dag.py

__author__ = "Johannes Köster"
__copyright__ = "Copyright 2022, Johannes Köster"
__email__ = "johannes.koester@uni-due.de"
__license__ = "MIT"

import html
import os
import sys
import shutil
import textwrap
import time
import tarfile
from collections import defaultdict, Counter, deque, namedtuple
from itertools import chain, filterfalse, groupby
from functools import partial
from pathlib import Path
import uuid
import math

from snakemake.io import (
    PeriodicityDetector,
    get_flag_value,
    is_callable,
    wait_for_files,
    is_flagged,
    IOFile,
)
from snakemake.jobs import Reason, JobFactory, GroupJobFactory, Job
from snakemake.exceptions import MissingInputException
from snakemake.exceptions import MissingRuleException, AmbiguousRuleException
from snakemake.exceptions import CyclicGraphException, MissingOutputException
from snakemake.exceptions import IncompleteFilesException, ImproperOutputException
from snakemake.exceptions import PeriodicWildcardError
from snakemake.exceptions import RemoteFileException, WorkflowError, ChildIOException
from snakemake.exceptions import InputFunctionException
from snakemake.logging import logger
from snakemake.common import DYNAMIC_FILL, ON_WINDOWS, group_into_chunks
from snakemake.deployment import conda, singularity
from snakemake.output_index import OutputIndex
from snakemake import workflow


PotentialDependency = namedtuple("PotentialDependency", ["file", "jobs", "known"])


class Batch:
    """Definition of a batch for calculating only a partial DAG."""

    def __init__(self, rulename: str, idx: int, batches: int):
        assert idx <= batches
        assert idx > 0
        self.rulename = rulename
        self.idx = idx
        self.batches = batches

    def get_batch(self, items: list):
        """Return the defined batch of the given items.
        Items are usually input files."""
        # make sure that we always consider items in the same order
        if len(items) < self.batches:
            raise WorkflowError(
                "Batching rule {} has less input files than batches. "
                "Please choose a smaller number of batches.".format(self.rulename)
            )
        items = sorted(items)
        batch_len = math.floor(len(items) / self.batches)
        # self.batch is one-based, hence we have to subtract 1
        idx = self.idx - 1
        i = idx * batch_len
        if self.is_final:
            # extend the last batch to cover rest of list
            return items[i:]
        else:
            return items[i : i + batch_len]

    @property
    def is_final(self):
        return self.idx == self.batches

    def __str__(self):
        return "{}/{} (rule {})".format(self.idx, self.batches, self.rulename)


class DAG:
    """Directed acyclic graph of jobs."""

    def __init__(
        self,
        workflow,
        rules=None,
        dryrun=False,
        targetfiles=None,
        targetrules=None,
        forceall=False,
        forcerules=None,
        forcefiles=None,
        priorityfiles=None,
        priorityrules=None,
        untilfiles=None,
        untilrules=None,
        omitfiles=None,
        omitrules=None,
        ignore_ambiguity=False,
        force_incomplete=False,
        ignore_incomplete=False,
        notemp=False,
        keep_remote_local=False,
        batch=None,
    ):
        self.dryrun = dryrun
        self.dependencies = defaultdict(partial(defaultdict, set))
        self.depending = defaultdict(partial(defaultdict, set))
        self._needrun = set()
        self._priority = dict()
        self._reason = defaultdict(Reason)
        self._finished = set()
        self._dynamic = set()
        self._len = 0
        self.workflow = workflow
        self.rules = set(rules)
        self.ignore_ambiguity = ignore_ambiguity
        self.targetfiles = targetfiles
        self.targetrules = targetrules
        self.priorityfiles = priorityfiles
        self.priorityrules = priorityrules
        self.targetjobs = set()
        self.prioritytargetjobs = set()
        self._ready_jobs = set()
        self.notemp = notemp
        self.keep_remote_local = keep_remote_local
        self._jobid = dict()
        self.job_cache = dict()
        self.conda_envs = dict()
        self.container_imgs = dict()
        self._progress = 0
        self._group = dict()
        self._n_until_ready = defaultdict(int)
        self._running = set()

        self.job_factory = JobFactory()
        self.group_job_factory = GroupJobFactory()

        self.forcerules = set()
        self.forcefiles = set()
        self.untilrules = set()
        self.untilfiles = set()
        self.omitrules = set()
        self.omitfiles = set()
        self.updated_subworkflow_files = set()
        if forceall:
            self.forcerules.update(self.rules)
        elif forcerules:
            self.forcerules.update(forcerules)
        if forcefiles:
            self.forcefiles.update(forcefiles)
        if untilrules:
            self.untilrules.update(set(rule.name for rule in untilrules))
        if untilfiles:
            self.untilfiles.update(untilfiles)
        if omitrules:
            self.omitrules.update(set(rule.name for rule in omitrules))
        if omitfiles:
            self.omitfiles.update(omitfiles)

        self.has_dynamic_rules = any(rule.dynamic_output for rule in self.rules)

        self.omitforce = set()

        self.batch = batch
        if batch is not None and not batch.is_final:
            # Since not all input files of a batching rule are considered, we cannot run
            # beyond that rule.
            # For the final batch, we do not need to omit anything.
            self.omitrules.add(batch.rulename)

        self.force_incomplete = force_incomplete
        self.ignore_incomplete = ignore_incomplete

        self.periodic_wildcard_detector = PeriodicityDetector()

        self.update_output_index()

    def init(self, progress=False):
        """Initialise the DAG."""
        for job in map(self.rule2job, self.targetrules):
            job = self.update([job], progress=progress, create_inventory=True)
            self.targetjobs.add(job)

        for file in self.targetfiles:
            job = self.update(
                self.file2jobs(file),
                file=file,
                progress=progress,
                create_inventory=True,
            )
            self.targetjobs.add(job)

        self.cleanup()

        self.check_incomplete()

        self.update_needrun(create_inventory=True)
        self.set_until_jobs()
        self.delete_omitfrom_jobs()
        self.update_jobids()

        self.check_directory_outputs()

        # check if remaining jobs are valid
        for i, job in enumerate(self.jobs):
            job.is_valid()

    def check_directory_outputs(self):
        """Check that no output file is contained in a directory output of the same or another rule."""
        outputs = sorted(
            {(os.path.abspath(f), job) for job in self.jobs for f in job.output}
        )
        for i in range(len(outputs) - 1):
            (a, job_a), (b, job_b) = outputs[i : i + 2]
            try:
                common = os.path.commonpath([a, b])
            except ValueError:
                # commonpath raises error if windows drives are different.
                continue
            if a != b and common == os.path.commonpath([a]) and job_a != job_b:
                raise ChildIOException(parent=outputs[i], child=outputs[i + 1])

    @property
    def checkpoint_jobs(self):
        for job in self.needrun_jobs:
            if job.is_checkpoint:
                yield job

    def update_checkpoint_outputs(self):
        workflow.checkpoints.future_output = set(
            f for job in self.checkpoint_jobs for f in job.output
        )

    def update_jobids(self):
        for job in self.jobs:
            if job not in self._jobid:
                self._jobid[job] = len(self._jobid)

    def cleanup_workdir(self):
        for job in self.jobs:
            if not self.is_edit_notebook_job(job):
                for io_dir in set(
                    os.path.dirname(io_file)
                    for io_file in chain(job.output, job.input)
                    if not os.path.exists(io_file)
                ):
                    if os.path.exists(io_dir) and not len(os.listdir(io_dir)):
                        os.removedirs(io_dir)

    def cleanup(self):
        self.job_cache.clear()
        final_jobs = set(self.bfs(self.dependencies, *self.targetjobs))
        todelete = [job for job in self.dependencies if job not in final_jobs]
        for job in todelete:
            try:
                self._needrun.remove(job)
            except KeyError:
                pass

            # delete all pointers from dependencies to this job
            for dep in self.dependencies[job]:
                try:
                    del self.depending[dep][job]
                except KeyError:
                    # In case the pointer has been deleted before or
                    # never created, we can simply continue.
                    pass

            # delete all dependencies
            del self.dependencies[job]
            try:
                # delete all pointers to downstream dependencies
                del self.depending[job]
            except KeyError:
                pass

    def create_conda_envs(
        self, dryrun=False, forceall=False, init_only=False, quiet=False
    ):
        # First deduplicate based on job.conda_env_spec
        jobs = self.jobs if forceall else self.needrun_jobs
        env_set = {
            (job.conda_env_spec, job.container_img_url)
            for job in jobs
            if job.conda_env_spec and (self.workflow.assume_shared_fs or job.is_local)
        }

