Avoid manipulating search path in table metadata fetch

[cwegrzyn]

We've run into some issues with transactions and the way that _get_all_column_info manipulates the search path. The search path manipulation only appears to be necessary because pg_table_def is used-- but pg_table_def is a pretty simple view that we can integrate into the query, minus its pesky search path filter.

For reference, pg_table_def is defined on Redshift as follows:

CREATE OR REPLACE VIEW pg_catalog.pg_table_def AS
SELECT n.nspname AS schemaname, c.relname AS tablename, a.attname AS "column", format_type(a.atttypid, a.atttypmod) AS "type", format_encoding(a.attencodingtype::integer) AS "encoding", a.attisdistkey AS "distkey", a.attsortkeyord AS "sortkey", a.attnotnull AS "notnull"
FROM pg_namespace n, pg_class c, pg_attribute a
WHERE n.oid = c.relnamespace AND c.oid = a.attrelid AND a.attnum > 0 AND NOT a.attisdropped AND pg_table_is_visible(c.oid)
ORDER BY n.nspname, c.relname, a.attnum

pg_table_is_visible is the bit that limits to tables in the search path.

Todos

• MIT compatible
• Tests
• Documentation
• Updated CHANGES.rst

[cwegrzyn]

Hey! I was just wondering about next steps for merging this in. I'd be happy to add an entry to CHANGES. Not sure if new tests/docs are relevant here. We're also bumping into the issue solved by #105, so I'd be happy to update that one to reflect the changes here.

[jklukas]

Passing integration tests, so this looks great! All we need is a line in CHANGES.rst that describes the change in behavior and links to this PR, and I'll go ahead and merge.

[cwegrzyn]

Does that look okay?

[jklukas]

small typo to fix

[cwegrzyn]

Oops, thanks! Committed

[jklukas]

Released in https://pypi.org/project/sqlalchemy-redshift/0.7.2/