bambuddy/backend/app/api/routes/print_queue.py

"""API routes for print queue management."""

import json
import logging
import zipfile
from datetime import datetime, timezone
from pathlib import Path
from typing import Literal

import defusedxml.ElementTree as ET
from fastapi import APIRouter, Depends, HTTPException, Query
from sqlalchemy import and_, case, func, inspect, or_, select, update
from sqlalchemy.exc import IntegrityError
from sqlalchemy.ext.asyncio import AsyncSession
from sqlalchemy.orm import aliased, selectinload

from backend.app.api.routes.library_variants import normalize_model_name, resolve_variant_model
from backend.app.core.auth import (
    ApiKeyActor,
    QueueReviewRequired,
    RequestActor,
    RequestPrinterScope,
    RequirePermissionIfAuthEnabled,
    RequirePrinterPermissionIfAuthEnabled,
    may_start_queue_item,
    require_ownership_permission,
)
from backend.app.core.config import settings
from backend.app.core.database import get_db
from backend.app.core.permissions import Permission
from backend.app.core.printer_scope import PrinterScope, ensure_model_target_allowed
from backend.app.models.archive import PrintArchive
from backend.app.models.library import LibraryFile
from backend.app.models.print_batch import PrintBatch, PrintBatchPlate
from backend.app.models.print_queue import PrintQueueItem, PrintQueueVariant
from backend.app.models.printer import Printer
from backend.app.models.project import Project
from backend.app.models.user import User
from backend.app.schemas.print_queue import (
    PrintBatchCreate,
    PrintBatchDispatchRequest,
    PrintBatchPlateProgress,
    PrintBatchPlateTarget,
    PrintBatchResponse,
    PrintBatchUngroupResponse,
    PrintBatchUpdate,
    PrintQueueBulkUpdate,
    PrintQueueBulkUpdateResponse,
    PrintQueueHistoryClearResponse,
    PrintQueueHistoryResponse,
    PrintQueueItemCreate,
    PrintQueueItemResponse,
    PrintQueueItemUpdate,
    PrintQueueReorder,
    QueueVariantCreate,
    QueueVariantSummary,
)
from backend.app.services.filament_deficit import compute_deficit_for_queue_item
from backend.app.services.filament_requirements import overrides_for_plate
from backend.app.services.finance_budget import release_budget_reservation, validate_print_budget
from backend.app.services.notification_service import notification_service
from backend.app.services.print_batch import (
    BatchDispatchError,
    dispatch_remaining,
    load_progress,
    refresh_batch_status,
    refresh_batch_status_for_item,
)
from backend.app.services.print_cost_estimate import estimate_queue_source_cost
from backend.app.services.queue_position import lock_queue_positions, max_queue_position
from backend.app.utils.printer_models import (
    is_gcode_compatible,
)
from backend.app.utils.threemf_tools import (
    extract_plate_metadata_from_3mf,
    extract_print_time_from_3mf,
)

logger = logging.getLogger(__name__)

router = APIRouter(prefix="/queue", tags=["queue"])

# Answer to a caller who may not start a job that waits for review (#1620)
_AWAITING_REVIEW = "This job waits for review: someone who can manage all queue jobs has to start it"


def _variant_summaries(item: PrintQueueItem) -> list[QueueVariantSummary]:
    """Cross-model candidates for display (#671), or [] if they weren't loaded.

    Every route that builds a queue response eager-loads ``variants``. Reading
    the attribute unguarded would still be a landmine for the next one that
    doesn't: a lazy load on an async session raises rather than degrading, so a
    forgotten ``selectinload`` would turn a card render into a 500.
    """
    if "variants" in inspect(item).unloaded:
        return []
    return [
        QueueVariantSummary(
            library_file_id=v.library_file_id,
            filename=v.library_file.filename if v.library_file else "",
            target_model=v.target_model,
            position=v.position,
        )
        for v in item.variants
    ]


def _extract_filament_types_from_3mf(file_path: Path, plate_id: int | None = None) -> list[str]:
    """Extract unique filament types from a 3MF file.

    Args:
        file_path: Path to the 3MF file
        plate_id: Optional plate index to filter for (for multi-plate files)

    Returns:
        List of unique filament types (e.g., ["PLA", "PETG"])
    """
    types: set[str] = set()

    try:
        with zipfile.ZipFile(file_path, "r") as zf:
            if "Metadata/slice_info.config" not in zf.namelist():
                return []

            content = zf.read("Metadata/slice_info.config").decode()
            root = ET.fromstring(content)

            if plate_id is not None:
                # Find the plate element with matching index
                for plate_elem in root.findall(".//plate"):
                    plate_index = None
                    for meta in plate_elem.findall("metadata"):
                        if meta.get("key") == "index":
                            try:
                                plate_index = int(meta.get("value", "0"))
                            except ValueError:
                                pass  # Skip plate with unparseable index
                            break

                    if plate_index == plate_id:
                        for filament_elem in plate_elem.findall("filament"):
                            filament_type = filament_elem.get("type", "")
                            used_g = filament_elem.get("used_g", "0")
                            try:
                                used_grams = float(used_g)
                            except (ValueError, TypeError):
                                used_grams = 0
                            if used_grams > 0 and filament_type:
                                types.add(filament_type)
                        break
            else:
                # No plate_id specified - extract all filaments with used_g > 0
                for filament_elem in root.findall(".//filament"):
                    filament_type = filament_elem.get("type", "")
                    used_g = filament_elem.get("used_g", "0")
                    try:
                        used_grams = float(used_g)
                    except (ValueError, TypeError):
                        used_grams = 0
                    if used_grams > 0 and filament_type:
                        types.add(filament_type)

    except Exception as e:
        logger.warning("Failed to extract filament types from %s: %s", file_path, e)

    return sorted(types)


# Local alias kept so existing call sites stay compact; the implementation lives
# in utils/threemf_tools.py so the notification path (main.py) can reuse it
# without importing from a routes module (#1785).
_extract_print_time_from_3mf = extract_print_time_from_3mf


def _assert_can_queue_archive(archive: PrintArchive, current_user: User | ApiKeyActor | None) -> None:
    """Gate turning *archive* into a print. Raises rather than returning a verdict.

    Shared by every route that creates queue items from an archive, so a new
    one can't quietly become a weaker door to the same action than
    ``POST /queue/`` is.

    Two separate checks:

    * IDOR fix (maziggy/bambuddy-security #2): without this, a caller with
      QUEUE_CREATE could queue any user's archive even without ARCHIVES_READ on
      it — Landon's PoC enumerated this on admin's archives as operator1. Gate
      on ARCHIVES_READ_ALL OR ownership. 404 (not 403) so we don't leak "this
      id exists but you can't queue it" for enumeration.
    * Reprint perm gate (#1625): the legacy ``/archives/{id}/reprint`` endpoint
      required ARCHIVES_REPRINT_OWN/ALL, and every route that replaces it must
      keep that gate or an operator with QUEUE_CREATE could reprint via a
      direct API call even when explicitly denied reprint perm. Mirrors the
      frontend ``canModify('archives', 'reprint', ...)`` helper: REPRINT_ALL
      allows any archive, REPRINT_OWN allows own only, ownerless archives
      require REPRINT_ALL (fail-closed).
    """
    # None is auth off. An API key arrives as its RequestActor, never None.
    if current_user is None:
        return
    if not current_user.has_permission(Permission.ARCHIVES_READ_ALL.value) and archive.created_by_id != current_user.id:
        raise HTTPException(404, "Archive not found")
    owns_archive = archive.created_by_id is not None and archive.created_by_id == current_user.id
    has_reprint = current_user.has_permission(Permission.ARCHIVES_REPRINT_ALL.value) or (
        owns_archive and current_user.has_permission(Permission.ARCHIVES_REPRINT_OWN.value)
    )
    if not has_reprint:
        raise HTTPException(
            status_code=403,
            detail="Permission archives:reprint_own or archives:reprint_all required",
        )


def _assert_can_queue_library_file(library_file: LibraryFile, current_user: User | ApiKeyActor | None) -> None:
    """Gate turning *library_file* into a print — LIBRARY_READ_ALL or ownership."""
    # None is auth off. An API key arrives as its RequestActor, never None.
    if current_user is None:
        return
    if (
        not current_user.has_permission(Permission.LIBRARY_READ_ALL.value)
        and library_file.created_by_id != current_user.id
    ):
        raise HTTPException(404, "Library file not found")


async def _is_orders_last_source(db: AsyncSession, item: PrintQueueItem) -> bool:
    """True when deleting *item* would leave an order owing runs it can't queue.

    Dispatch produces the runs an order still owes by cloning an existing
    queue item for the same plate — that row is the only record of the printer
    target, AMS mapping and print options the user chose. Delete the last one
    while the plate still has a target and the order is stranded: it goes on
    reporting work outstanding with no way left to produce it (#2960).
    """
    if item.batch_id is None:
        return False

    # A cancelled order can never dispatch again, so nothing about it can be
    # stranded and its leftover rows must stay deletable — tidying up after a
    # cancel is the most likely reason anyone deletes them.
    status = (await db.execute(select(PrintBatch.status).where(PrintBatch.id == item.batch_id))).scalar_one_or_none()
    if status == "cancelled":
        return False

    plate_scope = (
        PrintBatchPlate.plate_id == item.plate_id if item.plate_id is not None else PrintBatchPlate.plate_id.is_(None)
    )
    # first(), not scalar_one_or_none(): a UNIQUE(batch_id, plate_id) does not
    # constrain NULL plate_ids on either dialect, and a duplicate whole-file
    # row must not turn a delete into a 500.
    target = (
        (
            await db.execute(
                select(PrintBatchPlate.quantity_target)
                .where(PrintBatchPlate.batch_id == item.batch_id)
                .where(plate_scope)
                .order_by(PrintBatchPlate.quantity_target.desc())
                .limit(1)
            )
        )
        .scalars()
        .first()
    )
    # No target row: a grouping, or a plate assigned into the order by hand.
    # A target of 0 is legal and means the plate is not wanted. Neither owes
    # anything, so neither can be stranded.
    if not target:
        return False

    item_scope = (
        PrintQueueItem.plate_id == item.plate_id if item.plate_id is not None else PrintQueueItem.plate_id.is_(None)
    )
    survivor = (
        await db.execute(
            select(PrintQueueItem.id)
            .where(PrintQueueItem.batch_id == item.batch_id)
            .where(item_scope)
            .where(PrintQueueItem.id != item.id)
            .limit(1)
        )
    ).scalar_one_or_none()
    return survivor is None


async def _assert_can_dispatch_batch_sources(
    db: AsyncSession, batch_id: int, current_user: User | ApiKeyActor | None
) -> None:
    """Apply the ``POST /queue/`` source-file gates to everything a dispatch would print.

    Dispatching clones existing queue items, so without this it would be a
    weaker door to the same outcome: a caller holding QUEUE_CREATE and
    QUEUE_UPDATE_ALL but explicitly denied ``archives:reprint_*`` could start
    prints through an order that ``POST /queue/`` would have refused them.

    Every distinct source among the batch's items is checked, including the
    library files behind cross-model variants — those get cloned too, and any
    one of them may be the file that actually runs.
    """
    archive_ids = set(
        (
            await db.execute(
                select(PrintQueueItem.archive_id)
                .where(PrintQueueItem.batch_id == batch_id)
                .where(PrintQueueItem.archive_id.is_not(None))
                .distinct()
            )
        )
        .scalars()
        .all()
    )
    library_file_ids = set(
        (
            await db.execute(
                select(PrintQueueItem.library_file_id)
                .where(PrintQueueItem.batch_id == batch_id)
                .where(PrintQueueItem.library_file_id.is_not(None))
                .distinct()
            )
        )
        .scalars()
        .all()
    )
    library_file_ids |= set(
        (
            await db.execute(
                select(PrintQueueVariant.library_file_id)
                .join(PrintQueueItem, PrintQueueVariant.queue_item_id == PrintQueueItem.id)
                .where(PrintQueueItem.batch_id == batch_id)
                .distinct()
            )
        )
        .scalars()
        .all()
    )

    for archive_id in archive_ids:
        archive = (await db.execute(select(PrintArchive).where(PrintArchive.id == archive_id))).scalar_one_or_none()
        # A deleted source can't be printed; dispatch will fail on it anyway.
        if archive is not None:
            _assert_can_queue_archive(archive, current_user)

    for library_file_id in library_file_ids:
        library_file = (
            await db.execute(LibraryFile.active().where(LibraryFile.id == library_file_id))
        ).scalar_one_or_none()
        if library_file is not None:
            _assert_can_queue_library_file(library_file, current_user)


async def _resolve_source_path(db: AsyncSession, item: PrintQueueItem) -> Path | None:
    """Resolve an existing queue item's source 3MF on disk, or None."""
    if item.archive_id:
        result = await db.execute(select(PrintArchive).where(PrintArchive.id == item.archive_id))
        archive = result.scalar_one_or_none()
        if archive:
            return settings.base_dir / archive.file_path
    elif item.library_file_id:
        result = await db.execute(LibraryFile.active().where(LibraryFile.id == item.library_file_id))
        library_file = result.scalar_one_or_none()
        if library_file:
            lib_path = Path(library_file.file_path)
            return lib_path if lib_path.is_absolute() else settings.base_dir / library_file.file_path
    return None


async def _trusted_item_estimated_cost(
    db: AsyncSession,
    item: PrintQueueItem,
    *,
    printer_id: int | None,
    plate_id: int | None,
    ams_mapping: list[int] | str | None,
) -> float | None:
    """Recompute an existing queue item's cost from its persisted source."""

