Updated SqlAlchemy

This commit is contained in:
2017-04-15 16:27:12 +00:00
parent 2c790e1fe1
commit e3267d4bda
59 changed files with 30236 additions and 26049 deletions

View File

@@ -1,4 +1,11 @@
from sqlalchemy.sql.expression import (
# sql/__init__.py
# Copyright (C) 2005-2017 the SQLAlchemy authors and contributors
# <see AUTHORS file>
#
# This module is part of SQLAlchemy and is released under
# the MIT License: http://www.opensource.org/licenses/mit-license.php
from .expression import (
Alias,
ClauseElement,
ColumnCollection,
@@ -11,9 +18,12 @@ from sqlalchemy.sql.expression import (
Select,
Selectable,
TableClause,
TableSample,
Update,
alias,
and_,
any_,
all_,
asc,
between,
bindparam,
@@ -28,12 +38,16 @@ from sqlalchemy.sql.expression import (
except_all,
exists,
extract,
false,
False_,
func,
funcfilter,
insert,
intersect,
intersect_all,
join,
label,
lateral,
literal,
literal_column,
modifier,
@@ -42,17 +56,43 @@ from sqlalchemy.sql.expression import (
or_,
outerjoin,
outparam,
over,
select,
subquery,
table,
tablesample,
text,
true,
True_,
tuple_,
type_coerce,
union,
union_all,
update,
)
within_group
)
from sqlalchemy.sql.visitors import ClauseVisitor
from .visitors import ClauseVisitor
__tmp = locals().keys()
__all__ = sorted([i for i in __tmp if not i.startswith('__')])
def __go(lcls):
global __all__
from .. import util as _sa_util
import inspect as _inspect
__all__ = sorted(name for name, obj in lcls.items()
if not (name.startswith('_') or _inspect.ismodule(obj)))
from .annotation import _prepare_annotations, Annotated
from .elements import AnnotatedColumnElement, ClauseList
from .selectable import AnnotatedFromClause
_prepare_annotations(ColumnElement, AnnotatedColumnElement)
_prepare_annotations(FromClause, AnnotatedFromClause)
_prepare_annotations(ClauseList, Annotated)
_sa_util.dependencies.resolve_all("sqlalchemy.sql")
from . import naming
__go(locals())

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

View File

@@ -1,104 +1,813 @@
from sqlalchemy import types as sqltypes
from sqlalchemy.sql.expression import (
ClauseList, Function, _literal_as_binds, text, _type_from_args
)
from sqlalchemy.sql import operators
from sqlalchemy.sql.visitors import VisitableType
# sql/functions.py
# Copyright (C) 2005-2017 the SQLAlchemy authors and contributors
# <see AUTHORS file>
#
# This module is part of SQLAlchemy and is released under
# the MIT License: http://www.opensource.org/licenses/mit-license.php
"""SQL function API, factories, and built-in functions.
"""
from . import sqltypes, schema
from .base import Executable, ColumnCollection
from .elements import ClauseList, Cast, Extract, _literal_as_binds, \
literal_column, _type_from_args, ColumnElement, _clone,\
Over, BindParameter, FunctionFilter, Grouping, WithinGroup
from .selectable import FromClause, Select, Alias
from . import util as sqlutil
from . import operators
from .visitors import VisitableType
from .. import util
from . import annotation
_registry = util.defaultdict(dict)
def register_function(identifier, fn, package="_default"):
"""Associate a callable with a particular func. name.
This is normally called by _GenericMeta, but is also
available by itself so that a non-Function construct
can be associated with the :data:`.func` accessor (i.e.
CAST, EXTRACT).
"""
reg = _registry[package]
reg[identifier] = fn
class FunctionElement(Executable, ColumnElement, FromClause):
"""Base for SQL function-oriented constructs.
.. seealso::
:class:`.Function` - named SQL function.
:data:`.func` - namespace which produces registered or ad-hoc
:class:`.Function` instances.
:class:`.GenericFunction` - allows creation of registered function
types.
"""
packagenames = ()
def __init__(self, *clauses, **kwargs):
"""Construct a :class:`.FunctionElement`.
"""
args = [_literal_as_binds(c, self.name) for c in clauses]
self.clause_expr = ClauseList(
operator=operators.comma_op,
group_contents=True, *args).\
self_group()
def _execute_on_connection(self, connection, multiparams, params):
return connection._execute_function(self, multiparams, params)
@property
def columns(self):
"""The set of columns exported by this :class:`.FunctionElement`.
Function objects currently have no result column names built in;
this method returns a single-element column collection with
an anonymously named column.
