Two independent leaks of filament on the belt purge prism, plus the
replan safety net the second one needs.
1. The early-truncation scan bounded itself with the prism's own
toolchanges. ToolOrdering covers the whole print and the prism is a
printed object in it, so the "last toolchange" the scan found was on
the prism's own top layers -- it runs past every model object by
design -- and the truncation cancelled nothing. Bound the scan at the
tallest non-prism object (support layers included; on a belt they can
top the object). On MCTEST5 that was 197 toolchanges over 39.4 mm of
tower that no swap ever needed.
2. On a layer with no toolchange, the prism's entire fill printed as
solid infill in its own filament. Drop the fills no toolchange
claimed, right after the purge marking and before
ensure_perimeters_infills_order() force-overrides whatever is left.
Perimeters stay so the bar keeps a continuous wall. An earlier version
of this deleted the entities and had to be reverted: psWipeTower can
rerun without regenerating infill, and a later tool ordering may claim
what this one did not. The entities are now stashed with their layer,
region and index and put back exactly, the same reversibility contract
layer truncation already had.
3. Both stashes go stale if an object step reruns: make_fills() clears
and regenerates fills over m_layers only, so a stale stash would put
old fills back next to new ones, and truncated layers would keep old
perimeters/fills. Undo the plan's edits at the top of Print::process()
whenever psWipeTower is not done. Every object-step invalidation also
invalidates psWipeTower, so that condition is exactly "some object
step may rerun"; when it is done nothing regenerates and the edits
must stay. This also covers a prism left behind after belt mode is
turned off, which previously stayed truncated forever.
WipingExtrusions::is_entity_overridden() becomes public so the prism can
tell claimed fills from unclaimed ones.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SsuY8Laiyh7q2zPVVKV3HZ
Fixes the report in #12998 (comment 5465250754): first-layer speed and the
slow_down_layers ramp were ignored on a belt printer. The report reads as a
per-object problem, but neither applied to *any* object -- the reporter's first
part slowed down because slow_down_for_layer_cooling was on, which is
CoolingBuffer's time-per-layer mechanism, not initial_layer_speed.
FirstLayerPlane decides first-layer-ness by perpendicular distance to a plane it
derives by composing gcode_remap_* with compute_machine_z_affine(). The plane is
therefore a function of how G-code is *addressed*, not of where the belt is:
change the output axis convention and the plane moves. On MCBELT-TYPE2 the
first layer measured 86.2 mm from the plane and got effective index 431, far
past any slow_down_layers ramp.
on_first_layer(point) and effective_layer_index_for_point() now measure height
above the belt surface, using the belt description already carried in
SlicingParameters -- belt_floor_shear_factor / belt_floor_from_axis /
belt_floor_z_shift -- the same description the support generator uses. That is a
property of how the object was sliced, so no remap or back-transform can perturb
it.
Deliberately not via BeltFloorContext: its init() folds in
belt_support_floor_offset, a support-generator diagnostic, and letting that
option steer the model's first-layer speed band would be a surprising coupling
(a negative value would switch the slowdown off outright).
Preserving the existing first-layer-plane settings:
* first_layer_plane XY/YZ/XZ keeps the FirstLayerPlane evaluator, as those are
explicit opt-outs.
* A non-zero first_layer_plane_offset also keeps it. The offset is a machine-Z
shift that FirstLayerPlane converts into a perpendicular distance in the
slicing frame; this evaluator measures along slicing Z, so there is no
faithful translation. Deferring to the evaluator that implements the setting
beats silently ignoring it.
* The two thresholds stay separate, exactly as FirstLayerPlane keeps them:
the first-layer boolean tests initial_layer_print_height, while the
effective layer index counts bands of first_layer_plane_thickness.
Brim and coincident apron bands are emitted before m_layer is switched to their
object -- for an apron band there is no Layer at all -- so both paths publish the
belt-floor owner explicitly. Without that a brim's classification would borrow
whichever object was visited previously, making it depend on plate order.
