Installation on Windows 10

[gyanz]

I was able to install clawpack and run few python examples on a Windows 10 machine with the following steps:

• Defined CONDA_BUILD=1 environment variable to prevent error due to stdout redirection.
• Ran pull_all.sh in clawpack source directory.
• Commented out the line that adds the "weno" subpackage in clawpack\pyclaw\src\pyclaw\limiters\setup.py. This bypasses the "unresolved external PyInit_reconstruct" link error.
• Commented out line # 74 in clawpack\pyclaw\src\pyclaw\controller.py. This bypasses the issue with setting logging verbosity.
• I noticed problem with numpy.distutils.fcompiler.FCompiler class. The compiler_f90 attribute for this class instance was ['C:\Program', 'Files', '(x86)\Intel\oneAPI\compiler\latest\windows\bin\intel64\ifort.exe', '/nologo', '/MD', '/nbs', '/names:lowercase', '/assume:underscore', '/O1', '/assume:minus0']. The first entry in the list is supposed to be the fortran compiler path, but this path was split into multiple entries. Same was the issue with compiler_f77 and compiler_fix attributes. I was able to resolve this issue by revising the customize method of FCompiler class. I commented out the lines f77 = _shell_utils.NativeParser.split(f77), f90 = _shell_utils.NativeParser.split(f90), fix = _shell_utils.NativeParser.split(fix) and replaced them with f77 = [f77], f90 = [f90], fix = [fix] respectively. I am using Numpy 1.21.1.
• Ran python setup.py build_ext --inplace in clawpack source directory
• Ran pip install -e . in clawpack source directory

Initially, I had compiled clawpack source using Anaconda's Python 3.7.10 64-bit, but the extension modules gave "DLL load failed: The specified procedure could not be found" error while running the python examples. So I switched to python (same version) that was installed along with MS Visual Studio 2019 Community Edition. I used Intel Fortran Intel 64 Compiler (Version 2021.3.0, Build 20210609_000000) and MSVC x64 (Version 19.29.30133) for compilation.

I tested the installation using nose following this instruction. The results show few errors - see the attachment below.
nosetests_result.txt

[mandli]

Thanks @gyanz! I am curious if you tried the fix that is in the master branch for the weno package. I think that has been fixed but if there is another problem with the compilation we should fix that. Also, just to confirm you are running this outside of the WSL?

[gyanz]

@mandli Yes, I ran it outside of WSL on Windows 10 Build 19043.1165, and I am using master branch for the weno package. I think the module initialization in weno/reconstruct.c must be revised as follows. I was able to compile and run nosetests after this revision.

PyMODINIT_FUNC
#if PY_MAJOR_VERSION >= 3
PyInit_reconstruct (void)
#else
initreconstruct (void)
#endif
{
PyObject *module;
#if PY_MAJOR_VERSION >= 3
  static struct PyModuleDef moduledef = {
    PyModuleDef_HEAD_INIT,
    "reconstruct",       /* m_name */
    NULL,                /* m_doc */
    -1,                  /* m_size */
    reconstructmethods,  /* m_methods */
    NULL,                /* m_reload */
    NULL,                /* m_traverse */
    NULL,                /* m_clear */
    NULL                 /* m_free */
  };
  module = PyModule_Create(&moduledef);
  if (module == NULL)
      return NULL;
  return module;  
#else
  module = Py_InitModule ("reconstruct", reconstructmethods);
  if (module == NULL)
     return NULL;
#endif
}

[mandli]

So this is similar but not identical to #667. Do you understand the additional code here that makes the difference?

[gyanz]

I have further edited my code for initialization of reconstruct python module. Extension modules are initialized differently in Python 2 and 3. For module named "ABC", initialization function is initABC in Python 2 and PyInit_ABC in Python 3. Furthermore, the initABC function does not have return value while the PyInit_ABC returns the module PyObject. Both functions return NULL if there is an error. In my suggested code, I have removed the output_array() call because I thought it served no purpose for module initialization. Can you please check if this call should be placed somewhere else in the source code?

[mandli]

Beyond understanding the basics of the differences between the C API I cannot provide much myself as I was not the author of the code nor do I venture this far deep into the C of Python. The problem is clearly whether certain compilers ignore what should really be considered bugs. If we can come up with a fix though that seems to address this that would be awesome.

In terms of your actual question, given that this code was auto-generated from a now I think defunct package I think it's safe to make the assumption that it was not being used. There are tests in the test suite though so if this fix works on Travis, your own windows setup, and a modern macOS I think we may be good. I can test the mac part of that if you issue a PR with your changes.

[gyanz]

I have issued PR with revision to reconstruct.c. I have no problem compiling pyclaw and loading the reconstruct module in Python 3 on Windows 10. I don't know yet the best way to test the methods in this module as I am new to clawpack. I will look forward to seeing the automated tests results.