        # Then based on md5sum values
        self.conda_envs = dict()
        for (env_spec, simg_url) in env_set:
            simg = None
            if simg_url and self.workflow.use_singularity:
                assert (
                    simg_url in self.container_imgs
                ), "bug: must first pull singularity images"
                simg = self.container_imgs[simg_url]
            env = env_spec.get_conda_env(
                self.workflow,
                container_img=simg,
                cleanup=self.workflow.conda_cleanup_pkgs,
            )
            self.conda_envs[(env_spec, simg_url)] = env

        if not init_only:
            for env in self.conda_envs.values():
                if (not dryrun or not quiet) and not env.is_named:
                    env.create(dryrun)

    def pull_container_imgs(self, dryrun=False, forceall=False, quiet=False):
        # First deduplicate based on job.conda_env_spec
        jobs = self.jobs if forceall else self.needrun_jobs
        img_set = {
            (job.container_img_url, job.is_containerized)
            for job in jobs
            if job.container_img_url
        }

        for img_url, is_containerized in img_set:
            img = singularity.Image(img_url, self, is_containerized)
            if not dryrun or not quiet:
                img.pull(dryrun)
            self.container_imgs[img_url] = img

    def update_output_index(self):
        """Update the OutputIndex."""
        self.output_index = OutputIndex(self.rules)

    def check_incomplete(self):
        """Check if any output files are incomplete. This is done by looking up
        markers in the persistence module."""
        if not self.ignore_incomplete:
            incomplete = self.incomplete_files
            if incomplete:
                if self.force_incomplete:
                    logger.debug("Forcing incomplete files:")
                    logger.debug("\t" + "\n\t".join(incomplete))
                    self.forcefiles.update(incomplete)
                else:
                    raise IncompleteFilesException(incomplete)

    def incomplete_external_jobid(self, job):
        """Return the external jobid of the job if it is marked as incomplete.

        Returns None, if job is not incomplete, or if no external jobid has been
        registered or if force_incomplete is True.
        """
        if self.force_incomplete:
            return None
        jobids = self.workflow.persistence.external_jobids(job)
        if len(jobids) == 1:
            return jobids[0]
        elif len(jobids) > 1:
            raise WorkflowError(
                "Multiple different external jobids registered "
                "for output files of incomplete job {} ({}). This job "
                "cannot be resumed. Execute Snakemake with --rerun-incomplete "
                "to fix this issue.".format(job.jobid, jobids)
            )

    def check_dynamic(self):
        """Check dynamic output and update downstream rules if necessary."""
        if self.has_dynamic_rules:
            for job in filter(
                lambda job: (job.dynamic_output and not self.needrun(job)),
                list(self.jobs),
            ):
                self.update_dynamic(job)
            self.postprocess()

    def is_edit_notebook_job(self, job):
        return self.workflow.edit_notebook and job.targetfile in self.targetfiles

    def get_job_group(self, job):
        return self._group.get(job)

    @property
    def dynamic_output_jobs(self):
        """Iterate over all jobs with dynamic output files."""
        return (job for job in self.jobs if job.dynamic_output)

    @property
    def jobs(self):
        """All jobs in the DAG."""
        return self.dependencies.keys()

    @property
    def needrun_jobs(self):
        """Jobs that need to be executed."""
        return filterfalse(self.finished, self._needrun)

    @property
    def local_needrun_jobs(self):
        """Iterate over all jobs that need to be run and are marked as local."""
        return filter(lambda job: job.is_local, self.needrun_jobs)

    @property
    def finished_jobs(self):
        """Iterate over all jobs that have been finished."""
        return filter(self.finished, self.jobs)

    @property
    def ready_jobs(self):
        """Jobs that are ready to execute."""
        return self._ready_jobs

    def needrun(self, job):
        """Return whether a given job needs to be executed."""
        return job in self._needrun

    def priority(self, job):
        """Return priority of given job."""
        return self._priority[job]

    def noneedrun_finished(self, job):
        """
        Return whether a given job is finished or was not
        required to run at all.
        """
        return not self.needrun(job) or self.finished(job)

    def reason(self, job):
        """Return the reason of the job execution."""
        return self._reason[job]

    def finished(self, job):
        """Return whether a job is finished."""
        return job in self._finished

    def dynamic(self, job):
        """
        Return whether a job is dynamic (i.e. it is only a placeholder
        for those that are created after the job with dynamic output has
        finished.
        """
        if job.is_group():
            for j in job:
                if j in self._dynamic:
                    return True
        else:
            return job in self._dynamic

    def requested_files(self, job):
        """Return the files a job requests."""
        return set(*self.depending[job].values())

    @property
    def incomplete_files(self):
        """Return list of incomplete files."""
        return list(
            chain(
                *(
                    job.output
                    for job in filter(
                        self.workflow.persistence.incomplete,
                        filterfalse(self.needrun, self.jobs),
                    )
                )
            )
        )

    @property
    def newversion_files(self):
        """Return list of files where the current version is newer than the
        recorded version.
        """
        return list(
            chain(
                *(
                    job.output
                    for job in filter(self.workflow.persistence.newversion, self.jobs)
                )
            )
        )

    def missing_temp(self, job):
        """
        Return whether a temp file that is input of the given job is missing.
        """
        for job_, files in self.depending[job].items():
            if self.needrun(job_) and any(not f.exists for f in files):
                return True
        return False

    def handle_ensure(self, job, expanded_output):
        ensured_output = {
            f: get_flag_value(f, "ensure")
            for f in expanded_output
            if is_flagged(f, "ensure")
        }
        # handle non_empty
        empty_output = [
            f
            for f, ensure in ensured_output.items()
            if ensure["non_empty"] and f.size == 0
        ]
        if empty_output:
            raise WorkflowError(
                "Detected unexpected empty output files. "
                "Something went wrong in the rule without "
                "an error being reported:\n{}".format("\n".join(empty_output)),
                rule=job.rule,
            )

        # handle checksum
        def is_not_same_checksum(f, checksum):
            if checksum is None:
                return False
            if is_callable(checksum):
                try:
                    checksum = checksum(job.wildcards)
                except Exception as e:
                    raise WorkflowError(
                        "Error calling checksum function provided to ensure marker.",
                        e,
                        rule=job.rule,
                    )
            return not f.is_same_checksum(checksum, force=True)

        checksum_failed_output = [
            f
            for f, ensure in ensured_output.items()
            if is_not_same_checksum(f, ensure.get("sha256"))
        ]
        if checksum_failed_output:
            raise WorkflowError(
                "Output files have checksums that differ from the expected ones "
                "defined in the workflow:\n{}".format(
                    "\n".join(checksum_failed_output)
                ),
                rule=job.rule,
            )

    def check_and_touch_output(
        self,
        job,
        wait=3,
        ignore_missing_output=False,
        no_touch=False,
        force_stay_on_remote=False,
    ):
        """Raise exception if output files of job are missing."""
        expanded_output = [job.shadowed_path(path) for path in job.expanded_output]
        if job.benchmark:
            expanded_output.append(job.benchmark)

        if not ignore_missing_output:
            try:
                wait_for_files(
                    expanded_output,
                    latency_wait=wait,
                    force_stay_on_remote=force_stay_on_remote,
                    ignore_pipe_or_service=True,
                )
            except IOError as e:
                raise MissingOutputException(
                    str(e),
                    rule=job.rule,
                    jobid=self.jobid(job),
                )

        # Ensure that outputs are of the correct type (those flagged with directory()
        # are directories and not files and vice versa). We can't check for remote objects.
        for f in expanded_output:
            if (f.is_directory and not f.remote_object and not os.path.isdir(f)) or (
                not f.remote_object and os.path.isdir(f) and not f.is_directory
            ):
                raise ImproperOutputException(job, [f])

