base.py

# dialects/postgresql/base.py
# Copyright (C) 2005-2025 the SQLAlchemy authors and contributors
# <see AUTHORS file>
#
# This module is part of SQLAlchemy and is released under
# the MIT License: https://www.opensource.org/licenses/mit-license.php
# mypy: ignore-errors

r"""
.. dialect:: postgresql
    :name: PostgreSQL
    :normal_support: 9.6+
    :best_effort: 9+

.. _postgresql_sequences:

Sequences/SERIAL/IDENTITY
-------------------------

PostgreSQL supports sequences, and SQLAlchemy uses these as the default means
of creating new primary key values for integer-based primary key columns. When
creating tables, SQLAlchemy will issue the ``SERIAL`` datatype for
integer-based primary key columns, which generates a sequence and server side
default corresponding to the column.

To specify a specific named sequence to be used for primary key generation,
use the :func:`~sqlalchemy.schema.Sequence` construct::

    Table(
        "sometable",
        metadata,
        Column(
            "id", Integer, Sequence("some_id_seq", start=1), primary_key=True
        ),
    )

When SQLAlchemy issues a single INSERT statement, to fulfill the contract of
having the "last insert identifier" available, a RETURNING clause is added to
the INSERT statement which specifies the primary key columns should be
returned after the statement completes. The RETURNING functionality only takes
place if PostgreSQL 8.2 or later is in use. As a fallback approach, the
sequence, whether specified explicitly or implicitly via ``SERIAL``, is
executed independently beforehand, the returned value to be used in the
subsequent insert. Note that when an
:func:`~sqlalchemy.sql.expression.insert()` construct is executed using
"executemany" semantics, the "last inserted identifier" functionality does not
apply; no RETURNING clause is emitted nor is the sequence pre-executed in this
case.


PostgreSQL 10 and above IDENTITY columns
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

PostgreSQL 10 and above have a new IDENTITY feature that supersedes the use
of SERIAL. The :class:`_schema.Identity` construct in a
:class:`_schema.Column` can be used to control its behavior::

    from sqlalchemy import Table, Column, MetaData, Integer, Computed

    metadata = MetaData()

    data = Table(
        "data",
        metadata,
        Column(
            "id", Integer, Identity(start=42, cycle=True), primary_key=True
        ),
        Column("data", String),
    )

The CREATE TABLE for the above :class:`_schema.Table` object would be:

.. sourcecode:: sql

    CREATE TABLE data (
        id INTEGER GENERATED BY DEFAULT AS IDENTITY (START WITH 42 CYCLE),
        data VARCHAR,
        PRIMARY KEY (id)
    )

.. versionchanged::  1.4   Added :class:`_schema.Identity` construct
   in a :class:`_schema.Column` to specify the option of an autoincrementing
   column.

.. note::

   Previous versions of SQLAlchemy did not have built-in support for rendering
   of IDENTITY, and could use the following compilation hook to replace
   occurrences of SERIAL with IDENTITY::

       from sqlalchemy.schema import CreateColumn
       from sqlalchemy.ext.compiler import compiles


       @compiles(CreateColumn, "postgresql")
       def use_identity(element, compiler, **kw):
           text = compiler.visit_create_column(element, **kw)
           text = text.replace("SERIAL", "INT GENERATED BY DEFAULT AS IDENTITY")
           return text

   Using the above, a table such as::

       t = Table(
           "t", m, Column("id", Integer, primary_key=True), Column("data", String)
       )

   Will generate on the backing database as:

   .. sourcecode:: sql

       CREATE TABLE t (
           id INT GENERATED BY DEFAULT AS IDENTITY,
           data VARCHAR,
           PRIMARY KEY (id)
       )

.. _postgresql_ss_cursors:

Server Side Cursors
-------------------

Server-side cursor support is available for the psycopg2, asyncpg
dialects and may also be available in others.

Server side cursors are enabled on a per-statement basis by using the
:paramref:`.Connection.execution_options.stream_results` connection execution
option::

    with engine.connect() as conn:
        result = conn.execution_options(stream_results=True).execute(
            text("select * from table")
        )

Note that some kinds of SQL statements may not be supported with
server side cursors; generally, only SQL statements that return rows should be
used with this option.

.. deprecated:: 1.4  The dialect-level server_side_cursors flag is deprecated
   and will be removed in a future release.  Please use the
   :paramref:`_engine.Connection.stream_results` execution option for
   unbuffered cursor support.

.. seealso::

    :ref:`engine_stream_results`

.. _postgresql_isolation_level:

Transaction Isolation Level
---------------------------

Most SQLAlchemy dialects support setting of transaction isolation level
using the :paramref:`_sa.create_engine.isolation_level` parameter
at the :func:`_sa.create_engine` level, and at the :class:`_engine.Connection`
level via the :paramref:`.Connection.execution_options.isolation_level`
parameter.

For PostgreSQL dialects, this feature works either by making use of the
DBAPI-specific features, such as psycopg2's isolation level flags which will
embed the isolation level setting inline with the ``"BEGIN"`` statement, or for
DBAPIs with no direct support by emitting ``SET SESSION CHARACTERISTICS AS
TRANSACTION ISOLATION LEVEL <level>`` ahead of the ``"BEGIN"`` statement
emitted by the DBAPI.   For the special AUTOCOMMIT isolation level,
DBAPI-specific techniques are used which is typically an ``.autocommit``
flag on the DBAPI connection object.

To set isolation level using :func:`_sa.create_engine`::

    engine = create_engine(
        "postgresql+pg8000://scott:tiger@localhost/test",
        isolation_level="REPEATABLE READ",
    )

To set using per-connection execution options::

    with engine.connect() as conn:
        conn = conn.execution_options(isolation_level="REPEATABLE READ")
        with conn.begin():
            ...  # work with transaction

There are also more options for isolation level configurations, such as
"sub-engine" objects linked to a main :class:`_engine.Engine` which each apply
different isolation level settings.  See the discussion at
:ref:`dbapi_autocommit` for background.

Valid values for ``isolation_level`` on most PostgreSQL dialects include:

* ``READ COMMITTED``
* ``READ UNCOMMITTED``
* ``REPEATABLE READ``
* ``SERIALIZABLE``
* ``AUTOCOMMIT``

.. seealso::

    :ref:`dbapi_autocommit`

    :ref:`postgresql_readonly_deferrable`

    :ref:`psycopg2_isolation_level`

    :ref:`pg8000_isolation_level`

.. _postgresql_readonly_deferrable:

Setting READ ONLY / DEFERRABLE
------------------------------

Most PostgreSQL dialects support setting the "READ ONLY" and "DEFERRABLE"
characteristics of the transaction, which is in addition to the isolation level
setting. These two attributes can be established either in conjunction with or
independently of the isolation level by passing the ``postgresql_readonly`` and
``postgresql_deferrable`` flags with
:meth:`_engine.Connection.execution_options`.  The example below illustrates
passing the ``"SERIALIZABLE"`` isolation level at the same time as setting
"READ ONLY" and "DEFERRABLE"::

    with engine.connect() as conn:
        conn = conn.execution_options(
            isolation_level="SERIALIZABLE",
            postgresql_readonly=True,
            postgresql_deferrable=True,
        )
        with conn.begin():
            ...  # work with transaction

Note that some DBAPIs such as asyncpg only support "readonly" with
SERIALIZABLE isolation.

.. versionadded:: 1.4 added support for the ``postgresql_readonly``
   and ``postgresql_deferrable`` execution options.

.. _postgresql_reset_on_return:

Temporary Table / Resource Reset for Connection Pooling
-------------------------------------------------------

The :class:`.QueuePool` connection pool implementation used
by the SQLAlchemy :class:`.Engine` object includes
:ref:`reset on return <pool_reset_on_return>` behavior that will invoke
the DBAPI ``.rollback()`` method when connections are returned to the pool.
While this rollback will clear out the immediate state used by the previous
transaction, it does not cover a wider range of session-level state, including
temporary tables as well as other server state such as prepared statement
handles and statement caches.   The PostgreSQL database includes a variety
of commands which may be used to reset this state, including
``DISCARD``, ``RESET``, ``DEALLOCATE``, and ``UNLISTEN``.


To install
one or more of these commands as the means of performing reset-on-return,
the :meth:`.PoolEvents.reset` event hook may be used, as demonstrated
in the example below. The implementation
will end transactions in progress as well as discard temporary tables
using the ``CLOSE``, ``RESET`` and ``DISCARD`` commands; see the PostgreSQL
documentation for background on what each of these statements do.

The :paramref:`_sa.create_engine.pool_reset_on_return` parameter
is set to ``None`` so that the custom scheme can replace the default behavior
completely.   The custom hook implementation calls ``.rollback()`` in any case,
as it's usually important that the DBAPI's own tracking of commit/rollback
will remain consistent with the state of the transaction::


    from sqlalchemy import create_engine
    from sqlalchemy import event

    postgresql_engine = create_engine(
        "postgresql+psycopg2://scott:tiger@hostname/dbname",
        # disable default reset-on-return scheme
        pool_reset_on_return=None,
    )


    @event.listens_for(postgresql_engine, "reset")
    def _reset_postgresql(dbapi_connection, connection_record, reset_state):
        if not reset_state.terminate_only:
            dbapi_connection.execute("CLOSE ALL")
            dbapi_connection.execute("RESET ALL")
            dbapi_connection.execute("DISCARD TEMP")

        # so that the DBAPI itself knows that the connection has been
        # reset
        dbapi_connection.rollback()

.. versionchanged:: 2.0.0b3  Added additional state arguments to
   the :meth:`.PoolEvents.reset` event and additionally ensured the event
   is invoked for all "reset" occurrences, so that it's appropriate
   as a place for custom "reset" handlers.   Previous schemes which
   use the :meth:`.PoolEvents.checkin` handler remain usable as well.

.. seealso::

    :ref:`pool_reset_on_return` - in the :ref:`pooling_toplevel` documentation

.. _postgresql_alternate_search_path:

Setting Alternate Search Paths on Connect
------------------------------------------

The PostgreSQL ``search_path`` variable refers to the list of schema names
that will be implicitly referenced when a particular table or other
object is referenced in a SQL statement.  As detailed in the next section
:ref:`postgresql_schema_reflection`, SQLAlchemy is generally organized around
the concept of keeping this variable at its default value of ``public``,
however, in order to have it set to any arbitrary name or names when connections
are used automatically, the "SET SESSION search_path" command may be invoked
for all connections in a pool using the following event handler, as discussed
at :ref:`schema_set_default_connections`::

    from sqlalchemy import event
    from sqlalchemy import create_engine

    engine = create_engine("postgresql+psycopg2://scott:tiger@host/dbname")


    @event.listens_for(engine, "connect", insert=True)
    def set_search_path(dbapi_connection, connection_record):
        existing_autocommit = dbapi_connection.autocommit
        dbapi_connection.autocommit = True
        cursor = dbapi_connection.cursor()
        cursor.execute("SET SESSION search_path='%s'" % schema_name)
        cursor.close()
        dbapi_connection.autocommit = existing_autocommit

The reason the recipe is complicated by use of the ``.autocommit`` DBAPI
attribute is so that when the ``SET SESSION search_path`` directive is invoked,
it is invoked outside of the scope of any transaction and therefore will not
be reverted when the DBAPI connection has a rollback.

.. seealso::

  :ref:`schema_set_default_connections` - in the :ref:`metadata_toplevel` documentation

.. _postgresql_schema_reflection:

Remote-Schema Table Introspection and PostgreSQL search_path
------------------------------------------------------------

.. admonition:: Section Best Practices Summarized

    keep the ``search_path`` variable set to its default of ``public``, without
    any other schema names. Ensure the username used to connect **does not**
    match remote schemas, or ensure the ``"$user"`` token is **removed** from
    ``search_path``.  For other schema names, name these explicitly
    within :class:`_schema.Table` definitions. Alternatively, the
    ``postgresql_ignore_search_path`` option will cause all reflected
    :class:`_schema.Table` objects to have a :attr:`_schema.Table.schema`
    attribute set up.

The PostgreSQL dialect can reflect tables from any schema, as outlined in
:ref:`metadata_reflection_schemas`.

In all cases, the first thing SQLAlchemy does when reflecting tables is
to **determine the default schema for the current database connection**.
It does this using the PostgreSQL ``current_schema()``
function, illustated below using a PostgreSQL client session (i.e. using
the ``psql`` tool):

.. sourcecode:: sql

    test=> select current_schema();
    current_schema
    ----------------
    public
    (1 row)

Above we see that on a plain install of PostgreSQL, the default schema name
is the name ``public``.

However, if your database username **matches the name of a schema**, PostgreSQL's
default is to then **use that name as the default schema**.  Below, we log in
using the username ``scott``.  When we create a schema named ``scott``, **it
implicitly changes the default schema**:

.. sourcecode:: sql

    test=> select current_schema();
    current_schema
    ----------------
    public
    (1 row)

    test=> create schema scott;
    CREATE SCHEMA
    test=> select current_schema();
    current_schema
    ----------------
    scott
    (1 row)

The behavior of ``current_schema()`` is derived from the
`PostgreSQL search path
<https://www.postgresql.org/docs/current/static/ddl-schemas.html#DDL-SCHEMAS-PATH>`_
variable ``search_path``, which in modern PostgreSQL versions defaults to this:

.. sourcecode:: sql

    test=> show search_path;
    search_path
    -----------------
    "$user", public
    (1 row)

Where above, the ``"$user"`` variable will inject the current username as the
default schema, if one exists.   Otherwise, ``public`` is used.

When a :class:`_schema.Table` object is reflected, if it is present in the
schema indicated by the ``current_schema()`` function, **the schema name assigned
to the ".schema" attribute of the Table is the Python "None" value**.  Otherwise, the
".schema" attribute will be assigned the string name of that schema.

With regards to tables which these :class:`_schema.Table`
objects refer to via foreign key constraint, a decision must be made as to how
the ``.schema`` is represented in those remote tables, in the case where that
remote schema name is also a member of the current ``search_path``.

By default, the PostgreSQL dialect mimics the behavior encouraged by
PostgreSQL's own ``pg_get_constraintdef()`` builtin procedure.  This function
returns a sample definition for a particular foreign key constraint,
omitting the referenced schema name from that definition when the name is
also in the PostgreSQL schema search path.  The interaction below
illustrates this behavior:

.. sourcecode:: sql

    test=> CREATE TABLE test_schema.referred(id INTEGER PRIMARY KEY);
    CREATE TABLE
    test=> CREATE TABLE referring(
    test(>         id INTEGER PRIMARY KEY,
    test(>         referred_id INTEGER REFERENCES test_schema.referred(id));
    CREATE TABLE
    test=> SET search_path TO public, test_schema;
    test=> SELECT pg_catalog.pg_get_constraintdef(r.oid, true) FROM
    test-> pg_catalog.pg_class c JOIN pg_catalog.pg_namespace n
    test-> ON n.oid = c.relnamespace
    test-> JOIN pg_catalog.pg_constraint r  ON c.oid = r.conrelid
    test-> WHERE c.relname='referring' AND r.contype = 'f'
    test-> ;
                   pg_get_constraintdef
    ---------------------------------------------------
     FOREIGN KEY (referred_id) REFERENCES referred(id)
    (1 row)

Above, we created a table ``referred`` as a member of the remote schema
``test_schema``, however when we added ``test_schema`` to the
PG ``search_path`` and then asked ``pg_get_constraintdef()`` for the
``FOREIGN KEY`` syntax, ``test_schema`` was not included in the output of
the function.

On the other hand, if we set the search path back to the typical default
of ``public``:

.. sourcecode:: sql

    test=> SET search_path TO public;
    SET

The same query against ``pg_get_constraintdef()`` now returns the fully
schema-qualified name for us:

.. sourcecode:: sql

    test=> SELECT pg_catalog.pg_get_constraintdef(r.oid, true) FROM
    test-> pg_catalog.pg_class c JOIN pg_catalog.pg_namespace n
    test-> ON n.oid = c.relnamespace
    test-> JOIN pg_catalog.pg_constraint r  ON c.oid = r.conrelid
    test-> WHERE c.relname='referring' AND r.contype = 'f';
                         pg_get_constraintdef
    ---------------------------------------------------------------
     FOREIGN KEY (referred_id) REFERENCES test_schema.referred(id)
    (1 row)

SQLAlchemy will by default use the return value of ``pg_get_constraintdef()``
in order to determine the remote schema name.  That is, if our ``search_path``
were set to include ``test_schema``, and we invoked a table
reflection process as follows::

    >>> from sqlalchemy import Table, MetaData, create_engine, text
    >>> engine = create_engine("postgresql+psycopg2://scott:tiger@localhost/test")
    >>> with engine.connect() as conn:
    ...     conn.execute(text("SET search_path TO test_schema, public"))
    ...     metadata_obj = MetaData()
    ...     referring = Table("referring", metadata_obj, autoload_with=conn)
    <sqlalchemy.engine.result.CursorResult object at 0x101612ed0>

The above process would deliver to the :attr:`_schema.MetaData.tables`
collection
``referred`` table named **without** the schema::

    >>> metadata_obj.tables["referred"].schema is None
    True

To alter the behavior of reflection such that the referred schema is
maintained regardless of the ``search_path`` setting, use the
``postgresql_ignore_search_path`` option, which can be specified as a
dialect-specific argument to both :class:`_schema.Table` as well as
:meth:`_schema.MetaData.reflect`::

    >>> with engine.connect() as conn:
    ...     conn.execute(text("SET search_path TO test_schema, public"))
    ...     metadata_obj = MetaData()
    ...     referring = Table(
    ...         "referring",
    ...         metadata_obj,
    ...         autoload_with=conn,
    ...         postgresql_ignore_search_path=True,
    ...     )
    <sqlalchemy.engine.result.CursorResult object at 0x1016126d0>

We will now have ``test_schema.referred`` stored as schema-qualified::

    >>> metadata_obj.tables["test_schema.referred"].schema
    'test_schema'

.. sidebar:: Best Practices for PostgreSQL Schema reflection

    The description of PostgreSQL schema reflection behavior is complex, and
    is the product of many years of dealing with widely varied use cases and
    user preferences. But in fact, there's no need to understand any of it if
    you just stick to the simplest use pattern: leave the ``search_path`` set
    to its default of ``public`` only, never refer to the name ``public`` as
    an explicit schema name otherwise, and refer to all other schema names
    explicitly when building up a :class:`_schema.Table` object.  The options
    described here are only for those users who can't, or prefer not to, stay
    within these guidelines.

.. seealso::

    :ref:`reflection_schema_qualified_interaction` - discussion of the issue
    from a backend-agnostic perspective

    `The Schema Search Path
    <https://www.postgresql.org/docs/current/static/ddl-schemas.html#DDL-SCHEMAS-PATH>`_
    - on the PostgreSQL website.

INSERT/UPDATE...RETURNING
-------------------------

The dialect supports PG 8.2's ``INSERT..RETURNING``, ``UPDATE..RETURNING`` and
``DELETE..RETURNING`` syntaxes.   ``INSERT..RETURNING`` is used by default
for single-row INSERT statements in order to fetch newly generated
primary key identifiers.   To specify an explicit ``RETURNING`` clause,
use the :meth:`._UpdateBase.returning` method on a per-statement basis::

    # INSERT..RETURNING
    result = (
        table.insert().returning(table.c.col1, table.c.col2).values(name="foo")
    )
    print(result.fetchall())

    # UPDATE..RETURNING
    result = (
        table.update()
        .returning(table.c.col1, table.c.col2)
        .where(table.c.name == "foo")
        .values(name="bar")
    )
    print(result.fetchall())

    # DELETE..RETURNING
    result = (
        table.delete()
        .returning(table.c.col1, table.c.col2)
        .where(table.c.name == "foo")
    )
    print(result.fetchall())

.. _postgresql_insert_on_conflict:

INSERT...ON CONFLICT (Upsert)
------------------------------

Starting with version 9.5, PostgreSQL allows "upserts" (update or insert) of
rows into a table via the ``ON CONFLICT`` clause of the ``INSERT`` statement. A
candidate row will only be inserted if that row does not violate any unique
constraints.  In the case of a unique constraint violation, a secondary action
can occur which can be either "DO UPDATE", indicating that the data in the
target row should be updated, or "DO NOTHING", which indicates to silently skip
this row.

Conflicts are determined using existing unique constraints and indexes.  These
constraints may be identified either using their name as stated in DDL,
or they may be inferred by stating the columns and conditions that comprise
the indexes.