An interim approach to providing named columns for a function
as a FROM clause is to build a :func:`.select` with the
desired columns::
from sqlalchemy.sql import column
stmt = select([column('x'), column('y')]).\
select_from(func.myfunction())
"""
return ColumnCollection(self.label(None))
@util.memoized_property
def clauses(self):
"""Return the underlying :class:`.ClauseList` which contains
the arguments for this :class:`.FunctionElement`.
"""
return self.clause_expr.element
def over(self, partition_by=None, order_by=None, rows=None, range_=None):
"""Produce an OVER clause against this function.
Used against aggregate or so-called "window" functions,
for database backends that support window functions.
The expression::
func.row_number().over(order_by='x')
is shorthand for::
from sqlalchemy import over
over(func.row_number(), order_by='x')
See :func:`~.expression.over` for a full description.
.. versionadded:: 0.7
"""
return Over(
self,
partition_by=partition_by,
order_by=order_by,
rows=rows,
range_=range_
)
def within_group(self, *order_by):
"""Produce a WITHIN GROUP (ORDER BY expr) clause against this function.
Used against so-called "ordered set aggregate" and "hypothetical
set aggregate" functions, including :class:`.percentile_cont`,
:class:`.rank`, :class:`.dense_rank`, etc.
See :func:`~.expression.within_group` for a full description.
.. versionadded:: 1.1
"""
return WithinGroup(self, *order_by)
def filter(self, *criterion):
"""Produce a FILTER clause against this function.
Used against aggregate and window functions,
for database backends that support the "FILTER" clause.
The expression::
func.count(1).filter(True)
is shorthand for::
from sqlalchemy import funcfilter
funcfilter(func.count(1), True)
.. versionadded:: 1.0.0
.. seealso::
:class:`.FunctionFilter`
:func:`.funcfilter`
"""
if not criterion:
return self
return FunctionFilter(self, *criterion)
@property
def _from_objects(self):
return self.clauses._from_objects
def get_children(self, **kwargs):
return self.clause_expr,
def _copy_internals(self, clone=_clone, **kw):
self.clause_expr = clone(self.clause_expr, **kw)
self._reset_exported()
FunctionElement.clauses._reset(self)
def within_group_type(self, within_group):
"""For types that define their return type as based on the criteria
within a WITHIN GROUP (ORDER BY) expression, called by the
:class:`.WithinGroup` construct.
Returns None by default, in which case the function's normal ``.type``
is used.
"""
return None
def alias(self, name=None, flat=False):
r"""Produce a :class:`.Alias` construct against this
:class:`.FunctionElement`.
This construct wraps the function in a named alias which
is suitable for the FROM clause, in the style accepted for example
by PostgreSQL.
e.g.::
from sqlalchemy.sql import column
stmt = select([column('data_view')]).\
select_from(SomeTable).\
select_from(func.unnest(SomeTable.data).alias('data_view')
)
Would produce:
.. sourcecode:: sql
SELECT data_view
FROM sometable, unnest(sometable.data) AS data_view
.. versionadded:: 0.9.8 The :meth:`.FunctionElement.alias` method
is now supported. Previously, this method's behavior was
undefined and did not behave consistently across versions.
"""
return Alias(self, name)
def select(self):
"""Produce a :func:`~.expression.select` construct
against this :class:`.FunctionElement`.
This is shorthand for::
s = select([function_element])
"""
s = Select([self])
if self._execution_options:
s = s.execution_options(**self._execution_options)
return s
def scalar(self):
"""Execute this :class:`.FunctionElement` against an embedded
'bind' and return a scalar value.
This first calls :meth:`~.FunctionElement.select` to
produce a SELECT construct.
Note that :class:`.FunctionElement` can be passed to
the :meth:`.Connectable.scalar` method of :class:`.Connection`
or :class:`.Engine`.
"""
return self.select().execute().scalar()
def execute(self):
"""Execute this :class:`.FunctionElement` against an embedded
'bind'.
This first calls :meth:`~.FunctionElement.select` to
produce a SELECT construct.
Note that :class:`.FunctionElement` can be passed to
the :meth:`.Connectable.execute` method of :class:`.Connection`
or :class:`.Engine`.
