Switch to github-actions

Advantages:

- unified configuration format for linux/windows/mac
- simplified usage of artifacts (compared to travis): no more external
  manually managed artifact repository
- artifacts should work on pull requests even without secrets (wasn't
  the case previously which meant that all PR builds always failed)
- unified access to all artifacts on github actions tab
- cache control with explicit cache key
- better integration into github UI
- pipeline can include jobs on different OSes
- easier configuration of parallel workflows
- ...?

.github/build/manylinux1/cpymad.sh

@@ -0,0 +1,42 @@
+#! /usr/bin/env bash
+set -ex
+
+# Build cpymad from checked out sources.
+# Expects a built madx distribution in '../MAD-X/dist'.
+# Builds in './build' and places wheels in './dist'.
+
+: ${PY:=/opt/python/cp36-cp36m/bin}
+
+# Build variables:
+export MADXDIR=$(readlink -nf ../MAD-X/dist)
+export X11=0 BLAS=0 LAPACK=0
+export CFLAGS="-flto"
+export LDFLAGS="-flto"
+
+# Copy the cpymad source files to a build folder in order to avoid permission
+# issues with the host filesystem (on both sides):
+mkdir -p build
+$PY/python setup.py egg_info
+tar -c $(cat src/cpymad.egg-info/SOURCES.txt) |
+    tar -x -C build --no-same-owner
+
+# We create the wheels from the source distribution to verify that the
+# source distribution can be used as installation medium. We will later
+# upload this exact source distribution to PyPI:
+pushd build
+$PY/pip install cython
+$PY/python setup.py sdist
+$PY/pip uninstall cython -y
+
+for PYBIN in /opt/python/*/bin; do
+    "${PYBIN}/pip" install -U setuptools
+    "${PYBIN}/pip" wheel dist/*.tar.gz --no-deps -w dist/
+done
+popd
+
+# Bundle external shared libraries into the wheels
+mkdir -p dist
+for whl in build/dist/*.whl; do
+    auditwheel repair "$whl" -w dist/
+done
+cp build/dist/*.tar.gz dist/

.github/build/manylinux1/madx.sh

@@ -0,0 +1,27 @@
+#! /usr/bin/env bash
+set -ex
+
+# Build MAD-X static library from prepared sources.
+# Must be run from the root the directory of the MAD-X sources.
+# Builds in './build' and installs to './dist'.
+
+rm -rf build
+mkdir build
+cd build
+
+: ${PY:=/opt/python/cp36-cp36m/bin}
+
+$PY/pip install --upgrade cmake
+$PY/cmake .. \
+    -DBUILD_SHARED_LIBS=OFF \
+    -DMADX_STATIC=ON \
+    -DCMAKE_INSTALL_PREFIX=../dist \
+    -DCMAKE_BUILD_TYPE=Release \
+    -DCMAKE_C_FLAGS="-fvisibility=hidden -flto" \
+    -DCMAKE_CXX_FLAGS="-fvisibility=hidden -flto" \
+    -DCMAKE_Fortran_FLAGS="-fvisibility=hidden -flto" \
+    -DMADX_INSTALL_DOC=OFF \
+    -DMADX_ONLINE=OFF \
+    -DMADX_FORCE_32=OFF \
+    -DMADX_X11=OFF
+$PY/cmake --build . --target install

