Conversation
luhenry
marked this pull request as ready for review
September 6, 2026 15:32
luhenry
added a commit
that referenced
this pull request
Sep 6, 2026
…multi-interpreter test
CasADi's ExternalProject_Add calls (ipopt, mumps, metis) don't declare BYPRODUCTS, so Ninja's stricter build-graph check fails with "missing and no known rule to make it" on their externally-built .so files. Unix Makefiles (what upstream's own CI always uses) tolerates this.
CasADi's ExternalProject_Add(metis-external ...) patch step shells out to wget (not curl) to fetch a patch file. The manylinux_2_39_riscv64 image only ships curl, so the patch step fails at ~2% into the build with "Utility wget not found in your PATH" - invisible until the end of the log because the parallel make -j keeps building every other unrelated target for the next ~20 minutes before gmake reports the overall failure. See gotcha 289.
CasADi's top-level CMakeLists.txt only appends -fPIC to CMAKE_C/CXX_FLAGS when CMAKE_SYSTEM_PROCESSOR is x86_64 or aarch64, so on riscv64 its vendored casadi-sundials static library builds without PIC and fails to link into libcasadi_sundials_common.so/libcasadi_rootfinder_kinsol.so with "relocation R_RISCV_JAL ... recompile with -fPIC". Passing -DCMAKE_POSITION_INDEPENDENT_CODE=ON forces PIC on every target regardless of that arch allowlist, with no source patch needed. See gotcha 294.
Run 34048805981 hit 8 test failures/errors in upstream's test/python/alltests.py after the wheel built and linked cleanly. test_stubs_installed was already a known gap (no .pyi stubs without CasADi's patched SWIG). Tracing the other 7 back to CasADi's misc/Dockerfile.swig (which builds jaeandersson/swig@98c1840, not stock SWIG, into ghcr.io/casadi/ci-swig for its own official wheel bindings) and diffing that fork against swig/swig shows it patches Source/Modules/python.cxx to add customdoc-gated overload/NotImplemented-fallback and merged-autodoc codegen beyond just -stubs. Building casadi's plain git checkout against this image's stock SWIG instead produces bindings with different OUTPUT-typemap tuple shapes, operator dispatch fallback, and docstring content than upstream's own ci-swig-generated ones - none of which are riscv64 codegen/numerical bugs. Deselect all 8 via a small inline unittest.TestSuite pruning step in the test job rather than patching alltests.py itself. Documented as gotcha 307.
Gotcha 307 (SWIG-fork test behavior divergence) already exists on main under the same number with identical content, per repo convention that skill/gotcha edits never ride on a port PR branch.
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
1 participant
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.
casadi3.8.0Compiles CasADi's C++ core (symbolic AD + IPOPT/MUMPS/qpOASES) and its SWIG-generated Python binding. Upstream publishes no riscv64 wheel.
Mirrors the Linux portion of upstream's
binaries.ymlbuild_flags(WITH_IPOPT/WITH_MUMPS/WITH_QPOASESplus the threading/GIL flags it sets unconditionally for every platform it ships).Differs from upstream
ghcr.io/jgillis/<target>dockcross cross-compile - no riscv64 dockcross target exists.WITH_LAPACK=ONagainst Rocky 10's system openblas-devel/lapack-devel, notWITH_BUILD_LAPACK=ON(upstream downloads+builds OpenBLAS from source).WITH_PYTHON_LIMITED_APIabi3 wheel (floor cp312) instead of a cp312/cp313/cp314/cp314t matrix - upstream itself builds a single abi3 wheel per release and never builds a free-threaded wheel at all..pyitype stubs - stub generation needs casadi's own patched SWIG fork; stock SWIG (already bundled in the manylinux image) builds the bindings themselves fine.Matrix: cp312-abi3 only (see above).
Testing
test/python/alltests.pyunittest suite against the installed wheel; its solver-availability checks (has_nlpsol/requires_nlpsol) already skip solvers this build omits.License: OK - CasADi (LGPLv3+) globs and installs every dependency's LICENSE file (IPOPT/EPL-2.0, MUMPS/CeCILL-C, METIS/Apache-2.0, OpenBLAS-LAPACK/BSD) under
casadi/include/licenses/itself; no GPL component is linked in.