Please refer to Contributing 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:
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.
Think about what’s the next step to let the functional test go further (i.e. fail later).
Write a failing unit test for the new code that you want to write.
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
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.
Commit (optional). Commit together the (working) unit tests and the new code.
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.
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.
About merge requests and code reviews¶
Size of merge requests¶
Submitted Merge Requests should consist of a single commit and 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 MR to be smaller than that. The single commit approach ensures that the CI runs on all commits that we merge, and makes it easy to push fixup commits that are all squashed together prior to merge.
If you are working on a large feature, we expect you to submit it progressively in multiple independant 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 want to provide a merge request with multiple commits to make it easier to review, it’s possible but be aware that the commits will be squashed together: thus please write the global commit message in your first commit. If you don’t want your commits to be squashed together, please say so in the merge request description and explain your reasons.
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 adressed, 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 MR.
Use of ‘Draft’ status¶
Draft merge requests are not open for detailed review (unless the submitter explicitly asks for early review). They are usually created early to benefit from the full CI testing during development. Thus they can be freely rebased, refactored and can be subject to large changes. They also don’t have any assignee or reviewer set.
Once the draft status is dropped, the merge request is open for review and subsequent changes to respond to review comments will be restricted to adding new commits.
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
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).
Functions are documented using docstrings with Sphinx markup.
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 foostatements are grouped and sorted at the top, while the
from foo import barstatements are grouped and sorted at the end.
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
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 bugs by putting some pseudo-fields at the end of the description:
for a GitLab issue, use
Fixes: #XX(this is a standard GitLab feature)
for a Debian bug, use
Closes: #XXXXXX(this is implemented by a webhook)
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.
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:
And change the relevant fields from
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:
Instead create a
connected method in the Worker’s Manager and use it:
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¶
Push filtering to the model layer:
.get_or_create()) and add a method
in the ModelManager describing the operation, such as:
Add the suffix _at for the fields of type DateTime:
created_at = models.DateTimeField()
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:
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.assertFalse()for testing boolean expressions
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
instead of methods such as
use the correct underlying method.
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.
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.
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.