How to contribute to Octez¶
The purpose of this document is to help contributors participate to Octez.
There are several ways to get involved with the codebase, and you may want to start with some preliminary steps.
The simplest way to contribute to Octez is to report issues that you may find with the software on GitLab. If you are unsure about an issue consult the technical support sources first and always make sure to search the existing issues before reporting a new one. Some information that are probably important to include in the description: the architecture (e.g. ARM64), the operating system (e.g. Debian Stretch), the network you are connected to (e.g. Carthagenet), the binary or component (e.g. tezos-node crashes or rpc X returns Y while Z was expected).
You may also want to fix some typos and minor errors or incoherencies in the documentation, which is situated in the
docs/ subfolder of the code repository.
This kind of small contributions can be done without creating a merge request, by directly pushing commits to the
typo-doc branch, which is regularly merged into the master branch, e.g., every one or two weeks.
This periodic merging is implemented by a series of MRs named “the typo train”, created for you by a volunteer, and batching the currently pending fixes.
Of course, all these commits will be reviewed before being integrated.
The current edition of the typo train MR can be found in meta-issue #2329.
To directly contribute to the codebase, expertise in a few areas is necessary.
First, make sure that you are proficient enough in OCaml. The community website https://ocaml.org gives a few useful pointers for that. In particular, we use a lot of functors, and a few GADTs in the codebase, so you may want to make sure that you master these advanced concepts. For a more specific explanation of GADT usage in Tezos you can check out Generalized Algebraic Data Types (GADTs).
Then, if you don’t know much about the Lwt library, that’s what you want to learn next. This library is used extensively throughout the code base: we use it to handle concurrency. You can use the online documentation. The chapter on concurrency of the Real World OCaml book has also been ported to Lwt.
While you familiarize yourself with the basics as suggested above, you can have a look at the software architecture of Tezos. It will give you the main components and their interactions, and links to the documentation for the various parts.
You may also want to take a look to some developer tools that can make protocol development more convenient.
Now, that you’re ready to delve into the code, it is time to know how contributions to the code are submitted, reviewed, and finally accepted into the master branch.
Our git strategy¶
First of all, the repository is https://gitlab.com/tezos/tezos. So if you want to contribute, simply create an account there.
There are many ways to use Git, here is ours.
We mostly use merge requests (aka MRs) for contributing to the master branch, meaning that nobody should be pushing into the master branch directly. Once a merge request is ready, it is reviewed and approved, then merged with a merge commit.
We maintain a semi-linear history,
which means that merge requests are only
merged if they are direct suffixes of the master branch.
This means that merge requests are rebased on top of
master before they are merged.
This can only be done automatically if there is no conflict though.
origin/master changes, you should make sure that your branch
can still be rebased on it. In case of conflict, you need to rebase manually
master, checkout your branch and run
git rebase master).
You may have to edit your patches during the rebase.
push -f in your branch to rewrite the history.
Being proficient with interactive rebases is mandatory to avoid
mistakes and wasting time.
This Git strategy is a variant of the git rebase workflow.
Workflow of an MR¶
This section presents a global view of our MR workflow. Details about the individual steps in this workflow are described in the following sections.
Our code review process uses GitLab. First a developer creates a new
branch for preparing the MR.
As this is a private new branch, the developer is free to
rebase, squash commits, rewrite history (
git push --force), etc. at will.
Once the code is ready to be shared with the rest of the team, the developer opens a Merge Request. It is useful to explain why the MR is created, to add a precise description of the code changes, and to check if those are in line with the initial requirements (if responding to an issue), or to the stated reasons (otherwise). Dependencies on other merge requests, other relationships to MRs, to issues, etc, should also be mentioned.
While the code is still not ready to be peer reviewed, but it is merely a
work in progress, the developer prefixes the MR with
Draft:. This will tell everybody
they can look at the code, comment, but there is still work to be done and the
branch can change and history be rewritten.
