The filament-group golden harness landed with H2C/A2L support (#14685). Its
"FilamentGroup golden regression" / stress_66 case fails intermittently on
Windows x64, on main and on unrelated PRs alike. The test depends on how fast the
runner is.
The k-medoids clustering these goldens exercise is an anytime search bounded by a
3 second wall clock. Every restart is seeded from its own index, so nothing about
it is random. What varies is how many restarts fit in the budget, and the best
cost is a minimum over completed restarts, so a slower runner is never better.
Grading a score produced that way measures the machine as much as the code.
Add a ClusteringBudget struct and let the tests set it. The defaults are the
current 3 seconds and 30 restarts, so slicing behavior is unchanged. A
non-positive timeout removes the wall clock and bounds the search by restart
count alone.
The goldens are then graded under a fixed budget of four restarts, where every
one of them reaches the BambuStudio reference within 3%, so the score becomes a
property of the code. This retires the machine-specific 125103 lock on stress_66.
The default wall-clock path keeps its own test, asserting the grouping is valid
and the search does not run away. It makes no score assertion, because under a
wall clock that number is not a property of the code.
The golden test also checks the run fits in ten times the default wall clock.
Slicing quality depends on how many restarts fit in the budget, so a search an
order of magnitude slower would degrade real groupings while a fixed-budget score
gate stayed green.
The 3% tolerance stays as the parity allowance against the goldens. It also
covers a small spread across standard libraries: the k-medoids search seeds each
restart with std::shuffle, whose algorithm the C++ standard leaves unspecified,
so libstdc++, libc++ and the MSVC STL permute the same seed differently, start
from different medoids, and settle on slightly different groupings, about 3e-4
apart and only on the goldens heavy enough to reach the k-medoids search.
The new "Prime-tower visits..." test from #14685 (H2C/A2L support) throws
"Coordinate outside allowed range" at random on CI, on both Windows arm64 and
Linux x86_64. It's an uninitialized read of WipeTowerData::height.
#10780 (H2D/H2S) added a second wipe tower path, generate_new(), that fills in
depth, bbx, brim_width and rib_offset but not height. The older generate() sets
height, and clear() never did, so on the generate_new path it stays garbage.
first_layer_wipe_tower_corners() passes height to get_wipe_tower_cone_base() as
R = tan(cone_angle/2) * height. The stray bytes are usually zero, so R is zero
and the slice is fine, which is why it passes most runs on every platform. When
they aren't zero the cone radius runs past ClipperLib's limit and the slice
throws. Nothing selects for it, so it just flakes around.
#14685's test is the first to exercise this path, so that's when it started
showing up.
Initializing height in clear() fixes it, same as the m_origin fix in #13712. The
BBL generate_new path has no stabilization cone, so height = 0 is right.
Deleting an object's instance leaves a stale (obj_id, instance_id) pair in
PartPlate::obj_to_instance_set until it is pruned. object_list_changed() runs
during the delete path and calls has_printable_instances(), which guarded only
obj_id and then indexed object->instances[instance_id] on the now-shorter
vector, dereferencing a garbage ModelInstance* and crashing with SIGSEGV. The
reported fault address (0x12a) matches a member read on that bad pointer.
Reuse the existing valid_instance() helper, which bounds-checks both obj_id and
instance_id, at every obj_to_instance_set scan that was missing the instance-id
check: has_printable_instances(), printable_instance_size(),
is_all_instances_unprintable(), get_extruders_under_cli(),
duplicate_all_instance() and set_pos_and_size() (the last had no bounds check at
all). valid_instance() is made const so the const CLI scan can call it.
Fixes#14159
* fix: out-of-bounds read computing tool-ordering max layer height
calc_max_layer_height() loops over the extruder count (nozzle_diameter)
but indexes max_layer_height with the same counter, reading past the end
when that array is shorter. Silent on release builds, aborts under a
bounds-checked STL (_GLIBCXX_ASSERTIONS).
Read via get_at(), which falls back to the first entry when the index is
out of range, as Slicing.cpp already does for this option.
Add a fff_print regression test slicing a two-extruder printer with a
single-entry max_layer_height.
* docs: clarify how max_layer_height ends up short in the regression test
Normalization sizes it to the filament count under single_extruder_multi_material,
not "a mismatch a profile can ship" as the earlier comment guessed.
round() already had unit coverage; floor() and ceil() had none. Add the missing
positive cases for both signs, plus round()'s half-away-from-zero tie-break, and
one negative case asserting that a name outside the grammar's built-in function
set is treated as an undefined variable and throws, rather than being passed
through to a math library.
