See Good First Issues.
Create an issue with this form.
Generally we follows the Conventional Commits for pull request titles, since we will squash and merge the PR and use the PR title as the first line of commit message.
For example, here are good PR titles:
- feat(java): support xxx feature
- fix(c++): blablabla
- chore(python): remove useless yyy file
If the submitted PR affects the performance of Fury, we strongly recommend using the perf type, and need to provide benchmark data in the PR description. For how to run the benchmark, please check Fury Java Benchmark.
For more details, please check pr-lint.yml.
For environmental requirements, please check DEVELOPMENT.md.
cd python
pytest -v -s .
cd java
mvn -T10 clean test
bazel test $(bazel query //...)
cd go/fury
go test -v
go test -v fury_xlang_test.go
cd rust
cargo test
cd javascript
npm run test
Run all checks: bash ci/format.sh --all
.
docker run --rm -v $(pwd):/github/workspace ghcr.io/korandoru/hawkeye-native:v3 format
cd java
# code format
mvn spotless:apply
# code format check
mvn spotless:check
mvn checkstyle:check
cd python
# install dependencies fro styling
pip install black==22.1.0 flake8==3.9.1 flake8-quotes flake8-bugbear click==8.0.2
# format python code
black pyfury
git ls-files -- '*.cc' '*.h' | xargs -P 5 clang-format -i
cd go/fury
gofmt -s -w .
cd rust
cargo fmt
cd javascript
npm run lint
Fury supports dump jit generated code into local file for better debug by configuring environment variables:
FURY_CODE_DIR
:The directory for fury to dump generated code. Set to empty by default to skip dump code.ENABLE_FURY_GENERATED_CLASS_UNIQUE_ID
: Append an unique id for dynamically generated files by default to avoid serializer collision for different classes with same name. Set this tofalse
to keep serializer name same for multiple execution orAOT
codegen.
By using those environment variables, we can generate code to source directory and debug the generated code in next run.
cd python
python setup.py develop
- Use
cython --cplus -a pyfury/_serialization.pyx
to produce an annotated HTML file of the source code. Then you can analyze interaction between Python objects and Python's C API. - Read more: https://cython.readthedocs.io/en/latest/src/userguide/debugging.html
FURY_DEBUG=true python setup.py build_ext --inplace
# For linux
cygdb build
See the Debugging C++ doc.
Enable core dump on Macos Monterey 12.1:
/usr/libexec/PlistBuddy -c "Add :com.apple.security.get-task-allow bool true" tmp.entitlements
codesign -s - -f --entitlements tmp.entitlements /Users/chaokunyang/anaconda3/envs/py3.8/bin/python
ulimit -c unlimited
then run the code:
python fury_serializer.py
ls -al /cores
# Dtrace
sudo dtrace -x ustackframes=100 -n 'profile-99 /pid == 73485 && arg1/ { @[ustack()] = count(); } tick-60s { exit(0); }' -o out.stack
sudo stackcollapse.pl out.stack > out.folded
sudo flamegraph.pl out.folded > out.svg
Fury's website consists of static pages hosted at https://github.com/apache/fury-site.
All updates about docs under guide and benchmarks will be synced to the site repo automatically.
If you want write a blog, or update other contents about the website, please submit PR to the site repo.