Finally, when the code is ready for the code review, the developer removes the Draft status of the MR and freezes the branch. From this moment on, the developer will refrain to rewrite history, but he/she can add new commits and rebase the branch for syncing it with master (this can be done regularly to make sure the branch does not get stale). At this point the developer interacts with the reviewers to address their comments and suggestions.
GitLab allows both to comment on the code and to add general comments on the MR. Each comment should be addressed by the developer. He/she can add additional commits to address each comment. This incremental approach will make it easier for the reviewer to keep interacting till each discussion is resolved. When the reviewer is satisfied, he/she will mark the discussion resolved.
When all discussions are resolved, you should squash any fix-up commits that were applied (don’t forget to edit the commit message appropriately). Then, the reviewer will rebase the branch and merge the MR in the master branch.
Preparing a Merge Request¶
While working on your branch to prepare a Merge Request, make sure you respect the following rules:
Give a meaningful and consistent name to the branch
It is useful to prefix the name of the branch with the name of the developer to make it clear at a glance who is working on what: e.g.
Note that some extra CI jobs are only run on demand for branches other than master. You can (should) activate these jobs by including keywords in the branch name.
opamin the branch name if you want to explicitly trigger the OPAM packaging pipeline. Note that any OPAM related changes will automatically trigger it.
docin the branch name if you change the documentation.
arm64in the branch name if you need to build ARM64 artifacts.
dockerin the branch name if you need an automatic (instead of manual) CI job for building Docker images.
Suffix the branch name by
-releaseif it is a release branch.
Prefer small atomic commits over a large one that does many things.
Don’t mix refactoring, reindentation, whitespace deletion, or other style changes with new features or other real changes.
No peneloping: don’t do something in a commit just to undo it two commits later.
We expect every commit to compile and pass tests. Obviously, we require tests to pass between each MR.
Follow the format of commit names, <Component>: <message>, with message in indicative or imperative present mood e.g.
Shell: fix bug #13rather than
Shell: fixed bug #13. Use multiline commit messages for important commits.
Adhere to the coding guidelines.
Document your changes, in the MR description and commit messages. Imagine if somebody asked what your change was about in front of the coffee machine, write down your answer and put it in the MR.
If there is a design description at the top of the file, consider updating it to reflect the new version. Additionally, if you feel that your design changes are worth mentioning to help upcoming contributors (e.g. justify a non-obvious design choice), you should document them in this file header, but in a separate “History” section.
If you add new functions to an interface, don’t forget to document the function in the interface (in the corresponding .mli file; or, if there is no .mli file, directly in the .ml file)
If you add a new RPC endpoint or modify an existing one, be sure to take into account the impact on RPC security.
If you modify the user API (e.g. add or change a configuration parameter or a command-line option), update the corresponding documentation. In particular, for configuration parameters of the Tezos node, update the node configuration documentation and the documentation of the modified component(s), usually referred by that page.
If your MR introduces new dependencies, follow the additional instructions.
Check whether your changes need to be reflected in changes to the corresponding README file (the one in the directory of the patched files). If your changes concern several directories, check all the corresponding README files.
For parts that have specifications in the repository (e.g., Michelson), make sure to keep them in sync with the implementation.
Creating the Merge Request¶
Your goal is to help the reviewers convince themselves that your patch should be merged. Well-documented merge requests will receive feedback faster. Complicated patches with no comments to help the reviewer will cause the reviewer to make the wrong decision or will discourage the reviewer to work on the MR.
Therefore, when creating your MR, observe the following rules:
Give it an appropriate title.
Give context: why was this patch written?
Does it fix a bug, add a feature or refactor existing code?
Is there an open issue on GitLab, or a post from an angry user somewhere?
Must it be merged before another merge request?
Explain how you tested your patch (or why you didn’t).
The description of merge requests must include instructions for how to manually test them, when applicable.
Merge requests should include automated tests for new functionality and bug fixes.
Refer to the testing guide for more information.
Bug fixes should include a test that demonstrates that the bug has been fixed (i.e. that fails before the supplied fix).
The test coverage can be used to guide testing of the proposed MR. If the modified code lacks coverage, then this indicates that tests should be added.
If no tests are included, a justification should be given in the description. Possible justifications include that testing is prohibitively difficult, or that the modified code is already well-exercised by the existing test suite. The point of the justification is to stress the importance of testing and to guide improvements of the test framework.
Divide and conquer: it is easier to merge several simple commits than a big one.
Isolate complicated parts of your patch in their own commits.
Put simple, non-controversial commits first. For instance: commits that fix typos, improve documentation, or are simple enough that we may want to merge them even without the rest of the merge request. Even better put them in a separate MR which can be merged easily.
Split your commits so that each step is convincing on its own, like the proof of a big theorem which is split into several lemmas.
Avoid merge requests that are too large. They are harder to rebase and request a longer continuous time for reviewing, making them overall slower to merge. See favoring small merge requests below for more details.
Anticipate questions: explain anything which may look surprising, as comments in the code itself if it has value to future readers, or in the MR description.
MR Labels: Add GitLab labels to the MR, like
The following special labels can be used to trigger different parts of the CI pipeline. To take effect, the label must be added before any push action is made on the MR.
ci--opamis for triggering the opam packaging tests pipeline.
ci--docsis for testing some scripts in the documentation (e.g. Octez installation scenarios).
ci--dockeris for publishing the docker image of the MR.
ci--arm64is for building on the ARM64 architecture.
MR Options: When opening an MR you should probably tick the following options:
Delete source branch when merge request is accepted. Helps keeping the repository clean of old branches.
Squash commits when merge request is accepted. Sometimes it’s useful to have many small commits to ease the review and see the story of a branch, but they are not relevant for the history of the project. In this case they can be squashed and replaced with a single meaningful commit. Nevertheless, you should squash yourself all fix-up commits when all discussions are resolved, as described above in the MR workflow, in order to ease the reviewers’ task.
Allow commits from members who can merge to the target branch. This option is useful to allow members of the Octez merge team, who are not developers in your project, to commit to your branch. It helps to rebase and propose fixes.
Find reviewers: it is the responsibility of the author to find suitable reviewers. In this context, finding a reviewer means finding someone that agreed to review in the next days after the MR becomes ready. Opting for a reviewer that is not in the capacity to review your MR in the next days when others can is unfortunate, because the merge request will become unnecessarily blocked; which is bad for:
the author, as their work gets delayed, and
the health of the repository, as it gives the impression that a new contribution will land soon; while it is not the case.
To find reviewers that will review promptly, we refer to the documentation of the reviewer field below.
Check progress: It is important to maintain to a minimum the number of your MRs that are in Draft state, and to constantly check that the discussion is progressing.
Example of an MR with a good, clean history (each bullet is a commit, any subitems represent the longer description of that commit):
* Doc: mark bug #13 as fixed * Test_python: add test for p2p bug #13 * Flextesa: add test for p2p bug #13 * Shell: fix p2p bug #13 - fix bug in the shell - fix relative unit test - add docstrings
Beware: For MRs touching
src/proto_alpha/lib_protocol, see protocol MRs.
Favoring Small Merge Requests¶
Small merge requests are encouraged for multiple reasons:
They are faster to review, which encourage reviewers to pick them.
They are easier to rebase, hereby saving developers time.
They are reviewed more thoroughly.
If the merge request is not accepted, less work is lost; in particular less review time has been spent.
However, small merge requests also come with drawbacks:
They make it more difficult for reviewers to get the global picture of the intended change.
They may introduce intermediate states, during which a feature is not yet finished; or dead code is temporarily introduced.
They have to be reverted if the entire feature is ultimately cancelled.
tezos/tezos to evolve fast, however, we are convinced that the advantages
of small merge requests outweigh the drawbacks. If possible, drawbacks
must be mitigated as follows:
Have the entire piece of work described or done somewhere. For example in an issue, or a branch containing the entire change, or a large (unsplit) work as a draft merge request. For complex works, an external document may be referred in the issue/MR, detailing the design/implementation rationale; if such documents are only targeted to reviewers and/or are only describing a change, they should not go in the online documentation.
Include a link to the entire piece of work in the description of each small merge requests created by splitting the large piece of work. This will help reviewers get the big picture.
Explain why the intermediate state is harmless, if applicable.
To mitigate loss of work if the whole piece is not accepted, we advise to split the work so that improvements that are desirable on their own are the first ones to be merged in the sequence of small merge requests. A desirable standalone improvement is for example a refactoring that improves the quality of the code, or adds new tests, or fixes typos.
Merge Request “Assignees” Field¶
Merge requests targeted at
tezos/tezos master should in general
have exactly one assignee. The assignee is someone from which an
action is required to get the merge request moving. Example actions include:
respond to a comment thread;
update the code;
rebase (in particular in case of conflicts);
find someone else who can get the merge request moving.
The assignee will thus often be one of the reviewers (if he needs to review or respond to a comment) or one of the merge request authors (if they need to update the code or respond to a comment).
If a merge request has no assignee, it is implicitly the role of the merge coordinator to assign it to someone.
Even though merge requests could require action from several people to be merged, we avoid assigning more than one to avoid diluting responsibility.
Merge Request “Reviewers” Field¶
The reviewer field of GitLab is used to specify reviewers. Once the merge request is ready for review, assign one of the reviewers that you specified in the reviewers field. As mentioned previously, it is the responsibility of authors to find reviewers. To find reviewers, either:
Advertize your merge request on the
#mr-advertizingchannel of the tezos-dev Slack. Good advertisement consists of a link to the MR and a one sentence summary.
Look at authors of the code you are modifying using git blame.
Ask help to the merge coordinator, either by asking him/here on Slack or mentioning them in a comment (see next paragraph).
Depending on your GitLab role
you may or may not be able to use the Reviewers field for specifying
the reviewers. If you don’t have the right, mention the reviewers using
their GitLab handle (username prefixed with
@) in a comment.
It causes GitLab to send a notification to them.
Merge Request “Draft” Mode¶
A merge request that is not yet ready for review should be marked
by prefixing its title with
tezos/tezos draft merge requests are ignored by reviewers.
Marking merge requests as draft hence helps lower
the number of merge requests that require attention from the
Octez merge team.
Special case: MRs that introduce a new dependency¶
In the special case where your MR adds a new opam dependency or updates an existing opam dependency, you will need to follow this additional dedicated guide:
for Python, you can refer to the related section in the python testing documentation.
the Rust dependencies are located in the GitLab repository tezos-rust-libs and the instructions are listed there.
For others, there is currently no dedicated guide. Do not hesitate to ask for
help on the
#devteam channel on the tezos-dev Slack.
Protocol development MRs¶
Because of the amendment procedure that governs the protocol, the workflow for protocol development is significantly different from master.
Before a proposal, a new directory, e.g.
proto_alpha where the development continues.
The hash of each active or candidate protocol is computed from the directory
src/proto_0*/lib_protocol, so every change in these directories
Right before the activation of a new protocol, there is a migration of the context that takes place. This migration typically converts data structures from the old to the new format. Each migration works exclusively between two protocol hashes and it is useless otherwise. For this reason after the activation of a protocol the first step to start a new development cycle is to remove the migration code. In order to facilitate this, migration code is always in a different commit with respect to the protocol features it migrates. When submitting an MR which contains migration code, the author must also have tested the migration (see How to Test a Protocol Proposal) and write in the description what was tested and how so that reviewers can reproduce it.
At Tezos all the code is peer reviewed before getting committed in the master branch by the Octez merge team. Briefly, a code review is a discussion between two or more developers about changes to the code to address an issue.
Merge Request Approvals¶
Two approvals from different Octez Octez merge team members are required for merge requests to be merged. Both approvals must result from independent thorough reviews. After both reviews, the second approver will also typically merge.
However, for less critical parts of the code, an Octez merge team member may choose to trust the review of a developer who is not a member of the Octez merge team. In that case, the Octez merge team member may choose to count that review as their own, effectively “converting” the other developer’s passing review into an approval. This may be done even when the Octez merge team member is the author of the merge request. A second independent review from an Octez merge team member is still required, of course.
For this reason, if you make a partial review, for instance if you only reviewed part of the code, or only the general design, it is good practice to say so in a comment, so that other reviewers know what is left to review. If you manually tested the merge request or ran some benchmarks, you can add a comment with the results.
Code review can be challenging at times. These are suggestions and common pitfalls a code reviewer should avoid.
Ask questions: How does this function work? If this requirement changes, what else would have to change? How could we make this more maintainable?
Discuss in person for more detailed points: Online comments are useful for focused technical questions. On many occasions it is more productive to discuss it in person rather than in the comments. Similarly, if discussion about a point goes back and forth, it will be often more productive to pick it up in person and finish out the discussion.
Explain reasoning: Sometimes it is best to both ask if there is a better alternative and at the same time justify why a problem in the code is worth fixing. Sometimes it can feel like the changes suggested are nit-picky without context or explanation.
Make it about the code: It is easy to take notes from code reviews personally, especially if we take pride in our work. It is best to make discussions about the code than about the developer. It lowers resistance and it is not about the developer anyway, it is about improving the quality of the code.
Suggest importance of fixes: While offering many suggestions at once, it is important to also clarify that not all of them need to be acted upon and some are more important than others. It gives an important guidance to the developer to improve their work incrementally.
When you consider that a fix is important but should not prevent the current MR to be merged (e.g., because it adds a sufficient amount of useful new features), you may suggest creating a follow-up issue. If the place in the code that needs to be fixed later is clear, you may also suggest marking it with a TODO/FIXME comment.
Take the developer’s opinion into consideration: Imposing a particular design choice out of personal preferences and without a real explanation will incentivize the developer to be a passive executor instead of a creative agent.
Do not re-write, remove or re-do all the work: Sometimes it is easier to re-do the work yourself discarding the work of the developer. This can give the impression that the work of the developer is worthless and adds additional work for the reviewer that effectively takes responsibility for the code.
Consider the person you are reviewing: Each developer is a person. If you know the person, consider their personality and experience while reviewing their code. Sometimes it is possible with somebody to be more direct and terse, while other people require a more thorough explanation.
Avoid confrontational and authoritative language: The way we communicate has an impact on the receiver. If communicating a problem in the code or a suggestion is the goal, making an effort to remove all possible noise from the message is important. Consider these two statements to communicate about a problem in the code : “This operation is wrong. Please fix it.” and “Doing this operation might result in an error, can you please review it?”. The first one implies you made an error (confrontational), and you should fix it (authority). The second suggests to review the code because there might be a mistake. Despite the message being the same, the recipient might have a different reaction to it and impact on the quality of this work. This general remark is valid for any comment.
When reviewing MRs involving documentation, you may check the built documentation directly within the Gitlab interface, see Building the documentation in the CI.
The Merge-Request Bot¶
Every 6 hours, an automated process running as the Tezbocop 🤖 user, inspects recent MRs and posts or edits comments on them, giving an inspection report on the contents of the MR.
Some warnings/comments are for you to potentially improve your MR, other comments just help us in the assignment & review process.
The first time Tezbocop posts a message you should receive a notification; for the subsequent edits there won’t be notifications; feel free to check Tezbocop’s comments any time.
If you think some of the remarks/warnings do not apply to your MR feel free to add a comment to justify it.
In particular, the Merge-Request Bot may complain about TODO/FIXME comments without an issue number ensuring that the intended evolution is tracked.
The code for the bot is at oxheadalpha/merbocop. It is of course work-in-progress and new warnings and comments will appear little by little. We welcome specific issues or contributions there too.