You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
solidcore: Shell keeps cavity regions without ambiguous point-solve owners; retire offset_keep_cavity_regions (#17393) - #17452
An inset solid offset keeps only its cavity regions by default (WR-667), and asks which regions are cavity only when the merge leaves two or more finite regions. UpstreamFixes::offset_keep_cavity_regions retires: its fixed path is the only path.
Changes a default: no
The kernel default changes to what SolidCraft's preset (crates/smlib/src/core.rsfixes()) already ran. The one difference from that preset: a single-region inset no longer runs the cavity verdicts, so their side effects on the merged body (UV trims, caches, tolerance growth from the live point classification) are not made. Measured on the goldens and the app suites (below).
Test change: retirement of offset_keep_cavity_regions.
fork_37af643_inset_keeps_cavity_regions.rs loses its flag-off test of the old path (flag_off_refuses_the_sliver_and_retains_the_input, which pins behaviour that no longer exists; refusal-and-retention stays covered by issue_3094_offset_refuses_lost_solid, wr_531_shell_drilled_hole_tip, fork_1b016ca_cavity_floor_under_opening, wr_342_vase_bottom_clipped_right_side_out). Its thicker_floor_is_unchanged (flag off == flag on, bitwise) cannot compare two paths any more and becomes thicker_floor_is_the_closed_form (669 to 1e-9). The flag-on test runs on the default kernel with an unchanged expectation, plus check_brep at every level on the main frame. Its offset-fixes variant (flag_on_with_offset_fixes, flag plus WR-455's one-pass stitch) is dropped: both flags it set are retired (solidcore: offset merge stitch widens only where a trial shows it completes the solid; retire offset_stitch_at_body_tolerance #17531), so it is keeps_only_the_cavity exactly.
wr567_revolution_round_offset_collapses.rs: wr567_default_fails_open_at_the_other_end pinned !shelled(...), the old default's 1001 when the merge's cavity regions were not selected. With the selection always on, that shell is the closed form, so the test becomes wr567_without_collapse_open_at_the_other_end_keeps_the_cavity: it asserts the shell builds and passes check_brep at every level. This is a changed expectation and it is proven by that check and by the case's closed form (the cavity, the cylinder r 15 - t, 5 - t deep less the round; the same closed form the collapse-test sibling already asserts). No expectation is widened. That this merge leaves two or more finite regions is ARGUED: with fewer the selection is skipped and the result could not have changed.
With the flag on, do_extended_offset builds the operand's point-solve cache (p1::make_original_brep_cache) for the first finite region it tests. OriginalKeys::new refuses a body whose geometry owners don't name one edge or face each.
issue_1476_swept_bar_shell.rs used to read its operand with key_of = |_| ObjectKey(0): every edge and face had the same name, so every curve and surface had the same owner. The refusal is correct for such a body: the point solve can't tell its faces apart. check_brep refuses the same body ("topology identities alias distinct edges") (MEASURED).
The swept bar's merge leaves one finite region, though. With one finite region the cavity verdict cannot change anything: select is only set when one region is cavity and another is not. A lone cavity region is every region (select false); a lone non-cavity region leaves cavities empty (select false). So make_manifold gets None either way (ARGUED from the code; the block is 20 lines in p2.rs).
Fix: run the verdicts, and build the cache, only when there are two or more finite regions. Which regions make_manifold keeps cannot change (ARGUED, above). The verdicts also edit result (the live point classification leaves UV trims, caches and possibly tolerance growth on it); those side effects are no longer made for a single-region inset, so "no result change" is MEASURED, not proven: goldens and app suites below. It removes a refusal that had no bearing on which regions are kept, plus a body cache and 3 BrepPointSolves per face on every single-region inset.
A body with ambiguous owners whose merge leaves two or more regions is still refused with the same Unsupported. No fallback.
shell builds, closed form to 1e-6 (the fixture's tolerance, as the collapse sibling), check_brep L0–L2 valid; that the merge leaves ≥2 finite regions is deduced (ARGUED)
pinned the 1001 failure
The closed form for the 1.0 floor: body 2000 − 200 − 220·2 = 1360. Cavity: z 0.5..2.5 over the core plus band, (25 + 80.5)·2 = 211, and z 2.5..4.5 over (0.5..19.5)² less (4.5..15.5)², 240·2 = 480. Shell: 1360 − 691 = 669 (PROVEN by the asserted closed form).
