shrub/CONTRIBUTING.md

Contributing to Urbit

Thank you for your interest in contributing to Urbit.

See urbit.org/using/install for basic orientation and usage instructions. You may also want to subscribe to urbit-dev, the Urbit development mailing list. For specific information on contributing to the Urbit interface, see its contribution guidelines.

For information on Arvo's maintainers, see pkg/arvo.

For more extensive information on Urbit development, such as how to set up an environment and how to submit a grant, see the developer documentation.

Fake ships

You may have an identity on the live network, but doing all your development on the live network would be cumbersome and unnecessary. Standard practice in Urbit development is to work on a fake ~zod. Fake ships use deterministic keys (derived from the ship address) and don't talk to the live network. They can talk to each other over the local loopback.

To start a fake ship, simply specify the name with -F:

$ urbit -F zod

You can also pass a name for the pier (or ship directory):

$ urbit -F zod -c my-fake-zod

To resume a fake ship, just pass the name of the pier:

$ urbit my-fake-zod

Fake ships by default use the same pre-compiled kernelspace ('pills') as livenet ships do: boot pills, which are not always current with master. If you wish to develop using code off the master branch, run the following from the repo directory:

git lfs install
git lfs pull
urbit -F zod -B "bin/solid.pill" -A "pkg/arvo"

Git practice

Contributing

The canonical source tree is the master branch of https://github.com/urbit/urbit. You should typically branch off of master when commencing new work. Most pull requests should be merging into one of the next/* branches, depending on what part of the system the pull request is targeting.

Since we use GitHub, we request you contribute via a GitHub pull request. Tag the maintainer for the component. If you have a question for the maintainer, you can direct message them from your Urbit ship using that information.

When contributing changes, via whatever means, make sure you describe them appropriately. You should attach a reasonably high-level summary of what the changes are and what they do; reference any useful background material that may exist, e.g. a GitHub issue, a mailing list discussion, a UP, etc. Here is a good example of a pull request with a useful, concise description.

If your changes replace significant extant functionality, be sure to compare them with the thing you're replacing. You may also want to cc reviewers, or other parties who might have a particular interest in what you're contributing.

Hygiene

Commits should generally be relevant, atomic, and have descriptions formatted in the following manner:

component: short description

long description

The 'component' is a short prefix of what area of the codebase the commit applies to. If a commit patches %gall, for example, the description should be prefixed by 'gall'. If it touches :aqua, it should be prefixed by 'aqua'. If it touches multiple components, then separate these by commas, e.g. "gall, aqua, ph" -- but note that this may be a warning that too many changes are being packed into a single commit. The 'component' and 'short description' combined should be no more than 50 characters.

A lengthier description is encouraged, where useful, but is not always strictly required. You should use the longer description to give any useful background on or motivation for the commit, provide a summary of what it does, link to relevant issues, proposals, or other commits, and so on.

Here is an example of our commit format, taken from a commit in the history:

zuse: remove superfluous 'scup' and 'culm' types.

%zuse includes definitions for 'scup' and 'culm', both of which are superfluous. 'scup' is simply (pair ship desk) and is used only in the definition of 'culm', a tagged union in which three of the four branches are commented out (i.e. are unused).

This commit deletes 'scup' and 'culm' and refactors what little code made use of them.

Note that the short description is prefixed by zuse:, which is what the commit touches. Otherwise it just includes a summary of the change.

Here's another example:

build: give arvo a high priority

0bdced981e introduced the 'arvo-ropsten' derivation. Attempting to install both 'arvo' and 'arvo-ropsten' via nix-env will result in a priority error; this assigns a higher priority to 'arvo' to resolve the conflict.

Fixes #1912.

Note that it cites a previous relevant commit, 0bdced981e4, in its summary, and also points at the issue that it resolves.

If you're in doubt about how to format your commit descriptions, take a look at the recent history and try to mimic the style that you can see others broadly follow there.

When we say commits should be "atomic", we mean with respect to some distinct logical unit, e.g. a type definition used across many files, or a single file, or just a single function in a single file. Commits should be atomic at the level of code, not of entire features. You don't have to squash your commits into a single one that captures everything you're trying to do -- the history will never make for pleasant bedtime reading, so focus instead on making your commits useful for tools like git-blame and git-bisect.

Your contribution must apply cleanly to master in order to be considered mergeable. You may want to regularly rebase your changes onto master in order to both clean up any intermediate "development" commits you make and to ensure that you're up to date.

If you're making a GitHub pull request, it's good practice to make it from a topic branch, rather than master, on your fork.