    if item.archive_id:
        archive = await db.scalar(select(PrintArchive).where(PrintArchive.id == item.archive_id))
        return await estimate_queue_source_cost(
            db,
            archive=archive,
            plate_id=plate_id,
            ams_mapping=ams_mapping,
            printer_id=printer_id,
        )
    if item.library_file_id:
        library_file = await db.scalar(LibraryFile.active().where(LibraryFile.id == item.library_file_id))
        return await estimate_queue_source_cost(
            db,
            library_file=library_file,
            plate_id=plate_id,
            ams_mapping=ams_mapping,
            printer_id=printer_id,
        )

    result = await db.execute(
        select(PrintQueueVariant, LibraryFile)
        .join(LibraryFile, LibraryFile.id == PrintQueueVariant.library_file_id)
        .where(PrintQueueVariant.queue_item_id == item.id)
    )
    estimates = [
        await estimate_queue_source_cost(
            db,
            library_file=library_file,
            plate_id=variant.plate_id,
            ams_mapping=variant.ams_mapping,
            printer_id=printer_id,
        )
        for variant, library_file in result.all()
    ]
    if not estimates or any(cost is None for cost in estimates):
        return None
    return max(cost for cost in estimates if cost is not None)


def _enrich_response(item: PrintQueueItem) -> PrintQueueItemResponse:
    """Add nested archive/printer/library_file info to response."""
    # Parse ams_mapping from JSON string BEFORE model_validate
    ams_mapping_parsed = None
    if item.ams_mapping:
        try:
            ams_mapping_parsed = json.loads(item.ams_mapping)
        except json.JSONDecodeError:
            ams_mapping_parsed = None

    # Parse required_filament_types from JSON string
    required_filament_types_parsed = None
    if item.required_filament_types:
        try:
            required_filament_types_parsed = json.loads(item.required_filament_types)
        except json.JSONDecodeError:
            required_filament_types_parsed = None

    # Parse filament_overrides from JSON string
    filament_overrides_parsed = None
    if item.filament_overrides:
        try:
            filament_overrides_parsed = json.loads(item.filament_overrides)
        except json.JSONDecodeError:
            filament_overrides_parsed = None

    # Parse nozzle_mapping from JSON string (#1780 — H2C rack slicer-pick
    # preservation). Nullable opaque JSON blob stored verbatim from
    # BambuStudio's project_file; surface it parsed for the response model
    # and any future "edit print → nozzle" UI.
    nozzle_mapping_parsed = None
    if item.nozzle_mapping:
        try:
            nozzle_mapping_parsed = json.loads(item.nozzle_mapping)
        except json.JSONDecodeError:
            nozzle_mapping_parsed = None

    # The operator's rack-position pick (#1784), keyed by filament group. Sent
    # parsed so the print dialog can show which hotend each group will use.
    nozzle_rack_choice_parsed = None
    if item.nozzle_rack_choice:
        try:
            nozzle_rack_choice_parsed = json.loads(item.nozzle_rack_choice)
        except json.JSONDecodeError:
            nozzle_rack_choice_parsed = None

    nozzles_info_parsed = None
    if item.nozzles_info:
        try:
            nozzles_info_parsed = json.loads(item.nozzles_info)
        except json.JSONDecodeError:
            nozzles_info_parsed = None

    # Create response with parsed ams_mapping
    item_dict = {
        "id": item.id,
        "printer_id": item.printer_id,
        "target_model": item.target_model,
        "target_location": item.target_location,
        "required_filament_types": required_filament_types_parsed,
        "filament_overrides": filament_overrides_parsed,
        "waiting_reason": item.waiting_reason,
        "archive_id": item.archive_id,
        "library_file_id": item.library_file_id,
        "cost_center_id": item.cost_center_id,
        "estimated_cost": item.estimated_cost,
        "position": item.position,
        "scheduled_time": item.scheduled_time,
        "require_previous_success": item.require_previous_success,
        "auto_off_after": item.auto_off_after,
        "manual_start": item.manual_start,
        "filament_short": bool(item.filament_short),
        "skip_filament_check": bool(item.skip_filament_check),
        "ams_mapping": ams_mapping_parsed,
        "plate_id": item.plate_id,
        "bed_levelling": item.bed_levelling,
        "flow_cali": item.flow_cali,
        "vibration_cali": item.vibration_cali,
        "layer_inspect": item.layer_inspect,
        "timelapse": item.timelapse,
        "use_ams": item.use_ams,
        "nozzle_offset_cali": item.nozzle_offset_cali,
        "confirm_outcome": item.confirm_outcome,
        "preheat_override": item.preheat_override,
        "preheat_chamber_target_override": item.preheat_chamber_target_override,
        "status": item.status,
        "started_at": item.started_at,
        "completed_at": item.completed_at,
        "error_message": item.error_message,
        "created_at": item.created_at,
        # User tracking (Issue #206)
        "created_by_id": item.created_by_id,
        "created_by_username": item.created_by.username if item.created_by else None,
        # Batch grouping
        "batch_id": item.batch_id,
        "batch_name": item.batch.name if item.batch else None,
        # SJF scheduling
        "been_jumped": item.been_jumped,
        # Auto-print G-code injection
        "gcode_injection": item.gcode_injection,
        # H2C rack-swap nozzle pick (#1780)
        "nozzle_mapping": nozzle_mapping_parsed,
        "nozzle_rack_choice": nozzle_rack_choice_parsed,
        "nozzles_info": nozzles_info_parsed,
        "cleanup_library_after_dispatch": item.cleanup_library_after_dispatch,
        # Cross-model alternatives (#671). Guarded rather than read directly:
        # every route that reaches here eager-loads the relationship, but a
        # caller that forgets would trigger a lazy load, and a lazy load on an
        # async session raises rather than degrading. An empty list is the
        # correct answer for the ordinary item this would most likely be.
        "variants": _variant_summaries(item),
    }
    response = PrintQueueItemResponse(**item_dict)
    if item.archive:
        # Soft-deleted archive: files are gone from disk but the row stays
        # (its filament/cost contribution still flows into stats per #1343).
        # Suppress the archive-derived UI surface so the queue page doesn't
        # 404-storm the thumbnail / plates / plate-thumbnail endpoints — the
        # frontend's existing truthy gate on archive_thumbnail covers it
        # (#1348 follow-up). The archive_deleted flag lets the UI render a
        # "source deleted" badge on these rows.
        if item.archive.deleted_at is not None:
            response.archive_deleted = True
        else:
            response.archive_name = item.archive.print_name or item.archive.filename
            response.archive_thumbnail = item.archive.thumbnail_path
            response.print_time_seconds = item.archive.print_time_seconds
            response.filament_used_grams = item.archive.filament_used_grams
            response.filament_type = item.archive.filament_type
            response.filament_color = item.archive.filament_color
            response.layer_height = item.archive.layer_height
            response.nozzle_diameter = item.archive.nozzle_diameter
            response.sliced_for_model = item.archive.sliced_for_model
            response.bed_type = item.archive.bed_type
            # Marks history/reprint rows whose archive carries the slicer's own
            # live-resolved AMS-slot pick (extra_data.slicer_ams_mapping) — see
            # `_extract_slicer_ams_mapping_json` in virtual_printer/manager.py.
            #
            # Only when the saved mapping was resolved against *this* row's
            # printer: a global tray ID means nothing on another printer, so
            # that's the exact condition under which the mapping is reused. A
            # badge on a row where nothing gets reused would be a lie (#2700
            # review). Model-based rows (printer_id None) never match, which is
            # correct — the mapping is not reused there either.
            extra = item.archive.extra_data if isinstance(item.archive.extra_data, dict) else {}
            saved_mapping = extra.get("slicer_ams_mapping")
            response.archive_has_slicer_ams_mapping = (
                isinstance(saved_mapping, dict)
                and isinstance(saved_mapping.get("mapping"), list)
                and item.printer_id is not None
                and saved_mapping.get("printer_id") == item.printer_id
            )
            if item.plate_id:
                archive_path = settings.base_dir / item.archive.file_path
                if archive_path.exists():
                    # One cached parse for all three per-plate overrides (#2573).
                    plate_meta = extract_plate_metadata_from_3mf(archive_path, item.plate_id)
                    if plate_meta.print_time_seconds is not None:
                        response.print_time_seconds = plate_meta.print_time_seconds
                    if plate_meta.filament_used_grams > 0:
                        response.filament_used_grams = plate_meta.filament_used_grams
                    if plate_meta.bed_type:
                        response.bed_type = plate_meta.bed_type
    if item.library_file:
        response.library_file_name = (
            item.library_file.file_metadata.get("print_name") if item.library_file.file_metadata else None
        )
        if not response.library_file_name:
            response.library_file_name = item.library_file.filename
        response.library_file_thumbnail = item.library_file.thumbnail_path
        # Get metadata from library file if no archive
        if not item.archive and item.library_file.file_metadata:
            response.print_time_seconds = item.library_file.file_metadata.get("print_time_seconds")
            response.filament_used_grams = item.library_file.file_metadata.get("filament_used_grams")
            response.filament_type = item.library_file.file_metadata.get("filament_type")
            response.filament_color = item.library_file.file_metadata.get("filament_color")
            response.layer_height = item.library_file.file_metadata.get("layer_height")
            response.nozzle_diameter = item.library_file.file_metadata.get("nozzle_diameter")
            response.sliced_for_model = item.library_file.file_metadata.get("sliced_for_model")
            response.bed_type = item.library_file.file_metadata.get("bed_type")
        if item.plate_id:
            lib_path = Path(item.library_file.file_path)
            library_file_path = lib_path if lib_path.is_absolute() else settings.base_dir / item.library_file.file_path
            if library_file_path.exists():
                # One cached parse for all three per-plate overrides (#2573).
                plate_meta = extract_plate_metadata_from_3mf(library_file_path, item.plate_id)
                if plate_meta.print_time_seconds is not None:
                    response.print_time_seconds = plate_meta.print_time_seconds
                if plate_meta.filament_used_grams > 0:
                    response.filament_used_grams = plate_meta.filament_used_grams
                if plate_meta.bed_type:
                    response.bed_type = plate_meta.bed_type
    if item.printer:
        response.printer_name = item.printer.name
    return response


# What _enrich_response reads, loaded up front for a list of items.
_QUEUE_ITEM_LOADS = (
    selectinload(PrintQueueItem.archive),
    selectinload(PrintQueueItem.printer),
    selectinload(PrintQueueItem.library_file),
    selectinload(PrintQueueItem.created_by),
    selectinload(PrintQueueItem.batch),
    # Cross-model candidates (#671) and their files, for the card label.
    selectinload(PrintQueueItem.variants).selectinload(PrintQueueVariant.library_file),
)


@router.get("/", response_model=list[PrintQueueItemResponse])
async def list_queue(
    printer_id: int | None = Query(None, description="Filter by printer (-1 for unassigned)"),
    status: str | None = Query(None, description="Filter by status; several may be given comma-separated"),
    target_model: str | None = Query(
        None, description="Filter by target model (also includes model-based items when combined with printer_id)"
    ),
    db: AsyncSession = Depends(get_db),
    auth_result: tuple[User | None, bool] = Depends(
        require_ownership_permission(
            Permission.QUEUE_READ_ALL,
            Permission.QUEUE_READ_OWN,
        )
    ),
    printer_scope: PrinterScope = RequestPrinterScope,
):
    """List all queue items, optionally filtered by printer or status."""
    user, can_read_all = auth_result
    query = (
        select(PrintQueueItem)
        .options(*_QUEUE_ITEM_LOADS)
        # The order the scheduler dispatches in (#3200), so the first pending
        # item for a printer is the one it will start next -- which is what the
        # printer card's "Next in queue" shows. Sorting by printer first put
        # every "Any <model>" job (no printer_id) ahead of a job pinned to that
        # printer, whatever their positions.
        .order_by(PrintQueueItem.position, PrintQueueItem.id)
    )
    if user is not None and not can_read_all:
        query = query.where(PrintQueueItem.created_by_id == user.id)
    # Items bound to printers the caller can't see stay out (#1727)
    if (clause := printer_scope.where(PrintQueueItem.printer_id)) is not None:
        query = query.where(clause)

    if printer_id is not None:
        if printer_id == -1:
            # Special value: filter for unassigned items
            query = query.where(PrintQueueItem.printer_id.is_(None))
        else:
            # Resolve effective model: prefer explicit param, fall back to printer's DB model.
            # This ensures model-based "Any X" items are returned even when the frontend
            # doesn't send target_model (e.g. printer.model is NULL on the client side).
            effective_model = target_model
            if not effective_model:
                printer_row = (
                    await db.execute(select(Printer.model).where(Printer.id == printer_id))
                ).scalar_one_or_none()
                effective_model = printer_row

            if effective_model:
                # Include both printer-specific items AND model-based (unassigned) items
                query = query.where(
                    or_(
                        PrintQueueItem.printer_id == printer_id,
                        and_(
                            PrintQueueItem.printer_id.is_(None),
                            func.lower(PrintQueueItem.target_model) == effective_model.lower(),
                        ),
                    )
                )
            else:
                query = query.where(PrintQueueItem.printer_id == printer_id)
    elif target_model:
        query = query.where(func.lower(PrintQueueItem.target_model) == target_model.lower())
    if status:
        statuses = [s.strip() for s in status.split(",") if s.strip()]
        query = query.where(PrintQueueItem.status.in_(statuses))

    result = await db.execute(query)
    items = result.scalars().all()
    return [_enrich_response(item) for item in items]


HISTORY_STATUSES = ("completed", "failed", "skipped", "cancelled")


def _history_conditions(
    user: User | None,
    can_read_all: bool,
    printer_scope: PrinterScope,
    printer_id: int | None,
    status: str | None,
    location: str | None,
) -> list:
    """WHERE clauses for the history the caller sees under the page's filters.

