KNOWN_FAILING_TESTS is now solidcore's, the default; KNOWN_FAILING_TESTS.smlib keeps
the C++ kernel's. A face copied out for untrimming now carries its tags on solidcore.
The profile test accepts either kernel's spans; ssi_corpus skips a kernel the
environment overrides.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk
KNOWN_FAILING.solidcore lists the 24 cases that fail on the default kernel, 23 of them also on SMLib's
list. The ladder golden is the C++ kernel's and solidcore differs on one case, so that gate opts in.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk
The fork places a recognized plane's origin after squaring its axes (0286682, in SMLIB_REV;
SmPlane.cpp:2836-2852 at 2fc375f), and solidcore#2004 brings the same behind this flag. It closes
sweeps/4305's Bad Gap and moves sweeps/5037 to SMLib's volume.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk
--known-failing follows the run's kernel: KNOWN_FAILING.solidcore for the default and the CI gates,
KNOWN_FAILING for --kernel smlib. The solidcore lists come with the final pin. Also fixes the --route
test for the new default.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk
ssi::intersect, domain and evaluate run on solidcore (global and advanced; the basic intersector has
no public entry and says so), and ssi_corpus scores both kernels: six cases C++ passes are listed as
solidcore's misses. A failing solidcore call now leaves its B-rep inputs and what failed, and
current_call names the running call without a lock. Op::History and Op::Flat, read by nothing, are
gone from the table, so diagnostics.kernel reports only what is true.
A guard test keeps the list of entry points: with no C++ kernel each answers, but for the C++ kernel's
own arrays. route::on_cpp now fails a test that never reached C++.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk
`shell_brep_simple`, `offset_brep_simple` and `local_operation_simple` carry the same options and
the source's attributes, so the 120 lines of key domains and notification stores in core/offset.rs
go (gap-audit D2). The held-flags comment now says what each flag's fork code does. Ignored tests cite
solidcore#1980, #1462, #1973 and #1981 instead of local files.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk
Reading the fork at SMLIB_REV finds `perspective_sphere_offsets` is there (709c5be: the cull projects
each sphere's own radius from its centre), so it goes on. The new
`gap_samples_use_trimmed_spans` and `twist_simple_duw_sign` are held: upstream's defects are still in
the fork.
The tabbed plate and #1997's rounded box are compared with the C++ kernel. `sweeps::List` is visible
to the crate for that.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk
--kernel smlib and both remain. --known-failing still grades the SMLib side, whose lists these are,
so it needs that said, and the CI gates opt in. The ladder takes --kernel for its one backend.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk
Native files run on solidcore through its public api_files (each kernel reads the other's, in both
forms). face_points routes with Split, pcurves with Convert, and an ssi::Surface no longer needs the
C API: its arguments are checked here as the C API checks them, and the C side is made only when a
family on the C++ kernel asks. intersect, domain and evaluate have no solidcore path and refuse with
Error::NotOnSolidcore. Planar region booleans and the rig CLI no longer need the smlib feature.
Tests that exercise the C++ kernel opt in per thread (route::on_cpp), so parity tests keep comparing
two kernels instead of solidcore with itself.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk
The routing table now sends each family to solidcore; the C++ SMLib kernel is an opt-in
(SOLIDCRAFT_KERNEL=smlib, the Geometry kernel preference, SOLIDCRAFT_KERNEL_ROUTES). Native .smb
files stay on SMLib here; the next commit moves them. Fallout follows in separate commits.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk
On, their fork commits in SMLIB_REV: `fillet_variable_span_follows_fillet` (4438859),
`simple_coincidence_crossing_is_point` (c2e3d9d), and from solidcore#1775 the statuses whose fork
passes do what the flag does: `boolean_temporary_transform_statuses` (e7ebc8d),
`primitive_path_curve_statuses`, `planar_path_sweep_statuses` and
`planar_path_profile_placement_statuses` (c8f6a31), `fillet_trim_status` (9b7cd48). The last two
were held as having no fork commit, which was wrong.
