GraphQL is still an evolving language. This repository contains the specification text as well as Pull Requests with suggested improvements and contributions.
Contributions which do not change the interpretation of the spec but instead improve legibility, fix editorial errors, clear up ambiguity and improve examples are encouraged and are often merged by a spec editor with little process.
However, contributions which do meaningfully change the interpretation of the spec must follow an RFC (Request For Comments) process led by a champion through a series of stages intended to improve visibility, allow for discussion to reach the best solution, and arrive at consensus. This process becomes ever more important as GraphQL's community broadens.
When proposing or weighing-in on any issue or pull request, consider the Code of Conduct to better understand expected and unacceptable behavior.
A common point of confusion for those who wish to contribute to GraphQL is where to start. In fact, you may have found yourself here after attempting to make an improvement to a GraphQL library. Should a new addition be made to the GraphQL spec first or a GraphQL library first? Admittedly, this can become a bit of a chicken-or-egg dilemma.
GraphQL libraries seek to be "spec compliant", which means they discourage changes that cause them to behave differently from the spec as written. However, they also encourage pull requests for changes that accompany an RFC proposal or RFC draft. In fact, a spec contribution RFC won't be accepted until it has experience being implemented in a GraphQL library.
To allow a library to remain spec compliant while also implementing proposals and drafts, it may request that these features are built so they are disabled by default with opt-in option flags or it may simply wait to merge a well-tested pull request until the spec proposal is accepted.
GraphQL's evolution is guided by a few principles. Suggested contributions should use these principles to guide the details of an RFC and decisions to move forward. See editor Lee Byron talk about guiding principles at GraphQL Europe 2017.
-
Backwards compatibility
Once a query is written, it should always mean the same thing and return the same shaped result. Future changes should not change the meaning of existing schema or queries or in any other way cause an existing compliant GraphQL service to become non-compliant for prior versions of the spec.
-
Performance is a feature
GraphQL typically avoids syntax or behaviors which could place burden on runtime efficiency or make demands of a GraphQL service it cannot efficiently fulfill.
-
Favor no change
As GraphQL is implemented in over a dozen languages under the collaboration of hundreds of individuals, incorporating any change has a high cost. Accordingly, proposed changes must meet a very high bar of added value. The burden of proof is on the contributor to illustrate this value.
-
Enable new capabilities motivated by real use cases
Every change should intend on unlocking a real and reasonable use case. Real examples are always more interesting than theoretical ones, and common scenarios are more interesting than rare ones. RFCs should do more than offer a different way to reach an already achievable outcome.
-
Simplicity and consistency over expressiveness and terseness
There are plenty of behaviors and patterns found in other languages intentionally absent from GraphQL. "Possible but awkward" is often favored over more complex alternatives. Simplicity (e.g. fewer concepts) is more important than expressing more sophisticated ideas or writing less.
-
Preserve option value
It's hard to know what the future brings, so whenever possible decisions should be made which allow for more options in the future. Sometimes this is unintuitive: spec rules often begin more strict than necessary with a future option to loosen when motivated by a real use case.
-
Understandability is just as important as correctness
The GraphQL spec, despite describing technical behavior, is intended to be read by people. Use natural tone and include motivation and examples.
Contributing to GraphQL requires a lot of dedicated work. To set clear expectations and provide accountability, each proposed RFC (request for comments) must have a champion who is responsible for addressing feedback and completing next steps. An RFC may have multiple champions. The spec editors are not responsible for completing RFCs which lack a champion (though an editor may be a champion for an RFC).
An RFC which does not have a champion may not progress through stages, and can become stale. Stale proposals may be picked up by a new champion or may be rejected.
RFCs are guided by a champion through a series of stages: strawman, proposal, draft, and accepted (or rejected), each of which has suggested entrance criteria and next steps detailed below. RFCs typically advance one stage at a time, however may advance multiple stages at a time. Stage advancements typically occur during Working Group meetings, however may occur on GitHub.
All RFCs start as either a strawman or proposal.
