The Octave package for Windows ships with libgfortran from gfortran 9. This
version of libgfortran does not include symbol _gfortran_os_error_at, which is
needed by binaries created with gfortran 10 (used on our runners). The solution
is to statically link against libgfortran (as we already do for libgcc and
libstdc++).
– Windows 11 is supported
– macOS versions older than 11 are no longer supported
– Octave versions older than 6.2.0 are no longer supported
– do not claim compatibility with future (unknown) versions of MATLAB and Octave
- strip the new preprocessor binary
- include the new preprocessor binary in the .zip and .7z archives
- include the old (backward-compatible) preprocessor binary in the .exe installer
– preprocessor binary renamed to dynare-preprocessor
— the preprocessor binary is now installed under a preprocessor/ top-level directory
— a backward-compatibility symlink is still provided under
matlab/preprocessor64/dynare_m
— the make install rule now install PDF and HTML documentation when those have
been built
Closes: #1728
The runners no longer have an octave binary in the PATH, so the configure
script fails. We pass a dummy value for the location of the Octave
binary (which is actually not run, because we don’t execute the testsuite).
Add -std=legacy to FFLAGS.
Do the same for the Windows package, even though we don’t yet use version 10 of
gfortran (but that will of course happen at some point).
The detection of the slicot library failed because xerbla has been moved out of
the BLAS implementation shipped with Octave (probably since 5.1.0). This commit
fixes our fake implementation of mkoctfile so that it returns -lxerbla after
-lblas (as is done in the official Windows package for Octave).
Since 533a4ff7db, several MEX files were missing
in the Windows package (gensylv, k-order, dynare_simul, MS-SBVAR). The problem
was that the environment variable containing the directory containing MSYS2
libraries (incl. matio and gsl) was not correctly passed to parallelized
subshells.