CONTRIBUTING.md
June 1, 2026 ยท View on GitHub
Example Workflow for Contributing
(provided by @spytheman)
(If you don't already have a GitHub account, please create one. Your GitHub username will be referred to later as 'YOUR_GITHUB_USERNAME'. Change it accordingly in the steps below.)
-
Fork https://github.com/vlang/vsl using GitHub's interface to your own account. Let's say that the forked repository is at
https://github.com/YOUR_GITHUB_USERNAME/vsl. -
Clone the main vsl repository https://github.com/vlang/vsl to a local folder on your computer (
git clone https://github.com/vlang/vsl) -
cd vsl -
git remote add pullrequest https://github.com/YOUR_GITHUB_USERNAME/vslNB: the remote namedpullrequestshould point to YOUR own forked repo, not the main v repository! After this, your local cloned repository is prepared for making pullrequests, and you can just do normal git operations such as:git pullgit statusand so on. -
When finished with a feature/bugfix/change, you can:
git checkout -b fix_alabala -
git push pullrequest# (NOTE: thepullrequestremote was setup on step 4) -
On GitHub's web interface, go to: https://github.com/vlang/vsl/pulls
Here the UI shows a dialog with a button to make a new pull request based on the new pushed branch. (Example dialog: https://url4e.com/gyazo/images/364edc04.png)
-
After making your pullrequest (aka, PR), you can continue to work on the branch
fix_alabala... just do againgit push pullrequestwhen you have more commits. -
If there are merge conflicts, or a branch lags too much behind VSL's main, you can do the following:
git pull --rebase origin main# solve conflicts and dogit rebase --continuegit push pullrequest -f# this will overwrite your current remote branch with the updated version of your changes.
The point of doing the above steps, is to never directly push to the main VSL
repository, only to your own fork. Since your local main branch tracks the
main VSL repository's main, then git checkout main, as well as
git pull --rebase origin main will continue to work as expected
(these are actually used by v up) and git can always do it cleanly.
Git is very flexible, so there are other ways to accomplish the same thing. See the GitHub flow for more information.
Using Github's hub CLI tool
You can download the hub tool from https://hub.github.com/ . Using
hub, you will not need to go through the (sometimes) slow website
to make PRs. Most remote operations can be done through the hub CLI
command.
NB: You still need to have a GitHub account.
Preparation
(steps 1..3 need to be done just once):
-
hub clone vlang/vsl my_vsl -
cd my_vsl -
hub fork --remote-name pullrequest -
git checkout -b my_cool_feature# Step 4 is better done once per each new feature/bugfix that you make.
Improve VSL by making commits
git commit -am "math: add a new function copysign"
Testing your commits locally
You can test locally whether your changes have not broken something by
running: ./bin/test. See README.md for more details.
Publishing your commits to GitHub
git push pullrequest
Making a PR with hub
(so that your changes can be merged to the main VSL repository)
hub pull-request
Optionally, you can track the status of your PR CI tests with:
hub ci-status --verbose
Fixing failing tests
If everything is OK, after some minutes, the CI tests should pass for
all platforms. If not, visit the URLs for the failing CI jobs, see
which tests have failed and then fix them by making more changes. Just use
git push pullrequest to publish your changes. The CI tests will
run with your updated code. Use hub ci-status --verbose to monitor
their status.
CUDA Backend Development
VSL supports GPU acceleration via NVIDIA CUDA (cuBLAS + cuDNN). See
cuda/README.md for the full architecture and status.
Setup (Arch Linux)
# NVIDIA driver
sudo pacman -S nvidia nvidia-utils
# CUDA Toolkit (includes cuBLAS)
sudo pacman -S cuda
# cuDNN (must match CUDA version)
sudo pacman -S cudnn
# Verify
nvcc --version
nvidia-smi
Building with CUDA support
# Compile with CUDA backend (activates $if cuda ? blocks)
v -d cuda run your_app.v
# Run CUDA tests
./bin/test --use-cuda
# Lint with CUDA
v -d cuda vet .
Phase status
CUDA now has active cuBLAS/cuDNN kernels for the core ML path (GEMM/GEMV, activations, softmax, Conv2D, and optional LayerNorm). New CUDA contributions should keep CPU fallback behavior and add scoped numerical validation where possible.
See ADR-001 for the architecture decision record.