Coding practices

Please refer to Contribute to debusine for general contributing guidelines. Once you are setup to contribute and know what you want to work on, you should read this document to learn our expectations on the code that you will submit as well as on the process leading to integrate your contribution.

Introduction to Test Driven Development

We expect you to follow this methodology when developing new code so if you are not familiar with it, have a look at Test-Driven Web Development with Python.

The suggested workflow looks like this:

  1. Add a functional test that covers the new feature from the point of view of the user. This test will fail since the feature doesn’t exist yet.

  2. Think about what’s the next step to let the functional test go further (i.e. fail later).

  3. Write a failing unit test for the new code that you want to write.

  4. Write the minimal code to make your unit test pass. You will typically run this very often:

    $ ./manage.py test path-to-the-testing-folder

  5. Refactor (if needed). You might have introduced some duplication in your code or in your tests. Clean that up now before it’s too late.

  6. Commit (optional). Commit together the (working) unit tests and the new code.

  7. If you made progress from the functional tests point of view, go back to point 2, otherwise go back to point 3. If the functional test passes, continue.

  8. Commit. The functional tests are committed at this point to ensure that they are committed in a working state:

    $ git add .
$ git commit

When you don’t develop a new feature, your workflow is restricted to steps 3 to 6.

Git commit messages

Please invest some time to write good commit messages. Just like your code, you write it once but it will be read many times by different persons looking to understand why you made the change. So make it pleasant to read.

The first line is the “summary” (or title) and describes briefly what the commit changes. It’s followed by an empty line and a long description. The long description can be as long as you want and should explain why you implemented the change seen in the commit.

The long description can also be used to close GitLab issues by putting some Fixes: #XXX pseudo-fields at the end of the description.

Pre-commit hooks

The setup instructions for developers recommend setting up pre-commit. This runs a variety of linters on the files changed by the commit to catch some problems that can be found by static analysis. These are defined as a collection of “hooks” in .pre-commit-config.yaml.

To run linters over all files rather than just changed files, run pre-commit run -a. This is particularly useful after a rebase, since the git hooks that pre-commit uses by default don’t run on rebase (and the ones that do run on rebase don’t operate on specific files anyway).

pre-commit run -a is also useful if you want to check unstaged files, since pre-commit normally only checks committed and staged files, and temporarily stashes unstaged changes. Things will normally work better if you git add your changes first.

To skip a particular pre-commit hook, you can set the SKIP environment variable to a comma-separated list of hook IDs: for example, SKIP=mypy.

To bypass pre-commit hooks temporarily for a single commit, use git commit -n.

To upgrade hooks, run pre-commit autoupdate, and then pre-commit run -a to check for new errors. Note that some hooks may be deliberately pinned to older versions, in which case you will need to revert changes to their versions afterwards.

About merge requests

Size of merge requests

Submitted merge requests should have a reasonable size: that is no more than 1000 lines of changes (total of added and removed lines, excluding test data and other auto-generated files) for large changes. We expect the majority of merge requests to be smaller than that.

If you are working on a large feature, we expect you to submit it progressively in multiple independent steps. If the separate merge of those individual steps would break debusine, then ask us to create a staging branch for your new feature and target that separate branch in your merge requests.

If you figure out that some refactoring is needed before you can reasonably implement your feature, consider doing that refactoring in a separate merge request.

A merge request can have multiple independent commits to make it easier to review.

Updates to merge requests

When a merge request has multiple independent commits, we want to preserve that split and not squash all the commits together. This means that fixes and improvements done following the review have to be incorporated in their respective commit.

With regular rebase -i

Assuming that the merge request targets the devel branch, you can do that locally with:

$ git commit --fixup=$COMMIT
$ git rebase -i --autosquash $(git merge-base HEAD devel)
$ git push --force-with-lease origin $BRANCH

The rebase command above ensures that the merge request is not rebased against newer version of the devel branch: this is important for big merge requests that have already been reviewed, so that the diff associated to the push does not include changes coming from the devel branch.

Note

If you need to rebase such a branch against the latest version of the devel branch, please do that in a push where that is the only change. You can do that with:

$ git fetch origin
$ git rebase origin/devel
$ git push --force-with-lease origin $BRANCH

For small merge requests, this is not so important and you can rebase against devel or origin/devel directly:

$ git rebase -i --autosquash origin/devel

With push of fixup commits

Alternatively, you can simply push fixup commits to the remote branch while review is in progress, and let the final git rebase -i --autosquash be done by whoever gets to merge the branch. Note however that the presence of fixup commits in the history will change the status of the merge request back to “draft”.

Note

If the merge request has multiple commits, and if you use GitLab’s “Add suggestion” feature, please input a commit message fixup! $TARGET_COMMIT_TITLE so that it can be easily autosquashed by the person that will perform the final merge.