GCode::extrude_support declared its per-path speed helper as a function-local
static lambda that captures `this` by reference. The closure is built once, on
the first extrude_support call, and reused for the rest of the process, so a
second G-code export in the same process runs the helper against a `this` from
the first export's stack frame, which has already returned.
The stale `this` flows through NOZZLE_CONFIG(...) -> cur_extruder_index() ->
GCodeWriter::filament(), reading a garbage current-extruder id and indexing
with it. It is silent whenever the reused stack still holds a usable pointer,
and an order-dependent SIGSEGV otherwise; AddressSanitizer reports it as a
stack-use-after-return in GCodeWriter::filament(). It is the only static
capturing lambda in libslic3r.
Drop static so the closure is rebuilt each call against the live frame. Add an
fff_print regression test that slices a support object twice in one process; it
fails without the fix (stack-use-after-return under ASan) and passes with it.
* ci: run unit tests on Windows and macOS
The Unit Tests CI job only ran on Linux, so platform-specific bugs
invisible to a Linux build could land undetected (e.g. the MSVC-only
NfpPlacer crash fixed in #14267). The full non-[NotWorking] suite already
builds and passes on every shipped arch, so this wires them into CI.
A reusable unit_tests.yml, called once per built arch, downloads that
arch's test artifact and runs ctest. Each build leg builds the test
executables and uploads them; a single publish_test_results job on Linux
aggregates the JUnit results into one check.
Coverage: Linux x86_64 + aarch64, Windows x64 + arm64, macOS arm64.
macOS x86_64 is deferred (cross-built on arm64, needs Rosetta).
- build_release_vs.bat: "tests" token enables BUILD_TESTS
- build_release_macos.sh: -T builds and runs tests; ORCA_TESTS_BUILD_ONLY
builds them without running (used by CI)
- scripts/run_unit_tests.sh: parameterized test dir and build config
Addresses #11273.
* ci: bump actions/checkout to v7 in reusable unit_tests workflow
Match the actions/checkout v6->v7 bump (#14517) that upstream applied to
the inline test job this reusable workflow replaces.
* fix: isolate calibration temp paths per user
The calibration temp files under <temp>/calib were file-scope statics
initialized before set_temporary_dir() runs at startup, so they kept
using the shared system temp root and missed the per-user isolation
added in #14607. On Linux every account shares /tmp, so the first user
to calibrate owns /tmp/calib and later users fail to write there, the
same cross-user collision #14607 fixed for model backups, STEP import,
and part skip.
Build the paths lazily from temporary_dir() instead, through a
calib_temp_dir() accessor and a calib_temp_file() join helper. The base
becomes <temp>/orcaslicer_<uid>/calib on Linux and is unchanged on
Windows, where the temp dir is already per user.
Also make StoreParams::path a std::string rather than a non-owning
const char*. The calibration code had to keep a std::string alive
solely to feed that pointer, and the field was uninitialized by
default; owning the string removes the lifetime hazard for all three
callers and makes the entry guard a reliable empty() check. Confined to
the 3mf project exporter (three callers, two internal reads); no
on-disk, format, or ABI impact.
Follows up on #14607 per @Noisyfox's review suggestion.
* fix: prevent out-of-bounds crash in Arachne beading interpolation
SkeletalTrapezoidation::interpolate() derives an inset index from `left` but
uses it to index the merged beading, which follows the thicker of left/right.
When the thicker side has fewer insets, the index runs past the end and the
slicer crashes during "Generating walls".
Skip the adjustment when the index is out of range, as the adjacent guards
already do. interpolate() uses no instance state, so make it static and add a
regression test that exercises it directly.
Fixes#14584
feat: add regex_replace() string transform to filename templates
The filename template language could test strings (=~, !~, one_of) but
never rewrite one, so there was no supported way to reshape a placeholder
value, such as dropping a file extension from {first_object_name}.
Add regex_replace(subject, /pattern/, replacement), reusing the existing
regex-literal syntax and boost::regex engine. Every placeholder keeps
returning its exact value and the template does the transform explicitly:
{regex_replace(first_object_name, /\.[^.]*$/, "")} strip any extension
The replacement may reference capture groups ($1, $2, ...). It is one
grammar function mirroring digits(), with the name registered as a keyword
so it is not parsed as a variable.