Note that first-layer-ness drives more than speed: extrusion acceleration, jerk,
the first-layer flow ratio and eligibility for overhang speed/fan analysis all
read it, so all of them are corrected on belt printers by this change.
Classification still samples only each path's first point, as it did before.
Non-belt is unaffected by construction: belt_height_above_floor() returns false
when the belt floor is inactive and both call sites fall back to the previous
path. FirstLayerPlane stays in place for its other modes and for CoolingBuffer,
whose machine-coordinate probe is a separate outstanding bug.
Measured, MCTEST4 on MCBELT-TYPE2 (initial_layer_speed=5, slow_down_layers=40):
15 distinct feedrates with no gradient and F300 absent, becomes 70 including the
full ramp 300(5) 382(6) 465(8) 630(10) 795(13) ... Two bare cubes on a belt:
0 slow extrusions becomes 2378 across Z 32.36..95.18. The same two cubes on a
Cartesian printer keep their slow extrusions confined to Z 0.20..2.00.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011jgzj1sf53KMLPweZ8yeUQ
1. Belt tree support could not slice at all.
layer_initialize() hardcodes layer 0's bottom_z to 0, encoding "below layer 0 is
the build plate at z = 0". True for a flat bed; false for a belt, whose virtual
support layers legitimately extend below zero. The bottom-most belt layer
therefore got height = print_z - 0 = -9.8, which reached Flow::with_height() and
threw FlowErrorNegativeFlow.
A 3DBenchy, a mushroom, an L-bracket and an extruded L all failed identically
with negative flow / return -100. Only a bare cube sliced, because its support
never reached that far down.
The bottom is now taken from the previous layer's z, and only a layer 0 whose
print_z is itself negative gets a synthesised bottom below it. Every
non-negative print_z -- every non-belt configuration -- keeps exactly the
previous 0, so this is behaviour-preserving off a belt by construction. An
earlier form used min(0., layer_z(0) - layer_height), which regressed flat beds
whenever the initial layer was thinner than the layer height.
3DBenchy on a 45-degree belt with organic tree support: fails to slice ->
247 support blocks / 168,596 extrusions.
2. Support generated against the belt, and against belt-tilted walls.
A plain 20mm cube on a 45-degree belt generated 86 support blocks and 46,307
support extrusions. Three causes, all gated on the belt floor being active:
a. The build-plate tilt compensation shifted the lower layer the wrong way.
tan(build_plate_tilt_*) carries a magnitude but no direction, and the sign
chosen moved the lower layer away from the newly appearing material rather
than under it, doubling the mismatch. The shift now comes from
belt_floor_shear_factor / belt_floor_from_axis, which carry sign and axis
exactly. Non-belt tilted beds keep the previous behaviour.
b. Material resting on the belt was treated as unsupported. The belt surface
is now unioned into the effective lower layer, sampled at the bottom of the
layer -- a layer meets the belt across its thickness and print_z is the
top. The half-plane is clipped to the layer's bounding box first: unioning
a +/-1000mm half-plane raw with 20mm-scale geometry put a huge dynamic
range through Clipper and left intermittent artefacts every few layers.
c. The object's first slice can be empty on a belt (the bottom vertex is a
sub-extrudable sliver), leaving the layer above with an empty predecessor
even though it rests on the belt. (b) already covers that per island. What
did need fixing is sharp-tail detection, which tests each island against
the raw lower slices; with an empty predecessor that test is trivially true
and every belt-contact island read as a sharp tail. It now tests against
the same effective lower layer.
An earlier form instead skipped the whole layer when the point of
get_extents(curr_polys) -- the bounding box of the union of every island --
nearest the belt was in contact. That was wrong in a way worth recording:
one island resting on the belt could suppress overhang and sharp-tail
detection for a separate island floating well above it. Every decision here
is per-island.
Cube on belt: 46,307 -> 0 support extrusions. Same cube non-belt: 0 before and
after. Benchy on belt still 247 blocks / 168,596 extrusions and a mushroom
111 / 82,157, so false positives are removed without suppressing true ones.
