Thanks for your interest in contributing to Persistent! This file has some tips for developing Persistent and getting a pull request accepted.
Persistent is a mega-repo that contains many Haskell packages, each in a different directory. All the subprojects can be developed with Stack, using stack <command> <subproject>
, e.g.
stack build persistent-mysql
stack haddock persistent
stack test persistent-postgresql
If you'd like to test your changes in a full-fledged app, you can use Stack to build against it, e.g.:
packages:
- '/path/to/this/repo/persistent-postgresql'
Additional information can be found in development.md.
Avoid partial functions in pure code. Even if you know the partial function is safe in your instance, partial functions require more reasoning from the programmer and are not resilient to refactoring.
Keep coding style consistent with the rest of the file, but don't worry about style too much otherwise. PRs changing code style are viewed skeptically.
Avoid adding unnecessary dependencies. If a dependency provides only a minor convenience for your implementation, it's probably better to skip it.
If you do add a new dependency, try to support a wide range of versions of it.
Backwards incompatible changes are viewed skeptically—best to ask in an issue to see if a particular backwards incompatible change would be approved. If possible keep backwards compatibility by adding new APIs and deprecating old ones.
Keep backwards compatibility with old versions of dependencies when possible.
As much as possible, keep separate changes in separate PRs.
Tests are strongly recommended. If your patch does not include tests, then it must be trivially correct or have type-level guarantees.
If you're reporting an issue, contributing a failing test is a great way to kickstart development on it.
Each backend defines it's own test suite. The persistent-test
package defines a common set of helpers and scenarios that backends can implement.
All public APIs must be documented. Documenting private functions is optional, but may be nice depending on their complexity. Example documentation:
-- | Retrieve a list of a certain 'Entity' from the database
--
-- ==== __Examples__
--
-- @
-- selectUsers :: 'MonadIO' m => 'ReaderT' 'SqlBackend' m ['Entity' User]
-- selectUsers = 'selectList' [UserAge <-. [40, 41]] []
-- @
--
-- @since 1.5.4
selectList :: (MonadIO m, PersistQueryRead backend, PersistRecordBackend record backend)
=> [Filter record] -- ^ Options to filter the query result; see "Query filter combinators" in the Database.Persist documetation
-> [SelectOpt record] -- ^ Options to limit and sort the returned records.
-> ReaderT backend m [Entity record]
selectList = ...
Examples are recommended, but not required, in documentation. Marking new APIs with @since <version number>
is required.
Persistent packages roughly follow the Haskell Package Versioning Policy style of A.B.C.[D] (MAJOR.MAJOR.MINOR.[PATCH])
- A - Used for massive changes in the library. (Example: 1.2.3.4 becomes 2.0.0)
- B - Used for smaller breaking changes, like removing, renaming, or changing behavior of existing public API. (Example: 1.2.3.4 becomes 1.3.0)
- C - Used for new public APIs (Example: 1.2.3.4 becomes 1.2.4)
- D - Used for bug fixes (Example: 1.2.3.4 becomes 1.2.3.5).
- D is optional in the version number, so 2.0.0 is a valid version.
Documentation changes don't require a new version.
If you feel there is ambiguity to a change (e.g. fixing a bug in a function, when people may be relying on the old broken behavior), you can ask in an issue or pull request.
Unlike in the Package Versioning Policy, deprecations are not counted as MAJOR changes.
In some cases, dropping compatibility with a major version of a dependency (e.g. changing from transformers >= 0.3 to transformers >= 0.4), is considered a breaking change.
After you submit a PR, update the subproject's Changelog.md file with the new version number and a link to your PR. If your PR does not need to bump the version number, include the change in an "Unreleased" section at the top.
Releases should be done as soon as possible after a pull request is merged—don't be shy about reminding us to make a release if we forget.