feat: Spoolman filament bookings + upload table in the web GUI

[VGottselig]

Books filament consumption to Spoolman straight out of the upload path, and gives the web GUI an upload table where every booking can be corrected by hand.

Fork feature — not intended for upstream. Deliberately kept apart from the clean PR branches fix-fixed-gcode-filesize (macdylan#35) and fix-persist-token-on-connect (macdylan#36).

Design decisions were settled beforehand in SPOOLMAN-INTEGRATION.md (1–20), the G-code format in GCODE-FILAMENT.md.

How it works

Every upload is parsed before SMFix, from the original, reading only the tail of the file. Each used filament slot becomes one ledger row in /data/uploads.yaml:

• Upload + start → booked immediately.
• Plain upload → recorded only, but the spool is created so the remaining column has something to show.
• No consumption block (Luban, .nc, laser/CNC) → a row without filament data, not bookable. Nothing is guessed from E moves.

Bookings are always deltas against booked_g, never absolute values. "Book" sets the booked amount to the G-code amount, "Cancel" sets 0, and any correction sends the difference (booked 120, corrected to 50 → use_weight: -70). There is no separate cancelled state: a row corrected back to 0 is indistinguishable from one never booked.

A missing filament is created automatically — vendor, filament and spool — carrying density, diameter, colour and price per kg from the G-code so Spoolman and Orca compute alike. The tare weight stays empty because the G-code does not know it.

Table

Time	File	Filament	G-code g	booked g	remaining g	Print time	Cost	Action
30 Jul 12:53	ElefantMobile_0.2mm_4h25m.gcode	🟪 DEEPLEE PLA PRO Rosa	46	[46]	954	4h 24m 54s	0.63 €	Book · Set · Cancel

The amount appears twice on purpose: the immutable G-code value on the left, what Spoolman actually holds on the right. Cost follows the booked value, so a correction moves it. Remaining weight below SPOOLMAN_LOW_G turns orange, below zero red and bold.

Spoolman API findings

Verified against a running v0.25.0 instance and its openapi.json. Four of them shaped the client:

Finding	Consequence
extra values are JSON-encoded strings whatever the field type ("\"hello\"")	the preset mapping is marshalled on write, unmarshalled on read
GET /spool has no filter for extra	the mapping match runs client-side over all non-archived spools
remaining_weight is clamped to >= 0, hiding an over-booked spool	remaining is computed here from initial_weight − used_weight, so it can go negative
the API chains no prices or net weights	the client forms spool.price → filament.price and spool.initial_weight → filament.weight itself

Negative use_weight works (use_weight_safe adds and clamps at 0) — but that clamp is silent, which is exactly why booked_g is tracked locally. PUT /use returns the updated spool, so the frozen remaining weight needs no extra GET.

Robustness

• Spoolman must never block a print. Booking happens after a successful upload; a failure only logs, marks the row and leaves booked_g alone so a retry sends the same delta. The answer to Orca is untouched.
• Atomic ledger writes (temp file, fsync, rename). LocalStorage.Save() truncates then writes; a crash mid-write would destroy a consumption ledger.
• A mutex around read-modify-write. http.Serve gives every request its own goroutine, so an Orca upload appending a row and a "book" click changing one really can overlap.
• Bookings use their own mutex, not the ledger's — a booking makes HTTP calls that may take seconds, and holding the ledger lock across them would stall the next upload.
• /book is POST only, so no link and no browser prefetch can change inventory (:8844 is LAN-internal without auth).

Configuration

environment:
- TZ=Europe/Berlin
- SPOOLMAN_URL=http://spoolman:8000   # empty/absent = bookings off, table still works
- SPOOLMAN_LOW_G=100                  # low-stock threshold in grams
networks:
- default
- home_net                            # without it "spoolman" does not resolve

Each has a flag too (-spoolman, -lowg); the ledger path comes from -uploads/UPLOADS_FILE and otherwise defaults next to -knownhosts, landing in the persistent /data volume. time/tzdata is embedded so Europe/Berlin works on any base image.

Tests

First tests in this repo — 49, all green under -race. Parser coverage comes from four real Snapmaker Orca 2.3.4 files that happen to cover all four slot indices:

Job	Slot	G-code	rounded	€/kg
Assembly	0	37.57 g	38	13.59
ElefantMobile 4h25	1	45.93 g	46	13.59
Ivana	2	1.98 g	2	11.99
ElefantMobile 2h14	3	18.84 g	19	13.59

Ivana proves the index parallelism of filament_cost. Also covered: semicolon inside a quoted preset name, both delimiter styles, the key:/key = duplicate, a block beyond the first tail window, missing consumption lines, atomic and concurrent ledger writes, and the Spoolman client against httptest including the double-encoded extra, negative deltas and the silent clamp.

testdata/ holds only the key lines of those files; the multi-MB originals stay out of the repo behind SM2_GCODE_SAMPLES.

Verified against the live stack

Deployed and exercised end to end: all four spools auto-created with correct density, diameter, colour, price and preset mapping; book → 46 g / remaining 954 / 0.63 €; correct to 20 g → delta −26 / 980 / 0.27 €; cancel → 0 g / 1000. Over-booking on purpose: Spoolman reports remaining_weight: 0.0 while the table shows −6 in red. Low-stock warning fired and did not block. GET /book → 405. Survives a container restart. No leak: RSS flat at ~21 MB over 300 page loads, MemStats.Alloc sawtooths between 0.8 and 3.2 MB.

Deviations from the pre-agreed decisions

• 20 (cost): no fetch per render but a 30 s cache with a 2 s timeout; a missing price or an unreachable Spoolman yields 0.00 instead of an empty cell.
• 4 (mapping): no server-side extra filter exists, so the match is client-side.
• 5 + 18: auto-creation moved from booking to upload — decision 18 wants an unbooked row to show the spool's current level, which requires the spool to exist. A plain upload still consumes nothing.
• GUI is English throughout, including the ledger's status values (open/booked/failed).

SPOOLMAN-INTEGRATION.md still describes the pre-implementation state; it has not been rewritten to match.

🤖 Generated with Claude Code

https://claude.ai/code/session_01QHt1eWHFiSbJaNjKVhU38N