Skip to content

Latest commit

 

History

History
355 lines (273 loc) · 14.9 KB

DOCUMENTATION_GUIDELINES.md

File metadata and controls

355 lines (273 loc) · 14.9 KB

Substrate Documentation Guidelines

This document is focused on documenting parts of Substrate that relate to its external API. The list of such crates can be found in CODEOWNERS. Search for the crates auto-assigned to the docs-audit team.

These crates are used by external developers and need thorough documentation. They are the most concerned with FRAME development.


General/Non-Pallet Crates

First, consider the case for all such crates, except for those that are pallets.

What to Document?

The first question is, what should you document? Use this filter:

  1. In the crates assigned to docs-audit in CODEOWNERS,

  2. All pub items need to be documented. If not pub, it doesn't appear in the rust-docs, and is not public facing.

    • Within pub items, sometimes they are only pub to be used by another internal crate, and you can foresee that this won't be used by anyone else. These need not be documented thoroughly.

    • Reminder: trait items are public by definition if the trait is public.

  3. All public modules (mod) should have reasonable module-level documentation (//!).

Rust Docs vs. Code Comments

Note that anything starting with /// is an external rust-doc, and everything starting with // does not appear in the rust-docs. It's important to not confuse the two in your documentation.

/// Computes the square root of the input, returning `Ok(_)` if successful.
///
/// # Errors
/// ...
///
// Details about the complexity, how you implemented this, and some quirks that
// are NOT relevant to the external interface, so it starts with '//'.
// This can also be moved inside the function.
pub fn sqrt(x: u32) -> Result<u32, ()> {
    todo!();
}

How to Document?

There are good sources to look into:

As mentioned here and here, always start with a single sentence demonstrating what is documented. All additional documentation should be added after a newline. Strive to make the first sentence succinct and short.The reason for this is the first paragraph of docs about an item (everything before the first newline) is used as the excerpt that rust doc displays about this item when it appears in tables, such as the table listing all functions in a module. If this excerpt is too long, the module docs will be very difficult to read.

About special sections, we will most likely not need to think about panic and safety in any runtime related code. Our code is never unsafe, and will (almost) never panic.

Use `# Examples as much as possible. These are great ways to further demonstrate what your APIs are doing, and add free test coverage. As an additional benefit, any code in rust-docs is treated as an "integration tests", not unit tests, which tests your crate in a different way than unit tests. So, it is both a win for "more documentation" and a win for "more test coverage".

You can also consider having an # Error section optionally. Of course, this only applies if there is a Result being returned, and if the Error variants are overly complicated.

Strive to include correct links to other items in your written docs as much as possible. In other words, avoid `some_func` and instead use [`some_fund`]. Read more about how to correctly use links in your rust-docs here and here. Strive to include correct links to other items in your written docs as much as possible. In other words, avoid `some_func` and instead use [`some_func`].

While you are linking, you might become conscious of the fact that you are in need of linking to (too many) foreign items in order to explain your API. This is leaning more towards API-Design rather than documentation, but it is a warning that the subject API might be slightly wrong. For example, most "glue" traits1 in frame/support should be designed and documented without making hard assumptions about particular pallets that implement them.


TLDR

  1. Have the goal of enforcing #![deny(missing_docs)] mentally, even if it is not enforced by the compiler 🙈.
  2. Start with a single, clear and concise sentence. Follow up with more context, after a newline, if needed.
  3. Use examples as much as reasonably possible.
  4. Use links as much as possible.
  5. Think about context. If you are explaining a lot of foreign topics while documenting a trait that should not explicitly depend on them, you have likely not designed it properly.

Proc-Macros

Note that there are special considerations when documenting proc macros. Doc links will appear to function within your proc macro crate, but often will no longer function when these proc macros are re-exported elsewhere in your project. The exception is doc links to other proc macros which will function just fine if they are also being re-exported. It is also often necessary to disambiguate between a proc macro and a function of the same name, which can be done using the macro@my_macro_name syntax in your link. Read more about how to correctly use links in your rust-docs here and here.


Other Guidelines

The above five guidelines must always be reasonably respected in the documentation.

The following are a set of notes that may not necessarily hold in all circumstances:


Document Through Code

You should make sure that your code is properly-named and well-organized so that your code functions as a form of documentation. However, within the complexity of our projects in Polkadot/Substrate that is not enough. Particularly, things like examples, errors and panics cannot be documented only through properly- named and well-organized code.

Our north star is self-documenting code that also happens to be well-documented and littered with examples.

  • Your written documents should complement the code, not repeat it. As an example, a documentation on top of a code example should never look like the following:
 /// Sends request and handles the response.
 trait SendRequestAndHandleResponse {

 }

In the above example, the documentation has added no useful information not already contained within the properly-named trait and is redundant.


Formatting Matters

The way you format your documents (newlines, heading and so on) makes a difference. Consider the below examples:

/// This function works with input u32 x and multiplies it by two. If
/// we optimize the other variant of it, we would be able to achieve more
/// efficiency but I have to think about it. Probably can panic if the input
/// overflows u32.
fn multiply_by_2(x: u32) -> u32 { .. }
/// Multiplies an input of type [`u32`] by two.
///
/// # Panics
///
/// Panics if the input overflows.
///
/// # Complexity
///
/// Is implemented using some algorithm that yields complexity of O(1).
// More efficiency can be achieved if we improve this via such and such.
fn multiply_by_2(x: u32) -> u32 { .. }

They are both roughly conveying the same set of facts, but one is easier to follow because it was formatted cleanly. Especially for traits and types that you can foresee will be seen and used a lot, try and write a well formatted version.