"""
return self.select().execute()
def _bind_param(self, operator, obj, type_=None):
return BindParameter(None, obj, _compared_to_operator=operator,
_compared_to_type=self.type, unique=True,
type_=type_)
def self_group(self, against=None):
# for the moment, we are parenthesizing all array-returning
# expressions against getitem. This may need to be made
# more portable if in the future we support other DBs
# besides postgresql.
if against is operators.getitem and \
isinstance(self.type, sqltypes.ARRAY):
return Grouping(self)
else:
return super(FunctionElement, self).self_group(against=against)
class _FunctionGenerator(object):
"""Generate :class:`.Function` objects based on getattr calls."""
def __init__(self, **opts):
self.__names = []
self.opts = opts
def __getattr__(self, name):
# passthru __ attributes; fixes pydoc
if name.startswith('__'):
try:
return self.__dict__[name]
except KeyError:
raise AttributeError(name)
elif name.endswith('_'):
name = name[0:-1]
f = _FunctionGenerator(**self.opts)
f.__names = list(self.__names) + [name]
return f
def __call__(self, *c, **kwargs):
o = self.opts.copy()
o.update(kwargs)
tokens = len(self.__names)
if tokens == 2:
package, fname = self.__names
elif tokens == 1:
package, fname = "_default", self.__names[0]
else:
package = None
if package is not None:
func = _registry[package].get(fname)
if func is not None:
return func(*c, **o)
return Function(self.__names[-1],
packagenames=self.__names[0:-1], *c, **o)
func = _FunctionGenerator()
"""Generate SQL function expressions.
:data:`.func` is a special object instance which generates SQL
functions based on name-based attributes, e.g.::
>>> print(func.count(1))
count(:param_1)
The element is a column-oriented SQL element like any other, and is
used in that way::
>>> print(select([func.count(table.c.id)]))
SELECT count(sometable.id) FROM sometable
Any name can be given to :data:`.func`. If the function name is unknown to
SQLAlchemy, it will be rendered exactly as is. For common SQL functions
which SQLAlchemy is aware of, the name may be interpreted as a *generic
function* which will be compiled appropriately to the target database::
>>> print(func.current_timestamp())
CURRENT_TIMESTAMP
To call functions which are present in dot-separated packages,
specify them in the same manner::
>>> print(func.stats.yield_curve(5, 10))
stats.yield_curve(:yield_curve_1, :yield_curve_2)
SQLAlchemy can be made aware of the return type of functions to enable
type-specific lexical and result-based behavior. For example, to ensure
that a string-based function returns a Unicode value and is similarly
treated as a string in expressions, specify
:class:`~sqlalchemy.types.Unicode` as the type:
>>> print(func.my_string(u'hi', type_=Unicode) + ' ' +
... func.my_string(u'there', type_=Unicode))
my_string(:my_string_1) || :my_string_2 || my_string(:my_string_3)
The object returned by a :data:`.func` call is usually an instance of
:class:`.Function`.
This object meets the "column" interface, including comparison and labeling
functions. The object can also be passed the :meth:`~.Connectable.execute`
method of a :class:`.Connection` or :class:`.Engine`, where it will be
wrapped inside of a SELECT statement first::
print(connection.execute(func.current_timestamp()).scalar())
In a few exception cases, the :data:`.func` accessor
will redirect a name to a built-in expression such as :func:`.cast`
or :func:`.extract`, as these names have well-known meaning
but are not exactly the same as "functions" from a SQLAlchemy
perspective.
.. versionadded:: 0.8 :data:`.func` can return non-function expression
constructs for common quasi-functional names like :func:`.cast`
and :func:`.extract`.
Functions which are interpreted as "generic" functions know how to
calculate their return type automatically. For a listing of known generic
functions, see :ref:`generic_functions`.
.. note::
The :data:`.func` construct has only limited support for calling
standalone "stored procedures", especially those with special
parameterization concerns.
See the section :ref:`stored_procedures` for details on how to use
the DBAPI-level ``callproc()`` method for fully traditional stored
procedures.
"""
modifier = _FunctionGenerator(group=False)
class Function(FunctionElement):
"""Describe a named SQL function.
See the superclass :class:`.FunctionElement` for a description
of public methods.
.. seealso::
:data:`.func` - namespace which produces registered or ad-hoc
:class:`.Function` instances.
:class:`.GenericFunction` - allows creation of registered function
types.
"""
__visit_name__ = 'function'
def __init__(self, name, *clauses, **kw):
"""Construct a :class:`.Function`.
The :data:`.func` construct is normally used to construct
new :class:`.Function` instances.
