f90wrap 0.3.0

Fortran to Python interface generator with derived type support

f90wrap: Fortran to Python interface generator with derived type support

f90wrap is a tool to automatically generate Python extension modules which interface to Fortran code that makes use of derived types. It builds on the capabilities of the popular f2py utility by generating a simpler Fortran 90 interface to the original Fortran code which is then suitable for wrapping with f2py, together with a higher-level Pythonic wrapper that makes the existance of an additional layer transparent to the final user.

Copyright (C) James Kermode 2011-2021. Released under the GNU Lesser General Public License, version 3. Parts originally based on f90doc - automatic documentation generator for Fortran 90. Copyright (C) 2004 Ian Rutt.

If you would like to license the source code under different terms, please contact James Kermode [e-mail]

Dependencies

1. Python 3.9+ (Python 2.7 no longer supported)
2. Recent version of numpy which includes f2py
3. Fortran compiler - tested with gfortran 4.6+ and recent ifort 12+

Installation

For the latest stable release, install with pip:

pip install f90wrap

There is also a conda package on conda-forge:

conda install -c conda-forge f90wrap

For the development version, installation is as follows:

pip install git+https://github.com/jameskermode/f90wrap

Note that if your Fortran 90 compiler has a non-standard name (e.g. gfortran-9) then you need to set the F90 environment variable prior to installing f90wrap to ensure it uses the correct one, e.g.

F90=gfortran-9 pip install f90wrap

Examples and Testing

To test the installation, run make test from the examples/ directory. You may find the code in the various examples useful.

Citing f90wrap

If you find f90wrap useful in academic work, please cite the following (open access) publication:

J. R. Kermode, f90wrap: an automated tool for constructing deep Python interfaces to modern Fortran codes. J. Phys. Condens. Matter (2020) doi:10.1088/1361-648X/ab82d2

BibTeX entry:

@ARTICLE{Kermode2020-f90wrap,
  title    = "f90wrap: an automated tool for constructing deep Python
              interfaces to modern Fortran codes",
  author   = "Kermode, James R",
  journal  = "J. Phys. Condens. Matter",
  month    =  mar,
  year     =  2020,
  keywords = "Fortran; Interfacing; Interoperability; Python; Wrapping codes;
              f2py",
  language = "en",
  issn     = "0953-8984, 1361-648X",
  pmid     = "32209737",
  doi      = "10.1088/1361-648X/ab82d2"
}

Case studies

f90wrap has been used to wrap the following large-scale scientific applications:

• QUIP - molecular dynamics code
• CASTEP - CasPyTep wrappers for electronic structure code
• QEpy - Python wrapper for Quantum Espresso electronic structure code

See this Jupyter notebook from a recent seminar for more details.

Usage

To use f90wrap to wrap a set of Fortran 90 source files and produce wrappers suitable for input to f2py use:

f90wrap -m MODULE F90_FILES

where MODULE is the name of the Python module you want to produce (e.g. the name of the Fortran code you are wrapping) and F90_FILES is a list of Fortran 90 source files containing the modules, types and subroutines you would like to expose via Python.

This will produce two types of output: Fortran 90 wrapper files suitable for input to f2py to produce a low-level Python extension module, and a high-level Python module desinged to be used together with the f2py-generated module to give a more Pythonic interface.

One Fortran 90 wrapper file is written for each source file, named f90wrap_F90_FILE.f90, plus possibly an extra file named f90wrap_toplevel.f90 if there are any subroutines or functions defined outside of modules in F90_FILES.

To use f2py to compile these wrappers into an extension module, use:

f2py -c -m _MODULE OBJ_FILES f90wrap_*.f90 *.o

where _MODULE is the name of the low-level extension module.

Optionally, you can replace f2py with f2py-f90wrap, which is a slightly modified version of f2py included in this distribution that introduces the following features:

1. Allow the Fortran present() intrinsic function to work correctly with optional arguments. If an argument to an f2py wrapped function is optional and is not given, replace it with NULL.
2. Allow Fortran routines to raise a RuntimeError exception with a message by calling an external function f90wrap_abort(). This is implemented using a setjmp()/longjmp() trap.
3. Allow Fortran routines to be interrupted with Ctrl+C by installing a custom interrupt handler before the call into Fortran is made. After the Fortran routine returns, the previous interrupt handler is restored.