* fix: initialize Print::m_isBBLPrinter
Built outside the GUI/CLI (headless tests, embedded use) the member was read
uninitialized: is_BBL_printer()/wipe_tower_type() feed it into ToolOrdering,
which then non-deterministically dropped per-feature filament assignments.
Default it to false, the value the GUI and CLI already assign for non-Bambu
printers.
* docs(test): add the fff_print testing contract
tests/fff_print/README.md codifies how the suite is organized: one file per
subsystem (each owning both in-memory and emitted-G-code assertions), flat
behavioral test names with a single [Subsystem] tag, a robust-tests guide,
the shared helpers, and an add-a-test checklist. Linked from tests/CLAUDE.md.
* test(fff_print): reorganize the suite to the contract and add coverage
Bring every subsystem into one file per the README: rename the test_data
harness to test_helpers; consolidate skirt/brim; split multi-filament and
cooling into their own files; disperse the test_printgcode grab-bag and the
end-to-end smoke scenario into focused tests; fold test_gcode into
test_gcodewriter. Standardize names and tags, align cube tests on the cube()
helper, and de-qualify the flagship files.
New coverage: multi-filament per-feature and per-object routing; a skirt/brim
behavior matrix (the #14333 rework, including brim ears, with regression
coverage for #14319 and #14366); resolved extrusion-width and config
comments; custom-G-code placeholders; fan control and speed-marker
consumption.
Re-enable three slice tests previously tagged [NotWorking]: the clipper
"Coordinate outside allowed range" error that disabled them was specific to a
past CI runner environment and no longer reproduces.
* test(fff_print): tag arm64-flaky skirt/brim tests NotWorking
Four skirt/brim slice tests intermittently throw ClipperLib's "Coordinate
outside allowed range" on the macOS and Windows arm64 CI toolchains (an FP
divergence, not a slicing bug; see PR #14207). Linux x86_64 and aarch64 are
unaffected. Tag them [NotWorking] so ctest -LE NotWorking skips them.
* test(fff_print): re-enable the arm64 skirt/brim tests
These were tagged [NotWorking] as a stopgap when myfork's daily-driver build
combined them with the cross-platform CI on a base that predated upstream's
m_origin fix (99dea01cc3). With upstream merged in, Print::m_origin is
initialized and the "Coordinate outside allowed range" throw is gone, so the
tests pass on macOS/Windows arm64. Drop the tags.
On Linux every account shares /tmp, but slicing builds temp paths there under
fixed, app-owned names via temporary_dir() (model backups, STEP import,
part-skip). The first user to slice creates and owns those dirs, so the next
user cannot write under them and slicing crashes with "No such file or
directory".
Tag the app temp root with the user id at startup (<temp>/orcaslicer_<uid>)
so every temporary_dir() consumer is isolated at once. The id stays at the
top level of the world-writable system temp so each user's dir is created
directly there; a shared parent dir would be owned by whichever user made it
first. The root is pre-created because STEP import writes into it directly.
Windows keeps the plain temp dir since it is already per-user.
Fixes#10108. Same root cause as #5969.
* fix: crash in Measure tool when a plain edge is the first selection
The SPHERE_2 gripper raycaster called get_feature_offset() on
.first.feature instead of .second.feature (copy-pasted from the SPHERE_1
block). Plain planar-border edges store no extra point, so the Edge
branch dereferenced an empty optional behind a release-stripped assert,
aborting on Flatpak and undefined behavior elsewhere.
Point the SPHERE_2 raycaster at .second.feature and fall the Edge branch
back to the edge midpoint.
Fixes#14018
{input_filename_base} is meant to be the saved project's file name. Before
#13753 a bug made it fall back to the first object's name when a project was
saved; #13753 fixed it to use the project name. Some users relied on the old
behavior to get the part name into their output file name and had no
placeholder to recover it ({model_name} is the 3mf designer metadata, blank
for plain STL imports).
Add {first_object_name} as a dedicated placeholder for the first printable
object on the current plate, populated in update_object_placeholders()
independently of {input_filename_base}.
Closes#14493
Selecting the prime tower and rotating it (PageUp/PageDown) crashed.
