Gord Thompson [Thu, 3 Oct 2019 22:18:54 +0000 (18:18 -0400)]
fix error in test_round_trip for TimeTest with mysql+pyodbc
Fixes: #4879
### Description
create mysql+pyodbc-specific `_pyodbcTIME` class to avoid error thrown by `result_processor` in the (more) generic mysql/types.py.
### Checklist
<!-- go over following points. check them with an `x` if they do apply, (they turn into clickable checkboxes once the PR is submitted, so no need to do everything at once)
-->
This pull request is:
- [ ] A documentation / typographical error fix
- Good to go, no issue or tests are needed
- [x] A short code fix
- please include the issue number, and create an issue if none exists, which
must include a complete example of the issue. one line code fixes without an
issue and demonstration will not be accepted.
- Please include: `Fixes: #<issue number>` in the commit message
- please include tests. one line code fixes without tests will not be accepted.
- [ ] A new feature implementation
- please include the issue number, and create an issue if none exists, which must
include a complete example of how the feature would look.
- Please include: `Fixes: #<issue number>` in the commit message
- please include tests.
Mike Bayer [Mon, 7 Oct 2019 20:12:30 +0000 (16:12 -0400)]
Fix max_identifier_length for SQL server
Fixed bug in SQL Server dialect with new "max_identifier_length" feature
where the mssql dialect already featured this flag, and the implementation
did not accommodate for the new initialization hook correctly.
Mike Bayer [Fri, 4 Oct 2019 15:12:27 +0000 (11:12 -0400)]
Warn for object replaced in identity map during flush
A warning is emitted for a condition in which the :class:`.Session` may
implicitly swap an object out of the identity map for another one with the
same primary key, detaching the old one, which can be an observed result of
load operations which occur within the :meth:`.SessionEvents.after_flush`
hook. The warning is intended to notify the user that some special
condition has caused this to happen and that the previous object may not be
in the expected state.
Mike Bayer [Thu, 3 Oct 2019 18:09:54 +0000 (14:09 -0400)]
call setinputsizes for cx_Oracle.DATETIME
Restored adding cx_Oracle.DATETIME to the setinputsizes() call when a
SQLAlchemy :class:`.Date`, :class:`.DateTime` or :class:`.Time` datatype is
used, so that in the case where a bound parameter is passed as NULL
in some complex queries (in particular this happens with some lazy load
situations), the type is still present. This was removed
in the 1.2 series for arbitrary reasons.
Also adds a suite test for this generic situation.
What's not clear is that do we really need setinputsizes() for all
datatypes if we are supporting NULL in bound parameters.
Mike Bayer [Thu, 3 Oct 2019 15:18:06 +0000 (11:18 -0400)]
Apply quoting to SQL Server _switch_db
Added identifier quoting to the schema name applied to the "use" statement
which is invoked when a SQL Server multipart schema name is used within a
:class:`.Table` that is being reflected, as well as for :class:`.Inspector`
methods such as :meth:`.Inspector.get_table_names`; this accommodates for
special characters or spaces in the database name. Additionally, the "use"
statement is not emitted if the current database matches the target owner
database name being passed.
Mike Bayer [Wed, 2 Oct 2019 15:20:58 +0000 (11:20 -0400)]
Add max_identifier_length parameter; warn for Oracle
Added new :func:`.create_engine` parameter
:paramref:`.create_engine.max_identifier_length`. This overrides the
dialect-coded "max identifier length" in order to accommodate for databases
that have recently changed this length and the SQLAlchemy dialect has
not yet been adjusted to detect for that version. This parameter interacts
with the existing :paramref:`.create_engine.label_length` parameter in that
it establishes the maximum (and default) value for anonymously generated
labels.
The Oracle dialect now emits a warning if Oracle version 12.2 or greater is
used, and the :paramref:`.create_engine.max_identifier_length` parameter is
not set. The version in this specific case defaults to that of the
"compatibility" version set in the Oracle server configuration, not the
actual server version. In version 1.4, the default max_identifier_length
for 12.2 or greater will move to 128 characters. In order to maintain
forwards compatibility, applications should set
:paramref:`.create_engine.max_identifier_length` to 30 in order to maintain
the same length behavior, or to 128 in order to test the upcoming behavior.