.github/build/msys2/cpymad.sh

@@ -0,0 +1,112 @@
+#! /usr/bin/env bash
+set -ex
+
+# Build cpymad from checked out sources.
+# Expects a built madx distribution in '../MAD-X/dist'.
+# Builds in './build' and places wheels in './dist'.
+
+main()
+{
+    ARCH=$1
+    MADXDIR=${2:-$MADXDIR}
+
+    # Create python environments:
+    conda_ create -qyf -n py27 python=2.7 wheel cython -c anaconda
+    conda_ create -qyf -n py35 python=3.5 wheel cython -c anaconda
+    conda_ create -qyf -n py36 python=3.6 wheel cython -c anaconda
+    conda_ create -qyf -n py37 python=3.7 wheel cython -c anaconda
+    conda_ create -qyf -n py38 python=3.8 wheel cython -c anaconda
+
+    # Build cpymad wheels:
+    if [[ $ARCH == i686 ]]; then
+        CFLAGS=
+        build py27 27 win32-2.7 ''
+        build py35 35 win32-3.5 .cp35-win32
+        build py36 36 win32-3.6 .cp36-win32
+        build py37 37 win32-3.7 .cp37-win32
+        build py38 38 win32-3.8 .cp38-win32
+    else
+        CFLAGS=-DMS_WIN64
+        build py27 27 win-amd64-2.7 ''
+        build py35 35 win-amd64-3.5 .cp35-win_amd64
+        build py36 36 win-amd64-3.6 .cp36-win_amd64
+        build py37 37 win-amd64-3.7 .cp37-win_amd64
+        build py38 38 win-amd64-3.8 .cp38-win_amd64
+    fi
+}
+
+# We manually build the C extension using our msys gcc because setuptools is
+# not smart enough to figure out how to build it. The downside is that
+# we link a different C runtime than is natively used by python. This will
+# result in horrible evil should we ever mix C objects/memory between python
+# and cpymad!
+build()
+{
+    py_env=$1
+    py_ver=$2
+    dir_tag=$3
+    file_tag=$4
+
+    # Ensure that cython code and extension module will be rebuilt since the
+    # cython code is partially incompatible between python versions:
+    rm -f src/cpymad/libmadx.c \
+          src/cpymad/libmadx.pyd
+
+    # We use a two stage build with the exact filenames as `python setup.py
+    # build_ext` would do (compile `.c` to `.obj` in $tempdir, then link to
+    # `.pyd` in $libdir) to prevent the final `python setup.py bdist_wheel`
+    # command from trying trying to perform either of these steps with MSVC.
+
+    conda_ activate $py_env
+    tempdir=build/temp.$dir_tag/Release/src/cpymad
+    libdir=build/lib.$dir_tag/cpymad
+    mkdir -p $tempdir
+    mkdir -p $libdir
+
+    pythondir="$(python -c 'import sys; print(sys.prefix)')"
+
+    # This will cythonize `.pyx` to `.c`:
+    pip install -U setuptools
+    python setup.py build_py
+
+    # We turn back on the base environment for building in order to set the
+    # the path to the runtime DLLs required for running gcc. Without this
+    # the command errors with a windows error that is visible only via the
+    # remote desktop but doesn't get logged as console output.
+    conda_ deactivate
+
+    gcc -mdll -O -Wall -flto $CFLAGS \
+        -I$MADXDIR/include \
+        -I$pythondir/include \
+        -c src/cpymad/libmadx.c \
+        -o $tempdir/libmadx.obj \
+        -std=gnu99
+
+    # Linking directly against the `pythonXX.dll` is the only way I found to
+    # satisfy the linker in a conda python environment. The conventional
+    # command line `-L$pythondir/libs -lpython$py_ver` used to work fine on
+    # WinPython, but fails on conda with large number of complaints about
+    # about undefined references, such as `__imp__Py_NoneStruct`,
+    gcc -shared -s -flto \
+        $tempdir/libmadx.obj \
+        -L$MADXDIR/lib \
+        -static \
+        -lmadx -lDISTlib -lptc -lgc-lib \
+        -lstdc++ -lgfortran -lquadmath \
+        $pythondir/python$py_ver.dll \
+        -o $libdir/libmadx$file_tag.pyd
+
+    # Turn target python environment back on, see above:
+    conda_ activate $py_env
+    python setup.py bdist_wheel
+    conda_ deactivate
+}
+
+conda_() {
+    # Conda with disabled trace (really noisy otherwise):
+    { set +x; } 2>/dev/null
+    conda "$@"
+    { set -x; } 2>/dev/null
+}
+
+main "$@"

.github/build/msys2/madx.sh

@@ -0,0 +1,24 @@
+#! /usr/bin/env bash
+set -ex
+
+# Build MAD-X static library from prepared sources.
+# Must be run from the root the directory of the MAD-X sources.
+# Builds in './build' and installs to './dist'.
+
+rm -rf build
+mkdir build
+cd build
+
+# Build MAD-X as library:
+pip install --upgrade cmake
+cmake .. \
+    -G "MinGW Makefiles" \
+    -DBUILD_SHARED_LIBS=OFF \
+    -DMADX_STATIC=ON \
+    -DCMAKE_INSTALL_PREFIX=../dist \
+    -DCMAKE_BUILD_TYPE=Release \
+    -DMADX_INSTALL_DOC=OFF \
+    -DMADX_ONLINE=OFF \
+    -DMADX_FORCE_32=OFF \
+    -DMADX_X11=OFF
+cmake --build . --target install

.github/checkout-madx/action.yml

@@ -0,0 +1,14 @@
+name: Download and prepare MAD-X sources
+inputs:
+  madx_version:
+    description: 'MAD-X version'
+    required: true
+
+runs:
+  using: composite
+  steps:
+    - run: git clone https://github.com/MethodicalAcceleratorDesign/MAD-X
+        ../MAD-X -b "${{ inputs.madx_version }}" --depth 1
+      shell: bash
+    - run: patch -d ../MAD-X -p1 < .github/patch/fix-cmake-Fortran_FLAGS.patch
+      shell: bash