CI: Test GMT dev version on Windows by building from source

[seisman]

Description of proposed changes

In the "GMT Dev Tests" workflow, the way we get the GMT dev version is different on Linux/macOS and Windows. We build the GMT master branch source code on Linux/macOS, but on Windows, we install the dev version from the conda-forge/dev channel. The conda-forge/dev channel is updated every few weeks to months, so it's not always update-to-date. We choose to use the conda-forge/dev channel for Windows because building GMT on Windows is difficult.

However, it turns out that building GMT on Windows is much easier than I initially though, as long as we install all GMT dependencies via conda.

In this PR, I update the workflow to build GMT from source code on Windows, so that we're always using the same GMT dev version on all three platforms.

Supersedes #756

Reminders

• Run make format and make check to make sure the code follows the style guide.
• Add tests for new features or tests that would have caught the bug that you're fixing.
• Add new public functions/methods/classes to doc/api/index.rst.
• Write detailed docstrings for all functions/methods.
• If wrapping a new module, open a 'Wrap new GMT module' issue and submit reasonably-sized PRs.
• If adding new functionality, add an example to docstrings or tutorials.
• Use underscores (not hyphens) in names of Python files and directories.

Slash Commands

You can write slash commands (/command) in the first line of a comment to perform
specific operations. Supported slash commands are:

• /format: automatically format and lint the code
• /test-gmt-dev: run full tests on the latest GMT development version

[weiji14]

Nice! The compilation on Windows worked at https://github.com/GenericMappingTools/pygmt/actions/runs/6680586107/job/18153844748#step:8:405. I'll approve this since the build works, but I have one comment about setting more build flags (which can be done separately if needed).

[weiji14]

We're not using many of the build flags in https://github.com/conda-forge/gmt-feedstock/blob/8cf880960420215125ee43b1cbd24b3fc6d03291/recipe/bld.bat#L10-L28. Is that ok?

[seisman]

We set the CMAKE_PREFIX_PATH flag to tell CMake where to find these libraries, so these flags are not necessary:

      -D FFTW3_ROOT=%LIBRARY_PREFIX% ^
      -D GDAL_ROOT=%LIBRARY_PREFIX% ^
      -D NETCDF_ROOT=%LIBRARY_PREFIX% ^
      -D PCRE_ROOT=%LIBRARY_PREFIX% ^
      -D ZLIB_ROOT=%LIBRARY_PREFIX% ^
      -D CURL_ROOT=%LIBRARY_PREFIX% ^

Parallelization support. Can be enable if necessary.

      -D GMT_ENABLE_OPENMP=TRUE ^
      -D GMT_USE_THREADS=TRUE ^

Not necessary:

      -D GMT_LIBDIR=lib ^

GSHHG and DCW data are not installed, but they can be automatically downloaded when needed.

      -D DCW_ROOT=%DCW_DIR% ^
      -D GSHHG_ROOT=%GSHHG_DIR% ^
      -D COPY_GSHHG=TRUE ^
      -D COPY_DCW=TRUE ^

Unnecessary flags for our "GMT Dev Tests":

      -D GMT_INSTALL_TRADITIONAL_FOLDERNAMES=FALSE ^
      -D GMT_INSTALL_MODULE_LINKS=FALSE ^

[weiji14]

Parallelization support. Can be enable if necessary.

Adding this would be good!

[seisman]

Done in c6b1baa.

[seisman]

It's unclear to me why we have a dvc error in this branch, which lead to ~200 failures

ERROR: unexpected error - failed to load directory ('4d', 'd0ad31844cb0b0b451648cda314e2a.dir'): [Errno 2] No such file or directory: 'D:\\a\\pygmt\\pygmt\\.dvc\\cache\\files\\md5\\4d\\d0ad31844cb0b0b451648cda314e2a.dir'

The scheduled runs two days ago don't have the issue https://github.com/GenericMappingTools/pygmt/actions/runs/6661401871/job/18104207392.

[seisman]

/test-gmt-dev