Please file issues on the GitHub issue tracker: https://github.com/apache/arrow-adbc/issues
Some dependencies are required to build and test the various ADBC packages.
For C/C++, you will most likely want a Conda installation, with Mambaforge being the most convenient distribution. If you have Mambaforge installed, you can set up a development environment as follows:
$ mamba create -n adbc --file ci/conda_env_cpp.txt
$ mamba activate adbc
(For other Conda distributions, you will likely need create ... -c conda-forge --file ...
).
There are additional environment definitions for development on Python and GLib/Ruby packages.
Conda is not required; you may also use a package manager like Nix or Homebrew, the system package manager, etc. so long as you configure CMake or other build tool appropriately. However, we primarily develop and support Conda users.
All libraries here contained within one CMake project. To build any
library, pass the -DADBC_COMPONENT=ON
flag to your cmake invocation,
replacing _COMPONENT
with the name of the library/libraries.
Note: unlike the Arrow C++ build system, the CMake projects will not automatically download and build dependencies—you should configure CMake appropriately to find dependencies in system or package manager locations, if you are not using something like Conda.
You can use CMake presets to build and test:
$ mkdir build
$ cd build
$ cmake ../c --preset debug
# ctest reads presets from PWD
$ cd ../c
$ ctest --preset debug --test-dir ../build
You can also manually configure CMake. For example, the driver manager and postgres driver may be built together as follows:
$ mkdir build
$ cd build
$ export CMAKE_EXPORT_COMPILE_COMMANDS=ON
$ cmake ../c -DADBC_DRIVER_POSTGRESQL=ON -DADBC_DRIVER_MANAGER=ON
$ make -j
export CMAKE_EXPORT_COMPILE_COMMANDS=ON
is
not required, but is useful if you are using Visual Studio Code,
Emacs, or another editor that integrates with a C/C++ language server.
For information on what each library can do and their dependencies, see their individual READMEs.
To specify where dependencies are to the build, use standard CMake
options such as CMAKE_PREFIX_PATH
.
Some build options are supported:
ADBC_BUILD_SHARED
,ADBC_BUILD_STATIC
: toggle building the shared/static libraries.ADBC_BUILD_TESTS
: build the unit tests (requires googletest/gmock).ADBC_INSTALL_NAME_RPATH
: setinstall_name
to@rpath
on MacOS. Usually it is more convenient to explicitly set this toOFF
for development.
All libraries use the same build options to enable tests. For example, to build and run tests for the SQLite3 driver:
$ mkdir build
$ cd build
# You may need to set -DCMAKE_PREFIX_PATH such that googletest can be found
$ cmake ../c -DADBC_BUILD_TESTS=ON -DADBC_DRIVER_SQLITE=ON
$ make -j
$ ctest
Tests use Googletest. Some libraries may have additional test-time dependencies. For instance, the PostgreSQL and Flight SQL drivers require servers to test against. See their individual READMEs for details.
The documentation uses the Sphinx documentation generator.
A list of dependencies for Conda (conda-forge) is included, and can be used as follows:
$ conda create -n adbc -c conda-forge --file ci/conda_env_docs.txt
$ conda activate adbc
# Mermaid must be installed separately
# While "global", it will end up in your Conda environment
$ npm install -g @mermaid-js/mermaid-cli
To build the HTML documentation:
$ pushd c/apidoc
$ doxygen
$ popd
# Optionally: to also build the Python documentation
$ pushd python/adbc_driver_manager
$ pip install -e .[test]
$ popd
$ cd docs
$ make html
The output can be found in build/
.
Some documentations are maintained as Mermaid diagrams, which must be rendered and checked in. This can be done as follows:
cd docs
make -f mermaid.makefile -j all
# Check in the updated files
The GLib bindings use the Meson build system.
