Chez Scheme is a work in progress, and we invite contributions from anyone who is interested in putting forth the necessary effort. One or more of the committers will review pull requests for compatibility with the principles and guidance given below. Details on pull-request processing are given in the governance document.
Our core principles are pretty simple: we try to make Chez Scheme reliable and efficient.
Reliability means behaving as designed and documented. We have no "known bugs" list, preferring instead careful design, coding, and testing practices that prevent most bugs and fixing bugs that do occur as they are discovered. A Scheme program run using Chez Scheme can crash due to bugs in the program, but it should not crash due to bugs in the implementation.
Efficiency means performing at a high level, consuming minimal cpu time and memory. Performance should be a continuous function, with no cliffs or surprises, and should scale well as program or problem size grow. Performance should be balanced across features, not good in one area and bad in another. Compile time is important as well as run time, so compiler optimizations are generally expected to pay their own way, i.e., indirectly benefit compiler performance enough to cover the direct cost of the optimization.
We attempt to achieve the core principles through careful control over growth, testing, and documentation.
Like the Scheme language itself, a good implementation is not built by piling up features but by providing enabling building blocks. So when asked to add a new feature, we first look for a way to achieve the same effect with existing functionality or smaller extensions that are more generally applicable.
Chez Scheme is tested in two ways: implicitly by bootstrapping itself and explicitly via a suite of tests. The suite of tests is about as large as the code base for the implementation, and it is often the case that more lines of test code than implementation code are added to support a new feature. We also benchmark the system whenever we make a change that might materially affect performance.
This project also includes documentation for Chez Scheme in the form of a manual page, a user's guide, and release notes, and we try to set high standards for this documentation. A feature isn't fully implemented until it has been documented. Writing documentation often exposes unnecessary complexity in the design and bugs in the implementation, particularly in corner cases.
Consistent with these principles, we naturally want Chez Scheme to evolve in various useful ways to, among other things:
- increase utility
- improve user friendliness
- support new standards
- run on new platforms
Backward compatibility should be maintained whenever feasible but must sometimes take a back seat to progress in a system whose lifetime is measured in decades.
Please keep in mind the following guidance when preparing contributions:
-
Appropriate tests and documentation changes should be included with all code changes.
-
Coding structure (including indentation) should be consistent with the existing code base. This implies that contributors should study the existing code before contributing.
-
Spend the time required to make the code as clean, clear, and efficient as possible. All other things equal, shorter code is preferable to longer code. Although some people believe more klocs equals more value, code quality is in fact inversely proportional to code size.
-
All changes must be described in LOG as well as via git commit messages and/or github pull request logs. The revision-control system might change over time, but the LOG should always be present.
-
Some contributions may be more appropriately published as projects of their own. If you are contributing a significant extension built using Chez Scheme, consider whether your contribution is such an independent project. An example of such a project is the Nanopass Framework which is both used by Chez Scheme and was initially written using Chez Scheme, but evolves separately.
Before investing significant effort preparing a contribution, consider running the idea by one of the committers for additional guidance and advice.