SQLAlchemy provides ``ON CONFLICT`` support via the PostgreSQL-specific
:func:`_postgresql.insert()` function, which provides
the generative methods :meth:`_postgresql.Insert.on_conflict_do_update`
and :meth:`~.postgresql.Insert.on_conflict_do_nothing`:

.. sourcecode:: pycon+sql

    >>> from sqlalchemy.dialects.postgresql import insert
    >>> insert_stmt = insert(my_table).values(
    ...     id="some_existing_id", data="inserted value"
    ... )
    >>> do_nothing_stmt = insert_stmt.on_conflict_do_nothing(index_elements=["id"])
    >>> print(do_nothing_stmt)
    {printsql}INSERT INTO my_table (id, data) VALUES (%(id)s, %(data)s)
    ON CONFLICT (id) DO NOTHING
    {stop}

    >>> do_update_stmt = insert_stmt.on_conflict_do_update(
    ...     constraint="pk_my_table", set_=dict(data="updated value")
    ... )
    >>> print(do_update_stmt)
    {printsql}INSERT INTO my_table (id, data) VALUES (%(id)s, %(data)s)
    ON CONFLICT ON CONSTRAINT pk_my_table DO UPDATE SET data = %(param_1)s

.. seealso::

    `INSERT .. ON CONFLICT
    <https://www.postgresql.org/docs/current/static/sql-insert.html#SQL-ON-CONFLICT>`_
    - in the PostgreSQL documentation.

Specifying the Target
^^^^^^^^^^^^^^^^^^^^^

Both methods supply the "target" of the conflict using either the
named constraint or by column inference:

* The :paramref:`_postgresql.Insert.on_conflict_do_update.index_elements` argument
  specifies a sequence containing string column names, :class:`_schema.Column`
  objects, and/or SQL expression elements, which would identify a unique
  index:

  .. sourcecode:: pycon+sql

    >>> do_update_stmt = insert_stmt.on_conflict_do_update(
    ...     index_elements=["id"], set_=dict(data="updated value")
    ... )
    >>> print(do_update_stmt)
    {printsql}INSERT INTO my_table (id, data) VALUES (%(id)s, %(data)s)
    ON CONFLICT (id) DO UPDATE SET data = %(param_1)s
    {stop}

    >>> do_update_stmt = insert_stmt.on_conflict_do_update(
    ...     index_elements=[my_table.c.id], set_=dict(data="updated value")
    ... )
    >>> print(do_update_stmt)
    {printsql}INSERT INTO my_table (id, data) VALUES (%(id)s, %(data)s)
    ON CONFLICT (id) DO UPDATE SET data = %(param_1)s

* When using :paramref:`_postgresql.Insert.on_conflict_do_update.index_elements` to
  infer an index, a partial index can be inferred by also specifying the
  use the :paramref:`_postgresql.Insert.on_conflict_do_update.index_where` parameter:

  .. sourcecode:: pycon+sql

    >>> stmt = insert(my_table).values(user_email="a@b.com", data="inserted data")
    >>> stmt = stmt.on_conflict_do_update(
    ...     index_elements=[my_table.c.user_email],
    ...     index_where=my_table.c.user_email.like("%@gmail.com"),
    ...     set_=dict(data=stmt.excluded.data),
    ... )
    >>> print(stmt)
    {printsql}INSERT INTO my_table (data, user_email)
    VALUES (%(data)s, %(user_email)s) ON CONFLICT (user_email)
    WHERE user_email LIKE %(user_email_1)s DO UPDATE SET data = excluded.data

* The :paramref:`_postgresql.Insert.on_conflict_do_update.constraint` argument is
  used to specify an index directly rather than inferring it.  This can be
  the name of a UNIQUE constraint, a PRIMARY KEY constraint, or an INDEX:

  .. sourcecode:: pycon+sql

    >>> do_update_stmt = insert_stmt.on_conflict_do_update(
    ...     constraint="my_table_idx_1", set_=dict(data="updated value")
    ... )
    >>> print(do_update_stmt)
    {printsql}INSERT INTO my_table (id, data) VALUES (%(id)s, %(data)s)
    ON CONFLICT ON CONSTRAINT my_table_idx_1 DO UPDATE SET data = %(param_1)s
    {stop}

    >>> do_update_stmt = insert_stmt.on_conflict_do_update(
    ...     constraint="my_table_pk", set_=dict(data="updated value")
    ... )
    >>> print(do_update_stmt)
    {printsql}INSERT INTO my_table (id, data) VALUES (%(id)s, %(data)s)
    ON CONFLICT ON CONSTRAINT my_table_pk DO UPDATE SET data = %(param_1)s
    {stop}

* The :paramref:`_postgresql.Insert.on_conflict_do_update.constraint` argument may
  also refer to a SQLAlchemy construct representing a constraint,
  e.g. :class:`.UniqueConstraint`, :class:`.PrimaryKeyConstraint`,
  :class:`.Index`, or :class:`.ExcludeConstraint`.   In this use,
  if the constraint has a name, it is used directly.  Otherwise, if the
  constraint is unnamed, then inference will be used, where the expressions
  and optional WHERE clause of the constraint will be spelled out in the
  construct.  This use is especially convenient
  to refer to the named or unnamed primary key of a :class:`_schema.Table`
  using the
  :attr:`_schema.Table.primary_key` attribute:

  .. sourcecode:: pycon+sql

    >>> do_update_stmt = insert_stmt.on_conflict_do_update(
    ...     constraint=my_table.primary_key, set_=dict(data="updated value")
    ... )
    >>> print(do_update_stmt)
    {printsql}INSERT INTO my_table (id, data) VALUES (%(id)s, %(data)s)
    ON CONFLICT (id) DO UPDATE SET data = %(param_1)s

The SET Clause
^^^^^^^^^^^^^^^

``ON CONFLICT...DO UPDATE`` is used to perform an update of the already
existing row, using any combination of new values as well as values
from the proposed insertion.   These values are specified using the
:paramref:`_postgresql.Insert.on_conflict_do_update.set_` parameter.  This
parameter accepts a dictionary which consists of direct values
for UPDATE:

.. sourcecode:: pycon+sql

    >>> stmt = insert(my_table).values(id="some_id", data="inserted value")
    >>> do_update_stmt = stmt.on_conflict_do_update(
    ...     index_elements=["id"], set_=dict(data="updated value")
    ... )
    >>> print(do_update_stmt)
    {printsql}INSERT INTO my_table (id, data) VALUES (%(id)s, %(data)s)
    ON CONFLICT (id) DO UPDATE SET data = %(param_1)s

.. warning::

    The :meth:`_expression.Insert.on_conflict_do_update`
    method does **not** take into
    account Python-side default UPDATE values or generation functions, e.g.
    those specified using :paramref:`_schema.Column.onupdate`.
    These values will not be exercised for an ON CONFLICT style of UPDATE,
    unless they are manually specified in the
    :paramref:`_postgresql.Insert.on_conflict_do_update.set_` dictionary.

Updating using the Excluded INSERT Values
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

In order to refer to the proposed insertion row, the special alias
:attr:`~.postgresql.Insert.excluded` is available as an attribute on
the :class:`_postgresql.Insert` object; this object is a
:class:`_expression.ColumnCollection`
which alias contains all columns of the target
table:

.. sourcecode:: pycon+sql

    >>> stmt = insert(my_table).values(
    ...     id="some_id", data="inserted value", author="jlh"
    ... )
    >>> do_update_stmt = stmt.on_conflict_do_update(
    ...     index_elements=["id"],
    ...     set_=dict(data="updated value", author=stmt.excluded.author),
    ... )
    >>> print(do_update_stmt)
    {printsql}INSERT INTO my_table (id, data, author)
    VALUES (%(id)s, %(data)s, %(author)s)
    ON CONFLICT (id) DO UPDATE SET data = %(param_1)s, author = excluded.author

Additional WHERE Criteria
^^^^^^^^^^^^^^^^^^^^^^^^^

The :meth:`_expression.Insert.on_conflict_do_update` method also accepts
a WHERE clause using the :paramref:`_postgresql.Insert.on_conflict_do_update.where`
parameter, which will limit those rows which receive an UPDATE:

.. sourcecode:: pycon+sql

    >>> stmt = insert(my_table).values(
    ...     id="some_id", data="inserted value", author="jlh"
    ... )
    >>> on_update_stmt = stmt.on_conflict_do_update(
    ...     index_elements=["id"],
    ...     set_=dict(data="updated value", author=stmt.excluded.author),
    ...     where=(my_table.c.status == 2),
    ... )
    >>> print(on_update_stmt)
    {printsql}INSERT INTO my_table (id, data, author)
    VALUES (%(id)s, %(data)s, %(author)s)
    ON CONFLICT (id) DO UPDATE SET data = %(param_1)s, author = excluded.author
    WHERE my_table.status = %(status_1)s

Skipping Rows with DO NOTHING
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

``ON CONFLICT`` may be used to skip inserting a row entirely
if any conflict with a unique or exclusion constraint occurs; below
this is illustrated using the
:meth:`~.postgresql.Insert.on_conflict_do_nothing` method:

.. sourcecode:: pycon+sql

    >>> stmt = insert(my_table).values(id="some_id", data="inserted value")
    >>> stmt = stmt.on_conflict_do_nothing(index_elements=["id"])
    >>> print(stmt)
    {printsql}INSERT INTO my_table (id, data) VALUES (%(id)s, %(data)s)
    ON CONFLICT (id) DO NOTHING

If ``DO NOTHING`` is used without specifying any columns or constraint,
it has the effect of skipping the INSERT for any unique or exclusion
constraint violation which occurs:

.. sourcecode:: pycon+sql

    >>> stmt = insert(my_table).values(id="some_id", data="inserted value")
    >>> stmt = stmt.on_conflict_do_nothing()
    >>> print(stmt)
    {printsql}INSERT INTO my_table (id, data) VALUES (%(id)s, %(data)s)
    ON CONFLICT DO NOTHING

.. _postgresql_match:

Full Text Search
----------------

PostgreSQL's full text search system is available through the use of the
:data:`.func` namespace, combined with the use of custom operators
via the :meth:`.Operators.bool_op` method.    For simple cases with some
degree of cross-backend compatibility, the :meth:`.Operators.match` operator
may also be used.

.. _postgresql_simple_match:

Simple plain text matching with ``match()``
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

The :meth:`.Operators.match` operator provides for cross-compatible simple
text matching.   For the PostgreSQL backend, it's hardcoded to generate
an expression using the ``@@`` operator in conjunction with the
``plainto_tsquery()`` PostgreSQL function.

On the PostgreSQL dialect, an expression like the following::

    select(sometable.c.text.match("search string"))

would emit to the database:

.. sourcecode:: sql

    SELECT text @@ plainto_tsquery('search string') FROM table

Above, passing a plain string to :meth:`.Operators.match` will automatically
make use of ``plainto_tsquery()`` to specify the type of tsquery.  This
establishes basic database cross-compatibility for :meth:`.Operators.match`
with other backends.

.. versionchanged:: 2.0 The default tsquery generation function used by the
   PostgreSQL dialect with :meth:`.Operators.match` is ``plainto_tsquery()``.

   To render exactly what was rendered in 1.4, use the following form::

        from sqlalchemy import func

        select(sometable.c.text.bool_op("@@")(func.to_tsquery("search string")))

   Which would emit:

   .. sourcecode:: sql

        SELECT text @@ to_tsquery('search string') FROM table

Using PostgreSQL full text functions and operators directly
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

Text search operations beyond the simple use of :meth:`.Operators.match`
may make use of the :data:`.func` namespace to generate PostgreSQL full-text
functions, in combination with :meth:`.Operators.bool_op` to generate
any boolean operator.

For example, the query::

    select(func.to_tsquery("cat").bool_op("@>")(func.to_tsquery("cat & rat")))

would generate:

.. sourcecode:: sql

    SELECT to_tsquery('cat') @> to_tsquery('cat & rat')


The :class:`_postgresql.TSVECTOR` type can provide for explicit CAST::

    from sqlalchemy.dialects.postgresql import TSVECTOR
    from sqlalchemy import select, cast

    select(cast("some text", TSVECTOR))

produces a statement equivalent to:

.. sourcecode:: sql

    SELECT CAST('some text' AS TSVECTOR) AS anon_1

The ``func`` namespace is augmented by the PostgreSQL dialect to set up
correct argument and return types for most full text search functions.
These functions are used automatically by the :attr:`_sql.func` namespace
assuming the ``sqlalchemy.dialects.postgresql`` package has been imported,
or :func:`_sa.create_engine` has been invoked using a ``postgresql``
dialect.  These functions are documented at:

* :class:`_postgresql.to_tsvector`
* :class:`_postgresql.to_tsquery`
* :class:`_postgresql.plainto_tsquery`
* :class:`_postgresql.phraseto_tsquery`
* :class:`_postgresql.websearch_to_tsquery`
* :class:`_postgresql.ts_headline`

Specifying the "regconfig" with ``match()`` or custom operators
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

PostgreSQL's ``plainto_tsquery()`` function accepts an optional
"regconfig" argument that is used to instruct PostgreSQL to use a
particular pre-computed GIN or GiST index in order to perform the search.
When using :meth:`.Operators.match`, this additional parameter may be
specified using the ``postgresql_regconfig`` parameter, such as::

    select(mytable.c.id).where(
        mytable.c.title.match("somestring", postgresql_regconfig="english")
    )

Which would emit:

.. sourcecode:: sql

    SELECT mytable.id FROM mytable
    WHERE mytable.title @@ plainto_tsquery('english', 'somestring')

When using other PostgreSQL search functions with :data:`.func`, the
"regconfig" parameter may be passed directly as the initial argument::

    select(mytable.c.id).where(
        func.to_tsvector("english", mytable.c.title).bool_op("@@")(
            func.to_tsquery("english", "somestring")
        )
    )

produces a statement equivalent to:

.. sourcecode:: sql

    SELECT mytable.id FROM mytable
    WHERE to_tsvector('english', mytable.title) @@
        to_tsquery('english', 'somestring')

It is recommended that you use the ``EXPLAIN ANALYZE...`` tool from
PostgreSQL to ensure that you are generating queries with SQLAlchemy that
take full advantage of any indexes you may have created for full text search.

.. seealso::

    `Full Text Search <https://www.postgresql.org/docs/current/textsearch-controls.html>`_ - in the PostgreSQL documentation


FROM ONLY ...
-------------

The dialect supports PostgreSQL's ONLY keyword for targeting only a particular
table in an inheritance hierarchy. This can be used to produce the
``SELECT ... FROM ONLY``, ``UPDATE ONLY ...``, and ``DELETE FROM ONLY ...``
syntaxes. It uses SQLAlchemy's hints mechanism::

    # SELECT ... FROM ONLY ...
    result = table.select().with_hint(table, "ONLY", "postgresql")
    print(result.fetchall())

    # UPDATE ONLY ...
    table.update(values=dict(foo="bar")).with_hint(
        "ONLY", dialect_name="postgresql"
    )

    # DELETE FROM ONLY ...
    table.delete().with_hint("ONLY", dialect_name="postgresql")

.. _postgresql_indexes:

PostgreSQL-Specific Index Options
---------------------------------

Several extensions to the :class:`.Index` construct are available, specific
to the PostgreSQL dialect.

.. _postgresql_covering_indexes:

Covering Indexes
^^^^^^^^^^^^^^^^

The ``postgresql_include`` option renders INCLUDE(colname) for the given
string names::

    Index("my_index", table.c.x, postgresql_include=["y"])

would render the index as ``CREATE INDEX my_index ON table (x) INCLUDE (y)``

Note that this feature requires PostgreSQL 11 or later.

.. seealso::

  :ref:`postgresql_constraint_options`

.. versionadded:: 1.4

.. _postgresql_partial_indexes:

Partial Indexes
^^^^^^^^^^^^^^^

Partial indexes add criterion to the index definition so that the index is
applied to a subset of rows.   These can be specified on :class:`.Index`
using the ``postgresql_where`` keyword argument::

  Index("my_index", my_table.c.id, postgresql_where=my_table.c.value > 10)

.. _postgresql_operator_classes:

Operator Classes
^^^^^^^^^^^^^^^^

PostgreSQL allows the specification of an *operator class* for each column of
an index (see
https://www.postgresql.org/docs/current/interactive/indexes-opclass.html).
The :class:`.Index` construct allows these to be specified via the
``postgresql_ops`` keyword argument::

    Index(
        "my_index",
        my_table.c.id,
        my_table.c.data,
        postgresql_ops={"data": "text_pattern_ops", "id": "int4_ops"},
    )

Note that the keys in the ``postgresql_ops`` dictionaries are the
"key" name of the :class:`_schema.Column`, i.e. the name used to access it from
the ``.c`` collection of :class:`_schema.Table`, which can be configured to be
different than the actual name of the column as expressed in the database.

If ``postgresql_ops`` is to be used against a complex SQL expression such
as a function call, then to apply to the column it must be given a label
that is identified in the dictionary by name, e.g.::

    Index(
        "my_index",
        my_table.c.id,
        func.lower(my_table.c.data).label("data_lower"),
        postgresql_ops={"data_lower": "text_pattern_ops", "id": "int4_ops"},
    )

Operator classes are also supported by the
:class:`_postgresql.ExcludeConstraint` construct using the
:paramref:`_postgresql.ExcludeConstraint.ops` parameter. See that parameter for
details.

.. versionadded:: 1.3.21 added support for operator classes with
   :class:`_postgresql.ExcludeConstraint`.


Index Types
^^^^^^^^^^^

PostgreSQL provides several index types: B-Tree, Hash, GiST, and GIN, as well
as the ability for users to create their own (see
https://www.postgresql.org/docs/current/static/indexes-types.html). These can be
specified on :class:`.Index` using the ``postgresql_using`` keyword argument::

    Index("my_index", my_table.c.data, postgresql_using="gin")

The value passed to the keyword argument will be simply passed through to the
underlying CREATE INDEX command, so it *must* be a valid index type for your
version of PostgreSQL.

.. _postgresql_index_storage:

Index Storage Parameters
^^^^^^^^^^^^^^^^^^^^^^^^

PostgreSQL allows storage parameters to be set on indexes. The storage
parameters available depend on the index method used by the index. Storage
parameters can be specified on :class:`.Index` using the ``postgresql_with``
keyword argument::

    Index("my_index", my_table.c.data, postgresql_with={"fillfactor": 50})

PostgreSQL allows to define the tablespace in which to create the index.
The tablespace can be specified on :class:`.Index` using the
``postgresql_tablespace`` keyword argument::

    Index("my_index", my_table.c.data, postgresql_tablespace="my_tablespace")

Note that the same option is available on :class:`_schema.Table` as well.

.. _postgresql_index_concurrently:

Indexes with CONCURRENTLY
^^^^^^^^^^^^^^^^^^^^^^^^^

The PostgreSQL index option CONCURRENTLY is supported by passing the
flag ``postgresql_concurrently`` to the :class:`.Index` construct::

    tbl = Table("testtbl", m, Column("data", Integer))

    idx1 = Index("test_idx1", tbl.c.data, postgresql_concurrently=True)

The above index construct will render DDL for CREATE INDEX, assuming
PostgreSQL 8.2 or higher is detected or for a connection-less dialect, as:

.. sourcecode:: sql

    CREATE INDEX CONCURRENTLY test_idx1 ON testtbl (data)

For DROP INDEX, assuming PostgreSQL 9.2 or higher is detected or for
a connection-less dialect, it will emit:

.. sourcecode:: sql

    DROP INDEX CONCURRENTLY test_idx1

When using CONCURRENTLY, the PostgreSQL database requires that the statement
be invoked outside of a transaction block.   The Python DBAPI enforces that
even for a single statement, a transaction is present, so to use this
construct, the DBAPI's "autocommit" mode must be used::

    metadata = MetaData()
    table = Table("foo", metadata, Column("id", String))
    index = Index("foo_idx", table.c.id, postgresql_concurrently=True)

    with engine.connect() as conn:
        with conn.execution_options(isolation_level="AUTOCOMMIT"):
            table.create(conn)

.. seealso::

    :ref:`postgresql_isolation_level`

.. _postgresql_index_reflection:

PostgreSQL Index Reflection
---------------------------

The PostgreSQL database creates a UNIQUE INDEX implicitly whenever the
UNIQUE CONSTRAINT construct is used.   When inspecting a table using
:class:`_reflection.Inspector`, the :meth:`_reflection.Inspector.get_indexes`
and the :meth:`_reflection.Inspector.get_unique_constraints`
will report on these
two constructs distinctly; in the case of the index, the key
``duplicates_constraint`` will be present in the index entry if it is
detected as mirroring a constraint.   When performing reflection using
``Table(..., autoload_with=engine)``, the UNIQUE INDEX is **not** returned
in :attr:`_schema.Table.indexes` when it is detected as mirroring a
:class:`.UniqueConstraint` in the :attr:`_schema.Table.constraints` collection
.

Special Reflection Options
--------------------------

The :class:`_reflection.Inspector`
used for the PostgreSQL backend is an instance
of :class:`.PGInspector`, which offers additional methods::

    from sqlalchemy import create_engine, inspect

    engine = create_engine("postgresql+psycopg2://localhost/test")
    insp = inspect(engine)  # will be a PGInspector

    print(insp.get_enums())

.. autoclass:: PGInspector
    :members:

.. _postgresql_table_options:

PostgreSQL Table Options
------------------------

Several options for CREATE TABLE are supported directly by the PostgreSQL
dialect in conjunction with the :class:`_schema.Table` construct:

* ``INHERITS``::

    Table("some_table", metadata, ..., postgresql_inherits="some_supertable")

    Table("some_table", metadata, ..., postgresql_inherits=("t1", "t2", ...))