        # Handle ensure flags
        self.handle_ensure(job, expanded_output)

        # It is possible, due to archive expansion or cluster clock skew, that
        # the files appear older than the input.  But we know they must be new,
        # so touch them to update timestamps. This also serves to touch outputs
        # when using the --touch flag.
        # Note that if the input files somehow have a future date then this will
        # not currently be spotted and the job will always be re-run.
        if not no_touch:
            for f in expanded_output:
                # This won't create normal files if missing, but will create
                # the flag file for directories.
                if f.exists_local:
                    f.touch()

    def unshadow_output(self, job, only_log=False):
        """Move files from shadow directory to real output paths."""
        if not job.shadow_dir or not job.expanded_output:
            return

        files = job.log if only_log else chain(job.expanded_output, job.log)

        for real_output in files:
            shadow_output = job.shadowed_path(real_output).file
            # Remake absolute symlinks as relative
            if os.path.islink(shadow_output):
                dest = os.readlink(shadow_output)
                if os.path.isabs(dest):
                    rel_dest = os.path.relpath(dest, job.shadow_dir)
                    os.remove(shadow_output)
                    os.symlink(rel_dest, shadow_output)

            if os.path.realpath(shadow_output) == os.path.realpath(real_output):
                continue
            logger.debug(
                "Moving shadow output {} to destination {}".format(
                    shadow_output, real_output
                )
            )
            shutil.move(shadow_output, real_output)
        shutil.rmtree(job.shadow_dir)

    def check_periodic_wildcards(self, job):
        """Raise an exception if a wildcard of the given job appears to be periodic,
        indicating a cyclic dependency."""
        for wildcard, value in job.wildcards_dict.items():
            periodic_substring = self.periodic_wildcard_detector.is_periodic(value)
            if periodic_substring is not None:
                raise PeriodicWildcardError(
                    "The value {} in wildcard {} is periodically repeated ({}). "
                    "This would lead to an infinite recursion. "
                    "To avoid this, e.g. restrict the wildcards in this rule to certain values.".format(
                        periodic_substring, wildcard, value
                    ),
                    rule=job.rule,
                )

    def handle_protected(self, job):
        """Write-protect output files that are marked with protected()."""
        for f in job.expanded_output:
            if f in job.protected_output:
                logger.info("Write-protecting output file {}.".format(f))
                f.protect()

    def handle_touch(self, job):
        """Touches those output files that are marked for touching."""
        for f in job.expanded_output:
            if f in job.touch_output:
                f = job.shadowed_path(f)
                logger.info("Touching output file {}.".format(f))
                f.touch_or_create()
                assert os.path.exists(f)

    def temp_input(self, job):
        for job_, files in self.dependencies[job].items():
            for f in filter(job_.temp_output.__contains__, files):
                yield f

    def temp_size(self, job):
        """Return the total size of temporary input files of the job.
        If none, return 0.
        """
        return sum(f.size for f in self.temp_input(job))

    def handle_temp(self, job):
        """Remove temp files if they are no longer needed. Update temp_mtimes."""
        if self.notemp:
            return

        if job.is_group():
            for j in job:
                self.handle_temp(j)
            return

        is_temp = lambda f: is_flagged(f, "temp")

        # handle temp input
        needed = lambda job_, f: any(
            f in files
            for j, files in self.depending[job_].items()
            if not self.finished(j) and self.needrun(j) and j != job
        )

        def unneeded_files():
            # temp input
            for job_, files in self.dependencies[job].items():
                tempfiles = set(f for f in job_.expanded_output if is_temp(f))
                yield from filterfalse(partial(needed, job_), tempfiles & files)

            # temp output
            if (
                not job.dynamic_output
                and not job.is_checkpoint
                and (
                    job not in self.targetjobs
                    or job.rule.name == self.workflow.default_target
                )
            ):
                tempfiles = (
                    f
                    for f in job.expanded_output
                    if is_temp(f) and f not in self.targetfiles
                )
                yield from filterfalse(partial(needed, job), tempfiles)

        for f in unneeded_files():
            if self.dryrun:
                logger.info(f"Would remove temporary output {f}")
            else:
                logger.info("Removing temporary output {}.".format(f))
                f.remove(remove_non_empty_dir=True)

    def handle_log(self, job, upload_remote=True):
        for f in job.log:
            f = job.shadowed_path(f)
            if not f.exists_local:
                # If log file was not created during job, create an empty one.
                f.touch_or_create()

    def handle_remote(self, job, upload=True):
        """Remove local files if they are no longer needed and upload."""
        if upload:
            # handle output files
            files = job.expanded_output
            if job.benchmark:
                files = chain(job.expanded_output, (job.benchmark,))
            if job.log:
                files = chain(files, job.log)
            for f in files:
                if f.is_remote and not f.should_stay_on_remote:
                    f.upload_to_remote()
                    remote_mtime = f.mtime.remote()
                    # immediately force local mtime to match remote,
                    # since conversions from S3 headers are not 100% reliable
                    # without this, newness comparisons may fail down the line
                    f.touch(times=(remote_mtime, remote_mtime))

                    if not f.exists_remote:
                        raise RemoteFileException(
                            "The file upload was attempted, but it does not "
                            "exist on remote. Check that your credentials have "
                            "read AND write permissions."
                        )

        if not self.keep_remote_local:
            if not any(f.is_remote for f in job.input):
                return

            # handle input files
            needed = lambda job_, f: any(
                f in files
                for j, files in self.depending[job_].items()
                if not self.finished(j) and self.needrun(j) and j != job
            )

            def unneeded_files():
                putative = (
                    lambda f: f.is_remote
                    and not f.protected
                    and not f.should_keep_local
                )

                generated_input = set()
                for job_, files in self.dependencies[job].items():
                    generated_input |= files
                    for f in filter(putative, files):
                        if not needed(job_, f):
                            yield f
                for f, f_ in zip(job.output, job.rule.output):
                    if putative(f) and not needed(job, f) and not f in self.targetfiles:
                        if f in job.dynamic_output:
                            for f_ in job.expand_dynamic(f_):
                                yield f_
                        else:
                            yield f
                for f in filter(putative, job.input):
                    # TODO what about remote inputs that are used by multiple jobs?
                    if f not in generated_input:
                        yield f

            for f in unneeded_files():
                if f.exists_local:
                    logger.info("Removing local copy of remote file: {}".format(f))
                    f.remove()

    def jobid(self, job):
        """Return job id of given job."""
        if job.is_group():
            return job.jobid
        else:
            return self._jobid[job]

    def update(
        self,
        jobs,
        file=None,
        visited=None,
        known_producers=None,
        skip_until_dynamic=False,
        progress=False,
        create_inventory=False,
    ):
        """Update the DAG by adding given jobs and their dependencies."""
        if visited is None:
            visited = set()
        if known_producers is None:
            known_producers = dict()
        producers = []
        exceptions = list()
        cycles = list()

        # check if all potential producers are strictly ordered
        jobs = sorted(jobs, reverse=True)
        discarded_jobs = set()

        def is_strictly_higher_ordered(pivot_job):
            return all(
                pivot_job > job
                for job in jobs
                if job is not pivot_job and job not in discarded_jobs
            )

        for i, job in enumerate(jobs):
            logger.dag_debug(dict(status="candidate", job=job))
            if file in job.input:
                cycles.append(job)
                continue
            if job in visited:
                cycles.append(job)
                continue
            try:
                self.check_periodic_wildcards(job)
                self.update_(
                    job,
                    visited=set(visited),
                    known_producers=known_producers,
                    skip_until_dynamic=skip_until_dynamic,
                    progress=progress,
                    create_inventory=create_inventory,
                )
                producers.append(job)
                if is_strictly_higher_ordered(job):
                    # All other jobs are either discarded or strictly less given
                    # any defined ruleorder.
                    break
            except (
                MissingInputException,
                CyclicGraphException,
                PeriodicWildcardError,
                WorkflowError,
            ) as ex:
                exceptions.append(ex)
                discarded_jobs.add(job)
            except RecursionError as e:
                raise WorkflowError(
                    e,
                    "If building the DAG exceeds the recursion limit, "
                    "this is likely due to a cyclic dependency."
                    "E.g. you might have a sequence of rules that "
                    "can generate their own input. Try to make "
                    "the output files more specific. "
                    "A common pattern is to have different prefixes "
                    "in the output files of different rules."
                    + "\nProblematic file pattern: {}".format(file)
                    if file
                    else "",
                )
        if not producers:
            if cycles:
                job = cycles[0]
                raise CyclicGraphException(job.rule, file, rule=job.rule)
            if len(exceptions) > 1:
                raise WorkflowError(*exceptions)
            elif len(exceptions) == 1:
                raise exceptions[0]

        n = len(self.dependencies)
        if progress and n % 1000 == 0 and n and self._progress != n:
            logger.info("Processed {} potential jobs.".format(n))
            self._progress = n

        producers.sort(reverse=True)
        producer = producers[0]
        ambiguities = list(
            filter(lambda x: not x < producer and not producer < x, producers[1:])
        )
        if ambiguities and not self.ignore_ambiguity:
            raise AmbiguousRuleException(file, producer, ambiguities[0])
        logger.dag_debug(dict(status="selected", job=producer))
        logger.dag_debug(
            dict(
                file=file,
                msg="Producer found, hence exceptions are ignored.",
                exception=WorkflowError(*exceptions),
            )
        )
        return producer

    def update_(
        self,
        job,
        visited=None,
        known_producers=None,
        skip_until_dynamic=False,
        progress=False,
        create_inventory=False,
    ):
        """Update the DAG by adding the given job and its dependencies."""
        if job in self.dependencies:
            return
        if visited is None:
            visited = set()
        if known_producers is None:
            known_producers = dict()
        visited.add(job)
        dependencies = self.dependencies[job]
        potential_dependencies = self.collect_potential_dependencies(
            job, known_producers=known_producers
        )

        skip_until_dynamic = skip_until_dynamic and not job.dynamic_output

        missing_input = set()
        producer = dict()
        exceptions = dict()
        for res in potential_dependencies:
            if create_inventory:
                # If possible, obtain inventory information starting from
                # given file and store it in the IOCache.
                # This should provide faster access to existence and mtime information
                # than querying file by file. If the file type does not support inventory
                # information, this call is a no-op.
                res.file.inventory()

            if not res.jobs:
                # no producing job found
                if not res.file.exists:
                    # file not found, hence missing input
                    missing_input.add(res.file)
                known_producers[res.file] = None
                # file found, no problem
                continue

            if res.known:
                producer[res.file] = res.jobs[0]
            else:
                try:
                    selected_job = self.update(
                        res.jobs,
                        file=res.file,
                        visited=visited,
                        known_producers=known_producers,
                        skip_until_dynamic=skip_until_dynamic
                        or res.file in job.dynamic_input,
                        progress=progress,
                    )
                    producer[res.file] = selected_job
                except (
                    MissingInputException,
                    CyclicGraphException,
                    PeriodicWildcardError,
                    WorkflowError,
                ) as ex:
                    if not res.file.exists:
                        self.delete_job(job, recursive=False)  # delete job from tree
                        raise ex
                    else:
                        logger.dag_debug(
                            dict(
                                file=res.file,
                                msg="No producers found, but file is present on disk.",
                                exception=ex,
                            )
                        )
                        known_producers[res.file] = None

        for file, job_ in producer.items():
            dependencies[job_].add(file)
            self.depending[job_][job].add(file)

        if self.is_batch_rule(job.rule) and self.batch.is_final:
            # For the final batch, ensure that all input files from
            # previous batches are present on disk.
            if any((f not in producer and not f.exists) for f in job.input):
                raise WorkflowError(
                    "Unable to execute batch {} because not all previous batches "
                    "have been completed before or files have been deleted.".format(
                        self.batch
                    )
                )

        if missing_input:
            self.delete_job(job, recursive=False)  # delete job from tree
            raise MissingInputException(job, missing_input)

        if skip_until_dynamic:
            self._dynamic.add(job)

    def update_needrun(self, create_inventory=False):
        """Update the information whether a job needs to be executed."""

        if create_inventory:
            # Concurrently collect mtimes of all existing files.
            self.workflow.iocache.mtime_inventory(self.jobs)

        output_mintime = dict()

        def update_output_mintime(job):
            try:
                return output_mintime[job]
            except KeyError:
                for job_ in chain([job], self.depending[job]):
                    try:
                        t = output_mintime[job_]
                    except KeyError:
                        t = job_.output_mintime
                    if t is not None:
                        output_mintime[job] = t
                        return
                output_mintime[job] = None

        is_same_checksum_cache = dict()

        def is_same_checksum(f, job):
            try:
                return is_same_checksum_cache[(f, job)]
            except KeyError:
                if not f.is_checksum_eligible():
                    # no chance to compute checksum, cannot be assumed the same
                    is_same = False
                else:
                    # obtain the input checksums for the given file for all output files of the job
                    checksums = self.workflow.persistence.input_checksums(job, f)
                    if len(checksums) > 1:
                        # more than one checksum recorded, cannot be all the same
                        is_same = False
                    elif not checksums:
                        # no checksums recorded, we cannot assume them to be the same
                        is_same = False
                    else:
                        is_same = f.is_same_checksum(checksums.pop())

                is_same_checksum_cache[(f, job)] = is_same
                return is_same

        def update_needrun(job):
            reason = self.reason(job)
            noinitreason = not reason
            updated_subworkflow_input = self.updated_subworkflow_files.intersection(
                job.input
            )

            if (
                job not in self.omitforce
                and job.rule in self.forcerules
                or not self.forcefiles.isdisjoint(job.output)
            ):
                reason.forced = True
            elif updated_subworkflow_input:
                reason.updated_input.update(updated_subworkflow_input)
            elif job in self.targetjobs:
                # TODO find a way to handle added/removed input files here?
                if not job.output and not job.benchmark:
                    if job.input:
                        if job.rule.norun:
                            reason.updated_input_run.update(
                                f for f in job.input if not f.exists
                            )
                        else:
                            reason.nooutput = True
                    else:
                        reason.noio = True
                else:
                    if job.rule in self.targetrules:
                        files = set(job.output)
                        if job.benchmark:
                            files.add(job.benchmark)
                    else:
                        files = set(chain(*self.depending[job].values()))
                        if self.targetfiles:
                            files.update(
                                f
                                for f in chain(job.output, job.log)
                                if f in self.targetfiles
                            )
                            if job.benchmark and job.benchmark in self.targetfiles:
                                files.add(job.benchmark)
                    reason.missing_output.update(job.missing_output(files))
            if not reason:
                output_mintime_ = output_mintime.get(job)
                updated_input = None
                if output_mintime_:
                    # Input is updated if it is newer that the oldest output file
                    # and does not have the same checksum as the one previously recorded.
                    updated_input = [
                        f
                        for f in job.input
                        if f.exists
                        and f.is_newer(output_mintime_)
                        and not is_same_checksum(f, job)
                    ]
                    reason.updated_input.update(updated_input)
                if not updated_input:
                    # check for other changes like parameters, set of input files, or code
                    if "params" in self.workflow.rerun_triggers:
                        reason.params_changed = any(
                            self.workflow.persistence.params_changed(job)
                        )
                    if "input" in self.workflow.rerun_triggers:
                        reason.input_changed = any(
                            self.workflow.persistence.input_changed(job)
                        )
                    if "code" in self.workflow.rerun_triggers:
                        reason.code_changed = any(
                            job.outputs_older_than_script_or_notebook()
                        ) or any(self.workflow.persistence.code_changed(job))
                    if "software-env" in self.workflow.rerun_triggers:
                        reason.software_stack_changed = any(
                            self.workflow.persistence.conda_env_changed(job)
                        ) or any(self.workflow.persistence.container_changed(job))

            if noinitreason and reason:
                reason.derived = False
            return reason

        reason = self.reason
        _needrun = self._needrun
        dependencies = self.dependencies
        depending = self.depending
        _n_until_ready = self._n_until_ready

        _needrun.clear()
        _n_until_ready.clear()
        self._ready_jobs.clear()

        candidates = list(self.toposorted())

        # Update the output mintime of all jobs.
        # We traverse them in BFS (level order) starting from target jobs.
        # Then, we check output mintime of job itself and all direct descendants,
        # which have already been visited in the level before.
        # This way, we achieve a linear runtime.
        for level in reversed(candidates):
            for job in level:
                update_output_mintime(job)

        # Update prior reason for all candidate jobs
        # Move from the first level to the last of the toposorted candidates.
        # If a job is needrun, mask all downstream jobs, they will below
        # in the bi-directional BFS
        # be determined as needrun because they depend on them.
        masked = set()
        queue = deque()
        for level in candidates:
            for job in level:
                if job in masked:
                    # depending jobs of jobs that are needrun as a prior
                    # can be skipped
                    continue

                if update_needrun(job):
                    queue.append(job)
                    masked.update(self.bfs(self.depending, job))

        # bi-directional BFS to determine further needrun jobs
        visited = set(queue)
        candidates_set = set(job for level in candidates for job in level)
        while queue:
            job = queue.popleft()
            _needrun.add(job)

            for job_, files in dependencies[job].items():
                missing_output = list(job_.missing_output(files))
                reason(job_).missing_output.update(missing_output)
                if missing_output and job_ not in visited:
                    visited.add(job_)
                    queue.append(job_)

            for job_, files in depending[job].items():
                if job_ in candidates_set:
                    if job_ not in visited:
                        # TODO may it happen that order determines whether
                        # _n_until_ready is incremented for this job?
                        if all(f.is_ancient and f.exists for f in files):
                            # No other reason to run job_.
                            # Since all files are ancient, we do not trigger it.
                            continue
                        visited.add(job_)
                        queue.append(job_)
                    _n_until_ready[job_] += 1
                    reason(job_).updated_input_run.update(files)

