The Fortran Standard, as published by the ISO (https://wg5-fortran.org/), does
not have a Standard Library. The goal of this project is to provide a community
driven and agreed upon de facto “standard” library for Fortran, called a
Fortran Standard Library (stdlib
). We have a rigorous process how stdlib
is
developed as documented in our Workflow. stdlib
is both a
specification and a reference implementation. We are cooperating with the
Fortran Standards Committee (e.g., the effort
started at the J3
committee repository) and the plan is to continue working with the Committee in
the future (such as in the step 5. in the Workflow document), so
that if the Committee wants to standardize some feature already available in stdlib
, it would
base it on stdlib
’s implementation.
The goal of the Fortran Standard Library is to achieve the following general scope:
git clone https://github.com/fortran-lang/stdlib
cd stdlib
To build the Fortran standard library you need
If your system package manager does not provide the required build tools, all build dependencies can be installed with the Python command line installer pip
:
pip install --user fypp cmake ninja
Alternatively, you can install the build tools from the conda-forge channel with the conda package manager:
conda config --add channels conda-forge
conda create -n stdlib-tools fypp cmake ninja
conda activate stdlib-tools
You can install conda using the miniforge installer.
Also, you can install a Fortran compiler from conda-forge by installing the fortran-compiler
package, which installs GFortran.
The following combinations are tested on the default branch of stdlib:
Name | Version | Platform | Architecture |
---|---|---|---|
GCC Fortran | 10, 11, 12, 13 | Ubuntu 22.04.2 LTS | x86_64 |
GCC Fortran | 10, 11, 12, 13 | macOS 12.6.3 (21G419) | x86_64 |
GCC Fortran (MSYS) | 13 | Windows Server 2022 (10.0.20348 Build 1547) | x86_64 |
GCC Fortran (MinGW) | 13 | Windows Server 2022 (10.0.20348 Build 1547) | x86_64, i686 |
Intel oneAPI LLVM | 2024.0 | Ubuntu 22.04.2 LTS | x86_64 |
Intel oneAPI classic | 2023.1 | macOS 12.6.3 (21G419) | x86_64 |
The following combinations are known to work, but they are not tested in the CI:
Name | Version | Platform | Architecture |
---|---|---|---|
GCC Fortran (MinGW) | 9.3.0, 10.2.0, 11.2.0 | Windows 10 | x86_64, i686 |
We try to test as many available compilers and platforms as possible. A list of tested compilers which are currently not working and the respective issue are listed below.
Name | Version | Platform | Architecture | Status |
---|---|---|---|---|
GCC Fortran | <9 | any | any | #296, #430 |
NVIDIA HPC SDK | 20.7, 20.9, 20.11 | Manjaro Linux 20 | x86_64 | #107 |
NAG | 7.0 | RHEL | x86_64 | #108 |
Intel Parallel Studio XE | 16, 17, 18 | OpenSUSE | x86_64 | failed to compile |
Please share your experience with successful and failing builds for compiler/platform/architecture combinations not covered above.
Configure the build with
cmake -B build
You can pass additional options to CMake to customize the build. Important options are
-G Ninja
to use the Ninja backend instead of the default Make backend. Other build backends are available with a similar syntax.-DCMAKE_INSTALL_PREFIX
is used to provide the install location for the library. If not provided the defaults will depend on your operating system, see here.-DCMAKE_MAXIMUM_RANK
the maximum array rank procedures should be generated for.
The default value is chosen as 4.
The maximum is 15 for Fortran 2003 compliant compilers, otherwise 7 for compilers not supporting Fortran 2003 completely yet.
The minimum required rank to compile this project is 4.
Compiling with maximum rank 15 can be resource intensive and requires at least 16 GB of memory to allow parallel compilation or 4 GB memory for sequential compilation.-DBUILD_SHARED_LIBS
set to on
in case you want link your application dynamically against the standard library (default: off
).-DBUILD_TESTING
set to off
in case you want to disable the stdlib tests (default: on
).-DCMAKE_VERBOSE_MAKEFILE
is by default set to Off
, but if set to On
will show commands used to compile the code.-DCMAKE_BUILD_TYPE
is by default set to RelWithDebInfo
, which uses compiler flags suitable for code development (but with only -O2
optimization). Beware the compiler flags set this way will override any compiler flags specified via FFLAGS
. To prevent this, use -DCMAKE_BUILD_TYPE=NoConfig
in conjunction with FFLAGS
.For example, to configure a build using the Ninja backend while specifying compiler optimization via FFLAGS
, generating procedures up to rank 7, installing to your home directory, using the NoConfig
compiler flags, and printing the compiler commands, use
export FFLAGS="-O3"
cmake -B build -G Ninja -DCMAKE_MAXIMUM_RANK:String=7 -DCMAKE_INSTALL_PREFIX=$HOME/.local -DCMAKE_VERBOSE_MAKEFILE=On -DCMAKE_BUILD_TYPE=NoConfig
To build the standard library run
cmake --build build
To test your build, run the test suite and all example programs after the build has finished with
cmake --build build --target test
To test only the test suite, run
ctest --test-dir build/test
Please report failing tests on our issue tracker including details of the compiler used, the operating system and platform architecture.