* ``ON COMMIT``::

    Table("some_table", metadata, ..., postgresql_on_commit="PRESERVE ROWS")

*
  ``PARTITION BY``::

    Table(
        "some_table",
        metadata,
        ...,
        postgresql_partition_by="LIST (part_column)",
    )

  .. versionadded:: 1.2.6

*
  ``TABLESPACE``::

    Table("some_table", metadata, ..., postgresql_tablespace="some_tablespace")

  The above option is also available on the :class:`.Index` construct.

*
  ``USING``::

    Table("some_table", metadata, ..., postgresql_using="heap")

  .. versionadded:: 2.0.26

* ``WITH OIDS``::

    Table("some_table", metadata, ..., postgresql_with_oids=True)

* ``WITHOUT OIDS``::

    Table("some_table", metadata, ..., postgresql_with_oids=False)

.. seealso::

    `PostgreSQL CREATE TABLE options
    <https://www.postgresql.org/docs/current/static/sql-createtable.html>`_ -
    in the PostgreSQL documentation.

.. _postgresql_constraint_options:

PostgreSQL Constraint Options
-----------------------------

The following option(s) are supported by the PostgreSQL dialect in conjunction
with selected constraint constructs:

* ``NOT VALID``:  This option applies towards CHECK and FOREIGN KEY constraints
  when the constraint is being added to an existing table via ALTER TABLE,
  and has the effect that existing rows are not scanned during the ALTER
  operation against the constraint being added.

  When using a SQL migration tool such as `Alembic <https://alembic.sqlalchemy.org>`_
  that renders ALTER TABLE constructs, the ``postgresql_not_valid`` argument
  may be specified as an additional keyword argument within the operation
  that creates the constraint, as in the following Alembic example::

        def update():
            op.create_foreign_key(
                "fk_user_address",
                "address",
                "user",
                ["user_id"],
                ["id"],
                postgresql_not_valid=True,
            )

  The keyword is ultimately accepted directly by the
  :class:`_schema.CheckConstraint`, :class:`_schema.ForeignKeyConstraint`
  and :class:`_schema.ForeignKey` constructs; when using a tool like
  Alembic, dialect-specific keyword arguments are passed through to
  these constructs from the migration operation directives::

       CheckConstraint("some_field IS NOT NULL", postgresql_not_valid=True)

       ForeignKeyConstraint(
           ["some_id"], ["some_table.some_id"], postgresql_not_valid=True
       )

  .. versionadded:: 1.4.32

  .. seealso::

      `PostgreSQL ALTER TABLE options
      <https://www.postgresql.org/docs/current/static/sql-altertable.html>`_ -
      in the PostgreSQL documentation.

* ``INCLUDE``:  This option adds one or more columns as a "payload" to the
  unique index created automatically by PostgreSQL for the constraint.
  For example, the following table definition::

      Table(
          "mytable",
          metadata,
          Column("id", Integer, nullable=False),
          Column("value", Integer, nullable=False),
          UniqueConstraint("id", postgresql_include=["value"]),
      )

  would produce the DDL statement

  .. sourcecode:: sql

       CREATE TABLE mytable (
           id INTEGER NOT NULL,
           value INTEGER NOT NULL,
           UNIQUE (id) INCLUDE (value)
        )

  Note that this feature requires PostgreSQL 11 or later.

  .. versionadded:: 2.0.41

  .. seealso::

      :ref:`postgresql_covering_indexes`

  .. seealso::

      `PostgreSQL CREATE TABLE options
      <https://www.postgresql.org/docs/current/static/sql-createtable.html>`_ -
      in the PostgreSQL documentation.

* Column list with foreign key ``ON DELETE SET`` actions:  This applies to
  :class:`.ForeignKey` and :class:`.ForeignKeyConstraint`, the :paramref:`.ForeignKey.ondelete`
  parameter will accept on the PostgreSQL backend only a string list of column
  names inside parenthesis, following the ``SET NULL`` or ``SET DEFAULT``
  phrases, which will limit the set of columns that are subject to the
  action::

        fktable = Table(
            "fktable",
            metadata,
            Column("tid", Integer),
            Column("id", Integer),
            Column("fk_id_del_set_null", Integer),
            ForeignKeyConstraint(
                columns=["tid", "fk_id_del_set_null"],
                refcolumns=[pktable.c.tid, pktable.c.id],
                ondelete="SET NULL (fk_id_del_set_null)",
            ),
        )

  .. versionadded:: 2.0.40


.. _postgresql_table_valued_overview:

Table values, Table and Column valued functions, Row and Tuple objects
-----------------------------------------------------------------------

PostgreSQL makes great use of modern SQL forms such as table-valued functions,
tables and rows as values.   These constructs are commonly used as part
of PostgreSQL's support for complex datatypes such as JSON, ARRAY, and other
datatypes.  SQLAlchemy's SQL expression language has native support for
most table-valued and row-valued forms.

.. _postgresql_table_valued:

Table-Valued Functions
^^^^^^^^^^^^^^^^^^^^^^^

Many PostgreSQL built-in functions are intended to be used in the FROM clause
of a SELECT statement, and are capable of returning table rows or sets of table
rows. A large portion of PostgreSQL's JSON functions for example such as
``json_array_elements()``, ``json_object_keys()``, ``json_each_text()``,
``json_each()``, ``json_to_record()``, ``json_populate_recordset()`` use such
forms. These classes of SQL function calling forms in SQLAlchemy are available
using the :meth:`_functions.FunctionElement.table_valued` method in conjunction
with :class:`_functions.Function` objects generated from the :data:`_sql.func`
namespace.

Examples from PostgreSQL's reference documentation follow below:

* ``json_each()``:

  .. sourcecode:: pycon+sql

    >>> from sqlalchemy import select, func
    >>> stmt = select(
    ...     func.json_each('{"a":"foo", "b":"bar"}').table_valued("key", "value")
    ... )
    >>> print(stmt)
    {printsql}SELECT anon_1.key, anon_1.value
    FROM json_each(:json_each_1) AS anon_1

* ``json_populate_record()``:

  .. sourcecode:: pycon+sql

    >>> from sqlalchemy import select, func, literal_column
    >>> stmt = select(
    ...     func.json_populate_record(
    ...         literal_column("null::myrowtype"), '{"a":1,"b":2}'
    ...     ).table_valued("a", "b", name="x")
    ... )
    >>> print(stmt)
    {printsql}SELECT x.a, x.b
    FROM json_populate_record(null::myrowtype, :json_populate_record_1) AS x

* ``json_to_record()`` - this form uses a PostgreSQL specific form of derived
  columns in the alias, where we may make use of :func:`_sql.column` elements with
  types to produce them.  The :meth:`_functions.FunctionElement.table_valued`
  method produces  a :class:`_sql.TableValuedAlias` construct, and the method
  :meth:`_sql.TableValuedAlias.render_derived` method sets up the derived
  columns specification:

  .. sourcecode:: pycon+sql

    >>> from sqlalchemy import select, func, column, Integer, Text
    >>> stmt = select(
    ...     func.json_to_record('{"a":1,"b":[1,2,3],"c":"bar"}')
    ...     .table_valued(
    ...         column("a", Integer),
    ...         column("b", Text),
    ...         column("d", Text),
    ...     )
    ...     .render_derived(name="x", with_types=True)
    ... )
    >>> print(stmt)
    {printsql}SELECT x.a, x.b, x.d
    FROM json_to_record(:json_to_record_1) AS x(a INTEGER, b TEXT, d TEXT)

* ``WITH ORDINALITY`` - part of the SQL standard, ``WITH ORDINALITY`` adds an
  ordinal counter to the output of a function and is accepted by a limited set
  of PostgreSQL functions including ``unnest()`` and ``generate_series()``. The
  :meth:`_functions.FunctionElement.table_valued` method accepts a keyword
  parameter ``with_ordinality`` for this purpose, which accepts the string name
  that will be applied to the "ordinality" column:

  .. sourcecode:: pycon+sql

    >>> from sqlalchemy import select, func
    >>> stmt = select(
    ...     func.generate_series(4, 1, -1)
    ...     .table_valued("value", with_ordinality="ordinality")
    ...     .render_derived()
    ... )
    >>> print(stmt)
    {printsql}SELECT anon_1.value, anon_1.ordinality
    FROM generate_series(:generate_series_1, :generate_series_2, :generate_series_3)
    WITH ORDINALITY AS anon_1(value, ordinality)

.. versionadded:: 1.4.0b2

.. seealso::

    :ref:`tutorial_functions_table_valued` - in the :ref:`unified_tutorial`

.. _postgresql_column_valued:

Column Valued Functions
^^^^^^^^^^^^^^^^^^^^^^^

Similar to the table valued function, a column valued function is present
in the FROM clause, but delivers itself to the columns clause as a single
scalar value.  PostgreSQL functions such as ``json_array_elements()``,
``unnest()`` and ``generate_series()`` may use this form. Column valued functions are available using the
:meth:`_functions.FunctionElement.column_valued` method of :class:`_functions.FunctionElement`:

* ``json_array_elements()``:

  .. sourcecode:: pycon+sql

    >>> from sqlalchemy import select, func
    >>> stmt = select(
    ...     func.json_array_elements('["one", "two"]').column_valued("x")
    ... )
    >>> print(stmt)
    {printsql}SELECT x
    FROM json_array_elements(:json_array_elements_1) AS x

* ``unnest()`` - in order to generate a PostgreSQL ARRAY literal, the
  :func:`_postgresql.array` construct may be used:

  .. sourcecode:: pycon+sql

    >>> from sqlalchemy.dialects.postgresql import array
    >>> from sqlalchemy import select, func
    >>> stmt = select(func.unnest(array([1, 2])).column_valued())
    >>> print(stmt)
    {printsql}SELECT anon_1
    FROM unnest(ARRAY[%(param_1)s, %(param_2)s]) AS anon_1

  The function can of course be used against an existing table-bound column
  that's of type :class:`_types.ARRAY`:

  .. sourcecode:: pycon+sql

    >>> from sqlalchemy import table, column, ARRAY, Integer
    >>> from sqlalchemy import select, func
    >>> t = table("t", column("value", ARRAY(Integer)))
    >>> stmt = select(func.unnest(t.c.value).column_valued("unnested_value"))
    >>> print(stmt)
    {printsql}SELECT unnested_value
    FROM unnest(t.value) AS unnested_value

.. seealso::

    :ref:`tutorial_functions_column_valued` - in the :ref:`unified_tutorial`


Row Types
^^^^^^^^^

Built-in support for rendering a ``ROW`` may be approximated using
``func.ROW`` with the :attr:`_sa.func` namespace, or by using the
:func:`_sql.tuple_` construct:

.. sourcecode:: pycon+sql

    >>> from sqlalchemy import table, column, func, tuple_
    >>> t = table("t", column("id"), column("fk"))
    >>> stmt = (
    ...     t.select()
    ...     .where(tuple_(t.c.id, t.c.fk) > (1, 2))
    ...     .where(func.ROW(t.c.id, t.c.fk) < func.ROW(3, 7))
    ... )
    >>> print(stmt)
    {printsql}SELECT t.id, t.fk
    FROM t
    WHERE (t.id, t.fk) > (:param_1, :param_2) AND ROW(t.id, t.fk) < ROW(:ROW_1, :ROW_2)

.. seealso::

    `PostgreSQL Row Constructors
    <https://www.postgresql.org/docs/current/sql-expressions.html#SQL-SYNTAX-ROW-CONSTRUCTORS>`_

    `PostgreSQL Row Constructor Comparison
    <https://www.postgresql.org/docs/current/functions-comparisons.html#ROW-WISE-COMPARISON>`_

Table Types passed to Functions
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

PostgreSQL supports passing a table as an argument to a function, which is
known as a "record" type. SQLAlchemy :class:`_sql.FromClause` objects
such as :class:`_schema.Table` support this special form using the
:meth:`_sql.FromClause.table_valued` method, which is comparable to the
:meth:`_functions.FunctionElement.table_valued` method except that the collection
of columns is already established by that of the :class:`_sql.FromClause`
itself:

.. sourcecode:: pycon+sql

    >>> from sqlalchemy import table, column, func, select
    >>> a = table("a", column("id"), column("x"), column("y"))
    >>> stmt = select(func.row_to_json(a.table_valued()))
    >>> print(stmt)
    {printsql}SELECT row_to_json(a) AS row_to_json_1
    FROM a

.. versionadded:: 1.4.0b2



"""  # noqa: E501

from __future__ import annotations

from collections import defaultdict
from functools import lru_cache
import re
from typing import Any
from typing import cast
from typing import Dict
from typing import List
from typing import Optional
from typing import Tuple
from typing import TYPE_CHECKING
from typing import Union

from . import arraylib as _array
from . import json as _json
from . import pg_catalog
from . import ranges as _ranges
from .ext import _regconfig_fn
from .ext import aggregate_order_by
from .hstore import HSTORE
from .named_types import CreateDomainType as CreateDomainType  # noqa: F401
from .named_types import CreateEnumType as CreateEnumType  # noqa: F401
from .named_types import DOMAIN as DOMAIN  # noqa: F401
from .named_types import DropDomainType as DropDomainType  # noqa: F401
from .named_types import DropEnumType as DropEnumType  # noqa: F401
from .named_types import ENUM as ENUM  # noqa: F401
from .named_types import NamedType as NamedType  # noqa: F401
from .types import _DECIMAL_TYPES  # noqa: F401
from .types import _FLOAT_TYPES  # noqa: F401
from .types import _INT_TYPES  # noqa: F401
from .types import BIT as BIT
from .types import BYTEA as BYTEA
from .types import CIDR as CIDR
from .types import CITEXT as CITEXT
from .types import INET as INET
from .types import INTERVAL as INTERVAL
from .types import MACADDR as MACADDR
from .types import MACADDR8 as MACADDR8
from .types import MONEY as MONEY
from .types import OID as OID
from .types import PGBit as PGBit  # noqa: F401
from .types import PGCidr as PGCidr  # noqa: F401
from .types import PGInet as PGInet  # noqa: F401
from .types import PGInterval as PGInterval  # noqa: F401
from .types import PGMacAddr as PGMacAddr  # noqa: F401
from .types import PGMacAddr8 as PGMacAddr8  # noqa: F401
from .types import PGUuid as PGUuid
from .types import REGCLASS as REGCLASS
from .types import REGCONFIG as REGCONFIG  # noqa: F401
from .types import TIME as TIME
from .types import TIMESTAMP as TIMESTAMP
from .types import TSVECTOR as TSVECTOR
from ... import exc
from ... import schema
from ... import select
from ... import sql
from ... import util
from ...engine import characteristics
from ...engine import default
from ...engine import interfaces
from ...engine import ObjectKind
from ...engine import ObjectScope
from ...engine import reflection
from ...engine import URL
from ...engine.reflection import ReflectionDefaults
from ...sql import bindparam
from ...sql import coercions
from ...sql import compiler
from ...sql import elements
from ...sql import expression
from ...sql import roles
from ...sql import sqltypes
from ...sql import util as sql_util
from ...sql.compiler import InsertmanyvaluesSentinelOpts
from ...sql.visitors import InternalTraversal
from ...types import BIGINT
from ...types import BOOLEAN
from ...types import CHAR
from ...types import DATE
from ...types import DOUBLE_PRECISION
from ...types import FLOAT
from ...types import INTEGER
from ...types import NUMERIC
from ...types import REAL
from ...types import SMALLINT
from ...types import TEXT
from ...types import UUID as UUID
from ...types import VARCHAR
from ...util.typing import TypedDict

IDX_USING = re.compile(r"^(?:btree|hash|gist|gin|[\w_]+)$", re.I)

RESERVED_WORDS = {
    "all",
    "analyse",
    "analyze",
    "and",
    "any",
    "array",
    "as",
    "asc",
    "asymmetric",
    "both",
    "case",
    "cast",
    "check",
    "collate",
    "column",
    "constraint",
    "create",
    "current_catalog",
    "current_date",
    "current_role",
    "current_time",
    "current_timestamp",
    "current_user",
    "default",
    "deferrable",
    "desc",
    "distinct",
    "do",
    "else",
    "end",
    "except",
    "false",
    "fetch",
    "for",
    "foreign",
    "from",
    "grant",
    "group",
    "having",
    "in",
    "initially",
    "intersect",
    "into",
    "leading",
    "limit",
    "localtime",
    "localtimestamp",
    "new",
    "not",
    "null",
    "of",
    "off",
    "offset",
    "old",
    "on",
    "only",
    "or",
    "order",
    "placing",
    "primary",
    "references",
    "returning",
    "select",
    "session_user",
    "some",
    "symmetric",
    "table",
    "then",
    "to",
    "trailing",
    "true",
    "union",
    "unique",
    "user",
    "using",
    "variadic",
    "when",
    "where",
    "window",
    "with",
    "authorization",
    "between",
    "binary",
    "cross",
    "current_schema",
    "freeze",
    "full",
    "ilike",
    "inner",
    "is",
    "isnull",
    "join",
    "left",
    "like",
    "natural",
    "notnull",
    "outer",
    "over",
    "overlaps",
    "right",
    "similar",
    "verbose",
}


colspecs = {
    sqltypes.ARRAY: _array.ARRAY,
    sqltypes.Interval: INTERVAL,
    sqltypes.Enum: ENUM,
    sqltypes.JSON.JSONPathType: _json.JSONPATH,
    sqltypes.JSON: _json.JSON,
    sqltypes.Uuid: PGUuid,
}


ischema_names = {
    "_array": _array.ARRAY,
    "hstore": HSTORE,
    "json": _json.JSON,
    "jsonb": _json.JSONB,
    "int4range": _ranges.INT4RANGE,
    "int8range": _ranges.INT8RANGE,
    "numrange": _ranges.NUMRANGE,
    "daterange": _ranges.DATERANGE,
    "tsrange": _ranges.TSRANGE,
    "tstzrange": _ranges.TSTZRANGE,
    "int4multirange": _ranges.INT4MULTIRANGE,
    "int8multirange": _ranges.INT8MULTIRANGE,
    "nummultirange": _ranges.NUMMULTIRANGE,
    "datemultirange": _ranges.DATEMULTIRANGE,
    "tsmultirange": _ranges.TSMULTIRANGE,
    "tstzmultirange": _ranges.TSTZMULTIRANGE,
    "integer": INTEGER,
    "bigint": BIGINT,
    "smallint": SMALLINT,
    "character varying": VARCHAR,
    "character": CHAR,
    '"char"': sqltypes.String,
    "name": sqltypes.String,
    "text": TEXT,
    "numeric": NUMERIC,
    "float": FLOAT,
    "real": REAL,
    "inet": INET,
    "cidr": CIDR,
    "citext": CITEXT,
    "uuid": UUID,
    "bit": BIT,
    "bit varying": BIT,
    "macaddr": MACADDR,
    "macaddr8": MACADDR8,
    "money": MONEY,
    "oid": OID,
    "regclass": REGCLASS,
    "double precision": DOUBLE_PRECISION,
    "timestamp": TIMESTAMP,
    "timestamp with time zone": TIMESTAMP,
    "timestamp without time zone": TIMESTAMP,
    "time with time zone": TIME,
    "time without time zone": TIME,
    "date": DATE,
    "time": TIME,
    "bytea": BYTEA,
    "boolean": BOOLEAN,
    "interval": INTERVAL,
    "tsvector": TSVECTOR,
}


class PGCompiler(compiler.SQLCompiler):
    def visit_to_tsvector_func(self, element, **kw):
        return self._assert_pg_ts_ext(element, **kw)

    def visit_to_tsquery_func(self, element, **kw):
        return self._assert_pg_ts_ext(element, **kw)

    def visit_plainto_tsquery_func(self, element, **kw):
        return self._assert_pg_ts_ext(element, **kw)

    def visit_phraseto_tsquery_func(self, element, **kw):
        return self._assert_pg_ts_ext(element, **kw)

    def visit_websearch_to_tsquery_func(self, element, **kw):
        return self._assert_pg_ts_ext(element, **kw)

    def visit_ts_headline_func(self, element, **kw):
        return self._assert_pg_ts_ext(element, **kw)

    def _assert_pg_ts_ext(self, element, **kw):
        if not isinstance(element, _regconfig_fn):
            # other options here include trying to rewrite the function
            # with the correct types.  however, that means we have to
            # "un-SQL-ize" the first argument, which can't work in a
            # generalized way. Also, parent compiler class has already added
            # the incorrect return type to the result map.   So let's just
            # make sure the function we want is used up front.

