Changelog¶
Next 2.x feature release
[Support]: Remove upper bounds pinning on many deps; this makes it easier for related projects to test upgrades, run CI, etc. In general, we’re moving away from this tactic.
3.3.0 2023-05-12
[Feature]: Add mypy type-checking variant of the recently added import test, in
packaging.release.test_install
. This helps prove packages exposingpy.typed
in their source tree are including it in their distributions correctly.
3.2.0 2023-05-11
[Feature]: Minor enhancements to the
checks
module:blacken
now has aformat
alias (and will likely reverse the real name and the alias in 4.0)Added
lint
task which currently just runsflake8
, will likely learn how to be configurable later.Added
all_
default task for the collection, which runs bothblacken
(in regular, not diff-only mode - idea is to be useful for devs, not CI, which already does both independently) andlint
in series.
3.1.0 2023-05-02
[Feature]: Updated
packaging.release.test_install
to attempt imports of freshly test-installed packages, to catch import-time errors on top of install-time ones. This can be opted out of by giving theskip_import
kwarg (aka the--skip-import
flag on the CLI).[Support]: Unpin
tabulate
in our install requirements, it’s had many more releases since we instituted a defensive pin vs some bugs in its later 0.7 line!
3.0.2 2023-04-28
[Support]: Unpin
tabulate
in our install requirements, it’s had many more releases since we instituted a defensive pin vs some bugs in its later 0.7 line!
3.0.1 2023-01-06
[Bug]: We neglected to remove references to
six
in a few spots - including some that utilized Invoke’s old vendor of same; this causes issues when trying to use development and upcoming versions of Invoke. Six is now truly gone!
3.0.0 2022-12-31
[Support]: Drop Python 2 (and 3.5) support. We now support Python 3.6+ only. This naturally includes a number of dependency updates (direct and indirect) as well.
Warning
This is a backwards-incompatible change.
[Support]: The
invocations.travis
module has been removed. If you relied upon it, we may accept PRs to make the newerinvocations.ci
module more generic.Warning
This is a backwards-incompatible change.
[Support]: The
dual_wheels
,alt_python
, andcheck_desc
arguments/config options for theinvocations.packaging.release
module have been removed.Warning
This is a backwards-incompatible change.
Note
If you were using
check_desc
, note that the release tasks have been usingtwine check
for a few releases now, as a default part of execution, and will continue doing so;check_desc
only impacted the use of the oldersetup.py check
command.[Support]: Various fixes and doc updates re: the
autodoc
module’s compatibility with modern Sphinx versions.
2.6.1 2022-06-26
[Support]: Remove upper bounds pinning on many deps; this makes it easier for related projects to test upgrades, run CI, etc. In general, we’re moving away from this tactic.
2.6.0 2022-03-25
[Feature]: Update
packaging.release.publish
with a new config option,rebuild_with_env
, to support a downstream (Fabric) release use-case.[Feature]: Enhance
packaging.release.test-install
so it’s more flexible about the primary directory argument (re: adist
dir, or a parent of one) and errors usefully when you (probably) gave it an incorrect path.
2.5.0 2022-03-25
[Feature]: Port
make-sshable
from thetravis
module to the newci
one.
2.4.0 2022-03-17
[Feature]: Add additional CLI flags to the use of
gpg
when signing releases, to support headless passphrase entry. It was found that modern GPG versions require--batch
and--pinentry-mode=loopback
for--passphrase-fd
to function correctly.[Feature]: Add a new
invocations.ci
task module for somewhat-more-generic CI support than the now legacyinvocations.travis
tasks.[Feature]: Allow supplying additional test runners to
pytest.coverage
; primarily useful for setting up multiple additive test runs before publishing reports.
2.3.0 2021-09-24
[Bug]: The
packaging.release.upload
task wasn’t properly exposed externally, even though another task’s docstring referenced it. Fixed.[Bug]: Ensure that the venv used for
packaging.release.test_install
has itspip
upgraded to match the invoking interpreter’s version of same; this avoids common pitfalls where the “inner” pip is a bundled-with-venv, much-older version incapable of modern package installations.[Support]: Overhaul testing and release procedures to use CircleCI & modern Invocations.