"""
self.packagenames = kw.pop('packagenames', None) or []
self.name = name
self._bind = kw.get('bind', None)
self.type = sqltypes.to_instance(kw.get('type_', None))
FunctionElement.__init__(self, *clauses, **kw)
def _bind_param(self, operator, obj, type_=None):
return BindParameter(self.name, obj,
_compared_to_operator=operator,
_compared_to_type=self.type,
type_=type_,
unique=True)
class _GenericMeta(VisitableType):
def __call__(self, *args, **kwargs):
args = [_literal_as_binds(c) for c in args]
return type.__call__(self, *args, **kwargs)
def __init__(cls, clsname, bases, clsdict):
if annotation.Annotated not in cls.__mro__:
cls.name = name = clsdict.get('name', clsname)
cls.identifier = identifier = clsdict.get('identifier', name)
package = clsdict.pop('package', '_default')
# legacy
if '__return_type__' in clsdict:
cls.type = clsdict['__return_type__']
register_function(identifier, cls, package)
super(_GenericMeta, cls).__init__(clsname, bases, clsdict)
class GenericFunction(Function):
__metaclass__ = _GenericMeta
def __init__(self, type_=None, args=(), **kwargs):
class GenericFunction(util.with_metaclass(_GenericMeta, Function)):
"""Define a 'generic' function.
A generic function is a pre-established :class:`.Function`
class that is instantiated automatically when called
by name from the :data:`.func` attribute. Note that
calling any name from :data:`.func` has the effect that
a new :class:`.Function` instance is created automatically,
given that name. The primary use case for defining
a :class:`.GenericFunction` class is so that a function
of a particular name may be given a fixed return type.
It can also include custom argument parsing schemes as well
as additional methods.
Subclasses of :class:`.GenericFunction` are automatically
registered under the name of the class. For
example, a user-defined function ``as_utc()`` would
be available immediately::
from sqlalchemy.sql.functions import GenericFunction
from sqlalchemy.types import DateTime
class as_utc(GenericFunction):
type = DateTime
print select([func.as_utc()])
User-defined generic functions can be organized into
packages by specifying the "package" attribute when defining
:class:`.GenericFunction`. Third party libraries
containing many functions may want to use this in order
to avoid name conflicts with other systems. For example,
if our ``as_utc()`` function were part of a package
"time"::
class as_utc(GenericFunction):
type = DateTime
package = "time"
The above function would be available from :data:`.func`
using the package name ``time``::
print select([func.time.as_utc()])
A final option is to allow the function to be accessed
from one name in :data:`.func` but to render as a different name.
The ``identifier`` attribute will override the name used to
access the function as loaded from :data:`.func`, but will retain
the usage of ``name`` as the rendered name::
class GeoBuffer(GenericFunction):
type = Geometry
package = "geo"
name = "ST_Buffer"
identifier = "buffer"
The above function will render as follows::
>>> print func.geo.buffer()
ST_Buffer()
.. versionadded:: 0.8 :class:`.GenericFunction` now supports
automatic registration of new functions as well as package
and custom naming support.
.. versionchanged:: 0.8 The attribute name ``type`` is used
to specify the function's return type at the class level.
Previously, the name ``__return_type__`` was used. This
name is still recognized for backwards-compatibility.
"""
coerce_arguments = True
def __init__(self, *args, **kwargs):
parsed_args = kwargs.pop('_parsed_args', None)
if parsed_args is None:
parsed_args = [_literal_as_binds(c, self.name) for c in args]
self.packagenames = []
self.name = self.__class__.__name__
self._bind = kwargs.get('bind', None)
self.clause_expr = ClauseList(
operator=operators.comma_op,
group_contents=True, *args).self_group()
operator=operators.comma_op,
group_contents=True, *parsed_args).self_group()
self.type = sqltypes.to_instance(
type_ or getattr(self, '__return_type__', None))
kwargs.pop("type_", None) or getattr(self, 'type', None))
register_function("cast", Cast)
register_function("extract", Extract)
class next_value(GenericFunction):
"""Represent the 'next value', given a :class:`.Sequence`
as its single argument.
Compiles into the appropriate function on each backend,
or will raise NotImplementedError if used on a backend
that does not provide support for sequences.