            raise exc.CompileError(
                f'Can\'t compile "{element.name}()" full text search '
                f"function construct that does not originate from the "
                f'"sqlalchemy.dialects.postgresql" package.  '
                f'Please ensure "import sqlalchemy.dialects.postgresql" is '
                f"called before constructing "
                f'"sqlalchemy.func.{element.name}()" to ensure registration '
                f"of the correct argument and return types."
            )

        return f"{element.name}{self.function_argspec(element, **kw)}"

    def render_bind_cast(self, type_, dbapi_type, sqltext):
        if dbapi_type._type_affinity is sqltypes.String and dbapi_type.length:
            # use VARCHAR with no length for VARCHAR cast.
            # see #9511
            dbapi_type = sqltypes.STRINGTYPE
        return f"""{sqltext}::{
            self.dialect.type_compiler_instance.process(
                dbapi_type, identifier_preparer=self.preparer
            )
        }"""

    def visit_array(self, element, **kw):
        if not element.clauses and not element.type.item_type._isnull:
            return "ARRAY[]::%s" % element.type.compile(self.dialect)
        return "ARRAY[%s]" % self.visit_clauselist(element, **kw)

    def visit_slice(self, element, **kw):
        return "%s:%s" % (
            self.process(element.start, **kw),
            self.process(element.stop, **kw),
        )

    def visit_bitwise_xor_op_binary(self, binary, operator, **kw):
        return self._generate_generic_binary(binary, " # ", **kw)

    def visit_json_getitem_op_binary(
        self, binary, operator, _cast_applied=False, **kw
    ):
        if (
            not _cast_applied
            and binary.type._type_affinity is not sqltypes.JSON
        ):
            kw["_cast_applied"] = True
            return self.process(sql.cast(binary, binary.type), **kw)

        kw["eager_grouping"] = True

        if (
            not _cast_applied
            and isinstance(binary.left.type, _json.JSONB)
            and self.dialect._supports_jsonb_subscripting
        ):
            # for pg14+JSONB use subscript notation: col['key'] instead
            # of col -> 'key'
            return "%s[%s]" % (
                self.process(binary.left, **kw),
                self.process(binary.right, **kw),
            )
        else:
            # Fall back to arrow notation for older versions or when cast
            # is applied
            return self._generate_generic_binary(
                binary, " -> " if not _cast_applied else " ->> ", **kw
            )

    def visit_json_path_getitem_op_binary(
        self, binary, operator, _cast_applied=False, **kw
    ):
        if (
            not _cast_applied
            and binary.type._type_affinity is not sqltypes.JSON
        ):
            kw["_cast_applied"] = True
            return self.process(sql.cast(binary, binary.type), **kw)

        kw["eager_grouping"] = True
        return self._generate_generic_binary(
            binary, " #> " if not _cast_applied else " #>> ", **kw
        )

    def visit_getitem_binary(self, binary, operator, **kw):
        return "%s[%s]" % (
            self.process(binary.left, **kw),
            self.process(binary.right, **kw),
        )

    def visit_aggregate_order_by(self, element, **kw):
        return "%s ORDER BY %s" % (
            self.process(element.target, **kw),
            self.process(element.order_by, **kw),
        )

    def visit_match_op_binary(self, binary, operator, **kw):
        if "postgresql_regconfig" in binary.modifiers:
            regconfig = self.render_literal_value(
                binary.modifiers["postgresql_regconfig"], sqltypes.STRINGTYPE
            )
            if regconfig:
                return "%s @@ plainto_tsquery(%s, %s)" % (
                    self.process(binary.left, **kw),
                    regconfig,
                    self.process(binary.right, **kw),
                )
        return "%s @@ plainto_tsquery(%s)" % (
            self.process(binary.left, **kw),
            self.process(binary.right, **kw),
        )

    def visit_ilike_case_insensitive_operand(self, element, **kw):
        return element.element._compiler_dispatch(self, **kw)

    def visit_ilike_op_binary(self, binary, operator, **kw):
        escape = binary.modifiers.get("escape", None)

        return "%s ILIKE %s" % (
            self.process(binary.left, **kw),
            self.process(binary.right, **kw),
        ) + (
            " ESCAPE " + self.render_literal_value(escape, sqltypes.STRINGTYPE)
            if escape is not None
            else ""
        )

    def visit_not_ilike_op_binary(self, binary, operator, **kw):
        escape = binary.modifiers.get("escape", None)
        return "%s NOT ILIKE %s" % (
            self.process(binary.left, **kw),
            self.process(binary.right, **kw),
        ) + (
            " ESCAPE " + self.render_literal_value(escape, sqltypes.STRINGTYPE)
            if escape is not None
            else ""
        )

    def _regexp_match(self, base_op, binary, operator, kw):
        flags = binary.modifiers["flags"]
        if flags is None:
            return self._generate_generic_binary(
                binary, " %s " % base_op, **kw
            )
        if flags == "i":
            return self._generate_generic_binary(
                binary, " %s* " % base_op, **kw
            )
        return "%s %s CONCAT('(?', %s, ')', %s)" % (
            self.process(binary.left, **kw),
            base_op,
            self.render_literal_value(flags, sqltypes.STRINGTYPE),
            self.process(binary.right, **kw),
        )

    def visit_regexp_match_op_binary(self, binary, operator, **kw):
        return self._regexp_match("~", binary, operator, kw)

    def visit_not_regexp_match_op_binary(self, binary, operator, **kw):
        return self._regexp_match("!~", binary, operator, kw)

    def visit_regexp_replace_op_binary(self, binary, operator, **kw):
        string = self.process(binary.left, **kw)
        pattern_replace = self.process(binary.right, **kw)
        flags = binary.modifiers["flags"]
        if flags is None:
            return "REGEXP_REPLACE(%s, %s)" % (
                string,
                pattern_replace,
            )
        else:
            return "REGEXP_REPLACE(%s, %s, %s)" % (
                string,
                pattern_replace,
                self.render_literal_value(flags, sqltypes.STRINGTYPE),
            )

    def visit_empty_set_expr(self, element_types, **kw):
        # cast the empty set to the type we are comparing against.  if
        # we are comparing against the null type, pick an arbitrary
        # datatype for the empty set
        return "SELECT %s WHERE 1!=1" % (
            ", ".join(
                "CAST(NULL AS %s)"
                % self.dialect.type_compiler_instance.process(
                    INTEGER() if type_._isnull else type_
                )
                for type_ in element_types or [INTEGER()]
            ),
        )

    def render_literal_value(self, value, type_):
        value = super().render_literal_value(value, type_)

        if self.dialect._backslash_escapes:
            value = value.replace("\\", "\\\\")
        return value

    def visit_aggregate_strings_func(self, fn, **kw):
        return "string_agg%s" % self.function_argspec(fn)

    def visit_sequence(self, seq, **kw):
        return "nextval('%s')" % self.preparer.format_sequence(seq)

    def limit_clause(self, select, **kw):
        text = ""
        if select._limit_clause is not None:
            text += " \n LIMIT " + self.process(select._limit_clause, **kw)
        if select._offset_clause is not None:
            if select._limit_clause is None:
                text += "\n LIMIT ALL"
            text += " OFFSET " + self.process(select._offset_clause, **kw)
        return text

    def format_from_hint_text(self, sqltext, table, hint, iscrud):
        if hint.upper() != "ONLY":
            raise exc.CompileError("Unrecognized hint: %r" % hint)
        return "ONLY " + sqltext

    def get_select_precolumns(self, select, **kw):
        # Do not call super().get_select_precolumns because
        # it will warn/raise when distinct on is present
        if select._distinct or select._distinct_on:
            if select._distinct_on:
                return (
                    "DISTINCT ON ("
                    + ", ".join(
                        [
                            self.process(col, **kw)
                            for col in select._distinct_on
                        ]
                    )
                    + ") "
                )
            else:
                return "DISTINCT "
        else:
            return ""

    def for_update_clause(self, select, **kw):
        if select._for_update_arg.read:
            if select._for_update_arg.key_share:
                tmp = " FOR KEY SHARE"
            else:
                tmp = " FOR SHARE"
        elif select._for_update_arg.key_share:
            tmp = " FOR NO KEY UPDATE"
        else:
            tmp = " FOR UPDATE"

        if select._for_update_arg.of:
            tables = util.OrderedSet()
            for c in select._for_update_arg.of:
                tables.update(sql_util.surface_selectables_only(c))

            of_kw = dict(kw)
            of_kw.update(ashint=True, use_schema=False)
            tmp += " OF " + ", ".join(
                self.process(table, **of_kw) for table in tables
            )

        if select._for_update_arg.nowait:
            tmp += " NOWAIT"
        if select._for_update_arg.skip_locked:
            tmp += " SKIP LOCKED"

        return tmp

    def visit_substring_func(self, func, **kw):
        s = self.process(func.clauses.clauses[0], **kw)
        start = self.process(func.clauses.clauses[1], **kw)
        if len(func.clauses.clauses) > 2:
            length = self.process(func.clauses.clauses[2], **kw)
            return "SUBSTRING(%s FROM %s FOR %s)" % (s, start, length)
        else:
            return "SUBSTRING(%s FROM %s)" % (s, start)

    def _on_conflict_target(self, clause, **kw):
        if clause.constraint_target is not None:
            # target may be a name of an Index, UniqueConstraint or
            # ExcludeConstraint.  While there is a separate
            # "max_identifier_length" for indexes, PostgreSQL uses the same
            # length for all objects so we can use
            # truncate_and_render_constraint_name
            target_text = (
                "ON CONSTRAINT %s"
                % self.preparer.truncate_and_render_constraint_name(
                    clause.constraint_target
                )
            )
        elif clause.inferred_target_elements is not None:
            target_text = "(%s)" % ", ".join(
                (
                    self.preparer.quote(c)
                    if isinstance(c, str)
                    else self.process(c, include_table=False, use_schema=False)
                )
                for c in clause.inferred_target_elements
            )
            if clause.inferred_target_whereclause is not None:
                target_text += " WHERE %s" % self.process(
                    clause.inferred_target_whereclause,
                    include_table=False,
                    use_schema=False,
                )
        else:
            target_text = ""

        return target_text

    def visit_on_conflict_do_nothing(self, on_conflict, **kw):
        target_text = self._on_conflict_target(on_conflict, **kw)

        if target_text:
            return "ON CONFLICT %s DO NOTHING" % target_text
        else:
            return "ON CONFLICT DO NOTHING"

    def visit_on_conflict_do_update(self, on_conflict, **kw):
        clause = on_conflict

        target_text = self._on_conflict_target(on_conflict, **kw)

        action_set_ops = []

        set_parameters = dict(clause.update_values_to_set)
        # create a list of column assignment clauses as tuples

        insert_statement = self.stack[-1]["selectable"]
        cols = insert_statement.table.c
        for c in cols:
            col_key = c.key

            if col_key in set_parameters:
                value = set_parameters.pop(col_key)
            elif c in set_parameters:
                value = set_parameters.pop(c)
            else:
                continue

            # TODO: this coercion should be up front.  we can't cache
            # SQL constructs with non-bound literals buried in them
            if coercions._is_literal(value):
                value = elements.BindParameter(None, value, type_=c.type)

            else:
                if (
                    isinstance(value, elements.BindParameter)
                    and value.type._isnull
                ):
                    value = value._clone()
                    value.type = c.type
            value_text = self.process(value.self_group(), use_schema=False)

            key_text = self.preparer.quote(c.name)
            action_set_ops.append("%s = %s" % (key_text, value_text))

        # check for names that don't match columns
        if set_parameters:
            util.warn(
                "Additional column names not matching "
                "any column keys in table '%s': %s"
                % (
                    self.current_executable.table.name,
                    (", ".join("'%s'" % c for c in set_parameters)),
                )
            )
            for k, v in set_parameters.items():
                key_text = (
                    self.preparer.quote(k)
                    if isinstance(k, str)
                    else self.process(k, use_schema=False)
                )
                value_text = self.process(
                    coercions.expect(roles.ExpressionElementRole, v),
                    use_schema=False,
                )
                action_set_ops.append("%s = %s" % (key_text, value_text))

        action_text = ", ".join(action_set_ops)
        if clause.update_whereclause is not None:
            action_text += " WHERE %s" % self.process(
                clause.update_whereclause, include_table=True, use_schema=False
            )

        return "ON CONFLICT %s DO UPDATE SET %s" % (target_text, action_text)

    def update_from_clause(
        self, update_stmt, from_table, extra_froms, from_hints, **kw
    ):
        kw["asfrom"] = True
        return "FROM " + ", ".join(
            t._compiler_dispatch(self, fromhints=from_hints, **kw)
            for t in extra_froms
        )

    def delete_extra_from_clause(
        self, delete_stmt, from_table, extra_froms, from_hints, **kw
    ):
        """Render the DELETE .. USING clause specific to PostgreSQL."""
        kw["asfrom"] = True
        return "USING " + ", ".join(
            t._compiler_dispatch(self, fromhints=from_hints, **kw)
            for t in extra_froms
        )

    def fetch_clause(self, select, **kw):
        # pg requires parens for non literal clauses. It's also required for
        # bind parameters if a ::type casts is used by the driver (asyncpg),
        # so it's easiest to just always add it
        text = ""
        if select._offset_clause is not None:
            text += "\n OFFSET (%s) ROWS" % self.process(
                select._offset_clause, **kw
            )
        if select._fetch_clause is not None:
            text += "\n FETCH FIRST (%s)%s ROWS %s" % (
                self.process(select._fetch_clause, **kw),
                " PERCENT" if select._fetch_clause_options["percent"] else "",
                (
                    "WITH TIES"
                    if select._fetch_clause_options["with_ties"]
                    else "ONLY"
                ),
            )
        return text


class PGDDLCompiler(compiler.DDLCompiler):
    def get_column_specification(self, column, **kwargs):
        colspec = self.preparer.format_column(column)
        impl_type = column.type.dialect_impl(self.dialect)
        if isinstance(impl_type, sqltypes.TypeDecorator):
            impl_type = impl_type.impl

        has_identity = (
            column.identity is not None
            and self.dialect.supports_identity_columns
        )

        if (
            column.primary_key
            and column is column.table._autoincrement_column
            and (
                self.dialect.supports_smallserial
                or not isinstance(impl_type, sqltypes.SmallInteger)
            )
            and not has_identity
            and (
                column.default is None
                or (
                    isinstance(column.default, schema.Sequence)
                    and column.default.optional
                )
            )
        ):
            if isinstance(impl_type, sqltypes.BigInteger):
                colspec += " BIGSERIAL"
            elif isinstance(impl_type, sqltypes.SmallInteger):
                colspec += " SMALLSERIAL"
            else:
                colspec += " SERIAL"
        else:
            colspec += " " + self.dialect.type_compiler_instance.process(
                column.type,
                type_expression=column,
                identifier_preparer=self.preparer,
            )
            default = self.get_column_default_string(column)
            if default is not None:
                colspec += " DEFAULT " + default

        if column.computed is not None:
            colspec += " " + self.process(column.computed)
        if has_identity:
            colspec += " " + self.process(column.identity)

        if not column.nullable and not has_identity:
            colspec += " NOT NULL"
        elif column.nullable and has_identity:
            colspec += " NULL"
        return colspec

    def _define_constraint_validity(self, constraint):
        not_valid = constraint.dialect_options["postgresql"]["not_valid"]
        return " NOT VALID" if not_valid else ""

    def _define_include(self, obj):
        includeclause = obj.dialect_options["postgresql"]["include"]
        if not includeclause:
            return ""
        inclusions = [
            obj.table.c[col] if isinstance(col, str) else col
            for col in includeclause
        ]
        return " INCLUDE (%s)" % ", ".join(
            [self.preparer.quote(c.name) for c in inclusions]
        )

    def visit_check_constraint(self, constraint, **kw):
        if constraint._type_bound:
            typ = list(constraint.columns)[0].type
            if (
                isinstance(typ, sqltypes.ARRAY)
                and isinstance(typ.item_type, sqltypes.Enum)
                and not typ.item_type.native_enum
            ):
                raise exc.CompileError(
                    "PostgreSQL dialect cannot produce the CHECK constraint "
                    "for ARRAY of non-native ENUM; please specify "
                    "create_constraint=False on this Enum datatype."
                )

        text = super().visit_check_constraint(constraint)
        text += self._define_constraint_validity(constraint)
        return text

    def visit_foreign_key_constraint(self, constraint, **kw):
        text = super().visit_foreign_key_constraint(constraint)
        text += self._define_constraint_validity(constraint)
        return text

    def visit_primary_key_constraint(self, constraint, **kw):
        text = super().visit_primary_key_constraint(constraint)
        text += self._define_include(constraint)
        return text

    def visit_unique_constraint(self, constraint, **kw):
        text = super().visit_unique_constraint(constraint)
        text += self._define_include(constraint)
        return text

    @util.memoized_property
    def _fk_ondelete_pattern(self):
        return re.compile(
            r"^(?:RESTRICT|CASCADE|SET (?:NULL|DEFAULT)(?:\s*\(.+\))?"
            r"|NO ACTION)$",
            re.I,
        )

    def define_constraint_ondelete_cascade(self, constraint):
        return " ON DELETE %s" % self.preparer.validate_sql_phrase(
            constraint.ondelete, self._fk_ondelete_pattern
        )

    def visit_create_enum_type(self, create, **kw):
        type_ = create.element

        return "CREATE TYPE %s AS ENUM (%s)" % (
            self.preparer.format_type(type_),
            ", ".join(
                self.sql_compiler.process(sql.literal(e), literal_binds=True)
                for e in type_.enums
            ),
        )

    def visit_drop_enum_type(self, drop, **kw):
        type_ = drop.element

        return "DROP TYPE %s" % (self.preparer.format_type(type_))

    def visit_create_domain_type(self, create, **kw):
        domain: DOMAIN = create.element

        options = []
        if domain.collation is not None:
            options.append(f"COLLATE {self.preparer.quote(domain.collation)}")
        if domain.default is not None:
            default = self.render_default_string(domain.default)
            options.append(f"DEFAULT {default}")
        if domain.constraint_name is not None:
            name = self.preparer.truncate_and_render_constraint_name(
                domain.constraint_name
            )
            options.append(f"CONSTRAINT {name}")
        if domain.not_null:
            options.append("NOT NULL")
        if domain.check is not None:
            check = self.sql_compiler.process(
                domain.check, include_table=False, literal_binds=True
            )
            options.append(f"CHECK ({check})")

        return (
            f"CREATE DOMAIN {self.preparer.format_type(domain)} AS "
            f"{self.type_compiler.process(domain.data_type)} "
            f"{' '.join(options)}"
        )

    def visit_drop_domain_type(self, drop, **kw):
        domain = drop.element
        return f"DROP DOMAIN {self.preparer.format_type(domain)}"

    def visit_create_index(self, create, **kw):
        preparer = self.preparer
        index = create.element
        self._verify_index_table(index)
        text = "CREATE "
        if index.unique:
            text += "UNIQUE "

        text += "INDEX "

        if self.dialect._supports_create_index_concurrently:
            concurrently = index.dialect_options["postgresql"]["concurrently"]
            if concurrently:
                text += "CONCURRENTLY "

        if create.if_not_exists:
            text += "IF NOT EXISTS "

        text += "%s ON %s " % (
            self._prepared_index_name(index, include_schema=False),
            preparer.format_table(index.table),
        )

        using = index.dialect_options["postgresql"]["using"]
        if using:
            text += (
                "USING %s "
                % self.preparer.validate_sql_phrase(using, IDX_USING).lower()
            )

        ops = index.dialect_options["postgresql"]["ops"]
        text += "(%s)" % (
            ", ".join(
                [
                    self.sql_compiler.process(
                        (
                            expr.self_group()
                            if not isinstance(expr, expression.ColumnClause)
                            else expr
                        ),
                        include_table=False,
                        literal_binds=True,
                    )
                    + (
                        (" " + ops[expr.key])
                        if hasattr(expr, "key") and expr.key in ops
                        else ""
                    )
                    for expr in index.expressions
                ]
            )
        )

        text += self._define_include(index)

        nulls_not_distinct = index.dialect_options["postgresql"][
            "nulls_not_distinct"
        ]
        if nulls_not_distinct is True:
            text += " NULLS NOT DISTINCT"
        elif nulls_not_distinct is False:
            text += " NULLS DISTINCT"

        withclause = index.dialect_options["postgresql"]["with"]
        if withclause:
            text += " WITH (%s)" % (
                ", ".join(
                    [
                        "%s = %s" % storage_parameter
                        for storage_parameter in withclause.items()
                    ]
                )
            )

        tablespace_name = index.dialect_options["postgresql"]["tablespace"]
        if tablespace_name:
            text += " TABLESPACE %s" % preparer.quote(tablespace_name)

        whereclause = index.dialect_options["postgresql"]["where"]
        if whereclause is not None:
            whereclause = coercions.expect(
                roles.DDLExpressionRole, whereclause
            )

            where_compiled = self.sql_compiler.process(
                whereclause, include_table=False, literal_binds=True
            )
            text += " WHERE " + where_compiled

        return text

    def define_unique_constraint_distinct(self, constraint, **kw):
        nulls_not_distinct = constraint.dialect_options["postgresql"][
            "nulls_not_distinct"
        ]
        if nulls_not_distinct is True:
            nulls_not_distinct_param = "NULLS NOT DISTINCT "
        elif nulls_not_distinct is False:
            nulls_not_distinct_param = "NULLS DISTINCT "
        else:
            nulls_not_distinct_param = ""
        return nulls_not_distinct_param

    def visit_drop_index(self, drop, **kw):
        index = drop.element

        text = "\nDROP INDEX "

        if self.dialect._supports_drop_index_concurrently:
            concurrently = index.dialect_options["postgresql"]["concurrently"]
            if concurrently:
                text += "CONCURRENTLY "

        if drop.if_exists:
            text += "IF EXISTS "

        text += self._prepared_index_name(index, include_schema=True)
        return text

    def visit_exclude_constraint(self, constraint, **kw):
        text = ""
        if constraint.name is not None:
            text += "CONSTRAINT %s " % self.preparer.format_constraint(
                constraint
            )
        elements = []
        kw["include_table"] = False
        kw["literal_binds"] = True
        for expr, name, op in constraint._render_exprs:
            exclude_element = self.sql_compiler.process(expr, **kw) + (
                (" " + constraint.ops[expr.key])
                if hasattr(expr, "key") and expr.key in constraint.ops
                else ""
            )

            elements.append("%s WITH %s" % (exclude_element, op))
        text += "EXCLUDE USING %s (%s)" % (
            self.preparer.validate_sql_phrase(
                constraint.using, IDX_USING
            ).lower(),
            ", ".join(elements),
        )
        if constraint.where is not None:
            text += " WHERE (%s)" % self.sql_compiler.process(
                constraint.where, literal_binds=True
            )
        text += self.define_constraint_deferrability(constraint)
        return text

    def post_create_table(self, table):
        table_opts = []
        pg_opts = table.dialect_options["postgresql"]

        inherits = pg_opts.get("inherits")
        if inherits is not None:
            if not isinstance(inherits, (list, tuple)):
                inherits = (inherits,)
            table_opts.append(
                "\n INHERITS ( "
                + ", ".join(self.preparer.quote(name) for name in inherits)
                + " )"
            )

        if pg_opts["partition_by"]:
            table_opts.append("\n PARTITION BY %s" % pg_opts["partition_by"])

        if pg_opts["using"]:
            table_opts.append("\n USING %s" % pg_opts["using"])

        if pg_opts["with_oids"] is True:
            table_opts.append("\n WITH OIDS")
        elif pg_opts["with_oids"] is False:
            table_opts.append("\n WITHOUT OIDS")

        if pg_opts["on_commit"]:
            on_commit_options = pg_opts["on_commit"].replace("_", " ").upper()
            table_opts.append("\n ON COMMIT %s" % on_commit_options)

        if pg_opts["tablespace"]:
            tablespace_name = pg_opts["tablespace"]
            table_opts.append(
                "\n TABLESPACE %s" % self.preparer.quote(tablespace_name)
            )

        return "".join(table_opts)

    def visit_computed_column(self, generated, **kw):
        if generated.persisted is False:
            raise exc.CompileError(
                "PostrgreSQL computed columns do not support 'virtual' "
                "persistence; set the 'persisted' flag to None or True for "
                "PostgreSQL support."
            )

        return "GENERATED ALWAYS AS (%s) STORED" % self.sql_compiler.process(
            generated.sqltext, include_table=False, literal_binds=True
        )

    def visit_create_sequence(self, create, **kw):
        prefix = None
        if create.element.data_type is not None:
            prefix = " AS %s" % self.type_compiler.process(
                create.element.data_type
            )

        return super().visit_create_sequence(create, prefix=prefix, **kw)

    def _can_comment_on_constraint(self, ddl_instance):
        constraint = ddl_instance.element
        if constraint.name is None:
            raise exc.CompileError(
                f"Can't emit COMMENT ON for constraint {constraint!r}: "
                "it has no name"
            )
        if constraint.table is None:
            raise exc.CompileError(
                f"Can't emit COMMENT ON for constraint {constraint!r}: "
                "it has no associated table"
            )

    def visit_set_constraint_comment(self, create, **kw):
        self._can_comment_on_constraint(create)
        return "COMMENT ON CONSTRAINT %s ON %s IS %s" % (
            self.preparer.format_constraint(create.element),
            self.preparer.format_table(create.element.table),
            self.sql_compiler.render_literal_value(
                create.element.comment, sqltypes.String()
            ),
        )

    def visit_drop_constraint_comment(self, drop, **kw):
        self._can_comment_on_constraint(drop)
        return "COMMENT ON CONSTRAINT %s ON %s IS NULL" % (
            self.preparer.format_constraint(drop.element),
            self.preparer.format_table(drop.element.table),
        )


class PGTypeCompiler(compiler.GenericTypeCompiler):
    def visit_TSVECTOR(self, type_, **kw):
        return "TSVECTOR"

    def visit_TSQUERY(self, type_, **kw):
        return "TSQUERY"

    def visit_INET(self, type_, **kw):
        return "INET"

    def visit_CIDR(self, type_, **kw):
        return "CIDR"

    def visit_CITEXT(self, type_, **kw):
        return "CITEXT"

    def visit_MACADDR(self, type_, **kw):
        return "MACADDR"

    def visit_MACADDR8(self, type_, **kw):
        return "MACADDR8"

    def visit_MONEY(self, type_, **kw):
        return "MONEY"

    def visit_OID(self, type_, **kw):
        return "OID"

    def visit_REGCONFIG(self, type_, **kw):
        return "REGCONFIG"

    def visit_REGCLASS(self, type_, **kw):
        return "REGCLASS"

    def visit_FLOAT(self, type_, **kw):
        if not type_.precision:
            return "FLOAT"
        else:
            return "FLOAT(%(precision)s)" % {"precision": type_.precision}

    def visit_double(self, type_, **kw):
        return self.visit_DOUBLE_PRECISION(type, **kw)

    def visit_BIGINT(self, type_, **kw):
        return "BIGINT"

    def visit_HSTORE(self, type_, **kw):
        return "HSTORE"

    def visit_JSON(self, type_, **kw):
        return "JSON"

    def visit_JSONB(self, type_, **kw):
        return "JSONB"

    def visit_INT4MULTIRANGE(self, type_, **kw):
        return "INT4MULTIRANGE"

    def visit_INT8MULTIRANGE(self, type_, **kw):
        return "INT8MULTIRANGE"

    def visit_NUMMULTIRANGE(self, type_, **kw):
        return "NUMMULTIRANGE"

    def visit_DATEMULTIRANGE(self, type_, **kw):
        return "DATEMULTIRANGE"

    def visit_TSMULTIRANGE(self, type_, **kw):
        return "TSMULTIRANGE"

    def visit_TSTZMULTIRANGE(self, type_, **kw):
        return "TSTZMULTIRANGE"

    def visit_INT4RANGE(self, type_, **kw):
        return "INT4RANGE"

    def visit_INT8RANGE(self, type_, **kw):
        return "INT8RANGE"

    def visit_NUMRANGE(self, type_, **kw):
        return "NUMRANGE"

    def visit_DATERANGE(self, type_, **kw):
        return "DATERANGE"

    def visit_TSRANGE(self, type_, **kw):
        return "TSRANGE"

    def visit_TSTZRANGE(self, type_, **kw):
        return "TSTZRANGE"

    def visit_json_int_index(self, type_, **kw):
        return "INT"

    def visit_json_str_index(self, type_, **kw):
        return "TEXT"

    def visit_datetime(self, type_, **kw):
        return self.visit_TIMESTAMP(type_, **kw)

    def visit_enum(self, type_, **kw):
        if not type_.native_enum or not self.dialect.supports_native_enum:
            return super().visit_enum(type_, **kw)
        else:
            return self.visit_ENUM(type_, **kw)

    def visit_ENUM(self, type_, identifier_preparer=None, **kw):
        if identifier_preparer is None:
            identifier_preparer = self.dialect.identifier_preparer
        return identifier_preparer.format_type(type_)

    def visit_DOMAIN(self, type_, identifier_preparer=None, **kw):
        if identifier_preparer is None:
            identifier_preparer = self.dialect.identifier_preparer
        return identifier_preparer.format_type(type_)

    def visit_TIMESTAMP(self, type_, **kw):
        return "TIMESTAMP%s %s" % (
            (
                "(%d)" % type_.precision
                if getattr(type_, "precision", None) is not None
                else ""
            ),
            (type_.timezone and "WITH" or "WITHOUT") + " TIME ZONE",
        )

    def visit_TIME(self, type_, **kw):
        return "TIME%s %s" % (
            (
                "(%d)" % type_.precision
                if getattr(type_, "precision", None) is not None
                else ""
            ),
            (type_.timezone and "WITH" or "WITHOUT") + " TIME ZONE",
        )

    def visit_INTERVAL(self, type_, **kw):
        text = "INTERVAL"
        if type_.fields is not None:
            text += " " + type_.fields
        if type_.precision is not None:
            text += " (%d)" % type_.precision
        return text

    def visit_BIT(self, type_, **kw):
        if type_.varying:
            compiled = "BIT VARYING"
            if type_.length is not None:
                compiled += "(%d)" % type_.length
        else:
            compiled = "BIT(%d)" % type_.length
        return compiled

    def visit_uuid(self, type_, **kw):
        if type_.native_uuid:
            return self.visit_UUID(type_, **kw)
        else:
            return super().visit_uuid(type_, **kw)

    def visit_UUID(self, type_, **kw):
        return "UUID"

    def visit_large_binary(self, type_, **kw):
        return self.visit_BYTEA(type_, **kw)

    def visit_BYTEA(self, type_, **kw):
        return "BYTEA"

    def visit_ARRAY(self, type_, **kw):
        inner = self.process(type_.item_type, **kw)
        return re.sub(
            r"((?: COLLATE.*)?)$",
            (
                r"%s\1"
                % (
                    "[]"
                    * (type_.dimensions if type_.dimensions is not None else 1)
                )
            ),
            inner,
            count=1,
        )

    def visit_json_path(self, type_, **kw):
        return self.visit_JSONPATH(type_, **kw)

    def visit_JSONPATH(self, type_, **kw):
        return "JSONPATH"


class PGIdentifierPreparer(compiler.IdentifierPreparer):
    reserved_words = RESERVED_WORDS

    def _unquote_identifier(self, value):
        if value[0] == self.initial_quote:
            value = value[1:-1].replace(
                self.escape_to_quote, self.escape_quote
            )
        return value

    def format_type(self, type_, use_schema=True):
        if not type_.name:
            raise exc.CompileError(
                f"PostgreSQL {type_.__class__.__name__} type requires a name."
            )

        name = self.quote(type_.name)
        effective_schema = self.schema_for_object(type_)

        if (
            not self.omit_schema
            and use_schema
            and effective_schema is not None
        ):
            name = f"{self.quote_schema(effective_schema)}.{name}"
        return name


class ReflectedNamedType(TypedDict):
    """Represents a reflected named type."""

    name: str
    """Name of the type."""
    schema: str
    """The schema of the type."""
    visible: bool
    """Indicates if this type is in the current search path."""


class ReflectedDomainConstraint(TypedDict):
    """Represents a reflect check constraint of a domain."""

    name: str
    """Name of the constraint."""
    check: str
    """The check constraint text."""


class ReflectedDomain(ReflectedNamedType):
    """Represents a reflected enum."""

    type: str
    """The string name of the underlying data type of the domain."""
    nullable: bool
    """Indicates if the domain allows null or not."""
    default: Optional[str]
    """The string representation of the default value of this domain
    or ``None`` if none present.
    """
    constraints: List[ReflectedDomainConstraint]
    """The constraints defined in the domain, if any.
    The constraint are in order of evaluation by postgresql.
    """
    collation: Optional[str]
    """The collation for the domain."""


class ReflectedEnum(ReflectedNamedType):
    """Represents a reflected enum."""

    labels: List[str]
    """The labels that compose the enum."""


class PGInspector(reflection.Inspector):
    dialect: PGDialect

    def get_table_oid(
        self, table_name: str, schema: Optional[str] = None
    ) -> int:
        """Return the OID for the given table name.

        :param table_name: string name of the table.  For special quoting,
         use :class:`.quoted_name`.

        :param schema: string schema name; if omitted, uses the default schema
         of the database connection.  For special quoting,
         use :class:`.quoted_name`.

        """

        with self._operation_context() as conn:
            return self.dialect.get_table_oid(
                conn, table_name, schema, info_cache=self.info_cache
            )

    def get_domains(
        self, schema: Optional[str] = None
    ) -> List[ReflectedDomain]:
        """Return a list of DOMAIN objects.

        Each member is a dictionary containing these fields:

            * name - name of the domain
            * schema - the schema name for the domain.
            * visible - boolean, whether or not this domain is visible
              in the default search path.
            * type - the type defined by this domain.
            * nullable - Indicates if this domain can be ``NULL``.
            * default - The default value of the domain or ``None`` if the
              domain has no default.
            * constraints - A list of dict wit the constraint defined by this
              domain. Each element constaints two keys: ``name`` of the
              constraint and ``check`` with the constraint text.

        :param schema: schema name.  If None, the default schema
         (typically 'public') is used.  May also be set to ``'*'`` to
         indicate load domains for all schemas.

        .. versionadded:: 2.0

        """
        with self._operation_context() as conn:
            return self.dialect._load_domains(
                conn, schema, info_cache=self.info_cache
            )

    def get_enums(self, schema: Optional[str] = None) -> List[ReflectedEnum]:
        """Return a list of ENUM objects.

        Each member is a dictionary containing these fields:

            * name - name of the enum
            * schema - the schema name for the enum.
            * visible - boolean, whether or not this enum is visible
              in the default search path.
            * labels - a list of string labels that apply to the enum.

        :param schema: schema name.  If None, the default schema
         (typically 'public') is used.  May also be set to ``'*'`` to
         indicate load enums for all schemas.

        """
        with self._operation_context() as conn:
            return self.dialect._load_enums(
                conn, schema, info_cache=self.info_cache
            )

    def get_foreign_table_names(
        self, schema: Optional[str] = None
    ) -> List[str]:
        """Return a list of FOREIGN TABLE names.

        Behavior is similar to that of
        :meth:`_reflection.Inspector.get_table_names`,
        except that the list is limited to those tables that report a
        ``relkind`` value of ``f``.

        """
        with self._operation_context() as conn:
            return self.dialect._get_foreign_table_names(
                conn, schema, info_cache=self.info_cache
            )

    def has_type(
        self, type_name: str, schema: Optional[str] = None, **kw: Any
    ) -> bool:
        """Return if the database has the specified type in the provided
        schema.

        :param type_name: the type to check.
        :param schema: schema name.  If None, the default schema
         (typically 'public') is used.  May also be set to ``'*'`` to
         check in all schemas.

        .. versionadded:: 2.0

        """
        with self._operation_context() as conn:
            return self.dialect.has_type(
                conn, type_name, schema, info_cache=self.info_cache
            )


class PGExecutionContext(default.DefaultExecutionContext):
    def fire_sequence(self, seq, type_):
        return self._execute_scalar(
            (
                "select nextval('%s')"
                % self.identifier_preparer.format_sequence(seq)
            ),
            type_,
        )

    def get_insert_default(self, column):
        if column.primary_key and column is column.table._autoincrement_column:
            if column.server_default and column.server_default.has_argument:
                # pre-execute passive defaults on primary key columns
                return self._execute_scalar(
                    "select %s" % column.server_default.arg, column.type
                )

            elif column.default is None or (
                column.default.is_sequence and column.default.optional
            ):
                # execute the sequence associated with a SERIAL primary
                # key column. for non-primary-key SERIAL, the ID just
                # generates server side.

                try:
                    seq_name = column._postgresql_seq_name
                except AttributeError:
                    tab = column.table.name
                    col = column.name
                    tab = tab[0 : 29 + max(0, (29 - len(col)))]
                    col = col[0 : 29 + max(0, (29 - len(tab)))]
                    name = "%s_%s_seq" % (tab, col)
                    column._postgresql_seq_name = seq_name = name

                if column.table is not None:
                    effective_schema = self.connection.schema_for_object(
                        column.table
                    )
                else:
                    effective_schema = None

                if effective_schema is not None:
                    exc = 'select nextval(\'"%s"."%s"\')' % (
                        effective_schema,
                        seq_name,
                    )
                else:
                    exc = "select nextval('\"%s\"')" % (seq_name,)

                return self._execute_scalar(exc, column.type)

        return super().get_insert_default(column)


class PGReadOnlyConnectionCharacteristic(
    characteristics.ConnectionCharacteristic
):
    transactional = True

    def reset_characteristic(self, dialect, dbapi_conn):
        dialect.set_readonly(dbapi_conn, False)

    def set_characteristic(self, dialect, dbapi_conn, value):
        dialect.set_readonly(dbapi_conn, value)

    def get_characteristic(self, dialect, dbapi_conn):
        return dialect.get_readonly(dbapi_conn)


class PGDeferrableConnectionCharacteristic(
    characteristics.ConnectionCharacteristic
):
    transactional = True

    def reset_characteristic(self, dialect, dbapi_conn):
        dialect.set_deferrable(dbapi_conn, False)

    def set_characteristic(self, dialect, dbapi_conn, value):
        dialect.set_deferrable(dbapi_conn, value)

    def get_characteristic(self, dialect, dbapi_conn):
        return dialect.get_deferrable(dbapi_conn)


class PGDialect(default.DefaultDialect):
    name = "postgresql"
    supports_statement_cache = True
    supports_alter = True
    max_identifier_length = 63
    supports_sane_rowcount = True

    bind_typing = interfaces.BindTyping.RENDER_CASTS

    supports_native_enum = True
    supports_native_boolean = True
    supports_native_uuid = True
    supports_smallserial = True

    supports_sequences = True
    sequences_optional = True
    preexecute_autoincrement_sequences = True
    postfetch_lastrowid = False
    use_insertmanyvalues = True

    returns_native_bytes = True

    insertmanyvalues_implicit_sentinel = (
        InsertmanyvaluesSentinelOpts.ANY_AUTOINCREMENT
        | InsertmanyvaluesSentinelOpts.USE_INSERT_FROM_SELECT
        | InsertmanyvaluesSentinelOpts.RENDER_SELECT_COL_CASTS
    )

    supports_comments = True
    supports_constraint_comments = True
    supports_default_values = True

    supports_default_metavalue = True

    supports_empty_insert = False
    supports_multivalues_insert = True

    supports_identity_columns = True

    default_paramstyle = "pyformat"
    ischema_names = ischema_names
    colspecs = colspecs

    statement_compiler = PGCompiler
    ddl_compiler = PGDDLCompiler
    type_compiler_cls = PGTypeCompiler
    preparer = PGIdentifierPreparer
    execution_ctx_cls = PGExecutionContext
    inspector = PGInspector

    update_returning = True
    delete_returning = True
    insert_returning = True
    update_returning_multifrom = True
    delete_returning_multifrom = True

    connection_characteristics = (
        default.DefaultDialect.connection_characteristics
    )
    connection_characteristics = connection_characteristics.union(
        {
            "postgresql_readonly": PGReadOnlyConnectionCharacteristic(),
            "postgresql_deferrable": PGDeferrableConnectionCharacteristic(),
        }
    )

    construct_arguments = [
        (
            schema.Index,
            {
                "using": False,
                "include": None,
                "where": None,
                "ops": {},
                "concurrently": False,
                "with": {},
                "tablespace": None,
                "nulls_not_distinct": None,
            },
        ),
        (
            schema.Table,
            {
                "ignore_search_path": False,
                "tablespace": None,
                "partition_by": None,
                "with_oids": None,
                "on_commit": None,
                "inherits": None,
                "using": None,
            },
        ),
        (
            schema.CheckConstraint,
            {
                "not_valid": False,
            },
        ),
        (
            schema.ForeignKeyConstraint,
            {
                "not_valid": False,
            },
        ),
        (
            schema.PrimaryKeyConstraint,
            {"include": None},
        ),
        (
            schema.UniqueConstraint,
            {
                "include": None,
                "nulls_not_distinct": None,
            },
        ),
    ]

    reflection_options = ("postgresql_ignore_search_path",)

    _backslash_escapes = True
    _supports_create_index_concurrently = True
    _supports_drop_index_concurrently = True
    _supports_jsonb_subscripting = True

    def __init__(
        self,
        native_inet_types=None,
        json_serializer=None,
        json_deserializer=None,
        **kwargs,
    ):
        default.DefaultDialect.__init__(self, **kwargs)

        self._native_inet_types = native_inet_types
        self._json_deserializer = json_deserializer
        self._json_serializer = json_serializer

    def initialize(self, connection):
        super().initialize(connection)