Pills

Any contribution that touches the kernel (i.e., anything in pkg/arvo/sys), should be accompanied by an updated solid pill. Pills are tracked in the repository via git LFS.

Whenever you make a contribution to the kernel, please create a new solid pill via:

sh/update-solid-pill

and include it along with your contribution.

Historically, we've sometimes included these updated pills in separate, standalone commits (you will see plenty of "pills: update solid" and similar commits if you look through the history), but this practice is considered to be deprecated -- you should usually just include the updated pill in the same commit that updates the source.

Releases

We typically create releases by tagging appropriate commits on master, so any given commit in master may not actually be present in the latest release.

We perform updates by pushing releases over-the-air to ~zod approximately once per week, so any contribution that can be deployed OTA will usually find its way onto the network pretty rapidly.

If you want to propose a hotfix (i.e. a small, OTA-updateable change, usually a bugfix, to some currently-deployed release) then simply make it clear that your contribution -- whether it be a pull request, patch, or whatever -- is intended to be a hotfix. A maintainer can then deploy it to the network outside of the normal release schedule.

Less frequently we release new Vere versions, which requires users to download new binaries, and occasionally, while Urbit is still in early development, we breach network continuity in order to release large changes that are difficult to push out over-the-air. Contributions to Vere, or non-OTA-able updates to Arvo, will find their way into releases before terribly long.

Code style

The Urbit project uses two-space indentation and avoids tab characters. In C code, it should not be too difficult to mimic the style of the code around you, which is just fairly standard K&R with braces on every compound statement. One thing to watch out for is top-level sections in source files that are denoted by comments and are actually indented one level.

Hoon will be a less familiar language to many contributors. We've published some style guidelines for Hoon, but above all you should try to mimic the style of the code around you. With regards to the style used throughout the codebase: the more recently the code was written, the more standard and accepted its style is likely to be.

Kernel development

Working on either C or non-kernel Hoon should not bring any surprises, but the Hoon kernel (anything under pkg/arvo/sys/) is bootstrapped from a so-called pill, and must be recompiled if any changes are made. This should happen automatically when you make changes, but if it doesn't, the command to manually recompile and install the new kernel is |reset in dojo. This rebuilds from the sys directory in the home desk in %clay.

Currently, |reset does not reload apps like dojo itself, which will still reference the old kernel. To force them to reload, make a trivial edit to their main source file (under the app directory) in %clay.

The kernel and pills

Urbit bootstraps itself using a binary blob called a pill (you can see it being fetched from bootstrap.urbit.org on boot). This is the compiled version of the kernel (which you can find in the sys directory of Arvo), along with a complete copy of the Arvo source.

The procedure for creating a pill is often called "soliding." It is somewhat similar to |reset, but instead of replacing your running kernel, it writes the compiled kernel to a file. The command to solid is:

> .urbit/pill +solid

When the compilation finishes, your pill will be found in the [pier]/.urb/put/ directory as urbit.pill.

You can boot a new ship from your local pill with -B:

$ urbit -F zod -B path/to/urbit.pill my-fake-zod

Release pills, i.e. those corresponding to vere releases, are cached at https://bootstrap.urbit.org and are indexed by the vere version number, e.g. urbit-0.8.2.pill.

Pills are also cached in version control via git LFS. You can find the latest solid pill, as well as the latest so-called brass and ivory pills, in the bin/ directory at the repository root. Note that you'll need to initialise git LFS in order to check these pills out:

$ git lfs init
$ git lfs pull

Issues

The GitHub tracker is our canonical source of truth around issues, bugs, performance problems, feature requests, and so on. If you encounter any issues when developing on Urbit, feel free to submit a report about it here.

A good bug report, description of a crash, etc., should ideally be reproducible, with clear steps as to how another developer can replicate and examine your problem. That said, this isn't always possible -- some bugs depend on having created a complicated or unusual state, or can otherwise simply be difficult to trigger again (say, you encountered it in the last continuity era).

Your issue should thus at a minimum be informative. The best advice here is probably "don't write bad issues," where "bad" is a matter of judgment and taste. Issues that the maintainers don't judge to be sufficiently useful or informative may be closed.

Feature requests are welcome, but they should include sufficient detail and explanation, as well as a discussion of perceived benefits one could expect from them. "It would be cool if.." probably does not, in itself, constitute a good feature request; instead, try to be specific about what you're requesting, and what your desired feature would accomplish.

Staying in touch

Questions or other communications about contributing to Urbit can go to support@urbit.org.