LSST Applications  21.0.0-172-gfb10e10a+18fedfabac,22.0.0+297cba6710,22.0.0+80564b0ff1,22.0.0+8d77f4f51a,22.0.0+a28f4c53b1,22.0.0+dcf3732eb2,22.0.1-1-g7d6de66+2a20fdde0d,22.0.1-1-g8e32f31+297cba6710,22.0.1-1-geca5380+7fa3b7d9b6,22.0.1-12-g44dc1dc+2a20fdde0d,22.0.1-15-g6a90155+515f58c32b,22.0.1-16-g9282f48+790f5f2caa,22.0.1-2-g92698f7+dcf3732eb2,22.0.1-2-ga9b0f51+7fa3b7d9b6,22.0.1-2-gd1925c9+bf4f0e694f,22.0.1-24-g1ad7a390+a9625a72a8,22.0.1-25-g5bf6245+3ad8ecd50b,22.0.1-25-gb120d7b+8b5510f75f,22.0.1-27-g97737f7+2a20fdde0d,22.0.1-32-gf62ce7b1+aa4237961e,22.0.1-4-g0b3f228+2a20fdde0d,22.0.1-4-g243d05b+871c1b8305,22.0.1-4-g3a563be+32dcf1063f,22.0.1-4-g44f2e3d+9e4ab0f4fa,22.0.1-42-gca6935d93+ba5e5ca3eb,22.0.1-5-g15c806e+85460ae5f3,22.0.1-5-g58711c4+611d128589,22.0.1-5-g75bb458+99c117b92f,22.0.1-6-g1c63a23+7fa3b7d9b6,22.0.1-6-g50866e6+84ff5a128b,22.0.1-6-g8d3140d+720564cf76,22.0.1-6-gd805d02+cc5644f571,22.0.1-8-ge5750ce+85460ae5f3,master-g6e05de7fdc+babf819c66,master-g99da0e417a+8d77f4f51a,w.2021.48
LSST Data Management Base Package
lsst::pex::config: configuration data management


The lsst::pex::config package provides for configurations for the LSST Data Management System.

Configurations are hierarchical trees of parameters used to control the execution of code. They should not be confused with input data. lsst::pex::config configurations ( Config objects) are validatable, documentable, and recordable. The configuration parameters are stored in attributes (member variables) of type Field in the Config object. Their values can be recorded for provenance purposes. The history of their changes during program execution is also tracked, and can be recorded and examined for provenance and debugging purposes.

The package is implemented in Python. Python code is used both in defining the available Fields and in setting or modifying their values. This provides great flexibility and introspection while still being approachable to non-programmers who need to modify configuration settings.


Defining a configuration subclass::

class IsrTaskConfig(config):
doWrite = Field("Write output?", bool, True)
fwhm = Field("FWHM of PSF (arcsec)", float, 1.0)
saturatedMaskName = pexConfig.Field(
"Name of mask plane to use in saturation detection", str, "SAT")
flatScalingType = ChoiceField(
"The method for scaling the flat on the fly.", str,
"USER": "User defined scaling",
"MEAN": "Scale by the inverse of the mean",
"MEDIAN": "Scale by the inverse of the median",
keysToRemoveFromAssembledCcd = ListField(
"fields to remove from the metadata of the assembled ccd.",
dtype=str, default=[])

Here is a sample configuration override file to be used with the partial IsrTaskConfig above. Note that all field names are prefixed with "config" and that overrides should never be used to set already-existing default values::

config.doWrite = False
config.fwhm = 0.8
config.saturatedMaskName = 'SATUR'
config.flatScalingType = 'MEAN'
config.keysToRemoveFromAssembledCcd = ['AMPNAME']

Typical usage::

def doIsrTask(ccd, configOverrideFilename=None):
config = IsrTaskConfig()
if configOverrideFilename is not None:
# Note: config override files are Python code with .py extensions
# by convention
detectSaturation(ccd, config.fwhm, config.saturatedMaskName)
# Note: methods typically do not need the entire config; they should be
# passed only relevant parameters
for k in config.keysToRemoveFromAssembledCcd:
if config.doWrite:


lsst::pex::config arose from a desire to have a configuration object holding key-value pairs that also allows for (arbitrarily simple or complex) validation of configuration values.

To configure code using lsst::pex::config, a developer subclasses the Config class. The subclass definition specifies the available Fields, their default values (if any), and their validation, if necessary.

Configs are hierarchical (see ConfigField), so calling code can embed the configuration definitions of called code.

Configurations are not input data. They should not be used in place of function or method arguments, nor are they intended to replace ordinary dictionary data structures. A good rule of thumb is that if a particular parameter does not have a useful default, it is probably an input rather than a configuration parameter. Another rule of thumb is that configuration parameters should generally not be set in algorithmic code, only in initialization or user interface code.

A Config subclass is instantiated to create a configuration object. If any default Field values need to be overridden, they can be set by assignment to the object's Field attributes (e.g. config.param1 = 3.14), often in a parent Config.setDefaults() method, or by loading from an external file. The code then uses the configuration values by accessing the object's Field attributes (e.g., x = config.param1).

The Config object can also be frozen; attempting to change the field values of a frozen object will raise an exception. This is useful to expose bugs that change configuration values after none should happen.

Finally, the contents of Config objects may easily be dumped, for provenance or debugging purposes.


  • Enable configuration of plug-in algorithms provided at runtime.
  • Allow setting of one Field to affect the values and the validation of others.
  • Collocate the Config definition with the code using the Config.
  • Provide a "Pythonic" interface.
  • Record the file and line of Field definitions and all changes to Fields, including setting default values.
  • Set defaults before overriding with user-specified values.
  • Support parameters with no (nonexistent) values, including overriding existing default values.
  • Enable closely-related Configs to be represented efficiently, with a minimum of duplication.
  • Have all user-modifiable Configs be part of a hierarchical tree.
  • Validate the contents of Fields as soon as possible.
  • Be able to "freeze" a Config to make it read-only.
  • Be able to persist a Config to a file and restore it identically.
  • Allow C++ control objects to be created from Configs, with documentation and validation specified exactly once.
  • Support lists of parameter values.

See wiki:PolicyEnhancement and wiki:Winter2012/PolicyRedesign.


All Configs are (direct or indirect) subclasses of the Python class lsst.pex.config.Config. Configs may inherit from other Configs, in which case all of the Fields of the parent class are present in the subclass.

Each Field is required to have a doc string that describes the contents of the field. Doc strings can be verbose and should give users of the Config a good understanding of what the Field is and how it will be interpreted and used. A doc string should also be provided for the class as a whole. The doc strings for the class and its Fields may be inspected using "help(MyConfig)" or with the pydoc command.

Types of Fields

Attributes of the configuration object must be subclasses of Field. A number of these are predefined: Field, RangeField, ChoiceField, ListField, ConfigField, ConfigChoiceField, RegistryField and ConfigurableField.

Example of RangeField::

class BackgroundConfig(Config):
"""Parameters for controlling background estimation."""
binSize = RangeField(
doc="how large a region of the sky should be used for each background point",
dtype=int, default=256, min=10

Example of ListField and Config inheritance::

class OutlierRejectedCoaddConfig(CoaddTask.ConfigClass):
"""Additional parameters for outlier-rejected coadds."""
subregionSize = ListField(
dtype = int,
doc = """width, height of stack subregion size; make small enough that a full stack of images will fit into memory at once""",
length = 2,
default = (200, 200),
optional = None,

Examples of ChoiceField and ConfigField and the use of Config.setDefaults() and Config.validate() methods::

class InitialPsfConfig(Config):
"""Describes the initial PSF used for detection and measurement before
we do PSF determination."""
model = ChoiceField(
dtype = str,
doc = "PSF model type",
default = "SingleGaussian",
allowed = {
"SingleGaussian": "Single Gaussian model",
"DoubleGaussian": "Double Gaussian model",
class CalibrateConfig(Config):
"""Configure calibration of an exposure."""
initialPsf = ConfigField(
dtype=InitialPsfConfig, doc=InitialPsfConfig.__doc__)
detection = ConfigField(
doc="Initial (high-threshold) detection phase for calibration"
def setDefaults(self):
self.detection.includeThresholdMultiplier = 10.0
def validate(self):
if self.doComputeApCorr and not self.doPsf:
raise ValueError("Cannot compute aperture correction without doing PSF determination")
T raise(T... args)

Example of a RegistryField created from a Registry object and use of both the Registry.register() method and the registerConfigurable decorator::

psfDeterminerRegistry = makeRegistry("""A registry of PSF determiner factories""")
class PcaPsfDeterminerConfig(Config):
spatialOrder = Field(
"spatial order for PSF kernel creation", int, 2)
class PcaPsfDeterminer(object):
ConfigClass = PcaPsfDeterminerConfig
# associate this Configurable class with its Config class for use
# by the registry
def __init__(self, config, schema=None):
def determinePsf(self, exposure, psfCandidateList, metadata=None):
psfDeterminerRegistry.register("shapelet", ShapeletPsfDeterminer)
class MeasurePsfConfig(Config):
psfDeterminer = measAlg.psfDeterminerRegistry.makeField("PSF determination algorithm", default="pca")
table::Schema schema
Definition: python.h:134
def registerConfigurable(name, registry, ConfigClass=None)
def makeRegistry(doc, configBaseType=Config)
table::Key< int > order

Inspecting a Config Object

Iterating through a Config yields the names of the Fields it contains. The standard dictionary-like keys(), items(), iterkeys(), iteritems(), and itervalues() methods are also supported.

Config.history contains the history of all changes to the Config's fields. Each Field also has a history. The formatHistory(fieldName) method displays the history of a given Field in a more human-readable format.

help(configObject) can be used to inspect the Config's doc strings as well as those of its Fields.

Wrapping C++ Control Objects

C++ control objects defined using the LSST_CONTROL_FIELD macro in lsst/pex/config.h can be wrapped using SWIG and the functions in lsst.pex.config.wrap, creating an equivalent Python Config. The Config will automatically create and set values in the C++ object, will provide access to the doc strings from C++, and will even call the C++ class's validate() method, if one exists. This helps to minimize duplication of code. In C++:

struct FooControl {
LSST_CONTROL_FIELD(bar, int, "documentation for field 'bar'");
LSST_CONTROL_FIELD(baz, double, "documentation for field 'baz'");
FooControl() : bar(0), baz(0.0) {}
A preprocessor macro used to define fields in C++ "control object" structs.
Definition: config.h:43

Note that only bool, int, double, and std::string fields, along with std::list and std::vector containers of those types, are fully supported.

Nested control objects are not supported.

After using SWIG, the preferred way to create the Config is via the wrap decorator::

from lsst.pex.config import wrap, Config
class FooConfig(Config):
table::Key< int > from
def wrap(ctrl)
A base class for image defects.



Config uses a metaclass to record the Field attributes within each Config object in an internal dictionary. The storage and history for the fields is also maintained in the Config, not the Field itself. This allows Fields to be inherited without difficulty.