Non-belt is unchanged by measurement, not only by the belt_ovh_active gate:
the same mushroom sliced on a Cartesian printer before and after gives 65,866
support extrusions and 68,717 total extrusions both times, the two G-code
files differing in exactly one line -- the object's plate position.
3. m_anti_overhang was filled and read in different index spaces.
It is consumed in the same index space as m_layer_outlines, where object layer i
lives at num_raft_layers + i, but was filled in object-layer space. Every entry
landed num_raft_layers too low (50 for a 20mm cube at bed Y=50) and the topmost
object layers got none. The belt injection also ran before m_raft_layers was
extended, so it could not have known the offset.
The array is now shifted as a whole and the injection moved after the raft
extension. This also repairs user support blockers under a raft, which is not
belt-specific: it changes behaviour for any ordinary raft, not just the belt's
virtual one, and should be reviewed as a general fix. Measured effect on the
cube was small on its own (46,307 -> 46,334 before the other fixes) because
m_anti_overhang only feeds calculate_placable; kept as a correctness fix on its
own merits.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011jgzj1sf53KMLPweZ8yeUQ
Both change emitted G-code, which is why they were kept out of the extraction
commit. Both are wrong only where the machine mapping is non-identity, which is
the definition of each bug.
1. Suppress lifts commanded through an unknown position.
_travel_to_z() emits full XYZ whenever the mapping must emit every axis, because
the mapping can make machine Z depend on logical X/Y, and it builds that point
from m_pos. At print start, and after any custom G-code that invalidates
position, m_pos.xy is the uninitialised origin; mapping (0, 0, z) through a
non-identity remap produces a real but wrong machine point -- for a reverse
mapping, build_vol_max, i.e. the far corner of the bed. The subsequent full-XYZ
move corrects the position, but the lift has already commanded a rapid across
the whole bed at travel speed.
Belt kinematics already guarded this; the Cartesian path did not. The guard is
now applied at all three lift sites through must_skip_lift_now(), not just the
one the extraction covered: travel_to_xyz()'s pending-lift branch,
lazy_lift(spiral_vase=true), and eager_lift(). The latter two also needed the
state fix -- both recorded m_lifted = target_lift regardless, so suppressing
only the emission would leave a later unlift() descending from a height that was
never commanded.
2. Never emit a G2/G3 arc a mapping cannot represent.
extrude_arc_to_xy() emitted G2/G3 with logical X/Y and I/J and never consulted
the mapping. There is no general fix by transforming the arc: a permutation
moves it out of the XY plane that I/J describes, a negation reverses handedness,
and the belt shear maps a circle to an ellipse that G2/G3 cannot express at all.
So supports_arc_moves() gates generation through the existing
GCode::should_disable_arc_fitting() hook, and BeltGCode's special-case override
is deleted -- belt now gets the same behaviour from the general rule instead of
its own exception.
supports_arc_moves() is m_remap_x == 0 && m_remap_y == 1, not !has_axis_remap():
an arc emits only X/Y/I/J, so a mapping that merely negates or reverses Z leaves
every emitted word untouched and keeps its arcs.
The fallback for an unrepresentable arc tessellates it into linear segments at a
0.005mm chord tolerance rather than substituting a single chord, and splits dE
proportionally across the segments. The capability check is hoisted above every
extrusion mutation: an earlier form ran it after filament()->extrude(dE) and so
extruded 2*dE on the fallback path.
Known limits of that fallback, since it is worth stating rather than discovering:
emitted relative E is conserved only to per-segment rounding (a radius-5
semicircle with dE=1.5 emits 1.50012 across 36 segments); the 0.005mm bound is a
logical-frame bound, about 0.00855mm in machine space under a 45-degree belt
shear; unequal endpoint radii and non-finite inputs are unchecked. Ordinary
export takes the original polyline when the mapping rejects arcs, so this path
is a fallback rather than the normal route.