A RFC at the strawman stage captures a described problem or partially-considered solutions. A strawman does not need to meet any entrance criteria. A strawman's goal is to prove or disprove a problem and guide discussion towards either rejection or a preferred solution. A strawman may be an issue or a pull request (though an illustrative pull request is preferrable).
There is no entrance criteria for a Strawman
As implied by the name strawman, the goal at this stage is to knock it down (reject) by considering other possible related solutions, showing that the motivating problem can be solved with no change to the specification, or that it is not aligned with the guiding principles.
Once determined that the strawman is compelling, it should seek the entrance criteria for proposal.
An RFC at the proposal stage is a solution to a problem with enough fidelity to be discussed in detail. It must be backed by a willing champion. A proposal's goal is to make a compelling case for acceptance by describing both the problem and the solution via examples and spec edits. A proposal should be a pull request.
Entrance criteria:
- Identified champion
- Clear explanation of problem and solution
- Illustrative examples
- Incomplete spec edits
- Identification of potential concerns, challenges, and drawbacks
A proposal is subject to the same discussion as a strawman: ensuring that it is well aligned with the guiding principles, is a problem worth solving, and is the preferred solution to that problem. A champion is not expected to have confidence in every detail at this stage and should instead focus on identifying and resolving issues and edge-cases. To better understand the technical ramifications of the proposal, a champion is encouraged to implement it in a GraphQL library.
Most proposals are expected to evolve or change and may be rejected. Therefore, it is unwise to rely on a proposal in a production GraphQL service. GraphQL libraries may implement proposals, though are encouraged to not enable the proposed feature without explicit opt-in.
An RFC at the draft stage is a fully formed solution. There is working group consensus that the problem identified should be solved, and this particular solution is preferred. A draft's goal is to precisely and completely describe the solution and resolve any concerns through library implementations. A draft must be a pull request.
Entrance criteria:
- Consensus the solution is preferred (typically via Working Group)
- Resolution of identified concerns and challenges
- Precisely described with spec edits
- Compliant implementation in GraphQL.js (might not be merged)
A proposal becomes a draft when the set of problems or drawbacks have been fully considered and accepted or resolved, and the solution is deemed desirable. A draft's goal is to complete final spec edits that are ready to be merged and implement the draft in GraphQL libraries along with tests to gain confidence that the spec text is sufficient.
Drafts may continue to evolve and change, occasionally dramatically, and are not guaranteed to be accepted. Therefore, it is unwise to rely on a draft in a production GraphQL Service. GraphQL libraries should implement drafts to provide valuable feedback, though are encouraged not to enable the draft feature without explicit opt-in when possible.
An RFC at the accepted stage is a completed solution. According to a spec editor it is ready to be merged as-is into the spec document. The RFC is ready to be deployed in GraphQL libraries. An accepted RFC must be implemented in GraphQL.js.
Entrance criteria:
- Consensus the solution is complete (via editor or working group)
- Complete spec edits, including examples and prose
- Compliant implementation in GraphQL.js (fully tested and merged or ready to merge)
A draft is accepted when it has learned via implementation and tests that it appropriately handles all edge cases, that the spec edits do not only precisely describe the new syntax and semantics but include motivating prose, examples, and include edits to any other affected areas of the spec. Once accepted, a champion should encourage adoption of the RFC by opening issues or pull requests on other popular GraphQL libraries.
An accepted RFC is merged into the GraphQL spec's master branch by an editor and will be included in the next released revision.
An RFC may be rejected at any point and for any reason. Most often because a strawman was proven to be unnecessary, was not aligned with the guiding principles, or failed to meet the entrance criteria to become a proposal. A proposal may become rejected for similar reasons as well as if it fails to reach consensus or loses the confidence of its champion. Likewise a draft may encounter unforeseen issues during implementations which cause it to lose consensus or the confidence of its champion.
RFCs which have lost a champion will not be rejected immediately, but may become rejected if they fail to attract a new champion.
Once rejected, an RFC will typically not be reconsidered. Reconsideration is possible if a champion believes the original reason for rejection no longer applies due to new circumstances or new evidence.