        # https://www.postgresql.org/docs/9.3/static/release-9-2.html#AEN116689
        self.supports_smallserial = self.server_version_info >= (9, 2)

        self._set_backslash_escapes(connection)

        self._supports_drop_index_concurrently = self.server_version_info >= (
            9,
            2,
        )
        self.supports_identity_columns = self.server_version_info >= (10,)

        self._supports_jsonb_subscripting = self.server_version_info >= (14,)

    def get_isolation_level_values(self, dbapi_conn):
        # note the generic dialect doesn't have AUTOCOMMIT, however
        # all postgresql dialects should include AUTOCOMMIT.
        return (
            "SERIALIZABLE",
            "READ UNCOMMITTED",
            "READ COMMITTED",
            "REPEATABLE READ",
        )

    def set_isolation_level(self, dbapi_connection, level):
        cursor = dbapi_connection.cursor()
        cursor.execute(
            "SET SESSION CHARACTERISTICS AS TRANSACTION "
            f"ISOLATION LEVEL {level}"
        )
        cursor.execute("COMMIT")
        cursor.close()

    def get_isolation_level(self, dbapi_connection):
        cursor = dbapi_connection.cursor()
        cursor.execute("show transaction isolation level")
        val = cursor.fetchone()[0]
        cursor.close()
        return val.upper()

    def set_readonly(self, connection, value):
        raise NotImplementedError()

    def get_readonly(self, connection):
        raise NotImplementedError()

    def set_deferrable(self, connection, value):
        raise NotImplementedError()

    def get_deferrable(self, connection):
        raise NotImplementedError()

    def _split_multihost_from_url(self, url: URL) -> Union[
        Tuple[None, None],
        Tuple[Tuple[Optional[str], ...], Tuple[Optional[int], ...]],
    ]:
        hosts: Optional[Tuple[Optional[str], ...]] = None
        ports_str: Union[str, Tuple[Optional[str], ...], None] = None

        integrated_multihost = False

        if "host" in url.query:
            if isinstance(url.query["host"], (list, tuple)):
                integrated_multihost = True
                hosts, ports_str = zip(
                    *[
                        token.split(":") if ":" in token else (token, None)
                        for token in url.query["host"]
                    ]
                )

            elif isinstance(url.query["host"], str):
                hosts = tuple(url.query["host"].split(","))

                if (
                    "port" not in url.query
                    and len(hosts) == 1
                    and ":" in hosts[0]
                ):
                    # internet host is alphanumeric plus dots or hyphens.
                    # this is essentially rfc1123, which refers to rfc952.
                    # https://stackoverflow.com/questions/3523028/
                    # valid-characters-of-a-hostname
                    host_port_match = re.match(
                        r"^([a-zA-Z0-9\-\.]*)(?:\:(\d*))?$", hosts[0]
                    )
                    if host_port_match:
                        integrated_multihost = True
                        h, p = host_port_match.group(1, 2)
                        if TYPE_CHECKING:
                            assert isinstance(h, str)
                            assert isinstance(p, str)
                        hosts = (h,)
                        ports_str = cast(
                            "Tuple[Optional[str], ...]", (p,) if p else (None,)
                        )

        if "port" in url.query:
            if integrated_multihost:
                raise exc.ArgumentError(
                    "Can't mix 'multihost' formats together; use "
                    '"host=h1,h2,h3&port=p1,p2,p3" or '
                    '"host=h1:p1&host=h2:p2&host=h3:p3" separately'
                )
            if isinstance(url.query["port"], (list, tuple)):
                ports_str = url.query["port"]
            elif isinstance(url.query["port"], str):
                ports_str = tuple(url.query["port"].split(","))

        ports: Optional[Tuple[Optional[int], ...]] = None

        if ports_str:
            try:
                ports = tuple(int(x) if x else None for x in ports_str)
            except ValueError:
                raise exc.ArgumentError(
                    f"Received non-integer port arguments: {ports_str}"
                ) from None

        if ports and (
            (not hosts and len(ports) > 1)
            or (
                hosts
                and ports
                and len(hosts) != len(ports)
                and (len(hosts) > 1 or len(ports) > 1)
            )
        ):
            raise exc.ArgumentError("number of hosts and ports don't match")

        if hosts is not None:
            if ports is None:
                ports = tuple(None for _ in hosts)

        return hosts, ports  # type: ignore

    def do_begin_twophase(self, connection, xid):
        self.do_begin(connection.connection)

    def do_prepare_twophase(self, connection, xid):
        connection.exec_driver_sql("PREPARE TRANSACTION '%s'" % xid)

    def do_rollback_twophase(
        self, connection, xid, is_prepared=True, recover=False
    ):
        if is_prepared:
            if recover:
                # FIXME: ugly hack to get out of transaction
                # context when committing recoverable transactions
                # Must find out a way how to make the dbapi not
                # open a transaction.
                connection.exec_driver_sql("ROLLBACK")
            connection.exec_driver_sql("ROLLBACK PREPARED '%s'" % xid)
            connection.exec_driver_sql("BEGIN")
            self.do_rollback(connection.connection)
        else:
            self.do_rollback(connection.connection)

    def do_commit_twophase(
        self, connection, xid, is_prepared=True, recover=False
    ):
        if is_prepared:
            if recover:
                connection.exec_driver_sql("ROLLBACK")
            connection.exec_driver_sql("COMMIT PREPARED '%s'" % xid)
            connection.exec_driver_sql("BEGIN")
            self.do_rollback(connection.connection)
        else:
            self.do_commit(connection.connection)

    def do_recover_twophase(self, connection):
        return connection.scalars(
            sql.text("SELECT gid FROM pg_prepared_xacts")
        ).all()

    def _get_default_schema_name(self, connection):
        return connection.exec_driver_sql("select current_schema()").scalar()

    @reflection.cache
    def has_schema(self, connection, schema, **kw):
        query = select(pg_catalog.pg_namespace.c.nspname).where(
            pg_catalog.pg_namespace.c.nspname == schema
        )
        return bool(connection.scalar(query))

    def _pg_class_filter_scope_schema(
        self, query, schema, scope, pg_class_table=None
    ):
        if pg_class_table is None:
            pg_class_table = pg_catalog.pg_class
        query = query.join(
            pg_catalog.pg_namespace,
            pg_catalog.pg_namespace.c.oid == pg_class_table.c.relnamespace,
        )

        if scope is ObjectScope.DEFAULT:
            query = query.where(pg_class_table.c.relpersistence != "t")
        elif scope is ObjectScope.TEMPORARY:
            query = query.where(pg_class_table.c.relpersistence == "t")

        if schema is None:
            query = query.where(
                pg_catalog.pg_table_is_visible(pg_class_table.c.oid),
                # ignore pg_catalog schema
                pg_catalog.pg_namespace.c.nspname != "pg_catalog",
            )
        else:
            query = query.where(pg_catalog.pg_namespace.c.nspname == schema)
        return query

    def _pg_class_relkind_condition(self, relkinds, pg_class_table=None):
        if pg_class_table is None:
            pg_class_table = pg_catalog.pg_class
        # uses the any form instead of in otherwise postgresql complaings
        # that 'IN could not convert type character to "char"'
        return pg_class_table.c.relkind == sql.any_(_array.array(relkinds))

    @lru_cache()
    def _has_table_query(self, schema):
        query = select(pg_catalog.pg_class.c.relname).where(
            pg_catalog.pg_class.c.relname == bindparam("table_name"),
            self._pg_class_relkind_condition(
                pg_catalog.RELKINDS_ALL_TABLE_LIKE
            ),
        )
        return self._pg_class_filter_scope_schema(
            query, schema, scope=ObjectScope.ANY
        )

    @reflection.cache
    def has_table(self, connection, table_name, schema=None, **kw):
        self._ensure_has_table_connection(connection)
        query = self._has_table_query(schema)
        return bool(connection.scalar(query, {"table_name": table_name}))

    @reflection.cache
    def has_sequence(self, connection, sequence_name, schema=None, **kw):
        query = select(pg_catalog.pg_class.c.relname).where(
            pg_catalog.pg_class.c.relkind == "S",
            pg_catalog.pg_class.c.relname == sequence_name,
        )
        query = self._pg_class_filter_scope_schema(
            query, schema, scope=ObjectScope.ANY
        )
        return bool(connection.scalar(query))

    @reflection.cache
    def has_type(self, connection, type_name, schema=None, **kw):
        query = (
            select(pg_catalog.pg_type.c.typname)
            .join(
                pg_catalog.pg_namespace,
                pg_catalog.pg_namespace.c.oid
                == pg_catalog.pg_type.c.typnamespace,
            )
            .where(pg_catalog.pg_type.c.typname == type_name)
        )
        if schema is None:
            query = query.where(
                pg_catalog.pg_type_is_visible(pg_catalog.pg_type.c.oid),
                # ignore pg_catalog schema
                pg_catalog.pg_namespace.c.nspname != "pg_catalog",
            )
        elif schema != "*":
            query = query.where(pg_catalog.pg_namespace.c.nspname == schema)

        return bool(connection.scalar(query))

    def _get_server_version_info(self, connection):
        v = connection.exec_driver_sql("select pg_catalog.version()").scalar()
        m = re.match(
            r".*(?:PostgreSQL|EnterpriseDB) "
            r"(\d+)\.?(\d+)?(?:\.(\d+))?(?:\.\d+)?(?:devel|beta)?",
            v,
        )
        if not m:
            raise AssertionError(
                "Could not determine version from string '%s'" % v
            )
        return tuple([int(x) for x in m.group(1, 2, 3) if x is not None])

    @reflection.cache
    def get_table_oid(self, connection, table_name, schema=None, **kw):
        """Fetch the oid for schema.table_name."""
        query = select(pg_catalog.pg_class.c.oid).where(
            pg_catalog.pg_class.c.relname == table_name,
            self._pg_class_relkind_condition(
                pg_catalog.RELKINDS_ALL_TABLE_LIKE
            ),
        )
        query = self._pg_class_filter_scope_schema(
            query, schema, scope=ObjectScope.ANY
        )
        table_oid = connection.scalar(query)
        if table_oid is None:
            raise exc.NoSuchTableError(
                f"{schema}.{table_name}" if schema else table_name
            )
        return table_oid

    @reflection.cache
    def get_schema_names(self, connection, **kw):
        query = (
            select(pg_catalog.pg_namespace.c.nspname)
            .where(pg_catalog.pg_namespace.c.nspname.not_like("pg_%"))
            .order_by(pg_catalog.pg_namespace.c.nspname)
        )
        return connection.scalars(query).all()

    def _get_relnames_for_relkinds(self, connection, schema, relkinds, scope):
        query = select(pg_catalog.pg_class.c.relname).where(
            self._pg_class_relkind_condition(relkinds)
        )
        query = self._pg_class_filter_scope_schema(query, schema, scope=scope)
        return connection.scalars(query).all()

    @reflection.cache
    def get_table_names(self, connection, schema=None, **kw):
        return self._get_relnames_for_relkinds(
            connection,
            schema,
            pg_catalog.RELKINDS_TABLE_NO_FOREIGN,
            scope=ObjectScope.DEFAULT,
        )

    @reflection.cache
    def get_temp_table_names(self, connection, **kw):
        return self._get_relnames_for_relkinds(
            connection,
            schema=None,
            relkinds=pg_catalog.RELKINDS_TABLE_NO_FOREIGN,
            scope=ObjectScope.TEMPORARY,
        )

    @reflection.cache
    def _get_foreign_table_names(self, connection, schema=None, **kw):
        return self._get_relnames_for_relkinds(
            connection, schema, relkinds=("f",), scope=ObjectScope.ANY
        )

    @reflection.cache
    def get_view_names(self, connection, schema=None, **kw):
        return self._get_relnames_for_relkinds(
            connection,
            schema,
            pg_catalog.RELKINDS_VIEW,
            scope=ObjectScope.DEFAULT,
        )

    @reflection.cache
    def get_materialized_view_names(self, connection, schema=None, **kw):
        return self._get_relnames_for_relkinds(
            connection,
            schema,
            pg_catalog.RELKINDS_MAT_VIEW,
            scope=ObjectScope.DEFAULT,
        )

    @reflection.cache
    def get_temp_view_names(self, connection, schema=None, **kw):
        return self._get_relnames_for_relkinds(
            connection,
            schema,
            # NOTE: do not include temp materialzied views (that do not
            # seem to be a thing at least up to version 14)
            pg_catalog.RELKINDS_VIEW,
            scope=ObjectScope.TEMPORARY,
        )

    @reflection.cache
    def get_sequence_names(self, connection, schema=None, **kw):
        return self._get_relnames_for_relkinds(
            connection, schema, relkinds=("S",), scope=ObjectScope.ANY
        )

    @reflection.cache
    def get_view_definition(self, connection, view_name, schema=None, **kw):
        query = (
            select(pg_catalog.pg_get_viewdef(pg_catalog.pg_class.c.oid))
            .select_from(pg_catalog.pg_class)
            .where(
                pg_catalog.pg_class.c.relname == view_name,
                self._pg_class_relkind_condition(
                    pg_catalog.RELKINDS_VIEW + pg_catalog.RELKINDS_MAT_VIEW
                ),
            )
        )
        query = self._pg_class_filter_scope_schema(
            query, schema, scope=ObjectScope.ANY
        )
        res = connection.scalar(query)
        if res is None:
            raise exc.NoSuchTableError(
                f"{schema}.{view_name}" if schema else view_name
            )
        else:
            return res

    def _value_or_raise(self, data, table, schema):
        try:
            return dict(data)[(schema, table)]
        except KeyError:
            raise exc.NoSuchTableError(
                f"{schema}.{table}" if schema else table
            ) from None

    def _prepare_filter_names(self, filter_names):
        if filter_names:
            return True, {"filter_names": filter_names}
        else:
            return False, {}

    def _kind_to_relkinds(self, kind: ObjectKind) -> Tuple[str, ...]:
        if kind is ObjectKind.ANY:
            return pg_catalog.RELKINDS_ALL_TABLE_LIKE
        relkinds = ()
        if ObjectKind.TABLE in kind:
            relkinds += pg_catalog.RELKINDS_TABLE
        if ObjectKind.VIEW in kind:
            relkinds += pg_catalog.RELKINDS_VIEW
        if ObjectKind.MATERIALIZED_VIEW in kind:
            relkinds += pg_catalog.RELKINDS_MAT_VIEW
        return relkinds

    @reflection.cache
    def get_columns(self, connection, table_name, schema=None, **kw):
        data = self.get_multi_columns(
            connection,
            schema=schema,
            filter_names=[table_name],
            scope=ObjectScope.ANY,
            kind=ObjectKind.ANY,
            **kw,
        )
        return self._value_or_raise(data, table_name, schema)

    @lru_cache()
    def _columns_query(self, schema, has_filter_names, scope, kind):
        # NOTE: the query with the default and identity options scalar
        # subquery is faster than trying to use outer joins for them
        generated = (
            pg_catalog.pg_attribute.c.attgenerated.label("generated")
            if self.server_version_info >= (12,)
            else sql.null().label("generated")
        )
        if self.server_version_info >= (10,):
            # join lateral performs worse (~2x slower) than a scalar_subquery
            identity = (
                select(
                    sql.func.json_build_object(
                        "always",
                        pg_catalog.pg_attribute.c.attidentity == "a",
                        "start",
                        pg_catalog.pg_sequence.c.seqstart,
                        "increment",
                        pg_catalog.pg_sequence.c.seqincrement,
                        "minvalue",
                        pg_catalog.pg_sequence.c.seqmin,
                        "maxvalue",
                        pg_catalog.pg_sequence.c.seqmax,
                        "cache",
                        pg_catalog.pg_sequence.c.seqcache,
                        "cycle",
                        pg_catalog.pg_sequence.c.seqcycle,
                        type_=sqltypes.JSON(),
                    )
                )
                .select_from(pg_catalog.pg_sequence)
                .where(
                    # attidentity != '' is required or it will reflect also
                    # serial columns as identity.
                    pg_catalog.pg_attribute.c.attidentity != "",
                    pg_catalog.pg_sequence.c.seqrelid
                    == sql.cast(
                        sql.cast(
                            pg_catalog.pg_get_serial_sequence(
                                sql.cast(
                                    sql.cast(
                                        pg_catalog.pg_attribute.c.attrelid,
                                        REGCLASS,
                                    ),
                                    TEXT,
                                ),
                                pg_catalog.pg_attribute.c.attname,
                            ),
                            REGCLASS,
                        ),
                        OID,
                    ),
                )
                .correlate(pg_catalog.pg_attribute)
                .scalar_subquery()
                .label("identity_options")
            )
        else:
            identity = sql.null().label("identity_options")

        # join lateral performs the same as scalar_subquery here
        default = (
            select(
                pg_catalog.pg_get_expr(
                    pg_catalog.pg_attrdef.c.adbin,
                    pg_catalog.pg_attrdef.c.adrelid,
                )
            )
            .select_from(pg_catalog.pg_attrdef)
            .where(
                pg_catalog.pg_attrdef.c.adrelid
                == pg_catalog.pg_attribute.c.attrelid,
                pg_catalog.pg_attrdef.c.adnum
                == pg_catalog.pg_attribute.c.attnum,
                pg_catalog.pg_attribute.c.atthasdef,
            )
            .correlate(pg_catalog.pg_attribute)
            .scalar_subquery()
            .label("default")
        )
        relkinds = self._kind_to_relkinds(kind)
        query = (
            select(
                pg_catalog.pg_attribute.c.attname.label("name"),
                pg_catalog.format_type(
                    pg_catalog.pg_attribute.c.atttypid,
                    pg_catalog.pg_attribute.c.atttypmod,
                ).label("format_type"),
                default,
                pg_catalog.pg_attribute.c.attnotnull.label("not_null"),
                pg_catalog.pg_class.c.relname.label("table_name"),
                pg_catalog.pg_description.c.description.label("comment"),
                generated,
                identity,
            )
            .select_from(pg_catalog.pg_class)
            # NOTE: postgresql support table with no user column, meaning
            # there is no row with pg_attribute.attnum > 0. use a left outer
            # join to avoid filtering these tables.
            .outerjoin(
                pg_catalog.pg_attribute,
                sql.and_(
                    pg_catalog.pg_class.c.oid
                    == pg_catalog.pg_attribute.c.attrelid,
                    pg_catalog.pg_attribute.c.attnum > 0,
                    ~pg_catalog.pg_attribute.c.attisdropped,
                ),
            )
            .outerjoin(
                pg_catalog.pg_description,
                sql.and_(
                    pg_catalog.pg_description.c.objoid
                    == pg_catalog.pg_attribute.c.attrelid,
                    pg_catalog.pg_description.c.objsubid
                    == pg_catalog.pg_attribute.c.attnum,
                ),
            )
            .where(self._pg_class_relkind_condition(relkinds))
            .order_by(
                pg_catalog.pg_class.c.relname, pg_catalog.pg_attribute.c.attnum
            )
        )
        query = self._pg_class_filter_scope_schema(query, schema, scope=scope)
        if has_filter_names:
            query = query.where(
                pg_catalog.pg_class.c.relname.in_(bindparam("filter_names"))
            )
        return query

    def get_multi_columns(
        self, connection, schema, filter_names, scope, kind, **kw
    ):
        has_filter_names, params = self._prepare_filter_names(filter_names)
        query = self._columns_query(schema, has_filter_names, scope, kind)
        rows = connection.execute(query, params).mappings()

        # dictionary with (name, ) if default search path or (schema, name)
        # as keys
        domains = {
            ((d["schema"], d["name"]) if not d["visible"] else (d["name"],)): d
            for d in self._load_domains(
                connection, schema="*", info_cache=kw.get("info_cache")
            )
        }

        # dictionary with (name, ) if default search path or (schema, name)
        # as keys
        enums = dict(
            (
                ((rec["name"],), rec)
                if rec["visible"]
                else ((rec["schema"], rec["name"]), rec)
            )
            for rec in self._load_enums(
                connection, schema="*", info_cache=kw.get("info_cache")
            )
        )

        columns = self._get_columns_info(rows, domains, enums, schema)

        return columns.items()

    _format_type_args_pattern = re.compile(r"\((.*)\)")
    _format_type_args_delim = re.compile(r"\s*,\s*")
    _format_array_spec_pattern = re.compile(r"((?:\[\])*)$")

    def _reflect_type(
        self,
        format_type: Optional[str],
        domains: Dict[str, ReflectedDomain],
        enums: Dict[str, ReflectedEnum],
        type_description: str,
    ) -> sqltypes.TypeEngine[Any]:
        """
        Attempts to reconstruct a column type defined in ischema_names based
        on the information available in the format_type.