Similarly, make sure your comments are wrapped at 100 characters line-width (as defined by our rustfmt.toml), no more and no less! The more is fixed by rustfmt and our CI, but if you (for some unknown reason) wrap your lines at 59 characters, it will pass the CI, and it will not look good 🫣. Consider using a plugin like rewrap (for Visual Studio Code) to properly do this.


Pallet Crates

The guidelines so far have been general in nature, and are applicable to crates that are pallets and crates that're not pallets.

The following is relevant to how to document parts of a crate that is a pallet. See pallet-fast-unstake as one example of adhering these guidelines.


Top Level Pallet Docs (lib.rs)

For the top-level pallet docs, consider the following template:

//! # <Pallet Name>
//!
//! <single-liner about the pallet>.
//!
//! ## Pallet API
//!
//! <Reminder: inside the [`pallet`] module, a template that leads the reader to the relevant items is auto-generated. There is no need to repeat
//! things like "See Config trait for ...", which are generated inside [`pallet`] here anyways. You can use the line below as-is:>
//!
//! See the [`pallet`] module for more information about the interfaces this pallet exposes, including its
//! configuration trait, dispatchables, storage items, events and errors.
//!
//! ## Overview
//!
//! <should be high-level details that are relevant to the most broad audience>
//!
//! <The audience here is potentially non-coders who just want to know what this pallet does, not how it does it>
//!
//! <potentially a few paragraphs, focus on what external folks should know about the pallet>
//!
//! ### Example
//!
//! <Your pallet must have a few tests that cover important user journeys. Use https://crates.io/crates/docify to
//! reuse these as examples.>
//!
//! <The audience of this is those who want to know how this pallet works, to the extent of being able to build
//! something on top of it, like a DApp or another pallet. In some cases, you might want to add an example of how to
//! use this pallet in other pallets.>
//!
//! This section can most often be left as-is.
//!
//! ## Low Level / Implementation Details
//!
//! <The format of this section is up to you, but we suggest the Design-oriented approach that follows>
//!
//! <The audience of this would be your future self, or anyone who wants to gain a deep understanding of how the pallet
//! works so that they can eventually propose optimizations to it>
//!
//! ### Design Goals (optional)
//!
//! <Describe your goals with the pallet design.>
//!
//! ### Design (optional)
//!
//! <Describe how you've reached those goals. This should describe the storage layout of your pallet and what was your
//! approach in designing it that way.>
//!
//! ### Terminology (optional)
//!
//! <Optionally, explain any non-obvious terminology here. You can link to it if you want to use the terminology further
//! up>

This template's details (heading 3s and beyond) are left flexible, and at the discretion of the developer to make the best final choice about. For example, you might want to include ### Terminology or not. Moreover, you might find it more useful to include it in ## Overview.

Nonetheless, the high level flow of going from the most high level explanation to the most low level explanation is important to follow.

As a rule of thumb, the Heading 2s (##) in this template can be considered a strict rule, while the Heading 3s (###) and beyond are flexible.


Polkadot and Substrate

Optionally, in order to demonstrate the relation between the two, you can start the pallet documentation with:

//! > Made with *Substrate*, for *Polkadot*.
//!
//! [![github]](https://github.com/paritytech/polkadot-sdk/tree/master/substrate/frame/fast-unstake) -
//! [![polkadot]](https://polkadot.network)
//!
//! [polkadot]: https://img.shields.io/badge/polkadot-E6007A?style=for-the-badge&logo=polkadot&logoColor=white
//! [github]: https://img.shields.io/badge/github-8da0cb?style=for-the-badge&labelColor=555555&logo=github

Dispatchables

For each dispatchable (fn item inside #[pallet::call]), consider the following template:

/// <One-liner explaining what the dispatchable does>
///
/// ## Dispatch Origin
///
/// The dispatch origin of this call must be <details (e.g. Root, Signed, Unsigned)>
///
/// ## Details
///
/// <All other details, namely any errors that could occur within this dispatch and the events this dispatch could emit>
///
/// ## Errors (optional)
///
/// <If an extensive list of errors can be returned, list them individually instead of mentioning them in the section
/// above>
///
/// ## Events (optional)
///
/// <Events are akin to the "return type" of dispatchables, optionally mention them>
pub fn name_of_dispatchable(origin: OriginFor<T>, ...) -> DispatchResult {}

Consider the fact that these docs will be part of the metadata of the associated dispatchable, and might be used by wallets and explorers.


Storage Items

  1. If a map-like type is being used, always note the choice of your hashers as private code docs (// Hasher X chosen because ...). Recall that this is not relevant information to external people, so it must be documented as //.

  2. Consider explaining the crypto-economics of how a deposit is being taken in return of the storage being used.

  3. Consider explaining why it is safe for the storage item to be unbounded, if #[pallet::unbounded] or #[pallet::without_storage_info] is being used.


Errors and Events

Consider the fact that, similar to dispatchables, these docs will be part of the metadata of the associated event/error, and might be used by wallets and explorers.

Specifically for error, explain why the error has happened, and what can be done in order to avoid it.

Documenting Changes/PR

See PRDoc.

Footnotes

  1. Those that help two pallets talk to each other.