DEP-0001: The Dat Enhancement Proposal Process
Title: DEP-0001: The Dat Enhancement Proposal Process
Status: Draft (as of 2018-01-30)
Github PR: Draft
The Dat Enhancement Proposal ("DEP") process is how the Dat open source community comes to (distributed) consensus around technical protocol enhancements and organizational process.
The Dat protocol is still a living standard. A transparent process is needed for community members to understand what changes are in the pipeline and how new ideas might come to fruition.
The core protocol is being used and extended by several projects with differing priorities and use cases. At the same time, lead developer time is very scarce. There is a need to parallelize design and implementation work between projects, which requires better coordination (process) and communication of technical details (standards). There is also an increasing need to be legible to and accessible to parties outside the existing Dat ecosystem.
With growing use, the logistics of rolling out protocol changes and backwards-incompatible changes becomes more difficult, but at the same time more important to coordinate smoothly. Planning requires clear communication of change ahead of time.
A public DEP process is expected to improve coordination and planning by setting clear expectations for documentation of protocol changes and extensions. The technical quality of the protocol itself should be improved by increasing the number of people who can view and understand proposals at each step of the process. The barrier to entry for independent implementations should be lower, allowing new technical and user communities to adopt the protocol. The overall development and decision making process should be more transparent, accessible, and scalable to a growing group of application developers and end users
How To Submit a Proposal
As a first step, before drafting a DEP or implementing experimental new protocol features, it's helpful to informally pitch your idea to see if others in the community are already thinking something similar, or have discussed the same idea in the past. This discussion could happen over chat, Github issues, blog posts, or other channels. If you can recruit collaborators and work out some of the details, all the better. This period could be called pre-DEP.
Once your idea has been flushed out, the process for proposing and debating a new DEP is:
- Use git to fork the datprotocol/deps repository
proposals/0000-my-proposal.md(don't choose the "next" number, use zero;
my-proposalshould be a stub identifier for the proposal.)
- Fill in the DEP template. All proposals should have a Type and Status (see below for details). Feel free to tweak or expand the structure (headers, content) of the document to fit your needs, but your proposal should be "complete" before submission.
- You can create an informal WIP (work in progress) PR (pull-request) whenever you like for early feedback and discussion, but there is no expectation that your proposal will be given detailed review until it is complete.
- When you are ready, submit your complete proposal for review (this could be opening a PR or removing WIP status from an existing one), with the intent of being accepted with "Draft" status. An editor (somebody who is an owner of the DEPs repository) will look over your proposal for completeness; if acceptable, they will assign one or more reviewers. At this stage, two outcomes are the most likely: your proposal is merged with "Draft" status, or declined. This first stage of the review is expected to take 3 weeks at most from when reviewers were assigned. It is appropriate to propose specific community members to review your proposal. The submitter can withdraw a proposal at any time. If accepted, a DEP number will be assigned and the PR merged. If declined, reviewers may give feedback and/or invite proposers to "significantly revise and resubmit".
- While in draft status, proposals can be experimented with. Small corrections and clarifications can be submitted by PR expect to be merged quickly if they are reasonable and don't change the broad behavior or semantics of the proposal; larger changes should be re-submitted as Superseding proposals.
- When it seems appropriate, a PR can be submitted to upgrade the status of a "Draft" to "Active". At this time a final review will take place, with the outcome being that a proposal stays "Draft" or becomes "Active". This review period is shorter (2 weeks at most), as everybody is expected to be more familiar with the proposal at this point. Small changes to the DEP can be included, but it's expected that this is in broad strokes the same proposal that was reviewed earlier (if not, a new "Draft" should be proposed that Supersedes).
- Small tweaks (grammar, clarifications) to a merged DEP can take place as regular Github PRs; revisiting or significantly revising should take place as a new DEP. Draft, Process, and Informational DEPs have a lower bar for evolution over time via direct PR.