Known gap, not claimed fixed: classic wipe towers have their own
enable_arc_fitting and their own G2/G3 emitter in GCode/WipeTower.cpp, which
should_disable_arc_fitting() does not govern. Belt printers are barred from
classic wipe towers; a remapped Cartesian printer is not.
Tests in tests/fff_print/test_gcodewriter.cpp: reverse-X remap with unknown and
with known position plus an identity control; eager_lift emitting nothing and
recording nothing; the arc-capability matrix including the Z-only cases; and the
tessellated fallback. E accounting is asserted through used_filament() rather
than E(), which resets per line in relative-E mode.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011jgzj1sf53KMLPweZ8yeUQ
BeltGCodeWriter subclassed GCodeWriter and overrode seven methods, five of them
by copying the base body and changing the transform. The base writer already
carried an axis remap and already branched at each of its seven
coordinate-emission decisions; the subclass did the same branching with a
different transform, and the two copies had begun to drift.
Replace the inheritance with a strategy object owned by GCodeWriter:
CartesianKinematics to_machine = the existing apply_axis_remap; today's base
behaviour, moved rather than changed.
BeltKinematics to_machine = MachineFrameTransform o axis_remap o
BeltBackTransform, plus a world_coordinates variant for
the PA calibration generators.
New: src/libslic3r/GCode/MachineKinematics.{hpp,cpp}, GCode/BeltKinematics.{hpp,cpp}
Deleted: src/libslic3r/BeltGCodeWriter.{hpp,cpp} (341 lines)
Points worth a reviewer's attention:
* The predicate is must_emit_all_axes(), not couples_axes(). The base returns
true for any non-identity remap, including pure permutations that do not
physically couple axes, so the question is "must every axis word be
emitted", not a statement about kinematics.
* Every per-site word-omission branch is preserved. The base deliberately
emits X/Y only, or Z only, or drops Z when its quantised value is unchanged.
The strategy changes which transform applies, never whether words are
omitted.
* set_kinematics() replays the configured remap and build volume onto a newly
installed strategy, because BeltGCode::init_belt_writer runs before
GCode.cpp calls set_axis_remap/set_build_volume_max.
* uses_pointwise_travel_speed() preserves a pre-existing divergence rather
than introducing one: the base travel_to_xyz emits the raw configured travel
speed in its final branch, ignoring the first-layer value computed at the
top, whereas the belt path used the first-layer-aware value throughout. Both
are kept. Unifying them changes feedrates and belongs in its own change.
* The [BELT-DEBUG] block is deleted; it rate-limited itself with a
function-local static thread_local in the hot emission path, and this is the
commit that would otherwise have moved it into shared code.
This commit is intended to preserve existing export output. That is reviewed by
construction -- each emission site keeps its own omission branch and each policy
divergence is preserved -- and is NOT verified against a G-code diff corpus.
Building that corpus is the outstanding work here.
Two API-equivalence exceptions, neither reachable by any caller today:
* Belt kinematics with no plane pointer installed, m_is_first_layer true,
initial and normal travel speeds differing, travel_to_xyz() reaching its
final branch: the old belt writer selected the initial-layer speed, the new
writer selects the normal travel speed. The pending-lift and XY-only
branches keep their previous selection.
* Belt kinematics installed without set_force_normal_lift(true) and a
non-normal lift requested: the old belt writer forced a normal lift, the new
writer can take the slope branch.
The PA-pattern generator reaches the writer through explicit travel_to_z() /
travel_to_xy(), not travel_to_xyz() or the lazy/eager lift paths, and normal
belt export installs both the plane and the forced-normal-lift policy, so
neither exception changes output produced today. They are recorded because a
future caller could reach them.
tests/fff_print/test_gcodewriter.cpp was also not compiling before this branch:
it called writer.to_machine_coords(), a method that existed only on
BeltGCodeWriter. It never surfaced because the build targets OrcaSlicer, not
all, and BUILD_TESTS defaults to OFF, so that translation unit was outside every
compile path. Fixed here; the existing 30-degree coordinate assertions are kept
verbatim as the best available regression net.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011jgzj1sf53KMLPweZ8yeUQ
* build: remove std::move that blocks copy elision
std::move wrapped around a temporary, or around a local being returned,
stops the compiler constructing it in place. Each edit is the fix clang
suggests, which is to delete the std::move call and keep its argument.
