Details
-
Type:
Story
-
Status: Done
-
Resolution: Done
-
Fix Version/s: None
-
Component/s: meas_base
-
Labels:
-
Epic Link:
-
Team:Architecture
Description
test_PluginLogs.py fails if you run all the tests for meas_base in a single process (such as would happen if you type pytest in the build directory. I think this is because the test is relying on the set of registered plugins but that set has been modified by earlier tests.
============================================== FAILURES ==============================================
|
__________________________ RegisteredPluginsTestCase.testSingleFramePlugins __________________________
|
|
|
self = <test_PluginLogs.RegisteredPluginsTestCase testMethod=testSingleFramePlugins>
|
|
|
def testSingleFramePlugins(self):
|
center = lsst.afw.geom.Point2D(50, 50)
|
bbox = lsst.afw.geom.Box2I(lsst.afw.geom.Point2I(0, 0),
|
lsst.afw.geom.Extent2I(100,100))
|
dataset = lsst.meas.base.tests.TestDataset(bbox)
|
dataset.addSource(1000000.0, center)
|
registry = SingleFramePlugin.registry
|
dependencies = registry.keys()
|
> task = self.makeSingleFrameMeasurementTask("base_SdssCentroid", dependencies=dependencies)
|
|
|
tests/test_PluginLogs.py:132:
|
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
|
python/lsst/meas/base/tests.py:515: in makeSingleFrameMeasurementTask
|
return SingleFrameMeasurementTask(schema=schema, algMetadata=algMetadata, config=config)
|
python/lsst/meas/base/sfm.py:267: in __init__
|
self.initializePlugins(schema=self.schema)
|
python/lsst/meas/base/baseMeasurement.py:232: in initializePlugins
|
for executionOrder, name, config, PluginClass in sorted(self.config.plugins.apply()):
|
../../stack/DarwinX86/pex_config/13.0-3-g520d906/python/lsst/pex/config/registry.py:182: in apply
|
retvals.append(self._field.typemap.registry[c](*args, config=self[c], **kw))
|
python/lsst/meas/base/pluginRegistry.py:95: in __call__
|
return (self.PluginClass.getExecutionOrder(), self.name, config, self.PluginClass)
|
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
|
|
|
cls = <class 'test_ApCorrNameSet.ApCorrNameTestCase.testRegisterDecorator.<locals>.NonApCorrPlugin'>
|
|
|
@classmethod
|
def getExecutionOrder(cls):
|
"""Sets the relative order of plugins (smaller numbers run first).
|
|
In general, the following class constants should be used (other values
|
are also allowed, but should be avoided unless they are needed):
|
CENTROID_ORDER centroids and other algorithms that require only a Footprint
|
and its Peaks as input
|
SHAPE_ORDER shape measurements and other algorithms that require getCentroid() to return
|
a good centroid (in addition to a Footprint and its Peaks).
|
FLUX_ORDER flux algorithms that require both getShape() and getCentroid(),
|
in addition to a Footprint and its Peaks
|
DEFAULT_CATALOGCALCULATION plugins that only operate on the catalog
|
|
Must be reimplemented as a class method by concrete derived classes.
|
|
This approach was chosen instead of a full graph-based analysis of dependencies
|
because algorithm dependencies are usually both quite simple and entirely substitutable:
|
an algorithm that requires a centroid can typically make use of any centroid algorithms
|
outputs. That makes it relatively easy to figure out the correct value to use for any
|
particular algorithm.
|
"""
|
> raise NotImplementedError("All plugins must implement getExecutionOrder()")
|
E NotImplementedError: All plugins must implement getExecutionOrder()
|
|
|
python/lsst/meas/base/pluginsBase.py:78: NotImplementedError
|
=============================== 1 failed, 197 passed in 72.78 seconds ================================
|
If I reverse the order of tests all tests pass (but "P" is late in the alphabet so there are only ten or so tests after it).
Attachments
Activity
I agree with Kian-Tat Lim's proposal, though I think John Swinbank's proposal to fix the malformed plugins would also be worth doing.
Downside to having tests specify up-front which plugins they will access is that it limits your ability to write a test for "do something with all known plugins" functionality. But in this case, it should be an easy workaround.
John Swinbank I've made the suggested fix that ensures the plugins are valid. Also did some flake8 fixes so that when DM-11514 lands we can turn on flake8 testing.
Looks ok — minor comments on the PR.
I think the "proper" way to handle this is for this test to only check plugins that it knows are going to be present (either because they are explicitly added by this test or because they are installed by default).