        If the `format_type` cannot be associated with a known `ischema_names`,
        it is treated as a reference to a known PostgreSQL named `ENUM` or
        `DOMAIN` type.
        """
        type_description = type_description or "unknown type"
        if format_type is None:
            util.warn(
                "PostgreSQL format_type() returned NULL for %s"
                % type_description
            )
            return sqltypes.NULLTYPE

        attype_args_match = self._format_type_args_pattern.search(format_type)
        if attype_args_match and attype_args_match.group(1):
            attype_args = self._format_type_args_delim.split(
                attype_args_match.group(1)
            )
        else:
            attype_args = ()

        match_array_dim = self._format_array_spec_pattern.search(format_type)
        # Each "[]" in array specs corresponds to an array dimension
        array_dim = len(match_array_dim.group(1) or "") // 2

        # Remove all parameters and array specs from format_type to obtain an
        # ischema_name candidate
        attype = self._format_type_args_pattern.sub("", format_type)
        attype = self._format_array_spec_pattern.sub("", attype)

        schema_type = self.ischema_names.get(attype.lower(), None)
        args, kwargs = (), {}

        if attype == "numeric":
            if len(attype_args) == 2:
                precision, scale = map(int, attype_args)
                args = (precision, scale)

        elif attype == "double precision":
            args = (53,)

        elif attype == "integer":
            args = ()

        elif attype in ("timestamp with time zone", "time with time zone"):
            kwargs["timezone"] = True
            if len(attype_args) == 1:
                kwargs["precision"] = int(attype_args[0])

        elif attype in (
            "timestamp without time zone",
            "time without time zone",
            "time",
        ):
            kwargs["timezone"] = False
            if len(attype_args) == 1:
                kwargs["precision"] = int(attype_args[0])

        elif attype == "bit varying":
            kwargs["varying"] = True
            if len(attype_args) == 1:
                charlen = int(attype_args[0])
                args = (charlen,)

        # a domain or enum can start with interval, so be mindful of that.
        elif attype == "interval" or attype.startswith("interval "):
            schema_type = INTERVAL

            field_match = re.match(r"interval (.+)", attype)
            if field_match:
                kwargs["fields"] = field_match.group(1)

            if len(attype_args) == 1:
                kwargs["precision"] = int(attype_args[0])

        else:
            enum_or_domain_key = tuple(util.quoted_token_parser(attype))

            if enum_or_domain_key in enums:
                schema_type = ENUM
                enum = enums[enum_or_domain_key]

                kwargs["name"] = enum["name"]

                if not enum["visible"]:
                    kwargs["schema"] = enum["schema"]
                args = tuple(enum["labels"])
            elif enum_or_domain_key in domains:
                schema_type = DOMAIN
                domain = domains[enum_or_domain_key]

                data_type = self._reflect_type(
                    domain["type"],
                    domains,
                    enums,
                    type_description="DOMAIN '%s'" % domain["name"],
                )
                args = (domain["name"], data_type)

                kwargs["collation"] = domain["collation"]
                kwargs["default"] = domain["default"]
                kwargs["not_null"] = not domain["nullable"]
                kwargs["create_type"] = False

                if domain["constraints"]:
                    # We only support a single constraint
                    check_constraint = domain["constraints"][0]

                    kwargs["constraint_name"] = check_constraint["name"]
                    kwargs["check"] = check_constraint["check"]

                if not domain["visible"]:
                    kwargs["schema"] = domain["schema"]

            else:
                try:
                    charlen = int(attype_args[0])
                    args = (charlen, *attype_args[1:])
                except (ValueError, IndexError):
                    args = attype_args

        if not schema_type:
            util.warn(
                "Did not recognize type '%s' of %s"
                % (attype, type_description)
            )
            return sqltypes.NULLTYPE

        data_type = schema_type(*args, **kwargs)
        if array_dim >= 1:
            # postgres does not preserve dimensionality or size of array types.
            data_type = _array.ARRAY(data_type)

        return data_type

    def _get_columns_info(self, rows, domains, enums, schema):
        columns = defaultdict(list)
        for row_dict in rows:
            # ensure that each table has an entry, even if it has no columns
            if row_dict["name"] is None:
                columns[(schema, row_dict["table_name"])] = (
                    ReflectionDefaults.columns()
                )
                continue
            table_cols = columns[(schema, row_dict["table_name"])]

            coltype = self._reflect_type(
                row_dict["format_type"],
                domains,
                enums,
                type_description="column '%s'" % row_dict["name"],
            )

            default = row_dict["default"]
            name = row_dict["name"]
            generated = row_dict["generated"]
            nullable = not row_dict["not_null"]

            if isinstance(coltype, DOMAIN):
                if not default:
                    # domain can override the default value but
                    # cant set it to None
                    if coltype.default is not None:
                        default = coltype.default

                nullable = nullable and not coltype.not_null

            identity = row_dict["identity_options"]

            # If a zero byte or blank string depending on driver (is also
            # absent for older PG versions), then not a generated column.
            # Otherwise, s = stored. (Other values might be added in the
            # future.)
            if generated not in (None, "", b"\x00"):
                computed = dict(
                    sqltext=default, persisted=generated in ("s", b"s")
                )
                default = None
            else:
                computed = None

            # adjust the default value
            autoincrement = False
            if default is not None:
                match = re.search(r"""(nextval\(')([^']+)('.*$)""", default)
                if match is not None:
                    if issubclass(coltype._type_affinity, sqltypes.Integer):
                        autoincrement = True
                    # the default is related to a Sequence
                    if "." not in match.group(2) and schema is not None:
                        # unconditionally quote the schema name.  this could
                        # later be enhanced to obey quoting rules /
                        # "quote schema"
                        default = (
                            match.group(1)
                            + ('"%s"' % schema)
                            + "."
                            + match.group(2)
                            + match.group(3)
                        )

            column_info = {
                "name": name,
                "type": coltype,
                "nullable": nullable,
                "default": default,
                "autoincrement": autoincrement or identity is not None,
                "comment": row_dict["comment"],
            }
            if computed is not None:
                column_info["computed"] = computed
            if identity is not None:
                column_info["identity"] = identity

            table_cols.append(column_info)

        return columns

    @lru_cache()
    def _table_oids_query(self, schema, has_filter_names, scope, kind):
        relkinds = self._kind_to_relkinds(kind)
        oid_q = select(
            pg_catalog.pg_class.c.oid, pg_catalog.pg_class.c.relname
        ).where(self._pg_class_relkind_condition(relkinds))
        oid_q = self._pg_class_filter_scope_schema(oid_q, schema, scope=scope)

        if has_filter_names:
            oid_q = oid_q.where(
                pg_catalog.pg_class.c.relname.in_(bindparam("filter_names"))
            )
        return oid_q

    @reflection.flexi_cache(
        ("schema", InternalTraversal.dp_string),
        ("filter_names", InternalTraversal.dp_string_list),
        ("kind", InternalTraversal.dp_plain_obj),
        ("scope", InternalTraversal.dp_plain_obj),
    )
    def _get_table_oids(
        self, connection, schema, filter_names, scope, kind, **kw
    ):
        has_filter_names, params = self._prepare_filter_names(filter_names)
        oid_q = self._table_oids_query(schema, has_filter_names, scope, kind)
        result = connection.execute(oid_q, params)
        return result.all()

    @util.memoized_property
    def _constraint_query(self):
        if self.server_version_info >= (11, 0):
            indnkeyatts = pg_catalog.pg_index.c.indnkeyatts
        else:
            indnkeyatts = pg_catalog.pg_index.c.indnatts.label("indnkeyatts")

        if self.server_version_info >= (15,):
            indnullsnotdistinct = pg_catalog.pg_index.c.indnullsnotdistinct
        else:
            indnullsnotdistinct = sql.false().label("indnullsnotdistinct")

        con_sq = (
            select(
                pg_catalog.pg_constraint.c.conrelid,
                pg_catalog.pg_constraint.c.conname,
                sql.func.unnest(pg_catalog.pg_index.c.indkey).label("attnum"),
                sql.func.generate_subscripts(
                    pg_catalog.pg_index.c.indkey, 1
                ).label("ord"),
                indnkeyatts,
                indnullsnotdistinct,
                pg_catalog.pg_description.c.description,
            )
            .join(
                pg_catalog.pg_index,
                pg_catalog.pg_constraint.c.conindid
                == pg_catalog.pg_index.c.indexrelid,
            )
            .outerjoin(
                pg_catalog.pg_description,
                pg_catalog.pg_description.c.objoid
                == pg_catalog.pg_constraint.c.oid,
            )
            .where(
                pg_catalog.pg_constraint.c.contype == bindparam("contype"),
                pg_catalog.pg_constraint.c.conrelid.in_(bindparam("oids")),
                # NOTE: filtering also on pg_index.indrelid for oids does
                # not seem to have a performance effect, but it may be an
                # option if perf problems are reported
            )
            .subquery("con")
        )

        attr_sq = (
            select(
                con_sq.c.conrelid,
                con_sq.c.conname,
                con_sq.c.description,
                con_sq.c.ord,
                con_sq.c.indnkeyatts,
                con_sq.c.indnullsnotdistinct,
                pg_catalog.pg_attribute.c.attname,
            )
            .select_from(pg_catalog.pg_attribute)
            .join(
                con_sq,
                sql.and_(
                    pg_catalog.pg_attribute.c.attnum == con_sq.c.attnum,
                    pg_catalog.pg_attribute.c.attrelid == con_sq.c.conrelid,
                ),
            )
            .where(
                # NOTE: restate the condition here, since pg15 otherwise
                # seems to get confused on pscopg2 sometimes, doing
                # a sequential scan of pg_attribute.
                # The condition in the con_sq subquery is not actually needed
                # in pg15, but it may be needed in older versions. Keeping it
                # does not seems to have any inpact in any case.
                con_sq.c.conrelid.in_(bindparam("oids"))
            )
            .subquery("attr")
        )

        return (
            select(
                attr_sq.c.conrelid,
                sql.func.array_agg(
                    # NOTE: cast since some postgresql derivatives may
                    # not support array_agg on the name type
                    aggregate_order_by(
                        attr_sq.c.attname.cast(TEXT), attr_sq.c.ord
                    )
                ).label("cols"),
                attr_sq.c.conname,
                sql.func.min(attr_sq.c.description).label("description"),
                sql.func.min(attr_sq.c.indnkeyatts).label("indnkeyatts"),
                sql.func.bool_and(attr_sq.c.indnullsnotdistinct).label(
                    "indnullsnotdistinct"
                ),
            )
            .group_by(attr_sq.c.conrelid, attr_sq.c.conname)
            .order_by(attr_sq.c.conrelid, attr_sq.c.conname)
        )

    def _reflect_constraint(
        self, connection, contype, schema, filter_names, scope, kind, **kw
    ):
        # used to reflect primary and unique constraint
        table_oids = self._get_table_oids(
            connection, schema, filter_names, scope, kind, **kw
        )
        batches = list(table_oids)
        is_unique = contype == "u"

        while batches:
            batch = batches[0:3000]
            batches[0:3000] = []

            result = connection.execute(
                self._constraint_query,
                {"oids": [r[0] for r in batch], "contype": contype},
            ).mappings()

            result_by_oid = defaultdict(list)
            for row_dict in result:
                result_by_oid[row_dict["conrelid"]].append(row_dict)

            for oid, tablename in batch:
                for_oid = result_by_oid.get(oid, ())
                if for_oid:
                    for row in for_oid:
                        # See note in get_multi_indexes
                        all_cols = row["cols"]
                        indnkeyatts = row["indnkeyatts"]
                        if len(all_cols) > indnkeyatts:
                            inc_cols = all_cols[indnkeyatts:]
                            cst_cols = all_cols[:indnkeyatts]
                        else:
                            inc_cols = []
                            cst_cols = all_cols

                        opts = {}
                        if self.server_version_info >= (11,):
                            opts["postgresql_include"] = inc_cols
                        if is_unique:
                            opts["postgresql_nulls_not_distinct"] = row[
                                "indnullsnotdistinct"
                            ]
                        yield (
                            tablename,
                            cst_cols,
                            row["conname"],
                            row["description"],
                            opts,
                        )
                else:
                    yield tablename, None, None, None, None

    @reflection.cache
    def get_pk_constraint(self, connection, table_name, schema=None, **kw):
        data = self.get_multi_pk_constraint(
            connection,
            schema=schema,
            filter_names=[table_name],
            scope=ObjectScope.ANY,
            kind=ObjectKind.ANY,
            **kw,
        )
        return self._value_or_raise(data, table_name, schema)

    def get_multi_pk_constraint(
        self, connection, schema, filter_names, scope, kind, **kw
    ):
        result = self._reflect_constraint(
            connection, "p", schema, filter_names, scope, kind, **kw
        )

        # only a single pk can be present for each table. Return an entry
        # even if a table has no primary key
        default = ReflectionDefaults.pk_constraint

        def pk_constraint(pk_name, cols, comment, opts):
            info = {
                "constrained_columns": cols,
                "name": pk_name,
                "comment": comment,
            }
            if opts:
                info["dialect_options"] = opts
            return info

        return (
            (
                (schema, table_name),
                (
                    pk_constraint(pk_name, cols, comment, opts)
                    if pk_name is not None
                    else default()
                ),
            )
            for table_name, cols, pk_name, comment, opts in result
        )

    @reflection.cache
    def get_foreign_keys(
        self,
        connection,
        table_name,
        schema=None,
        postgresql_ignore_search_path=False,
        **kw,
    ):
        data = self.get_multi_foreign_keys(
            connection,
            schema=schema,
            filter_names=[table_name],
            postgresql_ignore_search_path=postgresql_ignore_search_path,
            scope=ObjectScope.ANY,
            kind=ObjectKind.ANY,
            **kw,
        )
        return self._value_or_raise(data, table_name, schema)

    @lru_cache()
    def _foreing_key_query(self, schema, has_filter_names, scope, kind):
        pg_class_ref = pg_catalog.pg_class.alias("cls_ref")
        pg_namespace_ref = pg_catalog.pg_namespace.alias("nsp_ref")
        relkinds = self._kind_to_relkinds(kind)
        query = (
            select(
                pg_catalog.pg_class.c.relname,
                pg_catalog.pg_constraint.c.conname,
                # NOTE: avoid calling pg_get_constraintdef when not needed
                # to speed up the query
                sql.case(
                    (
                        pg_catalog.pg_constraint.c.oid.is_not(None),
                        pg_catalog.pg_get_constraintdef(
                            pg_catalog.pg_constraint.c.oid, True
                        ),
                    ),
                    else_=None,
                ),
                pg_namespace_ref.c.nspname,
                pg_catalog.pg_description.c.description,
            )
            .select_from(pg_catalog.pg_class)
            .outerjoin(
                pg_catalog.pg_constraint,
                sql.and_(
                    pg_catalog.pg_class.c.oid
                    == pg_catalog.pg_constraint.c.conrelid,
                    pg_catalog.pg_constraint.c.contype == "f",
                ),
            )
            .outerjoin(
                pg_class_ref,
                pg_class_ref.c.oid == pg_catalog.pg_constraint.c.confrelid,
            )
            .outerjoin(
                pg_namespace_ref,
                pg_class_ref.c.relnamespace == pg_namespace_ref.c.oid,
            )
            .outerjoin(
                pg_catalog.pg_description,
                pg_catalog.pg_description.c.objoid
                == pg_catalog.pg_constraint.c.oid,
            )
            .order_by(
                pg_catalog.pg_class.c.relname,
                pg_catalog.pg_constraint.c.conname,
            )
            .where(self._pg_class_relkind_condition(relkinds))
        )
        query = self._pg_class_filter_scope_schema(query, schema, scope)
        if has_filter_names:
            query = query.where(
                pg_catalog.pg_class.c.relname.in_(bindparam("filter_names"))
            )
        return query

    @util.memoized_property
    def _fk_regex_pattern(self):
        # optionally quoted token
        qtoken = '(?:"[^"]+"|[A-Za-z0-9_]+?)'

        # https://www.postgresql.org/docs/current/static/sql-createtable.html
        return re.compile(
            r"FOREIGN KEY \((.*?)\) "
            rf"REFERENCES (?:({qtoken})\.)?({qtoken})\(((?:{qtoken}(?: *, *)?)+)\)"  # noqa: E501
            r"[\s]?(MATCH (FULL|PARTIAL|SIMPLE)+)?"
            r"[\s]?(ON UPDATE "
            r"(CASCADE|RESTRICT|NO ACTION|SET NULL|SET DEFAULT)+)?"
            r"[\s]?(ON DELETE "
            r"(CASCADE|RESTRICT|NO ACTION|"
            r"SET (?:NULL|DEFAULT)(?:\s\(.+\))?)+)?"
            r"[\s]?(DEFERRABLE|NOT DEFERRABLE)?"
            r"[\s]?(INITIALLY (DEFERRED|IMMEDIATE)+)?"
        )

    def get_multi_foreign_keys(
        self,
        connection,
        schema,
        filter_names,
        scope,
        kind,
        postgresql_ignore_search_path=False,
        **kw,
    ):
        preparer = self.identifier_preparer

        has_filter_names, params = self._prepare_filter_names(filter_names)
        query = self._foreing_key_query(schema, has_filter_names, scope, kind)
        result = connection.execute(query, params)

        FK_REGEX = self._fk_regex_pattern

        fkeys = defaultdict(list)
        default = ReflectionDefaults.foreign_keys
        for table_name, conname, condef, conschema, comment in result:
            # ensure that each table has an entry, even if it has
            # no foreign keys
            if conname is None:
                fkeys[(schema, table_name)] = default()
                continue
            table_fks = fkeys[(schema, table_name)]
            m = re.search(FK_REGEX, condef).groups()

            (
                constrained_columns,
                referred_schema,
                referred_table,
                referred_columns,
                _,
                match,
                _,
                onupdate,
                _,
                ondelete,
                deferrable,
                _,
                initially,
            ) = m

            if deferrable is not None:
                deferrable = True if deferrable == "DEFERRABLE" else False
            constrained_columns = [
                preparer._unquote_identifier(x)
                for x in re.split(r"\s*,\s*", constrained_columns)
            ]

            if postgresql_ignore_search_path:
                # when ignoring search path, we use the actual schema
                # provided it isn't the "default" schema
                if conschema != self.default_schema_name:
                    referred_schema = conschema
                else:
                    referred_schema = schema
            elif referred_schema:
                # referred_schema is the schema that we regexp'ed from
                # pg_get_constraintdef().  If the schema is in the search
                # path, pg_get_constraintdef() will give us None.
                referred_schema = preparer._unquote_identifier(referred_schema)
            elif schema is not None and schema == conschema:
                # If the actual schema matches the schema of the table
                # we're reflecting, then we will use that.
                referred_schema = schema

            referred_table = preparer._unquote_identifier(referred_table)
            referred_columns = [
                preparer._unquote_identifier(x)
                for x in re.split(r"\s*,\s", referred_columns)
            ]
            options = {
                k: v
                for k, v in [
                    ("onupdate", onupdate),
                    ("ondelete", ondelete),
                    ("initially", initially),
                    ("deferrable", deferrable),
                    ("match", match),
                ]
                if v is not None and v != "NO ACTION"
            }
            fkey_d = {
                "name": conname,
                "constrained_columns": constrained_columns,
                "referred_schema": referred_schema,
                "referred_table": referred_table,
                "referred_columns": referred_columns,
                "options": options,
                "comment": comment,
            }
            table_fks.append(fkey_d)
        return fkeys.items()

    @reflection.cache
    def get_indexes(self, connection, table_name, schema=None, **kw):
        data = self.get_multi_indexes(
            connection,
            schema=schema,
            filter_names=[table_name],
            scope=ObjectScope.ANY,
            kind=ObjectKind.ANY,
            **kw,
        )
        return self._value_or_raise(data, table_name, schema)

    @util.memoized_property
    def _index_query(self):
        # NOTE: pg_index is used as from two times to improve performance,
        # since extraing all the index information from `idx_sq` to avoid
        # the second pg_index use leads to a worse performing query in
        # particular when querying for a single table (as of pg 17)
        # NOTE: repeating oids clause improve query performance