2.2.0 2021-09-03
[Feature]: Added the
invocations.environment
module with top-level functions such asin_ci
.[Feature]:
packaging.release.push
, in dry-run mode, now dry-runs itsgit push
subcommand – meaning the subcommand itself is what is “dry-ran”, instead of truly executinggit push --dry-run
– when a CI environment is detected.This prevents spurious errors when the git remote (eg Github) bails out on read-only authentication credentials, which is common within CI systems.
It’s also just not very useful to dry-run a real git push within CI, since almost certainly the commands to generate git objects to get pushed will themselves not have truly run!
[Bug]:
packaging.release.status
(and its use elsewhere, egprepare
) didn’t adequately reload the local project’s version module during its second/final recheck; this causes that check to fail when said version was edited as part of aprepare
run. It now force-reloads said version module.
2.1.0 2021-08-27
[Feature]: Added
twine check
(which validates packaging metadata’slong_description
) as a pre-upload step withinpackaging.release.publish
.This includes some tweaking of
readme_renderer
behavior (used internally by twine) so it correctly spots more malformed RST, as Sphinx does.
[Feature]: Add
packaging.release.push
for pushing Git objects as part of a release.[Feature]: The
packaging.release.all_
task has been expanded to actually do “ALL THE THINGS!!!”, given adry_run
flag, and renamed on the CLI toall
(no trailing underscore).[Feature]:
packaging.release.prepare
grew adry_run
flag to match the rest of its friends.[Feature]: Add Codecov support to
pytest.coverage
.[Feature]: Add
packaging.release.test_install
task and call it just prior to the final step inpackaging.release.upload
(so one doesn’t upload packages which build OK but don’t actually install OK).[Bug]:
pytest.coverage
incorrectly concatenated itsopts
argument to internal options; this has been fixed.[Bug]: Correctly test for
html
report type inside ofpytest.coverage
when deciding whether to runopen
at the end.[Bug]:
packaging.release.publish
missed a spot when it grew “kwargs beat configuration” behavior - theindex
kwarg still got overwritten by the config value, if defined. This has been fixed.[Bug]:
packaging.release.prepare
now generates annotated Git tags instead of lightweight ones. This was a perplexing oversight (Git has always intended annotated tags to be used for release purposes) so we’re considering it a bugfix instead of a backwards incompatible feature change.[Support]:
packaging.release.prepare
now runs its internal status check twice, once at the start (as before) and again at the end (to prove that the actions taken did in fact satisfy needs).[Support]: Rely on Invoke 1.6+ for some of its new features.
2.0.0 2021-01-24
[Feature]: Add a
warnings
kwarg/flag topytest.test
, allowing one to call it with--no-warnings
as an inline ‘alias’ for pytest’s own--disable-warnings
flag.[Bug]: Fix minor display bug causing the
pytest
task module to append a trailing space to the invocation of pytest itself.[Bug]:
release.build
andrelease.publish
had bad kwargs-vs-config logic preventing flags such as--wheel
or--python
from actually working (config defaults always won out, leading to silent ignoring of user input). This has been fixed; config will now only be honored unless the CLI appears to be overriding it.[Bug]:
release.build
’s--clean
flag has been updated:It now honors configuration like the other flags in this task, specifically
packaging.clean
.It now defaults to
False
(rationale: most build operations in the wild tend to assume no cleaning by default, so defaulting to the opposite was sometimes surprising).Warning
This is a backwards incompatible change.
When
True
, it applies to both build and dist directories, instead of just build.Warning
This is a backwards incompatible change.
[Support]: Modify
release
task tree to look atmain
branches in addition tomaster
ones, for “are we on a feature release line or a bugfix one?” calculations, etc.[Support]: Replace some old Python 2.6-compatible syntax bits.
[Support]: Reverse the default value of
release.build
andrelease.publish
)’swheel
argument fromFalse
toTrue
. Included in this change is a new required runtime dependency on thewheel
package.Rationale: at this point in time, most users will be expecting wheels to be available, and not building wheels is likely to be the uncommon case.
Warning
This is a backwards incompatible change.
[Support] #21: Only require
enum34
under Python 2 to prevent it clashing with the stdlibenum
under Python 3. Credit: Alex Gaynor.[Support] #12: Upgrade our packaging manifest so tests (also docs, requirements files, etc) are included in the distribution archives. Thanks to Tomáš Chvátal for the report.