The #[ignore]d issue_1476_swept_bar_shells_with_solidcraft_fixes{,_and_edges_on_analytics_exact} already fail on main: 1299.482307072064 vs the 1e-9 pin 1299.482314966473 (6e-9 relative; MEASURED). This PR gives identical values and leaves the pins alone.
§7 checklist
The skip is decided on the region count, a topological count, with no sampling. 2–3. n/a (no numeric acceptance). 4. No tolerance or expectation loosened; the one changed expectation (wr567) is asserted by check_brep and the closed form. 5. The remaining refusal is unchanged. 6. No new loops or recursion. 7. Goldens: see the gate. 8. Real defect still caught: the frame's sliver region is still dropped (keeps_only_the_cavity).
Full planned gate (sc-gate.sh, all goldens + rig) on de806e8, exit 0:
dad68c0 (after the gate) changes docs/comments only.
App tests touching shell/offset (cargo nextest run --cargo-profile ci, engine shell_regressions, shell_opening_regressions; kernel shell_barrel, chamfer_curved_holder): 13 run, 13 passed. MEASURED: no app-visible result change on these; full engine/CLI suites run on CI.
Inset solid offsets now automatically evaluate cavity regions when a merge produces multiple finite regions. When only one finite region remains, it is kept without cavity classification.
Shelling succeeds for a previously failing revolution-offset case and passes additional geometry validation.
Documentation
Updated offset and shell guidance to explain when cavity regions are evaluated and how to interpret the results.
Inset solid offsets now select cavity regions by default when a merge leaves multiple finite regions. The kernel fix flag is removed. Tests and documentation are updated to reflect the behavior and validate related shell results.
Changes: Inset solid offset cavity selection
Layer / File(s)
Summary
Apply cavity selection without a flag crates/smlib/src/core.rs, solidcore/crates/solidcore/src/kernel.rs, solidcore/crates/solidcore/src/offset/offset_executive/p2.rs, solidcore/crates/solidcore/src/offset/offset_executive/cavity.rs
The public offset_keep_cavity_regions flag and its preset assignment are removed. Inset solid offsets now classify finite regions only when more than one remains; a single finite region skips cavity classification.
Tests validate default cavity selection and B-rep results for inset, swept-bar, and revolution-offset shells. Test flag lists remove the retired flag. Documentation and the addendum describe when cavity classification runs.
Inset offsets now keep cavity regions by default, matching the existing SolidCraft behavior, and single-region insets skip unnecessary cavity checks. The review found no concrete regression. The only open suggestion is an optional test of the single-region bypass.
Security architecture risk: Minimal · up to dad68
The change alters a public setting and default behavior, but the inspected paths retain their failure and input-ownership safeguards. No introduced security-boundary or failure-containment weakness was established.
Retained concerns
No architecture-level concerns identified.
Security review details
Security Blast Radius
inferred — The demonstrated exposure change is geometric: caller-supplied inset geometry with multiple finite regions now enters an existing classification path by default and can produce a different result or refusal. This branch does not introduce a new principal, credential, or external service destination. Broader application and tenant exposure was not established by the bounded dependency evidence.
Trust Boundaries and Controls
inferred — The skipped single-region owner-uniqueness check is a precondition of an omitted geometry-cache operation, not an observed authentication or authorization boundary. Its omission therefore does not establish an identity-control bypass. Downstream reliance on incidental classification side effects remains only partially traced.
Resilience and Maintainability Implications
observed — The routed test-file entrypoints are test functions, not new production access paths. Updated assertions cover the default cavity result, single-region closed-form volume, and validation levels; the swept-bar validation test is ignored by default because of its runtime. These assertions were inspected but not executed.
