset in utils/printer_models.py
enumerates A1, A1 Mini, and their internal codes (N1, N2S, A04, A11,
A12). has_external_storage() returns False for those, True for
everything else. Unknown models default to True so the check stays
active for future Bambu lineup additions - new no-slot models must be
added to the set explicitly.
The diagnostic now short-circuits to skip before reading
store_to_sdcard when printer.model is in the set. X1, P1, P2S, H2,
and X2D are unchanged - the bit-off -> fail signal is still the right
read for them.
The companion FTP-upload-timeout symptom in the same bug report (ftp
code 28 from BambuStudio when sending to the proxy VP) is a separate
Docker-bridge-mode networking constraint, not addressed here.
The slot card on PrintersPage shows slot_preset_mappings.preset_name
first in its display fallback chain. Three write paths swap which
spool occupies a given slot:
- internal manual assign (inventory.apply_spool_to_slot_via_mqtt)
- internal RFID auto-assign (spool_tag_matcher.auto_assign_spool)
- Spoolman RFID sync (main.auto_sync_spoolman_ams_trays)
Only the first one was reconciling the row. After an RFID-driven
spool change, the card kept surfacing the previous spool's preset
name until the user opened Configure Slot manually.
Reporter saw H2D-1 / AMS-B3 displaying "Bambu PLA Silk+" for a
freshly-inserted Bambu PLA-CF spool. The matching row in
slot_preset_mappings was last written in March when a PLA Silk+
spool had been in that slot - confirmed live in the database.
New backend/app/services/slot_preset_writer.py exposes a primitive
upsert_slot_preset plus two derivation wrappers: one for the
internal Spool ORM object, one for the Spoolman API dict shape.
All three call sites now go through the helper, so the row stays
in lockstep with the assigned spool regardless of inventory mode.
Bug shape exists in both inventory modes and the patch fixes both
per feedback_inventory_modes_parity. The Spoolman path was latent
for users who'd never manually picked a slot preset; the same
"stale row overrides correct catalog name" symptom appeared for
those who had.
Existing stale rows self-heal on the next RFID-driven swap.
datetime.fromtimestamp(ts) and datetime.now() return naive local
datetimes; .isoformat() then emits no tz marker. The frontend's
parseUTCDate helper appends 'Z' to bare strings, treats the value
as UTC, then converts to local for display — applying the local
offset twice. Reporter on UTC+3 saw boot_time +3h ahead while
uptime was correct (uptime is a backend-side delta of two
naive-local values, so the missing tz info cancels out).
Fix: pass tz=timezone.utc to datetime.fromtimestamp and
datetime.now in system.py's boot_time / uptime path, plus the two
adjacent generated_at sites in system.py and support.py.
Logs device.dev_model_name / dev_product_name / dev_id / project_name
at INFO level once per client session, falling back to device.keys()
if none of the known fields are present.
The MQTT push_status carries the model code in device.dev_model_name
on every message, but nothing in bambu_mqtt.py reads or logs that
field — so adding a new printer model meant chasing the code through
either Bambu cloud or a manual mosquitto_sub. A2L (#1684) was the
case that surfaced this: get_version also failed because the firmware
disconnected right after request topic subscription, so the support
bundle had no way to disclose the model.
INFO level so the line lands in support bundles without enabling
debug. One-shot via _device_id_logged, mirroring the existing
_nozzle_fields_logged flag at line 2095, so push_status spam is
avoided.
Future-proofs against Bambu renaming the field (the fallback dumps
device.keys() so a rename like model_name without the dev_ prefix
is still observable). 3 unit tests in TestDeviceIdentificationProbe
pin all three branches.
N1 and N2S were flipped relative to every other registry that names
them - firmware_check.py (N2S -> "a1"), virtual_printer/manager.py
(both the model map and the serial-prefix map: N2S -> 039 = A1,
N1 -> 030 = A1 Mini), and printer_manager.py A1_MODELS all agree on
N2S = A1, N1 = A1 Mini. Only printer_models.py had it backwards, so
any path that resolved an A1-family printer by internal code rather
than serial prefix would silently misclassify.
