[MERGE] unaccent support + domain parsing cleanup/improvements

bzr revid: odo@openerp.com-20110922140856-p8q0ib1lirgv1g02

openerp/addons/base/test/test_osv_expression.yml

@@ -177,6 +177,251 @@
         res_ids = self.search(cr, uid, [('company_id.partner_id', 'not in', [])])
         res_ids.sort()
         assert res_ids == all_ids, "Searching against empty set failed, returns %r" % res_ids
+-
+    Test the '(not) like/in' behavior. res.partner and its parent_id column are used because
+    parent_id is a many2one, allowing to test the Null value, and there are actually some
+    null and non-null values in the demo data.
+-
+    !python {model: res.partner }: |
+        partner_ids = self.search(cr, uid, [])
+        partner_ids.sort()
+        max_partner_id = max(partner_ids)
+
+        # Grab test sample data without using a normal
+        # search domain, because we want to test these later,
+        # so we can't rely on them!
+        partners = self.browse(cr, uid, partner_ids)
+        with_parent = []
+        without_parent = []
+        with_website = []
+        for x in partners:
+            if x.parent_id:
+                with_parent.append(x.id)
+            else:
+                without_parent.append(x.id)
+            if x.website:
+                with_website.append(x.id)
+        with_parent.sort()
+        without_parent.sort()
+        with_website.sort()
+
+        # We treat null values differently than in SQL. For instance in SQL:
+        #   SELECT id FROM res_partner WHERE parent_id NOT IN (0)
+        # will return only the records with non-null parent_id.
+        #   SELECT id FROM res_partner WHERE parent_id IN (0)
+        # will return expectedly nothing (our ids always begin at 1).
+        # This means the union of those two results will give only some
+        # records, but not all present in database.
+        #
+        # When using domains and the ORM's search method, we think it is
+        # more intuitive that the union returns all the records, and that
+        # a domain like ('parent_id', 'not in', [0]) will return all
+        # the records. For instance, if you perform a search for the companies
+        # that don't have OpenERP has a parent company, you expect to find,
+        # among others, the companies that don't have parent company.
+        #
+        # ('parent_id', 'not in', [0]) must give the same result than
+        # ('parent_id', 'not in', []), i.e. a empty set or a set with non-
+        # existing values be treated similarly if we simply check that some
+        # existing value belongs to them.
+
+        res_0 = self.search(cr, uid, [('parent_id', 'not like', 'probably_unexisting_name')]) # get all rows, included null parent_id
+        res_0.sort()
+        res_1 = self.search(cr, uid, [('parent_id', 'not in', [max_partner_id + 1])]) # get all rows, included null parent_id
+        res_1.sort()
+        res_2 = self.search(cr, uid, [('parent_id', 'not in', False)]) # get rows with not null parent_id, deprecated syntax
+        res_2.sort()
+        res_3 = self.search(cr, uid, [('parent_id', 'not in', [])]) # get all rows, included null parent_id
+        res_3.sort()
+        res_4 = self.search(cr, uid, [('parent_id', 'not in', [False])]) # get rows with not null parent_id
+        res_4.sort()
+        assert res_0 == partner_ids
+        assert res_1 == partner_ids
+        assert res_2 == with_parent
+        assert res_3 == partner_ids
+        assert res_4 == with_parent
+        # The results of these queries, when combined with queries 0..4 must
+        # give the whole set of ids.
+        res_5 = self.search(cr, uid, [('parent_id', 'like', 'probably_unexisting_name')])
+        res_5.sort()
+        res_6 = self.search(cr, uid, [('parent_id', 'in', [max_partner_id + 1])])
+        res_6.sort()
+        res_7 = self.search(cr, uid, [('parent_id', 'in', False)])
+        res_7.sort()
+        res_8 = self.search(cr, uid, [('parent_id', 'in', [])])
+        res_8.sort()
+        res_9 = self.search(cr, uid, [('parent_id', 'in', [False])])
+        res_9.sort()
+        assert res_5 == []
+        assert res_6 == []
+        assert res_7 == without_parent
+        assert res_8 == []
+        assert res_9 == without_parent
+        # These queries must return exactly the results than the queries 0..4,
+        # i.e. not ... in ... must be the same as ... not in ... .
+        res_10 = self.search(cr, uid, ['!', ('parent_id', 'like', 'probably_unexisting_name')])
+        res_10.sort()
+        res_11 = self.search(cr, uid, ['!', ('parent_id', 'in', [max_partner_id + 1])])
+        res_11.sort()
+        res_12 = self.search(cr, uid, ['!', ('parent_id', 'in', False)])
+        res_12.sort()
+        res_13 = self.search(cr, uid, ['!', ('parent_id', 'in', [])])
+        res_13.sort()
+        res_14 = self.search(cr, uid, ['!', ('parent_id', 'in', [False])])
+        res_14.sort()
+        assert res_0 == res_10
+        assert res_1 == res_11
+        assert res_2 == res_12
+        assert res_3 == res_13
+        assert res_4 == res_14
+
+        # Testing many2one field is not enough, a regular char field is tested
+        # with in [] and must not return any result.
+        res_15 = self.search(cr, uid, [('website', 'in', [])])
+        assert res_15 == []
+        # not in [] must return everything.
+        res_16 = self.search(cr, uid, [('website', 'not in', [])])
+        res_16.sort()
+        assert res_16 == partner_ids
+
+        res_17 = self.search(cr, uid, [('website', 'not in', False)])
+        res_17.sort()
+        assert res_17 == with_website
+-
+    Property of the query (one2many not in False).
+-
+    !python {model: res.currency }: |
+        ids = self.search(cr, uid, [])
+        referenced_companies = set([x.company_id.id for x in self.browse(cr, uid, ids)])
+        companies = set(self.pool.get('res.company').search(cr, uid, [('currency_ids', 'not in', False)]))
+        assert referenced_companies == companies
+-
+    Property of the query (one2many in False).
+-
+    !python {model: res.currency }: |
+        ids = self.search(cr, uid, [])
+        referenced_companies = set([x.company_id.id for x in self.browse(cr, uid, ids)])
+        unreferenced_companies = set(self.pool.get('res.company').search(cr, uid, [])).difference(referenced_companies)
+        companies = set(self.pool.get('res.company').search(cr, uid, [('currency_ids', 'in', False)]))
+        assert unreferenced_companies == companies
+-
+    Equivalent queries.
+-
+    !python {model: res.currency }: |
+        max_currency_id = max(self.search(cr, uid, []))
+        res_0 = self.search(cr, uid, [])
+        res_1 = self.search(cr, uid, [('name', 'not like', 'probably_unexisting_name')])
+        res_2 = self.search(cr, uid, [('id', 'not in', [max_currency_id + 1003])])
+        res_3 = self.search(cr, uid, [('id', 'not in', [])])
+        res_4 = self.search(cr, uid, [('id', 'not in', False)])
+        res_0.sort()
+        res_1.sort()
+        res_2.sort()
+        res_3.sort()
+        res_4.sort()
+        assert res_0 == res_1
+        assert res_0 == res_2
+        assert res_0 == res_3
+        assert res_0 == res_4
+-
+    Equivalent queries, integer and string.
+-
+    !python {model: res.partner }: |
+        all_ids = self.search(cr, uid, [])
+        if len(all_ids) > 1:
+            one = all_ids[0]
+            record = self.browse(cr, uid, one)
+            others = all_ids[1:]
+            res_1 = self.search(cr, uid, [('id', '=', one)])
+            # self.search(cr, uid, [('id', '!=', others)]) # not permitted
+            res_2 = self.search(cr, uid, [('id', 'not in', others)])
+            res_3 = self.search(cr, uid, ['!', ('id', '!=', one)])
+            res_4 = self.search(cr, uid, ['!', ('id', 'in', others)])
+            # res_5 = self.search(cr, uid, [('id', 'in', one)]) # TODO make it permitted, just like for child_of
+            res_6 = self.search(cr, uid, [('id', 'in', [one])])
+            res_7 = self.search(cr, uid, [('name', '=', record.name)])
+            res_8 = self.search(cr, uid, [('name', 'in', [record.name])])
+            # res_9 = self.search(cr, uid, [('name', 'in', record.name)]) # TODO
+            assert [one] == res_1
+            assert [one] == res_2
+            assert [one] == res_3
+            assert [one] == res_4
+            #assert [one] == res_5
+            assert [one] == res_6
+            assert [one] == res_7
+-
+    Need a company with a parent_id.
+-
+    !record {model: res.company, id: ymltest_company3}:
+        name: Acme 3
+-
+    Need a company with a parent_id.
+-
+    !record {model: res.company, id: ymltest_company4}:
+        name: Acme 4
+        parent_id: ymltest_company3
+-
+    Equivalent queries, one2many.
+-
+    !python {model: res.company }: |
+        # Search the company via its one2many (the one2many must point back at the company).
+        company = self.browse(cr, uid, ref('ymltest_company3'))
+        max_currency_id = max(self.pool.get('res.currency').search(cr, uid, []))
+        currency_ids1 = self.pool.get('res.currency').search(cr, uid, [('name', 'not like', 'probably_unexisting_name')])
+        currency_ids2 = self.pool.get('res.currency').search(cr, uid, [('id', 'not in', [max_currency_id + 1003])])
+        currency_ids3 = self.pool.get('res.currency').search(cr, uid, [('id', 'not in', [])])
+        assert currency_ids1 == currency_ids2 == currency_ids3, 'All 3 results should have be the same: all currencies'
+        default_company = self.browse(cr, uid, 1)
+        # one2many towards same model
+        res_1 = self.search(cr, uid, [('child_ids', 'in', [x.id for x in company.child_ids])]) # any company having a child of company3 as child
+        res_2 = self.search(cr, uid, [('child_ids', 'in', [company.child_ids[0].id])]) # any company having the first child of company3 as child
+        # one2many towards another model
+        res_3 = self.search(cr, uid, [('currency_ids', 'in', [x.id for x in default_company.currency_ids])]) # companies having a currency of main company
+        res_4 = self.search(cr, uid, [('currency_ids', 'in', [default_company.currency_ids[0].id])]) # companies having first currency of main company
+        res_5 = self.search(cr, uid, [('currency_ids', 'in', default_company.currency_ids[0].id)]) # companies having first currency of main company
+        # res_6 = self.search(cr, uid, [('currency_ids', 'in', [default_company.currency_ids[0].name])]) # TODO
+        res_7 = self.search(cr, uid, [('currency_ids', '=', default_company.currency_ids[0].name)])
+        res_8 = self.search(cr, uid, [('currency_ids', 'like', default_company.currency_ids[0].name)])
+        res_9 = self.search(cr, uid, [('currency_ids', 'like', 'probably_unexisting_name')])
+        # self.search(cr, uid, [('currency_ids', 'unexisting_op', 'probably_unexisting_name')]) # TODO expected exception
+        assert res_1 == [ref('ymltest_company3')]
+        assert res_2 == [ref('ymltest_company3')]
+        assert res_3 == [1]
+        assert res_4 == [1]
+        assert res_5 == [1]
+        assert res_7 == [1]
+        assert res_8 == [1]
+        assert res_9 == []
+
+        # get the companies referenced by some currency (this is normally the main company)
+        res_10 = self.search(cr, uid, [('currency_ids', 'not like', 'probably_unexisting_name')])
+        res_11 = self.search(cr, uid, [('currency_ids', 'not in', [max_currency_id + 1])])
+        res_12 = self.search(cr, uid, [('currency_ids', 'not in', False)])
+        res_13 = self.search(cr, uid, [('currency_ids', 'not in', [])])
+        res_10.sort()
+        res_11.sort()
+        res_12.sort()
+        res_13.sort()
+        assert res_10 == res_11
+        assert res_10 == res_12
+        assert res_10 == res_13
+
+        # child_of x returns x and its children (direct or not).
+        company = self.browse(cr, uid, ref('ymltest_company3'))
+        expected = [ref('ymltest_company3'), ref('ymltest_company4')]
+        expected.sort()
+        res_1 = self.search(cr, uid, [('id', 'child_of', [ref('ymltest_company3')])])
+        res_1.sort()
+        res_2 = self.search(cr, uid, [('id', 'child_of', ref('ymltest_company3'))])
+        res_2.sort()
+        res_3 = self.search(cr, uid, [('id', 'child_of', [company.name])])
+        res_3.sort()
+        res_4 = self.search(cr, uid, [('id', 'child_of', company.name)])
+        res_4.sort()
+        assert res_1 == expected
+        assert res_2 == expected
+        assert res_3 == expected
+        assert res_4 == expected
 -
     Verify that normalize_domain() works.
 -
@@ -187,6 +432,24 @@
         domain = [('x','in',['y','z']),('a.v','=','e'),'|','|',('a','=','b'),'!',('c','>','d'),('e','!=','f'),('g','=','h')]
         norm_domain = ['&','&','&'] + domain
         assert norm_domain == expression.normalize(domain), "Non-normalized domains should be properly normalized"
+-
+    Unaccent. Create a company with an accent in its name.
+-
+    !record {model: res.company, id: ymltest_unaccent_company}:
+        name: Hélène
+-
+    Test the unaccent-enabled 'ilike'.
+-
+    !python {model: res.company}: |
+        if self.pool.has_unaccent:
+            ids = self.search(cr, uid, [('name','ilike','Helene')], {})
+            assert ids == [ref('ymltest_unaccent_company')]
+            ids = self.search(cr, uid, [('name','ilike','hélène')], {})
+            assert ids == [ref('ymltest_unaccent_company')]
+            ids = self.search(cr, uid, [('name','not ilike','Helene')], {})
+            assert ref('ymltest_unaccent_company') not in ids
+            ids = self.search(cr, uid, [('name','not ilike','hélène')], {})
+            assert ref('ymltest_unaccent_company') not in ids
 -
     Check that =like/=ilike expressions (no wildcard variants of like/ilike) are working on an untranslated field.
 -