This length determines among other things how generated constraint names
are truncated for statements like ``CREATE CONSTRAINT`` and ``DROP
CONSTRAINT``, which means a the new length may produce a name-mismatch
against a name that was generated with the old length, impacting database
migrations.
Mike Bayer [Mon, 30 Sep 2019 14:54:50 +0000 (10:54 -0400)]
Deprecate plain string passed to session.query()
Passing a plain string expression to :meth:`.Session.query` is deprecated,
as all string coercions were removed in :ticket:`4481` and this one should
have been included. The :func:`.literal_column` function may be used to
produce a textual column expression.
Note this changeset is in the 1.3 branch only. The 1.4 (currently master)
branch will already have removed this behavior.
Mike Bayer [Mon, 30 Sep 2019 13:55:40 +0000 (09:55 -0400)]
Ensure states with null m2o FK value are still populated by selectinloader
Fixed regression in selectinload loader strategy caused by #4775 (released
in version 1.3.6) where a many-to-one attribute of None would no longer be
populated by the loader. While this was usually not noticeable due to the
lazyloader populating None upon get, it would lead to a detached instance
error if the object were detached.
Mike Bayer [Fri, 20 Sep 2019 13:25:40 +0000 (09:25 -0400)]
Support SQLite URIs
Added support for sqlite "URI" connections, which allow for sqlite-specific
flags to be passed in the query string such as "read only" for Python
sqlite3 drivers that support this.
In 14b1e6fe8e18d139846c1aba6761d4eea3dc25c3 we added
suite-level requirements but did not add them to the
base, causing failures in third party dialect test suites.
Mike Bayer [Wed, 18 Sep 2019 01:04:21 +0000 (21:04 -0400)]
Add missing suite level requirements from 2efd89d02941
In 2efd89d02941ab4267d6e2842963fd38b1539f6c we added
suite-level requirements but did not add them to the
base, causing failures in third party dialect test suites.
Mike Bayer [Tue, 17 Sep 2019 16:49:46 +0000 (12:49 -0400)]
Add missing suite level requirements from #4234
In 9d5e117f6fcc38d8773bc943c615888dc8a3a819 we added
suite-level requirements but did not add them to the
base, causing failures in third party dialect test suites.
Mike Bayer [Thu, 12 Sep 2019 16:37:38 +0000 (12:37 -0400)]
Document visitors module
As we are going to be adding a lot of new features to the visitors
module which may impact end-user custom constructs, start documenting
the package for now.
Mike Bayer [Fri, 6 Sep 2019 22:40:46 +0000 (18:40 -0400)]
Catch set_parent_w_dispatch missing
Added an explicit error message for the case when objects passed to
:class:`.Table` are not :class:`.SchemaItem` objects, rather than resolving
to an attribute error.
Mike Bayer [Thu, 5 Sep 2019 15:46:44 +0000 (11:46 -0400)]
Document Query deduplication
Users are frequently confused why count() does not return the same
number as the number of objects returned by all(). While we continue
to want to find a better solution to this problem, in the meantime
this has never been clearly documented. Add an FAQ section with
links from .count() , .all(), ORM tutorial.
Mike Bayer [Wed, 4 Sep 2019 22:46:53 +0000 (18:46 -0400)]
Strip special chars in anonymized bind names
Characters that interfere with "pyformat" or "named" formats in bound
parameters, namely ``%, (, )`` and the space character, as well as a few
other typically undesirable characters, are stripped early for a
:func:`.bindparam` that is using an anonymized name, which is typically
generated automatically from a named column which itself includes these
characters in its name and does not use a ``.key``, so that they do not
interfere either with the SQLAlchemy compiler's use of string formatting or
with the driver-level parsing of the parameter, both of which could be
demonstrated before the fix. The change only applies to anonymized
parameter names that are generated and consumed internally, not end-user
defined names, so the change should have no impact on any existing code.
Applies in particular to the psycopg2 driver which does not otherwise quote
special parameter names, but also strips leading underscores to suit Oracle
(but not yet leading numbers, as some anon parameters are currently
entirely numeric/underscore based); Oracle in any case continues to quote
parameter names that include special characters.
