This document is a work in progress
- The main branch should always be stable and ready to release. All development occurs on the main branch.
- All changes to the main branch occur via merge requests. Individual developers are free to choose their own development workflow in their own repositories.
- Merge requests will not be accepted unless:
- All existing tests pass (or have been updated as needed).
- New tests have been written to cover newly added features.
- Code coverage is as close to 100% as possible.
- Coding conventions are adhered to (unless there is good reason not to).
To aid readability, all commit messages should be prefixed with one or more of the following labels (this convention has been inherited from nibabel):
- BF : bug fix
- RF : refactoring
- ENH: enhancement/new feature
- BW : addresses backward-compatibility
- OPT : optimization
- BK : breaks something and/or tests fail
- PL : making pylint happier
- DOC : for all kinds of documentation related commits
- TEST: for adding or changing tests
- MNT : for administrative/maintenance changes
- CI : for continuous-integration changes
The fslpy
version number roughly follows semantic versioning rules, so that dependant projects are able to perform
compatibility testing. The full version number string consists of three
numbers:
major.minor.patch
- The
patch
number is incremented on bugfixes and minor (backwards-compatible) changes. - The
minor
number is incremented on feature additions and/or backwards-compatible changes. - The
major
number is incremented on major feature additions, and backwards-incompatible changes.
The version number in the main
branch should be of the form
major.minor.patch.dev0
, to indicate that any releases made from this
branch are development releases (although development releases are not part of
the release model).
A separate branch is created for each minor release. The name of the
branch is v[major.minor]
, where [major.minor]
is the first two
components of the release version number (see above). For example, the branch
name for minor release 1.0
would be v1.0
.
Patches and bugfixes may be added to these release branches as patch
releases. These changes should be made on the main branch like any other
change (i.e. via merge requests), and then cherry-picked onto the relevant
release branch(es).
Every release commit is also tagged with its full version number. For
example, the first release off the v1.0
branch would be tagged with
1.0.0
. Patch releases to the v1.0
branch would be tagged with
1.0.1
, 1.0.2
, etc.
Follow this process for major and minor releases. Steps 1 and 2 should be performed via a merge request onto the main branch, and step 4 via a merge request onto the relevant minor branch.
- Update the changelog on the main branch to include the new version number and release date.
- On the main branch, update the version number in
fsl/version.py
to a development version of the next minor release number. For example, if you are about to release version1.3.0
, the version in the main branch should be1.4.0.dev0
. - Create the new minor release branch off the main branch.
- Update the version number on the release branch. If CI tests fail on the release branch, postpone the release until they are fixed.
- Tag the new release on the minor release branch.
Follow this process for patch releases. Step 1 should be performed via a merge request onto the main branch, and step 2 via a merge request onto the relevant minor branch.
- Add the fix to the main branch, along with an updated changelog including the version number and date for the bugfix release.
- Cherry-pick the relevant commit(s) from the main branch onto the minor release branch, and update the version number on the minor release branch. If CI tests fail on the release branch, go back to step 1.
- Tag the new release on the minor release branch.
Unit and integration tests are currently run with py.test
and
coverage
.
- Aim for 100% code coverage.
- Tests must pass on python 3.5, 3.6, and 3.7.
- Clean, readable code is good
- White space and visual alignment is good (where it helps to make the code more readable)
- Clear and accurate documentation is good
- Document all modules, functions, classes, and methods using ReStructuredText.
Configure your text editor to use:
- flake8: This checks your code for adherence to the PEP8 coding standard.
- pylint: This checks that your code follows other good conventions.
Because I like whitespace and vertical alignment more than PEP8 does, the following violations of the PEP8 standard are accepted (see here for a list of error codes):
- E127: continuation line over-indented for visual indent
- E201: whitespace after '('
- E203: whitespace before ':'
- E221: multiple spaces before operator
- E222: multiple spaces after operator
- E241: multiple spaces after ','
- E271: multiple spaces after keyword
- E272: multiple spaces before keyword
- E301: expected 1 blank line, found 0
- E302: expected 2 blank lines, found 0
- E303: too many blank lines (3)
- E701: multiple statements on one line (colon)
- W504: line break after binary operator
The pylint
tool can be very opinionated about how you write your code,
and also checks many of the same things as flake8
. So I disable all
refactoring and convention messages, and a few select warnings (type pylint
--list-msgs
for a full list of codes):
- W0511 (
fixme
): Warn aboutTODO
andFIXME
comments - W0703 (
broad-except
): Warn about too-generalexcept
blocks (e.g.except Exception:
) - W1202 (
logging-format-interpolation
): Warn about usingformat
when calling a log function, instead of using%
string formatting.
To check code with flake8
and pylint
, I use the following commands:
flake8 --ignore=E127,E201,E203,E221,E222,E241,E271,E272,E301,E302,E303,E701,W504 fsl pylint --extension-pkg-whitelist=numpy,wx \ --generated-members=np.int8,np.uint8,np.int16,np.uint16,np.int32,np.uint32,np.int64,np.uint64,np.float32,np.float64,np.float128,wx.PyDeadObjectError \ --disable=R,C,W0511,W0703,W1202 fsl