"""plan execution of rql queries on multiple sources

For a system source and a ldap user source (only EUser and its attributes

is supported, no group or such):

:EUser X:

1. fetch EUser X from both sources and return concatenation of results

:EUser X WHERE X in_group G, G name 'users':

* catch 1

* catch 2

:EUser X,L WHERE X in_group G, X login L, G name 'users':

1. fetch Any X,L WHERE X is EUser, X login L from both sources, store

   concatenation of results into a temporary table

2. return the result of Any X, L WHERE X is TMP, X login LX in_group G,

   G name 'users' from the system source

:Any X WHERE X owned_by Y:

* catch 1

* catch 2

:organization: Logilab

:copyright: 2003-2009 LOGILAB S.A. (Paris, FRANCE), all rights reserved.

:contact: http://www.logilab.fr/ -- mailto:contact@logilab.fr

"""

__docformat__ = "restructuredtext en"

from copy import deepcopy

from itertools import imap, ifilterfalse

from logilab.common.compat import any

from logilab.common.decorators import cached

from rql.stmts import Union, Select

from rql.nodes import VariableRef, Comparison, Relation, Constant, Exists, Variable

from cubicweb import server

from cubicweb.common.utils import make_uid

from cubicweb.server.utils import cleanup_solutions

from cubicweb.server.mssteps import *

from cubicweb.server.sources import AbstractSource

Variable._ms_table_key = lambda x: x.name

Relation._ms_table_key = lambda x: x.r_type

# str() Constant.value to ensure generated table name won't be unicode

Constant._ms_table_key = lambda x: str(x.value)

AbstractSource.dont_cross_relations = ()

AbstractSource.cross_relations = ()

def need_aggr_step(select, sources, stepdefs=None):

    """return True if a temporary table is necessary to store some partial

    results to execute the given query

    """

    if len(sources) == 1:

        # can do everything at once with a single source

        return False

    if select.orderby or select.groupby or select.has_aggregat:

        # if more than one source, we need a temp table to deal with sort /

        # groups / aggregat if :

        # * the rqlst won't be splitted (in the other case the last query

        #   using partial temporary table can do sort/groups/aggregat without

        #   the need for a later AggrStep)

        # * the rqlst is splitted in multiple steps and there are more than one

        #   final step

        if stepdefs is None:

            return True

        has_one_final = False

        fstepsolindices = set()

        for stepdef in stepdefs:

            if stepdef[-1]:

                if has_one_final or frozenset(stepdef[2]) != fstepsolindices:

                    return True

                has_one_final = True

            else:

                fstepsolindices.update(stepdef[2])

    return False

def select_group_sort(select): # XXX something similar done in rql2sql

    # add variables used in groups and sort terms to the selection

    # if necessary

    if select.groupby:

        for vref in select.groupby:

            if not vref in select.selection:

                select.append_selected(vref.copy(select))

    for sortterm in select.orderby:

        for vref in sortterm.iget_nodes(VariableRef):

            if not vref in select.get_selected_variables():

                # we can't directly insert sortterm.term because it references

                # a variable of the select before the copy.

                # XXX if constant term are used to define sort, their value

                # may necessite a decay

                select.append_selected(vref.copy(select))

                if select.groupby and not vref in select.groupby:

                    select.add_group_var(vref.copy(select))

def is_ancestor(n1, n2):

    p = n1.parent

    while p is not None:

        if p is n2:

            return True

        p = p.parent

    return False

class PartPlanInformation(object):

    """regroups necessary information to execute some part of a "global" rql

    query ("global" means as received by the querier, which may result in

    multi-sources repository, modifying its internal structure during the

    """

    def __init__(self, plan, rqlst, rqlhelper=None):

        self.needsplit = False

        self.temptable = None

        self.finaltable = None

        self._session = plan.session

        self._solutions = rqlst.solutions

        self._solindices = range(len(self._solutions))

        # source : {relation: set(child variable and constant)}

        self._crossrelations = {}

        # processing

        self._compute_needsplit()

            for k2, v2 in v.iteritems():

        self._inputmaps = {}

        if rqlhelper is not None: # else test

            self._insert_identity_variable = rqlhelper._annotator.rewrite_shared_optional

    def copy_solutions(self, solindices):

        return [self._solutions[solidx].copy() for solidx in solindices]

    @property

    @cached

    def part_sources(self):

    @property

    @cached

    def _sys_source_set(self):

                         for solindex in self._solindices)        

    @cached

    def _norel_support_set(self, relation):

        """return a set of (source, solindex) where source doesn't support the

        relation

        """

                         for solidx in self._solindices

                                 or relation.r_type in source.dont_cross_relations))

        """

        # find for each source which variable/solution are supported

        for varname, varobj in self.rqlst.defined_vars.items():

            # if variable has an eid specified, we can get its source directly

            # NOTE: use uidrels and not constnode to deal with "X eid IN(1,2,3,4)"

            if varobj.stinfo['uidrels']:

                vrels = varobj.stinfo['relations'] - varobj.stinfo['uidrels']

                for rel in varobj.stinfo['uidrels']:

                        continue

                    for const in rel.children[1].get_nodes(Constant):

                        eid = const.eval(self.plan.args)

                        source = self._session.source_from_eid(eid)

                        if vrels and not any(source.support_relation(r.r_type)

                                             for r in vrels):

                        else:

                continue

            rels = varobj.stinfo['relations']

            if not rels and not varobj.stinfo['typerels']:

                # (rare) case where the variable has no type specified nor

                # relation accessed ex. "Any MAX(X)"

                continue

            for i, sol in enumerate(self._solutions):

                vartype = sol[varname]

                # skip final variable

                if eschema(vartype).is_final():

                    break

                for source in repo.sources:

                    if source.support_entity(vartype):

                        # the source support the entity type, though we will

                        # actually have to fetch from it only if

                        # * the variable isn't invariant

                        # * at least one supported relation specified

                        if not varobj._q_invariant or \

                               any(imap(source.support_relation,

                                        (r.r_type for r in rels if r.r_type != 'eid'))):

                        # if variable is not invariant and is used by a relation

                        # not supported by this source, we'll have to split the

                        # query

                        if not varobj._q_invariant and any(ifilterfalse(

                            source.support_relation, (r.r_type for r in rels))):

                            self.needsplit = True               

        for rel in self.rqlst.iget_nodes(Relation):

            # process non final relations only

            # note: don't try to get schema for 'is' relation (not available

            # during bootstrap)

            if not rel.is_types_restriction() and not rschema(rel.r_type).is_final():

                # nothing to do if relation is not supported by multiple sources

                # or if some source has it listed in its cross_relations

                # attribute

                #

                # XXX code below don't deal if some source allow relation

                #     crossing but not another one

                relsources = repo.rel_type_sources(rel.r_type)

                if len(relsources) < 2:

                    # filter out sources being there because they have this

                    # relation in their dont_cross_relations attribute

                    relsources = [source for source in relsources

                                  if source.support_relation(rel.r_type)]

                    if relsources:

                        # this means the relation is using a variable inlined as

                        # a constant and another unsupported variable, in which

                    continue

                lhs, rhs = rel.get_variable_parts()

                lhsv, rhsv = getattr(lhs, 'variable', lhs), getattr(rhs, 'variable', rhs)