openerp/modules/db.py

@@ -123,4 +123,14 @@ def create_categories(cr, categories):
         categories = categories[1:]
     return p_id
 
+def has_unaccent(cr):
+    """ Test if the database has an unaccent function.
+
+    The unaccent is supposed to be provided by the PostgreSQL unaccent contrib
+    module but any similar function will be picked by OpenERP.
+
+    """
+    cr.execute("SELECT proname FROM pg_proc WHERE proname='unaccent'")
+    return len(cr.fetchall()) > 0
+
 # vim:expandtab:smartindent:tabstop=4:softtabstop=4:shiftwidth=4:

openerp/modules/registry.py

@@ -24,9 +24,12 @@
 """
 import threading
 
+import logging
+
 import openerp.sql_db
 import openerp.osv.orm
-
+import openerp.modules.db
+import openerp.tools.config
 
 class Registry(object):
     """ Model registry for a particular database.
@@ -45,6 +48,14 @@ class Registry(object):
         self.db_name = db_name
         self.db = openerp.sql_db.db_connect(db_name)
 
+        cr = self.db.cursor()
+        has_unaccent = openerp.modules.db.has_unaccent(cr)
+        if openerp.tools.config['unaccent'] and not has_unaccent:
+            logger = logging.getLogger('unaccent')
+            logger.warning("The option --unaccent was given but no unaccent() function was found in database.")
+        self.has_unaccent = openerp.tools.config['unaccent'] and has_unaccent
+        cr.close()
+
     def do_parent_store(self, cr):
         for o in self._init_parent:
             self.get(o)._parent_store_compute(cr)

openerp/osv/expression.py

@@ -20,17 +20,145 @@
 #
 ##############################################################################
 