[Support]: Drop Python 3.4 support. We didn’t actually do anything to make the code not work on 3.4, but we’ve removed some 3.4 related runtime (and development) dependency limitations. Our CI will also no longer test on 3.4.
Warning
This is technically a backwards incompatible change.
1.4.0 2018-06-26
[Feature]: Add a
find_opts
argument tochecks.blacken
for improved control over what files get blackened.
1.3.1 2018-06-26
[Bug]:
checks.blacken
had a typo regarding its folder selection argument; the CLI/function arg wasfolder
while the configuration value wasfolders
(plural). It’s been made consistent: the CLI/function argument is nowfolders
.[Bug]: Was missing a ‘hide output’ flag on a subprocess shell call, the result of which was mystery git branch names appearing in the output of
inv release
and friends. Fixed now.
1.2.2 2018-06-26
[Bug]:
checks.blacken
had a typo regarding its folder selection argument; the CLI/function arg wasfolder
while the configuration value wasfolders
(plural). It’s been made consistent: the CLI/function argument is nowfolders
.[Bug]: Was missing a ‘hide output’ flag on a subprocess shell call, the result of which was mystery git branch names appearing in the output of
inv release
and friends. Fixed now.
1.1.1 2018-06-26
[Bug]:
checks.blacken
had a typo regarding its folder selection argument; the CLI/function arg wasfolder
while the configuration value wasfolders
(plural). It’s been made consistent: the CLI/function argument is nowfolders
.[Bug]: Was missing a ‘hide output’ flag on a subprocess shell call, the result of which was mystery git branch names appearing in the output of
inv release
and friends. Fixed now.[Support]: Remove some apparently non-functional
setup.py
logic around conditionally requiringenum34
; it was never getting selected and thus breaking a couple modules that relied on it.enum34
is now a hard requirement like the other semi-optional-but-not-really requirements.
1.0.1 2018-06-26
[Bug]:
checks.blacken
had a typo regarding its folder selection argument; the CLI/function arg wasfolder
while the configuration value wasfolders
(plural). It’s been made consistent: the CLI/function argument is nowfolders
.[Bug]: Was missing a ‘hide output’ flag on a subprocess shell call, the result of which was mystery git branch names appearing in the output of
inv release
and friends. Fixed now.[Support]: Remove some apparently non-functional
setup.py
logic around conditionally requiringenum34
; it was never getting selected and thus breaking a couple modules that relied on it.enum34
is now a hard requirement like the other semi-optional-but-not-really requirements.
1.3.0 2018-06-20
[Feature]: Bump Releases requirement up to 1.6 and leverage its new ability to load Sphinx extensions, in
packaging.release.prepare
(which parses Releases changelogs programmatically). Prior to this, projects which needed extensions to build their doctree would throw errors when using thepackaging.release
module.[Support]: Remove some apparently non-functional
setup.py
logic around conditionally requiringenum34
; it was never getting selected and thus breaking a couple modules that relied on it.enum34
is now a hard requirement like the other semi-optional-but-not-really requirements.
1.2.1 2018-06-18
[Support]: Remove some apparently non-functional
setup.py
logic around conditionally requiringenum34
; it was never getting selected and thus breaking a couple modules that relied on it.enum34
is now a hard requirement like the other semi-optional-but-not-really requirements.
1.2.0 2018-05-22
[Feature]: Add the
checks
module, containingchecks.blacken
which executes the black code formatter. Thanks to Chris Rose.[Feature]: Add ‘missing’ arguments to
pytest.integration
so its signature now largely matchespytest.test
, which it wraps.[Feature]: Break out a generic form of the
travis.sudo-coverage
task intotravis.sudo-run
which can be used for arbitrary commands run under the ssh/sudo capable user generated bytravis.make-sudouser
/travis.make-sshable
.[Feature]: Add
travis.blacken
which wraps the newchecks.blacken
(in diff+check mode, for test output useful for users who cannot themselves simply run black) in addition to performing Travis-oriented Python version checks and pip installation.This is necessary to remove boilerplate around the fact that
black
is not even visible to Python versions less than 3.6.
1.1.0 2018-05-14
[Feature]: Split out the body of the (sadly incomplete)
packaging.release.all
task into the better-namedpackaging.release.prepare
. (all
continues to behave as it did, it just now callsprepare
explicitly.)
1.0.0 2018-05-08
[Feature]: Pre-history / code primarily for internal consumption