To install the project to the declared prefix run
cmake --install build
Now you have a working version of stdlib you can use for your project.
If at some point you wish to recompile stdlib
with different options, you might
want to delete the build
folder. This will ensure that cached variables from
earlier builds do not affect the new build.
Fortran Package Manager (fpm) is a package manager and build system for Fortran.
You can build stdlib
using provided fpm.toml
:
Option 1: From root folder
As fpm
does not currently support fypp
natively, stdlib
now proposes a python script to preprocess and build it.
This script enables modification of the different fypp
macros available in stdlib
. The preprocessed files will be dumped at <current_folder>/temp/*.f90
or *.F90
.
Make sure to install the dependencies from the requirement.txt
pip install --upgrade -r config/requirements.txt
To build, you can use the following command line:
python config/fypp_deployment.py
fpm build --profile release
or the short-cut
python config/fypp_deployment.py --build
To modify the maxrank
macro for instance:
python config/fypp_deployment.py --maxrank 7 --build
To see all the options:
python config/fypp_deployment.py --help
Note: If you use a compiler different than GNU compilers, the script will try to catch it from the environment variables FPM_FC
, FPM_CC
, FPM_CXX
.
Option 2: From the stdlib-fpm
branch which has already been preprocessed with default macros:
git checkout stdlib-fpm
fpm build --profile release
You can run the examples with fpm
as:
fpm run --example prog
with prog
being the name of the example program (e.g., example_sort
).
The stdlib project exports CMake package files and pkg-config files to make stdlib usable for other projects. The package files are located in the library directory in the installation prefix.
For CMake builds of stdlib you can find a local installation with
find_package(fortran_stdlib REQUIRED)
...
target_link_libraries(
${PROJECT_NAME}
PRIVATE
fortran_stdlib::fortran_stdlib
)
To make the installed stdlib project discoverable add the stdlib directory to the CMAKE_PREFIX_PATH
.
The usual install location of the package files is $PREFIX/lib/cmake/fortran_stdlib
.
To use stdlib
within your fpm
project, add the following lines to your fpm.toml
file:
[dependencies]
stdlib = { git="https://github.com/fortran-lang/stdlib", branch="stdlib-fpm" }
Warning
Fpm 0.9.0 and later implements stdlib as a metapackage. To include the standard library metapackage, change the dependency to:
stdlib = "*"
.
After the library has been built, it can be included in a regular Makefile. The recommended way to do this is using the pkg-config tool, for which an example is shown below.
# Necessary if the installation directory is not in PKG_CONFIG_PATH
install_dir := path/to/install_dir
export PKG_CONFIG_PATH := $(install_dir)/lib/pkgconfig:$(PKG_CONFIG_PATH)
STDLIB_CFLAGS := `pkg-config --cflags fortran_stdlib`
STDLIB_LIBS := `pkg-config --libs fortran_stdlib`
# Example definition of Fortran compiler and flags
FC := gfortran
FFLAGS := -O2 -Wall -g
# Definition of targets etc.
...
# Example rule to compile object files from .f90 files
%.o: %.f90
$(FC) -c -o $@ $< $(FFLAGS) $(STDLIB_CFLAGS)
# Example rule to link an executable from object files
%: %.o
$(FC) -o $@ $^ $(FFLAGS) $(STDLIB_LIBS)
The same can also be achieved without pkg-config.
If the library has been installed in a directory inside the compiler’s search path,
only a flag -lfortran_stdlib
is required.
If the installation directory is not in the compiler’s search path, one can add for example
install_dir := path/to/install_dir
libdir := $(install_dir)/lib
moduledir := $(install_dir)/include/fortran_stdlib/<compiler name and version>
The linker should then look for libraries in libdir
(using e.g.-L$(libdir)
) and the compiler should look for module files in moduledir
(using e.g. -I$(moduledir)
).
Alternatively, the library can also be included from a build directory without installation with
build_dir := path/to/build_dir
libdir := $(build_dir)/src
moduledir := $(build_dir)/src/mod_files
Documentation is a work in progress (see issue #4) but already available at stdlib.fortran-lang.org. This includes API documentation automatically generated from static analysis and markup comments in the source files using the FORD tool, as well as a specification document or “spec” for each proposed feature.
Some discussions and prototypes of proposed APIs along with a list of popular open source Fortran projects are available on the wiki.