Three of the 39 sites save a move, the two return std::move(local) in
Print.cpp and TreeSupport.cpp:2749. The rest are equivalent either way
and match how the codebase already writes this elsewhere.
Clears 39 -Wpessimizing-move warnings.
* build: drop null checks on references and this
A reference cannot be bound to null and this cannot be null, so the
compiler folds these conditions to true and drops the guard. Seven are
if (&bitmap && bitmap.IsOk()), where IsOk() already does the work; two
test this directly. The guarded code runs either way, so removing the
dead operand changes nothing.
Clears 11 -Wundefined-bool-conversion warnings.
* Keep mixed-color filaments intact when the extruder count changes
The extruder-count spinner resized the filament arrays in bulk at the tail,
which is where mixed-color slots live, so a new filament landed behind the
mix and the sidebar skipped a slot number. It now adds and removes one slot
at a time through the same calls the sidebar's +/- buttons use, so a new
slot opens ahead of the mixed tail and a removal renumbers object filament
ids, painted facets, custom g-code and mixed components rather than
clamping them away.
Drops the vector overload of set_num_filaments(), which this leaves without
callers.
Update perimeter traversal to pass each extrusion's closed/open state into `apply_fuzzy_skin`. This lets fuzzy skin logic distinguish contours from closed loops when processing perimeters.
Co-authored-by: Rodrigo Faselli <162915171+RF47@users.noreply.github.com>
* Fix contour cleanup across coplanar triangles
Avoid generic collinear simplification after slicing. Skip only junctions created by shared edges between coplanar faces so contours stay stable without altering shallow geometry.
Fixes#15364
* Fix contour cleanup across coplanar triangles (code review fixes)
---------
Co-authored-by: Ian Bassi <ian.bassi@outlook.com>
* Skip straight-run splits in corner smoothing
Teach `CornerSmoother` to treat vertices that only continue a straight segment as part of the same leg instead of rounding them as corners. The smoother now keeps a three-point window so it can emit a corner only once both adjoining legs are known, which avoids unnecessary corner processing while preserving real turns such as hairpins.
* Add regression test for split-leg smoothing
Adds a FillCornerSmoothing regression test covering polylines with an extra collinear vertex in a straight run. The test ensures corner smoothing treats split and unsplit geometry identically, preventing inconsistent rounding radii in triangular/grid infill paths.
# Description
This PR ports the color mixing feature from BambuStudio.
The port is based on the previous work by @ianalexis in #15231.
This PR completes the port and fixes various bugs.
Several improvements were also made during the porting process.
WIP
# Screenshots/Recordings/Graphs
<!--
> Please attach relevant screenshots to showcase the UI changes.
> Please attach images that can help explain the changes.
-->
## Tests
<!--
> Please describe the tests that you have conducted to verify the
changes made in this PR.
-->
<!--
> A guide for users on how to download the artifacts from this PR.
-->
[How to Download Pull Requests Artifacts for
Testing](https://www.orcaslicer.com/wiki/how_to_download_pr_artifacts)
* chore: mark every declaration that overrides a base virtual
clang-cl reports 42 member functions across 28 files that override a
base virtual without being marked `override`, inside classes that
already mark their other overrides. That is every occurrence of
-Winconsistent-missing-override in the tree, so the category drops to
zero and -Werror=inconsistent-missing-override becomes available as a
guard against it coming back.
Behaviour is unchanged. Each keyword goes only where clang had already
resolved the declaration to a base virtual, so it records what the
compiler already worked out and cannot affect overload resolution or
dispatch. If any of these signatures had not really overridden a base
method, the build would have failed rather than warned.