Mike Bayer [Fri, 30 Aug 2019 17:11:05 +0000 (13:11 -0400)]
Don't create enum constraints in enum sortable tests
Fixed unit test regression released in 1.3.8 that would cause failure for
Oracle, SQL Server and other non-native ENUM platforms due to new
enumeration tests added as part of :ticket:`4285` enum sortability in the
unit of work; the enumerations created constraints that were duplicated on
name.
Nicolas CANIART [Thu, 22 Aug 2019 18:16:29 +0000 (14:16 -0400)]
Implement type-level sorting for Enum; apply to ORM primary keys
Added support for the use of an :class:`.Enum` datatype using Python
pep-435 enumeration objects as values for use as a primary key column
mapped by the ORM. As these values are not inherently sortable, as
required by the ORM for primary keys, a new
:attr:`.TypeEngine.sort_key_function` attribute is added to the typing
system which allows any SQL type to implement a sorting for Python objects
of its type which is consulted by the unit of work. The :class:`.Enum`
type then defines this using the database value of a given enumeration.
The sorting scheme can be also be redefined by passing a callable to the
:paramref:`.Enum.sort_key_function` parameter. Pull request courtesy
Nicolas Caniart.
Added support for reflection of CHECK constraints that include the special
PostgreSQL qualifier "NOT VALID", which can be present for CHECK
constraints that were added to an exsiting table with the directive that
they not be applied to existing data in the table. The PostgreSQL
dictionary for CHECK constraints as returned by
:meth:`.Inspector.get_check_constraints` may include an additional entry
``dialect_options`` which within will contain an entry ``"not_valid":
True`` if this symbol is detected. Pull request courtesy Bill Finn.
Mike Bayer [Mon, 26 Aug 2019 15:44:09 +0000 (11:44 -0400)]
Serialize the context dictionary in Load objects
Fixed bug where :class:`.Load` objects were not pickleable due to
mapper/relationship state in the internal context dictionary. These
objects are now converted to picklable using similar techniques as that of
other elements within the loader option system that have long been
serializable.
Mike Bayer [Mon, 26 Aug 2019 13:45:06 +0000 (09:45 -0400)]
Remove erroneous assertion from array._bind_param
Fixed bug where Postgresql operators such as
:meth:`.postgresql.ARRAY.Comparator.contains` and
:meth:`.postgresql.ARRAY.Comparator.contained_by` would fail to function
correctly for non-integer values when used against a
:class:`.postgresql.array` object, due to an erroneous assert statement.
Mike Bayer [Wed, 21 Aug 2019 21:29:55 +0000 (17:29 -0400)]
Add hide_parameters flag to create_engine
Added new parameter :paramref:`.create_engine.hide_parameters` which when
set to True will cause SQL parameters to no longer be logged, nor rendered
in the string representation of a :class:`.StatementError` object.
Mike Bayer [Tue, 20 Aug 2019 13:42:36 +0000 (09:42 -0400)]
Document how to work with reflection and custom datatypes
Describe the link between table reflection, datatype lookups,
and what approaches are needed in the case where in-Python datatypes
are needed in the reflected table metadata.
Mike Bayer [Sun, 18 Aug 2019 17:03:49 +0000 (13:03 -0400)]
Reflect PK of referred table if referred columns not present
Fixed bug where a FOREIGN KEY that was set up to refer to the parent table
by table name only without the column names would not correctly be
reflected as far as setting up the "referred columns", since SQLite's
PRAGMA does not report on these columns if they weren't given explicitly.
For some reason this was harcoded to assume the name of the local column,
which might work for some cases but is not correct. The new approach
reflects the primary key of the referred table and uses the constraint
columns list as the referred columns list, if the remote column(s) aren't
present in the reflected pragma directly.
Mike Bayer [Sat, 17 Aug 2019 02:38:51 +0000 (22:38 -0400)]
Revise psycopg2 execute_values approach
Revised the approach for the just added support for the psycopg2
"execute_values()" feature added in 1.3.7 for :ticket:`4623`. The approach
relied upon a regular expression that would fail to match for a more
complex INSERT statement such as one which had subqueries involved. The
new approach matches exactly the string that was rendered as the VALUES
clause.