"""
type = sqltypes.Integer()
name = "next_value"
def __init__(self, seq, **kw):
assert isinstance(seq, schema.Sequence), \
"next_value() accepts a Sequence object as input."
self._bind = kw.get('bind', None)
self.sequence = seq
@property
def _from_objects(self):
return []
class AnsiFunction(GenericFunction):
def __init__(self, **kwargs):
GenericFunction.__init__(self, **kwargs)
class ReturnTypeFromArgs(GenericFunction):
"""Define a function whose return type is the same as its arguments."""
def __init__(self, *args, **kwargs):
args = [_literal_as_binds(c, self.name) for c in args]
kwargs.setdefault('type_', _type_from_args(args))
GenericFunction.__init__(self, args=args, **kwargs)
kwargs['_parsed_args'] = args
super(ReturnTypeFromArgs, self).__init__(*args, **kwargs)
class coalesce(ReturnTypeFromArgs):
pass
class max(ReturnTypeFromArgs):
pass
class min(ReturnTypeFromArgs):
pass
class sum(ReturnTypeFromArgs):
pass
class now(GenericFunction):
__return_type__ = sqltypes.DateTime
type = sqltypes.DateTime
class concat(GenericFunction):
__return_type__ = sqltypes.String
def __init__(self, *args, **kwargs):
GenericFunction.__init__(self, args=args, **kwargs)
type = sqltypes.String
class char_length(GenericFunction):
__return_type__ = sqltypes.Integer
type = sqltypes.Integer
def __init__(self, arg, **kwargs):
GenericFunction.__init__(self, args=[arg], **kwargs)
GenericFunction.__init__(self, arg, **kwargs)
class random(GenericFunction):
def __init__(self, *args, **kwargs):
kwargs.setdefault('type_', None)
GenericFunction.__init__(self, args=args, **kwargs)
pass
class count(GenericFunction):
"""The ANSI COUNT aggregate function. With no arguments, emits COUNT \*."""
r"""The ANSI COUNT aggregate function. With no arguments,
emits COUNT \*.
__return_type__ = sqltypes.Integer
"""
type = sqltypes.Integer
def __init__(self, expression=None, **kwargs):
if expression is None:
expression = text('*')
GenericFunction.__init__(self, args=(expression,), **kwargs)
expression = literal_column('*')
super(count, self).__init__(expression, **kwargs)
class current_date(AnsiFunction):
__return_type__ = sqltypes.Date
type = sqltypes.Date
class current_time(AnsiFunction):
__return_type__ = sqltypes.Time
type = sqltypes.Time
class current_timestamp(AnsiFunction):
__return_type__ = sqltypes.DateTime
type = sqltypes.DateTime
class current_user(AnsiFunction):
__return_type__ = sqltypes.String
type = sqltypes.String
class localtime(AnsiFunction):
__return_type__ = sqltypes.DateTime
type = sqltypes.DateTime
class localtimestamp(AnsiFunction):
__return_type__ = sqltypes.DateTime
type = sqltypes.DateTime
class session_user(AnsiFunction):
__return_type__ = sqltypes.String
type = sqltypes.String
class sysdate(AnsiFunction):
__return_type__ = sqltypes.DateTime
type = sqltypes.DateTime
class user(AnsiFunction):
__return_type__ = sqltypes.String
type = sqltypes.String
class array_agg(GenericFunction):
"""support for the ARRAY_AGG function.
The ``func.array_agg(expr)`` construct returns an expression of
type :class:`.types.ARRAY`.
e.g.::
stmt = select([func.array_agg(table.c.values)[2:5]])
.. versionadded:: 1.1
.. seealso::
:func:`.postgresql.array_agg` - PostgreSQL-specific version that
returns :class:`.postgresql.ARRAY`, which has PG-specific operators added.
"""
type = sqltypes.ARRAY
def __init__(self, *args, **kwargs):
args = [_literal_as_binds(c) for c in args]
kwargs.setdefault('type_', self.type(_type_from_args(args)))
kwargs['_parsed_args'] = args
super(array_agg, self).__init__(*args, **kwargs)
class OrderedSetAgg(GenericFunction):
"""Define a function where the return type is based on the sort
expression type as defined by the expression passed to the
:meth:`.FunctionElement.within_group` method."""
array_for_multi_clause = False
def within_group_type(self, within_group):
func_clauses = self.clause_expr.element
order_by = sqlutil.unwrap_order_by(within_group.order_by)
if self.array_for_multi_clause and len(func_clauses.clauses) > 1:
return sqltypes.ARRAY(order_by[0].type)
else:
return order_by[0].type
class mode(OrderedSetAgg):
"""implement the ``mode`` ordered-set aggregate function.
This function must be used with the :meth:`.FunctionElement.within_group`
modifier to supply a sort expression to operate upon.
The return type of this function is the same as the sort expression.
.. versionadded:: 1.1
"""
class percentile_cont(OrderedSetAgg):
"""implement the ``percentile_cont`` ordered-set aggregate function.
This function must be used with the :meth:`.FunctionElement.within_group`
modifier to supply a sort expression to operate upon.
The return type of this function is the same as the sort expression,
or if the arguments are an array, an :class:`.types.ARRAY` of the sort
expression's type.
.. versionadded:: 1.1
"""
array_for_multi_clause = True
class percentile_disc(OrderedSetAgg):
"""implement the ``percentile_disc`` ordered-set aggregate function.
This function must be used with the :meth:`.FunctionElement.within_group`
modifier to supply a sort expression to operate upon.
The return type of this function is the same as the sort expression,
or if the arguments are an array, an :class:`.types.ARRAY` of the sort
expression's type.
.. versionadded:: 1.1
"""
array_for_multi_clause = True
class rank(GenericFunction):
"""Implement the ``rank`` hypothetical-set aggregate function.
This function must be used with the :meth:`.FunctionElement.within_group`
modifier to supply a sort expression to operate upon.
The return type of this function is :class:`.Integer`.
.. versionadded:: 1.1
"""
type = sqltypes.Integer()
class dense_rank(GenericFunction):
"""Implement the ``dense_rank`` hypothetical-set aggregate function.
This function must be used with the :meth:`.FunctionElement.within_group`
modifier to supply a sort expression to operate upon.
The return type of this function is :class:`.Integer`.
.. versionadded:: 1.1
"""
type = sqltypes.Integer()
class percent_rank(GenericFunction):
"""Implement the ``percent_rank`` hypothetical-set aggregate function.
This function must be used with the :meth:`.FunctionElement.within_group`
modifier to supply a sort expression to operate upon.
The return type of this function is :class:`.Numeric`.
.. versionadded:: 1.1
"""
type = sqltypes.Numeric()
class cume_dist(GenericFunction):
"""Implement the ``cume_dist`` hypothetical-set aggregate function.
This function must be used with the :meth:`.FunctionElement.within_group`
modifier to supply a sort expression to operate upon.
The return type of this function is :class:`.Numeric`.
.. versionadded:: 1.1
"""
type = sqltypes.Numeric()

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

View File

@@ -1,90 +1,137 @@
# sql/visitors.py
# Copyright (C) 2005-2017 the SQLAlchemy authors and contributors
# <see AUTHORS file>
#
# This module is part of SQLAlchemy and is released under
# the MIT License: http://www.opensource.org/licenses/mit-license.php
"""Visitor/traversal interface and library functions.
SQLAlchemy schema and expression constructs rely on a Python-centric
version of the classic "visitor" pattern as the primary way in which
they apply functionality. The most common use of this pattern
is statement compilation, where individual expression classes match
up to rendering methods that produce a string result. Beyond this,
the visitor system is also used to inspect expressions for various
information and patterns, as well as for usage in
they apply functionality. The most common use of this pattern
is statement compilation, where individual expression classes match
up to rendering methods that produce a string result. Beyond this,
the visitor system is also used to inspect expressions for various
information and patterns, as well as for usage in
some kinds of expression transformation. Other kinds of transformation
use a non-visitor traversal system.
For many examples of how the visit system is used, see the
For many examples of how the visit system is used, see the
sqlalchemy.sql.util and the sqlalchemy.sql.compiler modules.
For an introduction to clause adaption, see
http://techspot.zzzeek.org/?p=19 .
http://techspot.zzzeek.org/2008/01/23/expression-transformations/
"""
from collections import deque
import re
from sqlalchemy import util
from .. import util
import operator
from .. import exc
__all__ = ['VisitableType', 'Visitable', 'ClauseVisitor',
'CloningVisitor', 'ReplacingCloningVisitor', 'iterate',
'iterate_depthfirst', 'traverse_using', 'traverse',
'traverse_depthfirst',
'cloned_traverse', 'replacement_traverse']
__all__ = ['VisitableType', 'Visitable', 'ClauseVisitor',
'CloningVisitor', 'ReplacingCloningVisitor', 'iterate',
'iterate_depthfirst', 'traverse_using', 'traverse',
'cloned_traverse', 'replacement_traverse']
class VisitableType(type):
"""Metaclass which checks for a `__visit_name__` attribute and
applies `_compiler_dispatch` method to classes.
"""Metaclass which assigns a `_compiler_dispatch` method to classes
having a `__visit_name__` attribute.
The _compiler_dispatch attribute becomes an instance method which
looks approximately like the following::
def _compiler_dispatch (self, visitor, **kw):
'''Look for an attribute named "visit_" + self.__visit_name__
on the visitor, and call it with the same kw params.'''
visit_attr = 'visit_%s' % self.__visit_name__
return getattr(visitor, visit_attr)(self, **kw)
Classes having no __visit_name__ attribute will remain unaffected.