Where a declaration already carried `virtual` it is left alone and the
keyword appended, matching the surrounding declarations. Plain
`override` is used rather than the wxWidgets `wxOVERRIDE` macro, which
wx/defs.h defines as `override` beneath a comment marking it obsolete,
and which the rest of src/slic3r already avoids by 1742 occurrences to
113.
A full clang-cl build takes -Winconsistent-missing-override from 1,146
warning lines to 0. Those 42 declarations produce that many lines
because a header is re-diagnosed in every translation unit that
includes it. CalibrationWizardStartPage.hpp alone accounts for 336 of
them from 4 declarations.
* chore: drop unused lambda captures in GUI/Widgets
clang-cl reports 10 lambda captures in src/slic3r/GUI/Widgets that are
never read. Removing them changes nothing at runtime.
Every capture removed is `this` or a raw pointer. clang does not report
a capture whose type has a non-trivial destructor, since such a capture
can be held purely for its effect on an object's lifetime, so nothing
that owns or extends a lifetime is touched. The std::weak_ptr captured
beside the removed `this` in MultiNozzleSync.cpp stays.
This clears the category in GUI/Widgets only. A full clang-cl build
takes -Wunused-lambda-capture from 312 warning lines to 302, leaving
235 sites in other directories for a follow-up.
Introduced a shared `colinear_vertex_tolerance()` helper in `ExtrusionLine.hpp` and updated both simplify paths (`ExtrusionLine.cpp` and `WallToolPaths.cpp`) to use it instead of duplicated hardcoded `0.005` scaled thresholds. This keeps the near-colinear early-out tied to `SCALED_EPSILON` (rounding-noise scale) and avoids unintended curve decimation from larger tolerances, while documenting the geometric impact in code.
* Add caching system for presets
* Removing user\bundle serialization and keeping it only for system presets
* Integrate caching into WebGuideDialog which speeds up time of SetupWizzard and PrinterSelection dialog
* Add CI\CD step to prepare cache file in ahead of time so user does not need to wait
* Add partial cache generation when only one of the vendros is changed to speed up recalculation time
* Handle corrupted files
* Add cache to GuideDialog as previos version didn't work as expected
* Add inspecting tool and fix CI cache generation
* Generate cache per vendor
* Simplify code by mergin it in PresetBundle
* Simplify code a bit more
* Add cereal serialize() to VendorProfile, PrinterModel, Preset, and Semver
* Remove CachedPrinterModel/VendorProfile/Preset mirror structs from VendorCache
* Fix use-after-free in CallAfter lambda; replace raw thread pointer with unique_ptr
* Use get_vendor_cache_key() to match cache keys written by the app
* Remove BOM added by VSC
* Skip invalid vendors
* Remove leftover cache file
* Fix build for windows arm64
* Revert json cache back
* Update check for stale cache
* Serealize all value fields for Preset class to minimize regression later
* Minimize field duplication by moving Cache thing into PresetBundle
* Add tests for Cache system
* Add a bit more tests
* Merge branch 'main' into feature/cache_profiles_and_optimize_loading_speed
* Rvert from per-verndor to single cache file
Replace N per-vendor .cache files with a single system_presets.cache
that holds all vendors and presets in one serialized blob.
Cache load is now all-or-nothing: on hit all vendors are applied from
the bundle (sub-second); on miss all vendors are parsed from JSON and
a fresh bundle is written to the user cache dir.
Invalidation is driven by bundle_key - a sorted concatenation of all
vendor JSON version strings. Any vendor update invalidates the whole
cache and triggers re-parse on next launch.
Guide wizard (WebGuideDialog) loads the bundled cache into a plain
PresetBundle instead of a separate VendorGuideData struct, removing
the duplicate data model.
generate_system_cache simplified from a per-vendor loop to a single
save_system_presets_cache() call producing one output file.
