Migrate build dependencies from the dlr-sc conda channel to conda-forge - #1406
Open
joergbrech wants to merge 18 commits into
Open
Migrate build dependencies from the dlr-sc conda channel to conda-forge#1406joergbrech wants to merge 18 commits into
joergbrech wants to merge 18 commits into
Conversation
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #1406 +/- ##
==========================================
+ Coverage 73.28% 73.31% +0.02%
==========================================
Files 324 324
Lines 28642 28630 -12
==========================================
- Hits 20991 20990 -1
+ Misses 7651 7640 -11
Flags with carried forward coverage won't be shown. Click here to find out more.
🚀 New features to boost your workflow:
|
joergbrech
changed the title
Contributor
PR RemindersThis PR has changes in
This is an automated reminder from CI |
The failing wing W14_ChaDih has tip sections translated by exactly (0, 1, 1) and (0, 2, 2) — a perfect 45° dihedral. CTiglWingHelper::GetMajorDirection() picks the span axis by comparing accumulated |Y| vs |Z| with a plain >=, and for this wing those sums are mathematically equal, so the winner is decided by last-ULP rounding — which differs on arm64 (FMA contraction, Apple libm). The CI numbers prove it: 43.152389734005403 + 46.847610265994597 = 90.0 exactly — after SetSweep nudged the tip, the tie flipped the major axis and the dihedral was measured as its complement. Fix: a relative-tolerance tie-break (GreaterOrTied, 1e-8) in GetMajorDirection/GetDeepDirection (src/wing/CTiglWingHelper.cpp) so near-ties resolve deterministically to the conventional axis on every platform. 2. BSplineInterpolation.withKinksShield — over-strict test, fixed. ArraysMatch defaults to exact ==, but the 0.5 parameter is derived arithmetic (tigl::Interpolate rescaling between kink params), which lands 1 ULP off on arm64. The test now uses InTolerance(1e-14). 3. SIGABRT in makeLoft.nacelleInverted — narrowed, needs one more CI run. Three things the log tells us: (a) guideCurvePatches.nacelleInverted — same .brep inputs through MakePatches::Perform directly — passes on macOS, so the crash lives in the CTiglMakeLoft wrapper path (tolerance 1e-6 vs 1e-7, plus ConcatenateWire and CloseShape). That's why the MakePatches null-guard in your last commit couldn't fix it. (b) gtest catches C++ exceptions on POSIX, so a thrown CTiglError/Standard_Failure would show as a test failure, not an abort. (c) No libc++abi/assert message reached stderr — this is a silent abort (terminate-in-noexcept, heap check, or a direct abort()), which can't be pinpointed by reading code. So I added a temporary step to .github/workflows/build-test.yml that, when the macOS test job fails, reruns makeLoft.* under lldb --batch with thread backtrace all on crash — the next CI run will hand you the exact crash site.
joergbrech
marked this pull request as ready for review
July 29, 2026 09:37
svengoldberg
requested changes
Aug 5, 2026
svengoldberg
left a comment
svengoldberg
left a comment
Contributor
There was a problem hiding this comment.
From my end, everything looks good und consistent!
I have just some remarks regarding the unit tests.
And one additional general question (I did not dive too much into the math tbh): Don't we loose a strong feature when omitting the GeomFill_CoonsC2Style?
Contributor
|
@joergbrech Maybe, after merging, we should open up an issue reminding us to change the |
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Description
Moves TiGL's conda dependencies off the self-hosted
dlr-scchannel and ontoconda-forge, resolving #1286. This is a prerequisite to hosting tigl as a conda-forge feedstock itself.This PR also re-introduces MacOS to CI.
Concretely:
occtpackage (dynamically linked,novtkbuild), bumped 7.6.2 → 7.9.3. Theocct-staticbuild option and all static-linking machinery have been removed entirely.occt7.6.2 only ships a vtk-enabled build whose dependencies conflict withtixion every platform;novtkbuilds only exist from 7.7.2 onward, andosx-arm64support only exists from 7.9.0.creatorWing.DISABLED_wingCreateSectionInsideWithParam,WingSegmentGuideCurves.DISABLED_tiglWingGetSegmentUpperSurfaceAreaTrimmed) — most likely due to the missing G2-continuous Coons-patch patch the old dlr-sc build had, though this wasn't fully isolated from the version bump.tixipackage (sametixi3CMake target, no CMake changes needed).1.8.15.*pin.novtk). Since conda-forge's package doesn't ship the SWIG interface sources TiGL's internal Python bindings need, those are now vendored as a git submodule atthirdparty/pythonocc-core, pinned to match — TODO: drop this vendoring once conda-forge/pythonocc-feedstock ships the sources itself.4.4.*to match what conda-forge'spythonocc-coreis actually built with — a mismatch here silently breaks SWIG's cross-module type sharing (functions returning OCCT types likeTopoDS_Face/gp_Pntfell back to raw, un-destructed pointers instead of properOCC.Core.*instances).matlab-libspackage (no conda-forge equivalent exists) are now vendored atthirdparty/matlab-sdk/{win-64,osx-64}, used automatically as a fallback bycmake/FindMATLAB.cmake.tixifeedstock has been updated to build for it.occt-static; macOS moved tomacos-latest, resolving the CI breakage from themacos-13runner's retirement (CI is failing now that MacOS-13 Github runner is in retirement #1273).Also fixed two bugs discovered along the way, unrelated to the conda-forge migration itself but blocking verification of it:
cmake/GetGitRevisionDescription.cmake.in) that madetiglOpenCPACSConfigurationspuriously returnTIGL_WRONG_CPACS_VERSIONwhenever the current branch's git ref was packed (e.g. aftergit gc) rather than a loose file.cannot find -lGL) on hosts without a system-widelibGL.so/libX11.so/libfontconfig.so— conda-forge'socctexports these as bare linker names, so$CONDA_PREFIX/libnow needs to be onLIBRARY_PATHat link time.Dropping the old pinned
tbb=2019.5static dependency (now resolved entirely through conda's own dependency graph) may also help with #964.Closes #1286
Closes #1273
How Has This Been Tested?
Locally via
pixi, for thedefault,coverage, annts:pixi install) from scratch verified for all 4 platforms (linux-64,win-64,osx-64,osx-arm64) with only theconda-forgechannel.unittests+integrationtests+pythontestsrun on Linux: 902 unit tests, 46 integration tests, and 110 Python testsall pass (3 disabled, as noted above).occtpackage metadata (confirmedTKernel's CMake target points at the real.dll, not just the.libas available for a live build.Screenshots, that help to understand the changes(if applicable):
N/A (build/CI infrastructure change, no user-facing UI change)
Checklist for PR Author:
ChangeLog.mdupdated