__author__ = "Johannes Köster"
__copyright__ = "Copyright 2015-2019, Johannes Köster"
__email__ = "koester@jimmy.harvard.edu"
__license__ = "MIT"
import inspect
import itertools
import os
import tempfile
import textwrap
import sys
import pickle
import subprocess
import collections
import re
from abc import ABC, abstractmethod
from urllib.request import urlopen, pathname2url
from urllib.error import URLError
from snakemake.utils import format
from snakemake.logging import logger
from snakemake.exceptions import WorkflowError
from snakemake.shell import shell
from snakemake.common import MIN_PY_VERSION, SNAKEMAKE_SEARCHPATH
from snakemake.io import git_content, split_git_path
from snakemake.deployment import singularity
PY_VER_RE = re.compile("Python (?P<ver_min>\d+\.\d+).*")
# TODO use this to find the right place for inserting the preamble
PY_PREAMBLE_RE = re.compile(r"from( )+__future__( )+import.*?(?P<end>[;\n])")
[docs]class Snakemake:
def __init__(
self,
input_,
output,
params,
wildcards,
threads,
resources,
log,
config,
rulename,
bench_iteration,
scriptdir=None,
):
# convert input and output to plain strings as some remote objects cannot
# be pickled
self.input = input_._plainstrings()
self.output = output._plainstrings()
self.params = params
self.wildcards = wildcards
self.threads = threads
self.resources = resources
self.log = log._plainstrings()
self.config = config
self.rule = rulename
self.bench_iteration = bench_iteration
self.scriptdir = scriptdir
[docs] def log_fmt_shell(self, stdout=True, stderr=True, append=False):
"""
Return a shell redirection string to be used in `shell()` calls
This function allows scripts and wrappers support optional `log` files
specified in the calling rule. If no `log` was specified, then an
empty string "" is returned, regardless of the values of `stdout`,
`stderr`, and `append`.
Parameters
---------
stdout : bool
Send stdout to log
stderr : bool
Send stderr to log
append : bool
Do not overwrite the log file. Useful for sending output of
multiple commands to the same log. Note however that the log will
not be truncated at the start.
The following table describes the output:
-------- -------- -------- ----- -------------
stdout stderr append log return value
-------- -------- -------- ----- ------------
True True True fn >> fn 2>&1
True False True fn >> fn
False True True fn 2>> fn
True True False fn > fn 2>&1
True False False fn > fn
False True False fn 2> fn
any any any None ""
-------- -------- -------- ----- -----------
"""
if not self.log:
return ""
lookup = {
(True, True, True): " >> {0} 2>&1",
(True, False, True): " >> {0}",
(False, True, True): " 2>> {0}",
(True, True, False): " > {0} 2>&1",
(True, False, False): " > {0}",
(False, True, False): " 2> {0}",
}
return lookup[(stdout, stderr, append)].format(self.log)
[docs]class REncoder:
"""Encoding Pyton data structures into R."""
[docs] @classmethod
def encode_numeric(cls, value):
if value is None:
return "as.numeric(NA)"
return str(value)
[docs] @classmethod
def encode_value(cls, value):
if value is None:
return "NULL"
elif isinstance(value, str):
return repr(value)
elif isinstance(value, dict):
return cls.encode_dict(value)
elif isinstance(value, bool):
return "TRUE" if value else "FALSE"
elif isinstance(value, int) or isinstance(value, float):
return str(value)
elif isinstance(value, collections.abc.Iterable):
# convert all iterables to vectors
return cls.encode_list(value)
else:
# Try to convert from numpy if numpy is present
try:
import numpy as np
if isinstance(value, np.number):
return str(value)
except ImportError:
pass
raise ValueError("Unsupported value for conversion into R: {}".format(value))
[docs] @classmethod
def encode_list(cls, l):
return "c({})".format(", ".join(map(cls.encode_value, l)))
[docs] @classmethod
def encode_items(cls, items):
def encode_item(item):
name, value = item
return '"{}" = {}'.format(name, cls.encode_value(value))
return ", ".join(map(encode_item, items))
[docs] @classmethod
def encode_dict(cls, d):
d = "list({})".format(cls.encode_items(d.items()))
return d
[docs] @classmethod
def encode_namedlist(cls, namedlist):
positional = ", ".join(map(cls.encode_value, namedlist))
named = cls.encode_items(namedlist.items())
source = "list("
if positional:
source += positional
if named:
source += ", " + named
source += ")"
return source
[docs]class JuliaEncoder:
"""Encoding Pyton data structures into Julia."""
[docs] @classmethod
def encode_value(cls, value):
if value is None:
return "nothing"
elif isinstance(value, str):
return repr(value)
elif isinstance(value, dict):
return cls.encode_dict(value)
elif isinstance(value, bool):
return "true" if value else "false"
elif isinstance(value, int) or isinstance(value, float):
return str(value)
elif isinstance(value, collections.abc.Iterable):
# convert all iterables to vectors
return cls.encode_list(value)
else:
# Try to convert from numpy if numpy is present
try:
import numpy as np
if isinstance(value, np.number):
return str(value)
except ImportError:
pass
raise ValueError(
"Unsupported value for conversion into Julia: {}".format(value)
)
[docs] @classmethod
def encode_list(cls, l):
return "[{}]".format(", ".join(map(cls.encode_value, l)))
[docs] @classmethod
def encode_items(cls, items):
def encode_item(item):
name, value = item
return '"{}" => {}'.format(name, cls.encode_value(value))
return ", ".join(map(encode_item, items))
[docs] @classmethod
def encode_positional_items(cls, namedlist):
encoded = ""
for index, value in enumerate(namedlist):
encoded += "{} => {}, ".format(index + 1, cls.encode_value(value))
return encoded
[docs] @classmethod
def encode_dict(cls, d):
d = "Dict({})".format(cls.encode_items(d.items()))
return d
[docs] @classmethod
def encode_namedlist(cls, namedlist):
positional = cls.encode_positional_items(namedlist)
named = cls.encode_items(namedlist.items())
source = "Dict("
if positional:
source += positional
if named:
source += named
source += ")"
return source
[docs]class ScriptBase(ABC):
def __init__(
self,
path,
source,
basedir,
input_,
output,
params,
wildcards,
threads,
resources,
log,
config,
rulename,
conda_env,
container_img,
singularity_args,
env_modules,
bench_record,
jobid,
bench_iteration,
cleanup_scripts,
shadow_dir,
):
self.path = path
self.source = source
self.basedir = basedir
self.input = input_
self.output = output
self.params = params
self.wildcards = wildcards
self.threads = threads
self.resources = resources
self.log = log
self.config = config
self.rulename = rulename
self.conda_env = conda_env
self.container_img = container_img
self.singularity_args = singularity_args
self.env_modules = env_modules
self.bench_record = bench_record
self.jobid = jobid
self.bench_iteration = bench_iteration
self.cleanup_scripts = cleanup_scripts
self.shadow_dir = shadow_dir
[docs] def evaluate(self):
fd = None
try:
# generate preamble
preamble = self.get_preamble()
# write script
dir_ = ".snakemake/scripts"
os.makedirs(dir_, exist_ok=True)
with tempfile.NamedTemporaryFile(
suffix="." + os.path.basename(self.path), dir=dir_, delete=False
) as fd:
self.write_script(preamble, fd)
# execute script
self.execute_script(fd.name)
except URLError as e:
raise WorkflowError(e)
finally:
if fd and self.cleanup_scripts:
os.remove(fd.name)
else:
if fd:
logger.warning("Not cleaning up %s" % fd.name)
else:
# nothing to clean up (TODO: ??)
pass
[docs] @abstractmethod
def get_preamble(self):
...
[docs] @abstractmethod
def write_script(self, preamble, fd):
...
[docs] @abstractmethod
def execute_script(self, fname):
...
def _execute_cmd(self, cmd, **kwargs):
shell(
cmd,
bench_record=self.bench_record,
conda_env=self.conda_env,
container_img=self.container_img,
shadow_dir=self.shadow_dir,
env_modules=self.env_modules,
**kwargs
)
[docs]class PythonScript(ScriptBase):
[docs] @staticmethod
def generate_preamble(
path,
source,
basedir,
input_,
output,
params,
wildcards,
threads,
resources,
log,
config,
rulename,
conda_env,
container_img,
singularity_args,
env_modules,
bench_record,
jobid,
bench_iteration,
cleanup_scripts,
shadow_dir,
preamble_addendum="",
):
wrapper_path = path[7:] if path.startswith("file://") else path
snakemake = Snakemake(
input_,
output,
params,
wildcards,
threads,
resources,
log,
config,
rulename,
bench_iteration,
os.path.dirname(wrapper_path),
)
snakemake = pickle.dumps(snakemake)
# Obtain search path for current snakemake module.
# The module is needed for unpickling in the script.
# We append it at the end (as a fallback).
searchpath = SNAKEMAKE_SEARCHPATH
if container_img is not None:
searchpath = singularity.SNAKEMAKE_MOUNTPOINT
searchpath = repr(searchpath)
# For local scripts, add their location to the path in case they use path-based imports
if path.startswith("file://"):
searchpath += ", " + repr(os.path.dirname(path[7:]))
return textwrap.dedent(
"""
######## Snakemake header ########
import sys; sys.path.extend([{searchpath}]); import pickle; snakemake = pickle.loads({snakemake}); from snakemake.logging import logger; logger.printshellcmds = {printshellcmds}; {preamble_addendum}
######## Original script #########
"""
).format(
searchpath=searchpath,
snakemake=snakemake,
printshellcmds=logger.printshellcmds,
preamble_addendum=preamble_addendum,
)
[docs] def get_preamble(self):
wrapper_path = self.path[7:] if self.path.startswith("file://") else self.path
preamble_addendum = "__real_file__ = __file__; __file__ = {file_override};".format(
file_override=repr(os.path.realpath(wrapper_path))
)
return PythonScript.generate_preamble(
self.path,
self.source,
self.basedir,
self.input,
self.output,
self.params,
self.wildcards,
self.threads,
self.resources,
self.log,
self.config,
self.rulename,
self.conda_env,
self.container_img,
self.singularity_args,
self.env_modules,
self.bench_record,
self.jobid,
self.bench_iteration,
self.cleanup_scripts,
self.shadow_dir,
preamble_addendum=preamble_addendum,
)
[docs] def write_script(self, preamble, fd):
fd.write(preamble.encode())
fd.write(self.source)
[docs] def execute_script(self, fname):
py_exec = sys.executable
if self.conda_env is not None:
py = os.path.join(self.conda_env, "bin", "python")
if os.path.exists(py):
out = subprocess.check_output(
[py, "--version"], stderr=subprocess.STDOUT, universal_newlines=True
)
ver = tuple(map(int, PY_VER_RE.match(out).group("ver_min").split(".")))
if ver >= MIN_PY_VERSION:
# Python version is new enough, make use of environment
# to execute script
py_exec = "python"
else:
logger.warning(
"Conda environment defines Python "
"version < {0}.{1}. Using Python of the "
"master process to execute "
"script. Note that this cannot be avoided, "
"because the script uses data structures from "
"Snakemake which are Python >={0}.{1} "
"only.".format(*MIN_PY_VERSION)
)
if self.container_img is not None:
# use python from image
py_exec = "python"
if self.env_modules is not None:
# use python from environment module
py_exec = "python"
# use the same Python as the running process or the one from the environment
self._execute_cmd("{py_exec} {fname:q}", py_exec=py_exec, fname=fname)
[docs]class RScript(ScriptBase):
[docs] @staticmethod
def generate_preamble(
path,
source,
basedir,
input_,
output,
params,
wildcards,
threads,
resources,
log,
config,
rulename,
conda_env,
container_img,
singularity_args,
env_modules,
bench_record,
jobid,
bench_iteration,
cleanup_scripts,
shadow_dir,
preamble_addendum="",
):
return textwrap.dedent(
"""
######## Snakemake header ########
library(methods)
Snakemake <- setClass(
"Snakemake",
slots = c(
input = "list",
output = "list",
params = "list",
wildcards = "list",
threads = "numeric",
log = "list",
resources = "list",
config = "list",
rule = "character",
bench_iteration = "numeric",
scriptdir = "character",
source = "function"
)
)
snakemake <- Snakemake(
input = {},
output = {},
params = {},
wildcards = {},
threads = {},
log = {},
resources = {},
config = {},
rule = {},
bench_iteration = {},
scriptdir = {},
source = function(...){{
wd <- getwd()
setwd(snakemake@scriptdir)
source(...)
setwd(wd)
}}
)
{preamble_addendum}
######## Original script #########
"""
).format(
REncoder.encode_namedlist(input_),
REncoder.encode_namedlist(output),
REncoder.encode_namedlist(params),
REncoder.encode_namedlist(wildcards),
threads,
REncoder.encode_namedlist(log),
REncoder.encode_namedlist(
{
name: value
for name, value in resources.items()
if name != "_cores" and name != "_nodes"
}
),
REncoder.encode_dict(config),
REncoder.encode_value(rulename),
REncoder.encode_numeric(bench_iteration),
REncoder.encode_value(
os.path.dirname(path[7:])
if path.startswith("file://")
else os.path.dirname(path)
),
preamble_addendum=preamble_addendum,
)
[docs] def get_preamble(self):
return RScript.generate_preamble(
self.path,
self.source,
self.basedir,
self.input,
self.output,
self.params,
self.wildcards,
self.threads,
self.resources,
self.log,
self.config,
self.rulename,
self.conda_env,
self.container_img,
self.singularity_args,
self.env_modules,
self.bench_record,
self.jobid,
self.bench_iteration,
self.cleanup_scripts,
self.shadow_dir,
)
[docs] def write_script(self, preamble, fd):
fd.write(preamble.encode())
fd.write(self.source)
[docs] def execute_script(self, fname):
if self.conda_env is not None and "R_LIBS" in os.environ:
logger.warning(
"R script job uses conda environment but "
"R_LIBS environment variable is set. This "
"is likely not intended, as R_LIBS can "
"interfere with R packages deployed via "
"conda. Consider running `unset R_LIBS` or "
"remove it entirely before executing "
"Snakemake."
)
self._execute_cmd("Rscript --vanilla {fname:q}", fname=fname)
[docs]class RMarkdown(ScriptBase):
[docs] def get_preamble(self):
return textwrap.dedent(
"""
######## Snakemake header ########
library(methods)
Snakemake <- setClass(
"Snakemake",
slots = c(
input = "list",
output = "list",
params = "list",
wildcards = "list",
threads = "numeric",
log = "list",
resources = "list",
config = "list",
rule = "character",
bench_iteration = "numeric",
scriptdir = "character",
source = "function"
)
)
snakemake <- Snakemake(
input = {},
output = {},
params = {},
wildcards = {},
threads = {},
log = {},
resources = {},
config = {},
rule = {},
bench_iteration = {},
scriptdir = {},
source = function(...){{
wd <- getwd()
setwd(snakemake@scriptdir)
source(...)
setwd(wd)
}}
)
######## Original script #########
"""
).format(
REncoder.encode_namedlist(self.input),
REncoder.encode_namedlist(self.output),
REncoder.encode_namedlist(self.params),
REncoder.encode_namedlist(self.wildcards),
self.threads,
REncoder.encode_namedlist(self.log),
REncoder.encode_namedlist(
{
name: value
for name, value in self.resources.items()
if name != "_cores" and name != "_nodes"
}
),
REncoder.encode_dict(self.config),
REncoder.encode_value(self.rulename),
REncoder.encode_numeric(self.bench_iteration),
REncoder.encode_value(
os.path.dirname(self.path[7:])
if self.path.startswith("file://")
else os.path.dirname(self.path)
),
)
[docs] def write_script(self, preamble, fd):
# Insert Snakemake object after the RMarkdown header
code = self.source.decode()
pos = next(itertools.islice(re.finditer(r"---\n", code), 1, 2)).start() + 3
fd.write(str.encode(code[:pos]))
preamble = textwrap.dedent(
"""
```{r, echo=FALSE, message=FALSE, warning=FALSE}
%s
```
"""
% preamble
)
fd.write(preamble.encode())
fd.write(str.encode(code[pos:]))
[docs] def execute_script(self, fname):
if len(self.output) != 1:
raise WorkflowError(
"RMarkdown scripts (.Rmd) may only have a single output file."
)
out = os.path.abspath(self.output[0])
self._execute_cmd(
'Rscript --vanilla -e \'rmarkdown::render("{fname}", output_file="{out}", quiet=TRUE, knit_root_dir = "{workdir}", params = list(rmd="{fname}"))\'',
fname=fname,
out=out,
workdir=os.getcwd(),
)
[docs]class JuliaScript(ScriptBase):
[docs] def get_preamble(self):
return textwrap.dedent(
"""
######## Snakemake header ########
struct Snakemake
input::Dict
output::Dict
params::Dict
wildcards::Dict
threads::Int64
log::Dict
resources::Dict
config::Dict
rule::String
bench_iteration
scriptdir::String
#source::Any
end
snakemake = Snakemake(
{}, #input::Dict
{}, #output::Dict
{}, #params::Dict
{}, #wildcards::Dict
{}, #threads::Int64
{}, #log::Dict
{}, #resources::Dict
{}, #config::Dict
{}, #rule::String
{}, #bench_iteration::Int64
{}, #scriptdir::String
#, #source::Any
)
######## Original script #########
""".format(
JuliaEncoder.encode_namedlist(self.input),
JuliaEncoder.encode_namedlist(self.output),
JuliaEncoder.encode_namedlist(self.params),
JuliaEncoder.encode_namedlist(self.wildcards),
JuliaEncoder.encode_value(self.threads),
JuliaEncoder.encode_namedlist(self.log),
JuliaEncoder.encode_namedlist(
{
name: value
for name, value in self.resources.items()
if name != "_cores" and name != "_nodes"
}
),
JuliaEncoder.encode_dict(self.config),
JuliaEncoder.encode_value(self.rulename),
JuliaEncoder.encode_value(self.bench_iteration),
JuliaEncoder.encode_value(
os.path.dirname(self.path[7:])
if self.path.startswith("file://")
else os.path.dirname(self.path)
),
).replace(
"'", '"'
)
)
[docs] def write_script(self, preamble, fd):
fd.write(preamble.encode())
fd.write(self.source)
[docs] def execute_script(self, fname):
self._execute_cmd("julia {fname:q}", fname=fname)
[docs]def get_source(path, basedir="."):
import nbformat
source = None
if not path.startswith("http") and not path.startswith("git+file"):
if path.startswith("file://"):
path = path[7:]
elif path.startswith("file:"):
path = path[5:]
if not os.path.isabs(path):
path = os.path.abspath(os.path.join(basedir, path))
path = "file://" + path
path = format(path, stepout=1)
if path.startswith("file://"):
sourceurl = "file:" + pathname2url(path[7:])
elif path.startswith("git+file"):
source = git_content(path)
(root_path, file_path, version) = split_git_path(path)
path = path.rstrip("@" + version)
else:
sourceurl = path
if source is None:
with urlopen(sourceurl) as source:
source = source.read()
language = None
if path.endswith(".py"):
language = "python"
elif path.endswith(".ipynb"):
language = "jupyter"
elif path.endswith(".R"):
language = "r"
elif path.endswith(".Rmd"):
language = "rmarkdown"
elif path.endswith(".jl"):
language = "julia"
# detect kernel language for Jupyter Notebooks
if language == "jupyter":
nb = nbformat.reads(source, as_version=nbformat.NO_CONVERT)
kernel_language = nb["metadata"]["language_info"]["name"]
language += "_" + kernel_language.lower()
return path, source, language
[docs]def script(
path,
basedir,
input,
output,
params,
wildcards,
threads,
resources,
log,
config,
rulename,
conda_env,
container_img,
singularity_args,
env_modules,
bench_record,
jobid,
bench_iteration,
cleanup_scripts,
shadow_dir,
):
"""
Load a script from the given basedir + path and execute it.
"""
path, source, language = get_source(path, basedir)
ExecClass = {
"python": PythonScript,
"r": RScript,
"rmarkdown": RMarkdown,
"julia": JuliaScript,
}.get(language, None)
if ExecClass is None:
raise ValueError(
"Unsupported script: Expecting either Python (.py), R (.R), RMarkdown (.Rmd) or Julia (.jl) script."
)
executor = ExecClass(
path,
source,
basedir,
input,
output,
params,
wildcards,
threads,
resources,
log,
config,
rulename,
conda_env,
container_img,
singularity_args,
env_modules,
bench_record,
jobid,
bench_iteration,
cleanup_scripts,
shadow_dir,
)
executor.evaluate()