Also fixes the matching comments in LINEAR_RAIL_MODELS - cosmetic only
(both codes were already in the frozenset) but kept the file
self-consistent.
New TestA1SeriesModelIds regression test pins both directions so a
future re-flip fails loudly.
close_all_connections() only disposes the engine's connection pool —
asyncio tasks like print_scheduler.run() and the smart-plug snapshot
loop wake on their 30 s cadence and lazily reopen pool connections
holding RowExclusiveLock on print_queue / smart_plug_energy_snapshots.
The restore's DROP TABLE ... CASCADE pass needs AccessExclusiveLock on
every public table, producing an AB/BA deadlock that rolls back the
entire restore transaction.
Reproduced 2026-06-09 restoring a native install's backup into a fresh
Docker+Postgres deploy:
asyncpg.exceptions.DeadlockDetectedError: deadlock detected
Process X waits for AccessExclusiveLock on relation 109940
Process Y waits for RowExclusiveLock on relation 110182
Fix:
- Layer 1: pause print_scheduler / smart_plug_manager /
notification_service / background_dispatch via their existing stop
affordances before close_all_connections(), with a 1.0 s sleep for
in-flight loop iterations to release sessions. Restore handler
already requires a container restart on success, so the paused
services come back via the next lifespan startup.
- Layer 2: prepend SET LOCAL lock_timeout = '10s' to the begin-block
in _import_sqlite_to_postgres so any reactive writer (per-printer
MQTT, hourly AMS history recorder) that slips through the pause
window fails fast and visibly instead of producing a new deadlock.
close_all_connections() only disposes the engine's connection pool —
asyncio tasks like print_scheduler.run() and the smart-plug snapshot
loop wake on their 30 s cadence and lazily reopen pool connections
holding RowExclusiveLock on print_queue / smart_plug_energy_snapshots.
The restore's DROP TABLE ... CASCADE pass needs AccessExclusiveLock on
every public table, producing an AB/BA deadlock that rolls back the
entire restore transaction.
Reproduced 2026-06-09 restoring a native install's backup into a fresh
Docker+Postgres deploy:
asyncpg.exceptions.DeadlockDetectedError: deadlock detected
Process X waits for AccessExclusiveLock on relation 109940
Process Y waits for RowExclusiveLock on relation 110182
Fix:
- Layer 1: pause print_scheduler / smart_plug_manager /
notification_service / background_dispatch via their existing stop
affordances before close_all_connections(), with a 1.0 s sleep for
in-flight loop iterations to release sessions. Restore handler
already requires a container restart on success, so the paused
services come back via the next lifespan startup.
- Layer 2: prepend SET LOCAL lock_timeout = '10s' to the begin-block
in _import_sqlite_to_postgres so any reactive writer (per-printer
MQTT, hourly AMS history recorder) that slips through the pause
window fails fast and visibly instead of producing a new deadlock.
Two adversarial-input fixtures added on the 0.2.4.6 branch were
missing the # nosec annotation that 32b3a93e established for the
same pattern. New TestNotArmedDiagnosticLogging test for the #1429
defensive diagnostic uses bind_address="0.0.0.0"; new
test_queue_start_user_attribution.py (#1670 fix) uses a /tmp path
in a PrintArchive fixture. Same annotation-only convention as the
test_virtual_printer.py sites.
Two complementary surfaces for the most-missed install step ("Store sent
files on external storage"):
1. Connection diagnostic check (printer-side variant)
- Reads state.store_to_sdcard, parsed from MQTT home_flag bit 11.
- Pass / fail / skip; instant, no I/O.
- Catches the newer-firmware variant where the toggle moved onto the
printer itself (P2S 01.02 / Studio 2.6+).
An FTP upload-and-verify probe was tried first and rejected. /cache
is always writable from Bambuddy regardless of the slicer setting;
only BambuStudio's own behaviour changes when the toggle flips.
Empirically confirmed against X1C + H2D with the slicer option
toggled off: probe still succeeded, home_flag bit 11 stayed True.