Mike Bayer [Fri, 16 Aug 2019 22:07:06 +0000 (18:07 -0400)]
Add new "exec_once_unless_exception" system; apply to dialect.initialize
Fixed an issue whereby if the dialect "initialize" process which occurs on
first connect would encounter an unexpected exception, the initialize
process would fail to complete and then no longer attempt on subsequent
connection attempts, leaving the dialect in an un-initialized, or partially
initialized state, within the scope of parameters that need to be
established based on inspection of a live connection. The "invoke once"
logic in the event system has been reworked to accommodate for this
occurrence using new, private API features that establish an "exec once"
hook that will continue to allow the initializer to fire off on subsequent
connections, until it completes without raising an exception. This does not
impact the behavior of the existing ``once=True`` flag within the event
system.
Yuval Dinari [Mon, 12 Aug 2019 14:44:59 +0000 (10:44 -0400)]
Add new executemany_mode, support for psycopg2.extras.execute_values()
Added new dialect flag for the psycopg2 dialect, ``executemany_mode`` which
supersedes the previous experimental ``use_batch_mode`` flag.
``executemany_mode`` supports both the "execute batch" and "execute values"
functions provided by psycopg2, the latter which is used for compiled
:func:`.insert` constructs. Pull request courtesy Yuval Dinari.
Mike Bayer [Sun, 11 Aug 2019 19:24:13 +0000 (15:24 -0400)]
Rewrite pool reset_on_return parsing using a util function
Choosing a util.symbol() based on a user parameter is about to have
another use case added as part of #4623, so add a generalized solution
ahead of it.
Mike Bayer [Sun, 11 Aug 2019 14:54:49 +0000 (10:54 -0400)]
Use ternary when running conditional with Query._offset
Fixed bug where using :meth:`.Query.first` or a slice expression in
conjunction with a query that has an expression based "offset" applied
would raise TypeError, due to an "or" conditional against "offset" that did
not expect it to be a SQL expression as opposed to an integer or None.
Mike Bayer [Sun, 11 Aug 2019 21:42:59 +0000 (17:42 -0400)]
Emit SET NAMES for all MySQL connections w charset
The MySQL dialects will emit "SET NAMES" at the start of a connection when
charset is given to the MySQL driver, to appease an apparent behavior
observed in MySQL 8.0 that raises a collation error when a UNION includes
string columns unioned against columns of the form CAST(NULL AS CHAR(..)),
which is what SQLAlchemy's polymorphic_union function does. The issue
seems to have affected PyMySQL for at least a year, however has recently
appeared as of mysqlclient 1.4.4 based on changes in how this DBAPI creates
a connection. As the presence of this directive impacts three separate
MySQL charset settings which each have intricate effects based on their
presense, SQLAlchemy will now emit the directive on new connections to
ensure correct behavior.
Mike Bayer [Sat, 10 Aug 2019 22:17:59 +0000 (18:17 -0400)]
Clarify INSERT/UPDATE defaults vs. data marshalling
- Add a full introductory paragraph to INSERT/UPDATE defaults
stating what we mean when we talk about this concept. Add a note
differentiating what a default does, vs. a rule that intercepts
data as it moves into the database, providing links
- Add a quick section referring to TypeDecorator in the ORM
section on modifying attribute behavior
- Add an "ORM tip", a new thing that we can use in Core to link
to ORM concepts when useful, in the TypeDecorator section which
mentions that for more open ended conversion of arbitrary user
data based on business rules, @validates might be useful
(although this still does not suit the case of full blown form
validation).
- add glossary entries for DML, data marshalling since we already
use these terms and I'd like to refer to them more often.
Mike Bayer [Fri, 9 Aug 2019 03:34:20 +0000 (23:34 -0400)]
Correct name for json_serializer / json_deserializer, document and test
The dialects that support json are supposed to take arguments
``json_serializer`` and ``json_deserializer`` at the create_engine() level,
however the SQLite dialect calls them ``_json_serilizer`` and
``_json_deserilalizer``. The names have been corrected, the old names are
accepted with a change warning, and these parameters are now documented as
:paramref:`.create_engine.json_serializer` and
:paramref:`.create_engine.json_deserializer`.
Mike Bayer [Sat, 3 Aug 2019 03:20:06 +0000 (23:20 -0400)]
Strong reference listen function wrapped by "once"
Fixed issue in event system where using the ``once=True`` flag with
dynamically generated listener functions would cause event registration of
future events to fail if those listener functions were garbage collected
after they were used, due to an assumption that a listened function is
strongly referenced. The "once" wrapped is now modified to strongly
reference the inner function persistently, and documentation is updated
that using "once" does not imply automatic de-registration of listener
functions.
Mike Bayer [Fri, 2 Aug 2019 17:03:29 +0000 (13:03 -0400)]
Always include a schema name in SQLite PRAGMA
Fixed bug where usage of "PRAGMA table_info" in SQLite dialect meant that
reflection features to detect for table existence, list of table columns,
and list of foreign keys, would default to any table in any attached
database, when no schema name was given and the table did not exist in the
base schema. The fix explicitly runs PRAGMA for the 'main' schema and then
the 'temp' schema if the 'main' returned no rows, to maintain the behavior
of tables + temp tables in the "no schema" namespace, attached tables only
in the "schema" namespace.
Mike Bayer [Thu, 1 Aug 2019 15:45:34 +0000 (11:45 -0400)]
Don't assume key when matching cloned columns in _make_proxy
Fixed issue where internal cloning of SELECT constructs could lead to a key
error if the copy of the SELECT changed its state such that its list of
columns changed. This was observed to be occurring in some ORM scenarios
which may be unique to 1.3 and above, so is partially a regression fix.
For 1.4, the _is_clone_of key will be removed entirely as it seems to
have no purpose. This commit is the initial backport to 1.3 which
includes tests.
Mike Bayer [Mon, 29 Jul 2019 18:49:22 +0000 (14:49 -0400)]
Invoke column_expression() for subsequent SELECTs in CompoundSelect
Fixed bug where :meth:`.TypeEngine.column_expression` method would not be
applied to subsequent SELECT statements inside of a UNION or other
:class:`.CompoundSelect`, even though the SELECT statements are rendered at
the topmost level of the statement. New logic now differentiates between
rendering the column expression, which is needed for all SELECTs in the
list, vs. gathering the returned data type for the result row, which is
needed only for the first SELECT.
Mike Bayer [Mon, 29 Jul 2019 18:07:21 +0000 (14:07 -0400)]
Correct for MySQL 8.0 table and schema names in FK reflection
Added another fix for an upstream MySQL 8 issue where a case sensitive
table name is reported incorrectly in foreign key constraint reflection,
this is an extension of the fix first added for :ticket:`4344` which
affects a case sensitive column name. The new issue occurs through MySQL
8.0.17, so the general logic of the 88718 fix remains in place.
Mike Bayer [Tue, 23 Jul 2019 22:38:48 +0000 (18:38 -0400)]
Ensure all Index arguments are counted when matching expr/columns
Fixed issue where :class:`.Index` object which contained a mixture of
functional expressions which were not resolvable to a particular column,
in combination with string-based column names, would fail to initialize
its internal state correctly leading to failures during DDL compilation.
Mike Bayer [Tue, 23 Jul 2019 13:24:32 +0000 (09:24 -0400)]
Don't assume m2o key is present in the dictionary
Fixed regression caused by new selectinload for many-to-one logic where
a primaryjoin condition not based on real foreign keys would cause
KeyError if a related object did not exist for a given key value on the
parent object.
Mike Bayer [Fri, 19 Jul 2019 17:05:06 +0000 (13:05 -0400)]
Optimize out JOIN for selectinload with many to one
The optimzation applied to selectin loading in :ticket:`4340` where a JOIN
is not needed to eagerly load related items is now applied to many-to-one
relationships as well, so that only the related table is queried for a
simple join condition. In this case, the related items are queried
based on the value of a foreign key column on the parent; if these columns
are deferred or otherwise not loaded on any of the parent objects in
the collection, the loader falls back to the JOIN method.
Mike Bayer [Tue, 16 Jul 2019 16:41:09 +0000 (12:41 -0400)]
Support tuple IN VALUES for SQLite, others
Added support for composite (tuple) IN operators with SQLite, by rendering
the VALUES keyword for this backend. As other backends such as DB2 are
known to use the same syntax, the syntax is enabled in the base compiler
using a dialect-level flag ``tuple_in_values``. The change also includes
support for "empty IN tuple" expressions for SQLite when using "in_()"
between a tuple value and an empty set.