"""
def __init__(cls, clsname, bases, clsdict):
if cls.__name__ == 'Visitable' or not hasattr(cls, '__visit_name__'):
super(VisitableType, cls).__init__(clsname, bases, clsdict)
return
# set up an optimized visit dispatch function
# for use by the compiler
visit_name = cls.__visit_name__
if isinstance(visit_name, str):
getter = operator.attrgetter("visit_%s" % visit_name)
def _compiler_dispatch(self, visitor, **kw):
return getter(visitor)(self, **kw)
else:
def _compiler_dispatch(self, visitor, **kw):
return getattr(visitor, 'visit_%s' % self.__visit_name__)(self, **kw)
cls._compiler_dispatch = _compiler_dispatch
if clsname != 'Visitable' and \
hasattr(cls, '__visit_name__'):
_generate_dispatch(cls)
super(VisitableType, cls).__init__(clsname, bases, clsdict)
class Visitable(object):
def _generate_dispatch(cls):
"""Return an optimized visit dispatch function for the cls
for use by the compiler.
"""
if '__visit_name__' in cls.__dict__:
visit_name = cls.__visit_name__
if isinstance(visit_name, str):
# There is an optimization opportunity here because the
# the string name of the class's __visit_name__ is known at
# this early stage (import time) so it can be pre-constructed.
getter = operator.attrgetter("visit_%s" % visit_name)
def _compiler_dispatch(self, visitor, **kw):
try:
meth = getter(visitor)
except AttributeError:
raise exc.UnsupportedCompilationError(visitor, cls)
else:
return meth(self, **kw)
else:
# The optimization opportunity is lost for this case because the
# __visit_name__ is not yet a string. As a result, the visit
# string has to be recalculated with each compilation.
def _compiler_dispatch(self, visitor, **kw):
visit_attr = 'visit_%s' % self.__visit_name__
try:
meth = getattr(visitor, visit_attr)
except AttributeError:
raise exc.UnsupportedCompilationError(visitor, cls)
else:
return meth(self, **kw)
_compiler_dispatch.__doc__ = \
"""Look for an attribute named "visit_" + self.__visit_name__
on the visitor, and call it with the same kw params.
"""
cls._compiler_dispatch = _compiler_dispatch
class Visitable(util.with_metaclass(VisitableType, object)):
"""Base class for visitable objects, applies the
``VisitableType`` metaclass.
"""
__metaclass__ = VisitableType
class ClauseVisitor(object):
"""Base class for visitor objects which can traverse using
"""Base class for visitor objects which can traverse using
the traverse() function.
"""
__traverse_options__ = {}
def traverse_single(self, obj):
def traverse_single(self, obj, **kw):
for v in self._visitor_iterator:
meth = getattr(v, "visit_%s" % obj.__visit_name__, None)
if meth:
return meth(obj)
def iterate(self, obj):
"""traverse the given expression structure, returning an iterator of all elements."""
return meth(obj, **kw)
def iterate(self, obj):
"""traverse the given expression structure, returning an iterator
of all elements.
"""
return iterate(obj, self.__traverse_options__)
def traverse(self, obj):
"""traverse and visit the given expression structure."""
return traverse(obj, self.__traverse_options__, self._visitor_dict)
@util.memoized_property
def _visitor_dict(self):
visitors = {}
@@ -93,11 +140,11 @@ class ClauseVisitor(object):
if name.startswith('visit_'):
visitors[name[6:]] = getattr(self, name)
return visitors
@property
def _visitor_iterator(self):
"""iterate through this visitor and each 'chained' visitor."""
v = self
while v:
yield v
@@ -105,41 +152,46 @@ class ClauseVisitor(object):
def chain(self, visitor):
"""'chain' an additional ClauseVisitor onto this ClauseVisitor.
the chained visitor will receive all visit events after this one.
"""
tail = list(self._visitor_iterator)[-1]
tail._next = visitor
return self
class CloningVisitor(ClauseVisitor):
"""Base class for visitor objects which can traverse using
"""Base class for visitor objects which can traverse using
the cloned_traverse() function.
"""
def copy_and_process(self, list_):
"""Apply cloned traversal to the given list of elements, and return the new list."""
"""Apply cloned traversal to the given list of elements, and return
the new list.
"""
return [self.traverse(x) for x in list_]
def traverse(self, obj):
"""traverse and visit the given expression structure."""
return cloned_traverse(obj, self.__traverse_options__, self._visitor_dict)
return cloned_traverse(
obj, self.__traverse_options__, self._visitor_dict)
class ReplacingCloningVisitor(CloningVisitor):
"""Base class for visitor objects which can traverse using
"""Base class for visitor objects which can traverse using
the replacement_traverse() function.