Held: `fillet_corner_statuses`, `fillet_edge_curve_domain_statuses`,
`offset_tangent_gap_statuses`, `ssi_alignment_statuses` (the fork carries on softly where these
refuse) and `translation_sweep_regularize` (no fork commit).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk
`face_tight_box_searches_trim` goes on now that solidcore#1961 brings fork 326de73.
`ssi_low_angle_start_on_probation` goes on: fork 30d25bf was rebased to 1f5c0f7, in SMLIB_REV.
Three port fixes and the new `extent3d_contains_both_segment_endpoints` stay held: the defects are
still in the fork's source at 2fc375f.
`measure::check` no longer matches `check_brep`'s operation string. solidcore does not raise that
error (a Caller body without keys is checked under arena identities), so the arm was dead; the
refusal is decided on the typed `NamingScheme` alone, and a test holds it there.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk
On: fillet_mitre_gap_patch (fork 2223108), fillet_run_on_past_far_end_only (7224896), good_drop_result_distances_in_order (eaa998a). Held back: geometric_g2_continuity and solidcore's new flags with no fork commit in SMLIB_REV.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk
The ordering (fork 241f7ba) is ported but off (OVERFLOW_GOES_FIRST): solidcore's cliff overflow loses the attributes of target edges its subtraction replaces, so a shortened picked edge reads as removed and is silently not rounded (2 rig cases wrong). Tests: the capi's rounds_at_a_concave_face_corner_are_mitred; overflow_goes_first (ignored, gives 6107.529369565 with it on); two ignored solidcore-only repros.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk
L1: GetWireEdges uses the process kernel, not a fresh default one.
L3: a face whose surface the B-rep's list does not hold is refused, as smlib_flat.cpp:1520.
L4: a hyperbola or parabola without its NURB makes one on a copy, as GetOrCreateGwNurbPointer.
L5: every rejection of the C++ Validate is in ours, read line by line; a mutation test against
the C++ kernel holds it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk
Flags on: fillet_cliff_overflow, fillet_3x2_tangent_side_edge_roles, offset_extension_as_itself, fillet_mitre_concave_face_corners (fork commits in SMLIB_REV 2fc375f). Held back: the port's own fixes with no fork commit. Ports capi blend_call / overflow_first (fork 241f7ba) onto the hooks of solidcore #1929, with blend_dropping routed. From ws/cliff-1905: blend_failure on NoRoomReason, fillet_over_a_cliff un-ignored, a_cliff_overflow_the_fork_refuses_has_no_room.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk
Closed edges including moved()'s open.is_empty() branch, a curved face split across its
seam, union_all of three, merges with several origins on each side, sheet offset, a
curved sheet's thicken and a shell of a cylinder match C++. Two tests are ignored for
solidcore differences written up apart: boolean_2d carries no attributes, and a flat
sheet's thicken loses the edge tags on the sheet's side.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk
smlib_history.cpp:246 counts any failure of SmApiEdgeClosestPoint as off the piece, so
the face is Modified. The port passed on an error that was no status, which failed a
boolean that had succeeded. An edge without its curve shows it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk
Found comparing drafts while checking M2: the capi's status (smlib_shim.h:82) answers
SM_ERR_OFFSET_EDGE_VANISHES (1043) as SMLIB_ERR_UNSUPPORTED; the port passed on Status(1043).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk
Port the capi's moved_faces (smlib_ops.cpp:156-165, :1078), its mark on the picked
faces as solidcore#1875/#1877 lay out, since a surface has no identity apart from
its owning face. The reason it was skipped (solidcore#1462) was stale. No draft
found reaches SMLIB_ERR_NO_EFFECT on C++, so the parity test holds the same answer,
volume and counts over the nearest cases, and a unit test holds the reading.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk