def solve_r_iterate(dp0, r1, F, S, trace):
LF = LowerSets(F)
if do_extra_checks():
LF.belongs(S)
F2 = F[1]
converged = set() # subset of solutions for which they converged
nextit = set()
# find the set of all r2s
for fa in S.maximals:
hr = dp0.solve_r_trace((r1, fa[1]), trace)
for fb in hr.maximals:
# valid = R.leq(ra, rb) # ra <= rb
valid = F.leq(fb, fa) # fb <= fa
if valid:
nextit.add(fb)
feasible = F2.leq(fa[1], fb[1])
if feasible:
converged.add(fb)
nextit = F.Ls(poset_maxima(nextit, F.leq))
converged = F.Ls(poset_maxima(converged, F.leq))
return nextit, converged
def solve_r_iterate(dp0, r1, F, S, trace):
LF = LowerSets(F)
if do_extra_checks():
LF.belongs(S)
F2 = F[1]
converged = set() # subset of solutions for which they converged
nextit = set()
# find the set of all r2s
for fa in S.maximals:
hr = dp0.solve_r_trace((r1, fa[1]), trace)
for fb in hr.maximals:
# valid = R.leq(ra, rb) # ra <= rb
valid = F.leq(fb, fa) # fb <= fa
if valid:
nextit.add(fb)
feasible = F2.leq(fa[1], fb[1])
if feasible:
converged.add(fb)
nextit = F.Ls(poset_maxima(nextit, F.leq))
converged = F.Ls(poset_maxima(converged, F.leq))
return nextit, converged
def solve_r_all(self, r1, trace):
""" Returns an upperset in UR. You want to project
it to R1 to use as the output. """
dp0 = self.dp1
F = dp0.get_fun_space()
F1 = F[0]
LF = LowerSets(F)
# we consider a set of iterates
# we start from the bottom
trace.log('Iterating in LF = %s' % LF.__str__())
s0 = F.Ls(F.get_maximal_elements())
S = [
KleeneIteration(s=s0,
s_converged=F.Ls(set()),
r=lowerset_project(s0, 0),
r_converged=F1.Ls(set()))
]
for i in range(1, 1000000): # XXX
with trace.iteration(i) as t:
si_prev = S[-1].s
si_next, converged = solve_r_iterate(dp0, r1, F, si_prev, t)
iteration = KleeneIteration(s=si_next,
s_converged=converged,
r=lowerset_project(si_next, 0),
r_converged=lowerset_project(
converged, 0))
S.append(iteration)
t.log('si_next = %s' % LF.format(si_next))
if do_extra_checks():
try:
LF.check_leq(si_prev, si_next)
except NotLeq as e:
msg = 'Loop iteration invariant not satisfied.'
raise_wrapped(Exception,
e,
msg,
si_prev=si_prev,
si_next=si_next,
dp=self.dp1)
t.values(state=S[-1])
if LF.leq(si_next, si_prev):
t.log('Breaking because converged (iteration %s) ' % i)
break
trace.values(type='loop2r', LF=LF, F=F, dp=self, iterations=S)
res_all = S[-1].s
res_f1 = lowerset_project(res_all, 0)
result = dict(res_all=res_all, res_f1=res_f1)
return result
def solve_r_all(self, r1, trace):
""" Returns an upperset in UR. You want to project
it to R1 to use as the output. """
dp0 = self.dp1
F = dp0.get_fun_space()
F1 = F[0]
LF = LowerSets(F)
# we consider a set of iterates
# we start from the bottom
trace.log('Iterating in LF = %s' % LF.__str__())
s0 = F.Ls(F.get_maximal_elements())
S = [KleeneIteration(s=s0, s_converged=F.Ls(set()),
r=lowerset_project(s0, 0),
r_converged=F1.Ls(set()))]
for i in range(1, 1000000): # XXX
with trace.iteration(i) as t:
si_prev = S[-1].s
si_next, converged = solve_r_iterate(dp0, r1, F, si_prev, t)
iteration = KleeneIteration(s=si_next,
s_converged=converged,
r=lowerset_project(si_next, 0),
r_converged=lowerset_project(converged, 0))
S.append(iteration)
t.log('si_next = %s' % LF.format(si_next))
if do_extra_checks():
try:
LF.check_leq(si_prev, si_next)
except NotLeq as e:
msg = 'Loop iteration invariant not satisfied.'
raise_wrapped(Exception, e, msg, si_prev=si_prev,
si_next=si_next, dp=self.dp1)
t.values(state=S[-1])
if LF.leq(si_next, si_prev):
t.log('Breaking because converged (iteration %s) ' % i)
break
trace.values(type='loop2r', LF=LF, F=F, dp=self, iterations=S)
res_all = S[-1].s
res_f1 = lowerset_project(res_all, 0)
result = dict(res_all=res_all, res_f1=res_f1)
return result