        # update len including finished jobs (because they have already increased the job counter)
        self._len = len(self._finished | self._needrun)

    def in_until(self, job):
        """Return whether given job has been specified via --until."""
        return job.rule.name in self.untilrules or not self.untilfiles.isdisjoint(
            job.output
        )

    def in_omitfrom(self, job):
        """Return whether given job has been specified via --omit-from."""
        return job.rule.name in self.omitrules or not self.omitfiles.isdisjoint(
            job.output
        )

    def until_jobs(self):
        """Returns a generator of jobs specified by untiljobs."""
        return (job for job in self.jobs if self.in_until(job))

    def omitfrom_jobs(self):
        """Returns a generator of jobs specified by omitfromjobs."""
        return (job for job in self.jobs if self.in_omitfrom(job))

    def downstream_of_omitfrom(self):
        """Returns the downstream of --omit-from rules or files and themselves."""
        return self.bfs(self.depending, *self.omitfrom_jobs())

    def delete_omitfrom_jobs(self):
        """Removes jobs downstream of jobs specified by --omit-from."""
        if not self.omitrules and not self.omitfiles:
            return
        downstream_jobs = list(
            self.downstream_of_omitfrom()
        )  # need to cast as list before deleting jobs
        for job in downstream_jobs:
            self.delete_job(job, recursive=False, add_dependencies=True)

    def set_until_jobs(self):
        """Removes jobs downstream of jobs specified by --omit-from."""
        if not self.untilrules and not self.untilfiles:
            return
        self.targetjobs = set(self.until_jobs())

    def update_priority(self):
        """Update job priorities."""
        prioritized = (
            lambda job: job.rule in self.priorityrules
            or not self.priorityfiles.isdisjoint(job.output)
        )
        for job in self.needrun_jobs:
            self._priority[job] = job.rule.priority
        for job in self.bfs(
            self.dependencies,
            *filter(prioritized, self.needrun_jobs),
            stop=self.noneedrun_finished,
        ):
            self._priority[job] = Job.HIGHEST_PRIORITY

    def update_groups(self):
        groups = dict()
        for job in self.needrun_jobs:
            if job.group is None:
                continue
            stop = lambda j: j.group != job.group
            # BFS into depending needrun jobs if in same group
            # Note: never go up here (into depending), because it may contain
            # jobs that have been sorted out due to e.g. ruleorder.
            group = self.group_job_factory.new(
                job.group,
                (
                    job
                    for job in self.bfs(self.dependencies, job, stop=stop)
                    if self.needrun(job)
                ),
            )

            # merge with previously determined groups if present
            for j in group:
                if j in groups:
                    other = groups[j]
                    other.merge(group)
                    group = other
            # update assignment
            for j in group:
                # Since groups might have been merged, we need
                # to update each job j in group.
                groups[j] = group

        self._group = groups

        self._update_group_components()

    def _update_group_components(self):
        # span connected components if requested
        groups_by_id = defaultdict(set)
        for group in self._group.values():
            groups_by_id[group.groupid].add(group)
        for groupid, conn_components in groups_by_id.items():
            n_components = self.workflow.group_components.get(groupid, 1)
            if n_components > 1:
                for chunk in group_into_chunks(n_components, conn_components):
                    if len(chunk) > 1:
                        primary = chunk[0]
                        for secondary in chunk[1:]:
                            primary.merge(secondary)
                        for j in primary:
                            self._group[j] = primary

        for group in self._group.values():
            group.finalize()

    def update_incomplete_input_expand_jobs(self):
        """Update (re-evaluate) all jobs which have incomplete input file expansions.

        only filled in the second pass of postprocessing.
        """
        updated = False
        for job in list(self.jobs):
            if job.incomplete_input_expand:
                newjob = job.updated()
                self.replace_job(job, newjob, recursive=False)
                updated = True
        return updated

    def update_ready(self, jobs=None):
        """Update information whether a job is ready to execute.

        Given jobs must be needrun jobs!
        """

        if jobs is None:
            jobs = self.needrun_jobs

        potential_new_ready_jobs = False
        candidate_groups = set()
        for job in jobs:
            if job in self._ready_jobs or job in self._running:
                # job has been seen before or is running, no need to process again
                continue
            if not self.finished(job) and self._ready(job):
                potential_new_ready_jobs = True
                if job.group is None:
                    self._ready_jobs.add(job)
                else:
                    group = self._group[job]
                    if group not in self._running:
                        candidate_groups.add(group)

        self._ready_jobs.update(
            group
            for group in candidate_groups
            if all(self._ready(job) for job in group)
        )
        return potential_new_ready_jobs

    def get_jobs_or_groups(self):
        visited_groups = set()
        for job in self.jobs:
            if job.group is None:
                yield job
            else:
                group = self._group[job]
                if group in visited_groups:
                    continue
                visited_groups.add(group)
                yield group

    def close_remote_objects(self):
        """Close all remote objects."""
        for job in self.jobs:
            if not self.needrun(job):
                job.close_remote()

    def postprocess(
        self, update_needrun=True, update_incomplete_input_expand_jobs=True
    ):
        """Postprocess the DAG. This has to be invoked after any change to the
        DAG topology."""
        self.cleanup()
        self.update_jobids()
        if update_needrun:
            self.update_needrun()
        self.update_priority()
        self.handle_pipes_and_services()
        self.update_groups()

        if update_incomplete_input_expand_jobs:
            updated = self.update_incomplete_input_expand_jobs()
            if updated:

                # run a second pass, some jobs have been updated
                # with potentially new input files that have depended
                # on group ids.
                self.postprocess(
                    update_needrun=True,
                    update_incomplete_input_expand_jobs=False,
                )

                return

        self.update_ready()
        self.close_remote_objects()
        self.update_checkpoint_outputs()

    def handle_pipes_and_services(self):
        """Use pipes and services to determine job groups. Check if every pipe has exactly
        one consumer"""

        visited = set()
        for job in self.needrun_jobs:
            candidate_groups = set()
            if job.group is not None:
                candidate_groups.add(job.group)
            all_depending = set()
            has_pipe_or_service = False
            for f in job.output:
                is_pipe = is_flagged(f, "pipe")
                is_service = is_flagged(f, "service")
                if is_pipe or is_service:
                    if job.is_run:
                        raise WorkflowError(
                            "Rule defines pipe output but "
                            "uses a 'run' directive. This is "
                            "not possible for technical "
                            "reasons. Consider using 'shell' or "
                            "'script'.",
                            rule=job.rule,
                        )

                    has_pipe_or_service = True
                    depending = [
                        j for j, files in self.depending[job].items() if f in files
                    ]
                    if is_pipe and len(depending) > 1:
                        raise WorkflowError(
                            "Output file {} is marked as pipe "
                            "but more than one job depends on "
                            "it. Make sure that any pipe "
                            "output is only consumed by one "
                            "job".format(f),
                            rule=job.rule,
                        )
                    elif len(depending) == 0:
                        raise WorkflowError(
                            "Output file {} is marked as pipe or service "
                            "but it has no consumer. This is "
                            "invalid because it can lead to "
                            "a dead lock.".format(f),
                            rule=job.rule,
                        )
                    elif is_pipe and depending[0].is_norun:
                        raise WorkflowError(
                            f"Output file {f} is marked as pipe but is requested by a rule that "
                            "does not execute anything. This is not allowed because it would lead "
                            "to a dead lock."
                        )

                    for dep in depending:
                        if dep.is_run:
                            raise WorkflowError(
                                "Rule consumes pipe or service input but "
                                "uses a 'run' directive. This is "
                                "not possible for technical "
                                "reasons. Consider using 'shell' or "
                                "'script'.",
                                rule=job.rule,
                            )

                        all_depending.add(dep)
                        if dep.group is not None:
                            candidate_groups.add(dep.group)
            if not has_pipe_or_service:
                continue

            if len(candidate_groups) > 1:
                if all(isinstance(group, CandidateGroup) for group in candidate_groups):
                    # all candidates are newly created groups, merge them into one
                    group = candidate_groups.pop()
                    for g in candidate_groups:
                        group.merge(g)
                else:
                    raise WorkflowError(
                        "An output file is marked as "
                        "pipe or service, but consuming jobs "
                        "are part of conflicting "
                        "groups.",
                        rule=job.rule,
                    )
            elif candidate_groups:
                # extend the candidate group to all involved jobs
                group = candidate_groups.pop()
            else:
                # generate a random unique group name
                group = CandidateGroup()

            # set group for job and all downstreams
            job.group = group
            visited.add(job)
            for j in all_depending:
                j.group = group
                visited.add(j)

        # convert candidate groups to plain string IDs
        for job in visited:
            job.group = (
                job.group.id if isinstance(job.group, CandidateGroup) else job.group
            )

    def _ready(self, job):
        """Return whether the given job is ready to execute."""
        group = self._group.get(job, None)
        if group is None:
            return self._n_until_ready[job] == 0
        else:
            n_internal_deps = lambda job: sum(
                self._group.get(dep) == group for dep in self.dependencies[job]
            )
            return all(
                (self._n_until_ready[job] - n_internal_deps(job)) == 0 for job in group
            )

    def update_checkpoint_dependencies(self, jobs=None):
        """Update dependencies of checkpoints."""
        updated = False
        self.update_checkpoint_outputs()
        if jobs is None:
            jobs = [job for job in self.jobs if not self.needrun(job)]
        for job in jobs:
            if job.is_checkpoint:
                depending = list(self.depending[job])
                # re-evaluate depending jobs, replace and update DAG
                for j in depending:
                    logger.debug("Updating job {}.".format(j))
                    newjob = j.updated()
                    self.replace_job(j, newjob, recursive=False)
                    updated = True
        if updated:
            self.postprocess()
        return updated

    def register_running(self, jobs):
        self._running.update(jobs)
        self._ready_jobs -= jobs
        for job in jobs:
            try:
                del self._n_until_ready[job]
            except KeyError:
                # already gone
                pass

    def finish(self, job, update_dynamic=True):
        """Finish a given job (e.g. remove from ready jobs, mark depending jobs
        as ready)."""

        self._running.remove(job)

        # turn off this job's Reason
        if job.is_group():
            for j in job:
                self.reason(j).mark_finished()
        else:
            self.reason(job).mark_finished()

        try:
            self._ready_jobs.remove(job)
        except KeyError:
            pass

        if job.is_group():
            jobs = job
        else:
            jobs = [job]

        self._finished.update(jobs)

        updated_dag = False
        if update_dynamic:
            updated_dag = self.update_checkpoint_dependencies(jobs)

        depending = [
            j
            for job in jobs
            for j in self.depending[job]
            if not self.in_until(job) and self.needrun(j)
        ]

        if not updated_dag:
            # Mark depending jobs as ready.
            # Skip jobs that are marked as until jobs.
            # This is not necessary if the DAG has been fully updated above.
            for job in depending:
                self._n_until_ready[job] -= 1

        potential_new_ready_jobs = self.update_ready(depending)

        for job in jobs:
            if update_dynamic and job.dynamic_output:
                logger.info("Dynamically updating jobs")
                newjob = self.update_dynamic(job)
                if newjob:
                    # simulate that this job ran and was finished before
                    self.omitforce.add(newjob)
                    self._needrun.add(newjob)
                    self._finished.add(newjob)
                    updated_dag = True

                    self.postprocess()
                    self.handle_protected(newjob)
                    self.handle_touch(newjob)

        if updated_dag:
            # We might have new jobs, so we need to ensure that all conda envs
            # and singularity images are set up.
            if self.workflow.use_singularity:
                self.pull_container_imgs()
            if self.workflow.use_conda:
                self.create_conda_envs()
            potential_new_ready_jobs = True

        self.handle_temp(job)

        return potential_new_ready_jobs

    def new_job(self, rule, targetfile=None, format_wildcards=None):
        """Create new job for given rule and (optional) targetfile.
        This will reuse existing jobs with the same wildcards."""
        key = (rule, targetfile)
        if key in self.job_cache:
            assert targetfile is not None
            return self.job_cache[key]
        wildcards_dict = rule.get_wildcards(targetfile)
        job = self.job_factory.new(
            rule,
            self,
            wildcards_dict=wildcards_dict,
            format_wildcards=format_wildcards,
            targetfile=targetfile,
        )
        self.cache_job(job)
        return job

    def cache_job(self, job):
        for f in job.products:
            self.job_cache[(job.rule, f)] = job

    def update_dynamic(self, job):
        """Update the DAG by evaluating the output of the given job that
        contains dynamic output files."""
        dynamic_wildcards = job.dynamic_wildcards
        if not dynamic_wildcards:
            # this happens e.g. in dryrun if output is not yet present
            return

        depending = list(
            filter(lambda job_: not self.finished(job_), self.bfs(self.depending, job))
        )
        newrule, non_dynamic_wildcards = job.rule.dynamic_branch(
            dynamic_wildcards, input=False
        )
        self.specialize_rule(job.rule, newrule)

        # no targetfile needed for job
        newjob = self.new_job(newrule, format_wildcards=non_dynamic_wildcards)
        self.replace_job(job, newjob)
        for job_ in depending:
            needs_update = any(
                f.get_wildcard_names() & dynamic_wildcards.keys()
                for f in job_.rule.dynamic_input
            )

            if needs_update:
                newrule_ = job_.rule.dynamic_branch(dynamic_wildcards)
                if newrule_ is not None:
                    self.specialize_rule(job_.rule, newrule_)
                    if not self.dynamic(job_):
                        logger.debug("Updating job {}.".format(job_))
                        newjob_ = self.new_job(
                            newrule_, targetfile=job_.output[0] if job_.output else None
                        )

                        unexpected_output = self.reason(
                            job_
                        ).missing_output.intersection(newjob.existing_output)
                        if unexpected_output:
                            logger.warning(
                                "Warning: the following output files of rule {} were not "
                                "present when the DAG was created:\n{}".format(
                                    newjob_.rule, unexpected_output
                                )
                            )

                        self.replace_job(job_, newjob_)
        return newjob

    def delete_job(self, job, recursive=True, add_dependencies=False):
        """Delete given job from DAG."""
        if job in self.targetjobs:
            self.targetjobs.remove(job)
        if add_dependencies:
            for _job in self.dependencies[job]:
                self.targetjobs.add(_job)
        for job_ in self.depending[job]:
            del self.dependencies[job_][job]
        del self.depending[job]
        for job_ in self.dependencies[job]:
            depending = self.depending[job_]
            del depending[job]
            if not depending and recursive:
                self.delete_job(job_)
        del self.dependencies[job]
        if job in self._needrun:
            self._len -= 1
            self._needrun.remove(job)
            del self._reason[job]
        if job in self._finished:
            self._finished.remove(job)
        if job in self._dynamic:
            self._dynamic.remove(job)
        if job in self._ready_jobs:
            self._ready_jobs.remove(job)
        if job in self._n_until_ready:
            del self._n_until_ready[job]
        # remove from cache
        for f in job.output:
            try:
                del self.job_cache[(job.rule, f)]
            except KeyError:
                pass

    def replace_job(self, job, newjob, recursive=True):
        """Replace given job with new job."""
        add_to_targetjobs = job in self.targetjobs
        try:
            jobid = self.jobid(job)
        except KeyError:
            # Job has been added while updating another checkpoint,
            # jobid is not yet known.
            jobid = None

        depending = list(self.depending[job].items())
        if self.finished(job):
            self._finished.add(newjob)

        self.delete_job(job, recursive=recursive)
        if jobid is not None:
            self._jobid[newjob] = jobid

        if add_to_targetjobs:
            self.targetjobs.add(newjob)

        self.cache_job(newjob)

        self.update([newjob])

        logger.debug("Replace {} with dynamic branch {}".format(job, newjob))
        for job_, files in depending:
            # if not job_.dynamic_input:
            logger.debug("updating depending job {}".format(job_))
            self.dependencies[job_][newjob].update(files)
            self.depending[newjob][job_].update(files)

    def specialize_rule(self, rule, newrule):
        """Specialize the given rule by inserting newrule into the DAG."""
        assert newrule is not None
        self.rules.add(newrule)
        self.update_output_index()

    def is_batch_rule(self, rule):
        """Return True if the underlying rule is to be used for batching the DAG."""
        return self.batch is not None and rule.name == self.batch.rulename

    def collect_potential_dependencies(self, job, known_producers):
        """Collect all potential dependencies of a job. These might contain
        ambiguities. The keys of the returned dict represent the files to be considered."""
        # use a set to circumvent multiple jobs for the same file
        # if user specified it twice
        file2jobs = self.file2jobs

        input_files = list(job.unique_input)

        if self.is_batch_rule(job.rule):
            # only consider the defined partition of the input files
            input_batch = self.batch.get_batch(input_files)
            if len(input_batch) != len(input_files):
                logger.info(
                    "Considering only batch {} for DAG computation.\n"
                    "All jobs beyond the batching rule are omitted until the final batch.\n"
                    "Don't forget to run the other batches too.".format(self.batch)
                )
                input_files = input_batch

        for file in input_files:
            # omit the file if it comes from a subworkflow
            if file in job.subworkflow_input:
                continue

            try:
                yield PotentialDependency(file, known_producers[file], True)
            except KeyError:
                try:
                    if file in job.dependencies:
                        yield PotentialDependency(
                            file,
                            [self.new_job(job.dependencies[file], targetfile=file)],
                            False,
                        )
                    else:
                        yield PotentialDependency(file, file2jobs(file), False)
                except MissingRuleException as ex:
                    # no dependency found
                    yield PotentialDependency(file, None, False)

    def bfs(self, direction, *jobs, stop=lambda job: False):
        """Perform a breadth-first traversal of the DAG."""
        queue = deque(jobs)
        visited = set(queue)
        while queue:
            job = queue.popleft()
            if stop(job):
                # stop criterion reached for this node
                continue
            yield job
            for job_ in direction[job].keys():
                if not job_ in visited:
                    queue.append(job_)
                    visited.add(job_)

    def level_bfs(self, direction, *jobs, stop=lambda job: False):
        """Perform a breadth-first traversal of the DAG, but also yield the
        level together with each job."""
        queue = [(job, 0) for job in jobs]
        visited = set(jobs)
        while queue:
            job, level = queue.pop(0)
            if stop(job):
                # stop criterion reached for this node
                continue
            yield level, job
            level += 1
            for job_, _ in direction[job].items():
                if not job_ in visited:
                    queue.append((job_, level))
                    visited.add(job_)

    def dfs(self, direction, *jobs, stop=lambda job: False, post=True):
        """Perform depth-first traversal of the DAG."""
        visited = set()

        def _dfs(job):
            """Inner function for DFS traversal."""
            if stop(job):
                return
            if not post:
                yield job
            for job_ in direction[job]:
                if not job_ in visited:
                    visited.add(job_)
                    for j in _dfs(job_):
                        yield j
            if post:
                yield job

        for job in jobs:
            for job_ in self._dfs(direction, job, visited, stop=stop, post=post):
                yield job_

    def new_wildcards(self, job):
        """Return wildcards that are newly introduced in this job,
        compared to its ancestors."""
        new_wildcards = set(job.wildcards.items())
        for job_ in self.dependencies[job]:
            if not new_wildcards:
                return set()
            for wildcard in job_.wildcards.items():
                new_wildcards.discard(wildcard)
        return new_wildcards

    def rule2job(self, targetrule):
        """Generate a new job from a given rule."""
        if targetrule.has_wildcards():
            raise WorkflowError(
                "Target rules may not contain wildcards. "
                "Please specify concrete files or a rule without wildcards at the command line, "
                "or have a rule without wildcards at the very top of your workflow (e.g. the typical "
                '"rule all" which just collects all results you want to generate in the end).'
            )
        return self.new_job(targetrule)

    def file2jobs(self, targetfile):
        rules = self.output_index.match(targetfile)
        jobs = []
        exceptions = list()
        for rule in rules:
            if rule.is_producer(targetfile):
                try:
                    jobs.append(self.new_job(rule, targetfile=targetfile))
                except InputFunctionException as e:
                    exceptions.append(e)
        if not jobs:
            if exceptions:
                raise exceptions[0]
            raise MissingRuleException(targetfile)
        return jobs

    def rule_dot2(self):
        dag = defaultdict(list)
        visited = set()
        preselect = set()

        def preselect_parents(job):
            for parent in self.depending[job]:
                if parent in preselect:
                    continue
                preselect.add(parent)
                preselect_parents(parent)

        def build_ruledag(job, key=lambda job: job.rule.name):
            if job in visited:
                return
            visited.add(job)
            deps = sorted(self.dependencies[job], key=key)
            deps = [
                (
                    group[0]
                    if preselect.isdisjoint(group)
                    else preselect.intersection(group).pop()
                )
                for group in (list(g) for _, g in groupby(deps, key))
            ]
            dag[job].extend(deps)
            preselect_parents(job)
            for dep in deps:
                build_ruledag(dep)

        for job in self.targetjobs:
            build_ruledag(job)

        return self._dot(dag.keys())

    def rule_dot(self):
        graph = defaultdict(set)
        for job in self.jobs:
            graph[job.rule].update(dep.rule for dep in self.dependencies[job])
        return self._dot(graph)

    def dot(self):
        def node2style(job):
            if not self.needrun(job):
                return "rounded,dashed"
            if self.dynamic(job) or job.dynamic_input:
                return "rounded,dotted"
            return "rounded"

        def format_wildcard(wildcard):
            name, value = wildcard
            if DYNAMIC_FILL in value:
                value = "..."
            return "{}: {}".format(name, value)

        node2rule = lambda job: job.rule
        node2label = lambda job: "\\n".join(
            chain(
                [job.rule.name], sorted(map(format_wildcard, self.new_wildcards(job)))
            )
        )

        dag = {job: self.dependencies[job] for job in self.jobs}

        return self._dot(
            dag, node2rule=node2rule, node2style=node2style, node2label=node2label
        )

    def _dot(
        self,
        graph,
        node2rule=lambda node: node,
        node2style=lambda node: "rounded",
        node2label=lambda node: node,
    ):

        # color rules
        huefactor = 2 / (3 * len(self.rules))
        rulecolor = {
            rule: "{:.2f} 0.6 0.85".format(i * huefactor)
            for i, rule in enumerate(self.rules)
        }

        # markup
        node_markup = '\t{}[label = "{}", color = "{}", style="{}"];'.format
        edge_markup = "\t{} -> {}".format

        # node ids
        ids = {node: i for i, node in enumerate(graph)}

        # calculate nodes
        nodes = [
            node_markup(
                ids[node],
                node2label(node),
                rulecolor[node2rule(node)],
                node2style(node),
            )
            for node in graph
        ]
        # calculate edges
        edges = [
            edge_markup(ids[dep], ids[node])
            for node, deps in graph.items()
            for dep in deps
        ]

        return textwrap.dedent(
            """\
            digraph snakemake_dag {{
                graph[bgcolor=white, margin=0];
                node[shape=box, style=rounded, fontname=sans, \
                fontsize=10, penwidth=2];
                edge[penwidth=2, color=grey];
            {items}
            }}\
            """
        ).format(items="\n".join(nodes + edges))

    def filegraph_dot(
        self,
        node2rule=lambda node: node,
        node2style=lambda node: "rounded",
        node2label=lambda node: node,
    ):

        # NOTE: This is code from the rule_dot method.
        # This method could be split like there as well, however,
        # it cannot easily reuse the _dot method due to the different node type
        graph = defaultdict(set)
        for job in self.jobs:
            graph[job.rule].update(dep.rule for dep in self.dependencies[job])

        # node ids
        ids = {node: i for i, node in enumerate(graph)}

        # Compute colors for rules
        def hsv_to_htmlhexrgb(h, s, v):
            """Convert hsv colors to hex-encoded rgb colors usable by html."""
            import colorsys

            hex_r, hex_g, hex_b = (round(255 * x) for x in colorsys.hsv_to_rgb(h, s, v))
            return "#{hex_r:0>2X}{hex_g:0>2X}{hex_b:0>2X}".format(
                hex_r=hex_r, hex_g=hex_g, hex_b=hex_b
            )

        huefactor = 2 / (3 * len(self.rules))
        rulecolor = {
            rule: hsv_to_htmlhexrgb(i * huefactor, 0.6, 0.85)
            for i, rule in enumerate(self.rules)
        }

        def resolve_input_functions(input_files):
            """Iterate over all input files and replace input functions
            with a fixed string.
            """
            files = []
            for f in input_files:
                if callable(f):
                    files.append("<input function>")
                    # NOTE: This is a workaround. It would be more informative
                    # to show the code of the input function here (if it is
                    # short enough). This cannot be easily done with the inspect
                    # module, since the line numbers in the Snakefile do not
                    # behave as expected. One (complicated) solution for this
                    # would be to find the Snakefile and directly extract the
                    # code of the function.
                else:
                    files.append(repr(f).strip("'"))
            return files

        def html_node(node_id, node, color):
            """Assemble a html style node for graphviz"""
            input_files = resolve_input_functions(node._input)
            output_files = [repr(f).strip("'") for f in node._output]
            input_header = (
                '<b><font point-size="14">&#8618; input</font></b>'
                if input_files
                else ""
            )
            output_header = (
                '<b><font point-size="14">output &rarr;</font></b>'
                if output_files
                else ""
            )
            html_node = [
                '{node_id} [ shape=none, margin=0, label=<<table border="2" color="{color}" cellspacing="3" cellborder="0">'.format(
                    node_id=node_id, color=color
                ),
                "<tr><td>",
                '<b><font point-size="18">{node.name}</font></b>'.format(node=node),
                "</td></tr>",
                "<hr/>",
                '<tr><td align="left"> {input_header} </td></tr>'.format(
                    input_header=input_header
                ),
            ]

            for filename in sorted(input_files):
                # Escape html relevant chars like '<' and '>' in filenames
                # These can be added by input functions etc. and cannot be
                # displayed in graphviz HTML nodes.
                in_file = html.escape(filename)
                html_node.extend(
                    [
                        "<tr>",
                        '<td align="left"><font face="monospace">{in_file}</font></td>'.format(
                            in_file=in_file
                        ),
                        "</tr>",
                    ]
                )

            html_node.append("<hr/>")
            html_node.append(
                '<tr><td align="right"> {output_header} </td> </tr>'.format(
                    output_header=output_header
                )
            )

            for filename in sorted(output_files):
                out_file = html.escape(filename)
                html_node.extend(
                    [
                        "<tr>",
                        '<td align="left"><font face="monospace">{out_file}</font></td>'
                        "</tr>".format(out_file=out_file),
                    ]
                )

            html_node.append("</table>>]")
            return "\n".join(html_node)

        nodes = [
            html_node(ids[node], node, rulecolor[node2rule(node)]) for node in graph
        ]

        # calculate edges
        edge_markup = "\t{} -> {}".format
        edges = [
            edge_markup(ids[dep], ids[node], ids[dep], ids[node])
            for node, deps in graph.items()
            for dep in deps
        ]

        return textwrap.dedent(
            """\
            digraph snakemake_dag {{
                graph[bgcolor=white, margin=0];
                node[shape=box, style=rounded, fontname=sans, \
                fontsize=10, penwidth=2];
                edge[penwidth=2, color=grey];
            {items}
            }}\
            """
        ).format(items="\n".join(nodes + edges))

    def summary(self, detailed=False):
        if detailed:
            yield "output_file\tdate\trule\tversion\tlog-file(s)\tinput-file(s)\tshellcmd\tstatus\tplan"
        else:
            yield "output_file\tdate\trule\tversion\tlog-file(s)\tstatus\tplan"

        for job in self.jobs:
            output = job.rule.output if self.dynamic(job) else job.expanded_output
            for f in output:
                rule = self.workflow.persistence.rule(f)
                rule = "-" if rule is None else rule

                version = self.workflow.persistence.version(f)
                version = "-" if version is None else str(version)

                date = time.ctime(f.mtime.local_or_remote()) if f.exists else "-"

                pending = "update pending" if self.reason(job) else "no update"

                log = self.workflow.persistence.log(f)
                log = "-" if log is None else ",".join(log)

                input = self.workflow.persistence.input(f)
                input = "-" if input is None else ",".join(input)

                shellcmd = self.workflow.persistence.shellcmd(f)
                shellcmd = "-" if shellcmd is None else shellcmd
                # remove new line characters, leading and trailing whitespace
                shellcmd = shellcmd.strip().replace("\n", "; ")

                status = "ok"
                if not f.exists:
                    status = "missing"
                elif self.reason(job).updated_input:
                    status = "updated input files"
                elif self.workflow.persistence.version_changed(job, file=f):
                    status = "version changed to {}".format(job.rule.version)
                elif self.workflow.persistence.code_changed(job, file=f):
                    status = "rule implementation changed"
                elif self.workflow.persistence.input_changed(job, file=f):
                    status = "set of input files changed"
                elif self.workflow.persistence.params_changed(job, file=f):
                    status = "params changed"
                if detailed:
                    yield "\t".join(
                        (f, date, rule, version, log, input, shellcmd, status, pending)
                    )
                else:
                    yield "\t".join((f, date, rule, version, log, status, pending))

    def archive(self, path):
        """Archives workflow such that it can be re-run on a different system.

        Archiving includes git versioned files (i.e. Snakefiles, config files, ...),
        ancestral input files and conda environments.
        """
        if path.endswith(".tar"):
            mode = "x"
        elif path.endswith("tar.bz2"):
            mode = "x:bz2"
        elif path.endswith("tar.xz"):
            mode = "x:xz"
        elif path.endswith("tar.gz"):
            mode = "x:gz"
        else:
            raise WorkflowError(
                "Unsupported archive format "
                "(supported: .tar, .tar.gz, .tar.bz2, .tar.xz)"
            )
        if os.path.exists(path):
            raise WorkflowError("Archive already exists:\n" + path)

        self.create_conda_envs(forceall=True)

        try:
            workdir = Path(os.path.abspath(os.getcwd()))
            with tarfile.open(path, mode=mode, dereference=True) as archive:
                archived = set()

                def add(path):
                    if workdir not in Path(os.path.abspath(path)).parents:
                        logger.warning(
                            "Path {} cannot be archived: "
                            "not within working directory.".format(path)
                        )
                    else:
                        f = os.path.relpath(path)
                        if f not in archived:
                            archive.add(f)
                            archived.add(f)
                            logger.info("archived " + f)

                logger.info(
                    "Archiving snakefiles, scripts and files under "
                    "version control..."
                )
                for f in self.workflow.get_sources():
                    add(f)

                logger.info("Archiving external input files...")
                for job in self.jobs:
                    # input files
                    for f in job.input:
                        if not any(
                            f in files for files in self.dependencies[job].values()
                        ):
                            # this is an input file that is not created by any job
                            add(f)

                logger.info("Archiving conda environments...")
                envs = set()
                for job in self.jobs:
                    if job.conda_env_spec:
                        env_archive = job.archive_conda_env()
                        envs.add(env_archive)
                for env in envs:
                    add(env)

        except (Exception, BaseException) as e:
            os.remove(path)
            raise e

    def clean(self, only_temp=False, dryrun=False):
        """Removes files generated by the workflow."""
        for job in self.jobs:
            for f in job.output:
                if not only_temp or is_flagged(f, "temp"):
                    # The reason for the second check is that dangling
                    # symlinks fail f.exists.
                    if f.exists or os.path.islink(f):
                        if f.protected:
                            logger.error("Skipping write-protected file {}.".format(f))
                        else:
                            msg = "Deleting {}" if not dryrun else "Would delete {}"
                            logger.info(msg.format(f))
                            if not dryrun:
                                # Remove non-empty dirs if flagged as temp()
                                f.remove(remove_non_empty_dir=only_temp)

    def list_untracked(self):
        """List files in the workdir that are not in the dag."""
        used_files = set()
        files_in_cwd = set()
        for job in self.jobs:
            used_files.update(
                os.path.relpath(file)
                for file in chain(job.local_input, job.local_output, job.log)
            )
        for root, dirs, files in os.walk(os.getcwd()):
            # Ignore hidden files and don't traverse into hidden dirs
            files_in_cwd.update(
                [
                    os.path.relpath(os.path.join(root, f))
                    for f in files
                    if not f[0] == "."
                ]
            )
            dirs[:] = [d for d in dirs if not d[0] == "."]
        for f in sorted(list(files_in_cwd - used_files)):
            logger.info(f)

    def d3dag(self, max_jobs=10000):
        def node(job):
            jobid = self.jobid(job)
            return {
                "id": jobid,
                "value": {
                    "jobid": jobid,
                    "label": job.rule.name,
                    "rule": job.rule.name,
                },
            }

        def edge(a, b):
            return {"u": self.jobid(a), "v": self.jobid(b)}

        jobs = list(self.jobs)

        if len(jobs) > max_jobs:
            logger.info(
                "Job-DAG is too large for visualization (>{} jobs).".format(max_jobs)
            )
        else:
            logger.d3dag(
                nodes=[node(job) for job in jobs],
                edges=[
                    edge(dep, job)
                    for job in jobs
                    for dep in self.dependencies[job]
                    if self.needrun(dep)
                ],
            )

    def stats(self):
        from tabulate import tabulate

        rules = Counter()
        rules.update(job.rule for job in self.needrun_jobs)
        rules.update(job.rule for job in self.finished_jobs)

        max_threads = defaultdict(int)
        min_threads = defaultdict(lambda: sys.maxsize)
        for job in chain(self.needrun_jobs, self.finished_jobs):
            max_threads[job.rule] = max(max_threads[job.rule], job.threads)
            min_threads[job.rule] = min(min_threads[job.rule], job.threads)
        rows = [
            {
                "job": rule,
                "count": count,
                "min threads": min_threads[rule],
                "max threads": max_threads[rule],
            }
            for rule, count in sorted(
                rules.most_common(), key=lambda item: item[0].name
            )
        ]
        rows.append(
            {
                "job": "total",
                "count": sum(rules.values()),
                "min threads": min(min_threads.values()),
                "max threads": max(max_threads.values()),
            }
        )

        yield "Job stats:"
        yield tabulate(rows, headers="keys")
        yield ""

    def toposorted(self, jobs=None, inherit_pipe_dependencies=False):
        from toposort import toposort

        if jobs is None:
            jobs = set(self.jobs)

        def get_dependencies(job):
            for dep, files in self.dependencies[job].items():
                if dep in jobs:
                    yield dep
                    if inherit_pipe_dependencies and any(
                        is_flagged(f, "pipe") for f in files
                    ):
                        # In case of a pipe, inherit the dependencies of the producer,
                        # such that the two jobs end up on the same toposort level.
                        # This is important because they are executed simulataneously.
                        yield from get_dependencies(dep)

        dag = {job: set(get_dependencies(job)) for job in jobs}

        return toposort(dag)

    def get_outputs_with_changes(self, change_type, include_needrun=True):
        is_changed = lambda job: (
            getattr(self.workflow.persistence, f"{change_type}_changed")(job)
            if not job.is_group() and (include_needrun or not self.needrun(job))
            else []
        )
        changed = list(chain(*map(is_changed, self.jobs)))
        if change_type == "code":
            for job in self.jobs:
                if not job.is_group() and (include_needrun or not self.needrun(job)):
                    changed.extend(list(job.outputs_older_than_script_or_notebook()))
        return changed

    def __str__(self):
        return self.dot()

    def __len__(self):
        return self._len


class CandidateGroup:
    def __init__(self):
        self.id = str(uuid.uuid4())

    def __eq__(self, other):
        return self.id == other.id

    def __hash__(self):
        return hash(self.id)

    def merge(self, other):
        self.id = other.id