My PR add support for using Postgresql lockmode 'FOR NO KEY UPDATE', which gives good possibility for no locks/dead locks, where we have many tables with foreign keys and lock rows in tables, which is master of foreign keys, and then trying to insert data into linked tables.

First run at fixing https://bitbucket.org/zzzeek/sqlalchemy/issue/2891/support-materialized-views-in-postgresql#comment-11769903

• Tables now take the postresql_relkind parameter in their constructor. This reflection option gets passed to get_table_oid() to filter its search for table oids. Can be full names (e.g. "materialized", "foreign", "view") or short-hand ("m", "f", "v", "r").
• get_table_oid() also always searches through tables with 'v' for backwards compatibility... Any better ideas on how to handle this?
• Added some basic tests for 'm' and 'f' relkinds. Added the postgres_test_db_link configuration for testing.

Another question: All calls to get_table_oid() need to come with table.dialect_kwargs to pass postgresql_relkind. Unfortunately, I couldn't find a way to reference table.dialect_kwargs from the get_unique_constraints method of the Inspector class. Any ideas on how to solve this?

https://github.com/rclmenezes/sqlalchemy/blob/feature/postgres-relkind/lib/sqlalchemy/engine/reflection.py#L385

SqlAlchemy should render NUMBER of NULL precision and scale == 0 as INTEGER instead of NUMBER (Otherwise roundtrip reflection re-creation changes original datatype.)

See RoundTripNumericTypeTest for failure case.

This intends to resolve https://bitbucket.org/zzzeek/sqlalchemy/issues/3906

The idea is to use values_callable as a new kwarg to Enum. It returns the values to be used, allowing for a variety of uses, including the one mentioned in the above issue.

Tests have been added as well as updates to the documentation. This is my first PR to sqlalchemy so apologies in advance for things that are out of place. Particularly I wonder if this needs to be supported in postgres as the work and tests I did were for mysql.

Hello, i'am return with new shiny tool flake8-rst. As you proposed in old PR. Tool inspired with blacked-docs.

Idea is simple, crop with regex code snippets and put it in flake8.

You can try it youself, just add this to tox.ini:

[flake8-rst]
filename = *.rst *.py
show-source = True
ignore = E711,E712,E721,N806,D,F811,F821,F401
exclude=.venv,.git,.tox,dist,doc,*egg,build
bootstrap = from sqlalchemy import Table, Column, Sequence, DateTime, Integer, String, Index, Base, ForeignKey, create_engine
            from sqlalchemy import event, trigger, listen
            from sqlalchemy import select, func
            from sqlalchemy.ext.declarative import declared_attr
            from sqlalchemy.orm.util import polymorphic_union
            from sqlalchemy.orm import relationship
            from fake_import import engine, my_engine, meta, mytable
            from fake_core import employees, users, addresses, conn

$ pip install flake8-rst
$ flake8-rst lib doc

This first part of changes:

1. add sourcecode directive;
2. fix syntax error.

After your approving of this approach, i am can fix other pep8 issues.

As discussed in #3747 (BitBucket)

1. Create an AppVeyor account and enable the SQLAlchemy repository (presumably the bitbucket one).
2. Encrypt PyPI credentials on AppVeyor by clicking your name at the top right and clicking "Encrypt data" under "Tools".

3. Put those values in appveyor.yml.

I've included the appveyor/getwheels.py script, which may be useful if you want to download the latest wheels for a build. I can remove it if you don't want it in the repo.

Why

• SQLAlchemy's connection pool is using the Queue data structure to provide the connection pools to threads round-robin basis. Which leads to maintain all connections even though the server's load is not that high enough to maintain all connections.
• Using the Stack data structure could solve this problem as connections that are not working will be disconnected from the database.

Changes

• Added StackPool in lib/sqlalchemy/pool/impl.py.
• Added Stack in lib/sqlalchemy/util/queue.py.

At some point it would also be nice to review the limitations that are listed in this file (mostly FreeTDS limitations I think as you point out) and delete the ones that don't apply anymore with reasonably recent versions of FreeTDS and pymssql.

Cc: @rsyring, @ramiro, @damoxc, @sontek, @aconrad

I think it would be good idea to add error on empy in_ usage in version 1.1.* Because current implementation is not fully correct: Firstly, it's not optimised, because of sequential scan Secondly, in current implementation null in () will return null, when in sql it should return False, try this: select null in (select 1 from some_table where false); -> false but not null

I understant that this can break some legacy code, but current implementation is obscure and quite expensive

This is specifically needed to properly integrate SQLAlchemy with asyncpg library. This library uses positional arguments and PostgreSQL requires $1-like syntax to accept such positional params.

Maybe you (@zzzeek) know how to make this possible in a simpler way? Current state looks terrible: asyncpgsa/connection.py - people are using weird regexp/replace stuff just to change paramstyle :(

Here I tried to make things better, but it still is not perfect. Using even one re.sub doesn't feels right.

And one more question. Currently it is not possible to compile final query statement with final params, which can be sent to the driver/db. Libraries like aiopg and asyncpgsa have to partially copy DefaultExecutionContext._init_compiled and other methods, accessing non-public _bind_processors property and so on. So is it possible to made such low-level usage possible and public to help integrate SQAlchemy with other libraries? You can see how this looks right now: aiopg/sa/connection.py :(

The problem:

Currently, in Oracle dialect, there is no way to tell sqlalchemy to connect to a database via service_name (a very common situation in Oracle RAC instances). The exisiting solutions are:

• use a full DSN in the host part of the connection string (very verbose)
• configure tnsnames.ora on server side and use that configured TNS name in your connection string (not very transferable)

Proposal

This PR introduces a new bool argument use_service_name (currently only via query string arg). It would control how to create a DSN from the cx_Oracle.makedsn function i.e. it would treat database part of the connection string as a service_name rather than sid.

Further development

I would be more than happy to also update OracleDialect_cx_oracle.__init__ method, write a test case and update documentation after this proposal gets an initial acceptance.

Bumps pypa/cibuildwheel from 2.11.2 to 2.11.4.

Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.