    Mirrors what the Queue page used to filter in the browser: the printer and
    status selectors, and the location selector, which matches a model-based
    run by its target location and any other run by its printer's location.
    """
    statuses = [status] if status in HISTORY_STATUSES else list(HISTORY_STATUSES) if not status else []
    conditions = [PrintQueueItem.status.in_(statuses)]
    if user is not None and not can_read_all:
        conditions.append(PrintQueueItem.created_by_id == user.id)
    # Items bound to printers the caller can't see stay out (#1727)
    if (clause := printer_scope.where(PrintQueueItem.printer_id)) is not None:
        conditions.append(clause)
    if printer_id is not None:
        if printer_id == -1:
            conditions.append(PrintQueueItem.printer_id.is_(None))
        else:
            conditions.append(PrintQueueItem.printer_id == printer_id)
    if location:
        # Aliased so it can't correlate with a Printer joined by the caller
        located = aliased(Printer)
        no_target_location = or_(PrintQueueItem.target_location.is_(None), PrintQueueItem.target_location == "")
        conditions.append(
            or_(
                PrintQueueItem.target_location == location,
                and_(
                    no_target_location,
                    PrintQueueItem.printer_id.in_(select(located.id).where(located.location == location)),
                ),
            )
        )
    return conditions


@router.get("/history", response_model=PrintQueueHistoryResponse)
async def list_queue_history(
    printer_id: int | None = Query(None, description="Filter by printer (-1 for unassigned)"),
    status: str | None = Query(None, description="One of completed, failed, skipped, cancelled"),
    location: str | None = Query(None, description="Filter by location"),
    sort_by: Literal["date", "name", "printer"] = Query("date"),
    reverse: bool = Query(False, description="Flip the default order (newest first, A to Z)"),
    # No practical cap: "Show more" can grow the page to the whole history,
    # as the page always could.
    limit: int = Query(50, ge=1, le=1_000_000),
    offset: int = Query(0, ge=0),
    db: AsyncSession = Depends(get_db),
    auth_result: tuple[User | None, bool] = Depends(
        require_ownership_permission(
            Permission.QUEUE_READ_ALL,
            Permission.QUEUE_READ_OWN,
        )
    ),
    printer_scope: PrinterScope = RequestPrinterScope,
):
    """Finished, failed, skipped and cancelled runs, one page at a time.

    The Queue page used to load the whole queue, history included, every five
    seconds and page through it in the browser. On a farm that is thousands of
    rows, each enriched from its 3MF, and the request held up every other one
    for over a minute.
    """
    user, can_read_all = auth_result
    conditions = _history_conditions(user, can_read_all, printer_scope, printer_id, status, location)

    total = (await db.execute(select(func.count(PrintQueueItem.id)).where(*conditions))).scalar_one()

    if sort_by == "name":
        # The name the row shows: a live archive's print name or file name,
        # else the library file's (its metadata print name is not sortable in
        # SQL, so the file name stands in for it).
        archive_name = case(
            (
                PrintArchive.deleted_at.is_(None),
                func.coalesce(func.nullif(PrintArchive.print_name, ""), PrintArchive.filename),
            ),
            else_=None,
        )
        name = func.lower(func.coalesce(func.nullif(archive_name, ""), LibraryFile.filename, ""))
        order = [name.desc() if reverse else name.asc()]
    elif sort_by == "printer":
        name = func.lower(func.coalesce(Printer.name, ""))
        order = [name.desc() if reverse else name.asc()]
    else:
        when = func.coalesce(PrintQueueItem.completed_at, PrintQueueItem.created_at)
        order = [when.asc() if reverse else when.desc()]
    order.append(PrintQueueItem.id.asc() if reverse else PrintQueueItem.id.desc())

    page_ids = list(
        (
            await db.execute(
                select(PrintQueueItem.id)
                .outerjoin(PrintArchive, PrintArchive.id == PrintQueueItem.archive_id)
                .outerjoin(LibraryFile, LibraryFile.id == PrintQueueItem.library_file_id)
                .outerjoin(Printer, Printer.id == PrintQueueItem.printer_id)
                .where(*conditions)
                .order_by(*order)
                .limit(limit)
                .offset(offset)
            )
        ).scalars()
    )

    # The other runs of the batches on this page, so a batch row counts them all
    batch_ids = (
        (
            await db.execute(
                select(PrintQueueItem.batch_id)
                .where(PrintQueueItem.id.in_(page_ids), PrintQueueItem.batch_id.isnot(None))
                .distinct()
            )
        )
        .scalars()
        .all()
        if page_ids
        else []
    )
    sibling_ids: list[int] = []
    if batch_ids:
        sibling_ids = list(
            (
                await db.execute(
                    select(PrintQueueItem.id)
                    .where(*conditions, PrintQueueItem.batch_id.in_(batch_ids), PrintQueueItem.id.notin_(page_ids))
                    .order_by(PrintQueueItem.id)
                )
            ).scalars()
        )

    ids = page_ids + sibling_ids
    rows = {}
    if ids:
        result = await db.execute(select(PrintQueueItem).options(*_QUEUE_ITEM_LOADS).where(PrintQueueItem.id.in_(ids)))
        rows = {item.id: item for item in result.scalars().all()}

    locations = (
        (
            await db.execute(
                select(PrintQueueItem.target_location)
                .where(
                    *_history_conditions(user, can_read_all, printer_scope, None, None, None),
                    PrintQueueItem.target_location.isnot(None),
                    PrintQueueItem.target_location != "",
                )
                .distinct()
            )
        )
        .scalars()
        .all()
    )

    return {
        "items": [_enrich_response(rows[i]) for i in ids if i in rows],
        "total": total,
        "locations": sorted(locations),
    }


@router.post("/history/clear", response_model=PrintQueueHistoryClearResponse)
async def clear_queue_history(
    printer_id: int | None = Query(None, description="Filter by printer (-1 for unassigned)"),
    status: str | None = Query(None, description="One of completed, failed, skipped, cancelled"),
    location: str | None = Query(None, description="Filter by location"),
    db: AsyncSession = Depends(get_db),
    auth_result: tuple[User | None, bool] = Depends(
        require_ownership_permission(
            Permission.QUEUE_DELETE_ALL,
            Permission.QUEUE_DELETE_OWN,
        )
    ),
    read_result: tuple[User | None, bool] = Depends(
        require_ownership_permission(
            Permission.QUEUE_READ_ALL,
            Permission.QUEUE_READ_OWN,
        )
    ),
    printer_scope: PrinterScope = RequestPrinterScope,
):
    """Remove the history the Queue page shows under its current filters.

    The page used to send one DELETE per history row. Each row goes through
    the same rule as that DELETE: an order's last run for a plate is kept as
    cancelled rather than deleted (#2960). Rows are handled one at a time, in
    one transaction, so each check sees the rows removed before it.

    Reaches only rows the caller could both see and delete, as the page's
    one-by-one DELETEs did: someone who may delete every run but read only
    their own clears only their own.
    """
    user, can_modify_all = auth_result
    _, can_read_all = read_result
    conditions = _history_conditions(user, can_modify_all and can_read_all, printer_scope, printer_id, status, location)
    # The order the page listed them in when it deleted them one by one, which
    # decides which of an order's runs is the last one left to keep.
    items = (
        (
            await db.execute(
                select(PrintQueueItem).where(*conditions).order_by(PrintQueueItem.position, PrintQueueItem.id)
            )
        )
        .scalars()
        .all()
    )

    cleared = kept = 0
    item_ids = [item.id for item in items]
    for item in items:
        if await _remove_queue_item(db, item):
            cleared += 1
        else:
            kept += 1
        await db.flush()
    await db.commit()
    _notify_dispatch_cancelled(item_ids)

    logger.info("Cleared queue history: %s deleted, %s kept as cancelled for their order", cleared, kept)
    return {"cleared": cleared, "kept": kept}


async def _resolve_queue_variants(
    db: AsyncSession,
    specs: list[QueueVariantCreate],
    current_user: User | ApiKeyActor | None,
) -> list[tuple[QueueVariantCreate, LibraryFile, str]]:
    """Validate a cross-model candidate set and pair each file with its model (#671).

    Validated as a set, not file by file, because the failure modes are about the
    set: two candidates for the same printer give the resolver no basis to choose,
    and a set where nothing can ever run is a job that waits forever.

    At least one candidate must have an active printer — the rest may not, which
    is deliberate. Grouping the H2C slice before the H2C arrives is a reasonable
    thing to do, and refusing the whole queue action over it would be worse than
    letting that candidate simply never match.
    """
    file_ids = [s.library_file_id for s in specs]
    if len(set(file_ids)) != len(file_ids):
        raise HTTPException(400, "The same file cannot be listed twice as a variant")

    rows = (await db.execute(LibraryFile.active().where(LibraryFile.id.in_(file_ids)))).scalars().all()
    by_id = {f.id: f for f in rows}

    resolved: list[tuple[QueueVariantCreate, LibraryFile, str]] = []
    seen_models: dict[str, str] = {}
    any_active_printer = False

    for spec in specs:
        library_file = by_id.get(spec.library_file_id)
        # Same IDOR posture as the single-file path: a file the caller cannot read
        # is reported as missing rather than forbidden.
        if not library_file or (
            current_user
            and not current_user.has_permission(Permission.LIBRARY_READ_ALL.value)
            and library_file.created_by_id != current_user.id
        ):
            raise HTTPException(404, f"Library file not found: {spec.library_file_id}")

        from backend.app.utils.filename import InvalidFilenameError, validate_print_filename

        try:
            validate_print_filename(library_file.filename)
        except InvalidFilenameError as e:
            raise HTTPException(400, str(e)) from e

        model = resolve_variant_model(library_file, spec.target_model)
        if not model:
            raise HTTPException(
                400,
                f"{library_file.filename} does not say which printer it was sliced for — "
                "set its target model explicitly",
            )

        # Cross-model safety gate (#2578), per candidate. A set is only as safe as
        # its worst member, and model-based dispatch has no human in the loop.
        sliced_for = (library_file.file_metadata or {}).get("sliced_for_model")
        if not is_gcode_compatible(sliced_for, model):
            raise HTTPException(
                400,
                f"{library_file.filename} was sliced for {sliced_for} and cannot be dispatched to {model} printers",
            )

        if model in seen_models:
            raise HTTPException(
                400,
                f"{library_file.filename} and {seen_models[model]} are both for {model} — "
                "variants must target different printers",
            )
        seen_models[model] = library_file.filename

        has_printer = (
            (
                await db.execute(
                    select(Printer).where(Printer.model == model).where(Printer.is_active == True)  # noqa: E712
                )
            )
            .scalars()
            .first()
        )
        any_active_printer = any_active_printer or bool(has_printer)

        resolved.append((spec, library_file, model))

    if not any_active_printer:
        raise HTTPException(400, f"No active printers for any of: {', '.join(seen_models)}")

    return resolved


def _variant_values(
    spec: QueueVariantCreate,
    library_file: LibraryFile,
    model: str,
    position: int,
) -> dict:
    """Column values for one candidate, extracted from its own 3MF.

    Each candidate is a different slice, so its filament requirements and print
    time come from its own file rather than being inherited from the item.

    Returns values rather than a row so a quantity>1 batch can build one row per
    copy without re-opening the 3MF for each.
    """
    lib_path = Path(library_file.file_path)
    file_path = lib_path if lib_path.is_absolute() else settings.base_dir / library_file.file_path

    required_types = None
    filament_overrides_json = None
    print_time = (library_file.file_metadata or {}).get("print_time_seconds")

    if file_path.exists():
        types = _extract_filament_types_from_3mf(file_path, spec.plate_id)
        if types:
            required_types = json.dumps(types)
        if spec.plate_id:
            plate_time = _extract_print_time_from_3mf(file_path, spec.plate_id)
            if plate_time is not None:
                print_time = plate_time
        if spec.filament_overrides:
            plate_overrides = overrides_for_plate(spec.filament_overrides, file_path, spec.plate_id)
            if plate_overrides:
                filament_overrides_json = json.dumps(plate_overrides)
                override_types = sorted({o["type"] for o in plate_overrides if "type" in o})
                if override_types:
                    existing = set(json.loads(required_types)) if required_types else set()
                    required_types = json.dumps(sorted(existing | set(override_types)))

    return {
        "position": position,
        "library_file_id": library_file.id,
        "target_model": model,
        "plate_id": spec.plate_id,
        "ams_mapping": json.dumps(spec.ams_mapping) if spec.ams_mapping else None,
        "nozzle_mapping": json.dumps(spec.nozzle_mapping) if spec.nozzle_mapping else None,
        "nozzle_rack_choice": json.dumps(spec.nozzle_rack_choice) if spec.nozzle_rack_choice else None,
        "filament_overrides": filament_overrides_json,
        "required_filament_types": required_types,
        "print_time_seconds": print_time,
    }


@router.post("/", response_model=PrintQueueItemResponse)
async def add_to_queue(
    data: PrintQueueItemCreate,
    db: AsyncSession = Depends(get_db),
    current_user: User | None = RequirePermissionIfAuthEnabled(Permission.QUEUE_CREATE),
    printer_scope: PrinterScope = RequestPrinterScope,
    review_required: bool = QueueReviewRequired,
    actor: User | ApiKeyActor | None = RequestActor,
):
    """Add an item to the print queue."""
    # Normalize target_model (e.g., "Bambu Lab X1E" / "C13" -> "X1E").
    # normalize_model_name resolves internal codes first: the previous
    # `normalize_printer_model(x) or normalize_printer_model_id(x)` chain never
    # reached the code map, because the first call returns unknown input
    # unchanged — so a "C13" target stayed "C13", matched no printer row and
    # left the item waiting forever. Identical result for every other spelling.
    target_model_norm = normalize_model_name(data.target_model)