"""
def replace(self, elem):
"""receive pre-copied elements during a cloning traversal.
If the method returns a new element, the element is used
instead of creating a simple copy of the element. Traversal
If the method returns a new element, the element is used
instead of creating a simple copy of the element. Traversal
will halt on the newly returned element if it is re-encountered.
"""
return None
@@ -154,25 +206,39 @@ class ReplacingCloningVisitor(CloningVisitor):
return e
return replacement_traverse(obj, self.__traverse_options__, replace)
def iterate(obj, opts):
"""traverse the given expression structure, returning an iterator.
traversal is configured to be breadth-first.
"""
# fasttrack for atomic elements like columns
children = obj.get_children(**opts)
if not children:
return [obj]
traversal = deque()
stack = deque([obj])
while stack:
t = stack.popleft()
yield t
traversal.append(t)
for c in t.get_children(**opts):
stack.append(c)
return iter(traversal)
def iterate_depthfirst(obj, opts):
"""traverse the given expression structure, returning an iterator.
traversal is configured to be depth-first.
"""
# fasttrack for atomic elements like columns
children = obj.get_children(**opts)
if not children:
return [obj]
stack = deque([obj])
traversal = deque()
while stack:
@@ -182,75 +248,81 @@ def iterate_depthfirst(obj, opts):
stack.append(c)
return iter(traversal)
def traverse_using(iterator, obj, visitors):
"""visit the given expression structure using the given iterator of objects."""
def traverse_using(iterator, obj, visitors):
"""visit the given expression structure using the given iterator of
objects.
"""
for target in iterator:
meth = visitors.get(target.__visit_name__, None)
if meth:
meth(target)
return obj
def traverse(obj, opts, visitors):
"""traverse and visit the given expression structure using the default iterator."""
def traverse(obj, opts, visitors):
"""traverse and visit the given expression structure using the default
iterator.
"""
return traverse_using(iterate(obj, opts), obj, visitors)
def traverse_depthfirst(obj, opts, visitors):
"""traverse and visit the given expression structure using the depth-first iterator."""
def traverse_depthfirst(obj, opts, visitors):
"""traverse and visit the given expression structure using the
depth-first iterator.
"""
return traverse_using(iterate_depthfirst(obj, opts), obj, visitors)
def cloned_traverse(obj, opts, visitors):
"""clone the given expression structure, allowing modifications by visitors."""
cloned = util.column_dict()
"""clone the given expression structure, allowing
modifications by visitors."""
def clone(element):
if element not in cloned:
cloned[element] = element._clone()
return cloned[element]
cloned = {}
stop_on = set(opts.get('stop_on', []))
obj = clone(obj)
stack = [obj]
def clone(elem):
if elem in stop_on:
return elem
else:
if id(elem) not in cloned:
cloned[id(elem)] = newelem = elem._clone()
newelem._copy_internals(clone=clone)
meth = visitors.get(newelem.__visit_name__, None)
if meth:
meth(newelem)
return cloned[id(elem)]
while stack:
t = stack.pop()
if t in cloned:
continue
t._copy_internals(clone=clone)
meth = visitors.get(t.__visit_name__, None)
if meth:
meth(t)
for c in t.get_children(**opts):
stack.append(c)
if obj is not None:
obj = clone(obj)
return obj
def replacement_traverse(obj, opts, replace):
"""clone the given expression structure, allowing element replacement by a given replacement function."""
cloned = util.column_dict()
stop_on = util.column_set(opts.get('stop_on', []))
"""clone the given expression structure, allowing element
replacement by a given replacement function."""
def clone(element):
newelem = replace(element)
if newelem is not None:
stop_on.add(newelem)
return newelem
cloned = {}
stop_on = set([id(x) for x in opts.get('stop_on', [])])
if element not in cloned:
cloned[element] = element._clone()
return cloned[element]
def clone(elem, **kw):
if id(elem) in stop_on or \
'no_replacement_traverse' in elem._annotations:
return elem
else:
newelem = replace(elem)
if newelem is not None:
stop_on.add(id(newelem))
return newelem
else:
if elem not in cloned:
cloned[elem] = newelem = elem._clone()
newelem._copy_internals(clone=clone, **kw)
return cloned[elem]
obj = clone(obj)
stack = [obj]
while stack:
t = stack.pop()
if t in stop_on:
continue
t._copy_internals(clone=clone)
for c in t.get_children(**opts):
stack.append(c)
if obj is not None:
obj = clone(obj, **opts)
return obj