Direct-C mode extensions

Quick build: f90wrap --build -m mymodule source.f90

Manual compilation: f90wrap --direct-c -m mymodule source.f90

Python package (pyproject.toml + setup.py):

[build-system]
requires = ["setuptools", "numpy", "f90wrap"]

[project]
name = "mypackage"
version = "0.1.0"

[tool.setuptools.packages]
find = {}

# setup.py
from setuptools import setup
from f90wrap.setuptools_ext import F90WrapExtension, build_ext_cmdclass

setup(ext_modules=[F90WrapExtension("mymodule", ["src/mymodule.f90"])],
      cmdclass=build_ext_cmdclass())

Result: import mypackage then use mypackage.mymodule

Notes

• Unlike standard f2py, f90wrap converts all intent(out) arrays to intent(in, out). This was a deliberate design decision to allow allocatable and automatic arrays of unknown output size to be used. It is hard in general to work out what size array needs to be allocated, so relying on the the user to pre-allocate from Python is the safest solution.
• Scalar arguments without intent are treated as intent(in) by f2py. To have inout scalars, you need to call f90wrap with the --default-to-inout flag and declare the python variables as 1-length numpy arrays (numpy.zeros(1) for example).
• Pointer arguments are not supported.
• Arrays of derived types are currently not fully supported: a workaround is provided for 1D-fixed-length arrays, i.e. type(a), dimension(b) :: c. In this case, the super-type Type_a_Xb_Array will be created, and the array of types can be accessed through c.items. Note that dimension b can not be :, but can be a parameter.
• Doxygen documentation in Fortran sources are parsed and given as docstring in corresponding python interfaces. Doxygen support is partial and keyword support is limited to brief, details, file, author and copyright.

Troubleshooting

NumPy 2.0+ and Meson Backend Issues

Starting with NumPy 2.0, the f2py build system transitioned from distutils to meson. This can cause issues with f90wrap, particularly when Fortran module files (.mod files) cannot be found during compilation.

Symptom: Errors like "Cannot open module file 'modulename.mod' for reading" during the f2py-f90wrap compilation step.

Workarounds:

1. Set FFLAGS to include current directory (recommended):

   export FFLAGS="-I$(pwd)"
   f2py-f90wrap -c -m _mymodule f90wrap_*.f90 *.o
   

2. Use the --build-dir option with f2py-f90wrap:

   f2py-f90wrap --build-dir build -c -m _mymodule f90wrap_*.f90 mymodule.f90
   

   Note: When using --build-dir, pass the original .f90 source files instead of pre-compiled .o files, as the meson backend needs to compile them in the build directory where the .mod files will be generated.

3. Downgrade to NumPy < 2.0 (temporary):

   pip install 'numpy<2.0'
   

Note: The f2py-f90wrap script includes patches to help with the meson backend, including automatic handling of include paths and library directories when using --build-dir. If you encounter issues, please report them at https://github.com/jameskermode/f90wrap/issues

How f90wrap works

There are five steps in the process of wrapping a Fortran 90 routine to allow it to be called from Python.

1. The Fortran source files are scanned, building up an abstract symbol tree (AST) which describes all the modules, types, subroutines and functions found.

2. The AST is transformed to remove nodes which should not be wrapped (e.g. private symbols in modules, routines with arguments of a derived type not defined in the project, etc.)

3. The f90wrap.f90wrapgen.F90WrapperGenerator class is used to write a simplified Fortran 90 prototype for each routine, with derived type arguments replaced by integer arrays containing a representation of a pointer to the derived type, in the manner described in Pletzer2008. This allows opaque references to the true Fortran derived type data structures to be passed back and forth between Python and Fortran.

4. Standard mode (default): f2py is used to combine the F90 wrappers and the original compiled functions into a Python extension module (optionally, f2py can be replaced by f2py-f90wrap, a slightly modified version which adds support for exception handling and interruption during exceution of Fortran code).

   Direct-C mode (--direct-c): As an alternative to f2py, the f90wrap.directc_cgen package generates C code using the Python C API to call the F90 wrappers directly. This eliminates the f2py dependency.

5. The f90wrap.pywrapgen.PythonWrapperGenerator class is used to write a thin object-oriented layer on top of the f2py (or Direct-C) generated wrapper functions which handles conversion between Python object instances and Fortran derived-type variables, converting arguments back and forth automatically.