    # Cross-model alternatives (#671): several sliced files, whichever printer
    # frees up first. The whole candidate set is validated before anything is
    # written — a half-valid set would produce a job that can only reach some of
    # the printers the user asked for, with nothing to say which.
    variant_specs: list[tuple[QueueVariantCreate, LibraryFile, str]] = []
    if data.variants:
        if data.printer_id:
            raise HTTPException(
                400, "Cannot specify both printer_id and variants — pick a printer or offer alternatives"
            )
        if data.archive_id or data.library_file_id:
            raise HTTPException(
                400, "Cannot combine variants with archive_id or library_file_id — the variants are the files"
            )
        variant_specs = await _resolve_queue_variants(db, data.variants, actor)
        # Mirror the first candidate onto the item so the queue listing, the SJF
        # grouping and the "Any H2S" label have something before a printer is
        # picked. Resolution overwrites it with whichever candidate actually runs.
        target_model_norm = variant_specs[0][2]

    # Validate that either archive_id or library_file_id is provided.
    # A cross-model item deliberately holds neither: its files live on the
    # variant rows. Pointing library_file_id at one of them would be worse than
    # useless — that FK is ON DELETE CASCADE, so deleting a single alternative
    # would take the whole queue item with it.
    if not data.archive_id and not data.library_file_id and not data.variants:
        raise HTTPException(400, "Either archive_id or library_file_id must be provided")

    # Cannot specify both printer_id and target_model
    if data.printer_id and target_model_norm:
        raise HTTPException(400, "Cannot specify both printer_id and target_model")
    if target_model_norm:
        # The key itself, not its actor: the scheduler holds an any-printer job
        # to its creator's printers, and a key may be limited to fewer of them.
        ensure_model_target_allowed(current_user, printer_scope)

    # Validate printer exists (if assigned)
    if data.printer_id is not None:
        printer_scope.ensure(data.printer_id)
        result = await db.execute(select(Printer).where(Printer.id == data.printer_id))
        if not result.scalar_one_or_none():
            raise HTTPException(400, "Printer not found")

    # Validate target_model has active printers. Skipped for cross-model items:
    # target_model there is just the first candidate, and _resolve_queue_variants
    # has already required that *some* candidate has a printer.
    if target_model_norm and not data.variants:
        result = await db.execute(
            select(Printer).where(Printer.model == target_model_norm).where(Printer.is_active == True)  # noqa: E712
        )
        if not result.scalars().first():
            raise HTTPException(400, f"No active printers for model: {target_model_norm}")

    # Validate archive exists (if provided) and get it for filament extraction
    archive = None
    if data.archive_id:
        result = await db.execute(select(PrintArchive).where(PrintArchive.id == data.archive_id))
        archive = result.scalar_one_or_none()
        if not archive:
            raise HTTPException(400, "Archive not found")
        _assert_can_queue_archive(archive, actor)

    # Validate library file exists (if provided) and get it for filament extraction
    library_file = None
    if data.library_file_id:
        result = await db.execute(LibraryFile.active().where(LibraryFile.id == data.library_file_id))
        library_file = result.scalar_one_or_none()
        if not library_file:
            raise HTTPException(400, "Library file not found")
        _assert_can_queue_library_file(library_file, actor)
        # Bambu SD card is FAT32/exFAT — illegal filename chars would 553 at
        # FTP upload time (#1540). Reject at queue time so the user gets the
        # actionable error before waiting in queue.
        from backend.app.utils.filename import InvalidFilenameError, validate_print_filename

        try:
            validate_print_filename(library_file.filename)
        except InvalidFilenameError as e:
            raise HTTPException(400, str(e)) from e

    # Cross-model safety gate (#2578): a G-code 3MF sliced for one model must
    # not be queued for dispatch to an incompatible model. The UI can no longer
    # produce such rows, but API-created rows must be rejected here too — the
    # scheduler assigns model-based items to hardware with no human in the loop.
    if target_model_norm:
        sliced_for = None
        if archive:
            sliced_for = archive.sliced_for_model
        elif library_file and library_file.file_metadata:
            sliced_for = library_file.file_metadata.get("sliced_for_model")
        if not is_gcode_compatible(sliced_for, target_model_norm):
            raise HTTPException(
                400,
                f"File was sliced for {sliced_for} and cannot be dispatched to {target_model_norm} printers",
            )

    # Extract filament types for model-based assignment (used by scheduler for validation)
    required_filament_types = None
    file_path = None
    # Get file path from archive or library file
    if archive:
        file_path = settings.base_dir / archive.file_path
    elif library_file:
        lib_path = Path(library_file.file_path)
        file_path = lib_path if lib_path.is_absolute() else settings.base_dir / library_file.file_path
    if target_model_norm:
        if file_path and file_path.exists():
            filament_types = _extract_filament_types_from_3mf(file_path, data.plate_id)
            if filament_types:
                required_filament_types = json.dumps(filament_types)
                logger.info("Extracted filament types for model-based queue: %s", filament_types)

    # If filament overrides are provided, update required_filament_types to match override types.
    # A specific-printer job keeps its overrides too (#3133): an override chosen for
    # "Any P2S" survives the switch to one P2S in the print dialog, and when the
    # dialog could not resolve every tray the scheduler recomputes the mapping at
    # dispatch — against the 3MF's filament, unless the row still says otherwise.
    # The type list below stays model-only; it gates which printer of a model is
    # eligible, which a printer-targeted job has already settled.
    filament_overrides_json = None
    if data.filament_overrides and (target_model_norm or data.printer_id is not None):
        plate_overrides = overrides_for_plate(data.filament_overrides, file_path, data.plate_id)
        if plate_overrides:
            filament_overrides_json = json.dumps(plate_overrides)
            # Update required_filament_types from overrides so scheduler validates against overridden types
            override_types = sorted({o["type"] for o in plate_overrides if "type" in o})
            if override_types and target_model_norm:
                # Merge with existing types (overrides may only cover some slots)
                existing_types = set(json.loads(required_filament_types)) if required_filament_types else set()
                # Replace types for overridden slots, keep others
                all_types = existing_types | set(override_types)
                required_filament_types = json.dumps(sorted(all_types))

    # Validate quantity
    quantity = max(1, data.quantity)

    # Validate batch_id if provided. Client passes batch_id when adding items
    # into a pre-created batch (multi-plate auto-batch or "Group as batch" flow).
    # 404 keeps the existing-id leak surface low.
    batch = None
    batch_id = None
    if data.batch_id is not None:
        result = await db.execute(select(PrintBatch).where(PrintBatch.id == data.batch_id))
        existing_batch = result.scalar_one_or_none()
        if not existing_batch:
            raise HTTPException(404, "Batch not found")
        if existing_batch.status != "active":
            raise HTTPException(400, "Cannot add items to a non-active batch")
        if (
            actor is not None
            and existing_batch.created_by_id is not None
            and existing_batch.created_by_id != actor.id
            and not actor.has_permission(Permission.QUEUE_UPDATE_ALL.value)
        ):
            raise HTTPException(404, "Batch not found")
        batch = existing_batch
        batch_id = existing_batch.id

    # Create batch if quantity > 1 and no batch_id provided
    if batch_id is None and quantity > 1:
        # Derive batch name from source file
        batch_name_base = "Batch"
        if archive:
            batch_name_base = archive.print_name or archive.filename or "Batch"
        elif library_file:
            if library_file.file_metadata:
                batch_name_base = library_file.file_metadata.get("print_name") or library_file.filename
            else:
                batch_name_base = library_file.filename
        elif variant_specs:
            # A cross-model job carries neither archive_id nor library_file_id --
            # the candidates are the files (#671) -- so both branches above miss
            # and every such batch was named "Batch". Unreachable until the print
            # dialog could ask for more than one copy of one (#3101). Name it
            # after the first candidate, which is what the dialog names the job
            # after and what the resolver prefers when both printers are free.
            first_file = variant_specs[0][1]
            batch_name_base = (first_file.file_metadata or {}).get("print_name") or first_file.filename or "Batch"
        batch_name_base = batch_name_base.replace(".gcode.3mf", "").replace(".3mf", "")

        batch = PrintBatch(
            name=f"{batch_name_base} ×{quantity}",
            archive_id=data.archive_id,
            library_file_id=data.library_file_id,
            quantity=quantity,
            status="active",
            created_by_id=actor.id if actor else None,
        )
        db.add(batch)
        await db.flush()  # Get batch.id before creating items
        batch_id = batch.id

    # Positions are one sequence across every pending item (#3200), so a new
    # item lands relative to the whole list, not to its printer's share of it.
    queue_scope = (PrintQueueItem.status == "pending",)
    await lock_queue_positions(db)

    insert_position = max(1, data.insert_position or 1)
    if data.insert_at_top or data.insert_position is not None:
        max_pos = await max_queue_position(db)
        insert_position = min(insert_position, max_pos + 1)
        await db.execute(
            update(PrintQueueItem)
            .where(*queue_scope)
            .where(PrintQueueItem.position >= insert_position)
            .values(position=PrintQueueItem.position + quantity)
        )
        start_position = insert_position
    else:
        start_position = await max_queue_position(db) + 1

    # Resolve print_time_seconds for SJF scheduling (cache on item at creation)
    cached_print_time = None
    if archive:
        cached_print_time = archive.print_time_seconds
        if data.plate_id:
            archive_path = settings.base_dir / archive.file_path
            if archive_path.exists():
                plate_time = _extract_print_time_from_3mf(archive_path, data.plate_id)
                if plate_time is not None:
                    cached_print_time = plate_time
    elif library_file:
        if library_file.file_metadata:
            cached_print_time = library_file.file_metadata.get("print_time_seconds")
        if data.plate_id:
            lib_path = Path(library_file.file_path)
            library_file_path = lib_path if lib_path.is_absolute() else settings.base_dir / library_file.file_path
            if library_file_path.exists():
                plate_time = _extract_print_time_from_3mf(library_file_path, data.plate_id)
                if plate_time is not None:
                    cached_print_time = plate_time

    # Validate project exists before insert so a bogus ID yields 404, not an FK-constraint 500
    if data.project_id is not None:
        project_result = await db.execute(select(Project).where(Project.id == data.project_id))
        if not project_result.scalar_one_or_none():
            raise HTTPException(status_code=404, detail="Project not found")

    # Security boundary: the browser's estimated_cost is only a display hint.
    # Budget enforcement and the persisted reservation value must be derived
    # from the server-owned archive/library metadata and spool assignments.
    if variant_specs:
        variant_costs = [
            await estimate_queue_source_cost(
                db,
                library_file=variant_file,
                plate_id=spec.plate_id,
                ams_mapping=spec.ams_mapping,
                printer_id=data.printer_id,
            )
            for spec, variant_file, _model in variant_specs
        ]
        trusted_estimated_cost = (
            max(cost for cost in variant_costs if cost is not None)
            if variant_costs and all(cost is not None for cost in variant_costs)
            else None
        )
    else:
        trusted_estimated_cost = await estimate_queue_source_cost(
            db,
            archive=archive,
            library_file=library_file,
            plate_id=data.plate_id,
            ams_mapping=data.ams_mapping,
            printer_id=data.printer_id,
        )

    await validate_print_budget(
        db,
        cost_center_id=data.cost_center_id,
        estimated_cost=trusted_estimated_cost,
        current_user=actor,
        quantity=quantity,
    )

    ams_mapping_json = json.dumps(data.ams_mapping) if data.ams_mapping else None
    # Same Text-as-JSON convention for the rack-position pick (#1784).
    nozzle_rack_choice_json = json.dumps(data.nozzle_rack_choice) if data.nozzle_rack_choice else None
    # Reprint fallback: the caller didn't specify an explicit ams_mapping (no
    # per-slot filament-mapping edit was made), but the archive carries the
    # slicer's own live-resolved AMS-slot pick from the original print (see
    # `extra_data.slicer_ams_mapping`, written by the VP-queue path via
    # `_extract_slicer_ams_mapping_json`). Reuse it so the reprint dispatches
    # to the exact same physical spool instead of the scheduler re-deriving a
    # (possibly ambiguous) mapping from just the file's static type/color.
    #
    # Global tray IDs only mean something relative to the specific printer
    # they were resolved against, so this only fires when the reprint targets
    # that exact printer (`extra_data.slicer_ams_mapping.printer_id`) — never
    # for a model-based dispatch (data.printer_id is None) or a reprint aimed
    # at a different printer, where the same tray number can hold a
    # completely different spool (#2700 review).
    #
    # It also stands down when the request carries force-color-match overrides:
    # those are the caller asking the scheduler to match strictly against the
    # printer's live trays, and they are only ever applied inside
    # `_compute_ams_mapping_for_printer` — the function a stored mapping makes
    # the scheduler skip. Same precedence as the VP-side toggle pair (#2700
    # review).
    #
    # Note this is otherwise unconditional — it applies regardless of whether
    # the physical spool in that slot has changed since the original print.
    # #1308 covers re-verifying a stored mapping against live AMS state at
    # dispatch time; that check is a separate PR and, once merged, will also
    # catch a stale slot inherited through this fallback.
    wants_live_color_match = any(
        isinstance(o, dict) and o.get("force_color_match") for o in (data.filament_overrides or [])
    )
    if (
        ams_mapping_json is None
        and not wants_live_color_match
        and archive
        and archive.extra_data
        and data.printer_id is not None
    ):
        saved = archive.extra_data.get("slicer_ams_mapping")
        if (
            isinstance(saved, dict)
            and saved.get("printer_id") == data.printer_id
            and isinstance(saved.get("mapping"), list)
            and saved["mapping"]
        ):
            ams_mapping_json = json.dumps(saved["mapping"])
    items = []
    for i in range(quantity):
        item = PrintQueueItem(
            printer_id=data.printer_id,
            target_model=target_model_norm,
            target_location=data.target_location,
            required_filament_types=required_filament_types,
            filament_overrides=filament_overrides_json,
            archive_id=data.archive_id,
            library_file_id=data.library_file_id,
            cost_center_id=data.cost_center_id,
            estimated_cost=trusted_estimated_cost,
            scheduled_time=data.scheduled_time,
            require_previous_success=data.require_previous_success,
            auto_off_after=data.auto_off_after,
            # Without queue:start_unreviewed the job waits for someone to
            # start it, whatever the request asked for (#1620)
            manual_start=data.manual_start or review_required,
            skip_filament_check=data.skip_filament_check,
            ams_mapping=ams_mapping_json,
            nozzle_rack_choice=nozzle_rack_choice_json,
            plate_id=data.plate_id,
            bed_levelling=data.bed_levelling,
            flow_cali=data.flow_cali,
            vibration_cali=data.vibration_cali,
            layer_inspect=data.layer_inspect,
            timelapse=data.timelapse,
            use_ams=data.use_ams,
            nozzle_offset_cali=data.nozzle_offset_cali,
            confirm_outcome=data.confirm_outcome,
            preheat_override=data.preheat_override,
            preheat_chamber_target_override=data.preheat_chamber_target_override,
            gcode_injection=data.gcode_injection,
            cleanup_library_after_dispatch=data.cleanup_library_after_dispatch,
            project_id=data.project_id,
            position=start_position + i,
            status="pending",
            created_by_id=actor.id if actor else None,
            batch_id=batch_id,
            print_time_seconds=cached_print_time,
        )
        db.add(item)
        items.append(item)