A build of the driver manager is
required. For example, if the libraries are installed to
$HOME/local
:
$ meson setup \
--pkg-config-path=$HOME/local/lib/pkgconfig \
--prefix=$HOME/local \
build/glib \
glib
$ meson install -C build/glib
A list of dependencies for Conda (conda-forge) is included, and can be used as follows:
$ conda create -n adbc -c conda-forge --file ci/conda_env_glib.txt
$ conda activate adbc
Go libraries are a standard Go project.
$ cd go/adbc
$ go build -v ./...
$ go test -v ./...
The Java components are a standard Maven project.
$ cd java/
# Build and run tests
$ mvn clean install
Python libraries are managed with setuptools. See individual READMEs for additional dependencies. In general, that means all projects can be built as follows:
$ cd python/adbc_driver_manager
$ pip install -e .
Tests use pytest. Some libraries may have additional test-time dependencies. See their individual READMEs for details.
# Install dependencies
$ pip install -e .[test]
# Run tests
$ pytest -vvx
Type checking is done with pyright. There is a script to run the type checker:
# Build native libraries first
$ env ADBC_USE_ASAN=0 ADBC_USE_UBSAN=0 ./ci/scripts/cpp_build.sh $(pwd) $(pwd)/build
# Install Python packages
$ ./ci/scripts/python_build.sh $(pwd) $(pwd)/build
# Run type checker
$ ./ci/scripts/python_typecheck.sh $(pwd)
The Ruby libraries are bindings around the GLib libraries.
The Rust components are a standard Rust project.
$ cd rust
# Build and run tests
$ cargo test
Before opening a pull request, please run the static checks, which are
enforced via pre-commit
. This will run
linters, formatters, and other analysis. For example:
# Install pre-commit
$ pip install pre-commit
# or alternatively
$ conda install -c conda-forge --file ci/conda_env_dev.txt
# Set up hooks
$ pre-commit install
# Run manually
$ pre-commit run
Check Xml............................................(no files to check)Skipped
Check Yaml...........................................(no files to check)Skipped
Fix End of Files.....................................(no files to check)Skipped
Trim Trailing Whitespace.............................(no files to check)Skipped
clang-format.........................................(no files to check)Skipped
cmake-format.........................................(no files to check)Skipped
cpplint..............................................(no files to check)Skipped
Google Java Formatter................................(no files to check)Skipped
black................................................(no files to check)Skipped
flake8...............................................(no files to check)Skipped
isort................................................(no files to check)Skipped
# Hooks automatically run on commit
$ git commit
When committing, please follow Conventional Commits. This helps maintain semantic versioning of components.
Please use the following commit types: build
, chore
, ci
, docs
,
feat
, fix
, perf
, refactor
, revert
, style
, test
.
Please use the following scopes:
c/driver/postgresql
,java/driver-manager
, …: for a component and all its bindings. For example,c/driver-manager
covers the C/C++ driver manager and its GLib and Python bindings, whilejava/driver/flight-sql
covers only the Flight SQL driver for Java. (The scope names are derived from the filesystem paths.)c/format
,go/format
,java/format
: for the core API definitions (adbc.h for C/C++, adbc.go for Go, adbc-core for Java).
For example:
feat(c/driver/postgresql): implement prepared statements
ci(go/adbc/drivermgr): pass through DYLD_LIBRARY_PATH in tests
fix(java/driver/jdbc): adjust SQL type mapping for JDBC driver
In order to collect the licenses for our Go-dependencies we leverage the
tool github.com/google/go-licenses
. We have a template containing the
non-go licenses, and then you can install go-licenses
with:
$ go install github.com/google/go-licenses@latest
You can generate the LICENSE.txt with the following command:
$ cd go/adbc && go-licenses report ./... \
--ignore github.com/apache/arrow-adbc/go/adbc \
--ignore github.com/apache/arrow/go/v11 \
--ignore github.com/apache/arrow/go/v12 \
--ignore github.com/apache/arrow/go/v13 \
--ignore github.com/apache/arrow/go/v15 \
--template ../../license.tpl > ../../LICENSE.txt 2> /dev/null
You will have to manually fix up the license, since some packages do not fill out their metadata correctly and things like READMEs may end up in the license.