Advanced Features

Additional command line arguments can be passed to f90wrap to customize how the wrappers are generated. See the examples/ directory to see how some of the options are used:

  -h, --help            show this help message and exit
  -v, --verbose         set verbosity level [default: None]
  -V, --version         show program's version number and exit
  -p PREFIX, --prefix PREFIX
                        Prefix to prepend to arguments and subroutines.
  -c [CALLBACK [CALLBACK ...]], --callback [CALLBACK [CALLBACK ...]]
                        Names of permitted callback routines.
  -C [CONSTRUCTORS [CONSTRUCTORS ...]], --constructors [CONSTRUCTORS [CONSTRUCTORS ...]]
                        Names of constructor routines.
  -D [DESTRUCTORS [DESTRUCTORS ...]], --destructors [DESTRUCTORS [DESTRUCTORS ...]]
                        Names of destructor routines.
  -k KIND_MAP, --kind-map KIND_MAP
                        File containting Python dictionary in f2py_f2cmap
                        format
  -s STRING_LENGTHS, --string-lengths STRING_LENGTHS
                        "File containing Python dictionary mapping string
                        length names to values
  -S DEFAULT_STRING_LENGTH, --default-string-length DEFAULT_STRING_LENGTH
                        Default length of character strings
  -i INIT_LINES, --init-lines INIT_LINES
                        File containing Python dictionary mapping type names
                        to necessary initialisation code
  -I INIT_FILE, --init-file INIT_FILE
                        Python source file containing code to be added to
                        autogenerated __init__.py
  -A ARGUMENT_NAME_MAP, --argument-name-map ARGUMENT_NAME_MAP
                        File containing Python dictionary to rename Fortran
                        arguments
  --short-names SHORT_NAMES
                        File containing Python dictionary mapping full type
                        names to abbreviations
  --py-mod-names PY_MOD_NAMES
                        File containing Python dictionary mapping Fortran
                        module names to Python ones
  --class-names CLASS_NAMES
                        File containing Python dictionary mapping Fortran type
                        names to Python classes
  --joint-modules JOINT_MODULES
                        File containing Python dictionary mapping modules
                        defining times to list of additional modules defining
                        methods
  -m MOD_NAME, --mod-name MOD_NAME
                        Name of output extension module (without .so
                        extension).
  -M, --move-methods    Convert routines with derived type instance as first
                        agument into class methods
  --shorten-routine-names
                        Remove type name prefix from routine names, e.g.
                        cell_symmetrise() -> symmetrise()
  -P, --package         Generate a Python package instead of a single module
  -a ABORT_FUNC, --abort-func ABORT_FUNC
                        Name of Fortran subroutine to invoke if a fatal error
                        occurs
  --auto-raise-error AUTO_RAISE_ERROR
                        Generate calls to abort subroutine in f90wrap fortran wrappers: 'err_num_variable,err_msg_variable'
  --only [ONLY [ONLY ...]]
                        Subroutines to include in wrapper
  --skip [SKIP [SKIP ...]]
                        Subroutines to exclude modules and subroutines from
                        wrapper
  --skip-types [SKIP_TYPES [SKIP_TYPES ...]]
                        Subroutines to exclude types from wrapper
  --force-public [FORCE_PUBLIC [FORCE_PUBLIC ...]]
                        Names which are forced to be make public
  --default-to-inout    Sets all arguments without intent to intent(inout)
  --conf-file CONF_FILE
                        Use Python configuration script to set options
  --documentation-plugin DOCUMENTATION_PLUGIN
                        Use Python script for expanding the documentation of
                        functions and subroutines. All lines of the given tree
                        object are passed to it with a reference to its
                        documentation
  --py-max-line-length PY_MAX_LINE_LENGTH
                        Maximum length of lines in python files written.
                        Default: 80
  --f90-max-line-length F90_MAX_LINE_LENGTH
                        Maximum length of lines in fortan files written.
                        Default: 120
  --type-check          Check for type/shape matching of Python argument with the wrapped Fortran subroutine
  --direct-c            Generate direct-C extension instead of relying on f2py

Author

James Kermode jameskermode

Key Contributors

• Tamas Stenczel stenczelt
• Steven Murray steven-murray
• Greg Corbett gregcorbett
• Bob Fischer citibob
• David Verelst davidovitch
• James Orr jamesorr
• yvesch
• Matthias Cuntz
• Balthasar Reuter reuterbal
• Daniel Vanzo danbeibei
• Christopher Albert krystophny

Developer Notes

Triggering the wheel build

Wheels are built on push and pull requests to master using cibuildwheel with this workflow.

To make a release candidate create a tag with a suffix such as -rc1 for the first attempt, push to trigger the build:

git commit -m 'release v0.x.z.rc1'
git tag v0.x.y.rc1
git push --tags

If all goes well, the .whl files will show up as assets within a new GitHub release. The installation process can now be tested locally.

Release wheels to PyPI

Once everything works correctly, make a full release (i.e. create a tag named just v0.x.y without the -rc1 suffix). This will trigger the upload of wheels and source distribution to PyPI.