* Transfer all Preset fields from cache via move assignmet
apply_vendor_preset_group was copying fields manually and missed
bundle_id, user_id, base_id, sync_info, updated_time, key_values,
ini_str. Replace field-by-field copy with move assignment of the
fully-deserialized Preset, then restore the vendor pointer which
is excluded from serialization.
* Ignore cache for future
* Remove not used files
* Ship one preset cache per vendor in place of the profile JSONs
Each vendor's system presets serialize into a single <vendor>.opc built at
package time, and a shipped build carries that file alone — the profile JSON
and its sub-file tree are pruned. The vendor loader, the setup wizard's profile
list and the resource installer all read a vendor through its cache, falling
back to parsing whenever one is absent, stale or unreadable, so the cache stays
an optimization and never a source of truth. Caches hold presets in source form
and resolve inheritance at load, through the same code the JSON path uses.
* Make the preset cache self-describing and load each vendor from the system folder alone
The cached DynamicPrintConfig is keyed by name, through a per-file dictionary of the
distinct opt_keys, the type each was written as, and the distinct enum value names,
instead of by serialization_key_ordinal — a position assigned by declaration order at
static init, where inserting one option shifts every later ordinal and the lookup then
succeeds on the wrong option. Because a name-keyed payload drops the options this build
cannot place rather than being rejected wholesale, the schema fingerprint goes, and with
it the two fallbacks that existed only because an installed cache died on every app
upgrade: the second lookup tier into resources/profiles and the parse fallback to the
same place. A vendor is loaded from <data_dir>/system/ and nowhere else, as on main —
which is what makes the app write its .opc files there again.
* Simplify the preset cache internals after review
* Use the shared temp-dir helper in the preset bundle loading test
* Bound stamp string reads in the preset cache
* Speed up the setup wizard with a profile-data cache
The wizard's per-vendor fast path threw on vendors present only in
resources, falling back to a ~29 s raw JSON scan on every open. Each
vendor now loads from the directory it was found in, and the derived
model/machine/filament/process catalog is cached whole in
<data_dir>/cache/wizard_profile_data.json, stamped by each vendor's
name and version - a fresh cache makes an open one file read, with no
bundle built and no presets installed (~0.2 s vs ~2 s).
* Remove debug SVG dump from a geometry test
* Move the per-vendor cache file format into PresetCacheFormat
* Move the vendor install helpers from PresetBundle into Utils
* rename
* fix flatpak
* change cache version to 1
---------
Co-authored-by: SoftFever <softfeverever@gmail.com>
* Normalize the junction direction vector over XYZE
calc_vmax_junction_deviation() treats the dot product of two jd_unit_vec as a
cosine, but the vectors were scaled by 1 / block.distance, which is the XYZ
length. On an extruding move the E component then pushes the 4D norm above 1 and
the dot product below -1, so the corner reads as straighter than it is and is
planned too fast -- the more so the higher the flow. Measured on a 6 degree
corner at scv 5: 86.9mm/s with no extrusion, 94.4mm/s at 0.029mm/mm, 150.0mm/s
at 0.1mm/mm.
Neither firmware does that. Marlin normalizes over XYZE for any extruding move
(planner.cpp: `if (... || esteps > 0) normalize_junction_vector(unit_vec)`) and
Klipper leaves E out of the cosine entirely, dotting only axes_r[0..2]
(toolhead.py::Move.calc_junction). Normalizing satisfies both: with E normalized
in, the cosine differs from the XYZ-only one by ~1e-5 at printing flow rates.
This is a deliberate divergence from PrusaSlicer, which still scales by
1 / distance -- it carries an older Marlin's behaviour.
Travel moves are unaffected, their vector was already unit length.
Reported by Copilot in review of #15304.
* Test that extrusion rate does not change corner planning
The junction deviation tests were all travel-only, which is exactly why the E
component of the junction vector went unchecked. Cover it: the same corner has
to be planned the same whether nothing, an ordinary 0.42 x 0.2 line, or a fat
large-nozzle line is extruded through it, on both Klipper and Marlin 2.
Reported by Copilot in review of #15304.