Selection::notify_instance_update() indexed m_model->objects with the
wipe tower's synthetic id (>= 1000), which is not a ModelObject index,
so the lookup returned garbage and dereferencing it segfaulted.
do_rotate/do_scale/do_mirror already skip the wipe tower in their own
loops but all call this shared helper, so scale and mirror hit the same
fault. Selection::drop() had the same latent bug via a direct index.
Guard both with the >= 1000 check already used throughout the file.
Fixes#14498
WipeTowerIntegration::append_tcr processed filament_end_gcode with only
layer_num in its placeholder config, so a filament_end_gcode referencing
{layer_z} could not be evaluated and slicing aborted. This affects any
multi-filament print that routes tool changes through the prime/wipe tower
(for example a support filament on a Bambu printer); the same macro works
in machine_end_gcode and on the non-wipe-tower set_extruder path, which
both define layer_z.
Set layer_z to tcr.print_z, the value this function already provides to its
change_filament_gcode and tcr_rotated_gcode placeholders.
Fixes#10119
PR #13712 fixed the uninitialized Print::m_origin (commit 99dea01cc3, "Fix coord
out-of-range exception caused by m_origin memory not initialized to 0") that made
headless slice() intermittently throw ClipperLib's "Coordinate outside allowed
range". With that root cause fixed, the three tests disabled for it pass again,
so drop their [NotWorking] tags.
CalibPressureAdvancePattern::line_width_first_layer() returned the raw
initial_layer_line_width, so a value of 0 (which means "use the default")
fed 0 into the Flow spacing math and threw FlowErrorNegativeSpacing,
crashing the whole app when slicing a PA pattern calibration.
Mirror the guard already present in the sibling line_width(): when the
configured width is non-positive, fall back to auto_extrusion_width.
Add a libslic3r regression test covering the width resolution.
Fixes#13188
* fix(libnest2d): skip the excluded-region alignment pass when there are none
NfpPlacer::finalAlign(), run from clearItems() and the destructor, always
ran the "find a best position inside the NFP of fixed items" pass even when
no items are fixed. With nothing to avoid, calcnfp() computes the inner-fit
NFP of the pile and can feed clipper a coordinate outside its allowed range.
On Linux/clang the value stays in range so it went unnoticed; on MSVC the
clipper "Coordinate outside allowed range" exception escapes the noexcept
destructor and aborts the process (exit 0xC0000409).
Build the excluded set up front and only run the pass when it is non-empty.
The block exists solely to keep the pile clear of fixed items (excluded
regions / wipe tower), so it is a no-op when there are none and the
wipe-tower behaviour is unchanged.
* test(libnest2d): remove dead nesting tests and split the suite by feature
Seven of the suite's hidden [.] test cases drove code paths Orca abandoned
at the BambuStudio fork: BottomLeftPlacer (used nowhere in src/) and the
stock default NfpPlacer backend, which returns zero bins in Orca. They have
been red since the fork and are never registered with ctest. Remove them.
Split the 1,000-line libnest2d_tests_main.cpp into per-feature files, per the
repo convention, sharing a header for the no-fit-polygon backend setup that
every translation unit must agree on (ODR):
libnest2d_tests.cpp Item and nest() basics
test_geometry.cpp geometry primitives
test_nfp.cpp no-fit-polygon machinery
libnest2d_test_utils.hpp shared includes and the NFP backend specialisation
Along the way: drop a debug exportSVG() helper that only wrote a file on test
failure (so the suite never leaves stray assets), convert the deprecated
Catch::Approx to WithinRel/WithinAbs matchers, and give the tests descriptive
names.
* test(libnest2d): add NfpPlacer unit tests
NfpPlacer is the placement engine the arranger drives, but the suite only
covered the geometry primitives. Add a fixture and five tests that exercise
pack()/accept() directly: a single item lands in the bin, an oversized item
is rejected, the first item is seeded for every starting point, many items
pack without overlap, and the rotation candidates are searched. This lifts
nfpplacer.hpp line coverage from 42% to 87% in the libnest2d suite.
* test(libslic3r): add arrangement::arrange() integration coverage
The libnest2d suite cannot reach Orca's real nesting entry point because it
does not link libslic3r. Add test_arrange.cpp driving arrangement::arrange():
items land on the bed and within bounds, do not overlap, are spaced by their
inflation, an oversized item stays unplaced, overflow spills onto virtual beds,
an empty input is a no-op, and the DONT_ALIGN and USER_DEFINED final-alignment
paths are exercised. A self-test guards the overlap check the other cases use.
