features branches are not necessary now that Forgejo is a hard fork
2.7 KiB
title | license |
---|---|
Development workflow | CC-BY-SA-4.0 |
All Forgejo commits are merged into the forgejo
branch from which binary releases and packages are created and distributed. The development workflow is a set of conventions Forgejo developers are expected to follow to work together.
Naming conventions
- development:
forgejo
- stable releases:
vX.Y/forgejo
Cherry-picking
Gitea branches are regularly scrubbed, and changes useful to Forgejo are cherry picked.
Testing
Software is always tested with different methods which are more or less costly. Forgejo has a lot of technical debt and large portions of the codebase are very difficult to test. A reasonable effort should be made to improve the situation whenever a change is proposed.
Automated tests
They do not require manual intervention and can be run as often as necessary from the CI.
Ideally all aspects of a code can be verified with unit tests (e.g. make unit-test-coverage
). Integration tests may involve third party components such as a database (e.g. make test-sqlite
).
Tests that require launching a Forgejo instance are found in the end-to-end repository.
It is possible to run them on a pull request by setting the run-end-to-end-tests
label. It will trigger a workflow that:
- builds a Forgejo binary including the pull request and uploads it as an artifact of the PR
- creates a pull request in the end-to-end repository against the forgejo-pr branch
- run a workflow using the Forgejo binary found in the PR artifact
Manual tests
When the test infrastructure is lacking or the test to verify a change (such as a oneliner) is out of proportion with the effort to create and automated test (as a few days of work may be needed if this is in the web UI), the instructions to manually verify the change is good must be documented in the forgejo-manual-testing repository.
The person who wrote the instructions is expected to run them on a regular basis to ensure there are no regressions. If this is not done, the regression will always happen over time and will be discovered by end users which by far the most costly and frustrating method of testing.