Types and Statuses
DEPs should have a type:
- Standard for technical changes to the protocol, on-disk formats, or public APIs. These are intended to be prescriptive, and to clearly delineate which features and behaviors are mandatory or optional.
- Process for formalizing community processes or other (technical or non-technical) decisions. For example, a security vulnerability reporting policy, a process for handling conflicts of interest, or procedures for mentoring new developers.
- Informative for describing conventions, design patterns, existing norms, special considerations, etc.
The status of a DEP can be:
- Pre-Merge: a well-formed DEP has been written and a PR opened. The "Status" line can list Draft when in this state.
- Draft: PR has been merged and a number assigned, but additional time is needed for deeper discussion or more implementation before being "normative" and expected for implementation. It is acceptable to have "competing" proposals in this state at the same time.
- Active: adopted or intended for implementation in mainline libraries and clients as appropriate. Again, DEPs should clarify which aspects of themselves are optional or required for well-behaved clients.
- Closed: either consensus was against, a decision was postponed, or the authors withdrew their proposal. This could apply to any of: a proposal PR that was never merged, a merged Draft (which was never Active), or an Active DEP which there is now consensus against without a specific new DEP to replace it.
- Superseded: a formerly Active DEP has been made obsolete by a new Active DEP; the new DEP should specify specific old DEPs that it would supersede.
Content and Structure
A changelog should be kept in the DEP itself giving the date of any changes of status.
A template file is provided, but sections can be added or removed as appropriate for a specific DEP.
The DEP text itself should be permissively licensed; the convention is to use the Creative Commons Attribution License (CC-BY), with attribution to the major contributing authors listed.
For appropriate DEPs (including all Standards DEPs), authors should explicitly consider and note impacts on:
- Privacy and User Rights. Consider reading IETF [RFC 6973] ("Privacy Considerations for Internet Protocols") and [RFC 8280] ("Research into Human Rights Protocol Considerations".)
- Backwards compatibility of on-disk archives and older network clients. If a backwards-incompatible change is proposed, a migration plan should be sketched out in the proposal.
The criteria for a proposal being accepted as a Draft are, at a minimum, that the proposal is complete, understandable, unambiguous, and relevant. There is a good faith assumption that the submitter believes that the proposal could actually be adopted and put to beneficial use. An editor may screen proposals before passing on to the group for review.
For Standards and Process DEPs, Draft proposals should be specific enough that they can be prototyped and experimented with (e.g. in a pilot program or test network), but it isn't expected that all details have been worked out. Working code is helpful but not required.
For a Draft to migrate to Active, there is an expectation that the proposal has been demonstrated (e.g. in working code, though not necessarily in "official" libraries yet), that the proposal will be the "new normal" and expected behavior going forward, and that the chance of unforeseen issues arising during complete adoption is low.
Decision Making Process
There exists a Protocol Working Group (WG) (WG) which makes DEP status decisions; see the Github repository for a list of current members and the governance process.
By no means should working group members be the only people reviewing or giving feedback on proposals.
When deciding on Draft status, at least one WG member must review the entire proposal in detail, give feedback, and give informed approval. If no detailed review takes place in the fixed time window, the default is to close (reject) until a member is willing to commit to review. Any WG member can request revisions or clarifications (blocking acceptance until addressed), and any member can block. In other words, consensus requires that at least one member actively approve, and any member can block, but it isn't required to have every member review and actively given an opinion. This is referred to as the "loose consensus" model.
For Active status, the expectation is that all working group members will review the proposal and actively participate in consensus. In the event that not all members can participate, the default is again negative. Any member can block. This is referred to as the "complete consensus" model.
For all other status changes, at least one WG member must vouch for or approve the change ("loose consensus"). If there is unambiguous consensus (or, eg, a DEP is documenting already adopted practice), a DEP can move directly to Active status (following the "complete consensus" process).