The testing guide stated OrcaSlicer uses Catch2 v2 and advised the v2
`<catch2/catch.hpp>` include, but the vendored framework is v3.11.0
(tests/catch2/) and every test file includes `<catch2/catch_all.hpp>`.
The wrong version drove several incorrect claims: that SKIP() is
unavailable (it is, v3.3.0+), that the string matcher is "Contains"
rather than "ContainsSubstring", and that thread-safe assertions,
multiple reporters, STATIC_CHECK and built-in sharding do not exist.
Correct all version statements, the example include, and the
former "Version-Specific Limitations" section to reflect v3.11.0.
Re-enable [OrcaCloudServiceAgent] tests now that the headless crash is fixed
The two OrcaCloudServiceAgent display-name tests were tagged [NotWorking]
in #14175 because the agent constructor dereferenced a null wxTheApp when
run headless (no wxApp is created in the unit-test binary), crashing before
any assertion ran. That null dereference was fixed in 14d2dfdd4c, which
guards wxTheApp in compute_fallback_path() and skips file persistence when
no fallback path is available.
With the fix in place both tests build and pass headless, so drop the
[NotWorking] tag and the stale explanatory comments. Verified on Linux
clang-18 (the CI compiler), headless: 20 assertions in 2 test cases pass.
Closes#14193
* Disable fff_print tests that fail only in CI
Skirt height is honored, Scenario: Skirt and brim generation, and
Scenario: PrintGCode basic functionality slice geometry that makes clipper's
coordinate range check throw "Coordinate outside allowed range" in the Linux
CI environment, while the same tests pass in local builds. Tag them
[NotWorking] so the Unit Tests job (ctest -LE NotWorking) excludes them until
the underlying slicing issue is fixed in a follow-up PR.
* Trigger Build all workflow on tests/** changes
The push and pull_request path filters did not include tests/**, so a
test-only change never started the build and the Unit Tests job never ran.
Add tests/** to both filters so changes to the test suite are built and
exercised by CI.
Fix Unit Tests CI job silently running zero tests
scripts/run_unit_tests.sh selected tests with `ctest -L "Http|PlaceholderParser"`,
but catch_discover_tests() was called without ADD_TAGS_AS_LABELS, so Catch2 tags
were never registered as CTest labels. The -L filter matched nothing and the job
passed green while running no tests ("No tests were found!!!"). Tests have not run
in CI since PR #11485 added that -L line (2025-12-23).
Register tags as labels via a shared orcaslicer_discover_tests() wrapper in
tests/CMakeLists.txt (passing ADD_TAGS_AS_LABELS), routed through all five test
suites. Restore full-suite execution by replacing the narrow -L selection with a
`-LE NotWorking` exclusion, so all reliable tests gate PRs again (the suite ran in
full before #11485).
Tag the two OrcaCloudServiceAgent display-name tests [NotWorking]: their
constructor reaches wxStandardPaths::Get().GetUserDataDir(), which dereferences
the null wxTheApp in the headless test binary and segfaults on every platform.
Excluded until the agent can be constructed without the wx app context.
CI now runs 151 tests (was 0) and passes.
* Fix null-deref and arranger bugs that gate headless slicing tests
export_gcode dereferenced a null result out-param, enum serialization
dereferenced a null keys_map, and get_arrange_polys left bed_idx unseeded so
the arranger dropped items. All only affect the headless test/CLI path.
* Fix the headless test harness and add G-code test helpers
Use the real arranger, fix temp-file handling with an RAII guard, and add
layers_with_role / max_z for inspecting sliced G-code.
* Re-enable the Model construction test
* Re-enable SupportMaterial tests and add an enforced-support test
* Re-enable and extend PrintObject layer-height and perimeter tests
* Re-enable Print skirt, brim, and solid-surface tests
* Re-enable and extend PrintGCode tests
Un-hide the basic scenario (dead-key fixes, reframes, trimmed trivia) and add
initial-layer-height, sequential-order, and null-result export tests.
* Re-enable and reframe the skirt/brim tests
Detect skirt/brim by G-code role comment instead of a sentinel speed, and
resolve the previously-unfinished skirt-enclosure test.
* Replace the stale lift()/unlift() test with a z_hop test
* Delete the stub and broken Flow tests
Kris Austin