+""" Domain expression processing
+
+The main duty of this module is to compile a domain expression into a SQL
+query. A lot of things should be documented here, but as a first step in the
+right direction, some tests in test_osv_expression.yml might give you some
+additional information.
+
+For legacy reasons, a domain uses an inconsistent two-levels abstract syntax
+(domains are regular Python data structures). At the first level, a domain
+is an expression made of terms (sometimes called leaves) and (domain) operators
+used in prefix notation. The available operators at this level are '!', '&',
+and '|'. '!' is a unary 'not', '&' is a binary 'and', and '|' is a binary 'or'.
+For instance, here is a possible domain. (<term> stands for an arbitrary term,
+more on this later.)
+
+    ['&', '!', <term1>, '|', <term2>, <term3>]
+
+It is equivalent to this pseudo code using infix notation:
+
+    (not <term1>) and (<term2> or <term3>)
+
+The second level of syntax deals with the term representation. A term is
+a triple of the form (left, operator, right). That is, a term uses an infix
+notation, and the available operators, and possible left and right operands
+differ with those of the previous level. Here is a possible term:
+
+    ('company_id.name', '=', 'OpenERP')
+
+The left and right operand don't have the same possible values. The left
+operand is field name (related to the model for which the domain applies).
+Actually, the field name can use the dot-notation to traverse relationships.
+The right operand is a Python value whose type should match the used operator
+and field type. In the above example, a string is used because the name field
+of a company has type string, and because we use the '=' operator. When
+appropriate, a 'in' operator can be used, and thus the right operand should be
+a list.
+
+Note: the non-uniform syntax could have been more uniform, but this would hide
+an important limitation of the domain syntax. Say that the term representation
+was ['=', 'company_id.name', 'OpenERP']. Used in a complete domain, this would
+look like:
+
+  ['!', ['=', 'company_id.name', 'OpenERP']]
+
+and you would be tempted to believe something like this would be possible:
+
+  ['!', ['=', 'company_id.name', ['&', ..., ...]]]
+
+That is, a domain could be a valid operand. But this is not the case. A domain
+is really limited to a two-level nature, and can not takes a recursive form: a
+domain is not a valid second-level operand.
+
+Unaccent - Accent-insensitive search
+
+OpenERP will use the SQL function 'unaccent' when available for the 'ilike' and
+'not ilike' operators, and enabled in the configuration.
+Normally the 'unaccent' function is obtained from the PostgreSQL 'unaccent'
+contrib module[0]. 
+
+
+..todo: The following explanation should be moved in some external installation
+        guide 
+
+The steps to install the module might differ on specific PostgreSQL versions.
+We give here some instruction for PostgreSQL 9.x on a Ubuntu system.
+
+Ubuntu doesn't come yet with PostgreSQL 9.x, so an alternative package source
+is used. We use Martin Pitt's PPA available at ppa:pitti/postgresql[1]. See
+[2] for instructions. Basically:
+
+    > sudo add-apt-repository ppa:pitti/postgresql
+    > sudo apt-get update
+
+Once the package list is up-to-date, you have to install PostgreSQL 9.0 and
+its contrib modules.
+
+    > sudo apt-get install postgresql-9.0 postgresql-contrib-9.0
+
+When you want to enable unaccent on some database:
+
+    > psql9 <database> -f /usr/share/postgresql/9.0/contrib/unaccent.sql
+
+Here 'psql9' is an alias for the newly installed PostgreSQL 9.0 tool, together
+with the correct port if necessary (for instance if PostgreSQL 8.4 is running
+on 5432). (Other aliases can be used for createdb and dropdb.)
+
+    > alias psql9='/usr/lib/postgresql/9.0/bin/psql -p 5433'
+
+You can check unaccent is working:
+
+    > psql9 <database> -c"select unaccent('hélène')"
+
+Finally, to instruct OpenERP to really use the unaccent function, you have to
+start the server specifying the --unaccent flag.
+
+[0] http://developer.postgresql.org/pgdocs/postgres/unaccent.html
+[1] https://launchpad.net/~pitti/+archive/postgresql
+[2] https://launchpad.net/+help/soyuz/ppa-sources-list.html
+
+"""
+
+import logging
+
 from openerp.tools import flatten, reverse_enumerate
 import fields
+import openerp.modules
 
 #.apidoc title: Domain Expressions
 
+# Domain operators.
 NOT_OPERATOR = '!'
 OR_OPERATOR = '|'
 AND_OPERATOR = '&'
+DOMAIN_OPERATORS = (NOT_OPERATOR, OR_OPERATOR, AND_OPERATOR)
 
-TRUE_DOMAIN = [(1,'=',1)]
-FALSE_DOMAIN = [(0,'=',1)]
+# List of available term operators. It is also possible to use the '<>'
+# operator, which is strictly the same as '!='; the later should be prefered
+# for consistency. This list doesn't contain '<>' as it is simpified to '!='
+# by the normalize_operator() function (so later part of the code deals with
+# only one representation).
+# An internal (i.e. not available to the user) 'inselect' operator is also
+# used. In this case its right operand has the form (subselect, params).
+TERM_OPERATORS = ('=', '!=', '<=', '<', '>', '>=', '=?', '=like', '=ilike',
+                  'like', 'not like', 'ilike', 'not ilike', 'in', 'not in',
+                  'child_of')
+
+# A subset of the above operators, with a 'negative' semantic. When the
+# expressions 'in NEGATIVE_TERM_OPERATORS' or 'not in NEGATIVE_TERM_OPERATORS' are used in the code
+# below, this doesn't necessarily mean that any of those NEGATIVE_TERM_OPERATORS is
+# legal in the processed term.
+NEGATIVE_TERM_OPERATORS = ('!=', 'not like', 'not ilike', 'not in')
+
+TRUE_LEAF = (1, '=', 1)
+FALSE_LEAF = (0, '=', 1)
+
+TRUE_DOMAIN = [TRUE_LEAF]
+FALSE_DOMAIN = [FALSE_LEAF]
+
+_logger = logging.getLogger('expression')
 
 def normalize(domain):
     """Returns a normalized version of ``domain_expr``, where all implicit '&' operators
@@ -45,10 +173,10 @@ def normalize(domain):
     op_arity = {NOT_OPERATOR: 1, AND_OPERATOR: 2, OR_OPERATOR: 2}
     for token in domain:
         if expected == 0:                   # more than expected, like in [A, B]
-            result[0:0] = ['&']             # put an extra '&' in front
+            result[0:0] = [AND_OPERATOR]             # put an extra '&' in front
             expected = 1
         result.append(token)
-        if isinstance(token, (list,tuple)): # domain term
+        if isinstance(token, (list, tuple)): # domain term
             expected -= 1
         else:
             expected += op_arity.get(token, 0) - 1
@@ -57,7 +185,8 @@ def normalize(domain):
 
 def combine(operator, unit, zero, domains):
     """Returns a new domain expression where all domain components from ``domains``
-       have been added together using the binary operator ``operator``.
+       have been added together using the binary operator ``operator``. The given
+       domains must be normalized.
 
        :param unit: the identity element of the domains "set" with regard to the operation
                     performed by ``operator``, i.e the domain component ``i`` which, when
@@ -69,6 +198,7 @@ def combine(operator, unit, zero, domains):
                     combined with any domain ``x`` via ``operator``, yields ``z``. 
                     E.g. [(1,'=',1)] is the typical zero for OR_OPERATOR: as soon as
                     you see it in a domain component the resulting domain is the zero.
+       :param domains: a list of normalized domains.
     """
     result = []
     count = 0
@@ -84,13 +214,130 @@ def combine(operator, unit, zero, domains):
     return result
 
 def AND(domains):
-    """ AND([D1,D2,...]) returns a domain representing D1 and D2 and ... """
+    """AND([D1,D2,...]) returns a domain representing D1 and D2 and ... """
     return combine(AND_OPERATOR, TRUE_DOMAIN, FALSE_DOMAIN, domains)
 
 def OR(domains):
-    """ OR([D1,D2,...]) returns a domain representing D1 or D2 or ... """
+    """OR([D1,D2,...]) returns a domain representing D1 or D2 or ... """
     return combine(OR_OPERATOR, FALSE_DOMAIN, TRUE_DOMAIN, domains)
 
+def is_operator(element):
+    """Test whether an object is a valid domain operator. """
+    return isinstance(element, basestring) and element in DOMAIN_OPERATORS
+
+# TODO change the share wizard to use this function.
+def is_leaf(element, internal=False):
+    """ Test whether an object is a valid domain term.
+
+    :param internal: allow or not the 'inselect' internal operator in the term.
+    This normally should be always left to False.
+    """
+    INTERNAL_OPS = TERM_OPERATORS + ('inselect',)
+    return (isinstance(element, tuple) or isinstance(element, list)) \
+       and len(element) == 3 \
+       and (((not internal) and element[1] in TERM_OPERATORS + ('<>',)) \
+            or (internal and element[1] in INTERNAL_OPS + ('<>',)))
+
+def normalize_leaf(left, operator, right):
+    """ Change a term's operator to some canonical form, simplifying later
+    processing.
+    """
+    original = operator
+    operator = operator.lower()
+    if operator == '<>':
+        operator = '!='
+    if isinstance(right, bool) and operator in ('in', 'not in'):
+        _logger.warning("The domain term '%s' should use the '=' or '!=' operator." % ((left, original, right),))
+        operator = '=' if operator == 'in' else '!='
+    if isinstance(right, (list, tuple)) and operator in ('=', '!='):
+        _logger.warning("The domain term '%s' should use the 'in' or 'not in' operator." % ((left, original, right),))
+        operator = 'in' if operator == '=' else 'not in'
+    return left, operator, right
+
+def distribute_not(domain):
+    """ Distribute any '!' domain operators found inside a normalized domain.
+
+    Because we don't use SQL semantic for processing a 'left not in right'
+    query (i.e. our 'not in' is not simply translated to a SQL 'not in'),
+    it means that a '! left in right' can not be simply processed
+    by __leaf_to_sql by first emitting code for 'left in right' then wrapping
+    the result with 'not (...)', as it would result in a 'not in' at the SQL
+    level.
+
+    This function is thus responsible for pushing any '!' domain operators
+    inside the terms themselves. For example::
+
+         ['!','&',('user_id','=',4),('partner_id','in',[1,2])]
+            will be turned into:
+         ['|',('user_id','!=',4),('partner_id','not in',[1,2])]
+
+    """
+    def negate(leaf):
+        """Negates and returns a single domain leaf term,
+        using the opposite operator if possible"""
+        left, operator, right = leaf
+        mapping = {
+            '<': '>=',
+            '>': '<=',
+            '<=': '>',
+            '>=': '<',
+            '=': '!=',
+            '!=': '=',
+        }
+        if operator in ('in', 'like', 'ilike'):
+            operator = 'not ' + operator
+            return [(left, operator, right)]
+        if operator in ('not in', 'not like', 'not ilike'):
+            operator = operator[4:]
+            return [(left, operator, right)]
+        if operator in mapping:
+            operator = mapping[operator]
+            return [(left, operator, right)]
+        return [NOT_OPERATOR, (left, operator, right)]
+    def distribute_negate(domain):
+        """Negate the domain ``subtree`` rooted at domain[0],
+        leaving the rest of the domain intact, and return
+        (negated_subtree, untouched_domain_rest)
+        """
+        if is_leaf(domain[0]):
+            return negate(domain[0]), domain[1:]
+        if domain[0] == AND_OPERATOR:
+            done1, todo1 = distribute_negate(domain[1:])
+            done2, todo2 = distribute_negate(todo1)
+            return [OR_OPERATOR] + done1 + done2, todo2
+        if domain[0] == OR_OPERATOR:
+            done1, todo1 = distribute_negate(domain[1:])
+            done2, todo2 = distribute_negate(todo1)
+            return [AND_OPERATOR] + done1 + done2, todo2
+    if not domain:
+        return []
+    if domain[0] != NOT_OPERATOR:
+        return [domain[0]] + distribute_not(domain[1:])
+    if domain[0] == NOT_OPERATOR:
+        done, todo = distribute_negate(domain[1:])
+        return done + distribute_not(todo)
+
+def select_from_where(cr, select_field, from_table, where_field, where_ids, where_operator):
+    # todo: merge into parent query as sub-query
+    res = []
+    if where_ids:
+        if where_operator in ['<','>','>=','<=']:
+            cr.execute('SELECT "%s" FROM "%s" WHERE "%s" %s %%s' % \
+                (select_field, from_table, where_field, where_operator),
+                (where_ids[0],)) # TODO shouldn't this be min/max(where_ids) ?
+            res = [r[0] for r in cr.fetchall()]
+        else: # TODO where_operator is supposed to be 'in'? It is called with child_of...
+            for i in range(0, len(where_ids), cr.IN_MAX):
+                subids = where_ids[i:i+cr.IN_MAX]
+                cr.execute('SELECT "%s" FROM "%s" WHERE "%s" IN %%s' % \
+                    (select_field, from_table, where_field), (tuple(subids),))
+                res.extend([r[0] for r in cr.fetchall()])
+    return res
+
+def select_distinct_from_where_not_null(cr, select_field, from_table):
+    cr.execute('SELECT distinct("%s") FROM "%s" where "%s" is not null' % \
+               (select_field, from_table, select_field))
+    return [r[0] for r in cr.fetchall()]
 
 class expression(object):
     """
@@ -100,127 +347,103 @@ class expression(object):
     For more info: http://christophe-simonis-at-tiny.blogspot.com/2008/08/new-new-domain-notation.html
     """
 
-    @classmethod
-    def _is_operator(cls, element):
-        return isinstance(element, (str, unicode)) and element in [AND_OPERATOR, OR_OPERATOR, NOT_OPERATOR]
-
-    @classmethod
-    def _is_leaf(cls, element, internal=False):
-        OPS = ('=', '!=', '<>', '<=', '<', '>', '>=', '=?', '=like', '=ilike', 'like', 'not like', 'ilike', 'not ilike', 'in', 'not in', 'child_of')
-        INTERNAL_OPS = OPS + ('inselect',)
-        return (isinstance(element, tuple) or isinstance(element, list)) \
-           and len(element) == 3 \
-           and (((not internal) and element[1] in OPS) \
-                or (internal and element[1] in INTERNAL_OPS))
-
-    def __execute_recursive_in(self, cr, s, f, w, ids, op, type):
-        # todo: merge into parent query as sub-query
-        res = []
-        if ids:
-            if op in ['<','>','>=','<=']:
-                cr.execute('SELECT "%s"'    \
-                               '  FROM "%s"'    \
-                               ' WHERE "%s" %s %%s' % (s, f, w, op), (ids[0],))
-                res.extend([r[0] for r in cr.fetchall()])
-            else:
-                for i in range(0, len(ids), cr.IN_MAX):
-                    subids = ids[i:i+cr.IN_MAX]
-                    cr.execute('SELECT "%s"'    \
-                               '  FROM "%s"'    \
-                               '  WHERE "%s" IN %%s' % (s, f, w),(tuple(subids),))
-                    res.extend([r[0] for r in cr.fetchall()])
-        else:
-            cr.execute('SELECT distinct("%s")'    \
-                           '  FROM "%s" where "%s" is not null'  % (s, f, s)),
-            res.extend([r[0] for r in cr.fetchall()])
-        return res
-
-    def __init__(self, exp):
-        # check if the expression is valid
-        if not reduce(lambda acc, val: acc and (self._is_operator(val) or self._is_leaf(val)), exp, True):
-            raise ValueError('Bad domain expression: %r' % (exp,))
-        self.__exp = exp
+    def __init__(self, cr, uid, exp, table, context):
+        self.has_unaccent = openerp.modules.registry.RegistryManager.get(cr.dbname).has_unaccent
         self.__field_tables = {}  # used to store the table to use for the sql generation. key = index of the leaf
         self.__all_tables = set()
         self.__joins = []
         self.__main_table = None # 'root' table. set by parse()
-        self.__DUMMY_LEAF = (1, '=', 1) # a dummy leaf that must not be parsed or sql generated
+        # assign self.__exp with the normalized, parsed domain.
+        self.parse(cr, uid, distribute_not(normalize(exp)), table, context)
 
+    # TODO used only for osv_memory
     @property
     def exp(self):
         return self.__exp[:]
 
-    def parse(self, cr, uid, table, context):
-        """ transform the leafs of the expression """
-        if not self.__exp:
-            return self
+    def parse(self, cr, uid, exp, table, context):
+        """ transform the leaves of the expression """
+        self.__exp = exp
+        self.__main_table = table
+        self.__all_tables.add(table)
 
-        def _rec_get(ids, table, parent=None, left='id', prefix=''):
-            if table._parent_store and (not table.pool._init):
-# TODO: Improve where joins are implemented for many with '.', replace by:
-# doms += ['&',(prefix+'.parent_left','<',o.parent_right),(prefix+'.parent_left','>=',o.parent_left)]
+        def child_of_domain(left, ids, left_model, parent=None, prefix=''):
+            """Returns a domain implementing the child_of operator for [(left,child_of,ids)],
+            either as a range using the parent_left/right tree lookup fields (when available),
+            or as an expanded [(left,in,child_ids)]"""
+            if left_model._parent_store and (not left_model.pool._init):
+                # TODO: Improve where joins are implemented for many with '.', replace by:
+                # doms += ['&',(prefix+'.parent_left','<',o.parent_right),(prefix+'.parent_left','>=',o.parent_left)]
                 doms = []
-                for o in table.browse(cr, uid, ids, context=context):
+                for o in left_model.browse(cr, uid, ids, context=context):
                     if doms:
                         doms.insert(0, OR_OPERATOR)
                     doms += [AND_OPERATOR, ('parent_left', '<', o.parent_right), ('parent_left', '>=', o.parent_left)]
                 if prefix:
-                    return [(left, 'in', table.search(cr, uid, doms, context=context))]
+                    return [(left, 'in', left_model.search(cr, uid, doms, context=context))]
                 return doms
             else:
-                def rg(ids, table, parent):
+                def recursive_children(ids, model, parent_field):
                     if not ids:
                         return []
-                    ids2 = table.search(cr, uid, [(parent, 'in', ids)], context=context)
-                    return ids + rg(ids2, table, parent)
-                return [(left, 'in', rg(ids, table, parent or table._parent_name))]
+                    ids2 = model.search(cr, uid, [(parent_field, 'in', ids)], context=context)
+                    return ids + recursive_children(ids2, model, parent_field)
+                return [(left, 'in', recursive_children(ids, left_model, parent or left_model._parent_name))]
 
-        def child_of_right_to_ids(value):
-            """ Normalize a single id, or a string, or a list of ids to a list of ids.
-
-            This function is always used with _rec_get() above, so it should be
-            called directly from _rec_get instead of repeatedly before _rec_get.
-
-            """
+        def to_ids(value, field_obj):
+            """Normalize a single id or name, or a list of those, into a list of ids"""
+            names = []
             if isinstance(value, basestring):
-                return [x[0] for x in field_obj.name_search(cr, uid, value, [], 'ilike', context=context, limit=None)]
+                names = [value]
+            if value and isinstance(value, (tuple, list)) and isinstance(value[0], basestring):
+                names = value
+            if names:
+                return flatten([[x[0] for x in field_obj.name_search(cr, uid, n, [], 'ilike', context=context, limit=None)] \
+                                    for n in names])
             elif isinstance(value, (int, long)):
                 return [value]
-            else:
-                return list(value)
-
-        self.__main_table = table
-        self.__all_tables.add(table)
+            return list(value)
 
         i = -1
         while i + 1<len(self.__exp):
             i += 1
             e = self.__exp[i]
-            if self._is_operator(e) or e == self.__DUMMY_LEAF:
+            if is_operator(e) or e == TRUE_LEAF or e == FALSE_LEAF:
                 continue
+
+            # check if the expression is valid
+            if not is_leaf(e):
+                raise ValueError('Bad domain expression: %r, %r is not a valid term.' % (exp, e))
+
+            # normalize the leaf's operator
+            e = normalize_leaf(*e)
+            self.__exp[i] = e
             left, operator, right = e
-            operator = operator.lower()
-            working_table = table
-            main_table = table
+
+            working_table = table # The table containing the field (the name provided in the left operand)
             fargs = left.split('.', 1)
+
+            # If the field is _inherits'd, search for the working_table,
+            # and extract the field.
             if fargs[0] in table._inherit_fields:
                 while True:
-                    field = main_table._columns.get(fargs[0], False)
+                    field = working_table._columns.get(fargs[0])
                     if field:
-                        working_table = main_table
                         self.__field_tables[i] = working_table
                         break
-                    working_table = main_table.pool.get(main_table._inherit_fields[fargs[0]][0])
-                    if working_table not in self.__all_tables:
-                        self.__joins.append('%s."%s"=%s."%s"' % (working_table._table, 'id', main_table._table, main_table._inherits[working_table._name]))
-                        self.__all_tables.add(working_table)
-                    main_table = working_table
+                    next_table = working_table.pool.get(working_table._inherit_fields[fargs[0]][0])
+                    if next_table not in self.__all_tables:
+                        self.__joins.append('%s."%s"=%s."%s"' % (next_table._table, 'id', working_table._table, working_table._inherits[next_table._name]))
+                        self.__all_tables.add(next_table)
+                    working_table = next_table
+            # Or (try to) directly extract the field.
+            else:
+                field = working_table._columns.get(fargs[0])
 
-            field = working_table._columns.get(fargs[0], False)
             if not field:
                 if left == 'id' and operator == 'child_of':
-                    ids2 = child_of_right_to_ids(right)
-                    dom = _rec_get(ids2, working_table)
+                    ids2 = to_ids(right, table)
+                    dom = child_of_domain(left, ids2, working_table)
                     self.__exp = self.__exp[:i] + dom + self.__exp[i+1:]
                 continue
 
@@ -228,20 +451,14 @@ class expression(object):
             if len(fargs) > 1:
                 if field._type == 'many2one':
                     right = field_obj.search(cr, uid, [(fargs[1], operator, right)], context=context)
-                    if right == []:
-                        self.__exp[i] = ( 'id', '=', 0 )
-                    else:
-                        self.__exp[i] = (fargs[0], 'in', right)
+                    self.__exp[i] = (fargs[0], 'in', right)
                 # Making search easier when there is a left operand as field.o2m or field.m2m
-                if field._type in ['many2many','one2many']:
+                if field._type in ['many2many', 'one2many']:
                     right = field_obj.search(cr, uid, [(fargs[1], operator, right)], context=context)
-                    right1 = table.search(cr, uid, [(fargs[0],'in', right)], context=context)
-                    if right1 == []:
-                        self.__exp[i] = ( 'id', '=', 0 )
-                    else:
-                        self.__exp[i] = ('id', 'in', right1)
+                    right1 = table.search(cr, uid, [(fargs[0], 'in', right)], context=context)
+                    self.__exp[i] = ('id', 'in', right1)
 
-                if not isinstance(field,fields.property):
+                if not isinstance(field, fields.property):
                     continue
 
             if field._properties and not field.store:
@@ -249,16 +466,16 @@ class expression(object):
                 if not field._fnct_search:
                     # the function field doesn't provide a search function and doesn't store
                     # values in the database, so we must ignore it : we generate a dummy leaf
-                    self.__exp[i] = self.__DUMMY_LEAF
+                    self.__exp[i] = TRUE_LEAF
                 else:
                     subexp = field.search(cr, uid, table, left, [self.__exp[i]], context=context)
                     if not subexp:
-                        self.__exp[i] = self.__DUMMY_LEAF
+                        self.__exp[i] = TRUE_LEAF
                     else:
                         # we assume that the expression is valid
                         # we create a dummy leaf for forcing the parsing of the resulting expression
                         self.__exp[i] = AND_OPERATOR
-                        self.__exp.insert(i + 1, self.__DUMMY_LEAF)
+                        self.__exp.insert(i + 1, TRUE_LEAF)
                         for j, se in enumerate(subexp):
                             self.__exp.insert(i + 2 + j, se)
             # else, the value of the field is store in the database, so we search on it
@@ -266,11 +483,11 @@ class expression(object):
             elif field._type == 'one2many':
                 # Applying recursivity on field(one2many)
                 if operator == 'child_of':
-                    ids2 = child_of_right_to_ids(right)
+                    ids2 = to_ids(right, field_obj)
                     if field._obj != working_table._name:
-                        dom = _rec_get(ids2, field_obj, left=left, prefix=field._obj)
+                        dom = child_of_domain(left, ids2, field_obj, prefix=field._obj)
                     else:
-                        dom = _rec_get(ids2, working_table, parent=left)
+                        dom = child_of_domain('id', ids2, working_table, parent=left)
                     self.__exp = self.__exp[:i] + dom + self.__exp[i+1:]
 
                 else:
@@ -282,7 +499,7 @@ class expression(object):
                             if ids2:
                                 operator = 'in'
                         else:
-                            if not isinstance(right,list):
+                            if not isinstance(right, list):
                                 ids2 = [right]
                             else:
                                 ids2 = right
@@ -290,22 +507,16 @@ class expression(object):
                             if operator in ['like','ilike','in','=']:
                                 #no result found with given search criteria
                                 call_null = False
-                                self.__exp[i] = ('id','=',0)
-                            else:
-                                call_null = True
-                                operator = 'in' # operator changed because ids are directly related to main object
+                                self.__exp[i] = FALSE_LEAF
                         else:
-                            call_null = False
-                            o2m_op = 'in'
-                            if operator in  ['not like','not ilike','not in','<>','!=']:
-                                o2m_op = 'not in'
-                            self.__exp[i] = ('id', o2m_op, self.__execute_recursive_in(cr, field._fields_id, field_obj._table, 'id', ids2, operator, field._type))
+                            ids2 = select_from_where(cr, field._fields_id, field_obj._table, 'id', ids2, operator)
+                            if ids2:
+                                call_null = False
+                                self.__exp[i] = ('id', 'in', ids2)
 
                     if call_null:
-                        o2m_op = 'not in'
-                        if operator in  ['not like','not ilike','not in','<>','!=']:
-                            o2m_op = 'in'
-                        self.__exp[i] = ('id', o2m_op, self.__execute_recursive_in(cr, field._fields_id, field_obj._table, 'id', [], operator, field._type) or [0])
+                        o2m_op = 'in' if operator in NEGATIVE_TERM_OPERATORS else 'not in'
+                        self.__exp[i] = ('id', o2m_op, select_distinct_from_where_not_null(cr, field._fields_id, field_obj._table))
 
             elif field._type == 'many2many':
                 #FIXME
@@ -313,10 +524,10 @@ class expression(object):
                     def _rec_convert(ids):
                         if field_obj == table:
                             return ids
-                        return self.__execute_recursive_in(cr, field._id1, field._rel, field._id2, ids, operator, field._type)
+                        return select_from_where(cr, field._id1, field._rel, field._id2, ids, operator)
 
-                    ids2 = child_of_right_to_ids(right)
-                    dom = _rec_get(ids2, field_obj)
+                    ids2 = to_ids(right, field_obj)
+                    dom = child_of_domain('id', ids2, field_obj)
                     ids2 = field_obj.search(cr, uid, dom, context=context)
                     self.__exp[i] = ('id', 'in', _rec_convert(ids2))
                 else:
@@ -335,34 +546,28 @@ class expression(object):
                             if operator in ['like','ilike','in','=']:
                                 #no result found with given search criteria
                                 call_null_m2m = False
-                                self.__exp[i] = ('id','=',0)
+                                self.__exp[i] = FALSE_LEAF
                             else:
-                                call_null_m2m = True
                                 operator = 'in' # operator changed because ids are directly related to main object
                         else:
                             call_null_m2m = False
-                            m2m_op = 'in'
-                            if operator in  ['not like','not ilike','not in','<>','!=']:
-                                m2m_op = 'not in'
+                            m2m_op = 'not in' if operator in NEGATIVE_TERM_OPERATORS else 'in'
+                            self.__exp[i] = ('id', m2m_op, select_from_where(cr, field._id1, field._rel, field._id2, res_ids, operator) or [0])
 
-                            self.__exp[i] = ('id', m2m_op, self.__execute_recursive_in(cr, field._id1, field._rel, field._id2, res_ids, operator, field._type) or [0])
                     if call_null_m2m:
-                        m2m_op = 'not in'
-                        if operator in  ['not like','not ilike','not in','<>','!=']:
-                            m2m_op = 'in'
-                        self.__exp[i] = ('id', m2m_op, self.__execute_recursive_in(cr, field._id1, field._rel, field._id2, [], operator,  field._type) or [0])
+                        m2m_op = 'in' if operator in NEGATIVE_TERM_OPERATORS else 'not in'
+                        self.__exp[i] = ('id', m2m_op, select_distinct_from_where_not_null(cr, field._id1, field._rel))
 
             elif field._type == 'many2one':
                 if operator == 'child_of':
-                    ids2 = child_of_right_to_ids(right)
-                    self.__operator = 'in'
+                    ids2 = to_ids(right, field_obj)
                     if field._obj != working_table._name:
-                        dom = _rec_get(ids2, field_obj, left=left, prefix=field._obj)
+                        dom = child_of_domain(left, ids2, field_obj, prefix=field._obj)
                     else:
-                        dom = _rec_get(ids2, working_table, parent=left)
+                        dom = child_of_domain('id', ids2, working_table, parent=left)
                     self.__exp = self.__exp[:i] + dom + self.__exp[i+1:]
                 else:
-                    def _get_expression(field_obj,cr, uid, left, right, operator, context=None):
+                    def _get_expression(field_obj, cr, uid, left, right, operator, context=None):
                         if context is None:
                             context = {}
                         c = context.copy()
@@ -370,46 +575,35 @@ class expression(object):
                         #Special treatment to ill-formed domains
                         operator = ( operator in ['<','>','<=','>='] ) and 'in' or operator
 