2. Archives-page banner (slicer-side variant)
- The slicer-side toggle is invisible to the printer — older
BambuStudio doesn't push the change to the printer. The diagnostic
can't see it.
- Symptom is deterministic: archiver creates rows with
extra_data.no_3mf_available=True (main.py:2770) when it can't pull
the 3MF from /cache after a slicer-initiated print.
- New endpoint GET /archives/no-3mf-warning returns whether any
archive in the last 30 days has the flag (excluding soft-deleted).
- Amber dismissible banner at the top of /archives; one-shot
localStorage dismissal (matches Layout.tsx update-banner pattern,
but persistent across sessions).
- React-Query disabled after dismissal so the endpoint isn't polled
once the user has been told.
Detects the printer-side variant of install step 4 — many users (esp. on
clean installs) forget to enable this and only notice when their archive
cards have no thumbnails. The diagnostic now catches it upfront.
Detection: read state.store_to_sdcard, which Bambuddy already parses from
MQTT push_status home_flag bit 11 (bambu_mqtt.py:153). Instant, no I/O.
An FTP upload-and-verify probe was tried first and rejected. /cache is
always writable from Bambuddy regardless of the slicer setting — only
BambuStudio's own behaviour changes when the toggle flips, not the
printer's acceptance policy. Confirmed empirically against X1C + H2D
with the slicer option toggled off: probe succeeded, home_flag bit 11
stayed True. So the only reliable signal is what the printer actually
reports about its own state.
Limitation: the printer-side variant only exists on newer firmware
(P2S 01.02 / Bambu Studio 2.6+). On older versions the toggle lives
only in the slicer and the printer never hears about it, so this check
will pass even when the user is missing step 4 in BambuStudio. The
skip-text and the wiki call this out explicitly. A reactive banner on
the no-3MF archive-fallback path is planned as a follow-up to cover
that case.
Statuses:
- pass: state.store_to_sdcard is True
- fail: state.store_to_sdcard is False (-> overall escalates to problems)
- skip: no live state, disconnected, or field never populated
(#1687 part 4, reported by @IndividualGhost1905)
Reporter clarified after part 1 shipped that point 2 wasn't about
archive `tags` (which describe the model — home decor, toys), but
about failure-cause classification on the *log* row itself:
spaghetti, jam, bed-adhesion, etc. Different surface, different
lifetime.
The data field he wanted already existed. PrintLogEntry.failure_reason
is a String(100); the Failure Analysis widget already groups by it;
the Archive Edit modal already mirrors archive.failure_reason into
the most recent log entry (archives.py:1421, shipped with #1444).
The only gaps were:
1. The GET endpoint silently dropped failure_reason (and archive_id
and created_by_id) from PrintLogEntrySchema construction even
when set in the DB — so the Print Log table couldn't render what
the Failure Analysis widget grouped by. Fixed independently of
the editor; regression test added.
2. Orphan log entries (no archive — dispatch errors, aborts before
archive creation, manual entries) had no edit path at all because
the Archive Edit modal cannot reach them. The new endpoint is
the only way to classify those rows.
Changes:
- Backend: new PATCH /print-log/{entry_id} taking
{failure_reason, status}, gated on require_ownership_permission(
ARCHIVES_UPDATE_ALL, ARCHIVES_UPDATE_OWN) — same ownership shape
as the per-row DELETE. Validates against the same 11-key failure
vocabulary and 5-key status set the Archive Edit modal uses;
unknown values return 400 rather than getting stored as raw text
(the i18n layer maps the value back through the vocabulary,
unrecognised values would render as literal strings).
Empty-string failure_reason stores back as NULL so the column's
nullable=True intent is preserved end-to-end. GET endpoint now
surfaces failure_reason, archive_id, created_by_id.
- Frontend: FAILURE_REASON_KEYS moved to an export from
EditArchiveModal.tsx so the new editor reuses the exact same
vocabulary — backend and frontend stay in lockstep. Pencil icon
beside the existing trash icon on every Print Log row, opens a
compact two-field modal (status + failure reason). Save
invalidates print-log and archives-stats query keys so the
Failure Analysis widget reflects the re-classification on the
same response cycle. Failure reason rendered as a sub-label under
the status badge, matching PrintLogTable.tsx's convention.
