def enumerate(self,
cudd=True,
export_pdf=None,
return_count=False,
return_solutions=False):
if cudd:
from dd.cudd import BDD
else:
from dd.autoref import BDD
kconfig_file = f"{self.cwd}/{self.kconfig}"
kconfig_hash = self.file_hash(kconfig_file)
with cd(self.cwd):
kconf = kconfiglib.Kconfig(kconfig_file)
pre_variables = list()
pre_expressions = list()
for choice in kconf.choices:
var_name = f"_choice_{choice.name}"
pre_variables.append(var_name)
# Build "exactly_one", expressing that exactly one of the symbols managed by this choice must be selected
symbols = list(map(lambda sym: sym.name, choice.syms))
exactly_one = list()
for sym1 in symbols:
subexpr = list()
for sym2 in symbols:
if sym1 == sym2:
subexpr.append(sym2)
else:
subexpr.append(f"!{sym2}")
exactly_one.append("(" + " & ".join(subexpr) + ")")
exactly_one = " | ".join(exactly_one)
# If the choice is selected, exactly one choice element must be selected
pre_expressions.append(f"{var_name} -> ({exactly_one})")
# Each choice symbol in exactly_once depends on the choice itself, which will lead to "{symbol} -> {var_name}" rules being generated later on. This
# ensures that if the choice is false, each symbol is false too. We do not need to handle that case here.
# The choice may depend on other variables
depends_on = self._dependencies_to_bdd_expr(choice.direct_dep)
if depends_on:
if choice.is_optional:
pre_expressions.append(f"{var_name} -> {depends_on}")
else:
pre_expressions.append(f"{var_name} <-> {depends_on}")
elif not choice.is_optional:
# Always active
pre_expressions.append(var_name)
for symbol in kconf.syms.values():
if not self._can_be_handled_by_bdd(symbol):
continue
pre_variables.append(symbol.name)
depends_on = self._dependencies_to_bdd_expr(symbol.direct_dep)
if depends_on:
pre_expressions.append(f"{symbol.name} -> {depends_on}")
for selected_symbol, depends_on in symbol.selects:
depends_on = self._dependencies_to_bdd_expr(depends_on)
if depends_on:
pre_expressions.append(
f"({symbol.name} & ({depends_on})) -> {selected_symbol.name}"
)
else:
pre_expressions.append(
f"{symbol.name} -> {selected_symbol.name}")
logger.debug("Variables:")
logger.debug("\n".join(pre_variables))
logger.debug("Expressions:")
logger.debug("\n".join(pre_expressions))
variables = list()
expressions = list()
bdd = BDD()
variable_count = 0
for variable in pre_variables:
if variable[0] != "#":
variables.append(variable)
variable_count += 1
bdd.declare(variable)
logger.debug(f"Got {variable_count} variables")
constraint = "True"
expression_count = 0
for expression in pre_expressions:
if expression[0] != "#":
expressions.append(expression)
expression_count += 1
constraint += f" & ({expression})"
logger.debug(f"Got {expression_count} rules")
logger.debug(constraint)
constraint = bdd.add_expr(constraint)
if cudd:
# Egal?
logger.debug("Reordering ...")
BDD.reorder(bdd)
else:
# Wichtig! Lesbarkeit++ falls gedumpt wird, Performance vermutlich auch.
logger.debug("Collecting Garbage ...")
bdd.collect_garbage()
# See <http://www.ecs.umass.edu/ece/labs/vlsicad/ece667/reading/somenzi99bdd.pdf> for how to read the graphical representation.
# A solid line is followed if the origin node is 1
# A dashed line is followed if the origin node is 0
# A path from a top node to 1 satisfies the function iff the number of negations ("-1" annotations) is even
if export_pdf is not None:
logger.info(f"Dumping to {export_pdf} ...")
bdd.dump(export_pdf)
logger.debug("Solving ...")
# still need to be set, otherwise autoref and cudd complain and set them anyways.
# care_vars = list(filter(lambda x: "meta_" not in x and "_choice_" not in x, variables))
if return_solutions:
return bdd.pick_iter(constraint, care_vars=variables)
if return_count:
return len(bdd.pick_iter(constraint, care_vars=variables))
config_file = f"{self.cwd}/.config"
for solution in bdd.pick_iter(constraint, care_vars=variables):
logger.debug(f"Set {solution}")
with open(config_file, "w") as f:
for k, v in solution.items():
if v:
print(f"CONFIG_{k}=y", file=f)
else:
print(f"# CONFIG_{k} is not set", file=f)
with cd(self.cwd):
kconf = kconfiglib.Kconfig(kconfig_file)
kconf.load_config(config_file)
int_values = list()
int_names = list()
for symbol in kconf.syms.values():
if (kconfiglib.TYPE_TO_STR[symbol.type] == "int"
and symbol.visibility and symbol.ranges):
for min_val, max_val, condition in symbol.ranges:
if condition.tri_value:
int_names.append(symbol.name)
min_val = int(min_val.str_value, 0)
max_val = int(max_val.str_value, 0)
step_size = (max_val - min_val) // 8
if step_size == 0:
step_size = 1
int_values.append(
list(range(min_val, max_val + 1, step_size)))
continue
for int_config in itertools.product(*int_values):
for i, int_name in enumerate(int_names):
val = int_config[i]
symbol = kconf.syms[int_name]
logger.debug(f"Set {symbol.name} to {val}")
symbol.set_value(str(val))
self._run_explore_experiment(kconf, kconfig_hash, config_file)
class BDDModel:
"""
A Binary Decision Diagram (BDD) representation of the feature model given as a CNF formula.
It relies on the dd module: https://pypi.org/project/dd/
"""
AND = '&'
OR = '|'
NOT = '!'
def __init__(self, feature_model: FeatureModel, cnf_formula: str):
self.feature_model = feature_model
self.cnf = cnf_formula.replace('-', '')
self.variables = self._extract_variables(self.cnf)
self.bdd = BDD() # Instantiate a manager
self.declare_variables(self.variables) # Declare variables
self.expression = self.bdd.add_expr(self.cnf)
def _extract_variables(self, cnf_formula: str) -> list[str]:
variables = set()
for v in cnf_formula.split():
if BDDModel.AND not in v and BDDModel.OR not in v:
var = v.strip().replace('(', '').replace(')', '').replace(
BDDModel.NOT, '')
variables.add(var)
return list(variables)
def declare_variables(self, variables: list[str]):
for v in variables:
self.bdd.declare(v)
def serialize(self, filepath: str, filetype: str = 'png'):
self.bdd.dump(filename=filepath,
roots=[self.expression],
filetype=filetype)
def get_number_of_configurations(
self,
selected_features: list[Feature] = None,
deselected_features: list[Feature] = None) -> int:
if not selected_features:
expr = self.cnf
else:
expr = f' {BDDModel.AND} '.join(
[f.name for f in selected_features])
if deselected_features:
expr += f' {BDDModel.AND} ' + f' {BDDModel.AND} !'.join(
[f.name for f in deselected_features])
expr += f' {BDDModel.AND} ' + '{x}'.format(x=self.expression)
u = self.bdd.add_expr(expr)
return self.bdd.count(u, nvars=len(self.variables))
def get_configurations(
self,
selected_features: list[Feature] = None,
deselected_features: list[Feature] = None
) -> list[FMConfiguration]:
if not selected_features:
expr = self.cnf
else:
expr = f' {BDDModel.AND} '.join(
[f.name for f in selected_features])
if deselected_features:
expr += f' {BDDModel.AND} ' + f' {BDDModel.AND} '.join(
['!' + f.name for f in deselected_features])
expr += f' {BDDModel.AND} ' + '{x}'.format(x=self.expression)
u = self.bdd.add_expr(expr)
configs = []
for c in self.bdd.pick_iter(u, care_vars=self.variables):
elements = {
self.feature_model.get_feature_by_name(f): True
for f in c.keys() if c[f]
}
configs.append(FMConfiguration(elements))
return configs
class BDDModel:
"""
A Binary Decision Diagram (BDD) representation of the feature model given as a CNF formula.
It relies on the dd module: https://pypi.org/project/dd/
"""
AND = '&'
OR = '|'
NOT = '!'
def __init__(self, cnf_formula: str):
self.cnf = cnf_formula.replace('-', '')
self.variables = self._extract_variables(self.cnf)
self.bdd = BDD() # Instantiate a manager
self.declare_variables(self.variables) # Declare variables
self.expression = self.bdd.add_expr(self.cnf)
def _extract_variables(self, cnf_formula: str) -> list[str]:
variables = set()
for v in cnf_formula.split():
if BDDModel.AND not in v and BDDModel.OR not in v:
var = v.strip().replace('(', '').replace(')', '').replace(
BDDModel.NOT, '')
variables.add(var)
return list(variables)
def declare_variables(self, variables: list[str]):
for v in variables:
self.bdd.declare(v)
def serialize(self, filepath: str, filetype: str = 'png'):
self.bdd.dump(filename=filepath,
roots=[self.expression],
filetype=filetype)
def get_number_of_configurations(self,
features: list[Feature] = None) -> int:
if features is None:
return self.bdd.count(self.expression, nvars=len(self.variables))
expr = f' {BDDModel.AND} '.join([
f.name for f in features
]) + f' {BDDModel.AND} ' + '{x}'.format(x=self.expression)
u = self.bdd.add_expr(expr)
return self.bdd.count(u, nvars=len(self.variables))
def get_configurations(self,
features: list[Feature] = None
) -> list[FMConfiguration]:
if features is None:
expr = self.cnf
else:
expr = f' {BDDModel.AND} '.join([
f.name for f in features
]) + f' {BDDModel.AND} ' + '{x}'.format(x=self.expression)
u = self.bdd.add_expr(expr)
configs = []
for c in self.bdd.pick_iter(u, care_vars=self.variables):
configs.append(sorted([f for f in c.keys() if c[f]]))
return configs
def get_uniform_random_sample(self, size: int) -> list[list[str]]:
"""This generates all configurations."""
configs = self.get_configurations()
if size > len(configs):
size = len(configs)
return random.sample(configs, size)
def get_random_configuration(self) -> list[str]:
"""This follows the Knut algorithm, but needs to be optimized"""
solutions = self.bdd.count(self.expression, nvars=len(self.variables))
expr = ""
variables = list(self.variables)
while solutions > 1:
feature = random.choice(variables)
variables.remove(feature)
possible_expr = expr + f' {BDDModel.AND} '.join(
[feature]) + f' {BDDModel.AND} '
formula = possible_expr + '{x}'.format(x=self.expression)
u = self.bdd.add_expr(formula)
solutions = self.bdd.count(u, nvars=len(self.variables))
if solutions <= 0:
possible_expr = expr + f' {BDDModel.AND} '.join(
['!' + feature]) + f' {BDDModel.AND} '
formula = possible_expr + '{x}'.format(x=self.expression)
u = self.bdd.add_expr(formula)
solutions = self.bdd.count(u, nvars=len(self.variables))
expr = possible_expr
config = self.bdd.pick(u)
return sorted([f for f in config.keys() if config[f]])
def get_sample_of_configurations(self, size: int) -> list[list[str]]:
"""
Bad implementation, we need to:
The original algorithm by Knuth is specified on BDDs very efficiently, as the probabilities required for all the possible SAT solutions are computed just once with a single BDD traversal,
and then reused every time a solution is generated.
"""
nof_configs = self.get_number_of_configurations()
if size > nof_configs:
size = nof_configs
sample = list()
while len(sample) < size:
config = self.get_random_configuration()
if config not in sample:
sample.append(config)
return sample