| 1 |
========= |
| 2 |
Changelog |
| 3 |
========= |
| 4 |
|
| 5 |
- :release:`2.2.1 <2025-10-10>` |
| 6 |
- :release:`2.1.4 <2025-10-10>` |
| 7 |
- :bug:`1038` (fixed in :issue:`1040`) Python 3.14 tweaked the behavior of |
| 8 |
`fcntl` to raise `SystemError` on buffer overflows, which our interpretation |
| 9 |
of `termios.TIOCGWINSZ` technically was (we care only about the first two |
| 10 |
fields in what is technically a four-field struct with half the fields |
| 11 |
unused). This has been fixed by unpacking all 4 fields and then discarding |
| 12 |
the unused fields during processing. |
| 13 |
|
| 14 |
Thanks to ``@kasium`` for the original bug report and to ``@rathann`` and |
| 15 |
``@BradleyKirton`` for workshopping the patch into its final shape. |
| 16 |
- :release:`2.2.0 <2023-07-12>` |
| 17 |
- :feature:`-` Remove the somewhat inaccurate subclass requirement around |
| 18 |
`~invoke.config.Config`'s ``.clone(into=...)`` constructor call. It was |
| 19 |
broken for certain use cases (such as trying to clone one subclass into a |
| 20 |
sibling subclass, which would yield a ``TypeError``) and is irrelevant if one |
| 21 |
is using the new type annotations. |
| 22 |
- :release:`2.1.3 <2023-06-14>` |
| 23 |
- :bug:`944` After the release of 2.1, package-style task modules started |
| 24 |
looking in the wrong place for project-level config files (inside one's eg |
| 25 |
``tasks/`` dir, instead of *next to* that dir) due to a subtlety in the new |
| 26 |
import/discovery mechanism used. This has been fixed. Thanks to Arnaud V. and |
| 27 |
Hunter Kelly for the reports and to Jesse P. Johnson for initial |
| 28 |
debugging/diagnosis. |
| 29 |
- :release:`2.1.2 <2023-05-15>` |
| 30 |
- :support:`936 backported` Make sure ``py.typed`` is in our packaging |
| 31 |
manifest; without it, users working from a regular installation |
| 32 |
can't perform type checks. Thanks to Nikita Sobolev for catch & patch. |
| 33 |
- :release:`2.1.1 <2023-05-01>` |
| 34 |
- :bug:`934` The `importlib` upgrade in 2.1 had a corner case bug (regarding |
| 35 |
``from . import <submodule>`` functionality within package-like task trees) |
| 36 |
which in turn exposed a false-pass in our test suite. Both have now been |
| 37 |
fixed. Thanks to Greg Meyer and Robert J. Berger for the bug reports. |
| 38 |
- :release:`2.0.1 <2023-04-29>` |
| 39 |
- :bug:`910` Add more rigor around subprocess/runner shutdown to avoid spurious |
| 40 |
exceptions & also fix downstream issues in libraries like Fabric. Reported by |
| 41 |
Orlando Rodríguez. |
| 42 |
- :release:`2.1.0 <2023-04-28>` |
| 43 |
- :support:`675` Implement `importlib` and deprecate `imp` module. Patches |
| 44 |
provided by Jesse P. Johnson |
| 45 |
- :bug:`376 major` Resolve equality comparison bug for non-collections. Patch |
| 46 |
via Jesse P. Johnson |
| 47 |
- :support:`906` Implement type hints and type checking tests with mypy to |
| 48 |
reduce errors and impove code documentation. Patches by Jesse P. Johnson and |
| 49 |
review by Sam Bull. |
| 50 |
- :support:`901 backported` (via :issue:`903`) Tweak test suite ``setup`` |
| 51 |
methods to be named ``setup_method`` so pytest stops whining about it. Patch |
| 52 |
via Jesse P. Johnson. |
| 53 |
- :release:`2.0.0 <2023-01-16>` |
| 54 |
- :support:`-` Remove support for, and imports related to, all Python versions |
| 55 |
less than 3.6 - including Python 2. This also includes updates to vendored |
| 56 |
packages, such as removing ``six`` and upgrading ``lexicon`` to the latest |
| 57 |
version; and also treatment of things like ``Mock`` use within |
| 58 |
`invoke.context.MockContext` (which now expects stdlib's ``unittest.mock`` |
| 59 |
instead of hunting for the old standalone ``mock`` library). |
| 60 |
|
| 61 |
Thanks to various folks for patches related to some of this work, including |
| 62 |
Jesse P. Johnson who supplied multiple PRs whose commits made it in. |
| 63 |
|
| 64 |
.. warning:: |
| 65 |
This change is backwards incompatible in the following scenarios: |
| 66 |
|
| 67 |
- You use Python <3.6. Shouldn't be an issue as we now specify |
| 68 |
``python_requires`` in packaging metadata. |
| 69 |
- You call ``invoke.util.encode_output`` manually for some reason. (This |
| 70 |
became a noop under Python 3, so just...remove it!) |
| 71 |
- You use `invoke.context.MockContext`; its ``repeat`` init kwarg changed |
| 72 |
its default value from ``False`` to ``True``. This probably won't bite |
| 73 |
you, but we mention it just in case you somehow relied upon the legacy |
| 74 |
behavior. |
| 75 |
- You subclass `invoke.runners.Runner` and/or have had to interact with its |
| 76 |
``stop`` or ``stop_timer`` methods. The latter has been merged into the |
| 77 |
former, and if you are overriding ``stop``, you'll want to make sure you |
| 78 |
now call ``super()`` somewhere if you were not already. |
| 79 |
|
| 80 |
- :support:`-` `Task.argspec <invoke.tasks.Task.argspec>` has changed its |
| 81 |
return value; it now returns an `inspect.Signature` derived from that of the |
| 82 |
task's body callable. |
| 83 |
|
| 84 |
.. warning:: |
| 85 |
This change is backwards incompatible if you were using this method |
| 86 |
directly. |
| 87 |
|
| 88 |
- :release:`1.7.3 <2022-09-30>` |
| 89 |
- :support:`- backported` Fix a non-fatal bug in our setup.py |
| 90 |
``long_description`` generation causing 1.7.0-1.7.2 to have malformed |
| 91 |
description text on PyPI. |
| 92 |
- :release:`1.7.2 <2022-09-30>` |
| 93 |
- :bug:`876` Refactor CLI parser instantiation such that the |
| 94 |
``tasks.ignore_unknown_help`` feature (added in 1.7) works when Invoke is run |
| 95 |
in ``--complete`` mode, i.e. in tab-completion scripts. |
| 96 |
- :bug:`-` Fix errors thrown when comparing `~invoke.tasks.Task` objects to |
| 97 |
non-Task objects; such comparisons are now always false. |
| 98 |
- :release:`1.7.1 <2022-05-11>` |
| 99 |
- :bug:`659` Improve behavior under ``nohup``, which causes stdin to become an |
| 100 |
undetectably-unreadable (but otherwise legit) file descriptor. Previously |
| 101 |
this led to `OSError` even if you weren't expecting anything on stdin; we now |
| 102 |
trap this specific case and silently ignore it, allowing execution to |
| 103 |
continue. Thanks to ``@kingkisskill`` for initial report and to Ryan Stoner |
| 104 |
for followup and workshopping. |
| 105 |
- :release:`1.7.0 <2022-03-18>` |
| 106 |
- :feature:`793` Add a new ``tasks.ignore_unknown_help`` config option for |
| 107 |
users who hand their tasks centrally-defined argument help dictionaries; it |
| 108 |
defaults to ``False`` but such users may set it to ``True`` to avoid |
| 109 |
exceptions. Thanks to ``@Allu2`` for the report. |
| 110 |
- :support:`-` Switch to using ``yaml.safe_load`` for loading config files. |
| 111 |
This avoids some warnings under newer PyYAML versions and is also, in a |
| 112 |
shocking twist, more secure. |
| 113 |
- :support:`803` Upgrade our vendored PyYAML from 3.11 to 5.4.1; this should |
| 114 |
both supply a number of security fixes, and address problems loading |
| 115 |
project-level YAML config files under Python 3.10. Fix via Andreas Rammhold. |
| 116 |
- :feature:`845` Env vars explicitly supplied to `~invoke.context.Context.sudo` |
| 117 |
(via its ``env`` kwarg) are now explicitly preserved via ``sudo``'s |
| 118 |
``--preserve-env`` argument. Patch courtesy of Benno Rice. |
| 119 |
- :support:`-` Switch our continuous integration service from Travis-CI to |
| 120 |
Circle-CI, plus related and necessary updates to various administrative |
| 121 |
config files, management tasks and metadata. Including but not limited to: |
| 122 |
|
| 123 |
- Enhanced PyPI-level metadata/links |
| 124 |
- Split out tool config data from ``setup.cfg`` |
| 125 |
- Enhance execution & coverage of unit vs integration tests under CI |
| 126 |
|
| 127 |
.. warning:: |
| 128 |
Due to various factors, this release will **not** include a Python |
| 129 |
2-compatible wheel archive. Users of Python 2 can still install from the |
| 130 |
sdist, and are strongly encouraged to check `the roadmap |
| 131 |
<https://bitprophet.org/projects/#roadmap>`_ as the next release will |
| 132 |
likely be the one that removes Python 2 entirely! |
| 133 |
|
| 134 |
- :release:`1.6.0 <2021-07-09>` |
| 135 |
- :release:`1.5.1 <2021-07-09>` |
| 136 |
- :support:`- backported` (Mostly) alphabetize the list of ``run()`` params in |
| 137 |
the runners API docs. Previously they were an unordered mess. |
| 138 |
- :support:`- backported` Document the ``dry`` keyword argument to ``run``, |
| 139 |
which had been added in version 1.3 but only documented as a CLI flag; it |
| 140 |
does also work as a kwarg (and as a config option). |
| 141 |
- :bug:`751` Don't explode on empty-but-for-whitespace task docstrings. Thanks |
| 142 |
to Matt Hayden for the report & initial patches. |
| 143 |
- :feature:`791` Add a new ``run.echo_format`` configuration option allowing |
| 144 |
control over the format of echoed commands. It defaults to the previously |
| 145 |
hardcoded value (wrap in ANSI bold) and is thus backwards compatible. Thanks |
| 146 |
to David JM Emmett for the patch. |
| 147 |
- :release:`1.5.0 <2020-12-30>` |
| 148 |
- :feature:`454` (also :issue:`577`/:issue:`658`, via |
| 149 |
:issue:`583`/:issue:`681`/:issue:`607`) Allow any string-compatible object to |
| 150 |
be passed to `Context.cd <invoke.context.Context.cd>`, enabling use of eg |
| 151 |
``pathlib.Path`` instances. Thanks to Jimm Domingo for the original report |
| 152 |
and Ludovico Bianchi, Mario César, and Floris Lambrechts for patches. |
| 153 |
- :bug:`409 major` (via :issue:`611`/:issue:`580`) Don't silently discard help |
| 154 |
text for task arguments whose names happen to contain underscores. Reported |
| 155 |
by ``@iago1460``, original patches by Hayden Flinner and Floris Lambrechts. |
| 156 |
- :bug:`398 major` (via :issue:`611`/:issue:`580`) Don't silently ignore task |
| 157 |
help specifiers which don't actually map to the decorated task's arguments |
| 158 |
(eg ``@task(help={"foo": "help for foo"})`` wrapping a task without a ``foo`` |
| 159 |
argument). Reported by Sohaib Farooqi, with original patches by Hayden |
| 160 |
Flinner and Floris Lambrechts. |
| 161 |
- :feature:`197` Allow subcollections to act as the default 'tasks' of their |
| 162 |
parent collections (via the new ``default`` kwarg to |
| 163 |
`~invoke.collection.Collection.add_collection`). This means that nontrivial |
| 164 |
task trees can specify eg "use my test subcollection's default task as the |
| 165 |
global default task" and similar. Thanks to Tye Wang for the request and |
| 166 |
initial patch. |
| 167 |
- :support:`-` Enhanced test coverage in a handful of modules whose coverage |
| 168 |
was under 90%. |
| 169 |
- :feature:`-` `~invoke.context.MockContext` now populates its |
| 170 |
``NotImplementedError`` exception instances (typically raised when a command |
| 171 |
is executed which had no pre-prepared result) with the command string that |
| 172 |
triggered them; this makes it much easier to tell what exactly in a test |
| 173 |
caused the error. |
| 174 |
- :feature:`-` `~invoke.context.MockContext` now accepts a few quality-of-life |
| 175 |
shortcuts as keys and values in its ``run``/``sudo`` arguments: |
| 176 |
|
| 177 |
- Keys may be compiled regular expression objects, as well as strings, and |
| 178 |
will match any calls whose commands match the regex. |
| 179 |
- Values may be ``True`` or ``False`` as shorthand for otherwise empty |
| 180 |
`~invoke.runners.Result` objects with exit codes of ``0`` or ``1`` |
| 181 |
respectively. |
| 182 |
- Values may also be strings, as shorthand for otherwise empty |
| 183 |
`~invoke.runners.Result` objects with those strings given as the |
| 184 |
``stdout`` argument. |
| 185 |
|
| 186 |
- :feature:`441` Add a new ``repeat`` kwarg to `~invoke.context.MockContext` |
| 187 |
which, when True (default: False) causes stored results for its methods to be |
| 188 |
yielded repeatedly instead of consumed. Feature request courtesy of |
| 189 |
``@SwampFalc``. |
| 190 |
- :bug:`- major` Immutable iterable result values handed to |
| 191 |
`~invoke.context.MockContext` would yield errors (due to the use of |
| 192 |
``pop()``). The offending logic has been retooled to be more iterator-focused |
| 193 |
and now works for tuples and etc. |
| 194 |
- :support:`-` Update the :ref:`testing documentation <testing-user-code>` a |
| 195 |
bit: cleaned up existing examples and added new sections for the other |
| 196 |
updates in the 1.5 release. |
| 197 |
- :feature:`700` Automatically populate the ``command`` attribute of |
| 198 |
`~invoke.runners.Result` objects returned by `~invoke.context.MockContext` |
| 199 |
methods, with the command string triggering that result. Previously users had |
| 200 |
to do this by hand or otherwise suffered inaccurate result objects. Thanks to |
| 201 |
``@SwampFalc`` for the report & initial patch. |
| 202 |
- :feature:`-` Upgrade `~invoke.context.MockContext` to wrap its methods in |
| 203 |
``Mock`` objects if the ``(unittest.)mock`` library is importable. This makes |
| 204 |
testing Invoke-using codebases even easier. |
| 205 |
- :release:`1.4.1 <2020-01-29>` |
| 206 |
- :release:`1.3.1 <2020-01-29>` |
| 207 |
- :support:`586 backported` Explicitly strip out ``__pycache__`` (and for good |
| 208 |
measure, ``.py[co]``, which previously we only stripped from the ``tests/`` |
| 209 |
folder) in our ``MANIFEST.in``, since at least some earlier releases |
| 210 |
erroneously included such. Credit to Martijn Pieters for the report and |
| 211 |
Floris Lambrechts for the patch. |
| 212 |
- :bug:`660` Fix an issue with `~invoke.run` & friends having intermittent |
| 213 |
problems at exit time (symptom was typically about the exit code value being |
| 214 |
``None`` instead of an integer; often with an exception trace). Thanks to |
| 215 |
Frank Lazzarini for the report and to the numerous others who provided |
| 216 |
reproduction cases. |
| 217 |
- :bug:`518` Close pseudoterminals opened by the `~invoke.runners.Local` class |
| 218 |
during ``run(..., pty=True)``. Previously, these were only closed |
| 219 |
incidentally at process shutdown, causing file descriptor leakage in |
| 220 |
long-running processes. Thanks to Jonathan Paulson for the report. |
| 221 |
- :release:`1.4.0 <2020-01-03>` |
| 222 |
- :bug:`637 major` A corner case in `~invoke.context.Context.run` caused |
| 223 |
overridden streams to be unused if those streams were also set to be hidden |
| 224 |
(eg ``run(command, hide=True, out_stream=StringIO())`` would result in no |
| 225 |
writes to the ``StringIO`` object). |
| 226 |
|
| 227 |
This has been fixed - hiding for a given stream is now ignored if that stream |
| 228 |
has been set to some non-``None`` (and in the case of ``in_stream``, |
| 229 |
non-``False``) value. |
| 230 |
- :bug:`- major` As part of feature work on :issue:`682`, we noticed that the |
| 231 |
`~invoke.runners.Result` return value from `~invoke.context.Context.run` was |
| 232 |
inconsistent between dry-run and regular modes; for example, the dry-run |
| 233 |
version of the object lacked updated values for ``hide``, ``encoding`` and |
| 234 |
``env``. This has been fixed. |
| 235 |
- :feature:`682` (originally reported as :issue:`194`) Add asynchronous |
| 236 |
behavior to `~invoke.runners.Runner.run`: |
| 237 |
|
| 238 |
- Basic asynchronicity, where the method returns as soon as the subprocess |
| 239 |
has started running, and that return value is an object with methods |
| 240 |
allowing access to the final result. |
| 241 |
- "Disowning" subprocesses entirely, which not only returns immediately but |
| 242 |
also omits background threading, allowing the subprocesses to outlive |
| 243 |
Invoke's own process. |
| 244 |
|
| 245 |
See the updated API docs for the `~invoke.runners.Runner` for details on the |
| 246 |
new ``asynchronous`` and ``disown`` kwargs enabling this behavior. Thanks to |
| 247 |
``@MinchinWeb`` for the original report. |
| 248 |
- :feature:`-` Never accompanied the top-level singleton `~invoke.run` (which |
| 249 |
simply wraps an anonymous `~invoke.context.Context`'s ``run`` method) with |
| 250 |
its logical sibling, `~invoke.sudo` - this has been remedied. |
| 251 |
- :release:`1.3.0 <2019-08-06>` |
| 252 |
- :feature:`324` Add basic dry-run support, in the form of a new |
| 253 |
:option:`--dry` CLI option and matching ``run.dry`` config setting, which |
| 254 |
causes command runners (eg `~invoke.run`, `Context.run |
| 255 |
<invoke.context.Context.run>`) to: |
| 256 |
|
| 257 |
- Act as if the ``echo`` option has been turned on, printing the |
| 258 |
command-to-be-run to stdout; |
| 259 |
- Skip actual subprocess invocation (returning before any of that machinery |
| 260 |
starts running); |
| 261 |
- Return a dummy `~invoke.runners.Result` object with 'blank' values (empty |
| 262 |
stdout/err strings, ``0`` exit code, etc). |
| 263 |
|
| 264 |
This allows quickly seeing what a given task or series of tasks might do, |
| 265 |
without actually running any shell commands (though naturally, any |
| 266 |
state-modifying Python code will still run). |
| 267 |
|
| 268 |
Thanks to Monty Hindman for the feature request and ``@thebjorn`` for the |
| 269 |
initial patch. |
| 270 |
|
| 271 |
- :bug:`384 major` (via :issue:`653`) Modify config file loading so it detects |
| 272 |
missing-file IOErrors via their ``errno`` attribute instead of their string |
| 273 |
rendering (eg ``"No such file"``). This should improve compatibility for |
| 274 |
non-English locales. Thanks to Patrick Massot for the report and Github user |
| 275 |
``@cybiere`` for the patch. |
| 276 |
- :feature:`539` (via :issue:`645`) Add support for command timeouts, i.e. the |
| 277 |
ability to add an upper bound on how long a call to |
| 278 |
`~invoke.context.Context.run` may take to execute. Specifically: |
| 279 |
|
| 280 |
- A ``timeout`` argument to `~invoke.context.Context.run`. |
| 281 |
- The ``timeouts.command`` config setting mapping to that argument. |
| 282 |
- The :option:`-T/--command-timeout <-T>` CLI flag. |
| 283 |
|
| 284 |
Thanks to Israel Fruchter for the request & an early version of the patchset. |
| 285 |
- :bug:`552 major` (also :issue:`553`) Add a new `~invoke.runners.Runner` |
| 286 |
method, `~invoke.runners.Runner.close_proc_stdin`, and call it when standard |
| 287 |
input processing detects an EOF. Without this, subprocesses that read their |
| 288 |
stdin until EOF would block forever, hanging the program. Thanks to |
| 289 |
``@plockc`` for the report & initial patch. |
| 290 |
|
| 291 |
.. note:: |
| 292 |
This fix only applies when ``pty=False`` (the default); PTYs complicate the |
| 293 |
situation greatly (but also mean the issue is less likely to occur). |
| 294 |
|
| 295 |
- :bug:`557 major` (with assist from :issue:`640`) Fix the |
| 296 |
`~invoke.context.Context.cd` and `~invoke.context.Context.prefix` context |
| 297 |
managers so that ``with cd`` and ``with prefix`` correctly revert their state |
| 298 |
manipulations after they exit, when exceptions occur. Thanks to Jon Walsh and |
| 299 |
Artur Puzio for their respective patches. |
| 300 |
- :bug:`466 major` Update the parsing and CLI-program mechanisms so that all |
| 301 |
core arguments may be given within task CLI contexts; previously this |
| 302 |
functionality only worked for the ``--help`` flag, and other core arguments |
| 303 |
given after task names (such as ``--echo``) were silently ignored. |
| 304 |
- :feature:`-` Allow the configuration system to override which |
| 305 |
`~invoke.executor.Executor` subclass to use when executing tasks (via an |
| 306 |
import-oriented string). |
| 307 |
|
| 308 |
Specifically, it's now possible to alter execution by distributing such a |
| 309 |
subclass alongside, for example, a repository-local config file which sets |
| 310 |
``tasks.executor_class``; previously, this sort of thing required use of |
| 311 |
:ref:`custom binaries <reusing-as-a-binary>`. |
| 312 |
- :release:`1.2.0 <2018-09-13>` |
| 313 |
- :feature:`301` (via :issue:`414`) Overhaul tab completion mechanisms so users |
| 314 |
can :ref:`print a completion script <print-completion-script>` which |
| 315 |
automatically matches the emitting binary's configured names (compared to the |
| 316 |
previous hardcoded scripts, which only worked for ``inv``/``invoke`` by |
| 317 |
default). Thanks to Nicolas Höning for the foundational patchset. |
| 318 |
- :release:`1.1.1 <2018-07-31>` |
| 319 |
- :release:`1.0.2 <2018-07-31>` |
| 320 |
- :bug:`556` (also `fabric/fabric#1823 |
| 321 |
<https://github.com/fabric/fabric/issues/1823>`_) Pre-emptively check for an |
| 322 |
error condition involving an unpicklable config file value (Python config |
| 323 |
files and imported module objects) and raise a useful exception instead of |
| 324 |
allowing a confusing ``TypeError`` to bubble up later. Reported by Pham Cong |
| 325 |
Dinh. |
| 326 |
- :bug:`559` (also `fabric/fabric#1812 |
| 327 |
<https://github.com/fabric/fabric/issues/1812>`_) Modify how |
| 328 |
`~invoke.runners.Runner` performs stdin terminal mode changes, to avoid |
| 329 |
incorrect terminal state restoration when run concurrently (which could lead |
| 330 |
to things like terminal echo becoming disabled after the Python process |
| 331 |
exits). |
| 332 |
|
| 333 |
Thanks to Adam Jensen and Nick Timkovich for the detailed bug reports & |
| 334 |
reproduction assistance. |
| 335 |
- :release:`1.1.0 <2018-07-12>` |
| 336 |
- :release:`1.0.1 <2018-07-12>` |
| 337 |
- :feature:`-` Enhance `~invoke.tasks.Call` with a new method |
| 338 |
(``clone_data``) and new kwarg to an existing method (``clone`` grew |
| 339 |
``with_``) to assist subclassers when extending. |
| 340 |
- :bug:`270` (also :issue:`551`) ``None`` values in config levels (most |
| 341 |
commonly caused by empty configuration files) would raise ``AttributeError`` |
| 342 |
when `~invoke.config.merge_dicts` was used to merge config levels together. |
| 343 |
This has been fixed. Thanks to Tyler Hoffman and Vlad Frolov for the reports. |
| 344 |
- :feature:`-` Refactor `~invoke.tasks.Call` internals slightly, exposing some |
| 345 |
previously internal logic as the ``clone_data`` method; this is useful for |
| 346 |
client codebases when extending `~invoke.tasks.Call` and friends. |
| 347 |
- :feature:`-` Remove overzealous argument checking in `@task |
| 348 |
<invoke.tasks.task>`, instead just handing any extra kwargs into the task |
| 349 |
class constructor. The high level behavior for truly invalid kwargs is the |
| 350 |
same (``TypeError``) but now extending codebases can add kwargs to their |
| 351 |
versions of ``@task`` without issue. |
| 352 |
- :feature:`-` Add a ``klass`` kwarg to `@task <invoke.tasks.task>` to allow |
| 353 |
extending codebases the ability to create their own variants on |
| 354 |
``@task``/``Task``. |
| 355 |
- :bug:`-` Fix up the ``__repr__`` of `~invoke.tasks.Call` to reference dynamic |
| 356 |
class name instead of hardcoding ``"Call"``; this allows subclasses' |
| 357 |
``__repr__`` output to be correct instead of confusing. |
| 358 |
- :support:`- backported` Fixed some inaccuracies in the API docs around |
| 359 |
`~invoke.executor.Executor` and its ``core`` kwarg (was erroneously referring |
| 360 |
to `~invoke.parser.context.ParserContext` instead of |
| 361 |
`~invoke.parser.parser.ParseResult`). Includes related cleaning-up of |
| 362 |
docstrings and tests. |
| 363 |
- :support:`- backported` Apply the `black <https://black.readthedocs.io/>`_ |
| 364 |
code formatter to our codebase and our CI configuration. |
| 365 |
- :support:`- backported` Fix some test-suite-only failures preventing |
| 366 |
successful testing on Python 3.7 and PyPy3, and move them out of the 'allowed |
| 367 |
failures' test matrix quarantine now that they pass. |
| 368 |
- :support:`- backported` Implemented some minor missing tests, such as testing |
| 369 |
the ``INVOKE_DEBUG`` low-level env var. |
| 370 |
- :feature:`543` Implemented support for using ``INVOKE_RUNTIME_CONFIG`` env |
| 371 |
var as an alternate method of supplying a runtime configuration file path |
| 372 |
(effectively, an env var based version of using the ``-f``/``--config`` |
| 373 |
option). Feature request via Kevin J. Qiu. |
| 374 |
- :bug:`528` Around Invoke 0.23 we broke the ability to weave in subcollections |
| 375 |
via keyword arguments to `~invoke.collection.Collection`, though it primarily |
| 376 |
manifests as ``NoneType`` related errors during ``inv --list``. This was |
| 377 |
unintentional and has been fixed. Report submitted by Tuukka Mustonen. |
| 378 |
- :bug:`-` As part of solving :issue:`528` we found a related bug, where |
| 379 |
unnamed subcollections also caused issues with ``inv --list |
| 380 |
--list-format=json``. Specifically, `Collection.serialized |
| 381 |
<invoke.collection.Collection.serialized>` sorts subcollections by name, |
| 382 |
which is problematic when that name is ``None``. This is now fixed. |
| 383 |
- :release:`1.0.0 <2018-05-09>` |
| 384 |
- :feature:`-` Added the :ref:`--prompt-for-sudo-password |
| 385 |
<prompt-for-sudo-password>` CLI option for getpass-based up-front prompting |
| 386 |
of a sensitive configuration value. |
| 387 |
- :feature:`-` Updated `~invoke.tasks.Task` to mimic the wrapped function's |
| 388 |
``__module__`` attribute, allowing for better interaction with things like |
| 389 |
Sphinx autodoc that attempt to filter out imported objects from a module. |
| 390 |
- :bug:`- major` Removed an old, unused and untested (but, regrettably, |
| 391 |
documented and public) method that doesn't seem to be much use: |
| 392 |
``invoke.config.Config.paths``. Please reach out if you were actually using |
| 393 |
it and we may consider adding some form of it back. |
| 394 |
|
| 395 |
.. warning:: |
| 396 |
This is a backwards incompatible change if you were using ``Config.paths``. |
| 397 |
|
| 398 |
- :bug:`- major` Tweaked the innards of |
| 399 |
`~invoke.config.Config`/`~invoke.config.DataProxy` to prevent accessing |
| 400 |
properties & other attributes' values during ``__setattr__`` (the code in |
| 401 |
question only needed the names). This should have no noticeable effect on |
| 402 |
user code (besides a marginal speed increase) but fixed some minor test |
| 403 |
coverage issues. |
| 404 |
- :release:`0.23.0 <2018-04-29>` |
| 405 |
- :bug:`- major` Previously, some error conditions (such as invalid task or |
| 406 |
collection names being supplied by the user) printed to standard output, |
| 407 |
instead of standard error. Standard error seems more appropriate here, so |
| 408 |
this has been fixed. |
| 409 |
|
| 410 |
.. warning:: |
| 411 |
This is backwards incompatible if you were explicitly checking the standard |
| 412 |
output of the ``inv[oke]`` program for some of these error messages. |
| 413 |
|
| 414 |
.. warning:: |
| 415 |
If your code is manually raising or introspecting instances of |
| 416 |
`~invoke.exceptions.Exit`, note that its signature has changed from |
| 417 |
``Exit(code=0)`` to ``Exit(message=None, code=None)``. (Thus, this will |
| 418 |
only impact you if you were calling its constructor instead of raising the |
| 419 |
class object itself.) |
| 420 |
|
| 421 |
- :bug:`- major` `~invoke.collection.Collection` had some minor bugs or |
| 422 |
oversights in how it responds to things like ``repr()``, ``==``; boolean |
| 423 |
behavior; how docstrings appear when created from a Python module; etc. All |
| 424 |
are now fixed. If you're not sure whether this affects you, it does not :) |
| 425 |
- :bug:`- major` Integer-type CLI arguments were not displaying placeholder |
| 426 |
text in ``--help`` output (i.e. they appeared as ``--myint`` instead of |
| 427 |
``--myint=INT``.) This has been fixed. |
| 428 |
- :feature:`33` Overhaul task listing (formerly just a simple, boolean |
| 429 |
``--list``) to make life easier for users with nontrivial task trees: |
| 430 |
|
| 431 |
- Limit display to a specific namespace by giving an optional argument to |
| 432 |
``--list``, e.g. ``--list build``; |
| 433 |
- Additional output formats besides the default (now known as ``flat``) such |
| 434 |
as a nested view with ``--list-format nested`` or script-friendly output |
| 435 |
with ``--list-format json``. |
| 436 |
- The default ``flat`` format now sorts a bit differently - the previous |
| 437 |
algorithm would break up trees of tasks. |
| 438 |
- Limit listing depth, so it's easier to view only the first level or two |
| 439 |
(i.e. the overall namespaces) of a large tree, e.g. ``--list --list-depth |
| 440 |
1``; |
| 441 |
|
| 442 |
Thanks to the many users who submitted various requests under this ticket's |
| 443 |
umbrella, and to Dave Burkholder in particular for detailed use case analysis |
| 444 |
& feedback. |
| 445 |
|
| 446 |
- :support:`-` (partially re: :issue:`33`) Renamed the ``--root`` CLI flag to |
| 447 |
``--search-root``, partly for clarity (:issue:`33` will be adding namespace |
| 448 |
display-root related flags, which would make ``--root`` ambiguous) and partly |
| 449 |
for consistency with the config option, which was already named |
| 450 |
``search_root``. (The short version of the flag, ``-r``, is unchanged.) |
| 451 |
|
| 452 |
.. warning:: |
| 453 |
This is a backwards incompatible change. To fix, simply use |
| 454 |
``--search-root`` anywhere you were previously using ``--root``. |
| 455 |
- :bug:`516 major` Remove the CLI parser ambiguity rule regarding flag-like |
| 456 |
tokens which are seen after an optional-value flag (e.g. ``inv task |
| 457 |
--optionally-takes-a-value --some-other-flag``.) Previously, any flag-like |
| 458 |
value in such a spot was considered ambiguous and raised a |
| 459 |
`~invoke.exceptions.ParseError`. Now, the surrounding parse context is used |
| 460 |
to resolve the ambiguity, and no error is raised. |
| 461 |
|
| 462 |
.. warning:: |
| 463 |
This behavior is backwards incompatible, but only if you had the minority |
| 464 |
case where users frequently *and erroneously* give otherwise-legitimate |
| 465 |
flag-like values to optional-value arguments, and you rely on the parse |
| 466 |
errors to notify them of their mistake. (If you don't understand what this |
| 467 |
means, don't worry, you almost certainly don't need to care!) |
| 468 |
|
| 469 |
- :support:`515` Ported the test suite from `spec |
| 470 |
<https://github.com/bitprophet/spec>`_ (`nose |
| 471 |
<https://nose.readthedocs.io>`_) to `pytest-relaxed |
| 472 |
<https://github.com/bitprophet/pytest-relaxed>`_ (`pytest |
| 473 |
<https://pytest.org>`_) as pytest basically won the test-runner war against |
| 474 |
nose & has greater mindshare, more shiny toys, etc. |
| 475 |
- :support:`-` Rename ``invoke.platform`` to ``invoke.terminals``; it was |
| 476 |
inadvertently shadowing the ``platform`` standard library builtin module. |
| 477 |
This was not causing any bugs we are aware of, but it is still poor hygiene. |
| 478 |
|
| 479 |
.. warning:: |
| 480 |
This change is technically backwards incompatible. We don't expect many |
| 481 |
users import ``invoke.platform`` directly, but if you are, take note. |
| 482 |
|
| 483 |
- :bug:`- major` (partially re: :issue:`449`) Update error message around |
| 484 |
missing positional arguments so it actually lists them. Includes a minor |
| 485 |
tweak to the API of `~invoke.parser.context.ParserContext`, namely changing |
| 486 |
``needs_positional_arguments`` (bool) to ``missing_positional_arguments`` |
| 487 |
(list). |
| 488 |
- :release:`0.22.1 <2018-01-29>` |
| 489 |
- :bug:`342` Accidentally hardcoded ``Collection`` instead of ``cls`` in |
| 490 |
`Collection.from_module <invoke.collection.Collection.from_module>` (an |
| 491 |
alternate constructor and therefore a classmethod.) This made it rather hard |
| 492 |
to properly subclass `~invoke.collection.Collection`. Report and initial |
| 493 |
patch courtesy of Luc Saffre. |
| 494 |
- :support:`433 backported` Add -dev and -nightly style Python versions to our |
| 495 |
Travis builds. Thanks to ``@SylvainDe`` for the contribution. |
| 496 |
- :bug:`437` When merging configuration levels together (which uses |
| 497 |
`copy.copy` by default), pass file objects by reference so they don't get |
| 498 |
closed. Catch & patch by Paul Healy. |
| 499 |
- :support:`469 backported` Fix up the :ref:`doc/example |
| 500 |
<customizing-config-defaults>` re: subclassing `~invoke.config.Config`. |
| 501 |
Credit: ``@Aiky30``. |
| 502 |
- :bug:`488` Account for additional I/O related ``OSError`` error strings |
| 503 |
when attempting to capture only this specific subtype of error. This should |
| 504 |
fix some issues with less common libc implementations such as ``musl`` (as |
| 505 |
found on e.g. Alpine Linux.) Thanks to Rajitha Perera for the report. |
| 506 |
- :release:`0.22.0 <2017-11-29>` |
| 507 |
- :bug:`407 major` (also :issue:`494`, :issue:`67`) Update the default value of |
| 508 |
the ``run.shell`` config value so that it reflects a Windows-appropriate |
| 509 |
value (specifically, the ``COMSPEC`` env var or a fallback of ``cmd.exe``) on |
| 510 |
Windows platforms. This prevents Windows users from being forced to always |
| 511 |
ship around configuration-level overrides. |
| 512 |
|
| 513 |
Thanks to Maciej 'maQ' Kusz for the original patchset, and to ``@thebjorn`` |
| 514 |
and Garrett Jenkins for providing lots of feedback. |
| 515 |
- :bug:`- major` Iterable-type CLI args were actually still somewhat broken & |
| 516 |
were 'eating' values after themselves in the parser stream (thus e.g. |
| 517 |
preventing parsing of subsequent tasks or flags.) This has been fixed. |
| 518 |
- :support:`364` Drop Python 2.6 and Python 3.3 support, as these versions now |
| 519 |
account for only very low percentages of the userbase and are unsupported (or |
| 520 |
about to be unsupported) by the rest of the ecosystem, including ``pip``. |
| 521 |
|
| 522 |
This includes updating documentation & packaging metadata as well as taking |
| 523 |
advantage of basic syntax additions like set literals/comprehensions (``{1, |
| 524 |
2, 3}`` instead of ``set([1, 2, 3])``) and removing positional string |
| 525 |
argument specifiers (``"{}".format(val)`` instead of ``"{0}".format(val)``). |
| 526 |
|
| 527 |
- :release:`0.21.0 <2017-09-18>` |
| 528 |
- :feature:`132` Implement 'iterable' and 'incrementable' CLI flags, allowing |
| 529 |
for invocations like ``inv mytask --listy foo --listy bar`` (resulting in a |
| 530 |
call like ``mytask(listy=['foo', 'bar'])``) or ``inv mytask -vvv`` (resulting |
| 531 |
in e.g. ``mytask(verbose=3)``. Specifically, these require use of the new |
| 532 |
:ref:`iterable <iterable-flag-values>` and :ref:`incrementable |
| 533 |
<incrementable-flag-values>` arguments to `@task <invoke.tasks.task>` - see |
| 534 |
those links to the conceptual docs for details. |
| 535 |
- :release:`0.20.4 <2017-08-14>` |
| 536 |
- :bug:`-` The behavior of `Config <invoke.config.Config>` when ``lazy=True`` |
| 537 |
didn't match that described in the API docs, after the recent updates to its |
| 538 |
lifecycle. (Specifically, any config data given to the constructor was not |
| 539 |
visible in the resulting instance until ``merge()`` was explicitly called.) |
| 540 |
This has been fixed, along with other related minor issues. |
| 541 |
- :release:`0.20.3 <2017-08-04>` |
| 542 |
- :bug:`467` (Arguably also a feature, but since it enables behavior users |
| 543 |
clearly found intuitive, we're considering it a bug.) Split up the parsing |
| 544 |
machinery of `Program <invoke.program.Program>` and pushed the `Collection |
| 545 |
<invoke.collection.Collection>`-making out of `Loader |
| 546 |
<invoke.loader.Loader>`. Combined, this allows us to honor the project-level |
| 547 |
config file *before* the second (task-oriented) CLI parsing step, instead of |
| 548 |
after. |
| 549 |
|
| 550 |
For example, this means you can turn off ``auto_dash_names`` in your |
| 551 |
per-project configs and not only in your system or user configs. |
| 552 |
|
| 553 |
Report again courtesy of Luke Orland. |
| 554 |
|
| 555 |
.. warning:: |
| 556 |
This is a backwards incompatible change *if* you were subclassing and |
| 557 |
overriding any of the affected methods in the ``Program`` or ``Loader`` |
| 558 |
classes. |
| 559 |
|
| 560 |
- :release:`0.20.2 <2017-08-02>` |
| 561 |
- :bug:`465` The ``tasks.auto_dash_names`` config option added in ``0.20.0`` |
| 562 |
wasn't being fully honored when set to ``False``; this has been fixed. Thanks |
| 563 |
to Luke Orland for the report. |
| 564 |
- :release:`0.20.1 <2017-07-27>` |
| 565 |
- :bug:`-` Fix a broken ``six.moves`` import within ``invoke.util``; was |
| 566 |
causing ``ImportError`` in environments without an external copy of ``six`` |
| 567 |
installed. |
| 568 |
|
| 569 |
The dangers of one's local and CI environments all pulling down packages that |
| 570 |
use ``six``! It's everywhere! |
| 571 |
- :release:`0.20.0 <2017-07-27>` |
| 572 |
- :feature:`-` (required to support :issue:`310` and :issue:`329`) Break up the |
| 573 |
`~invoke.config.Config` lifecycle some more, allowing it to gradually load |
| 574 |
configuration vectors; this allows the CLI machinery |
| 575 |
(`~invoke.executor.Executor`) to honor configuration settings from config |
| 576 |
files which impact how CLI parsing and task loading behaves. |
| 577 |
|
| 578 |
Specifically, this adds more public ``Config.load_*`` methods, which in |
| 579 |
tandem with the ``lazy`` kwarg to ``__init__`` (formerly ``defer_post_init``, |
| 580 |
see below) allow full control over exactly when each config level is loaded. |
| 581 |
|
| 582 |
.. warning:: |
| 583 |
This change may be backwards incompatible if you were using or subclassing |
| 584 |
the `~invoke.config.Config` class in any of the following ways: |
| 585 |
|
| 586 |
- If you were passing ``__init__`` kwargs such as ``project_home`` or |
| 587 |
``runtime_path`` and expecting those files to auto-load, they no longer |
| 588 |
do; you must explicitly call `~invoke.config.Config.load_project` and/or |
| 589 |
`~invoke.config.Config.load_runtime` explicitly. |
| 590 |
- The ``defer_post_init`` keyword argument to ``Config.__init__`` has been |
| 591 |
renamed to ``lazy``, and controls whether system/user config files are |
| 592 |
auto-loaded. |
| 593 |
- ``Config.post_init`` has been removed, in favor of explicit/granular use |
| 594 |
of the ``load_*`` family of methods. |
| 595 |
- All ``load_*`` methods now call ``Config.merge`` automatically by default |
| 596 |
(previously, merging was deferred to the end of most config related |
| 597 |
workflows.) |
| 598 |
|
| 599 |
This should only be a problem if your config contents are extremely large |
| 600 |
(it's an entirely in-memory dict-traversal operation) and can be avoided |
| 601 |
by specifying ``merge=False`` to any such method. (Note that you must, at |
| 602 |
some point, call `~invoke.config.Config.merge` in order for the config |
| 603 |
object to work normally!) |
| 604 |
|
| 605 |
- :feature:`310` (also :issue:`455`, :issue:`291`) Allow configuring collection |
| 606 |
root directory & module name via configuration files (previously, they were |
| 607 |
only configurable via CLI flags or generating a custom |
| 608 |
`~invoke.program.Program`.) |
| 609 |
- :feature:`329` All task and collection names now have underscores turned into |
| 610 |
dashes automatically, as task parameters have been for some time. This |
| 611 |
impacts ``--list``, ``--help``, and of course the parser. For details, see |
| 612 |
:ref:`dashes-vs-underscores`. |
| 613 |
|
| 614 |
This behavior is controlled by a new config setting, |
| 615 |
``tasks.auto_dash_names``, which can be set to ``False`` to go back to the |
| 616 |
classic behavior. |
| 617 |
|
| 618 |
Thanks to Alexander Artemenko for the initial feature request. |
| 619 |
- :bug:`396 major` ``Collection.add_task(task, aliases=('other', 'names')`` was |
| 620 |
listed in the conceptual documentation, but not implemented (technically, it |
| 621 |
was removed at some point and never reinstated.) It has been (re-)added and |
| 622 |
now exists. Thanks to ``@jenisys`` for the report. |
| 623 |
|
| 624 |
.. warning:: |
| 625 |
This technically changes argument order for `Collection.add_task |
| 626 |
<invoke.collection.Collection.add_task>`, so be aware if you were using |
| 627 |
positional arguments! |
| 628 |
|
| 629 |
- :bug:`- major` Display of hidden subprocess output when a command |
| 630 |
execution failed (end-of-session output starting with ``Encountered a bad |
| 631 |
command exit code!``) was liable to display encoding errors (e.g. ``'ascii' |
| 632 |
codec can't encode character ...``) when that output was not |
| 633 |
ASCII-compatible. |
| 634 |
|
| 635 |
This problem was previously solved for *non-hidden* (mirrored) subprocess |
| 636 |
output, but the fix (encode the data with the local encoding) had not been |
| 637 |
applied to exception display. Now it's applied in both cases. |
| 638 |
- :feature:`322` Allow users to completely disable mirroring of stdin to |
| 639 |
subprocesses, by specifying ``False`` for the ``run.in_stream`` config |
| 640 |
setting and/or keyword argument. |
| 641 |
|
| 642 |
This can help prevent problems when running Invoke under systems that have no |
| 643 |
useful standard input and which otherwise defeat our pty/fileno related |
| 644 |
detection. |
| 645 |
- :release:`0.19.0 <2017-06-19>` |
| 646 |
- :feature:`-` Add `MockContext.set_result_for |
| 647 |
<invoke.context.MockContext.set_result_for>` to allow massaging a mock |
| 648 |
Context's configured results after instantiation. |
| 649 |
- :release:`0.18.1 <2017-06-07>` |
| 650 |
- :bug:`-` Update Context internals re: command execution & configuration of |
| 651 |
runner subclasses, to work better in client libraries such as Fabric 2. |
| 652 |
|
| 653 |
.. note:: |
| 654 |
If you were using the undocumented ``runner`` configuration value added |
| 655 |
in :issue:`446`, it is now ``runners.local``. |
| 656 |
|
| 657 |
.. warning:: |
| 658 |
This change modifies the internals of methods like |
| 659 |
`~invoke.context.Context.run` and `~invoke.context.Context.sudo`; users |
| 660 |
maintaining their own subclasses should be aware of possible breakage. |
| 661 |
|
| 662 |
- :release:`0.18.0 <2017-06-02>` |
| 663 |
- :feature:`446` Implement `~invoke.context.Context.cd` and |
| 664 |
`~invoke.context.Context.prefix` context managers (as methods on the |
| 665 |
not-that-one-the-other-one `~invoke.context.Context` class.) These are based |
| 666 |
on similar functionality in Fabric 1.x. Credit: Ryan P Kilby. |
| 667 |
- :support:`448` Fix up some config-related tests that have been failing on |
| 668 |
Windows for some time. Thanks to Ryan P Kilby. |
| 669 |
- :feature:`205` Allow giving core flags like ``--help`` after tasks to trigger |
| 670 |
per-task help. Previously, only ``inv --help taskname`` worked. |
| 671 |
|
| 672 |
.. note:: |
| 673 |
Tasks with their own ``--help`` flags won't be able to leverage this |
| 674 |
feature - the parser will still interpret the flag as being per-task and |
| 675 |
not global. This may change in the future to simply throw an exception |
| 676 |
complaining about the ambiguity. (Feedback welcome.) |
| 677 |
|
| 678 |
- :feature:`444` Add support for being used as ``python -m invoke <args>`` on |
| 679 |
Python 2.7 and up. Thanks to Pekka Klärck for the feature request. |
| 680 |
- :release:`0.17.0 <2017-05-05>` |
| 681 |
- :bug:`439 major` Avoid placing stdin into bytewise read mode when it looks |
| 682 |
like Invoke has been placed in the background by a shell's job control |
| 683 |
system; doing so was causing the shell to pause the Invoke process (e.g. with |
| 684 |
a message like ``suspended (tty output)``.) Reported by Tuukka Mustonen. |
| 685 |
- :bug:`425 major` Fix ``Inappropriate ioctl for device`` errors (usually |
| 686 |
``OSError``) when running Invoke without a tty-attached stdin (i.e. when run |
| 687 |
under 'headless' continuous integration systems or simply as e.g. ``inv |
| 688 |
sometask < /dev/null`` (redirected stdin.) Thanks to Javier Domingo Cansino |
| 689 |
for the report & Tuukka Mustonen for troubleshooting assistance. |
| 690 |
- :feature:`-` Add a ``user`` kwarg & config parameter to |
| 691 |
`Context.sudo <invoke.context.Context.sudo>`, which corresponds roughly to |
| 692 |
``sudo -u <user> <command>``. |
| 693 |
- :bug:`440 major` Make sure to skip a call to ``struct``/``ioctl`` on Windows |
| 694 |
platforms; otherwise certain situations inside ``run`` calls would trigger |
| 695 |
import errors. Thanks to ``@chrisc11`` for the report. |
| 696 |
- :release:`0.16.3 <2017-04-18>` |
| 697 |
- :bug:`-` Even more setup.py related tomfoolery. |
| 698 |
- :release:`0.16.2 <2017-04-18>` |
| 699 |
- :bug:`-` Deal with the fact that PyPI's rendering of Restructured Text has no |
| 700 |
idea about our fancy new use of Sphinx's doctest module. Sob. |
| 701 |
- :release:`0.16.1 <2017-04-18>` |
| 702 |
- :bug:`-` Fix a silly typo preventing proper rendering of the packaging |
| 703 |
``long_description`` (causing an effectively blank PyPI description.) |
| 704 |
- :release:`0.16.0 <2017-04-18>` |
| 705 |
- :feature:`232` Add support for ``.yml``-suffixed config files (in addition to |
| 706 |
``.yaml``, ``.json`` and ``.py``.) Thanks to Matthias Lehmann for the |
| 707 |
original request & Greg Back for an early patch. |
| 708 |
- :feature:`418` Enhance ability of client libraries to override config |
| 709 |
filename prefixes. This includes modifications to related functionality, such |
| 710 |
as how env var prefixes are configured. |
| 711 |
|
| 712 |
.. warning:: |
| 713 |
**This is a backwards incompatible change** if: |
| 714 |
|
| 715 |
- you were relying on the ``env_prefix`` keyword argument to |
| 716 |
`Config.__init__ <invoke.config.Config.__init__>`; it is now the |
| 717 |
``prefix`` or ``env_prefix`` class attribute, depending. |
| 718 |
- or the kwarg/attribute of the same name in `Program.__init__ |
| 719 |
<invoke.program.Program.__init__>`; you should now be subclassing |
| 720 |
``Config`` and using its ``env_prefix`` attribute; |
| 721 |
- or if you were relying on how standalone ``Config`` objects defaulted to |
| 722 |
having a ``None`` value for ``env_prefix``, and thus loaded env vars |
| 723 |
without an ``INVOKE_`` style prefix. |
| 724 |
|
| 725 |
See new documentation for this functionality at |
| 726 |
:ref:`customizing-config-defaults` for details. |
| 727 |
|
| 728 |
- :feature:`309` Overhaul how task execution contexts/configs are handled, such |
| 729 |
that all contexts in a session now share the same config object, and thus |
| 730 |
user modifications are preserved between tasks. This has been done in a |
| 731 |
manner that should not break things like collection-based config (which may |
| 732 |
still differ from task to task.) |
| 733 |
|
| 734 |
.. warning:: |
| 735 |
**This is a backwards incompatible change** if you were relying on the |
| 736 |
post-0.12 behavior of cloning config objects between each task execution. |
| 737 |
Make sure to investigate if you find tasks affecting one another in |
| 738 |
unexpected ways! |
| 739 |
|
| 740 |
- :support:`-` Fixed some Python 2.6 incompatible string formatting that snuck |
| 741 |
in recently. |
| 742 |
- :feature:`-` Switched the order of the first two arguments of |
| 743 |
`Config.__init__ <invoke.config.Config.__init__>`, so that the ``overrides`` |
| 744 |
kwarg becomes the first positional argument. |
| 745 |
|
| 746 |
This supports the common use case of making a `Config <invoke.config.Config>` |
| 747 |
object that honors the system's core/global defaults; previously, because |
| 748 |
``defaults`` was the first argument, you'd end up replacing those core |
| 749 |
defaults instead of merging with them. |
| 750 |
|
| 751 |
.. warning:: |
| 752 |
**This is a backwards incompatible change** if you were creating custom |
| 753 |
``Config`` objects via positional, instead of keyword, arguments. It should |
| 754 |
have no effect otherwise. |
| 755 |
|
| 756 |
- :feature:`-` `Context.sudo <invoke.context.Context.sudo>` no longer prompts |
| 757 |
the user when the configured sudo password is empty; thus, an empty sudo |
| 758 |
password and a ``sudo`` program configured to require one will result in an |
| 759 |
exception. |
| 760 |
|
| 761 |
The runtime prompting for a missing password was a temporary holdover from |
| 762 |
Fabric v1, and in retrospect is undesirable. We may add it back in as an |
| 763 |
opt-in behavior (probably via subclassing) in the future if anybody misses |
| 764 |
it. |
| 765 |
|
| 766 |
.. warning:: |
| 767 |
**This is a backwards incompatible change**, if you were relying on |
| 768 |
``sudo()`` prompting you for your password (vs configuring it). If you |
| 769 |
*were* doing that, you can simply switch to ``run("sudo <command>")`` and |
| 770 |
respond to the subprocess' sudo prompt by hand instead. |
| 771 |
|
| 772 |
- :feature:`-` `Result <invoke.runners.Result>` and `UnexpectedExit |
| 773 |
<invoke.exceptions.UnexpectedExit>` objects now have a more useful ``repr()`` |
| 774 |
(and in the case of ``UnexpectedExit``, a distinct ``repr()`` from their |
| 775 |
preexisting ``str()``.) |
| 776 |
- :bug:`432 major` Tighten application of IO thread ``join`` timeouts (in `run |
| 777 |
<invoke.runners.Runner.run>`) to only happen when :issue:`351` appears |
| 778 |
actually present. Otherwise, slow/overworked IO threads had a chance of being |
| 779 |
joined before truly reading all data from the subprocess' pipe. |
| 780 |
- :bug:`430 major` Fallback importing of PyYAML when Invoke has been installed |
| 781 |
without its vendor directory, was still trying to import the vendorized |
| 782 |
module names (e.g. ``yaml2`` or ``yaml3`` instead of simply ``yaml``). This |
| 783 |
has been fixed, thanks to Athmane Madjoudj. |
| 784 |
- :release:`0.15.0 <2017-02-14>` |
| 785 |
- :bug:`426 major` `DataProxy <invoke.config.DataProxy>` based classes like |
| 786 |
`Config <invoke.config.Config>` and `Context <invoke.context.Context>` didn't |
| 787 |
like being `pickled <pickle>` or `copied <copy.copy>` and threw |
| 788 |
``RecursionError``. This has been fixed. |
| 789 |
- :feature:`-` `Config <invoke.config.Config>`'s internals got cleaned up |
| 790 |
somewhat; end users should not see much of a difference, but advanced |
| 791 |
users or authors of extension code may notice the following: |
| 792 |
|
| 793 |
- Direct modification of config data (e.g. ``myconfig.section.subsection.key |
| 794 |
= 'value'`` in user/task code) is now stored in its own config 'level'/data |
| 795 |
structure; previously such modifications simply mutated the central, |
| 796 |
'merged' config cache. This makes it much easier to determine where a final |
| 797 |
observed value came from, and prevents accidental data loss. |
| 798 |
- Ditto for deleted values. |
| 799 |
- Merging/reconciliation of the config levels now happens automatically when |
| 800 |
data is loaded or modified, which not only simplifies the object's |
| 801 |
lifecycle a bit but allows the previous change to function without |
| 802 |
requiring users to call ``.merge()`` after every modification. |
| 803 |
|
| 804 |
- :bug:`- major` Python 3's hashing rules differ from Python 2, specifically: |
| 805 |
|
| 806 |
A class that overrides ``__eq__()`` and does not define ``__hash__()`` will |
| 807 |
have its ``__hash__()`` implicitly set to None. |
| 808 |
|
| 809 |
`Config <invoke.config.Config>` (specifically, its foundational class |
| 810 |
`DataProxy <invoke.config.DataProxy>`) only defined ``__eq__`` which, |
| 811 |
combined with the above behavior, meant that ``Config`` objects appeared to |
| 812 |
hash successfully on Python 2 but yielded ``TypeErrors`` on Python 3. |
| 813 |
|
| 814 |
This has been fixed by explicitly setting ``__hash__ = None`` so that the |
| 815 |
objects do not hash on either interpreter (there are no good immutable |
| 816 |
attributes by which to define hashability). |
| 817 |
- :bug:`- major` Configuration keys named ``config`` were inadvertently |
| 818 |
exposing the internal dict representation of the containing config object, |
| 819 |
instead of displaying the actual value stored in that key. (Thus, a set |
| 820 |
config of ``mycontext.foo.bar.config`` would act as if it was the key/value |
| 821 |
contents of the ``mycontext.foo.bar`` subtree.) This has been fixed. |
| 822 |
- :feature:`421` Updated `Config.clone <invoke.config.Config.clone>` (and a few |
| 823 |
other related areas) to replace use of `copy.deepcopy` with a less-rigorous |
| 824 |
but also less-likely-to-explode recursive dict copier. This prevents |
| 825 |
frustrating ``TypeErrors`` while still preserving barriers between different |
| 826 |
tasks' configuration values. |
| 827 |
- :feature:`-` `Config.clone <invoke.config.Config.clone>` grew a new ``into`` |
| 828 |
kwarg allowing client libraries with their own `~invoke.config.Config` |
| 829 |
subclasses to easily "upgrade" vanilla Invoke config objects into their local |
| 830 |
variety. |
| 831 |
- :bug:`419 major` Optional parser arguments had a few issues: |
| 832 |
|
| 833 |
- The :ref:`conceptual docs about CLI parsing <optional-values>` mentioned |
| 834 |
them, but didn't actually show via example how to enable the feature, |
| 835 |
implying (incorrectly) that they were active always by default. An example |
| 836 |
has been added. |
| 837 |
- Even when enabled, they did not function correctly when their default |
| 838 |
values were of type ``bool``; in this situation, trying to give a value (vs |
| 839 |
just giving the flag name by itself) caused a parser error. This has been |
| 840 |
fixed. |
| 841 |
|
| 842 |
Thanks to ``@ouroboroscoding`` for the report. |
| 843 |
- :support:`204` (via :issue:`412`) Fall back to globally-installed copies of |
| 844 |
our vendored dependencies, if the import from the ``vendor`` tree fails. In |
| 845 |
normal situations this won't happen, but it allows advanced users or |
| 846 |
downstream maintainers to nuke ``vendor/`` and prefer explicitly installed |
| 847 |
packages of e.g. ``six``, ``pyyaml`` or ``fluidity``. Thanks to Athmane |
| 848 |
Madjoudj for the patch. |
| 849 |
- :bug:`- major` Fix configuration framework such that nested or dict-like |
| 850 |
config values may be compared with regular dicts. Previously, doing so caused |
| 851 |
an ``AttributeError`` (as regular dicts lack a ``.config``). |
| 852 |
- :bug:`413 major` Update behavior of ``DataProxy`` (used within |
| 853 |
`~invoke.context.Context` and `~invoke.config.Config`) again, fixing two |
| 854 |
related issues: |
| 855 |
|
| 856 |
- Creating new configuration keys via attribute access wasn't possible: one |
| 857 |
had to do ``config['foo'] = 'bar'`` because ``config.foo = 'bar'`` would |
| 858 |
set a real attribute instead of touching configuration. |
| 859 |
- Supertypes' attributes weren't being considered during the "is this a real |
| 860 |
attribute on ``self``?" test, leading to different behavior between a |
| 861 |
nested config-value-as-attribute and a top-level Context/Config one. |
| 862 |
|
| 863 |
- :release:`0.14.0 <2016-12-05>` |
| 864 |
- :bug:`349 major` Display the string representation of |
| 865 |
`~invoke.exceptions.UnexpectedExit` when handling it inside of |
| 866 |
`~invoke.program.Program` (including regular ``inv``), if any output was |
| 867 |
hidden during the ``run`` that generated it. |
| 868 |
|
| 869 |
Previously, we only exited with the exception's stored exit code, meaning |
| 870 |
failures of ``run(..., hide=True)`` commands were unexpectedly silent. |
| 871 |
(Library-style use of the codebase didn't have this problem, since tracebacks |
| 872 |
aren't muted.) |
| 873 |
|
| 874 |
While implementing this change, we also tweaked the overall display of |
| 875 |
``UnexpectedExit`` so it's a bit more consistent & useful: |
| 876 |
|
| 877 |
- noting "hey, you ran with ``pty=True``, so there's no stderr"; |
| 878 |
- showing only the last 10 lines of captured output in the error message |
| 879 |
(users can, of course, always manually handle the error & access the full |
| 880 |
thing if desired); |
| 881 |
- only showing a given stream when it was not already printed to the user's |
| 882 |
terminal (i.e. if ``hide=False``, no captured output is shown in the error |
| 883 |
text; if ``hide='stdout'``, only stdout is shown in the error text; etc.) |
| 884 |
|
| 885 |
Thanks to Patrick Massot for the original bug report. |
| 886 |
|
| 887 |
- :feature:`-` Expose the (normalized) value of `~invoke.runners.Runner.run`'s |
| 888 |
``hide`` parameter in its return-value `~invoke.runners.Result` objects. |
| 889 |
- :bug:`288 major` Address a bug preventing reuse of Invoke as a custom |
| 890 |
binstub, by moving ``--list`` into the "core args" set of flags present on |
| 891 |
all Invoke-derived binstubs. Thanks to Jordon Mears for catch & patch. |
| 892 |
- :bug:`283 major` Fix the concepts/library docs so the example of an explicit |
| 893 |
``namespace=`` argument correctly shows wrapping an imported task module in a |
| 894 |
`~invoke.collection.Collection`. Thanks to ``@zaiste`` for the report. |
| 895 |
- :bug:`- major` Fix ``DataProxy`` (used within `~invoke.context.Context` and |
| 896 |
`~invoke.config.Config`) so that real attributes and methods which are |
| 897 |
shadowed by configuration keys, aren't proxied to the config during regular |
| 898 |
attribute get/set. (Such config keys are thus required to be accessed via |
| 899 |
dict-style only, or (on `~invoke.context.Context`) via the explicit |
| 900 |
``.config`` attribute.) |
| 901 |
- :bug:`58 major` Work around bugs in ``select()`` when handling subprocess |
| 902 |
stream reads, which was causing poor behavior in many nontrivial interactive |
| 903 |
programs (such as ``vim`` and other fullscreen editors, ``python`` and other |
| 904 |
REPLs/shells, etc). Such programs should now be largely indistinguishable |
| 905 |
from their behavior when run directly from a user's shell. |
| 906 |
- :feature:`406` Update handling of Ctrl-C/``KeyboardInterrupt``, and |
| 907 |
subprocess exit status pass-through, to be more correct than before: |
| 908 |
|
| 909 |
- Submit the interrupt byte sequence ``\x03`` to stdin of all subprocesses, |
| 910 |
instead of sending ``SIGINT``. |
| 911 |
|
| 912 |
- This results in behavior closer to that of truly pressing Ctrl-C when |
| 913 |
running subprocesses directly; for example, interactive programs like |
| 914 |
``vim`` or ``python`` now behave normally instead of prematurely |
| 915 |
exiting. |
| 916 |
- Of course, programs that would normally exit on Ctrl-C will still do |
| 917 |
so! |
| 918 |
|
| 919 |
- The exit statuses of subprocesses run with ``pty=True`` are more rigorously |
| 920 |
checked (using `os.WIFEXITED` and friends), allowing us to surface the real |
| 921 |
exit values of interrupted programs instead of manually assuming exit code |
| 922 |
``130``. |
| 923 |
|
| 924 |
- Typically, this will be exit code ``-2``, but it is system dependent. |
| 925 |
- Other, non-Ctrl-C-driven signal-related exits under PTYs should behave |
| 926 |
better now as well - previously they could appear to exit ``0``! |
| 927 |
|
| 928 |
- Non-subprocess-related ``KeyboardInterrupt`` (i.e. those generated when |
| 929 |
running top level Python code outside of any ``run`` function calls) |
| 930 |
will now trigger exit code ``1``, as that is how the Python interpreter |
| 931 |
typically behaves if you ``KeyboardInterrupt`` it outside of a live |
| 932 |
REPL. |
| 933 |
|
| 934 |
.. warning:: |
| 935 |
These changes are **backwards incompatible** if you were relying on the |
| 936 |
"exits ``130``" behavior added in version 0.13, or on the (incorrect) |
| 937 |
``SIGINT`` method of killing pty-driven subprocesses on Ctrl-C. |
| 938 |
|
| 939 |
- :bug:`- major` Correctly raise ``TypeError`` when unexpected keyword |
| 940 |
arguments are given to `~invoke.runners.Runner.run`. |
| 941 |
- :feature:`-` Add a `~invoke.context.MockContext` class for easier testing of |
| 942 |
user-written tasks and related client code. Includes adding a |
| 943 |
:ref:`conceptual document on how to test Invoke-using code |
| 944 |
<testing-user-code>`. |
| 945 |
- :feature:`-` Update implementation of `~invoke.runners.Result` so it has |
| 946 |
default values for all parameters/attributes. This allows it to be more |
| 947 |
easily used when mocking ``run`` calls in client libraries' tests. |
| 948 |
|
| 949 |
.. warning:: |
| 950 |
This is a backwards incompatible change if you are manually instantiating |
| 951 |
`~invoke.runners.Result` objects with positional arguments: positional |
| 952 |
argument order has changed. (Compare the API docs between versions to see |
| 953 |
exactly how.) |
| 954 |
|
| 955 |
- :feature:`294` Implement `Context.sudo <invoke.context.Context.sudo>`, which |
| 956 |
wraps `~invoke.context.Context.run` inside a ``sudo`` command. It is capable |
| 957 |
of auto-responding to ``sudo``'s password prompt with a configured password, |
| 958 |
and raises a specific exception (`~invoke.exceptions.AuthFailure`) if that |
| 959 |
password is rejected. |
| 960 |
- :feature:`369` Overhaul the autoresponse functionality for `~invoke.run` so |
| 961 |
it's significantly more extensible, both for its own sake and as part of |
| 962 |
implementing :issue:`294` (see its own changelog entry for details). |
| 963 |
|
| 964 |
.. warning:: |
| 965 |
This is a backwards incompatible change: the ``responses`` kwarg to |
| 966 |
``run()`` is now ``watchers``, and accepts a list of |
| 967 |
`~invoke.watchers.StreamWatcher` objects (such as |
| 968 |
`~invoke.watchers.Responder`) instead of a dict. |
| 969 |
|
| 970 |
If you were using ``run(..., responses={'pattern': 'response'}`` |
| 971 |
previously, just update to instead use ``run(..., |
| 972 |
watchers=[Responder('pattern', 'response')])``. |
| 973 |
|
| 974 |
- :bug:`- major` Fix a bug in `Config.clone <invoke.config.Config.clone>` where |
| 975 |
it was instantiating a new ``Config`` instead of a member of the subclass. |
| 976 |
- :release:`0.13.0 <2016-06-09>` |
| 977 |
- :feature:`114` Ripped off the band-aid and removed non-contextualized tasks |
| 978 |
as an option; all tasks must now be contextualized (defined as ``def |
| 979 |
mytask(context, ...)`` - see :ref:`defining-and-running-task-functions`) even |
| 980 |
if not using the context. This simplifies the implementation as well as |
| 981 |
users' conceptual models. Thanks to Bay Grabowski for the patch. |
| 982 |
|
| 983 |
.. warning:: This is a backwards incompatible change! |
| 984 |
|
| 985 |
- :bug:`350 major` (also :issue:`274`, :issue:`241`, :issue:`262`, |
| 986 |
:issue:`242`, :issue:`321`, :issue:`338`) Clean up and reorganize |
| 987 |
encoding-related parts of the code to avoid some of the more common or |
| 988 |
egregious encode/decode errors surrounding clearly non-ASCII-compatible text. |
| 989 |
Bug reports, assistance, feedback and code examples courtesy of Paul Moore, |
| 990 |
Vlad Frolov, Christian Aichinger, Fotis Gimian, Daniel Nunes, and others. |
| 991 |
- :bug:`351 major` Protect against ``run`` deadlocks involving exceptions in |
| 992 |
I/O threads & nontrivial amounts of unread data in the corresponding |
| 993 |
subprocess pipe(s). This situation should now always result in exceptions |
| 994 |
instead of hangs. |
| 995 |
- :feature:`259` (also :issue:`280`) Allow updating (or replacing) subprocess |
| 996 |
shell environments, via the ``env`` and ``replace_env`` kwargs to |
| 997 |
`~invoke.runners.Runner.run`. Thanks to Fotis Gimian for the report, |
| 998 |
``@philtay`` for an early version of the final patch, and Erich Heine & Vlad |
| 999 |
Frolov for feedback. |
| 1000 |
- :feature:`67` Added ``shell`` option to `~invoke.runners.Runner.run`, |
| 1001 |
allowing control of the shell used when invoking commands. Previously, |
| 1002 |
``pty=True`` used ``/bin/bash`` and ``pty=False`` (the default) used |
| 1003 |
``/bin/sh``; the new unified default value is ``/bin/bash``. |
| 1004 |
|
| 1005 |
Thanks to Jochen Breuer for the report. |
| 1006 |
- :bug:`152 major` (also :issue:`251`, :issue:`331`) Correctly handle |
| 1007 |
``KeyboardInterrupt`` during `~invoke.runners.Runner.run`, re: both mirroring |
| 1008 |
the interrupt signal to the subprocess *and* capturing the local exception |
| 1009 |
within Invoke's CLI handler (so there's no messy traceback, just exiting with |
| 1010 |
code ``130``). |
| 1011 |
|
| 1012 |
Thanks to Peter Darrow for the report, and to Mika Eloranta & Máté Farkas for |
| 1013 |
early versions of the patchset. |
| 1014 |
- :support:`319` Fixed an issue resulting from :issue:`255` which |
| 1015 |
caused problems with how we generate release wheels (notably, some releases |
| 1016 |
such as 0.12.1 fail when installing from wheels on Python 2). |
| 1017 |
|
| 1018 |
.. note:: |
| 1019 |
As part of this fix, the next release will distribute individual Python 2 |
| 1020 |
and Python 3 wheels instead of one 'universal' wheel. This change should be |
| 1021 |
transparent to users. |
| 1022 |
|
| 1023 |
Thanks to ``@ojos`` for the initial report and Frazer McLean for some |
| 1024 |
particularly useful feedback. |
| 1025 |
- :release:`0.12.2 <2016-02-07>` |
| 1026 |
- :support:`314 backported` (Partial fix.) Update ``MANIFEST.in`` so source |
| 1027 |
distributions include some missing project-management files (e.g. our |
| 1028 |
internal ``tasks.py``). This makes unpacked sdists more useful for things |
| 1029 |
like running the doc or build tasks. |
| 1030 |
- :bug:`303` Make sure `~invoke.run` waits for its IO worker threads to cleanly |
| 1031 |
exit (such as allowing a ``finally`` block to revert TTY settings) when |
| 1032 |
``KeyboardInterrupt`` (or similar) aborts execution in the main thread. |
| 1033 |
Thanks to Tony S Yu and Máté Farkas for the report. |
| 1034 |
- :release:`0.12.1 <2016-02-03>` |
| 1035 |
- :bug:`308` Earlier changes to TTY detection & its use in determining features |
| 1036 |
such as stdin pass-through, were insufficient to handle edge cases such as |
| 1037 |
nested Invoke sessions or piped stdin to Invoke processes. This manifested as |
| 1038 |
hangs and ``OSError`` messages about broken pipes. |
| 1039 |
|
| 1040 |
The issue has been fixed by overhauling all related code to use more specific |
| 1041 |
and accurate checks (e.g. examining just ``fileno`` and/or just ``isatty``). |
| 1042 |
|
| 1043 |
Thanks to Tuukka Mustonen and Máté Farkas for the report (and for enduring |
| 1044 |
the subsequent flood of the project maintainer's stream-of-consciousness |
| 1045 |
ticket updates). |
| 1046 |
- :bug:`305` (also :issue:`306`) Fix up some test-suite issues causing failures |
| 1047 |
on Windows/Appveyor. Thanks to Paul Moore. |
| 1048 |
- :bug:`289` Handful of issues, all fallout from :issue:`289`, which failed to |
| 1049 |
make it out the door for 0.12.0. More are on the way but these should address |
| 1050 |
blockers for some users: |
| 1051 |
|
| 1052 |
* Windows support for the new stdin replication functionality (this was |
| 1053 |
totally blocking Windows users, as reported in :issue:`302` - sorry!); |
| 1054 |
* Stdin is now mirrored to stdout when no PTY is present, so you can see |
| 1055 |
what you're typing (plus a new `~invoke.runners.Runner.run` option and |
| 1056 |
config param, ``echo_stdin``, allowing user override of this behavior); |
| 1057 |
* Exposed the stdin read loop's sleep time as `Runner.input_sleep |
| 1058 |
<invoke.runners.Runner.input_sleep>`; |
| 1059 |
* Sped up some tests a bit. |
| 1060 |
|
| 1061 |
- :release:`0.12.0 <2016-01-12>` |
| 1062 |
- :bug:`257 major` Fix a RecursionError under Python 3 due to lack of |
| 1063 |
``__deepcopy__`` on `~invoke.tasks.Call` objects. Thanks to Markus |
| 1064 |
Zapke-Gründemann for initial report and Máté Farkas for the patch. |
| 1065 |
- :support:`265` Update our Travis config to select its newer build |
| 1066 |
infrastructure and also run on PyPy3. Thanks to Omer Katz. |
| 1067 |
- :support:`254` Add an ``exclude`` option in our ``setup.py`` so setuptools |
| 1068 |
doesn't try loading our vendored PyYAML's Python 2 sub-package under Python 3 |
| 1069 |
(or vice versa - though all reports were from Python 3 users). Thanks to |
| 1070 |
``@yoshiya0503`` for catch & initial patch. |
| 1071 |
- :feature:`68` Disable Python's bytecode caching by default, as it complicates |
| 1072 |
our typical use case (frequently-changing .py files) and offers little |
| 1073 |
benefit for human-facing startup times. Bytecode caching can be explicitly |
| 1074 |
re-enabled by specifying ``--write-pyc`` at runtime. Thanks to Jochen Breuer |
| 1075 |
for feature request and ``@brutus`` for initial patchset. |
| 1076 |
- :support:`144` Add code-coverage reporting to our CI builds (albeit `CodeCov |
| 1077 |
<https://codecov.io>`_ instead of `coveralls.io <https://coveralls.io>`_). |
| 1078 |
Includes rejiggering our project-specific coverage-generating tasks. Thanks |
| 1079 |
to David Baumgold for the original request/PR and to Justin Abrahms for the |
| 1080 |
tipoff re: CodeCov. |
| 1081 |
- :bug:`297 major` Ignore leading and trailing underscores when turning task |
| 1082 |
arguments into CLI flag names. |
| 1083 |
- :bug:`296 major` Don't mutate ``sys.path`` on collection load if task's |
| 1084 |
parent directory is already on ``sys.path``. |
| 1085 |
- :bug:`295 major` Make sure that `~invoke.run`'s ``hide=True`` also disables |
| 1086 |
echoing. Otherwise, "hidden" helper ``run`` calls will still pollute output |
| 1087 |
when run as e.g. ``invoke --echo ...``. |
| 1088 |
- :feature:`289` (also :issue:`263`) Implement :ref:`autoresponding |
| 1089 |
<autoresponding>` for `~invoke.run`. |
| 1090 |
- :support:`-` Removed official Python 3.2 support; sibling projects also did |
| 1091 |
this recently, it's simply not worth the annoyance given the userbase size. |
| 1092 |
- :feature:`228` (partial) Modified and expanded implementation of |
| 1093 |
`~invoke.executor.Executor`, `~invoke.tasks.Task` and `~invoke.tasks.Call` to |
| 1094 |
make implementing task parameterization easier. |
| 1095 |
- :support:`-` Removed the ``-H`` short flag, leaving just ``--hide``. This was |
| 1096 |
done to avoid conflicts with Fabric's host-oriented ``-H`` flag. Favoritism |
| 1097 |
is real! Apologies. |
| 1098 |
|
| 1099 |
.. warning:: This change is backwards compatible if you used ``-H``. |
| 1100 |
|
| 1101 |
- :feature:`173` Overhauled top level CLI functionality to allow reusing |
| 1102 |
Invoke for distinct binaries, optionally with bundled task namespaces as |
| 1103 |
subcommands. As a side effect, this functionality is now much more extensible |
| 1104 |
to boot. Thanks to Erich Heine for feedback/suggestions during development. |
| 1105 |
|
| 1106 |
.. warning:: |
| 1107 |
This change is backwards incompatible if you imported anything from the |
| 1108 |
``invoke.cli`` module (which is now rearchitected as |
| 1109 |
`~invoke.program.Program`). It should be transparent to everybody else. |
| 1110 |
|
| 1111 |
.. seealso:: :ref:`reusing-as-a-binary` |
| 1112 |
|
| 1113 |
- :bug:`- major` Fixed a bug in the parser where ``invoke --takes-optional-arg |
| 1114 |
avalue --anotherflag`` was incorrectly considering ``--anotherflag`` to be an |
| 1115 |
ambiguity error (as if ``avalue`` had not been given to |
| 1116 |
``--takes-optional-arg``. |
| 1117 |
- :release:`0.11.1 <2015-09-07>` |
| 1118 |
- :support:`- backported` Fix incorrect changelog URL in package metadata. |
| 1119 |
- :release:`0.11.0 <2015-09-07>` |
| 1120 |
- :feature:`-` Add a ``.command`` attribute to `~invoke.runners.Result` to |
| 1121 |
preserve the command executed for post-execution introspection. |
| 1122 |
- :feature:`-` Detect local controlling terminal size |
| 1123 |
(`~invoke.terminals.pty_size`) and apply that information when creating |
| 1124 |
pseudoterminals in `~invoke.run` when ``pty=True``. |
| 1125 |
- :bug:`- major` Display stdout instead of stderr in the ``repr()`` of |
| 1126 |
`~invoke.exceptions.Failure` objects, when a pseudo-terminal was used. |
| 1127 |
Previously, failure display focused on the stderr stream, which is always |
| 1128 |
empty under pseudo-terminals. |
| 1129 |
- :bug:`- major` Correctly handle situations where `sys.stdin` has been |
| 1130 |
replaced with an object lacking ``.fileno`` (e.g., some advanced Python |
| 1131 |
shells, headless code execution tools, etc). Previously, this situation |
| 1132 |
resulted in an ``AttributeError``. |
| 1133 |
- :bug:`- major` Capture & reraise exceptions generated by command execution I/O |
| 1134 |
threads, in the main thread, as a `~invoke.exceptions.ThreadException`. |
| 1135 |
- :feature:`235` Allow custom stream objects to be used in `~invoke.run` calls, |
| 1136 |
to be used instead of the defaults of ``sys.stdout``/``sys.stderr``. |
| 1137 |
|
| 1138 |
.. warning:: |
| 1139 |
This change required a major cleanup/rearchitecture of the command |
| 1140 |
execution implementation. The vendored ``pexpect`` module has been |
| 1141 |
completely removed and the API of the `~invoke.runners.Runner` class has |
| 1142 |
changed dramatically (though **the API for run() itself has not**). |
| 1143 |
|
| 1144 |
Be aware there may be edge-case terminal behaviors which have changed or |
| 1145 |
broken as a result of removing ``pexpect``. Please report these as bugs! We |
| 1146 |
expect to crib small bits of what ``pexpect`` does but need concrete test |
| 1147 |
cases first. |
| 1148 |
|
| 1149 |
- :bug:`234 major` (also :issue:`243`) Preserve task-module load location when |
| 1150 |
creating explicit collections with |
| 1151 |
`~invoke.collection.Collection.from_module`; when this was not done, |
| 1152 |
project-local config files were not loading correctly. Thanks to ``@brutus`` |
| 1153 |
and Jan Willems for initial report & troubleshooting, and to Greg Back for |
| 1154 |
identifying the fix. |
| 1155 |
- :bug:`237 major` Completion output lacked "inverse" flag names (e.g. |
| 1156 |
``--no-myoption`` as a boolean negative version of a defaulting-to-True |
| 1157 |
boolean ``myoption``). This has been corrected. |
| 1158 |
- :bug:`239 major` Completion erroneously presented core flags instead of |
| 1159 |
per-task flags when both are present in the invocation being completed (e.g. |
| 1160 |
``inv --debug my_task -<tab>``). This has been fixed. |
| 1161 |
- :bug:`238 major` (partial fix) Update the ``zsh`` completion script to |
| 1162 |
account for use of the ``--collection`` core flag. |
| 1163 |
- :support:`-` Additional rearranging of ``run``/``Runner`` related concerns |
| 1164 |
for improved subclassing, organization, and use in other libraries, |
| 1165 |
including: |
| 1166 |
|
| 1167 |
* Changed the name of the ``runner`` module to ``runners``. |
| 1168 |
* Moved the top level ``run`` function from its original home in |
| 1169 |
``invoke.runner`` to `invoke.__init__ <invoke>`, to reflect the fact that |
| 1170 |
it's now simply a convenience wrapper around ``Runner``. |
| 1171 |
* Tweaked the implementation of `~invoke.runners.Runner` so it can |
| 1172 |
reference `~invoke.context.Context` objects (useful for anticipated |
| 1173 |
subclasses). |
| 1174 |
|
| 1175 |
.. warning:: |
| 1176 |
These are backwards incompatible changes if your code was doing any imports |
| 1177 |
from the ``invoke.runner`` module (including especially |
| 1178 |
``invoke.runner.run``, which is now only ``invoke.run``). Function |
| 1179 |
signatures have **not** changed. |
| 1180 |
|
| 1181 |
- :support:`224` Add a completion script for the ``fish`` shell, courtesy of |
| 1182 |
Jaime Marquínez Ferrándiz. |
| 1183 |
- :release:`0.10.1 <2015-03-17>` |
| 1184 |
- :support:`- backported` Tweak README to reflect recent(-ish) changes in |
| 1185 |
``pip`` re: users who install the development version via ``pip`` instead of |
| 1186 |
using git. |
| 1187 |
- :release:`0.10.0 <2015-03-17>` |
| 1188 |
- :feature:`104` Add core CLI flag ``--complete`` to support shell tab |
| 1189 |
completion scripts, and add some 'blessed' such scripts for bash (3 and 4) |
| 1190 |
and zsh. Thanks to Ivan Malison and Andrew Roberts for providing discussion & |
| 1191 |
early patchsets. |
| 1192 |
- :support:`-` Reorganize `~invoke.runners.Runner`, `~invoke.runners.Local` and |
| 1193 |
``invoke.runner.run`` for improved distribution of responsibilities & |
| 1194 |
downstream subclassing. |
| 1195 |
|
| 1196 |
.. warning:: |
| 1197 |
This includes backwards incompatible changes to the API signature of most |
| 1198 |
members of the ``invoke.runner`` module, including ``invoke.runner.run``. |
| 1199 |
(However, in the case of ``invoke.runner.run``, the changes are mostly in |
| 1200 |
the later, optional keyword arguments.) |
| 1201 |
|
| 1202 |
- :feature:`219` Fall back to non-PTY command execution in situations where |
| 1203 |
``pty=True`` but no PTY appears present. See `~invoke.runners.Local` for |
| 1204 |
details. |
| 1205 |
- :support:`212` Implement basic linting support using ``flake8``, and apply |
| 1206 |
formatting changes to satisfy said linting. As part of this shakeup, also |
| 1207 |
changed all old-style (``%s``) string formatting to new-style (``{0}``). |
| 1208 |
Thanks to Collin Anderson for the foundational patch. |
| 1209 |
- :support:`215` (also :issue:`213`, :issue:`214`) Tweak tests & configuration |
| 1210 |
sections of the code to include Windows compatibility. Thanks to Paul Moore. |
| 1211 |
- :bug:`201 major` (also :issue:`211`) Replace the old, first-draft gross |
| 1212 |
monkeypatched Popen code used for ``invoke.runner.run`` with a |
| 1213 |
non-monkeypatched approach that works better on non-POSIX platforms like |
| 1214 |
Windows, and also attempts to handle encoding and locale issues more |
| 1215 |
gracefully (meaning: at all gracefully). |
| 1216 |
|
| 1217 |
Specifically, the new approach uses threading instead of ``select.select``, |
| 1218 |
and performs explicit encoding/decoding based on detected or explicitly |
| 1219 |
expressed encodings. |
| 1220 |
|
| 1221 |
Major thanks to Paul Moore for an enormous amount of |
| 1222 |
testing/experimentation/discussion, as well as the bulk of the code changes |
| 1223 |
themselves. |
| 1224 |
|
| 1225 |
.. warning:: |
| 1226 |
The top level ``invoke.runner.run`` function has had a minor signature |
| 1227 |
change: the sixth positional argument used to be ``runner`` and is now |
| 1228 |
``encoding`` (with ``runner`` now being the seventh positional argument). |
| 1229 |
|
| 1230 |
- :feature:`147` Drastically overhaul/expand the configuration system to |
| 1231 |
account for multiple configuration levels including (but not limited to) file |
| 1232 |
paths, environment variables, and Python-level constructs (previously the |
| 1233 |
only option). See :ref:`configuration` for details. Thanks to Erich Heine for |
| 1234 |
his copious feedback on this topic. |
| 1235 |
|
| 1236 |
.. warning:: |
| 1237 |
This is technically a backwards incompatible change, though some existing |
| 1238 |
user config-setting code may continue to work as-is. In addition, this |
| 1239 |
system may see further updates before 1.0. |
| 1240 |
|
| 1241 |
- :bug:`191 major` Bypass ``pexpect``'s automatic command splitting to avoid |
| 1242 |
issues running complex nested/quoted commands under a pty. Credit to |
| 1243 |
``@mijikai`` for noticing the problem. |
| 1244 |
- :bug:`183 major` Task docstrings whose first line started on the same line as |
| 1245 |
the opening quote(s) were incorrectly presented in ``invoke --help <task>``. |
| 1246 |
This has been fixed by using `inspect.getdoc`. Thanks to Pekka Klärck for the |
| 1247 |
catch & suggested fix. |
| 1248 |
- :bug:`180 major` Empty invocation (e.g. just ``invoke`` with no flags or |
| 1249 |
tasks, and when no default task is defined) no longer printed help output, |
| 1250 |
instead complaining about the lack of default task. It now prints help again. |
| 1251 |
Thanks to Brent O'Connor for the catch. |
| 1252 |
- :bug:`175 major` ``autoprint`` did not function correctly for tasks stored |
| 1253 |
in sub-collections; this has been fixed. Credit: Matthias Lehmann. |
| 1254 |
- :release:`0.9.0 <2014-08-26>` |
| 1255 |
- :bug:`165 major` Running ``inv[oke]`` with no task names on a collection |
| 1256 |
containing a default task should (intuitively) have run that default task, |
| 1257 |
but instead did nothing. This has been fixed. |
| 1258 |
- :bug:`167 major` Running the same task multiple times in one CLI session was |
| 1259 |
horribly broken; it works now. Thanks to Erich Heine for the report. |
| 1260 |
- :bug:`119 major` (also :issue:`162`, :issue:`113`) Better handle |
| 1261 |
platform-sensitive operations such as pty size detection or use, either |
| 1262 |
replacing with platform-specific implementations or raising useful |
| 1263 |
exceptions. Thanks to Gabi Davar and (especially) Paul Moore, for feedback & |
| 1264 |
original versions of the final patchset. |
| 1265 |
- :feature:`136` Added the ``autoprint`` flag to |
| 1266 |
`invoke.tasks.Task`/`@task <invoke.tasks.task>`, allowing users to set up |
| 1267 |
tasks which act as both subroutines & "print a result" CLI tasks. Thanks to |
| 1268 |
Matthias Lehmann for the original patch. |
| 1269 |
- :bug:`162 major` Adjust platform-sensitive imports so Windows users don't |
| 1270 |
encounter import-time exceptions. Thanks to Paul Moore for the patch. |
| 1271 |
- :support:`169` Overhaul the Sphinx docs into two trees, one for main project |
| 1272 |
info and one for versioned API docs. |
| 1273 |
- :bug:`- major` Fixed a sub-case of the already-mostly-fixed :issue:`149` so |
| 1274 |
the error message works usefully even with no explicit collection name given. |
| 1275 |
- :release:`0.8.2 <2014-06-15>` |
| 1276 |
- :bug:`149` Print a useful message to stderr when Invoke can't find the |
| 1277 |
requested collection/tasks file, instead of displaying a traceback. |
| 1278 |
- :bug:`145` Ensure a useful message is displayed (instead of a confusing |
| 1279 |
exception) when listing empty task collections. |
| 1280 |
- :bug:`142` The refactored Loader class failed to account for the behavior of |
| 1281 |
`imp.find_module` when run against packages (vs modules) and was exploding at |
| 1282 |
load time. This has been fixed. Thanks to David Baumgold for catch & patch. |
| 1283 |
- :release:`0.8.1 <2014-06-09>` |
| 1284 |
- :bug:`140` Revert incorrect changes to our ``setup.py`` regarding detection |
| 1285 |
of sub-packages such as the vendor tree & the parser. Also add additional |
| 1286 |
scripting to our Travis-CI config to catch this class of error in future. |
| 1287 |
Thanks to Steven Loria and James Cox for the reports. |
| 1288 |
- :release:`0.8.0 <2014-06-08>` |
| 1289 |
- :feature:`135` (also bugs :issue:`120`, :issue:`123`) Implement post-tasks to |
| 1290 |
match pre-tasks, and allow control over the arguments passed to both (via |
| 1291 |
`invoke.tasks.call`). For details, see :ref:`pre-post-tasks`. |
| 1292 |
|
| 1293 |
.. warning:: |
| 1294 |
Pre-tasks were overhauled a moderate amount to implement this feature; |
| 1295 |
they now require references to **task objects** instead of **task |
| 1296 |
names**. This is a backwards incompatible change. |
| 1297 |
|
| 1298 |
- :support:`25` Trim a bunch of time off the test suite by using mocking and |
| 1299 |
other tools instead of dogfooding a bunch of subprocess spawns. |
| 1300 |
- :bug:`128 major` Positional arguments containing underscores were not |
| 1301 |
exporting to the parser correctly; this has been fixed. Thanks to J. Javier |
| 1302 |
Maestro for catch & patch. |
| 1303 |
- :bug:`121 major` Add missing help output denoting inverse Boolean options |
| 1304 |
(i.e. ``--[no-]foo`` for a ``--foo`` flag whose value defaults to true.) |
| 1305 |
Thanks to Andrew Roberts for catch & patch. |
| 1306 |
- :support:`118` Update the bundled ``six`` plus other minor tweaks to support |
| 1307 |
files. Thanks to Matt Iversen. |
| 1308 |
- :feature:`115` Make it easier to reuse Invoke's primary CLI machinery in |
| 1309 |
other (non-Invoke-distributed) bin-scripts. Thanks to Noah Kantrowitz. |
| 1310 |
- :feature:`110` Add task docstrings' 1st lines to ``--list`` output. Thanks to |
| 1311 |
Hiroki Kiyohara for the original PR (with assists from Robert Read and James |
| 1312 |
Thigpen.) |
| 1313 |
- :support:`117` Tidy up ``setup.py`` a bit, including axing the (broken) |
| 1314 |
`distutils` support. Thanks to Matt Iversen for the original PR & followup |
| 1315 |
discussion. |
| 1316 |
- :feature:`87` (also :issue:`92`) Rework the loader module such that recursive |
| 1317 |
filesystem searching is implemented, and is used instead of searching |
| 1318 |
`sys.path`. |
| 1319 |
|
| 1320 |
This adds the behavior most users expect or are familiar with from Fabric 1 |
| 1321 |
or similar tools; and it avoids nasty surprise collisions with other |
| 1322 |
installed packages containing files named ``tasks.py``. |
| 1323 |
|
| 1324 |
Thanks to Michael Hahn for the original report & PR, and to Matt Iversen for |
| 1325 |
providing the discovery algorithm used in the final version of this change. |
| 1326 |
|
| 1327 |
.. warning:: |
| 1328 |
This is technically a backwards incompatible change (reminder: we're not |
| 1329 |
at 1.0 yet!). You'll only notice if you were relying on adding your tasks |
| 1330 |
module to ``sys.path`` and then calling Invoke elsewhere on the |
| 1331 |
filesystem. |
| 1332 |
|
| 1333 |
- :support:`-` Refactor the `invoke.runners.Runner` module to differentiate |
| 1334 |
what it means to run a command in the abstract, from execution specifics. Top |
| 1335 |
level API is unaffected. |
| 1336 |
- :bug:`131 major` Make sure one's local tasks module is always first in |
| 1337 |
``sys.path``, even if its parent directory was already somewhere else in |
| 1338 |
``sys.path``. This ensures that local tasks modules never become hidden by |
| 1339 |
third-party ones. Thanks to ``@crccheck`` for the early report and to Dorian |
| 1340 |
Puła for assistance fixing. |
| 1341 |
- :bug:`116 major` Ensure nested config overrides play nicely with default |
| 1342 |
tasks and pre-tasks. |
| 1343 |
- :bug:`127 major` Fill in tasks' exposed ``name`` attribute with body name if |
| 1344 |
explicit name not given. |
| 1345 |
- :feature:`124` Add a ``--debug`` flag to the core parser to enable easier |
| 1346 |
debugging (on top of existing ``INVOKE_DEBUG`` env var.) |
| 1347 |
- :feature:`125` Improve output of Failure exceptions when printed. |
| 1348 |
- :release:`0.7.0 <2014.01.28>` |
| 1349 |
- :feature:`109` Add a ``default`` kwarg to |
| 1350 |
`invoke.collection.Collection.add_task` allowing per-collection control over |
| 1351 |
default tasks. |
| 1352 |
- :feature:`108` Update `invoke.collection.Collection.from_module` to accept |
| 1353 |
useful shorthand arguments for tweaking the `invoke.collection.Collection` |
| 1354 |
objects it creates (e.g. name, configuration.) |
| 1355 |
- :feature:`107` Update configuration merging behavior for more flexible reuse |
| 1356 |
of imported task modules, such as parameterizing multiple copies of a module |
| 1357 |
within a task tree. |
| 1358 |
- :release:`0.6.1 <2013.11.21>` |
| 1359 |
- :bug:`96` Tasks in subcollections which set explicit names (via e.g. |
| 1360 |
``@task(name='foo')``) were not having those names honored. This is fixed. |
| 1361 |
Thanks to Omer Katz for the report. |
| 1362 |
- :bug:`98` **BACKWARDS INCOMPATIBLE CHANGE!** Configuration merging has been |
| 1363 |
reversed so outer collections' config settings override inner collections. |
| 1364 |
This makes distributing reusable modules significantly less silly. |
| 1365 |
- :release:`0.6.0 <2013.11.21>` |
| 1366 |
- :bug:`86 major` Task arguments named with an underscore broke the help feature; |
| 1367 |
this is now fixed. Thanks to Stéphane Klein for the catch. |
| 1368 |
- :feature:`89` Implemented configuration for distributed task modules: can set |
| 1369 |
config options in `invoke.collection.Collection` objects and they are made |
| 1370 |
available to contextualized tasks. |
| 1371 |
- :release:`0.5.1 <2013.09.15>` |
| 1372 |
- :bug:`81` Fall back to sane defaults for PTY sizes when autodetection gives |
| 1373 |
insane results. Thanks to ``@akitada`` for the patch. |
| 1374 |
- :bug:`83` Fix a bug preventing underscored keyword arguments from working |
| 1375 |
correctly as CLI flags (e.g. ``mytask --my-arg`` would not map back correctly |
| 1376 |
to ``mytask(my_arg=...)``.) Credit: ``@akitada``. |
| 1377 |
- :release:`0.5.0 <2013.08.16>` |
| 1378 |
- :feature:`57` Optional-value flags added - e.g. ``--foo`` tells the parser to |
| 1379 |
set the ``foo`` option value to True; ``--foo myval`` sets the value to |
| 1380 |
"myval". The built-in ``--help`` option now leverages this feature for |
| 1381 |
per-task help (e.g. ``--help`` displays global help, ``--help mytask`` |
| 1382 |
displays help for ``mytask`` only.) |
| 1383 |
- :bug:`55 major` A bug in our vendored copy of ``pexpect`` clashed with a |
| 1384 |
Python 2->3 change in import behavior to prevent Invoke from running on |
| 1385 |
Python 3 unless the ``six`` module was installed in one's environment. This |
| 1386 |
was fixed - our vendored ``pexpect`` now always loads its sibling vendored |
| 1387 |
``six`` correctly. |