    if variant_specs:
        variant_values = [
            _variant_values(spec, library_file, model, position)
            for position, (spec, library_file, model) in enumerate(variant_specs)
        ]
        # SJF orders pending items before any printer is known, so the row carries
        # the shortest candidate's estimate. Resolution replaces it with the one
        # that actually runs.
        estimates = [v["print_time_seconds"] for v in variant_values if v["print_time_seconds"]]
        for item in items:
            # Each copy in a quantity>1 batch gets its own candidate rows —
            # attempt counts are per-item, and two copies must be free to land on
            # different printers.
            item.variants.extend(PrintQueueVariant(**values) for values in variant_values)
            item.print_time_seconds = min(estimates) if estimates else None

    await db.commit()

    # Refresh the first item for the response
    item = items[0]
    await db.refresh(item)
    await db.refresh(item, ["archive", "printer", "library_file", "created_by", "batch"])

    source_name = f"archive {data.archive_id}" if data.archive_id else f"library file {data.library_file_id}"
    target_desc = data.printer_id or (f"model {target_model_norm}" if target_model_norm else "unassigned")
    qty_desc = f" (×{quantity})" if quantity > 1 else ""
    logger.info("Added %s to queue for %s%s", source_name, target_desc, qty_desc)

    # MQTT relay - publish queue job added
    try:
        from backend.app.services.mqtt_relay import mqtt_relay

        await mqtt_relay.on_queue_job_added(
            job_id=item.id,
            filename=item.archive.filename if item.archive else "",
            printer_id=item.printer_id,
            printer_name=item.printer.name if item.printer else None,
        )
    except Exception:
        pass  # Don't fail queue add if MQTT fails

    # Send notification for job added
    try:
        job_name = (
            item.archive.filename
            if item.archive
            else item.library_file.filename
            if item.library_file
            else f"Job #{item.id}"
        )
        job_name = job_name.replace(".gcode.3mf", "").replace(".3mf", "")
        if quantity > 1:
            job_name = f"{job_name} ×{quantity}"
        target = (
            item.printer.name if item.printer else (f"Any {item.target_model}" if target_model_norm else "Unassigned")
        )
        await notification_service.on_queue_job_added(
            job_name=job_name,
            target=target,
            db=db,
            printer_id=item.printer_id,
            printer_name=item.printer.name if item.printer else None,
        )
    except Exception:
        pass  # Don't fail queue add if notification fails

    return _enrich_response(item)


@router.patch("/bulk", response_model=PrintQueueBulkUpdateResponse)
async def bulk_update_queue_items(
    data: PrintQueueBulkUpdate,
    db: AsyncSession = Depends(get_db),
    auth_result: tuple[User | None, bool] = Depends(
        require_ownership_permission(
            Permission.QUEUE_UPDATE_ALL,
            Permission.QUEUE_UPDATE_OWN,
        )
    ),
    printer_scope: PrinterScope = RequestPrinterScope,
    actor: User | ApiKeyActor | None = RequestActor,
):
    """Bulk update multiple queue items with the same values.

    Only pending items can be updated. Non-pending items are skipped.
    Items not owned by the user are also skipped (unless user has *_all permission).
    """
    user, can_modify_all = auth_result

    if not data.item_ids:
        raise HTTPException(400, "No item IDs provided")

    # Get fields to update (exclude item_ids and unset fields)
    update_data = data.model_dump(exclude={"item_ids"}, exclude_unset=True)
    if not update_data:
        raise HTTPException(400, "No fields to update")

    # Validate printer_id if being changed
    if "printer_id" in update_data and update_data["printer_id"] is not None:
        printer_scope.ensure(update_data["printer_id"])
        result = await db.execute(select(Printer).where(Printer.id == update_data["printer_id"]))
        if not result.scalar_one_or_none():
            raise HTTPException(400, "Printer not found")

    # Fetch all items, minus any bound to a printer the caller can't see (#1727)
    result = await db.execute(select(PrintQueueItem).where(PrintQueueItem.id.in_(data.item_ids)))
    items = [item for item in result.scalars().all() if printer_scope.allows(item.printer_id)]

    updated_count = 0
    skipped_count = 0
    validates_billing_fields = "cost_center_id" in update_data or "estimated_cost" in update_data

    for item in items:
        # Skip non-pending rows and rows a dispatch worker has claimed (#2615) —
        # editing a claimed row mid-upload would split it from the in-flight
        # dispatch, so it's excluded from the bulk change (cancel to move it).
        if item.status != "pending" or item.dispatching_at is not None:
            skipped_count += 1
            continue

        # Ownership check
        if not can_modify_all and item.created_by_id != user.id:
            skipped_count += 1
            continue

        # Clearing "wait for manual start" starts the job, which needs the
        # same right as the start button (#1620)
        if (
            item.manual_start
            and "manual_start" in update_data
            and not update_data["manual_start"]
            and not may_start_queue_item(user, can_modify_all, item.created_by_id)
        ):
            skipped_count += 1
            continue

        item_update_data = update_data.copy()
        if validates_billing_fields:
            trusted_estimated_cost = await _trusted_item_estimated_cost(
                db,
                item,
                printer_id=item_update_data.get("printer_id", item.printer_id),
                plate_id=item.plate_id,
                ams_mapping=item.ams_mapping,
            )
            item_update_data["estimated_cost"] = trusted_estimated_cost
            await validate_print_budget(
                db,
                cost_center_id=item_update_data.get("cost_center_id", item.cost_center_id),
                estimated_cost=trusted_estimated_cost,
                current_user=actor,
                exclude_queue_item_id=item.id,
            )

        for field, value in item_update_data.items():
            setattr(item, field, value)
        updated_count += 1

    await db.commit()

    logger.info("Bulk updated %s queue items, skipped %s", updated_count, skipped_count)
    return PrintQueueBulkUpdateResponse(
        updated_count=updated_count,
        skipped_count=skipped_count,
        message=f"Updated {updated_count} items"
        + (f", skipped {skipped_count} non-pending/not-owned" if skipped_count else ""),
    )


# --- Batch endpoints ---


def _validate_plate_targets(
    plates: list[PrintBatchPlateTarget] | None,
) -> list[PrintBatchPlateTarget] | None:
    """Reject duplicate plates and orders that ask for nothing at all.

    A duplicate would violate the (batch_id, plate_id) unique constraint at
    flush time — and on SQLite/PostgreSQL a NULL plate_id slips past that
    constraint entirely, so the check has to happen here to catch two
    "whole file" rows in one order.
    """
    if plates is None:
        return None
    if not plates:
        raise HTTPException(400, "plates must contain at least one plate")

    seen: set[int | None] = set()
    for target in plates:
        if target.plate_id in seen:
            label = target.plate_id if target.plate_id is not None else "whole file"
            raise HTTPException(400, f"Duplicate plate in order: {label}")
        seen.add(target.plate_id)

    if all(target.quantity_target == 0 for target in plates):
        raise HTTPException(400, "Order must request at least one print")
    return plates


async def _validate_batch_project(
    db: AsyncSession, project_id: int | None, current_user: User | ApiKeyActor | None
) -> None:
    """404 on a bogus project id rather than letting the FK blow up as a 500."""
    if project_id is None:
        return
    result = await db.execute(select(Project).where(Project.id == project_id))
    if not result.scalar_one_or_none():
        raise HTTPException(404, "Project not found")


async def _load_batch_for_write(
    db: AsyncSession, batch_id: int, current_user: User | ApiKeyActor | None, permission: Permission
) -> PrintBatch:
    """Fetch a batch the caller is allowed to modify, or 404.

    404 rather than 403 on the ownership miss, matching the rest of this
    module: a 403 would confirm the id exists to someone enumerating.
    """
    result = await db.execute(
        select(PrintBatch).options(selectinload(PrintBatch.plates)).where(PrintBatch.id == batch_id)
    )
    batch = result.scalar_one_or_none()
    if not batch:
        raise HTTPException(404, "Batch not found")
    if (
        current_user is not None
        and batch.created_by_id is not None
        and batch.created_by_id != current_user.id
        and not current_user.has_permission(permission.value)
    ):
        raise HTTPException(404, "Batch not found")
    return batch


# Deliberately without the existing batch's id: the caller may not be allowed
# to read it. They can look it up by the pair, which applies the usual
# ownership rules.
_EXTERNAL_REF_TAKEN = "A batch for this external_source and external_ref already exists"


async def _ensure_batch_in_scope(db: AsyncSession, batch_id: int, printer_scope: PrinterScope) -> None:
    """404 a batch holding jobs on printers the caller can't see (#1727).

    Cancelling or ungrouping acts on every member, so doing it to only the
    visible ones would leave a batch that reads cancelled with jobs still
    pending, and refusing outright is the honest answer.
    """
    if printer_scope.is_unrestricted:
        return
    result = await db.execute(select(PrintQueueItem.printer_id).where(PrintQueueItem.batch_id == batch_id))
    if not all(printer_scope.allows(pid) for (pid,) in result.all()):
        raise HTTPException(404, "Batch not found")


@router.post("/batches", response_model=PrintBatchResponse)
async def create_batch(
    data: PrintBatchCreate,
    db: AsyncSession = Depends(get_db),
    current_user: User | None = RequirePermissionIfAuthEnabled(Permission.QUEUE_CREATE),
    printer_scope: PrinterScope = RequestPrinterScope,
    actor: User | ApiKeyActor | None = RequestActor,
):
    """Create a batch.

    Two modes:
    * ``item_ids`` provided: assign the listed pending queue items to a new
      batch ("Group as batch" UI action).
    * ``item_ids`` omitted/empty: create an empty batch so the client can
      pass the returned ``id`` on subsequent ``POST /queue/`` calls. Used by
      the multi-plate auto-batch flow in PrintModal.

    ``plates`` turns the batch into an order with per-plate targets (#342):
    progress is then measured against what was asked for rather than against
    what happened to be queued, so a failed run still counts as owed. Omitting
    it keeps the pre-#342 behaviour exactly.
    """
    if not data.name or not data.name.strip():
        raise HTTPException(400, "Batch name is required")

    plate_targets = _validate_plate_targets(data.plates)
    await _validate_batch_project(db, data.project_id, actor)
    if data.external_source is not None:
        existing = await db.execute(
            select(PrintBatch.id).where(
                PrintBatch.external_source == data.external_source,
                PrintBatch.external_ref == data.external_ref,
            )
        )
        if existing.scalar_one_or_none() is not None:
            raise HTTPException(409, _EXTERNAL_REF_TAKEN)

    batch = PrintBatch(
        name=data.name.strip()[:255],
        archive_id=data.archive_id,
        library_file_id=data.library_file_id,
        quantity=len(data.item_ids) if data.item_ids else 1,
        status="active",
        created_by_id=actor.id if actor else None,
        project_id=data.project_id,
        due_date=data.due_date,
        notes=data.notes,
        external_source=data.external_source,
        external_ref=data.external_ref,
    )
    db.add(batch)
    try:
        await db.flush()  # Need batch.id before assigning to items
    except IntegrityError:
        # Lost a race with a concurrent create for the same external record:
        # the unique index caught what the lookup above could not.
        await db.rollback()
        raise HTTPException(409, _EXTERNAL_REF_TAKEN) from None

    if plate_targets is not None:
        for target in plate_targets:
            db.add(
                PrintBatchPlate(
                    batch_id=batch.id,
                    plate_id=target.plate_id,
                    plate_name=target.plate_name,
                    quantity_target=target.quantity_target,
                    sort_order=target.sort_order,
                )
            )
        # The legacy `quantity` column is display-only; keep it meaningful for
        # anything still reading it by making it the order's total.
        batch.quantity = max(1, sum(t.quantity_target for t in plate_targets))

    assigned = 0
    if data.item_ids:
        result = await db.execute(select(PrintQueueItem).where(PrintQueueItem.id.in_(data.item_ids)))
        items = result.scalars().all()
        for item in items:
            if item.status != "pending":
                continue
            if item.batch_id is not None:
                continue
            if not printer_scope.allows(item.printer_id):
                continue
            if (
                actor is not None
                and item.created_by_id != actor.id
                and not actor.has_permission(Permission.QUEUE_UPDATE_ALL.value)
            ):
                continue
            item.batch_id = batch.id
            assigned += 1
        batch.quantity = max(assigned, 1)

    await db.commit()
    await db.refresh(batch)

    logger.info("Created batch %s '%s' with %s assigned items", batch.id, batch.name, assigned)
    return await _build_batch_response(db, batch)


@router.patch("/batches/{batch_id}", response_model=PrintBatchResponse)
async def update_batch(
    batch_id: int,
    data: PrintBatchUpdate,
    db: AsyncSession = Depends(get_db),
    current_user: User | None = RequirePermissionIfAuthEnabled(Permission.QUEUE_UPDATE_OWN),
    actor: User | ApiKeyActor | None = RequestActor,
):
    """Edit an order's header or its per-plate targets while it runs (#342).