-                        dict_op = {'not in':'!=','in':'=','=':'in','!=':'not in','<>':'not in'}
-                        if isinstance(right,tuple):
+                        dict_op = {'not in':'!=','in':'=','=':'in','!=':'not in'}
+                        if isinstance(right, tuple):
                             right = list(right)
-                        if (not isinstance(right,list)) and operator in ['not in','in']:
+                        if (not isinstance(right, list)) and operator in ['not in','in']:
                             operator = dict_op[operator]
-                        elif isinstance(right,list) and operator in ['<>','!=','=']: #for domain (FIELD,'=',['value1','value2'])
+                        elif isinstance(right, list) and operator in ['!=','=']: #for domain (FIELD,'=',['value1','value2'])
                             operator = dict_op[operator]
-                        res_ids = field_obj.name_search(cr, uid, right, [], operator, limit=None, context=c)
-                        if not res_ids:
-                           return ('id','=',0)
-                        else:
-                            right = map(lambda x: x[0], res_ids)
-                            return (left, 'in', right)
+                        res_ids = [x[0] for x in field_obj.name_search(cr, uid, right, [], operator, limit=None, context=c)]
+                        if operator in NEGATIVE_TERM_OPERATORS:
+                            res_ids.append(False) # TODO this should not be appended if False was in 'right'
+                        return (left, 'in', res_ids)
 
                     m2o_str = False
                     if right:
                         if isinstance(right, basestring): # and not isinstance(field, fields.related):
                             m2o_str = True
-                        elif isinstance(right,(list,tuple)):
+                        elif isinstance(right, (list, tuple)):
                             m2o_str = True
                             for ele in right:
                                 if not isinstance(ele, basestring):
                                     m2o_str = False
                                     break
+                        if m2o_str:
+                            self.__exp[i] = _get_expression(field_obj, cr, uid, left, right, operator, context=context)
                     elif right == []:
-                        m2o_str = False
-                        if operator in ('not in', '!=', '<>'):
-                            # (many2one not in []) should return all records
-                            self.__exp[i] = self.__DUMMY_LEAF
-                        else:
-                            self.__exp[i] = ('id','=',0)
-                    else:
-                        new_op = '='
-                        if operator in  ['not like','not ilike','not in','<>','!=']:
-                            new_op = '!='
-                        #Is it ok to put 'left' and not 'id' ?
-                        self.__exp[i] = (left,new_op,False)
+                        pass # Handled by __leaf_to_sql().
+                    else: # right is False
+                        pass # Handled by __leaf_to_sql().
 
-                    if m2o_str:
-                        self.__exp[i] = _get_expression(field_obj,cr, uid, left, right, operator, context=context)
             else:
                 # other field type
                 # add the time part to datetime field when it's not there:
@@ -430,7 +624,7 @@ class expression(object):
                     if need_wildcard:
                         right = '%%%s%%' % right
 
-                    query1 = '( SELECT res_id'          \
+                    subselect = '( SELECT res_id'          \
                              '    FROM ir_translation'  \
                              '   WHERE name = %s'       \
                              '     AND lang = %s'       \
@@ -439,112 +633,143 @@ class expression(object):
                     #Covering in,not in operators with operands (%s,%s) ,etc.
                     if sql_operator in ['in','not in']:
                         instr = ','.join(['%s'] * len(right))
-                        query1 += '     AND value ' + sql_operator +  ' ' +" (" + instr + ")"   \
+                        subselect += '     AND value ' + sql_operator +  ' ' +" (" + instr + ")"   \
                              ') UNION ('                \
                              '  SELECT id'              \
                              '    FROM "' + working_table._table + '"'       \
                              '   WHERE "' + left + '" ' + sql_operator + ' ' +" (" + instr + "))"
                     else:
-                        query1 += '     AND value ' + sql_operator + instr +   \
+                        subselect += '     AND value ' + sql_operator + instr +   \
                              ') UNION ('                \
                              '  SELECT id'              \
                              '    FROM "' + working_table._table + '"'       \
                              '   WHERE "' + left + '" ' + sql_operator + instr + ")"
 
-                    query2 = [working_table._name + ',' + left,
+                    params = [working_table._name + ',' + left,
                               context.get('lang', False) or 'en_US',
                               'model',
                               right,
                               right,
                              ]
 
-                    self.__exp[i] = ('id', 'inselect', (query1, query2))
-        return self
+                    self.__exp[i] = ('id', 'inselect', (subselect, params))
 
     def __leaf_to_sql(self, leaf, table):
-        if leaf == self.__DUMMY_LEAF:
-            return ('(1=1)', [])
         left, operator, right = leaf
 
-        if operator == 'inselect':
-            query = '(%s."%s" in (%s))' % (table._table, left, right[0])
-            params = right[1]
-        elif operator in ['in', 'not in']:
-            params = right and right[:] or []
-            len_before = len(params)
-            for i in range(len_before)[::-1]:
-                if params[i] == False:
-                    del params[i]
-
-            len_after = len(params)
-            check_nulls = len_after != len_before
-            query = '(1=0)'
-
-            if len_after:
-                if left == 'id':
-                    instr = ','.join(['%s'] * len_after)
-                else:
-                    instr = ','.join([table._columns[left]._symbol_set[0]] * len_after)
-                query = '(%s."%s" %s (%s))' % (table._table, left, operator, instr)
-            else:
-                # the case for [field, 'in', []] or [left, 'not in', []]
-                if operator == 'in':
-                    query = '(%s."%s" IS NULL)' % (table._table, left)
-                else:
-                    query = '(%s."%s" IS NOT NULL)' % (table._table, left)
-            if check_nulls:
-                query = '(%s OR %s."%s" IS NULL)' % (query, table._table, left)
-        else:
+        if leaf == TRUE_LEAF:
+            query = 'TRUE'
             params = []
 
-            if right == False and (leaf[0] in table._columns)  and table._columns[leaf[0]]._type=="boolean"  and (operator == '='):
-                query = '(%s."%s" IS NULL or %s."%s" = false )' % (table._table, left,table._table, left)
-            elif (((right == False) and (type(right)==bool)) or (right is None)) and (operator == '='):
-                query = '%s."%s" IS NULL ' % (table._table, left)
-            elif right == False and (leaf[0] in table._columns)  and table._columns[leaf[0]]._type=="boolean"  and (operator in ['<>', '!=']):
-                query = '(%s."%s" IS NOT NULL and %s."%s" != false)' % (table._table, left,table._table, left)
-            elif (((right == False) and (type(right)==bool)) or right is None) and (operator in ['<>', '!=']):
-                query = '%s."%s" IS NOT NULL' % (table._table, left)
-            elif (operator == '=?'):
-                op = '='
-                if (right is False or right is None):
-                    return ( 'TRUE',[])
-                if left in table._columns:
-                        format = table._columns[left]._symbol_set[0]
-                        query = '(%s."%s" %s %s)' % (table._table, left, op, format)
-                        params = table._columns[left]._symbol_set[1](right)
-                else:
-                        query = "(%s.\"%s\" %s '%%s')" % (table._table, left, op)
-                        params = right
+        elif leaf == FALSE_LEAF:
+            query = 'FALSE'
+            params = []
 
-            else:
-                if left == 'id':
-                    query = '%s.id %s %%s' % (table._table, operator)
-                    params = right
+        elif operator == 'inselect':
+            query = '(%s."%s" in (%s))' % (table._table, left, right[0])
+            params = right[1]
+
+        elif operator in ['in', 'not in']:
+            # Two cases: right is a boolean or a list. The boolean case is an
+            # abuse and handled for backward compatibility.
+            if isinstance(right, bool):
+                _logger.warning("The domain term '%s' should use the '=' or '!=' operator." % (leaf,))
+                if operator == 'in':
+                    r = 'NOT NULL' if right else 'NULL'
                 else:
-                    need_wildcard = operator in ('like', 'ilike', 'not like', 'not ilike')
-                    sql_operator = {'=like':'like','=ilike':'ilike'}.get(operator,operator)
-                    if left in table._columns:
-                        format = need_wildcard and '%s' or table._columns[left]._symbol_set[0]
-                        query = '(%s."%s" %s %s)' % (table._table, left, sql_operator, format)
+                    r = 'NULL' if right else 'NOT NULL'
+                query = '(%s."%s" IS %s)' % (table._table, left, r)
+                params = []
+            elif isinstance(right, (list, tuple)):
+                params = right[:]
+                check_nulls = False
+                for i in range(len(params))[::-1]:
+                    if params[i] == False:
+                        check_nulls = True
+                        del params[i]
+
+                if params:
+                    if left == 'id':
+                        instr = ','.join(['%s'] * len(params))
                     else:
-                        query = "(%s.\"%s\" %s '%s')" % (table._table, left, sql_operator, right)
+                        instr = ','.join([table._columns[left]._symbol_set[0]] * len(params))
+                    query = '(%s."%s" %s (%s))' % (table._table, left, operator, instr)
+                else:
+                    # The case for (left, 'in', []) or (left, 'not in', []).
+                    query = 'FALSE' if operator == 'in' else 'TRUE'
 
-                    add_null = False
-                    if need_wildcard:
-                        if isinstance(right, str):
-                            str_utf8 = right
-                        elif isinstance(right, unicode):
-                            str_utf8 = right.encode('utf-8')
-                        else:
-                            str_utf8 = str(right)
-                        params = '%%%s%%' % str_utf8
-                        add_null = not str_utf8
-                    elif left in table._columns:
-                        params = table._columns[left]._symbol_set[1](right)
+                if check_nulls and operator == 'in':
+                    query = '(%s OR %s."%s" IS NULL)' % (query, table._table, left)
+                elif not check_nulls and operator == 'not in':
+                    query = '(%s OR %s."%s" IS NULL)' % (query, table._table, left)
+                elif check_nulls and operator == 'not in':
+                    query = '(%s AND %s."%s" IS NOT NULL)' % (query, table._table, left) # needed only for TRUE.
+            else: # Must not happen.
+                pass
 
-                    if add_null:
-                        query = '(%s OR \"%s\" IS NULL)' % (query, left)
+        elif right == False and (left in table._columns) and table._columns[left]._type=="boolean" and (operator == '='):
+            query = '(%s."%s" IS NULL or %s."%s" = false )' % (table._table, left, table._table, left)
+            params = []
+
+        elif (right is False or right is None) and (operator == '='):
+            query = '%s."%s" IS NULL ' % (table._table, left)
+            params = []
+
+        elif right == False and (left in table._columns) and table._columns[left]._type=="boolean" and (operator == '!='):
+            query = '(%s."%s" IS NOT NULL and %s."%s" != false)' % (table._table, left, table._table, left)
+            params = []
+
+        elif (right is False or right is None) and (operator == '!='):
+            query = '%s."%s" IS NOT NULL' % (table._table, left)
+            params = []
+
+        elif (operator == '=?'):
+            if (right is False or right is None):
+                query = 'TRUE'
+                params = []
+            elif left in table._columns:
+                format = table._columns[left]._symbol_set[0]
+                query = '(%s."%s" = %s)' % (table._table, left, format)
+                params = table._columns[left]._symbol_set[1](right)
+            else:
+                query = "(%s.\"%s\" = '%%s')" % (table._table, left)
+                params = right
+
+        elif left == 'id':
+            query = '%s.id %s %%s' % (table._table, operator)
+            params = right
+
+        else:
+            need_wildcard = operator in ('like', 'ilike', 'not like', 'not ilike')
+            sql_operator = {'=like':'like','=ilike':'ilike'}.get(operator,operator)
+
+            if left in table._columns:
+                format = need_wildcard and '%s' or table._columns[left]._symbol_set[0]
+                if self.has_unaccent and sql_operator in ('ilike', 'not ilike'):
+                    query = '(unaccent(%s."%s") %s unaccent(%s))' % (table._table, left, sql_operator, format)
+                else:
+                    query = '(%s."%s" %s %s)' % (table._table, left, sql_operator, format)
+            else:
+                if self.has_unaccent and sql_operator in ('ilike', 'not ilike'):
+                    query = "(unaccent(%s.\"%s\") %s unaccent('%s'))" % (table._table, left, sql_operator, right)
+                else:
+                    query = "(%s.\"%s\" %s '%s')" % (table._table, left, sql_operator, right)
+
+            add_null = False
+            if need_wildcard:
+                if isinstance(right, str):
+                    str_utf8 = right
+                elif isinstance(right, unicode):
+                    str_utf8 = right.encode('utf-8')
+                else:
+                    str_utf8 = str(right)
+                params = '%%%s%%' % str_utf8
+                add_null = not str_utf8
+            elif left in table._columns:
+                params = table._columns[left]._symbol_set[1](right)
+
+            if add_null:
+                query = '(%s OR %s."%s" IS NULL)' % (query, table._table, left)
 
         if isinstance(params, basestring):
             params = [params]
@@ -554,25 +779,26 @@ class expression(object):
     def to_sql(self):
         stack = []
         params = []
+        # Process the domain from right to left, using a stack, to generate a SQL expression.
         for i, e in reverse_enumerate(self.__exp):
-            if self._is_leaf(e, internal=True):
+            if is_leaf(e, internal=True):
                 table = self.__field_tables.get(i, self.__main_table)
                 q, p = self.__leaf_to_sql(e, table)
                 params.insert(0, p)
                 stack.append(q)
+            elif e == NOT_OPERATOR:
+                stack.append('(NOT (%s))' % (stack.pop(),))
             else:
-                if e == NOT_OPERATOR:
-                    stack.append('(NOT (%s))' % (stack.pop(),))
-                else:
-                    ops = {AND_OPERATOR: ' AND ', OR_OPERATOR: ' OR '}
-                    q1 = stack.pop()
-                    q2 = stack.pop()
-                    stack.append('(%s %s %s)' % (q1, ops[e], q2,))
+                ops = {AND_OPERATOR: ' AND ', OR_OPERATOR: ' OR '}
+                q1 = stack.pop()
+                q2 = stack.pop()
+                stack.append('(%s %s %s)' % (q1, ops[e], q2,))
 
-        query = ' AND '.join(reversed(stack))
+        assert len(stack) == 1
+        query = stack[0]
         joins = ' AND '.join(self.__joins)
         if joins:
-            query = '(%s) AND (%s)' % (joins, query)
+            query = '(%s) AND %s' % (joins, query)
         return (query, flatten(params))
 
     def get_tables(self):

openerp/osv/orm.py

@@ -2459,8 +2459,7 @@ class orm_memory(orm_template):
                 args = [('active', '=', 1)]
         if args:
             import expression
-            e = expression.expression(args)
-            e.parse(cr, user, self, context)
+            e = expression.expression(cr, user, args, self, context)
             res = e.exp
         return res or []
 
@@ -2490,6 +2489,9 @@ class orm_memory(orm_template):
                     break
                 f = True
                 for arg in result:
+                    if len(arg) != 3:
+                       # Amazing hack: orm_memory handles only simple domains.
+                       continue
                     if arg[1] == '=':
                         val = eval('data[arg[0]]'+'==' +' arg[2]', locals())
                     elif arg[1] in ['<', '>', 'in', 'not in', '<=', '>=', '<>']:
@@ -4418,8 +4420,7 @@ class orm(orm_template):
 
         if domain:
             import expression
-            e = expression.expression(domain)
-            e.parse(cr, user, self, context)
+            e = expression.expression(cr, user, domain, self, context)
             tables = e.get_tables()
             where_clause, where_params = e.to_sql()
             where_clause = where_clause and [where_clause] or []

openerp/tools/config.py

@@ -250,6 +250,8 @@ class configmanager(object):
                               "osv_memory tables. This is a decimal value expressed in hours, "
                               "and the default is 1 hour.",
                          type="float")
+        group.add_option("--unaccent", dest="unaccent", my_default=False, action="store_true",
+                         help="Use the unaccent function provided by the database when available.")
         parser.add_option_group(group)
 
         # Copy all optparse options (i.e. MyOption) into self.options.
@@ -335,7 +337,7 @@ class configmanager(object):
             'stop_after_init', 'logrotate', 'without_demo', 'netrpc', 'xmlrpc', 'syslog',
             'list_db', 'xmlrpcs',
             'test_file', 'test_disable', 'test_commit', 'test_report_directory',
-            'osv_memory_count_limit', 'osv_memory_age_limit',
+            'osv_memory_count_limit', 'osv_memory_age_limit', 'unaccent',
         ]
 
         for arg in keys: