Hello,
Thanks for your concerns. What specific issues are you seeing when you say “code suddenly stops working after no changes or updating by me” ?
Code repository code is editable and can be run in preview. This runs only for you and is subject to the preview settings (sampling, etc.) as well as the transform you preview.
Once you commit your code on master, “checks” are running which validate the changes (e.g. new output datasets being produced, new inputs being read from, no circular dependency, etc.) and then publish those changes if successful.
Example of checks passing and failing on a code repository. Only the latest version of the code will be used for the builds, when triggered.
Once published, the code in the code repository become irrelevant. You can create a schedule on your dataset and trigger build from the dataset preview UI. The version of the code used is the latest published by successful Checks of the code repository “owning” this dataset.
In the rawest format, the versions of the code used are visible in the jobspecs of the transaction produced on the output datasets, leading to a fully auditable pipeline: you know which inputs, which transactions ranges were consumed, which code version was used for this.
Code repository upgrade Pull-requests (Upgrade PRs) can be automatically opened on your master branch but requires your manual approval before merging. Those can include dependencies upgrades, which can break the logic in your code if it was depending on a no-longer-available syntax (rare, but possible). Hence why they require your approval before merging.
Example upgrade PR
If no upgrade PR is available on a code repository, you can trigger one to open by clicking “upgrade repo” top right, in code repository.