Pre-merge checks: 5
Check name
Status
Explanation
Title check
The title names the main behavior change: shell keeps cavity regions by default, with the ambiguous point-solve owner handling, and the retired flag. The changeset removes the flag and gates cavity ve…
Description check
The description covers the change, the default-behavior line, the test-change justification required for kernel test edits, the root cause, and a validation section with measured results. It does not …
Linked Issues check
Directly linked issue #17393 requires the issue_1476 swept-bar shell to build a solid of volume 40L - 28(L - 0.5), L = 60 + 15π, with the cavity-region path active (the SolidCraft preset flag). Releva…
Out of Scope Changes check
All changed files relate to retiring UpstreamFixes::offset_keep_cavity_regions and making the region-count rule the fixed inset path: the removed kernel field (kernel.rs), the removed preset line (s…
Existing Defaults Unchanged
The check asks whether the PR changes a UX default that a user already relies on. It excludes kernel and geometry results, including anything under solidcore/, and treats them as not a changed default…
Full details: Title check
Explanation
The title names the main behavior change: shell keeps cavity regions by default, with the ambiguous point-solve owner handling, and the retired flag. The changeset removes the flag and gates cavity verdicts on two or more finite regions, which the title reflects. The title is long (118 characters) and dense, but it is related to the main change.
Full details: Description check
Explanation
The description covers the change, the default-behavior line, the test-change justification required for kernel test edits, the root cause, and a validation section with measured results. It does not follow the template's screenshot section, which is required only for observable CAD behavior. Here the change is a kernel result change with no UI-visible effect, and the author states app-visible results are unchanged. The Integration status section is missing. The template asks for the upstream commit and PR status checked before handoff, any conflicts resolved, and post-integration validation. The description also lacks the template's Title line, which is an HTML comment and not a required section. The description is largely complete but omits the Integration status section.
Resolution
Consider adding the Integration status section from the template (upstream commit and PR status checked, conflicts resolved, validation after integration). Optionally state why no screenshots apply, since the change has no UI-visible effect.
Full details: Linked Issues check
Explanation
Directly linked issue #17393 requires the issue_1476 swept-bar shell to build a solid of volume 40L - 28(L - 0.5), L = 60 + 15π, with the cavity-region path active (the SolidCraft preset flag). Relevant evidence: (1) The default-kernel test issue_1476_swept_bar_shells asserts Ok, manifold, and |V - closed form| ≤ 1e-5·V with 16 line edges. The PR reports this test passes on main and at the reviewed head, and the PR's measured V = 1299.487679574382 matches the closed form (≈1299.48768). Since the flag is retired, the default kernel runs the cavity path for two or more finite regions. (2) The new ignored SolidCraft-set test issue_1476_swept_bar_shells_with_solidcraft_fixes_validates asserts the same volume bound and check_brep Levels 0–2. The PR reports SolidCraft-set V = 1299.482307, a deviation of about 5.4e-3 (about 4e-6 relative), inside the 1e-5 bound. (3) The reported ambiguous-owner refusal is stated unreachable on main, because the reader refuses zero-named identities since #17064 and the fixture names entities apart. The PR also skips the verdicts and cache for single-region insets. Limits: the code gate in p2.rs was not readable in the inspection, so the single-region skip is supported by the PR's argument and the goldens and app-suite results, not by code read here. The ignored _fixes pin tests (1e-9) already fail on main, and this PR leaves them unchanged; they do not establish a requirement failure against the issue's closed-form expectation.
Full details: Out of Scope Changes check
Explanation
All changed files relate to retiring UpstreamFixes::offset_keep_cavity_regions and making the region-count rule the fixed inset path: the removed kernel field (kernel.rs), the removed preset line (smlib core.rs), the gated selection in p2.rs, the module and API docs (cavity.rs, offset-and-shell.md), the removed flag references in issue_2260 and fixes.json, the reworked fork_37af643 tests (flag-off/on comparison replaced by the closed form), the wr567 expectation change (the shell now builds under the always-on selection; the PR states the closed form), the swept-bar SolidCraft-set validation test, and the WR-667 plan addendum. The check_brep additions in issue_1476 and wr567 support the changed claims. No file was shown to change behavior unrelated to the flag or the swept-bar and cavity cases.
