Made some adjustments to :class:`.Pool` and :class:`.Connection` such
that recovery logic is not run underneath exception catches for
``pool.Empty``, ``AttributeError``, since when the recovery operation
itself fails, Python 3 creates a misleading stack trace referring to the
``Empty`` / ``AttributeError`` as the cause, when in fact these exception
catches are part of control flow.

Change-Id: Id3ed9a8f96ce4ccb4009c94af30ddc2ddb9818b9
Fixes: #4028

Changed the mechanics of :class:`.ResultProxy` to unconditionally
delay the "autoclose" step until the :class:`.Connection` is done
with the object; in the case where Postgresql ON CONFLICT with
RETURNING returns no rows, autoclose was occurring in this previously
non-existent use case, causing the usual autocommit behavior that
occurs unconditionally upon INSERT/UPDATE/DELETE to fail.

Change-Id: I235a25daf4381b31f523331f810ea04450349722
Fixes: #3955
(cherry picked from commit 8ee363e4917b0dcd64a83b6d26e465c9e61e0ea5)
(cherry picked from commit f52fb5282a046d26b6ee2778e03b995eb117c2ee)

Added an exception handler that will warn for the "cause" exception on
Py2K when the "autorollback" feature of :class:`.Connection` itself
raises an exception. In Py3K, the two exceptions are naturally reported
by the interpreter as one occurring during the handling of the other.
This is continuing with the series of changes for rollback failure
handling that were last visited as part of :ticket:`2696` in 1.0.12.

Change-Id: I600ba455a14ebaea27c6189889181f97c632f179
Fixes: #3946

Corrects some warnings and adds tox config.  Adds DeprecationWarning
to the error category.   Large sweep for string literals w/ backslashes
as this is common in docstrings

Co-authored-by: Andrii Soldatenko
Fixes: #3886
Change-Id: Ia7c838dfbbe70b262622ed0803d581edc736e085
Pull-request: https://github.com/zzzeek/sqlalchemy/pull/337

Change-Id: I4e8c2aa8fe817bb2af8707410fa0201f938781de

This allows to skip buffering of the results on the client side, e.g.
the following snippet:

    table = sa.Table(
        'testtbl', sa.MetaData(),
        sa.Column('id', sa.Integer, primary_key=True),
        sa.Column('a', sa.Integer),
        sa.Column('b', sa.String(512))
    )
    table.create(eng, checkfirst=True)

    with eng.connect() as conn:
        result = conn.execute(table.select().limit(1)).fetchone()
        if result is None:
            for _ in range(1000):
                conn.execute(
                    table.insert(),
                    [{'a': random.randint(1, 100000),
                      'b': ''.join(random.choice(string.ascii_letters) for _ in range(100))}
                      for _ in range(1000)]
                )

    with eng.connect() as conn:
        for row in conn.execution_options(stream_results=True).execute(table.select()):
            pass

now uses ~23 MB of memory instead of ~327 MB on CPython 3.5.2 and
PyMySQL 0.7.9.

psycopg2 implementation and execution options (stream_results,
server_side_cursors) are reused.

Change-Id: I4dc23ce3094f027bdff51b896b050361991c62e2

The method `engine.base.Connection.default_schema_name` is broken since
4b532e2 (this was when `engine.dialect.get_default_schema_name` was
replaced with `engine.dialect.default_schema_name`).  The method is
unused and also is not functional for end-user use, so remove it
for now.

Change-Id: I6ffbd9ce1b9b54ff6df63ed45d7d33591f183933
Pull-request: https://github.com/zzzeek/sqlalchemy/pull/321