path: root/gaphas/solver.py

"""
Constraint solver allows to define constraint between two or more different
variables and keep this constraint always true when one or more of the
constrained variables change. For example, one may want to keep two
variables always equal.

Variables change and at some point of time we want to make all constraints
valid again. This process is called solving constraints.

Gaphas' solver allows to define constraints between Variable instances.
Constraint classes are defined in `gaphas.constraint` module.

How It Works
------------
Every constraint contains list of variables and has to be registered in
solver object. Variables change (`Variable.dirty()`, `Solver.request_resolve()`
methods) and their constraints are marked by solver as dirty. To solve
constraints, solver loops through dirty constraints and asks constraint for
a variable (called weakest variable), which

- has the lowest strength
- or if there are many variables with the same, lowest strength value
  return first unchanged variable with lowest strength
- or if there is no unchanged, then return the first changed with the
  lowest strength

(weakest variable invariants defined above)

Having weakest variable (`constraint.Constraint.weakest()` method) every
constraint is being asked to solve itself (`constraint.Constraint.solve_for()`
method) changing appropriate variables to make the constraint valid again.
"""

from __future__ import division

__version__ = "$Revision$"
# $HeadURL$

from operator import isCallable
from state import observed, reversible_pair, reversible_property

# epsilon for float comparison
# is simple abs(x - y) > EPSILON enough for canvas needs?
EPSILON = 1e-10

# Variable Strengths:
VERY_WEAK = 0
WEAK = 10
NORMAL = 20
STRONG = 30
VERY_STRONG = 40
REQUIRED = 100


class Variable(object):
    """
    Representation of a variable in the constraint solver.
    Each Variable has a @value and a @strength. Ina constraint the
    weakest variables are changed.
    
    You can even do some calculating with it. The Variable always represents
    a float variable.
    """

    def __init__(self, value=0.0, strength=NORMAL):
        self._value = float(value)
        self._strength = strength

        # These variables are set by the Solver:
        self._solver = None
        self._constraints = set()

    @observed
    def _set_strength(self, strength):
        self._strength = strength
        for c in self._constraints:
            c.create_weakest_list()

    strength = reversible_property(lambda s: s._strength, _set_strength)

    def dirty(self):
        """
        Mark the variable dirty in both the constraint solver and attached
        constraints.
        
        Variables are marked dirty also during constraints solving to
        solve all dependent constraints, i.e. two equals constraints
        between 3 variables.
        """
        solver = self._solver
        if not solver:
            return

        solver.request_resolve(self)

    @observed
    def set_value(self, value):
        self._value = float(value)
        self.dirty()

    value = reversible_property(lambda s: s._value, set_value)

    def __str__(self):
        return 'Variable(%g, %d)' % (self._value, self._strength)
    __repr__ = __str__

    def __float__(self):
        return float(self._value)

    def __eq__(self, other):
        """
        >>> Variable(5) == 5
        True
        >>> Variable(5) == 4
        False
        >>> Variable(5) != 5
        False
        """
        return abs(self._value - other) < EPSILON

    def __ne__(self, other):
        """
        >>> Variable(5) != 4
        True
        >>> Variable(5) != 5
        False
        """
        return abs(self._value - other) > EPSILON

    def __gt__(self, other):
        """
        >>> Variable(5) > 4
        True
        >>> Variable(5) > 5
        False
        """
        return self._value.__gt__(float(other))

    def __lt__(self, other):
        """
        >>> Variable(5) < 4
        False
        >>> Variable(5) < 6
        True
        """
        return self._value.__lt__(float(other))

    def __ge__(self, other):
        """
        >>> Variable(5) >= 5
        True
        """
        return self._value.__ge__(float(other))

    def __le__(self, other):
        """
        >>> Variable(5) <= 5
        True
        """
        return self._value.__le__(float(other))

    def __add__(self, other):
        """
        >>> Variable(5) + 4
        9.0
        """
        return self._value.__add__(float(other))

    def __sub__(self, other):
        """
        >>> Variable(5) - 4
        1.0
        >>> Variable(5) - Variable(4)
        1.0
        """
        return self._value.__sub__(float(other))

    def __mul__(self, other):
        """
        >>> Variable(5) * 4
        20.0
        >>> Variable(5) * Variable(4)
        20.0
        """
        return self._value.__mul__(float(other))

    def __floordiv__(self, other):
        """
        >>> Variable(21) // 4
        5.0
        >>> Variable(21) // Variable(4)
        5.0
        """
        return self._value.__floordiv__(float(other))

    def __mod__(self, other):
        """
        >>> Variable(5) % 4
        1.0
        >>> Variable(5) % Variable(4)
        1.0
        """
        return self._value.__mod__(float(other))

    def __divmod__(self, other):
        """
        >>> divmod(Variable(21), 4)
        (5.0, 1.0)
        >>> divmod(Variable(21), Variable(4))
        (5.0, 1.0)
        """
        return self._value.__divmod__(float(other))

    def __pow__(self, other):
        """
        >>> pow(Variable(5), 4)
        625.0
        >>> pow(Variable(5), Variable(4))
        625.0
        """
        return self._value.__pow__(float(other))

    def __div__(self, other):
        """
        >>> Variable(5) / 4.
        1.25
        >>> Variable(5) / Variable(4)
        1.25
        """
        return self._value.__div__(float(other))

    def __truediv__(self, other):
        """
        >>> Variable(5.) / 4
        1.25
        >>> 10 / Variable(5.)
        2.0
        """
        return self._value.__truediv__(float(other))

    # .. And the other way around:

    def __radd__(self, other):
        """
        >>> 4 + Variable(5)
        9.0
        >>> Variable(4) + Variable(5)
        9.0
        """
        return self._value.__radd__(float(other))

    def __rsub__(self, other):
        """
        >>> 6 - Variable(5)
        1.0
        """
        return self._value.__rsub__(other)

    def __rmul__(self, other):
        """
        >>> 4 * Variable(5)
        20.0
        """
        return self._value.__rmul__(other)

    def __rfloordiv__(self, other):
        """
        >>> 21 // Variable(4)
        5.0
        """
        return self._value.__rfloordiv__(other)

    def __rmod__(self, other):
        """
        >>> 5 % Variable(4)
        1.0
        """
        return self._value.__rmod__(other)

    def __rdivmod__(self, other):
        """
        >>> divmod(21, Variable(4))
        (5.0, 1.0)
        """
        return self._value.__rdivmod__(other)

    def __rpow__(self, other):
        """
        >>> pow(4, Variable(5))
        1024.0
        """
        return self._value.__rpow__(other)

    def __rdiv__(self, other):
        """
        >>> 5 / Variable(4.)
        1.25
        """
        return self._value.__rdiv__(other)

    def __rtruediv__(self, other):
        """
        >>> 5. / Variable(4)
        1.25
        """
        return self._value.__rtruediv__(other)


class Projection(object):
    """
    Projections are used to convert values from one space to another,
    e.g. from Canvas to Item space or visa versa.

    In order to be a Projection the ``value`` and ``strength`` properties
    should be implemented and a method named ``variable()`` should be present.

    Projections should inherit from this class.

    Projections may be nested.

    This default implementation projects a variable to it's own:

    >>> v = Variable(4.0)
    >>> v
    Variable(4, 20)
    >>> p = Projection(v)
    >>> p.value
    4.0
    >>> p.value = -1
    >>> p.value
    -1.0
    >>> v.value
    -1.0
    >>> p.strength
    20
    >>> p.variable()
    Variable(-1, 20)
    """

    def __init__(self, var):
        self._var = var

    def _set_value(self, value):
        self._var.value = value

    value = property(lambda s: s._var.value, _set_value)

    strength = property(lambda s: s._var.strength)

    def variable(self):
        """
        Return the variable owned by the projection.
        """
        return self._var

    def __float__(self):
        return float(self.variable()._value)

    def __str__(self):
        return '%s(%s)' % (self.__class__.__name__, self.variable())
    __repr__ = __str__


class Solver(object):
    """
    Solve constraints. A constraint should have accompanying
    variables.
    """

    def __init__(self):
        # a dict of constraint -> name/variable mappings
        self._constraints = set()
        self._marked_cons = []
        self._solving = False

    def request_resolve(self, variable, projections_only=False):
        """
        Mark a variable as "dirty". This means it it solved the next time
        the constraints are resolved.

        If projections_only is set to True, only constraints using the
        variable through a Projection instance (e.i. variable itself is not
        in `constraint.Constraint.variables()`) are marked.

        Example:

        >>> from constraint import EquationConstraint
        >>> a, b, c = Variable(1.0), Variable(2.0), Variable(3.0)
        >>> s = Solver()
        >>> c_eq = EquationConstraint(lambda a,b: a+b, a=a, b=b)
        >>> s.add_constraint(c_eq)
        EquationConstraint(<lambda>, a=Variable(1, 20), b=Variable(2, 20))
        >>> c_eq._weakest
        [Variable(1, 20), Variable(2, 20)]
        >>> s._marked_cons
        [EquationConstraint(<lambda>, a=Variable(1, 20), b=Variable(2, 20))]
        >>> a.value=5.0
        >>> c_eq.weakest()
        Variable(2, 20)
        >>> b.value=2.0
        >>> c_eq.weakest()
        Variable(5, 20)
        >>> a.value=5.0
        >>> c_eq.weakest()
        Variable(2, 20)
        """
        # Peel of Projections:
        while isinstance(variable, Projection):
            variable = variable.variable()
        for c in variable._constraints:
            if not projections_only or c._solver_has_projections:
                if not self._solving:
                    if c in self._marked_cons:
                        self._marked_cons.remove(c)
                    c.mark_dirty(variable)
                    self._marked_cons.append(c)
                else:
                    c.mark_dirty(variable)
                    self._marked_cons.append(c)
                    if __debug__:
                        if self._marked_cons.count(c) > 100:
                            raise JuggleError, 'Variable juggling detected, constraint %s' % c


    @observed
    def add_constraint(self, constraint):
        """
        Add a constraint.
        The actual constraint is returned, so the constraint can be removed
        later on.

        Example:

        >>> from constraint import EquationConstraint
        >>> s = Solver()
        >>> a, b = Variable(), Variable(2.0)
        >>> s.add_constraint(EquationConstraint(lambda a, b: a -b, a=a, b=b))
        EquationConstraint(<lambda>, a=Variable(0, 20), b=Variable(2, 20))
        >>> len(s._constraints)
        1
        >>> a.value
        0.0
        >>> b.value
        2.0
        >>> len(s._constraints)
        1
        """
        assert constraint, 'No constraint (%s)' % (constraint,)
        self._constraints.add(constraint)
        self._marked_cons.append(constraint)
        constraint._solver_has_projections = False
        for v in constraint.variables():
            while isinstance(v, Projection):
                v = v.variable()
                constraint._solver_has_projections = True
            v._constraints.add(constraint)
            v._solver = self
        #print 'added constraint', constraint
        return constraint

    @observed
    def remove_constraint(self, constraint):
        """
        Remove a constraint from the solver

        >>> from constraint import EquationConstraint
        >>> s = Solver()
        >>> a, b = Variable(), Variable(2.0)
        >>> c = s.add_constraint(EquationConstraint(lambda a, b: a -b, a=a, b=b))
        >>> c
        EquationConstraint(<lambda>, a=Variable(0, 20), b=Variable(2, 20))
        >>> s.remove_constraint(c)
        >>> s._marked_cons
        []
        >>> s._constraints
        set([])

        Removing a constraint twice has no effect:

        >>> s.remove_constraint(c)
        """
        assert constraint, 'No constraint (%s)' % (constraint,)
        for v in constraint.variables():
            while isinstance(v, Projection):
                v = v.variable()
            v._constraints.discard(constraint)
        self._constraints.discard(constraint)
        while constraint in self._marked_cons:
            self._marked_cons.remove(constraint)

    reversible_pair(add_constraint, remove_constraint)


    def request_resolve_constraint(self, c):
        """
        Request resolving a constraint.
        """
        self._marked_cons.append(c)


    def constraints_with_variable(self, *variables):
        """
        Return an iterator of constraints that work with variable.
        The variable in question should be exposed by the constraints
        `constraint.Constraint.variables()` method.

        >>> from constraint import EquationConstraint
        >>> s = Solver()
        >>> a, b, c = Variable(), Variable(2.0), Variable(4.0)
        >>> eq_a_b = s.add_constraint(EquationConstraint(lambda a, b: a -b, a=a, b=b))
        >>> eq_a_b
        EquationConstraint(<lambda>, a=Variable(0, 20), b=Variable(2, 20))
        >>> eq_a_c = s.add_constraint(EquationConstraint(lambda a, b: a -b, a=a, b=c))
        >>> eq_a_c
        EquationConstraint(<lambda>, a=Variable(0, 20), b=Variable(4, 20))

        And now for some testing:

        >>> eq_a_b in s.constraints_with_variable(a)
        True
        >>> eq_a_c in s.constraints_with_variable(a)
        True
        >>> eq_a_b in s.constraints_with_variable(a, b)
        True
        >>> eq_a_c in s.constraints_with_variable(a, b)
        False

        Using another variable with the same value does not work:

        >>> d = Variable(2.0)
        >>> eq_a_b in s.constraints_with_variable(a, d)
        False

        This also works for projections:

        >>> eq_pr_a_b = s.add_constraint(EquationConstraint(lambda a, b: a -b, a=Projection(a), b=Projection(b)))
        >>> eq_pr_a_b   # doctest: +ELLIPSIS
        EquationConstraint(<lambda>, a=Projection(Variable(0, 20)), b=Projection(Variable(2, 20)))

        >>> eq_pr_a_b in s.constraints_with_variable(a, b)
        True
        >>> eq_pr_a_b in s.constraints_with_variable(a, c)
        False
        >>> eq_pr_a_b in s.constraints_with_variable(a, d)
        False
        """
        # Use a copy of the original set, so constraints may be
        # deleted in the meantime.
        variables = set(variables)
        for c in set(self._constraints):
            if variables.issubset(set(c.variables())):
                yield c
            elif c._solver_has_projections:
                found = True
                for v in c.variables():
                    if v in variables:
                        continue
                    while isinstance(v, Projection):
                        v = v.variable()
                        if v in variables:
                            break
                    else:
                        found = False
                    if not found:
                        break # quit for loop, variable not in constraint
                else:
                    # All iteration have completed succesfully,
                    # so all variables are in the constraint
                    yield c
                    

    def solve(self):
        """
        Example:

        >>> from constraint import EquationConstraint
        >>> a, b, c = Variable(1.0), Variable(2.0), Variable(3.0)
        >>> s = Solver()
        >>> s.add_constraint(EquationConstraint(lambda a,b: a+b, a=a, b=b))
        EquationConstraint(<lambda>, a=Variable(1, 20), b=Variable(2, 20))
        >>> a.value = 5.0
        >>> s.solve()
        >>> len(s._marked_cons)
        0
        >>> b._value
        -5.0
        >>> s.add_constraint(EquationConstraint(lambda a,b: a+b, a=b, b=c))
        EquationConstraint(<lambda>, a=Variable(-5, 20), b=Variable(3, 20))
        >>> len(s._constraints)
        2
        >>> len(s._marked_cons)
        1
        >>> b._value
        -5.0
        >>> s.solve()
        >>> b._value
        -3.0
        >>> a.value = 10
        >>> s.solve()
        >>> c._value
        10.0
        """
        marked_cons = self._marked_cons
        try:
            self._solving = True

            # Solve each constraint. Using a counter makes it
            # possible to also solve constraints that are marked as
            # a result of other variabled being solved.
            n = 0
            while n < len(marked_cons):
                c = marked_cons[n]
                if not c.disabled:
                    wvar = c.weakest()
                    c.solve_for(wvar)
                n += 1

            self._marked_cons = []
        finally:
            self._solving = False

class solvable(object):
    """
    Easy-to-use drop Variable descriptor.

    >>> class A(object):
    ...     x = solvable(varname='_v_x')
    ...     y = solvable(STRONG)
    ...     def __init__(self):
    ...         self.x = 12
    >>> a = A()
    >>> a.x
    Variable(12, 20)
    >>> a._v_x
    Variable(12, 20)
    >>> a.x = 3
    >>> a.x 
    Variable(3, 20)
    >>> a.y 
    Variable(0, 30)
    """

    def __init__(self, strength=NORMAL, varname=None):
        self._strength = strength
        self._varname = varname or '_variable_%x' % id(self)

    def __get__(self, obj, class_=None):
        if not obj:
            return self
        try:
            return getattr(obj, self._varname)
        except AttributeError:
            setattr(obj, self._varname, Variable(strength=self._strength))
            return getattr(obj, self._varname)

    def __set__(self, obj, value):
        try:
            getattr(obj, self._varname).value = float(value)
        except AttributeError:
            v = Variable(strength=self._strength)
            setattr(obj, self._varname, v)
            v.value = value




class JuggleError(AssertionError):
    """
    Variable juggling exception. Raised when constraint's variables are
    marking each other dirty forever.
    """


__test__ = {
    'Solver.add_constraint': Solver.add_constraint,
    'Solver.remove_constraint': Solver.remove_constraint,
    }


# vim:sw=4:et:ai