    Production requirements change mid-run, so targets are editable. Lowering a
    target below what has already been dispatched is allowed and simply leaves
    ``remaining`` at zero — cancelling the surplus queue items is a separate,
    explicit action, because silently deleting queued work on a number change
    would be a nasty surprise.
    """
    batch = await _load_batch_for_write(db, batch_id, actor, Permission.QUEUE_UPDATE_ALL)

    plate_targets = _validate_plate_targets(data.plates)
    if data.project_id is not None:
        await _validate_batch_project(db, data.project_id, actor)

    if data.name is not None:
        if not data.name.strip():
            raise HTTPException(400, "Batch name is required")
        batch.name = data.name.strip()[:255]
    if data.project_id is not None:
        batch.project_id = data.project_id
    if data.due_date is not None:
        batch.due_date = data.due_date
    if data.notes is not None:
        batch.notes = data.notes
    if data.status is not None:
        batch.status = data.status

    if plate_targets is not None:
        existing = {row.plate_id: row for row in batch.plates}
        for target in plate_targets:
            row = existing.pop(target.plate_id, None)
            if row is None:
                db.add(
                    PrintBatchPlate(
                        batch_id=batch.id,
                        plate_id=target.plate_id,
                        plate_name=target.plate_name,
                        quantity_target=target.quantity_target,
                        sort_order=target.sort_order,
                    )
                )
            else:
                row.quantity_target = target.quantity_target
                row.sort_order = target.sort_order
                if target.plate_name is not None:
                    row.plate_name = target.plate_name
        # Plates absent from the payload are dropped — the list is the order.
        for orphan in existing.values():
            await db.delete(orphan)
        batch.quantity = max(1, sum(t.quantity_target for t in plate_targets))

    await db.flush()
    await db.refresh(batch)
    # Raising a target on a finished order reopens it; lowering one on a
    # running order can complete it.
    await refresh_batch_status(db, batch)
    await db.commit()
    await db.refresh(batch)

    logger.info("Updated batch %s", batch.id)
    return await _build_batch_response(db, batch)


@router.post("/batches/{batch_id}/dispatch", response_model=PrintBatchResponse)
async def dispatch_batch(
    batch_id: int,
    data: PrintBatchDispatchRequest,
    db: AsyncSession = Depends(get_db),
    current_user: User | None = RequirePermissionIfAuthEnabled(Permission.QUEUE_CREATE),
    printer_scope: PrinterScope = RequestPrinterScope,
    review_required: bool = QueueReviewRequired,
    actor: User | ApiKeyActor | None = RequestActor,
):
    """Queue the runs this order still owes (#342).

    Each new item is cloned from the most recent item for the same plate in
    this batch, so it inherits the printer/model target, AMS mapping, filament
    overrides and print options the user already chose — and the validation
    those went through at creation time.
    """
    batch = await _load_batch_for_write(db, batch_id, actor, Permission.QUEUE_UPDATE_ALL)
    if batch.status == "cancelled":
        raise HTTPException(400, "Cannot dispatch a cancelled batch")

    # Dispatch starts prints, so it must not be a weaker door than POST /queue/.
    await _assert_can_dispatch_batch_sources(db, batch.id, actor)

    try:
        created = await dispatch_remaining(
            db,
            batch,
            plate_id=data.plate_id,
            only_plate=data.only_plate,
            limit=data.limit,
            created_by_id=actor.id if actor else None,
        )
    except BatchDispatchError as exc:
        raise HTTPException(400, str(exc)) from exc
    # The clones inherit their templates' targets, which whoever queued them
    # was allowed; the dispatcher has to be allowed them too (#1727).
    try:
        for item in created:
            printer_scope.ensure(item.printer_id)
            if item.printer_id is None:
                ensure_model_target_allowed(current_user, printer_scope)
    except HTTPException:
        await db.rollback()
        raise
    # A clone copies its template's "wait for manual start", which is off once
    # staff started the template; the dispatcher's own jobs still wait (#1620)
    if review_required:
        for item in created:
            item.manual_start = True

    await db.commit()
    await db.refresh(batch)

    logger.info("Batch %s dispatched %d item(s)", batch.id, len(created))
    return await _build_batch_response(db, batch)


@router.post("/batches/{batch_id}/ungroup", response_model=PrintBatchUngroupResponse)
async def ungroup_batch(
    batch_id: int,
    db: AsyncSession = Depends(get_db),
    current_user: User | None = RequirePermissionIfAuthEnabled(Permission.QUEUE_UPDATE_OWN),
    printer_scope: PrinterScope = RequestPrinterScope,
    actor: User | ApiKeyActor | None = RequestActor,
):
    """Disband a batch: clear batch_id from all members and delete the batch row.

    Items stay in the queue. Only ungroups items the caller owns (unless they
    hold QUEUE_UPDATE_ALL). A batch with all members ungrouped is deleted.
    """
    result = await db.execute(select(PrintBatch).where(PrintBatch.id == batch_id))
    batch = result.scalar_one_or_none()
    if not batch:
        raise HTTPException(404, "Batch not found")

    can_modify_all = actor is None or actor.has_permission(Permission.QUEUE_UPDATE_ALL.value)
    if not can_modify_all and batch.created_by_id != (actor.id if actor else None):
        raise HTTPException(404, "Batch not found")
    await _ensure_batch_in_scope(db, batch_id, printer_scope)

    result = await db.execute(select(PrintQueueItem).where(PrintQueueItem.batch_id == batch_id))
    items = result.scalars().all()
    ungrouped = 0
    remaining = 0
    for item in items:
        if not can_modify_all and item.created_by_id != (actor.id if actor else None):
            remaining += 1
            continue
        item.batch_id = None
        ungrouped += 1

    # Delete the batch row only when all members were ungrouped — otherwise it
    # still owns the items the caller couldn't touch.
    if remaining == 0:
        await db.delete(batch)

    await db.commit()

    logger.info("Ungrouped batch %s (%s items)", batch_id, ungrouped)
    return PrintBatchUngroupResponse(
        ungrouped_count=ungrouped,
        message=f"Ungrouped {ungrouped} item(s)",
    )


@router.get("/batches", response_model=list[PrintBatchResponse])
async def list_batches(
    status: str | None = Query(None, description="Filter by status (active, completed, cancelled)"),
    external_source: str | None = Query(None, description="Filter by the integration that created the batch"),
    external_ref: str | None = Query(None, description="Filter by the external record the batch fulfils"),
    db: AsyncSession = Depends(get_db),
    auth_result: tuple[User | None, bool] = Depends(
        require_ownership_permission(
            Permission.QUEUE_READ_ALL,
            Permission.QUEUE_READ_OWN,
        )
    ),
):
    """List print batches with progress stats.

    Batches with neither queue items nor per-plate targets are omitted. Those
    are empty shells — a grouping whose items were deleted with their source
    archive, or a create that never got as far as adding any — and they carry
    nothing to show, track or dispatch. A brand-new order is still listed
    before its first dispatch, because its targets say what it owes.
    """
    current_user, can_read_all = auth_result
    query = (
        select(PrintBatch)
        .where(
            select(PrintQueueItem.id).where(PrintQueueItem.batch_id == PrintBatch.id).exists()
            | select(PrintBatchPlate.id).where(PrintBatchPlate.batch_id == PrintBatch.id).exists()
        )
        .order_by(PrintBatch.created_at.desc())
    )
    if status:
        query = query.where(PrintBatch.status == status)
    if external_source is not None:
        query = query.where(PrintBatch.external_source == external_source)
    if external_ref is not None:
        query = query.where(PrintBatch.external_ref == external_ref)
    if current_user is not None and not can_read_all:
        query = query.where(PrintBatch.created_by_id == current_user.id)
    result = await db.execute(query)
    batches = result.scalars().all()

    # Resolve creator names in one query rather than one per batch.
    creator_ids = {b.created_by_id for b in batches if b.created_by_id is not None}
    usernames: dict[int, str] = {}
    if creator_ids:
        rows = await db.execute(select(User.id, User.username).where(User.id.in_(creator_ids)))
        usernames = {row[0]: row[1] for row in rows.all()}

    return [await _build_batch_response(db, batch, usernames=usernames) for batch in batches]


@router.get("/batches/{batch_id}", response_model=PrintBatchResponse)
async def get_batch(
    batch_id: int,
    db: AsyncSession = Depends(get_db),
    auth_result: tuple[User | None, bool] = Depends(
        require_ownership_permission(
            Permission.QUEUE_READ_ALL,
            Permission.QUEUE_READ_OWN,
        )
    ),
):
    """Get a print batch with progress stats."""
    current_user, can_read_all = auth_result
    result = await db.execute(select(PrintBatch).where(PrintBatch.id == batch_id))
    batch = result.scalar_one_or_none()
    if not batch:
        raise HTTPException(404, "Batch not found")
    if (
        current_user is not None
        and not can_read_all
        and (batch.created_by_id is None or batch.created_by_id != current_user.id)
    ):
        raise HTTPException(404, "Batch not found")
    return await _build_batch_response(db, batch)


@router.delete("/batches/{batch_id}")
async def cancel_batch(
    batch_id: int,
    db: AsyncSession = Depends(get_db),
    current_user: User | None = RequirePermissionIfAuthEnabled(Permission.QUEUE_DELETE_ALL),
    printer_scope: PrinterScope = RequestPrinterScope,
):
    """Cancel all pending items in a batch and mark batch as cancelled."""
    result = await db.execute(select(PrintBatch).where(PrintBatch.id == batch_id))
    batch = result.scalar_one_or_none()
    if not batch:
        raise HTTPException(404, "Batch not found")
    await _ensure_batch_in_scope(db, batch_id, printer_scope)

    # Cancel all pending queue items in this batch
    result = await db.execute(
        select(PrintQueueItem).where(and_(PrintQueueItem.batch_id == batch_id, PrintQueueItem.status == "pending"))
    )
    pending_items = result.scalars().all()
    cancelled_count = 0
    cancelled_ids: list[int] = []
    for item in pending_items:
        item.status = "cancelled"
        await release_budget_reservation(
            db,
            source_type="print_queue",
            source_id=item.id,
            status="released",
        )
        cancelled_ids.append(item.id)
        cancelled_count += 1

    batch.status = "cancelled"
    await db.commit()

    # Same as the single-item path: a dispatch already preheating for one of
    # these cannot see the status change on its own (#2727).
    from backend.app.services.print_scheduler import scheduler as _scheduler

    for _cancelled_id in cancelled_ids:
        _scheduler.notify_dispatch_cancelled(_cancelled_id)

    return {"message": f"Batch cancelled, {cancelled_count} pending items cancelled"}


async def _build_batch_response(
    db: AsyncSession, batch: PrintBatch, *, usernames: dict[int, str] | None = None
) -> PrintBatchResponse:
    """Build a batch response with per-plate progress derived from queue items.