In all cases, if there is a deadlock (a block can not be overcome after further discussion), there is an option to override the block by a vote. The details of this process are left to description in the Working Group repository, and are expected to be used only in exceptional cases (eg, are not invoked by default if a member blocks).
Proposals are expected to be open for at least three days (72 hours) for comment (and longer to accommodate special circumstances, like holidays). Vetoes (blocks) can happen up to a week after initially being submitted for review, which might be retroactive if the proposal was accepted (by other WG members) very quickly.
In the design space of RFC processes, there are two decision points determining the formality and maturity of accepted standards. The first is, does draft status mean could be implemented, or has been implemented? We chose "could". Secondly, for active (or "final") status, is the proposal expected to be dominant in the wild, or is it already dominant in the wild? We chose "expected". In both cases we are emphasizing clarification and stabilization of new ideas, as opposed to enforcing interoperability of competing formulations of the same idea. In short, we expect DEPs to lead (rather than tail) implementation.
Time limits and default outcomes are used to prevent proposals could get stuck in an ambiguous indefinite state anywhere along the process. "Draft" status is considered a stable state to linger in.
Setting expectations for "completeness" of proposals, having an editor quickly skim proposals before jumping in to a full review, and acknowledging an explicit "revise and re-submit" workflow are all attempts to head off the situation of partial proposals being submitted and then significantly revised, which places extra (time) burden on reviewers.
There are already multiple sources of technical documentation: the Dat
protocol website, the Dat whitepaper, Dat
website documentation section, the discussion repo
issues, and the datprotocol github group (containing, eg, the
dat.json repo/spec). Without consensus and consolidation, this would be "yet
another" place to look.
Background and References
The following standards processes were referenced and considered while designing the DEP process:
- BitTorrent Enhancement Process as described in BEP 1. The Bittorrent protocol has a lot of technical similarities to Dat, and as a single protocol family (not a language or full-stack system) is one of the most similar in scope. The de facto BEP model is that Drafts are very stable and widely adopted; only the most universal core components are Final. The DEP process bases it's type categories on the BEP process. There is a single editor and an explicit BDFL in the BEP process.
- The Rust Language RFC Process is relatively new, but has had a huge volume of proposals, rivaling even the IETF. The process is relatively lightweight and happens entirely on Github; it is the most similar to the DEP process proposed here, in terms of Draft/Active distinction. Rust has strong organizational backing with defined leadership roles; proposals are reviewed by specific sub-teams.
- IETF RFC Process: perhaps the oldest and best known RFC process, under the motto of "rough consensus and working code". The process is very bespoke (involving custom file formats and software) and heavy on process (with working groups and in-person meetings).
- XMPP Standards Process: has the interesting sub-pattern of regularly updated (annual) standards. XMPP is also a protocol, like Bittorrent. The protocol was designed for easy extension, and at various points has seen adoption, extension, and pressure from powerful entities.
- Python Enhancement Process documented in PEP 1. PEPs are relatively broad in scope (they often speak to process and organizational dynamics), and are widely cited directly by name. Proposals are usually debated in great detail on mailing lists before being proposed. Python has a BDFL (benevolent dictator for life) who has final say over proposals, though he sometimes delegates to deputies.
- The W3C is a paid membership organization which, like the IETF, is made up of entities large and small, for-profit and altruistic, with decent regional diversity. W3C standards are often rather large and verbose documents, and tend to tail (rather than lead) implementation.
How mutable should "Draft" Standards DEPs be over time? What about Informational DEPs? Should there be an additional status ("Living"?) for DEPs that are expected to evolve, or is this against the whole philosophy of having specific stable documents to reference? This is expected to be clarified while this DEP itself is in Draft status.
Does "Active" status mean that implementation is mandatory, and that features must be implemented unless they are explicitly optional? How would this expectation be enforced for third-party software? This is expected to be clarified when concrete examples arise.
- 2018-01-24: DEP process and governance model is discussed by Working Group. First full draft of DEP-0001 submitted for review.