This crate follows the standard for developing a Rust library via cargo
.
The CI is our "ground truth" over the state of the library. Check out the different parts of
the CI to understand how to test the different parts of this library locally.
The crate comes with additional submodules to aid with testing, to ensure you have them if you plan on testing, using --recurse-submodules
will clone the submodules alongside the repository.
git clone --recurse-submodules https://github.com/jorgecarleitao/arrow2
The simplest way to test the crate is to run
cargo test --tests
This runs the tests of the crate without features. To run all features, use
cargo test --features full
If you get warnings about parquet files not existing, you can generate the test files by using Python:
# Ubuntu: sudo apt install python3-pip python3-venv
# Mac: brew install python3
# Archlinux: sudo pacman -Syu python
# Create a virtual environment for python, to keep dependencies contained
python3 -m venv venv
# Activate the virtual environment
source venv/bin/activate
# Make sure pip is up to date
pip install pip --upgrade
# Install pyarrow, version 6
pip install pyarrow==6 pyorc
# Generate the parquet files (this might take some time, depending on your computer setup)
python parquet_integration/write_parquet.py
# generate ORC files
python tests/it/io/orc/write.py
# Get out of venv, back to normal terminal
deactivate
If you receive warnings about other files not found (IPC), ensure you have all submodules:
# If you didn't clone with `git clone --recurse-submodules https://github.com/jorgecarleitao/arrow2`
git submodule update --init --recursive
# Update to the latest submodules
git submodule update --recursive --remote
during development of particular parts of the crate, it is usually faster to reduce the feature set - the tests are gated to only the relevant tests of that feature set. For example, if improving JSON, you can use
cargo test --tests --features io_json
to only run both non-feature-specific tests and tests for json.
We currently do not have maintenance versions and thus only PR and merge to main
.
We use labels to build a changelog - it is very important to label issues and/or PRs accordingly. Because our changelog can contain both issues and PRs, when a PR closes an issue, we favor having the PR on the changelog, since it includes a reference to the author (credits).
Summary:
- pull requests with both backward-incompatible changes and new features/enhancements MUST close at least one issue (the one documenting the backward-incompatible change)
- Every other pull request MAY close one issue
issues are only used to document situations whose a single PR adds two entries to the changelog (e.g. a backward incompatible change + a new enhancement).
Merging a PR to main has the following checklist:
- Does it close an issue? If yes, add the label
no-changelog
to the issue. - Label the PR accordingly (
Testing
,Documentation
,Enhancement
,Feature
,Bug
) - If the PR is backward incompatible:
- create a new issue labeled
backward-incompatible
with what changed and how to migrate from the old API to the new API - Edit the PR's description with
Closes #...
- create a new issue labeled
- Adjust the PRs title with a description suitable for the changelog
- In the past tense
- backtick code names (e.g.
MutableArray
, not MutableArray) - place "(-X%)" or "(Yx)" at the end if it is a performance improvement
- Press merge (squash merge) and adjust any items added by github to your liking
This will reduce the burden of a release, where we go through every item from the changelog and adjust titles, labels, etc.