pipelineIR.py

Go to the documentation of this file.

# This file is part of pipe_base.
#
# Developed for the LSST Data Management System.
# This product includes software developed by the LSST Project
# (http://www.lsst.org).
# See the COPYRIGHT file at the top-level directory of this distribution
# for details of code ownership.
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program.  If not, see <http://www.gnu.org/licenses/>.
from __future__ import annotations
 
__all__ = ("ConfigIR", "ContractError", "ContractIR", "InheritIR", "PipelineIR", "TaskIR", "LabeledSubset")
 
from collections import Counter
from collections.abc import Iterable as abcIterable
from dataclasses import dataclass, field
from typing import Any, List, Set, Union, Generator, MutableMapping, Optional, Dict
 
import copy
import re
import os
import yaml
import warnings
 
 
class PipelineYamlLoader(yaml.SafeLoader):
    """This is a specialized version of yaml's SafeLoader. It checks and raises
    an exception if it finds that there are multiple instances of the same key
    found inside a pipeline file at a given scope.
    """
    def construct_mapping(self, node, deep=False):
        # do the call to super first so that it can do all the other forms of
        # checking on this node. If you check the uniqueness of keys first
        # it would save the work that super does in the case of a failure, but
        # it might fail in the case that the node was the incorrect node due
        # to a parsing error, and the resulting exception would be difficult to
        # understand.
        mapping = super().construct_mapping(node, deep)
        # Check if there are any duplicate keys
        all_keys = Counter(key_node.value for key_node, _ in node.value)
        duplicates = {k for k, i in all_keys.items() if i != 1}
        if duplicates:
            raise KeyError("Pipeline files must not have duplicated keys, "
                           f"{duplicates} appeared multiple times")
        return mapping
 
 
class ContractError(Exception):
    """An exception that is raised when a pipeline contract is not satisfied
    """
    pass
 
 
@dataclass
class ContractIR:
    """Intermediate representation of contracts read from a pipeline yaml file.
    """
    contract: str
    """A string of python code representing one or more conditions on configs
    in a pipeline. This code-as-string should, once evaluated, should be True
    if the configs are fine, and False otherwise.
    """
    msg: Union[str, None] = None
    """An optional message to be shown to the user if a contract fails
    """
 
    def to_primitives(self) -> dict:
        """Convert to a representation used in yaml serialization
        """
        accumulate = {"contract": self.contract}
        if self.msg is not None:
            accumulate['msg'] = self.msg
        return accumulate
 
    def __eq__(self, other: "ContractIR"):
        if not isinstance(other, ContractIR):
            return False
        elif self.contract == other.contract and self.msg == other.msg:
            return True
        else:
            return False
 
 
@dataclass
class LabeledSubset:
    """Intermediate representation of named subset of task labels read from
    a pipeline yaml file.
    """
    label: str
    """The label used to identify the subset of task labels.
    """
    subset: Set[str]
    """A set of task labels contained in this subset.
    """
    description: Optional[str]
    """A description of what this subset of tasks is intended to do
    """
 
    @staticmethod
    def from_primatives(label: str, value: Union[List[str], dict]) -> LabeledSubset:
        """Generate `LabeledSubset` objects given a properly formatted object
        that as been created by a yaml loader.
 
        Parameters
        ----------
        label : `str`
            The label that will be used to identify this labeled subset.
        value : `list` of `str` or `dict`
            Object returned from loading a labeled subset section from a yaml
            document.
 
        Returns
        -------
        labeledSubset : `LabeledSubset`
            A `LabeledSubset` object build from the inputs.
 
        Raises
        ------
        ValueError
            Raised if the value input is not properly formatted for parsing
        """
        if isinstance(value, MutableMapping):
            subset = value.pop("subset", None)
            if subset is None:
                raise ValueError("If a labeled subset is specified as a mapping, it must contain the key "
                                 "'subset'")
            description = value.pop("description", None)
        elif isinstance(value, abcIterable):
            subset = value
            description = None
        else:
            raise ValueError(f"There was a problem parsing the labeled subset {label}, make sure the "
                             "definition is either a valid yaml list, or a mapping with keys "
                             "(subset, description) where subset points to a yaml list, and description is "
                             "associated with a string")
        return LabeledSubset(label, set(subset), description)
 
    def to_primitives(self) -> dict:
        """Convert to a representation used in yaml serialization
        """
        accumulate: Dict[str, Any] = {"subset": list(self.subset)}
        if self.description is not None:
            accumulate["description"] = self.description
        return accumulate
 
 
@dataclass
class ConfigIR:
    """Intermediate representation of configurations read from a pipeline yaml
    file.
    """
    python: Union[str, None] = None
    """A string of python code that is used to modify a configuration. This can
    also be None if there are no modifications to do.
    """
    dataId: Union[dict, None] = None
    """A dataId that is used to constrain these config overrides to only quanta
    with matching dataIds. This field can be None if there is no constraint.
    This is currently an unimplemented feature, and is placed here for future
    use.
    """
    file: List[str] = field(default_factory=list)
    """A list of paths which points to a file containing config overrides to be
    applied. This value may be an empty list if there are no overrides to
    apply.
    """
    rest: dict = field(default_factory=dict)
    """This is a dictionary of key value pairs, where the keys are strings
    corresponding to qualified fields on a config to override, and the values
    are strings representing the values to apply.
    """
 
    def to_primitives(self) -> dict:
        """Convert to a representation used in yaml serialization
        """
        accumulate = {}
        for name in ("python", "dataId", "file"):
            # if this attribute is thruthy add it to the accumulation
            # dictionary
            if getattr(self, name):
                accumulate[name] = getattr(self, name)
        # Add the dictionary containing the rest of the config keys to the
        # # accumulated dictionary
        accumulate.update(self.rest)
        return accumulate
 
    def maybe_merge(self, other_config: "ConfigIR") -> Generator["ConfigIR", None, None]:
        """Merges another instance of a `ConfigIR` into this instance if
        possible. This function returns a generator that is either self
        if the configs were merged, or self, and other_config if that could
        not be merged.
 
        Parameters
        ----------
        other_config : `ConfigIR`
            An instance of `ConfigIR` to merge into this instance.
 
        Returns
        -------
        Generator : `ConfigIR`
            A generator containing either self, or self and other_config if
            the configs could be merged or not respectively.
        """
        # Verify that the config blocks can be merged
        if self.dataId != other_config.dataId or self.python or other_config.python or\
                self.file or other_config.file:
            yield from (self, other_config)
            return
 
        # create a set of all keys, and verify two keys do not have different
        # values
        key_union = self.rest.keys() & other_config.rest.keys()
        for key in key_union:
            if self.rest[key] != other_config.rest[key]:
                yield from (self, other_config)
                return
        self.rest.update(other_config.rest)
 
        # Combine the lists of override files to load
        self_file_set = set(self.file)
        other_file_set = set(other_config.file)
        self.file = list(self_file_set.union(other_file_set))
 
        yield self
 
    def __eq__(self, other: "ConfigIR"):
        if not isinstance(other, ConfigIR):
            return False
        elif all(getattr(self, attr) == getattr(other, attr) for attr in
                 ("python", "dataId", "file", "rest")):
            return True
        else:
            return False
 
 
@dataclass
class TaskIR:
    """Intermediate representation of tasks read from a pipeline yaml file.
    """
    label: str
    """An identifier used to refer to a task.
    """
    klass: str
    """A string containing a fully qualified python class to be run in a
    pipeline.
    """
    config: Union[List[ConfigIR], None] = None
    """List of all configs overrides associated with this task, and may be
    `None` if there are no config overrides.
    """
 
    def to_primitives(self) -> dict:
        """Convert to a representation used in yaml serialization
        """
        accumulate = {'class': self.klass}
        if self.config:
            accumulate['config'] = [c.to_primitives() for c in self.config]
        return accumulate
 
    def add_or_update_config(self, other_config: ConfigIR):
        """Adds a `ConfigIR` to this task if one is not present. Merges configs
        if there is a `ConfigIR` present and the dataId keys of both configs
        match, otherwise adds a new entry to the config list. The exception to
        the above is that if either the last config or other_config has a
        python block, then other_config is always added, as python blocks can
        modify configs in ways that cannot be predicted.
 
        Parameters
        ----------
        other_config : `ConfigIR`
            A `ConfigIR` instance to add or merge into the config attribute of
            this task.
        """
        if not self.config:
            self.config = [other_config]
            return
        self.config.extend(self.config.pop().maybe_merge(other_config))
 
    def __eq__(self, other: "TaskIR"):
        if not isinstance(other, TaskIR):
            return False
        elif all(getattr(self, attr) == getattr(other, attr) for attr in
                 ("label", "klass", "config")):
            return True
        else:
            return False
 
 
@dataclass
class InheritIR:
    """An intermediate representation of inherited pipelines
    """
    location: str
    """This is the location of the pipeline to inherit. The path should be
    specified as an absolute path. Environment variables may be used in the
    path and should be specified as a python string template, with the name of
    the environment variable inside braces.
    """
    include: Union[List[str], None] = None
    """List of tasks that should be included when inheriting this pipeline.
    Either the include or exclude attributes may be specified, but not both.
    """
    exclude: Union[List[str], None] = None
    """List of tasks that should be excluded when inheriting this pipeline.
    Either the include or exclude attributes may be specified, but not both.
    """
    importContracts: bool = True
    """Boolean attribute to dictate if contracts should be inherited with the
    pipeline or not.
    """
 
    def toPipelineIR(self) -> "PipelineIR":
        """Convert to a representation used in yaml serialization
        """
        if self.include and self.exclude:
            raise ValueError("Both an include and an exclude list cant be specified"
                             " when declaring a pipeline import")
        tmp_pipeline = PipelineIR.from_file(os.path.expandvars(self.location))
        if tmp_pipeline.instrument is not None:
            warnings.warn("Any instrument definitions in imported pipelines are ignored. "
                          "if an instrument is desired please define it in the top most pipeline")
 
        included_labels = set()
        for label in tmp_pipeline.tasks:
            if (self.include and label in self.include) or (self.exclude and label not in self.exclude)\
                    or (self.include is None and self.exclude is None):
                included_labels.add(label)
 
        # Handle labeled subsets being specified in the include or exclude
        # list, adding or removing labels.
        if self.include is not None:
            subsets_in_include = tmp_pipeline.labeled_subsets.keys() & self.include
            for label in subsets_in_include:
                included_labels.update(tmp_pipeline.labeled_subsets[label].subset)
 
        elif self.exclude is not None:
            subsets_in_exclude = tmp_pipeline.labeled_subsets.keys() & self.exclude
            for label in subsets_in_exclude:
                included_labels.difference_update(tmp_pipeline.labeled_subsets[label].subset)
 
        tmp_pipeline = tmp_pipeline.subset_from_labels(included_labels)
 
        if not self.importContracts:
            tmp_pipeline.contracts = []
 
        return tmp_pipeline
 
    def __eq__(self, other: "InheritIR"):
        if not isinstance(other, InheritIR):
            return False
        elif all(getattr(self, attr) == getattr(other, attr) for attr in
                 ("location", "include", "exclude", "importContracts")):
            return True
        else:
            return False
 
 
class PipelineIR:
    """Intermediate representation of a pipeline definition
 
    Parameters
    ----------
    loaded_yaml : `dict`
        A dictionary which matches the structure that would be produced by a
        yaml reader which parses a pipeline definition document
 
    Raises
    ------
    ValueError :
        - If a pipeline is declared without a description
        - If no tasks are declared in a pipeline, and no pipelines are to be
          inherited
        - If more than one instrument is specified
        - If more than one inherited pipeline share a label
    """
    def __init__(self, loaded_yaml):
        # Check required fields are present
        if "description" not in loaded_yaml:
            raise ValueError("A pipeline must be declared with a description")
        if "tasks" not in loaded_yaml and "inherits" not in loaded_yaml:
            raise ValueError("A pipeline must be declared with one or more tasks")
 
        # These steps below must happen in this call order
 
        # Process pipeline description
        self.description = loaded_yaml.pop("description")
 
        # Process tasks
        self._read_tasks(loaded_yaml)
 
        # Process instrument keys
        inst = loaded_yaml.pop("instrument", None)
        if isinstance(inst, list):
            raise ValueError("Only one top level instrument can be defined in a pipeline")
        self.instrument = inst
 
        # Process any contracts
        self._read_contracts(loaded_yaml)
 
        # Process any named label subsets
        self._read_labeled_subsets(loaded_yaml)
 
        # Process any inherited pipelines
        self._read_inherits(loaded_yaml)
 
        # verify named subsets, must be done after inheriting
        self._verify_labeled_subsets()
 
    def _read_contracts(self, loaded_yaml):
        """Process the contracts portion of the loaded yaml document
 
        Parameters
        ---------
        loaded_yaml : `dict`
            A dictionary which matches the structure that would be produced by
            a yaml reader which parses a pipeline definition document
        """
        loaded_contracts = loaded_yaml.pop("contracts", [])
        if isinstance(loaded_contracts, str):
            loaded_contracts = [loaded_contracts]
        self.contracts = []
        for contract in loaded_contracts:
            if isinstance(contract, dict):
                self.contracts.append(ContractIR(**contract))
            if isinstance(contract, str):
                self.contracts.append(ContractIR(contract=contract))
 
    def _read_labeled_subsets(self, loaded_yaml: dict):
        """Process the subsets portion of the loaded yaml document
 
        Parameters
        ----------
        loaded_yaml: `MutableMapping`
            A dictionary which matches the structure that would be produced
            by a yaml reader which parses a pipeline definition document
        """
        loaded_subsets = loaded_yaml.pop("subsets", {})
        self.labeled_subsets = {}
        if not loaded_subsets and "subset" in loaded_yaml:
            raise ValueError("Top level key should be subsets and not subset, add an s")
        for key, value in loaded_subsets.items():
            self.labeled_subsets[key] = LabeledSubset.from_primatives(key, value)
 
    def _verify_labeled_subsets(self):
        """Verifies that all the labels in each named subset exist within the
        pipeline.
        """
        # Verify that all labels defined in a labeled subset are in the
        # Pipeline
        for labeled_subset in self.labeled_subsets.values():
            if not labeled_subset.subset.issubset(self.tasks.keys()):
                raise ValueError(f"Labels {labeled_subset.subset - self.tasks.keys()} were not found in the "
                                 "declared pipeline")
        # Verify subset labels are not already task labels
        label_intersection = self.labeled_subsets.keys() & self.tasks.keys()
        if label_intersection:
            raise ValueError(f"Labeled subsets can not use the same label as a task: {label_intersection}")
 
    def _read_inherits(self, loaded_yaml):
        """Process the inherits portion of the loaded yaml document
 
        Parameters
        ---------
        loaded_yaml : `dict`
            A dictionary which matches the structure that would be produced by
            a yaml reader which parses a pipeline definition document
        """
        def process_args(argument: Union[str, dict]) -> dict:
            if isinstance(argument, str):
                return {"location": argument}
            elif isinstance(argument, dict):
                if "exclude" in argument and isinstance(argument["exclude"], str):
                    argument["exclude"] = [argument["exclude"]]
                if "include" in argument and isinstance(argument["include"], str):
                    argument["include"] = [argument["include"]]
                return argument
        tmp_inherit = loaded_yaml.pop("inherits", None)
        if tmp_inherit is None:
            self.inherits = []
        elif isinstance(tmp_inherit, list):
            self.inherits = [InheritIR(**process_args(args)) for args in tmp_inherit]
        else:
            self.inherits = [InheritIR(**process_args(tmp_inherit))]
 
        # integrate any imported pipelines
        accumulate_tasks = {}
        accumulate_labeled_subsets = {}
        for other_pipeline in self.inherits:
            tmp_IR = other_pipeline.toPipelineIR()
            if accumulate_tasks.keys() & tmp_IR.tasks.keys():
                raise ValueError("Task labels in the imported pipelines must "
                                 "be unique")
            accumulate_tasks.update(tmp_IR.tasks)
            self.contracts.extend(tmp_IR.contracts)
            # verify that tmp_IR has unique labels for named subset among
            # existing labeled subsets, and with existing task labels.
            overlapping_subsets = accumulate_labeled_subsets.keys() & tmp_IR.labeled_subsets.keys()
            task_subset_overlap = ((accumulate_labeled_subsets.keys() | tmp_IR.labeled_subsets.keys())
                                   & accumulate_tasks.keys())
            if overlapping_subsets or task_subset_overlap:
                raise ValueError("Labeled subset names must be unique amongst imports in both labels and "
                                 f" named Subsets. Duplicate: {overlapping_subsets | task_subset_overlap}")
            accumulate_labeled_subsets.update(tmp_IR.labeled_subsets)
 
        # verify that any accumulated labeled subsets dont clash with a label
        # from this pipeline
        if accumulate_labeled_subsets.keys() & self.tasks.keys():
            raise ValueError("Labeled subset names must be unique amongst imports in both labels and "
                             " named Subsets")
        # merge in the named subsets for self so this document can override any
        # that have been delcared
        accumulate_labeled_subsets.update(self.labeled_subsets)
        self.labeled_subsets = accumulate_labeled_subsets
 
        # merge the dict of label:TaskIR objects, preserving any configs in the
        # imported pipeline if the labels point to the same class
        for label, task in self.tasks.items():
            if label not in accumulate_tasks:
                accumulate_tasks[label] = task
            elif accumulate_tasks[label].klass == task.klass:
                if task.config is not None:
                    for config in task.config:
                        accumulate_tasks[label].add_or_update_config(config)
            else:
                accumulate_tasks[label] = task
        self.tasks = accumulate_tasks
 
    def _read_tasks(self, loaded_yaml):
        """Process the tasks portion of the loaded yaml document
 
        Parameters
        ---------
        loaded_yaml : `dict`
            A dictionary which matches the structure that would be produced by
            a yaml reader which parses a pipeline definition document
        """
        self.tasks = {}
        tmp_tasks = loaded_yaml.pop("tasks", None)
        if tmp_tasks is None:
            tmp_tasks = {}
 
        for label, definition in tmp_tasks.items():
            if isinstance(definition, str):
                definition = {"class": definition}
            config = definition.get('config', None)
            if config is None:
                task_config_ir = None
            else:
                if isinstance(config, dict):
                    config = [config]
                task_config_ir = []
                for c in config:
                    file = c.pop("file", None)
                    if file is None:
                        file = []
                    elif not isinstance(file, list):
                        file = [file]
                    task_config_ir.append(ConfigIR(python=c.pop("python", None),
                                                   dataId=c.pop("dataId", None),
                                                   file=file,
                                                   rest=c))
            self.tasks[label] = TaskIR(label, definition["class"], task_config_ir)
 
    def _remove_contracts(self, label: str):
        """Remove any contracts that contain the given label
 
        String comparison used in this way is not the most elegant and may
        have issues, but it is the only feasible way when users can specify
        contracts with generic strings.
        """
        new_contracts = []
        for contract in self.contracts:
            # match a label that is not preceded by an ASCII identifier, or
            # is the start of a line and is followed by a dot
            if re.match(f".*([^A-Za-z0-9_]|^){label}[.]", contract.contract):
                continue
            new_contracts.append(contract)
        self.contracts = new_contracts
 
    def subset_from_labels(self, labelSpecifier: Set[str]) -> PipelineIR:
        """Subset a pipelineIR to contain only labels specified in
        labelSpecifier.
 
        Parameters
        ----------
        labelSpecifier : `set` of `str`
            Set containing labels that describes how to subset a pipeline.
 
        Returns
        -------
        pipeline : `PipelineIR`
            A new pipelineIR object that is a subset of the old pipelineIR
 
        Raises
        ------
        ValueError
            Raised if there is an issue with specified labels
 
        Notes
        -----
        This method attempts to prune any contracts that contain labels which
        are not in the declared subset of labels. This pruning is done using a
        string based matching due to the nature of contracts and may prune more
        than it should. Any labeled subsets defined that no longer have all
        members of the subset present in the pipeline will be removed from the
        resulting pipeline.
        """
 
        pipeline = copy.deepcopy(self)
 
        # update the label specifier to expand any named subsets
        toRemove = set()
        toAdd = set()
        for label in labelSpecifier:
            if label in pipeline.labeled_subsets:
                toRemove.add(label)
                toAdd.update(pipeline.labeled_subsets[label].subset)
        labelSpecifier.difference_update(toRemove)
        labelSpecifier.update(toAdd)
        # verify all the labels are in the pipeline
        if not labelSpecifier.issubset(pipeline.tasks.keys()
                                       | pipeline.labeled_subsets):
            difference = labelSpecifier.difference(pipeline.tasks.keys())
            raise ValueError("Not all supplied labels (specified or named subsets) are in the pipeline "
                             f"definition, extra labels: {difference}")
        # copy needed so as to not modify while iterating
        pipeline_labels = set(pipeline.tasks.keys())
        # Remove the labels from the pipelineIR, and any contracts that contain
        # those labels (see docstring on _remove_contracts for why this may
        # cause issues)
        for label in pipeline_labels:
            if label not in labelSpecifier:
                pipeline.tasks.pop(label)
                pipeline._remove_contracts(label)
 
        # create a copy of the object to iterate over
        labeled_subsets = copy.copy(pipeline.labeled_subsets)
        # remove any labeled subsets that no longer have a complete set
        for label, labeled_subset in labeled_subsets.items():
            if labeled_subset.subset - pipeline.tasks.keys():
                pipeline.labeled_subsets.pop(label)
 
        return pipeline
 
    @classmethod
    def from_string(cls, pipeline_string: str):
        """Create a `PipelineIR` object from a string formatted like a pipeline
        document
 
        Parameters
        ----------
        pipeline_string : `str`
            A string that is formatted according like a pipeline document
        """
        loaded_yaml = yaml.load(pipeline_string, Loader=PipelineYamlLoader)
        return cls(loaded_yaml)
 
    @classmethod
    def from_file(cls, filename: str):
        """Create a `PipelineIR` object from the document specified by the
        input path.
 
        Parameters
        ----------
        filename : `str`
            Location of document to use in creating a `PipelineIR` object.
        """
        with open(filename, 'r') as f:
            loaded_yaml = yaml.load(f, Loader=PipelineYamlLoader)
        return cls(loaded_yaml)
 
    def to_file(self, filename: str):
        """Serialize this `PipelineIR` object into a yaml formatted string and
        write the output to a file at the specified path.
 
        Parameters
        ----------
        filename : `str`
            Location of document to write a `PipelineIR` object.
        """
        with open(filename, 'w') as f:
            yaml.dump(self.to_primitives(), f, sort_keys=False)
 
    def to_primitives(self):
        """Convert to a representation used in yaml serialization
        """
        accumulate = {"description": self.description}
        if self.instrument is not None:
            accumulate['instrument'] = self.instrument
        accumulate['tasks'] = {m: t.to_primitives() for m, t in self.tasks.items()}
        if len(self.contracts) > 0:
            accumulate['contracts'] = [c.to_primitives() for c in self.contracts]
        if self.labeled_subsets:
            accumulate['subsets'] = {k: v.to_primitives() for k, v in self.labeled_subsets.items()}
        return accumulate
 
    def __str__(self) -> str:
        """Instance formatting as how it would look in yaml representation
        """
        return yaml.dump(self.to_primitives(), sort_keys=False)
 
    def __repr__(self) -> str:
        """Instance formatting as how it would look in yaml representation
        """
        return str(self)
 
    def __eq__(self, other: "PipelineIR"):
        if not isinstance(other, PipelineIR):
            return False
        elif all(getattr(self, attr) == getattr(other, attr) for attr in
                 ("contracts", "tasks", "instrument")):
            return True
        else:
            return False