Pytest 6.0.0rc1 attempts to log a repr of our mocked entry_point,
but it fails, becasue the repr metohod expects a valua attribute:
Reproducer:
$ tox -e py38 --force-dep 'pytest==6.0.0rc1'
...
Traceback (most recent call last):
File "/usr/lib/python3.8/logging/__init__.py", line 1081, in emit
msg = self.format(record)
File "/usr/lib/python3.8/logging/__init__.py", line 925, in format
return fmt.format(record)
File "/usr/lib/python3.8/logging/__init__.py", line 664, in format
record.message = record.getMessage()
File "/usr/lib/python3.8/logging/__init__.py", line 369, in getMessage
msg = msg % self.args
File "/tmp/flake8/.tox/py38/lib/python3.8/site-packages/flake8/plugins/manager.py", line 44, in __repr__
self.name, self.entry_point.value
File "/tmp/flake8/.tox/py38/lib/python3.8/site-packages/mock/mock.py", line 632, in __getattr__
raise AttributeError("Mock object has no attribute %r" % name)
AttributeError: Mock object has no attribute 'value'
Paths specified in configuration files should be relative to the
directory where the configuration file resides. Formerly, paths were
normalized relative to the current working directory where `flake8` was
invoked. The former behavior was not expected, especially for directory
structures with subprojects each having their own configuration.
This fixes a regression introduced in daca2c8 and b14d47b.
The --output-file` option was consumed by the preliminary option parser.
However, the `BaseFormatter` class needs the option for setting the
output filename. This special cases this option to ensure it gets
re-parsed and respected when specified on the CLI.
Preemptively determine the path of the user configuration file during
the construction of the `ConfigFileFinder` object. The user
configuration path will always be the same, regardless of when it gets
obtained by a run of `flake8`.
This isolates the logic of determining the user configuration path into
a static helper method to be called to set the `.user_config_file`
attribute. The helper method leverages `utils.is_windows()`, instead of
reimplementing the check, and decomposes clearly the directory name and
the base name to construct the path with a single `return` path.
Additionally, this avoids reconstructing the path on demand of obtaining
the user configuration file path.
Remove the checks to see if a configuration file has already been seen
and parsed. These checks aren't necessary because the entire run of
`flake8` calls these methods *only* once per configuration provided.
Now that `ConfigFileFinder.config_file` attribute is used everywhere and
is constructed from the `--config` CLI option, the now unused
`cli_config` parameters can be safely removed.
Now that the `ConfigFileFinder` has the `.config_file` attribute, switch
the relevant code paths to utilize this public attribute.
Tests have been updated to either construct `ConfigFileFinder` or mock
the object appropriately.
The `--config` flag is passed into `MergedConfigParser.parse()` and the
module-level function `config.get_local_plugins()`. Since both of these
places utilize the `ConfigFileFinder` object and the configuration file
override pertains to how configuration behaves, this incremental change
directly associates the `ConfigFileFinder` and the configuration file
override.
These attributes are only needed within the
`.generate_possible_local_files()` method. Therefore, just obtain the
current working directory at the beginning of the method and reduce the
lifetime state of the `ConfigFileFinder` object.
`os.getcwd()` returns an absolute path; thus, the patched paths should
be absolute as well.
This is an incremental change towards removing the `ConfigFileFinder`
attributes `.parent` and `.tail` to be localized to
`.generate_possible_local_files()`. Without this, the tests fail when
moving the patching because `os.path.abspath()` calls `os.getcwd()`,
expecting `os.getcwd()` to be an absolute path.
Now that `ConfigFileFinder.ignore_config_files` attribute is used
everywhere and is constructed from the `--isolated` CLI option, the now
unused `isolated` parameters can be safely removed.
Now that the `ConfigFileFinder` has the `.ignore_config_files`
attribute, switch the relevant code paths to utilize this public
attribute.
Tests have been updated to either construct `ConfigFileFinder` or mock
the object appropriately.
The `--isolated` flag is passed into `MergedConfigParser.parse()` and
the module-level function `config.get_local_plugins()`. Since both of
these places utilize the `ConfigFileFinder` object and isolation
pertains to how the `ConfigFileFinder` should behave with respect to
isolation, this incremental change more directly associates the
`ConfigFileFinder` and configuration file isolate.
This simplifies `._read_config()` by removing a conditional branch in
the situation where it is called with one file to process. Now the
contract accepts any number of arguments of the same type.
Where callers invoke `._read_config()` with a `Sequence`, the call site
has been changed to unpack arguments (i.e., `*`).
The tests in `test_merged_config_parser.py` needed to return a string
for the user configuration path instead of an empty list since
`ConfigFileFinder.user_config_file()` returns a string.
This fixes a regression introduced in !346 to ensure that `--config` and
`--isolated` are recognized in `aggregate_options()`.
The regression manifested because `aggregate_options()` was relying on
re-parsing `argv` to obtain the option values. However, !346 changed
the preliminary parsing logic to only parse and "eat" what is necessary
and forward along the options needed before all the configuration was
loaded. This code path was overlooked because the tests in
`test_aggregator()` were passing but the call from the `Application`
object would never have these options in the remaining `argv` list to be
passed long.
Simplify the initialization code path by invoking
`config.ConfigFileFinder` directly instead of the extra hop by calling
the static class-level `Application.make_config_finder()` method.
This change removes the `.config_finder` object from the `Application`.
Since the configuration finder is only needed during initialization, we
constrain the finder to be returned and passed to other methods
necessary for initialization.
Positional arguments aren't necessary for determining where to
load configuration anymore and is safe to keep both options and
arguments to be forwarded for later parsing after configuration is
loaded.
Now that preliminary options are registered with the preliminary parser
on the `Application`, leverage that for parsing known options.
This important change removes the `Application.option_manager` from
being responsible for pre-configuration parsing and the workarounds
needed in the `Application.parse_preliminary_options_and_args()` to
account for the fact that `Application.option_manager` was aware of
*all* options, not just the options necessary for pre-configuration
loading. A following commit will address removing these workarounds.