Mike Bayer [Thu, 18 Jul 2019 15:12:49 +0000 (11:12 -0400)]
Fix regression for self-ref join to self error message
Fixed regression caused by :ticket:`4365` where a join from an entity to
itself without using aliases no longer raises an informative error message,
instead failing on an assertion. The informative error condition has been
restored.
Mike Bayer [Thu, 18 Jul 2019 14:58:24 +0000 (10:58 -0400)]
Fixes for uselist=True with m2o relationships
Fixed bug where a many-to-one relationship that specified ``uselist=True``
would fail to update correctly during a primary key change where a related
column needs to change.
Fixed bug where the detection for many-to-one or one-to-one use with a
"dynamic" relationship, which is an invalid configuration, would fail to
raise if the relationship were configured with ``uselist=True``. The
current fix is that it warns, instead of raises, as this would otherwise be
backwards incompatible, however in a future release it will be a raise.
Mike Bayer [Tue, 16 Jul 2019 17:02:16 +0000 (13:02 -0400)]
Intercept unresolveable comparator attrbute error for attr access
Fixed bug where a synonym created against a mapped attribute that does not
exist yet, as is the case when it refers to backref before mappers are
configured, would raise recursion errors when trying to test for attributes
on it which ultimately don't exist (as occurs when the classes are run
through Sphinx autodoc), as the unconfigured state of the synonym would put
it into an attribute not found loop.
Mike Bayer [Sat, 13 Jul 2019 02:43:31 +0000 (22:43 -0400)]
self_group() for FunctionFilter
Fixed issue where the :class:`.array_agg` construct in combination with
:meth:`.FunctionElement.filter` would not produce the correct operator
precedence between the FILTER keyword and the array index operator.
Mike Bayer [Wed, 10 Jul 2019 20:12:48 +0000 (16:12 -0400)]
Add performance improvement for Enum w/ Python 2 enum library
Adjusted the initialization for :class:`.Enum` to minimize how often it
invokes the ``.__members__`` attribute of a given PEP-435 enumeration
object, to suit the case where this attribute is expensive to invoke, as is
the case for some popular third party enumeration libraries.
Mike Bayer [Mon, 8 Jul 2019 19:46:35 +0000 (15:46 -0400)]
Support multidimensional array literals in Postgresql
Added support for multidimensional Postgresql array literals via nesting
the :class:`.postgresql.array` object within another one. The
multidimensional array type is detected automatically.
Mike Bayer [Mon, 8 Jul 2019 14:42:20 +0000 (10:42 -0400)]
Ensure .engine is part of Connectable interface, implement as descriptor
Fixed bug where using reflection function such as :meth:`.MetaData.reflect`
with an :class:`.Engine` object that had execution options applied to it
would fail, as the resulting :class:`.OptionEngine` proxy object failed to
include a ``.engine`` attribute used within the reflection routines.
Mike Bayer [Tue, 23 Apr 2019 22:33:08 +0000 (17:33 -0500)]
Add additional parsing to extract the "MariaDB" keyword from version string
Enhanced MySQL/MariaDB version string parsing to accommodate for exotic
MariaDB version strings where the "MariaDB" word is embedded among other
alphanumeric characters such as "MariaDBV1". This detection is critical in
order to correctly accomodate for API features that have split between MySQL
and MariaDB such as the "transaction_isolation" system variable.
Mike Bayer [Fri, 5 Jul 2019 16:19:55 +0000 (12:19 -0400)]
Note DBAPIs and dialects that we don't support
Since we have strong CI for the DBAPIs and dialects that are actively
supported, this indicates that those DBAPIs that aren't in CI are
continuing to fall behind in support, to the point where we can
not address issues that may arise. As such, the Sybase and Firebird
dialects overall are moving into an explicit "not supported" zone
where we would like to eventually remove them. Additionally,
a pass is made through legacy MySQL and PostgreSQL DBAPI dialects
as well as those which we aren't able to include in CI to note
that these DBAPIs aren't actively supported by the project.