        # subquery to get the columns
        idx_sq = (
            select(
                pg_catalog.pg_index.c.indexrelid,
                pg_catalog.pg_index.c.indrelid,
                sql.func.unnest(pg_catalog.pg_index.c.indkey).label("attnum"),
                sql.func.unnest(pg_catalog.pg_index.c.indclass).label(
                    "att_opclass"
                ),
                sql.func.generate_subscripts(
                    pg_catalog.pg_index.c.indkey, 1
                ).label("ord"),
            )
            .where(
                ~pg_catalog.pg_index.c.indisprimary,
                pg_catalog.pg_index.c.indrelid.in_(bindparam("oids")),
            )
            .subquery("idx")
        )

        attr_sq = (
            select(
                idx_sq.c.indexrelid,
                idx_sq.c.indrelid,
                idx_sq.c.ord,
                # NOTE: always using pg_get_indexdef is too slow so just
                # invoke when the element is an expression
                sql.case(
                    (
                        idx_sq.c.attnum == 0,
                        pg_catalog.pg_get_indexdef(
                            idx_sq.c.indexrelid, idx_sq.c.ord + 1, True
                        ),
                    ),
                    # NOTE: need to cast this since attname is of type "name"
                    # that's limited to 63 bytes, while pg_get_indexdef
                    # returns "text" so its output may get cut
                    else_=pg_catalog.pg_attribute.c.attname.cast(TEXT),
                ).label("element"),
                (idx_sq.c.attnum == 0).label("is_expr"),
                pg_catalog.pg_opclass.c.opcname,
                pg_catalog.pg_opclass.c.opcdefault,
            )
            .select_from(idx_sq)
            .outerjoin(
                # do not remove rows where idx_sq.c.attnum is 0
                pg_catalog.pg_attribute,
                sql.and_(
                    pg_catalog.pg_attribute.c.attnum == idx_sq.c.attnum,
                    pg_catalog.pg_attribute.c.attrelid == idx_sq.c.indrelid,
                ),
            )
            .outerjoin(
                pg_catalog.pg_opclass,
                pg_catalog.pg_opclass.c.oid == idx_sq.c.att_opclass,
            )
            .where(idx_sq.c.indrelid.in_(bindparam("oids")))
            .subquery("idx_attr")
        )

        cols_sq = (
            select(
                attr_sq.c.indexrelid,
                sql.func.min(attr_sq.c.indrelid),
                sql.func.array_agg(
                    aggregate_order_by(attr_sq.c.element, attr_sq.c.ord)
                ).label("elements"),
                sql.func.array_agg(
                    aggregate_order_by(attr_sq.c.is_expr, attr_sq.c.ord)
                ).label("elements_is_expr"),
                sql.func.array_agg(
                    aggregate_order_by(attr_sq.c.opcname, attr_sq.c.ord)
                ).label("elements_opclass"),
                sql.func.array_agg(
                    aggregate_order_by(attr_sq.c.opcdefault, attr_sq.c.ord)
                ).label("elements_opdefault"),
            )
            .group_by(attr_sq.c.indexrelid)
            .subquery("idx_cols")
        )

        if self.server_version_info >= (11, 0):
            indnkeyatts = pg_catalog.pg_index.c.indnkeyatts
        else:
            indnkeyatts = pg_catalog.pg_index.c.indnatts.label("indnkeyatts")

        if self.server_version_info >= (15,):
            nulls_not_distinct = pg_catalog.pg_index.c.indnullsnotdistinct
        else:
            nulls_not_distinct = sql.false().label("indnullsnotdistinct")

        return (
            select(
                pg_catalog.pg_index.c.indrelid,
                pg_catalog.pg_class.c.relname,
                pg_catalog.pg_index.c.indisunique,
                pg_catalog.pg_constraint.c.conrelid.is_not(None).label(
                    "has_constraint"
                ),
                pg_catalog.pg_index.c.indoption,
                pg_catalog.pg_class.c.reloptions,
                pg_catalog.pg_am.c.amname,
                # NOTE: pg_get_expr is very fast so this case has almost no
                # performance impact
                sql.case(
                    (
                        pg_catalog.pg_index.c.indpred.is_not(None),
                        pg_catalog.pg_get_expr(
                            pg_catalog.pg_index.c.indpred,
                            pg_catalog.pg_index.c.indrelid,
                        ),
                    ),
                    else_=None,
                ).label("filter_definition"),
                indnkeyatts,
                nulls_not_distinct,
                cols_sq.c.elements,
                cols_sq.c.elements_is_expr,
                cols_sq.c.elements_opclass,
                cols_sq.c.elements_opdefault,
            )
            .select_from(pg_catalog.pg_index)
            .where(
                pg_catalog.pg_index.c.indrelid.in_(bindparam("oids")),
                ~pg_catalog.pg_index.c.indisprimary,
            )
            .join(
                pg_catalog.pg_class,
                pg_catalog.pg_index.c.indexrelid == pg_catalog.pg_class.c.oid,
            )
            .join(
                pg_catalog.pg_am,
                pg_catalog.pg_class.c.relam == pg_catalog.pg_am.c.oid,
            )
            .outerjoin(
                cols_sq,
                pg_catalog.pg_index.c.indexrelid == cols_sq.c.indexrelid,
            )
            .outerjoin(
                pg_catalog.pg_constraint,
                sql.and_(
                    pg_catalog.pg_index.c.indrelid
                    == pg_catalog.pg_constraint.c.conrelid,
                    pg_catalog.pg_index.c.indexrelid
                    == pg_catalog.pg_constraint.c.conindid,
                    pg_catalog.pg_constraint.c.contype
                    == sql.any_(_array.array(("p", "u", "x"))),
                ),
            )
            .order_by(
                pg_catalog.pg_index.c.indrelid, pg_catalog.pg_class.c.relname
            )
        )

    def get_multi_indexes(
        self, connection, schema, filter_names, scope, kind, **kw
    ):
        table_oids = self._get_table_oids(
            connection, schema, filter_names, scope, kind, **kw
        )

        indexes = defaultdict(list)
        default = ReflectionDefaults.indexes

        batches = list(table_oids)

        while batches:
            batch = batches[0:3000]
            batches[0:3000] = []

            result = connection.execute(
                self._index_query, {"oids": [r[0] for r in batch]}
            ).mappings()

            result_by_oid = defaultdict(list)
            for row_dict in result:
                result_by_oid[row_dict["indrelid"]].append(row_dict)

            for oid, table_name in batch:
                if oid not in result_by_oid:
                    # ensure that each table has an entry, even if reflection
                    # is skipped because not supported
                    indexes[(schema, table_name)] = default()
                    continue

                for row in result_by_oid[oid]:
                    index_name = row["relname"]

                    table_indexes = indexes[(schema, table_name)]

                    all_elements = row["elements"]
                    all_elements_is_expr = row["elements_is_expr"]
                    all_elements_opclass = row["elements_opclass"]
                    all_elements_opdefault = row["elements_opdefault"]
                    indnkeyatts = row["indnkeyatts"]
                    # "The number of key columns in the index, not counting any
                    # included columns, which are merely stored and do not
                    # participate in the index semantics"
                    if len(all_elements) > indnkeyatts:
                        # this is a "covering index" which has INCLUDE columns
                        # as well as regular index columns
                        inc_cols = all_elements[indnkeyatts:]
                        idx_elements = all_elements[:indnkeyatts]
                        idx_elements_is_expr = all_elements_is_expr[
                            :indnkeyatts
                        ]
                        # postgresql does not support expression on included
                        # columns as of v14: "ERROR: expressions are not
                        # supported in included columns".
                        assert all(
                            not is_expr
                            for is_expr in all_elements_is_expr[indnkeyatts:]
                        )
                        idx_elements_opclass = all_elements_opclass[
                            :indnkeyatts
                        ]
                        idx_elements_opdefault = all_elements_opdefault[
                            :indnkeyatts
                        ]
                    else:
                        idx_elements = all_elements
                        idx_elements_is_expr = all_elements_is_expr
                        inc_cols = []
                        idx_elements_opclass = all_elements_opclass
                        idx_elements_opdefault = all_elements_opdefault

                    index = {"name": index_name, "unique": row["indisunique"]}
                    if any(idx_elements_is_expr):
                        index["column_names"] = [
                            None if is_expr else expr
                            for expr, is_expr in zip(
                                idx_elements, idx_elements_is_expr
                            )
                        ]
                        index["expressions"] = idx_elements
                    else:
                        index["column_names"] = idx_elements

                    dialect_options = {}

                    if not all(idx_elements_opdefault):
                        dialect_options["postgresql_ops"] = {
                            name: opclass
                            for name, opclass, is_default in zip(
                                idx_elements,
                                idx_elements_opclass,
                                idx_elements_opdefault,
                            )
                            if not is_default
                        }

                    sorting = {}
                    for col_index, col_flags in enumerate(row["indoption"]):
                        col_sorting = ()
                        # try to set flags only if they differ from PG
                        # defaults...
                        if col_flags & 0x01:
                            col_sorting += ("desc",)
                            if not (col_flags & 0x02):
                                col_sorting += ("nulls_last",)
                        else:
                            if col_flags & 0x02:
                                col_sorting += ("nulls_first",)
                        if col_sorting:
                            sorting[idx_elements[col_index]] = col_sorting
                    if sorting:
                        index["column_sorting"] = sorting
                    if row["has_constraint"]:
                        index["duplicates_constraint"] = index_name

                    if row["reloptions"]:
                        dialect_options["postgresql_with"] = dict(
                            [
                                option.split("=", 1)
                                for option in row["reloptions"]
                            ]
                        )
                    # it *might* be nice to include that this is 'btree' in the
                    # reflection info.  But we don't want an Index object
                    # to have a ``postgresql_using`` in it that is just the
                    # default, so for the moment leaving this out.
                    amname = row["amname"]
                    if amname != "btree":
                        dialect_options["postgresql_using"] = row["amname"]
                    if row["filter_definition"]:
                        dialect_options["postgresql_where"] = row[
                            "filter_definition"
                        ]
                    if self.server_version_info >= (11,):
                        # NOTE: this is legacy, this is part of
                        # dialect_options now as of #7382
                        index["include_columns"] = inc_cols
                        dialect_options["postgresql_include"] = inc_cols
                    if row["indnullsnotdistinct"]:
                        # the default is False, so ignore it.
                        dialect_options["postgresql_nulls_not_distinct"] = row[
                            "indnullsnotdistinct"
                        ]

                    if dialect_options:
                        index["dialect_options"] = dialect_options

                    table_indexes.append(index)
        return indexes.items()

    @reflection.cache
    def get_unique_constraints(
        self, connection, table_name, schema=None, **kw
    ):
        data = self.get_multi_unique_constraints(
            connection,
            schema=schema,
            filter_names=[table_name],
            scope=ObjectScope.ANY,
            kind=ObjectKind.ANY,
            **kw,
        )
        return self._value_or_raise(data, table_name, schema)

    def get_multi_unique_constraints(
        self,
        connection,
        schema,
        filter_names,
        scope,
        kind,
        **kw,
    ):
        result = self._reflect_constraint(
            connection, "u", schema, filter_names, scope, kind, **kw
        )

        # each table can have multiple unique constraints
        uniques = defaultdict(list)
        default = ReflectionDefaults.unique_constraints
        for table_name, cols, con_name, comment, options in result:
            # ensure a list is created for each table. leave it empty if
            # the table has no unique cosntraint
            if con_name is None:
                uniques[(schema, table_name)] = default()
                continue

            uc_dict = {
                "column_names": cols,
                "name": con_name,
                "comment": comment,
            }
            if options:
                uc_dict["dialect_options"] = options

            uniques[(schema, table_name)].append(uc_dict)
        return uniques.items()

    @reflection.cache
    def get_table_comment(self, connection, table_name, schema=None, **kw):
        data = self.get_multi_table_comment(
            connection,
            schema,
            [table_name],
            scope=ObjectScope.ANY,
            kind=ObjectKind.ANY,
            **kw,
        )
        return self._value_or_raise(data, table_name, schema)

    @lru_cache()
    def _comment_query(self, schema, has_filter_names, scope, kind):
        relkinds = self._kind_to_relkinds(kind)
        query = (
            select(
                pg_catalog.pg_class.c.relname,
                pg_catalog.pg_description.c.description,
            )
            .select_from(pg_catalog.pg_class)
            .outerjoin(
                pg_catalog.pg_description,
                sql.and_(
                    pg_catalog.pg_class.c.oid
                    == pg_catalog.pg_description.c.objoid,
                    pg_catalog.pg_description.c.objsubid == 0,
                    pg_catalog.pg_description.c.classoid
                    == sql.func.cast("pg_catalog.pg_class", REGCLASS),
                ),
            )
            .where(self._pg_class_relkind_condition(relkinds))
        )
        query = self._pg_class_filter_scope_schema(query, schema, scope)
        if has_filter_names:
            query = query.where(
                pg_catalog.pg_class.c.relname.in_(bindparam("filter_names"))
            )
        return query

    def get_multi_table_comment(
        self, connection, schema, filter_names, scope, kind, **kw
    ):
        has_filter_names, params = self._prepare_filter_names(filter_names)
        query = self._comment_query(schema, has_filter_names, scope, kind)
        result = connection.execute(query, params)

        default = ReflectionDefaults.table_comment
        return (
            (
                (schema, table),
                {"text": comment} if comment is not None else default(),
            )
            for table, comment in result
        )

    @reflection.cache
    def get_check_constraints(self, connection, table_name, schema=None, **kw):
        data = self.get_multi_check_constraints(
            connection,
            schema,
            [table_name],
            scope=ObjectScope.ANY,
            kind=ObjectKind.ANY,
            **kw,
        )
        return self._value_or_raise(data, table_name, schema)

    @lru_cache()
    def _check_constraint_query(self, schema, has_filter_names, scope, kind):
        relkinds = self._kind_to_relkinds(kind)
        query = (
            select(
                pg_catalog.pg_class.c.relname,
                pg_catalog.pg_constraint.c.conname,
                # NOTE: avoid calling pg_get_constraintdef when not needed
                # to speed up the query
                sql.case(
                    (
                        pg_catalog.pg_constraint.c.oid.is_not(None),
                        pg_catalog.pg_get_constraintdef(
                            pg_catalog.pg_constraint.c.oid, True
                        ),
                    ),
                    else_=None,
                ),
                pg_catalog.pg_description.c.description,
            )
            .select_from(pg_catalog.pg_class)
            .outerjoin(
                pg_catalog.pg_constraint,
                sql.and_(
                    pg_catalog.pg_class.c.oid
                    == pg_catalog.pg_constraint.c.conrelid,
                    pg_catalog.pg_constraint.c.contype == "c",
                ),
            )
            .outerjoin(
                pg_catalog.pg_description,
                pg_catalog.pg_description.c.objoid
                == pg_catalog.pg_constraint.c.oid,
            )
            .order_by(
                pg_catalog.pg_class.c.relname,
                pg_catalog.pg_constraint.c.conname,
            )
            .where(self._pg_class_relkind_condition(relkinds))
        )
        query = self._pg_class_filter_scope_schema(query, schema, scope)
        if has_filter_names:
            query = query.where(
                pg_catalog.pg_class.c.relname.in_(bindparam("filter_names"))
            )
        return query

    def get_multi_check_constraints(
        self, connection, schema, filter_names, scope, kind, **kw
    ):
        has_filter_names, params = self._prepare_filter_names(filter_names)
        query = self._check_constraint_query(
            schema, has_filter_names, scope, kind
        )
        result = connection.execute(query, params)

        check_constraints = defaultdict(list)
        default = ReflectionDefaults.check_constraints
        for table_name, check_name, src, comment in result:
            # only two cases for check_name and src: both null or both defined
            if check_name is None and src is None:
                check_constraints[(schema, table_name)] = default()
                continue
            # samples:
            # "CHECK (((a > 1) AND (a < 5)))"
            # "CHECK (((a = 1) OR ((a > 2) AND (a < 5))))"
            # "CHECK (((a > 1) AND (a < 5))) NOT VALID"
            # "CHECK (some_boolean_function(a))"
            # "CHECK (((a\n < 1)\n OR\n (a\n >= 5))\n)"
            # "CHECK (a NOT NULL) NO INHERIT"
            # "CHECK (a NOT NULL) NO INHERIT NOT VALID"

            m = re.match(
                r"^CHECK *\((.+)\)( NO INHERIT)?( NOT VALID)?$",
                src,
                flags=re.DOTALL,
            )
            if not m:
                util.warn("Could not parse CHECK constraint text: %r" % src)
                sqltext = ""
            else:
                sqltext = re.compile(
                    r"^[\s\n]*\((.+)\)[\s\n]*$", flags=re.DOTALL
                ).sub(r"\1", m.group(1))
            entry = {
                "name": check_name,
                "sqltext": sqltext,
                "comment": comment,
            }
            if m:
                do = {}
                if " NOT VALID" in m.groups():
                    do["not_valid"] = True
                if " NO INHERIT" in m.groups():
                    do["no_inherit"] = True
                if do:
                    entry["dialect_options"] = do

            check_constraints[(schema, table_name)].append(entry)
        return check_constraints.items()

    def _pg_type_filter_schema(self, query, schema):
        if schema is None:
            query = query.where(
                pg_catalog.pg_type_is_visible(pg_catalog.pg_type.c.oid),
                # ignore pg_catalog schema
                pg_catalog.pg_namespace.c.nspname != "pg_catalog",
            )
        elif schema != "*":
            query = query.where(pg_catalog.pg_namespace.c.nspname == schema)
        return query

    @lru_cache()
    def _enum_query(self, schema):
        lbl_agg_sq = (
            select(
                pg_catalog.pg_enum.c.enumtypid,
                sql.func.array_agg(
                    aggregate_order_by(
                        # NOTE: cast since some postgresql derivatives may
                        # not support array_agg on the name type
                        pg_catalog.pg_enum.c.enumlabel.cast(TEXT),
                        pg_catalog.pg_enum.c.enumsortorder,
                    )
                ).label("labels"),
            )
            .group_by(pg_catalog.pg_enum.c.enumtypid)
            .subquery("lbl_agg")
        )

        query = (
            select(
                pg_catalog.pg_type.c.typname.label("name"),
                pg_catalog.pg_type_is_visible(pg_catalog.pg_type.c.oid).label(
                    "visible"
                ),
                pg_catalog.pg_namespace.c.nspname.label("schema"),
                lbl_agg_sq.c.labels.label("labels"),
            )
            .join(
                pg_catalog.pg_namespace,
                pg_catalog.pg_namespace.c.oid
                == pg_catalog.pg_type.c.typnamespace,
            )
            .outerjoin(
                lbl_agg_sq, pg_catalog.pg_type.c.oid == lbl_agg_sq.c.enumtypid
            )
            .where(pg_catalog.pg_type.c.typtype == "e")
            .order_by(
                pg_catalog.pg_namespace.c.nspname, pg_catalog.pg_type.c.typname
            )
        )

        return self._pg_type_filter_schema(query, schema)

    @reflection.cache
    def _load_enums(self, connection, schema=None, **kw):
        if not self.supports_native_enum:
            return []

        result = connection.execute(self._enum_query(schema))

        enums = []
        for name, visible, schema, labels in result:
            enums.append(
                {
                    "name": name,
                    "schema": schema,
                    "visible": visible,
                    "labels": [] if labels is None else labels,
                }
            )
        return enums

    @lru_cache()
    def _domain_query(self, schema):
        con_sq = (
            select(
                pg_catalog.pg_constraint.c.contypid,
                sql.func.array_agg(
                    pg_catalog.pg_get_constraintdef(
                        pg_catalog.pg_constraint.c.oid, True
                    )
                ).label("condefs"),
                sql.func.array_agg(
                    # NOTE: cast since some postgresql derivatives may
                    # not support array_agg on the name type
                    pg_catalog.pg_constraint.c.conname.cast(TEXT)
                ).label("connames"),
            )
            # The domain this constraint is on; zero if not a domain constraint
            .where(pg_catalog.pg_constraint.c.contypid != 0)
            .group_by(pg_catalog.pg_constraint.c.contypid)
            .subquery("domain_constraints")
        )

        query = (
            select(
                pg_catalog.pg_type.c.typname.label("name"),
                pg_catalog.format_type(
                    pg_catalog.pg_type.c.typbasetype,
                    pg_catalog.pg_type.c.typtypmod,
                ).label("attype"),
                (~pg_catalog.pg_type.c.typnotnull).label("nullable"),
                pg_catalog.pg_type.c.typdefault.label("default"),
                pg_catalog.pg_type_is_visible(pg_catalog.pg_type.c.oid).label(
                    "visible"
                ),
                pg_catalog.pg_namespace.c.nspname.label("schema"),
                con_sq.c.condefs,
                con_sq.c.connames,
                pg_catalog.pg_collation.c.collname,
            )
            .join(
                pg_catalog.pg_namespace,
                pg_catalog.pg_namespace.c.oid
                == pg_catalog.pg_type.c.typnamespace,
            )
            .outerjoin(
                pg_catalog.pg_collation,
                pg_catalog.pg_type.c.typcollation
                == pg_catalog.pg_collation.c.oid,
            )
            .outerjoin(
                con_sq,
                pg_catalog.pg_type.c.oid == con_sq.c.contypid,
            )
            .where(pg_catalog.pg_type.c.typtype == "d")
            .order_by(
                pg_catalog.pg_namespace.c.nspname, pg_catalog.pg_type.c.typname
            )
        )
        return self._pg_type_filter_schema(query, schema)

    @reflection.cache
    def _load_domains(self, connection, schema=None, **kw):
        result = connection.execute(self._domain_query(schema))

        domains: List[ReflectedDomain] = []
        for domain in result.mappings():
            # strip (30) from character varying(30)
            attype = re.search(r"([^\(]+)", domain["attype"]).group(1)
            constraints: List[ReflectedDomainConstraint] = []
            if domain["connames"]:
                # When a domain has multiple CHECK constraints, they will
                # be tested in alphabetical order by name.
                sorted_constraints = sorted(
                    zip(domain["connames"], domain["condefs"]),
                    key=lambda t: t[0],
                )
                for name, def_ in sorted_constraints:
                    # constraint is in the form "CHECK (expression)"
                    # or "NOT NULL". Ignore the "NOT NULL" and
                    # remove "CHECK (" and the tailing ")".
                    if def_.casefold().startswith("check"):
                        check = def_[7:-1]
                        constraints.append({"name": name, "check": check})
            domain_rec: ReflectedDomain = {
                "name": domain["name"],
                "schema": domain["schema"],
                "visible": domain["visible"],
                "type": attype,
                "nullable": domain["nullable"],
                "default": domain["default"],
                "constraints": constraints,
                "collation": domain["collname"],
            }
            domains.append(domain_rec)

        return domains

    def _set_backslash_escapes(self, connection):
        # this method is provided as an override hook for descendant
        # dialects (e.g. Redshift), so removing it may break them
        std_string = connection.exec_driver_sql(
            "show standard_conforming_strings"
        ).scalar()
        self._backslash_escapes = std_string == "off"