Finally, if the merge request has a single commit (and is expected to stay that way), you should be free to select “Squash commits when merge request is accepted” so that the final merge automatically squashes any fixup commit.

Discussions in merge requests

Discussions within merge requests should be limited to code reviews: pointing out mistakes and inconsistencies with the associated issue, suggesting improvements, etc. If there are architecture or design issues that need to be addressed, or if there are disagreements between the coder and the reviewer, then those discussions should be moved to a separate issue and be resolved there before getting back to the merge request.

Work in progress merge requests

In order to benefit from the full CI during development, developers can open merge requests early while they are still working on the feature.

Those merge requests should have the draft status and the MR: Work in progress label. They can be freely rebased, refactored and can be subject to large changes. They also don’t have any assignee or reviewer set.

Such merge requests are not open for detailed review (unless the submitter explicitly asks for early review).

Once the MR: Work in progress label is dropped, the merge request is open for review and subsequent changes should follow the rules described in Updates to merge requests.

Note

We rely on the label rather than only on the “draft” status, because push of fixup commits might bring back the status of a good merge request back to draft. Hence you can’t rely solely on the presence of the draft flag to decide to not review a merge request.

Merge requests labels

To help keep track of the status of the various merge requests, we have a few labels all starting with “MR:” (for easy auto-completion):

  • MR: Work in progress: the work is not yet ready to be reviewed

  • MR: Needs work: a reviewer found issues, they have to be addressed. This label doesn’t need to be immediately set during review, it can be set later if we realize that the submitter is not dealing with comments in a timely fashion. The label can be dropped by the submitter after having handled all the issues that have been identified.

  • MR: UI review needed: the merge request contains a prototype UI that needs usability feedback (usually via a playground-vm machine) before proceeding with further work

Note that any merge request without any “MR” label is thus implicitly in a status where the branch is assumed to be ready and where the submitter is thus seeking reviews.

Usage of ‘Assignee’ and ‘Reviewer’

As a submitter of a merge request, you don’t have to set those fields, by default every contributor is encouraged to review open merge requests. There are two cases where we might set the Reviewer field:

  • when the submitter would like to ask someone specific to review the MR (due to former experience on the code being modified for example)

  • when a reviewer wants to notify other team members that he is currently reviewing that merge request

Final merge and approvals

The final merge, if not done by a core developer, should only happen after a core developer has approved the merge request and after all comments left by the core developer (if any) have been addressed.

Prior to merge, one should always ensure that all fixup commits have been properly squashed in their respective commits.

If such an operation is required, you should take the opportunity to rebase the branch against the latest version of the devel branch. The rebase is also recommended if the base of the merge request is far behind the current tip of the devel branch: the rebase will trigger a final pipeline run on a codebase that will be much closer to the result of the merge, thus limiting the risk of failed pipeline on devel after the merge.

Coding style

  1. In regard to coding style, we observe PEP8 with a few exceptions. And we format the code with black by running make black (we use a different line length and don’t impose the single type of quote for strings).

  2. Functions are documented using docstrings with Sphinx markup.

  3. Imports are sorted in multiple groups separated by one empty line: first a group for __future__ imports, then a single group for all the Python standard modules, then one group for each third-party module (and groups are sorted between them as well), followed by groups for the project modules, and last, one group for relative imports.

    Within each group the import foo statements are grouped and sorted at the top, while the from foo import bar statements are grouped and sorted at the end.

    Example:

from __future__ import print_function

import datetime
import os
from datetime import timedelta
from email.utils import getaddresses, parseaddr

from django.conf import settings
from django.db import connection, models
from django.utils.safestring import mark_safe

import requests
from requests.structures import CaseInsensitiveDict

from debusine.artifacts.models import Artifact

Good code, good design

This section documents different decisions about implementation, naming, etc. that happened during merge requests. It is not an exclusive list of all the discussions and is subject to change.

Those rules are meant to help improve consistency and to obtain a cleaner overall design.

Models

Avoid changing fields from outside the model

Avoid changing fields in the models from their users. Do not do:

worker.connected_at = timezone.now()

Instead, create a method in Worker describing the action that you are doing:

worker.mark_connected()

And change the relevant fields from mark_connected().

This allows the model’s fields or logic to change without having to change the code which accesses it.

Read more in Push actions to the model layer.

Push filtering to the model layer

In order to encapsulate logic for filter and other queries, add a ModelManager to the Model and do the filtering there. Do not do:

worker.objects.filter(connected_at__isnull=False)

Instead create a connected method in the Worker’s Manager and use it:

worker.objects.connected()

This allows the code base to be consistent in the filtering.

Read more in Push filtering to the model layer.

Push Model.objects.create() to the model layer

Similar to Push filtering to the model layer: avoid using Model.objects.create() (or .get_or_create()) and add a method in the ModelManager describing the operation, such as:

Worker.objects.create_with_fqdn(fqdn, token)

Naming fields

Add the suffix _at for the fields of type DateTime:

created_at = models.DateTimeField()

Model method order

Follow the Code Review Doctor model method order:

  1. Field choice tuples

  2. Database fields

  3. Custom manager attributes

  4. class Meta

  5. def __str__()

  6. def save()

  7. def delete

  8. def get_absolute_url()

  9. Any custom methods

The Code Review Doctor method order is compatible with order in the Django documentation making the choice tuples and delete order explicit.

Error handling

Management commands that create model instances should show helpful error messages for any resulting constraint violations. PostgreSQL’s error messages are informative but can be difficult for non-experts to read. Since management commands aren’t very performance-sensitive, it is usually best to ask Django to perform its own constraint validation before inserting data into the database so that it can produce better error messages. For example:

try:
    collection = Collection(...)
    collection.full_clean()
    collection.save()
except ValidationError as exc:
    raise CommandError(
        "Error creating collection: " + "\n".join(exc.messages),
        returncode=3,
    )

Templates

Naming templates

If a Django template file is used to render a full page (including the HTML header, footer, etc.) it should follow a similar structure to work_request-list.html.

If a template file is meant to be included from another template, add a _. For example: _work_request-list.html is meant to be included from another template.

Include templates: specify the context explicitly

When including a template, specify the context made available to it:

{% include "web/_workspace-list.html" with workspace_list=workspace_list user=user only %}

It helps the reader to know which context is used by the template and also avoids using context that might be available from one {% include ... %} but not from another {% include ... %}.

Tests

Private methods

To facilitate Test-Driven Development and localised tests, it is ok to call private methods from the tests.

Assert function: order of the parameters

In the assert methods, put the “expected” value as second parameter, for example:

self.assertEqual(actual, expected)

Reason: some test methods such as assertQuerySetEqual expect “actual” to be the first parameter. Always using this order helps the tests to be more easily read.

Assert functions: assertEqual or specific

When there is a TestCase method with specific semantics, use them:

  • self.assertQuerySetEqual() for testing querysets

  • self.assertTrue() or self.assertFalse() for testing boolean expressions

  • self.assertIn() or self.assertNotIn()

Using the specific methods such as self.assertIn() helps to have a better test output compared with constructions such as self.assertTrue('john' in people).

When possible (actual and expected are the same type), use self.assertEqual() instead of methods such as self.assertDictEqual(). self.assertEqual() will use the correct underlying method.

Populating the database: Playground

If, while writing test code, you end up creating factory methods to create pydantic or database model objects, for example creating source or binary packages, check if suitable code already exists in the various Playground classes.

If you need to create new ones, consider adding them to Playground classes so they can be reused, both in tests and in creating scenarios for testing UI prototypes.

Playground integrates well with Django’s TestCase.setUpTestData, so if you are testing a nontrivial scenario you can create it once per test class, and have it rolled back to its pristine state before every test method. Playground takes care to ensure this also works for file store changes.

Evaluating UI prototypes

There is a bin/playground-vm script that will deploy the branch from a merge request to a newly created VM, explicitly intended to be used to publish a draft UI prototype for evaluation by other developers.

The script runs examples/playground/populate during provisioning, so, to discuss a UI prototype, you can:

  1. Code the prototype views and templates; at this stage your branch doesn’t need to pass CI or have full code coverage.

  2. Make sure examples/playground/populate populates the database with enough information to show the features you are coding. Feel free to extend it otherwise.

  3. Push your branch to a merge request. You can push without running the CI by using git push -o ci.skip

  4. Use bin/playground-vm to deploy the merge request to a publicly accessible VM.

Naming data keys

Names of data keys in artifact, collection, and task definitions can be expected to be used in pydantic models. As such, they must be valid Python identifiers: in particular, use underscores rather than hyphens as separators.

General guidelines

Constants

If one of our dependencies provides defined public constants, use them instead of re-defining them or using magic numbers.

# Use:
from rest_framework import status
code = status.HTTP_501_NOT_IMPLEMENTED

# Instead of:
code = 501

This helps readability for readers that might not know all the internal codes, might avoid typos and if the “constants” depended on versions, environment, etc. the library will take care of them.

Type hints

If you want to indicate the type of a variable, type hints are preferable to adding suffixes to the variable.

# Use:
def method(information: dict):
   ...

# Instead of:
def method(information_dict):
    ...

This helps (IDEs, mypy) to give hints to the programmer and it keeps the variable names shorter, avoiding the type repetition.

Early exit

To exit early:

# Use:
raise SystemExit(3)

# Instead of:
sys.exit(3)
exit(3)

It says explicitly what it does and there is no need to import the sys module.

If any utility in Debusine must exit early:

  • Use exit code 3. Exit code 1 is used by the Python interpreter for unhandled exceptions and exit code 2 by the argparse module for invalid command syntax.

  • Make sure to log or print (depending on the circumstances) why an early exit has happened so the user or admin can fix the situation.