    ``usernames`` lets the list endpoint resolve every creator in one query
    instead of one per batch.
    """
    progress = await load_progress(db, batch)

    created_by_username = None
    if batch.created_by_id:
        if usernames is not None:
            created_by_username = usernames.get(batch.created_by_id)
        else:
            result = await db.execute(select(User).where(User.id == batch.created_by_id))
            user = result.scalar_one_or_none()
            if user:
                created_by_username = user.username

    return PrintBatchResponse(
        id=batch.id,
        name=batch.name,
        archive_id=batch.archive_id,
        library_file_id=batch.library_file_id,
        quantity=batch.quantity,
        status=batch.status,
        created_at=batch.created_at,
        completed_at=batch.completed_at,
        created_by_id=batch.created_by_id,
        created_by_username=created_by_username,
        project_id=batch.project_id,
        due_date=batch.due_date,
        notes=batch.notes,
        external_source=batch.external_source,
        external_ref=batch.external_ref,
        pending_count=progress.pending,
        printing_count=progress.printing,
        completed_count=progress.completed,
        failed_count=progress.failed,
        cancelled_count=progress.cancelled,
        skipped_count=progress.skipped,
        has_targets=progress.has_targets,
        target_count=progress.target,
        remaining_count=progress.remaining,
        dispatchable_count=progress.dispatchable_remaining,
        actual_cost=progress.actual_cost,
        estimated_remaining_cost=progress.estimated_remaining_cost,
        filament_used_grams=progress.filament_used_grams,
        print_time_seconds=progress.print_time_seconds,
        plates=[
            PrintBatchPlateProgress(
                plate_id=plate.plate_id,
                plate_name=plate.plate_name,
                quantity_target=plate.quantity_target,
                dispatched=plate.dispatched,
                remaining=plate.remaining,
                pending_count=plate.pending,
                printing_count=plate.printing,
                completed_count=plate.completed,
                failed_count=plate.failed,
                cancelled_count=plate.cancelled,
                skipped_count=plate.skipped,
                actual_cost=plate.actual_cost,
                estimated_remaining_cost=plate.estimated_remaining_cost,
                filament_used_grams=plate.filament_used_grams,
                print_time_seconds=plate.print_time_seconds,
                can_dispatch=plate.can_dispatch,
            )
            for plate in progress.plates
        ],
    )


@router.get("/{item_id}", response_model=PrintQueueItemResponse)
async def get_queue_item(
    item_id: int,
    db: AsyncSession = Depends(get_db),
    auth_result: tuple[User | None, bool] = Depends(
        require_ownership_permission(
            Permission.QUEUE_READ_ALL,
            Permission.QUEUE_READ_OWN,
        )
    ),
    printer_scope: PrinterScope = RequestPrinterScope,
):
    """Get a specific queue item."""
    current_user, can_read_all = auth_result
    result = await db.execute(
        select(PrintQueueItem)
        .options(
            selectinload(PrintQueueItem.archive),
            selectinload(PrintQueueItem.printer),
            selectinload(PrintQueueItem.library_file),
            selectinload(PrintQueueItem.created_by),
            selectinload(PrintQueueItem.batch),
            # Cross-model candidates (#671) and their files, for the card label.
            selectinload(PrintQueueItem.variants).selectinload(PrintQueueVariant.library_file),
        )
        .where(PrintQueueItem.id == item_id)
    )
    item = result.scalar_one_or_none()
    if not item or not printer_scope.allows(item.printer_id):
        raise HTTPException(404, "Queue item not found")
    if (
        current_user is not None
        and not can_read_all
        and (item.created_by_id is None or item.created_by_id != current_user.id)
    ):
        raise HTTPException(404, "Queue item not found")
    return _enrich_response(item)


@router.patch("/{item_id}", response_model=PrintQueueItemResponse)
async def update_queue_item(
    item_id: int,
    data: PrintQueueItemUpdate,
    db: AsyncSession = Depends(get_db),
    auth_result: tuple[User | None, bool] = Depends(
        require_ownership_permission(
            Permission.QUEUE_UPDATE_ALL,
            Permission.QUEUE_UPDATE_OWN,
        )
    ),
    printer_scope: PrinterScope = RequestPrinterScope,
    actor: User | ApiKeyActor | None = RequestActor,
):
    """Update a queue item."""
    user, can_modify_all = auth_result

    result = await db.execute(
        select(PrintQueueItem)
        # Needed by the cross-model guard below, and by the response builder —
        # without it _variant_summaries falls back to [] and a PATCH would strip
        # the alternatives out of the payload it echoes back.
        .options(selectinload(PrintQueueItem.variants).selectinload(PrintQueueVariant.library_file))
        .where(PrintQueueItem.id == item_id)
    )
    item = result.scalar_one_or_none()
    if not item or not printer_scope.allows(item.printer_id):
        raise HTTPException(404, "Queue item not found")

    # Ownership check
    if not can_modify_all:
        if item.created_by_id != user.id:
            raise HTTPException(403, "You can only update your own queue items")

    if item.status != "pending":
        raise HTTPException(400, "Can only update pending items")

    # Dispatch claim (#2615): the row is pending but a scheduler worker has
    # already claimed it and is uploading to its printer. Editing now (e.g.
    # reassigning printer_id) would split the queue row from the in-flight
    # archive/expected-print/physical command. Reject until dispatch finishes;
    # to move it, cancel first (the coordinated escape) and re-queue.
    if item.dispatching_at is not None:
        raise HTTPException(409, "Item is being dispatched — cancel it first to make changes")

    update_data = data.model_dump(exclude_unset=True)

    # Clearing "wait for manual start" starts the job, which needs the same
    # right as the start button (#1620)
    if (
        item.manual_start
        and "manual_start" in update_data
        and not update_data["manual_start"]
        and not may_start_queue_item(user, can_modify_all, item.created_by_id)
    ):
        raise HTTPException(403, _AWAITING_REVIEW)

    # Normalize target_model if being updated (see add_to_queue for why the
    # code map has to run first).
    if "target_model" in update_data and update_data["target_model"]:
        update_data["target_model"] = normalize_model_name(update_data["target_model"])
        ensure_model_target_allowed(user, printer_scope)

    # A cross-model item (#671) owns its own printer decision: each candidate
    # carries its model, and the resolver folds the winner onto the row at
    # dispatch. Assigning a printer here would leave a row with variants *and* a
    # printer_id, and the fixed-printer branch of the scheduler wins that race —
    # so it would dispatch a row whose library_file_id is still null and die in
    # the upload. Narrowing target_model is refused for the same reason: it
    # would silently discard every alternative the user queued.
    #
    # Compared against the current value rather than merely present, because the
    # edit dialog re-sends target_model unchanged on every save.
    if item.variants:
        for field in ("printer_id", "target_model"):
            if field in update_data and update_data[field] != getattr(item, field):
                raise HTTPException(
                    400,
                    "This job has printer alternatives — remove them before assigning a printer or model",
                )

    # Cannot specify both printer_id and target_model
    new_printer_id = update_data.get("printer_id", item.printer_id)
    new_target_model = update_data.get("target_model", item.target_model)
    if new_printer_id and new_target_model:
        raise HTTPException(400, "Cannot specify both printer_id and target_model")

    # Validate new printer_id if being changed (and not None)
    if "printer_id" in update_data and update_data["printer_id"] is not None:
        printer_scope.ensure(update_data["printer_id"])
        result = await db.execute(select(Printer).where(Printer.id == update_data["printer_id"]))
        if not result.scalar_one_or_none():
            raise HTTPException(400, "Printer not found")

    # Validate target_model has active printers
    if "target_model" in update_data and update_data["target_model"]:
        result = await db.execute(
            select(Printer).where(Printer.model == update_data["target_model"]).where(Printer.is_active == True)  # noqa: E712
        )
        if not result.scalars().first():
            raise HTTPException(400, f"No active printers for model: {update_data['target_model']}")

        # Cross-model safety gate (#2578) — same check as the create route, so
        # a mismatched target can't be introduced by editing either.
        sliced_for = None
        if item.archive_id:
            result = await db.execute(select(PrintArchive.sliced_for_model).where(PrintArchive.id == item.archive_id))
            sliced_for = result.scalar_one_or_none()
        elif item.library_file_id:
            result = await db.execute(select(LibraryFile).where(LibraryFile.id == item.library_file_id))
            lib = result.scalar_one_or_none()
            if lib and lib.file_metadata:
                sliced_for = lib.file_metadata.get("sliced_for_model")
        if not is_gcode_compatible(sliced_for, update_data["target_model"]):
            raise HTTPException(
                400,
                f"File was sliced for {sliced_for} and cannot be dispatched to {update_data['target_model']} printers",
            )

    # A job that now waits for any printer of a model keeps nothing that was
    # resolved against one printer: the scheduler picks the printer and maps
    # its trays then (#3239). The edit dialog omits both fields on such a move.
    if new_target_model and not new_printer_id:
        update_data["ams_mapping"] = None
        if item.printer_id and "skip_filament_check" not in update_data:
            update_data["skip_filament_check"] = False

    # Serialize ams_mapping to JSON for TEXT column storage
    if "ams_mapping" in update_data:
        update_data["ams_mapping"] = json.dumps(update_data["ams_mapping"]) if update_data["ams_mapping"] else None

    # Serialize filament_overrides to JSON for TEXT column storage, keeping only
    # the slots this item's plate actually prints (#2551 — same shared-override
    # list the create path narrows).
    if "filament_overrides" in update_data:
        overrides = update_data["filament_overrides"]
        if overrides:
            overrides = overrides_for_plate(
                overrides,
                await _resolve_source_path(db, item),
                update_data.get("plate_id", item.plate_id),
            )
        update_data["filament_overrides"] = json.dumps(overrides) if overrides else None

    # Serialize H2C rack-swap nozzle pick (#1780) to JSON for TEXT column
    # storage; same Text-as-opaque-blob convention as ams_mapping above.
    if "nozzle_mapping" in update_data:
        update_data["nozzle_mapping"] = (
            json.dumps(update_data["nozzle_mapping"]) if update_data["nozzle_mapping"] else None
        )

    # Same Text-as-JSON convention for the rack-position pick (#1784). An empty
    # object clears it, which is how the UI says "assign these for me again".
    if "nozzle_rack_choice" in update_data:
        update_data["nozzle_rack_choice"] = (
            json.dumps(update_data["nozzle_rack_choice"]) if update_data["nozzle_rack_choice"] else None
        )

    trusted_estimated_cost = await _trusted_item_estimated_cost(
        db,
        item,
        printer_id=update_data.get("printer_id", item.printer_id),
        plate_id=update_data.get("plate_id", item.plate_id),
        ams_mapping=update_data.get("ams_mapping", item.ams_mapping),
    )
    update_data["estimated_cost"] = trusted_estimated_cost

    await validate_print_budget(
        db,
        cost_center_id=update_data.get("cost_center_id", item.cost_center_id),
        estimated_cost=trusted_estimated_cost,
        current_user=actor,
        exclude_queue_item_id=item.id,
    )

    # Re-check the dispatch claim right before mutating (#2615). Several awaited
    # validations ran since the guard above, and a scheduler worker may have
    # claimed the row in that gap. A fresh read (item isn't dirty yet, so no
    # autoflush races the check) narrows the window to effectively nothing.
    claimed = (
        await db.execute(select(PrintQueueItem.dispatching_at).where(PrintQueueItem.id == item_id))
    ).scalar_one_or_none()
    if claimed is not None:
        raise HTTPException(409, "Item is being dispatched — cancel it first to make changes")

    for field, value in update_data.items():
        setattr(item, field, value)

    await db.commit()
    await db.refresh(item, ["archive", "printer", "library_file", "created_by", "batch"])

    logger.info("Updated queue item %s", item_id)
    return _enrich_response(item)


async def _remove_queue_item(db: AsyncSession, item: PrintQueueItem) -> bool:
    """Delete *item*, or cancel it when its order still needs it (#2960).

    Returns True when the row was deleted. The caller commits. A completed run
    is always deleted: it is the record of something that was actually made,
    and rewriting it as cancelled would falsify the order's progress.
    """
    item_id = item.id
    keep_as_cancelled = item.status != "completed" and await _is_orders_last_source(db, item)

    await release_budget_reservation(
        db,
        source_type="print_queue",
        source_id=item_id,
        status="released",
    )
    if keep_as_cancelled:
        item.status = "cancelled"
        await db.flush()
        # The order may have been sitting on "completed" if this run's target
        # was met by it; cancelling reopens it.
        await refresh_batch_status_for_item(db, item_id)
    else:
        await db.delete(item)
    return not keep_as_cancelled


def _notify_dispatch_cancelled(item_ids: list[int]) -> None:
    """Stop an in-flight preheat for these items, once their change is committed.

    The dispatch coroutine is parked in a sleep and cannot see the status we
    just wrote (#2727); woken before the commit, it would read the old one.
    """
    from backend.app.services.print_scheduler import scheduler as _scheduler

    for item_id in item_ids:
        _scheduler.notify_dispatch_cancelled(item_id)


@router.delete("/{item_id}")
async def delete_queue_item(
    item_id: int,
    db: AsyncSession = Depends(get_db),
    auth_result: tuple[User | None, bool] = Depends(
        require_ownership_permission(
            Permission.QUEUE_DELETE_ALL,
            Permission.QUEUE_DELETE_OWN,
        )
    ),
    printer_scope: PrinterScope = RequestPrinterScope,
):
    """Remove an item from the queue.

    An order's last surviving run for a plate is cancelled instead of deleted
    (#2960). The row is what a later dispatch clones, so removing it would
    leave the order reporting work outstanding that nothing could ever
    produce. A completed run is exempt: it is the record of something that was
    actually made, and rewriting it as cancelled would falsify the order's
    progress.
    """
    user, can_modify_all = auth_result

    result = await db.execute(select(PrintQueueItem).where(PrintQueueItem.id == item_id))
    item = result.scalar_one_or_none()
    if not item or not printer_scope.allows(item.printer_id):
        raise HTTPException(404, "Queue item not found")

    # Ownership check
    if not can_modify_all:
        if item.created_by_id != user.id:
            raise HTTPException(403, "You can only delete your own queue items")

    if item.status == "printing":
        raise HTTPException(400, "Cannot delete item that is currently printing")

    keep_as_cancelled = not await _remove_queue_item(db, item)
    await db.commit()
    _notify_dispatch_cancelled([item_id])

    if keep_as_cancelled:
        logger.info("Kept queue item %s as cancelled — last source for its batch order plate", item_id)
        return {
            "message": "Item cancelled rather than deleted: it is the only run the order can re-queue this plate from",
            "deleted": False,
        }

    logger.info("Deleted queue item %s", item_id)
    return {"message": "Queue item deleted", "deleted": True}


@router.post("/reorder")
async def reorder_queue(
    data: PrintQueueReorder,
    db: AsyncSession = Depends(get_db),
    _: User | None = RequirePermissionIfAuthEnabled(Permission.QUEUE_UPDATE_ALL),
    printer_scope: PrinterScope = RequestPrinterScope,
):
    """Bulk update positions for queue items."""
    for reorder_item in data.items:
        result = await db.execute(select(PrintQueueItem).where(PrintQueueItem.id == reorder_item.id))
        item = result.scalar_one_or_none()
        # Jobs on printers the caller can't see keep their place (#1727)
        if item and item.status == "pending" and printer_scope.allows(item.printer_id):
            item.position = reorder_item.position

    await db.commit()
    logger.info("Reordered %s queue items", len(data.items))
    return {"message": f"Reordered {len(data.items)} items"}


@router.post("/printer/{printer_id}/resume")
async def resume_queue_after_failure(
    printer_id: int,
    db: AsyncSession = Depends(get_db),
    _: User | None = RequirePrinterPermissionIfAuthEnabled(Permission.QUEUE_UPDATE_ALL),
):
    """Clear the previous-success gate for a printer and restore skipped items.

