Fixes spurious make check / CI failures on the intensive driver tests (#78).
On the difficult Moré/Garbow/Hillstrom problems, iteration counts and last
digits differ between compilers because of floating-point contraction (FMA),
even though every problem still converges. The non-portable
driver-vs-reference comparison was dropped from make check; driver validation
is now an informational cross-check of how the pure-C, f2c and FORTRAN
implementations compare. The pass/fail gates are the standard example tests and
the driver smoke tests.
- Replace
examples/crosscheck.shwithexamples/crosscheck.py(no/bin/sh
dependency); addexamples/compare.pyandexamples/driver_check.py. Wire an
informational cross-check into CMake/CTest via theCMINPACK_CROSSCHECK
option. Python 3 is optional in both build systems: when absent only the
cross-check is skipped, with a disclaimer. - Fix a compile error in
examples/tenorm_.c(passdpmpar_its index by
pointer, the FORTRAN/f2c calling convention). - Document the numerical differences from FORTRAN MINPACK and the test setup in
the README and HTML docs; correct theenorm.ccomments (the divergence is
FMA, not the rdwarf/rgiant scaling constants).
Full changelog: v1.3.13...v1.3.14