- i18n: 10 new keys (editEntryTitle, editEntryDescription,
entryUpdated, entryUpdateFailed, archives.permission.noEdit, plus
a 5-key statuses block) translated across all 11 locales. No
English fallbacks.
- Wiki: features/print-log.md gains per-row actions section,
updated permissions table, PATCH/single-DELETE endpoint docs.
When the user clicked Print Anyway on a filament-deficit warning, the
acknowledgement was one-shot. The route cleared manual_start and
filament_short, then the next scheduler tick re-ran
compute_deficit_for_queue_item against identical spool state, found
the same deficit, and re-set both flags. The item bounced between
"user said anyway" and "scheduler re-blocked" — every Play click
returned 409, every confirm got rolled back on the next tick.
Add a persistent acknowledgement flag on the queue item:
- New column `skip_filament_check` on print_queue. SQLite + Postgres
migration branched on is_sqlite() so Postgres doesn't reject
DEFAULT 0 on BOOLEAN.
- PrintQueueItemCreate + PrintQueueItemResponse schemas + the
TypeScript types carry the field.
- POST /print-queue/{id}/start with skip_filament_check=true now
ALSO sets item.skip_filament_check = True (not just clearing
manual_start / filament_short).
- PrintScheduler._block_on_filament_deficit short-circuits to
False — no compute, no flag-setting, no notification — when
item.skip_filament_check is True. We trust the operator's
decision and stop fighting them.
- PrintModal at queue-creation time threads
skip_filament_check=true into the create payload when the user
clicks Print Anyway on the frontend deficit warning, so a print
that was warned-then-acknowledged at add-to-queue time goes in
pre-acknowledged — scheduler never blocks it on first tick.
Flag is not auto-cleared on spool swap by design: if remaining is
now sufficient, the check returns no deficit anyway, so the flag
is moot. Auto-clearing would add lifecycle complexity without
changing behaviour.
AMS Backup awareness (the other half of the discussion) intentionally
NOT included — verified the H2D's bit-26 of print.cfg toggles with
the printer-side AMS Backup setting, but the X1C's cfg has a
different shape entirely and verifying every model family isn't
realistic. Silently under-warning would be worse than always
per-slot. The check stays single-slot for now.
When a print targets a single plate from a multi-plate 3MF, both the
internal Filament Inventory tracker and the Spoolman-mode tracker parsed
the 3MF without a plate filter and summed every plate's filament — so a
single lid print debited the spool the entire file's grey + black totals.
The 3MF parser already supports plate_id (queue pre-flight uses it at
print_queue.py:254/:286). Plumbed it through both dispatch paths:
Queue path:
- PrintSession gains a plate_id field; on_print_start queries the
printer's currently-printing queue row and records queue_item.plate_id
onto the session.
- _track_from_3mf accepts plate_id and passes it to the extractor.
- store_print_data moves its existing queue-item lookup above the
extract and uses queue_item.plate_id as the plate filter.
Direct-Print path (reprintArchive / printLibraryFile — never goes
through the queue):
- _print_plate_ids dict added in main.py, parallel to _print_ams_mappings.
- register_expected_print accepts plate_id and stores it; the 2 sites in
background_dispatch.py and the 1 site in print_scheduler.py now pass
it (resolve was already happening, just needed reordering before the
register call so the value is available).
- Expected-print promotion in main.py injects _print_plate_ids[archive_id]
into the session, guarded so a queue capture wins over the dict.
- _get_start_plate_id helper feeds plate_id into all 3
_store_spoolman_print_data call sites; spoolman_tracking.store_print_data
takes the caller value first, falls back to queue_item.plate_id.
PrintArchive.filament_used_grams stays file-level summed by design
(#1593's contract — the archive describes the file, not the run); only
the per-run usage attribution becomes plate-aware. Single-plate direct
prints resolve to plate_id=1 → plate 1 = whole file, identical to the
prior no-filter behaviour.
maziggy