    Single atomic op (#1818):

    * Sets ``gate_acknowledged=True`` on every ``failed`` / ``aborted`` queue
      item for this printer that's still in the scheduler's lookback window,
      so the next ``_check_previous_success`` call ignores them.
    * Restores ``skipped`` items whose ``error_message`` matches the
      scheduler's exact "Previous print failed or was aborted" gate string
      back to ``pending`` (clears ``error_message`` + ``completed_at``).

    Returns counts so the UI can render a precise toast. No-op endpoint
    (zero counts) when called against a printer with no gate to clear.
    """
    result = await db.execute(select(Printer).where(Printer.id == printer_id))
    printer = result.scalar_one_or_none()
    if not printer:
        raise HTTPException(404, "Printer not found")

    ack_result = await db.execute(
        select(PrintQueueItem)
        .where(PrintQueueItem.printer_id == printer_id)
        .where(PrintQueueItem.status.in_(["failed", "aborted"]))
        .where(PrintQueueItem.gate_acknowledged == False)  # noqa: E712
    )
    to_ack = ack_result.scalars().all()
    for failed_item in to_ack:
        failed_item.gate_acknowledged = True

    restore_result = await db.execute(
        select(PrintQueueItem)
        .where(PrintQueueItem.printer_id == printer_id)
        .where(PrintQueueItem.status == "skipped")
        .where(PrintQueueItem.error_message == "Previous print failed or was aborted")
    )
    to_restore = restore_result.scalars().all()
    for skipped_item in to_restore:
        skipped_item.status = "pending"
        skipped_item.error_message = None
        skipped_item.completed_at = None

    await db.commit()

    logger.info(
        "Resume after failure on printer %s: acknowledged %d failure(s), restored %d skipped item(s)",
        printer_id,
        len(to_ack),
        len(to_restore),
    )
    return {"acknowledged": len(to_ack), "restored": len(to_restore)}


@router.post("/{item_id}/cancel")
async def cancel_queue_item(
    item_id: int,
    db: AsyncSession = Depends(get_db),
    auth_result: tuple[User | None, bool] = Depends(
        require_ownership_permission(
            Permission.QUEUE_UPDATE_ALL,
            Permission.QUEUE_UPDATE_OWN,
        )
    ),
    printer_scope: PrinterScope = RequestPrinterScope,
):
    """Cancel a pending queue item."""
    user, can_modify_all = auth_result

    result = await db.execute(select(PrintQueueItem).where(PrintQueueItem.id == item_id))
    item = result.scalar_one_or_none()
    if not item or not printer_scope.allows(item.printer_id):
        raise HTTPException(404, "Queue item not found")

    # Ownership check
    if not can_modify_all:
        if item.created_by_id != user.id:
            raise HTTPException(403, "You can only cancel your own queue items")

    if item.status not in ("pending",):
        raise HTTPException(400, f"Cannot cancel item with status '{item.status}'")

    item.status = "cancelled"
    item.completed_at = datetime.now(timezone.utc)
    await release_budget_reservation(
        db,
        source_type="print_queue",
        source_id=item.id,
        status="released",
    )
    await db.commit()

    # Stop an in-flight preheat for this item: the dispatch coroutine is
    # parked in a sleep and cannot see the status we just wrote (#2727).
    from backend.app.services.print_scheduler import scheduler as _scheduler

    _scheduler.notify_dispatch_cancelled(item_id)

    logger.info("Cancelled queue item %s", item_id)
    return {"message": "Queue item cancelled"}


@router.post("/{item_id}/stop")
async def stop_queue_item(
    item_id: int,
    db: AsyncSession = Depends(get_db),
    auth_result: tuple[User | None, bool] = Depends(
        require_ownership_permission(
            Permission.QUEUE_UPDATE_ALL,
            Permission.QUEUE_UPDATE_OWN,
        )
    ),
    printer_scope: PrinterScope = RequestPrinterScope,
):
    """Stop an actively printing queue item.

    Ownership-scoped (#1625-followup): callers with QUEUE_UPDATE_OWN can stop
    their own items; callers with QUEUE_UPDATE_ALL can stop any item. Mirrors
    the /cancel shape. Pre-fix this required QUEUE_UPDATE_ALL — Operators
    holding only _OWN saw the Stop button in the queue UI but got 403 on click.
    """

    from backend.app.services.printer_manager import printer_manager

    user, can_modify_all = auth_result

    result = await db.execute(select(PrintQueueItem).where(PrintQueueItem.id == item_id))
    item = result.scalar_one_or_none()
    if not item or not printer_scope.allows(item.printer_id):
        raise HTTPException(404, "Queue item not found")

    # Ownership check — mirrors /cancel. Ownerless items (created_by_id IS NULL)
    # require _ALL: stop is destructive and an _OWN holder can't claim "they
    # own it" the way /start does (#1670).
    if not can_modify_all and user is not None:
        if item.created_by_id is None or item.created_by_id != user.id:
            raise HTTPException(403, "You can only stop your own queue items")

    if item.status != "printing":
        raise HTTPException(400, f"Can only stop items that are printing, current status: '{item.status}'")

    # Capture values we need for background task
    printer_id = item.printer_id
    auto_off_after = item.auto_off_after

    # Try to send stop command to printer
    stop_sent = False
    try:
        stop_sent = printer_manager.stop_print(printer_id)
        if not stop_sent:
            logger.warning("stop_print returned False for printer %s - printer may not be connected", printer_id)
    except Exception as e:
        logger.error("Error sending stop command for queue item %s: %s", item_id, e)

    # Mark this printer as user-stopped BEFORE the first await so that if the
    # MQTT on_print_complete callback fires during the db.commit() yield the flag
    # is already set and the "failed" status will be correctly overridden to
    # "cancelled" (preventing a spurious "print failed" notification).
    try:
        from backend.app.main import mark_printer_stopped_by_user

        mark_printer_stopped_by_user(printer_id)
    except Exception as _mark_err:
        logger.warning("Failed to mark printer %s as user-stopped: %s", printer_id, _mark_err)

    # Update queue item status regardless - if printer is off, print is already stopped
    item.status = "cancelled"
    item.completed_at = datetime.now(timezone.utc)
    item.error_message = "Stopped by user" if stop_sent else "Stopped by user (printer was offline)"

    # Reconcile the linked archive when the printer is offline (#2603). When the
    # stop command reaches the printer it later reports the stop over MQTT and
    # on_print_complete flips the archive to cancelled/failed. When the printer is
    # offline no such event ever arrives, so the archive would stay "printing"
    # forever (queue row cancelled, archive still printing — the reporter's
    # archive 436). Close it out here, mirroring what the MQTT path would have
    # done. Only touch a still-"printing" archive so we never overwrite a real
    # completion that raced in.
    if not stop_sent and item.archive_id:
        archive = await db.get(PrintArchive, item.archive_id)
        if archive and archive.status == "printing":
            archive.status = "cancelled"
            archive.completed_at = datetime.now(timezone.utc)
            archive.failure_reason = "Stopped by user (printer was offline)"

    await db.commit()

    logger.info("Stopped printing queue item %s (stop command sent: %s)", item_id, stop_sent)

    # Schedule power-off if the queue item opted in. Delegates to the smart-plug
    # manager so the off honours each plug's configured strategy (time delay or
    # temperature threshold), is cancelled if the printer starts printing again,
    # and never cuts power on a loaded print (#1890). Previously an inline block
    # hardcoded a 50°C / 600s cooldown wait and powered off on the timeout
    # regardless of print state.
    if auto_off_after:
        from backend.app.services.smart_plug_manager import smart_plug_manager

        try:
            await smart_plug_manager.schedule_off_after_queue_job(printer_id, db)
        except Exception as e:
            logger.warning("Auto-off: Failed to schedule power-off for printer %s: %s", printer_id, e)

    return {"message": "Print stopped" if stop_sent else "Queue item cancelled (printer was offline)"}


@router.post("/{item_id}/start")
async def start_queue_item(
    item_id: int,
    skip_filament_check: bool = Query(default=False),
    db: AsyncSession = Depends(get_db),
    auth_result: tuple[User | None, bool] = Depends(
        require_ownership_permission(
            Permission.QUEUE_UPDATE_ALL,
            Permission.QUEUE_UPDATE_OWN,
        )
    ),
    printer_scope: PrinterScope = RequestPrinterScope,
    actor: User | ApiKeyActor | None = RequestActor,
):
    """Manually start a staged (manual_start) queue item.

    Ownership-scoped (#1625-followup): callers with QUEUE_UPDATE_OWN can
    start their own items + claim ownership of NULL-owner items (VP-uploaded
    items arrive unattributed per #1670). Callers with QUEUE_UPDATE_ALL can
    start any item. Pre-fix this required QUEUE_UPDATE_OWN with no ownership
    check, so _OWN holders could start anyone's queue items via direct API.

    Clears the manual_start flag so the scheduler picks it up. When
    ``skip_filament_check`` is false (the default) the live filament
    deficit (#1496) is checked first — if the assigned spool can't satisfy
    a slot's required grams, the route returns ``409`` with the deficit
    payload so the caller can show a confirm dialog and retry with
    ``skip_filament_check=true``. The same goes for a filament the printer has
    no tray for at all (#2799): ``409`` with ``code=unmatched_filament`` and
    the missing filaments.
    """
    user, can_modify_all = auth_result

    result = await db.execute(
        select(PrintQueueItem)
        .options(
            selectinload(PrintQueueItem.archive),
            selectinload(PrintQueueItem.printer),
            selectinload(PrintQueueItem.library_file),
            selectinload(PrintQueueItem.batch),
            selectinload(PrintQueueItem.variants).selectinload(PrintQueueVariant.library_file),
        )
        .where(PrintQueueItem.id == item_id)
    )
    item = result.scalar_one_or_none()
    if not item or not printer_scope.allows(item.printer_id):
        raise HTTPException(404, "Queue item not found")

    # Ownership check — softer than /cancel because /start is the entry point
    # for #1670's VP-import flow: an unowned item is claimable by the first
    # _OWN holder who clicks ▶, and the route below credits them as owner.
    # An item with a DIFFERENT owner → 403.
    if not can_modify_all and user is not None:
        if item.created_by_id is not None and item.created_by_id != user.id:
            raise HTTPException(403, "You can only start your own queue items")
    if not may_start_queue_item(user, can_modify_all, item.created_by_id):
        raise HTTPException(403, _AWAITING_REVIEW)

    if item.status != "pending":
        raise HTTPException(400, f"Can only start pending items, current status: '{item.status}'")

    item.estimated_cost = await _trusted_item_estimated_cost(
        db,
        item,
        printer_id=item.printer_id,
        plate_id=item.plate_id,
        ams_mapping=item.ams_mapping,
    )

    await validate_print_budget(
        db,
        cost_center_id=item.cost_center_id,
        estimated_cost=item.estimated_cost,
        current_user=actor,
        exclude_queue_item_id=item.id,
    )

    # Live deficit check — re-evaluated against current spool state, so a
    # spool swap between scheduler flagging and the user clicking ▶ clears
    # the block automatically.
    if not skip_filament_check:
        deficit = await compute_deficit_for_queue_item(db, item)
        if deficit:
            raise HTTPException(
                status_code=409,
                detail={
                    "code": "insufficient_filament",
                    "deficit": [d.to_dict() for d in deficit],
                },
            )

        # A filament the printer has no tray for at all (#2799). Without this,
        # Start on an item the scheduler held for exactly that would release it
        # only for the next pass to hold it again, and nothing would ever offer
        # "Print Anyway".
        from backend.app.services.print_scheduler import scheduler as _scheduler

        # A convenience, not a gate: the scheduler holds the item again if the
        # filament is still missing, so a failure here must not break Start.
        try:
            missing = await _scheduler.missing_filament_for_start(db, item)
        except Exception:
            logger.exception("Queue item %s: filament check before start failed", item_id)
            missing = None
        if missing:
            raise HTTPException(
                status_code=409,
                detail={
                    "code": "unmatched_filament",
                    "missing": missing,
                },
            )

    # Print Anyway / no deficit: clear the flags and let the scheduler dispatch.
    item.manual_start = False
    item.filament_short = False
    # Persist the user's "Print Anyway" decision so the scheduler does not
    # immediately re-flag this item on the next tick (#1698-followup). The
    # pre-fix behaviour bounced between "user said anyway" and
    # "scheduler re-blocked on same deficit" forever.
    if skip_filament_check:
        item.skip_filament_check = True
    # Credit the clicker as the item's owner when no prior owner is set —
    # VP-uploaded queue items arrive over FTP unattributed, so without this
    # the print log's User column stays blank even when auth is on
    # (#1670). An item that already has a creator (UI-added queue items)
    # keeps that attribution; the dispatcher is not promoted over the
    # original uploader.
    if user is not None and item.created_by_id is None:
        item.created_by_id = user.id
    await db.commit()
    await db.refresh(item, ["archive", "printer", "library_file", "created_by", "batch"])

    logger.info(
        "Manually started queue item %s (cleared manual_start; skip_filament_check=%s)",
        item_id,
        skip_filament_check,
    )
    return _enrich_response(item)