Proposed in the July 2026 AI-assisted code review (Claude Code), from investigation of the v1.4 development branch (octave).
-fmex emits a pure Fortran 90 MEX gateway (.f90) plus mwrap-fmex-support.f90 — no C shim, attractive for Fortran-centric projects (FMM libraries): src/mwrap-fmexgen.cc + src/mwrap-fmex-support.f90, mirrored in python/mwrap_fmexgen.py + python/mwrap_fmex_support.f90. -i8 maps to integer*8; -fmex -gpu is rejected in both implementations; a scalar-return fix is already on the branch. The branch has a CMake generation smoke test, testing/test_fmex1.mw, and the larger testing/nufft1d suite.
As with the other backends, the branch's experimental nocopy modifier is excluded from this port (see #45 and #21).
Before it lands on master:
- a compile/run test with a Fortran compiler in CI (gfortran is available on ubuntu runners)
- C++/Python byte-parity coverage in
testing/test_python.sh
- user-guide/README documentation including the supported type subset
See #45 for dispatch-table compatibility and merge scope.
Proposed in the July 2026 AI-assisted code review (Claude Code), from investigation of the v1.4 development branch (
octave).-fmexemits a pure Fortran 90 MEX gateway (.f90) plusmwrap-fmex-support.f90— no C shim, attractive for Fortran-centric projects (FMM libraries):src/mwrap-fmexgen.cc+src/mwrap-fmex-support.f90, mirrored inpython/mwrap_fmexgen.py+python/mwrap_fmex_support.f90.-i8maps tointeger*8;-fmex -gpuis rejected in both implementations; a scalar-return fix is already on the branch. The branch has a CMake generation smoke test,testing/test_fmex1.mw, and the largertesting/nufft1dsuite.As with the other backends, the branch's experimental
nocopymodifier is excluded from this port (see #45 and #21).Before it lands on master:
testing/test_python.shSee #45 for dispatch-table compatibility and merge scope.