Full details: Existing Defaults Unchanged
Explanation
The check asks whether the PR changes a UX default that a user already relies on. It excludes kernel and geometry results, including anything under solidcore/, and treats them as not a changed default. The changed-file inventory (refs/coderabbit/pre-merge-check/base..head) has 11 files: - crates/smlib/src/core.rs: removes one internal kernel preset line, f.offset_keep_cavity_regions = true;, from fixes(). This is a kernel flag, not a user-facing setting, preference, shortcut, layout, or command default. - solidcore/... (kernel.rs, offset/offset_executive/cavity.rs, p2.rs, tests, fixes.json, docs, plan addendum): retires the UpstreamFixes::offset_keep_cavity_regions flag. The kernel default inset behavior changes, so an inset keeps only its cavity regions by default. This is a changed kernel/geometry result and falls under the explicit exclusion. No UI, CLI, preference, shortcut, or command-default files change. The PR description says Changes a default: no. That label is inaccurate for the kernel default. The failure clause only applies when the code changes a UX default, and this PR does not. The kernel-default change is excluded by the check's own terms. Verdict: PASS. No UX default changes. The kernel default change is a solidcore geometry result, which the check excludes.
Finishing Touches
Generate unit tests (beta)
Create a new PR
Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts
…n two or more regions
Retire `UpstreamFixes::offset_keep_cavity_regions` (WR-667): an inset solid offset
always passes make_manifold only the finite regions that are cavity. The cavity
verdicts (and the body's point-solve cache they need) are now asked only when the
merge leaves two or more finite regions: with one, make_manifold keeps it whatever
the verdict, so the selection cannot change the result.
The swept bar of issue_1476 (read with every entity under one name, so no geometry
owner tells its edges or faces apart) failed with Unsupported ("point-solve geometry
owners must identify one edge or face each") building that cache under the flag,
although its merge has one finite region. It now shells to its closed form on the
default kernel.
Tests: the frame of fork_37af643 runs on the default kernel with check_brep at every
level, its 1.0-floor control is checked against its closed form, and its flag-off test
of the old path is deleted; issue_1476 gains a distinct-names test and a SolidCraft-set
test validated at every level. The flag leaves SolidCraft's preset, fixes.json and
issue_2260's list.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SdcgEtsWyH5fjgiavRsRdt
…ffset
Review fixes: the region list for the cavity selection is built only for an inset
solid offset (outsets and sheet offsets skip it); the docs say the selection is asked
only with two or more finite regions and keeps every region when all or none are
cavity; the swept-bar test's doc says the point solves ran under the former opt-in.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SdcgEtsWyH5fjgiavRsRdt
Overlap note: #17567 makes issue_1476_swept_bar_shell.rs read the preset from kernel_cases/fixes.json, and its two SolidCraft-fixes tests turn offset_keep_cavity_regions off there, citing #17393. Whichever of the two PRs lands second drops that one line (solidcraft_fixes_without_cavity_regions becomes solidcraft_fixes) and removes the flag's name from fixes.json. #17567 also adds a smlib test that keeps fixes.json equal to core::fixes().
Written by Claude Code (Anthropic) on behalf of @brett19.
Update to my earlier note: after merging main (#17064, typed reader identities), #17567 no longer turns offset_keep_cavity_regions off in issue_1476_swept_bar_shell.rs. At its head the flag makes no difference there. The only remaining overlap is the flag's name in kernel_cases/fixes.json, which #17567's smlib test keeps equal to core::fixes().
Written by Claude Code (Anthropic) on behalf of @brett19.
Conflict in issue_1476_swept_bar_shell.rs: main's reader now refuses identities that
name no edge (tangency-labs#17064), so the operand is read with notify_ref there; the branch's
one-name variant goes, keeping its check_brep validation helper and the SolidCraft-set
validation test on main's fixture.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SdcgEtsWyH5fjgiavRsRdt
Resolves crates/smlib/src/core.rs: main retired ellipse_step_interval_checks,
this branch retires offset_keep_cavity_regions; both lines go.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SdcgEtsWyH5fjgiavRsRdt
Resolves the fork_37af643 test conflict with tangency-labs#17522: the retired
offset_resolve_crossing_loops line goes; the default-kernel test with the
one-pass stitch stays.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BUqX4Moqx5bbkSocgEbsSV
tangency-labs#17531 merged (offset_stitch_at_body_tolerance retired): the shell keeps
main's stitch_merged_offset call and this branch's unconditional cavity
selection. fork_37af643's offset-fixes variant is dropped: with both
flags it set retired, it is keeps_only_the_cavity exactly.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BUqX4Moqx5bbkSocgEbsSV
…ts cavity
wr567_default_fails_open_at_the_other_end pinned the 1001 the inset gave
while it kept every finite region. With the cavity selection (WR-667)
always on, that shell builds; the test becomes
wr567_without_collapse_open_at_the_other_end_keeps_the_cavity and asserts
the shell builds and passes check_brep at levels 0, 1 and 2.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BUqX4Moqx5bbkSocgEbsSV
Unverified finding
Verification ran but could not confirm this finding. It is shown for review, not as a verified issue.
Add a regression that detects single-region classification.
The closed-form assertions do not detect whether a single-region inset builds the cache or runs cavity verdicts. The PR reports that the swept-bar result already matches before this change. Add a focused test that observes the bypass and fails when the region-count guard is removed. This protects the cache and mutation behavior that this guard changes.
As per path instructions, “Each fix needs a test that fails without it.”
Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 5 remain after this review.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
approved-for-ciReviewed: run the full PR CI (build, tests, goldens, Clippy)ready to mergereviewed, every ask addressed; merge once CI is green
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
An inset solid offset keeps only its cavity regions by default (WR-667), and asks which regions are cavity only when the merge leaves two or more finite regions.
UpstreamFixes::offset_keep_cavity_regionsretires: its fixed path is the only path.Fixes #17393
Part of #14246
Changes a default: no
The kernel default changes to what SolidCraft's preset (
crates/smlib/src/core.rsfixes()) already ran. The one difference from that preset: a single-region inset no longer runs the cavity verdicts, so their side effects on the merged body (UV trims, caches, tolerance growth from the live point classification) are not made. Measured on the goldens and the app suites (below).Test change: retirement of
offset_keep_cavity_regions.fork_37af643_inset_keeps_cavity_regions.rsloses its flag-off test of the old path (flag_off_refuses_the_sliver_and_retains_the_input, which pins behaviour that no longer exists; refusal-and-retention stays covered byissue_3094_offset_refuses_lost_solid,wr_531_shell_drilled_hole_tip,fork_1b016ca_cavity_floor_under_opening,wr_342_vase_bottom_clipped_right_side_out). Itsthicker_floor_is_unchanged(flag off == flag on, bitwise) cannot compare two paths any more and becomesthicker_floor_is_the_closed_form(669 to 1e-9). The flag-on test runs on the default kernel with an unchanged expectation, pluscheck_brepat every level on the main frame. Its offset-fixes variant (flag_on_with_offset_fixes, flag plus WR-455's one-pass stitch) is dropped: both flags it set are retired (solidcore: offset merge stitch widens only where a trial shows it completes the solid; retire offset_stitch_at_body_tolerance #17531), so it iskeeps_only_the_cavityexactly.wr567_revolution_round_offset_collapses.rs:wr567_default_fails_open_at_the_other_endpinned!shelled(...), the old default's 1001 when the merge's cavity regions were not selected. With the selection always on, that shell is the closed form, so the test becomeswr567_without_collapse_open_at_the_other_end_keeps_the_cavity: it asserts the shell builds and passescheck_brepat every level. This is a changed expectation and it is proven by that check and by the case's closed form (the cavity, the cylinder r 15 - t, 5 - t deep less the round; the same closed form the collapse-test sibling already asserts). No expectation is widened. That this merge leaves two or more finite regions is ARGUED: with fewer the selection is skipped and the result could not have changed.Root cause (#17393)
do_extended_offsetbuilds the operand's point-solve cache (p1::make_original_brep_cache) for the first finite region it tests.OriginalKeys::newrefuses a body whose geometry owners don't name one edge or face each.issue_1476_swept_bar_shell.rsused to read its operand withkey_of = |_| ObjectKey(0): every edge and face had the same name, so every curve and surface had the same owner. The refusal is correct for such a body: the point solve can't tell its faces apart.check_breprefuses the same body ("topology identities alias distinct edges") (MEASURED).selectis only set when one region is cavity and another is not. A lone cavity region is every region (selectfalse); a lone non-cavity region leavescavitiesempty (selectfalse). Somake_manifoldgetsNoneeither way (ARGUED from the code; the block is 20 lines inp2.rs).make_manifoldkeeps cannot change (ARGUED, above). The verdicts also editresult(the live point classification leaves UV trims, caches and possibly tolerance growth on it); those side effects are no longer made for a single-region inset, so "no result change" is MEASURED, not proven: goldens and app suites below. It removes a refusal that had no bearing on which regions are kept, plus a body cache and 3BrepPointSolves per face on every single-region inset.Unsupported. No fallback.issue_1476fixture names entities apart (notify_ref). The reported failure is therefore unreachable on main today; what remains of this PR for solidcore: shelling the issue_1476 swept bar fails with offset_keep_cavity_regions on (point-solve geometry owners) #17393 is the skip above, a performance and robustness change that removes a refusal and a body cache with no bearing on the result. Read with distinct names the swept bar builds both before and after this PR: 27 faces, V = 1299.487679574382 (MEASURED). SolidCraft names entities apart (crates/smlib/src/core/convert.rskey_of), so SolidCraft was probably not hit by this exact failure.Tests
issue_1476_swept_bar_shells(main's fixture, distinct names)issue_1476_swept_bar_shells_with_solidcraft_fixes_validates(new,#[ignore]like its slow siblings)check_brepLevel0/1/2 validfork_37af643keeps_only_the_cavity(the cavity-keeping case: two finite regions, sliver dropped)check_brepLevel0/1/2 validfork_37af643thicker_floor_is_the_closed_formwr567_without_collapse_open_at_the_other_end_keeps_the_cavitycheck_brepL0–L2 valid; that the merge leaves ≥2 finite regions is deduced (ARGUED)The closed form for the 1.0 floor: body 2000 − 200 − 220·2 = 1360. Cavity: z 0.5..2.5 over the core plus band, (25 + 80.5)·2 = 211, and z 2.5..4.5 over (0.5..19.5)² less (4.5..15.5)², 240·2 = 480. Shell: 1360 − 691 = 669 (PROVEN by the asserted closed form).
Found on the way (not changed here)
check_brepTestsolvecraft-cli snapshotignores body and face colours #4 and Debug builds: split_body fails with "Two same vertices cannot construct an edge" #13 (MEASURED). This PR doesn't change that, because the selection doesn't run for that body. It is the same on main. Under SolidCraft's set, the shell is valid at every level. So the default-kernel test doesn't assertcheck_brep; the SolidCraft-set test does.#[ignore]dissue_1476_swept_bar_shells_with_solidcraft_fixes{,_and_edges_on_analytics_exact}already fail on main: 1299.482307072064 vs the 1e-9 pin 1299.482314966473 (6e-9 relative; MEASURED). This PR gives identical values and leaves the pins alone.§7 checklist
check_brepand the closed form. 5. The remaining refusal is unchanged. 6. No new loops or recursion. 7. Goldens: see the gate. 8. Real defect still caught: the frame's sliver region is still dropped (keeps_only_the_cavity).Full planned gate (
sc-gate.sh, all goldens + rig) on de806e8, exit 0:dad68c0 (after the gate) changes docs/comments only.
App tests touching shell/offset (
cargo nextest run --cargo-profile ci, engineshell_regressions,shell_opening_regressions; kernelshell_barrel,chamfer_curved_holder): 13 run, 13 passed. MEASURED: no app-visible result change on these; full engine/CLI suites run on CI.🤖 Generated with Claude Code
https://claude.ai/code/session_01BUqX4Moqx5bbkSocgEbsSV
Summary by CodeRabbit