Fix: preserve quoted_name(quote=...) semantics when rendering schema in autogenerate

[pranjalm37]

Summary

Fixes #1526.

When a table/index/constraint's schema is set using an explicit
quoted_name(name, quote=True) or quoted_name(name, quote=False), the
autogenerate render code (_ident() in alembic/autogenerate/render.py)
collapsed the value down to a plain string before formatting it with %r,
so the caller's explicit quoting preference was silently dropped from the
generated migration file.

Fix

Adds _render_schema_ident(), used at every schema= render site
(create/drop table, index, column, check/unique constraint, drop_constraint,
alter_column, and the batch_alter_table header):

• If the schema is a quoted_name with quote explicitly set (True/False),
  render it as quoted_name('name', True/False) and add the necessary import
  to the generated migration file.
• Otherwise, falls back to repr() of the original value, preserving prior
  behavior — including for objects with a custom __repr__ (there's an
  existing test, test_drop_unique_constraint_schema_reprobj, covering that
  case, which still passes).

Test plan

• Added test_render_table_w_quoted_name_schema, matching the repro from
  Migration autogeneration should respect schemas defined using quoted_name #1526, asserting both the rendered code and that the quoted_name import
  is added.
• Full test suite passes locally: pytest tests/ --ignore=tests/test_post_write.py --ignore=tests/test_mssql.py --ignore=tests/test_oracle.py --ignore=tests/test_mysql.py --ignore=tests/test_postgresql.py → 1562 passed, 133 skipped (DB-specific suites skipped, no local DB available; test_post_write.py failures are pre-existing/unrelated — they fail identically on main without this change, due to black not being installed in the hook environment).
• black --check and flake8 pass on both changed files.
• Added changelog fragment at docs/build/unreleased/1526.rst.

[MohammedAlkindi]

Ran this on Windows 11 / Python 3.13.13, since alembic's CI matrix is ubuntu-22.04 / ubuntu-latest only.

No new failures. Full suite on this branch:

$ python -m pytest tests/ -q
5 failed, 1814 passed, 132 skipped in 39.87s

The 5 are all ScriptNamingTest timezone tests (test_custom_tz, test_custom_tz_lowercase, test_custom_tz_utc, test_custom_tzdata_tz, test_generates_a_date) failing on zoneinfo with no system tzdata, which is a Windows-only environment gap. They fail identically on a clean upstream/main checkout on this machine — same count, same five names — so the delta from this PR is zero.

The new test has teeth. Reverting only the source file and keeping the test:

$ git checkout upstream/main -- alembic/autogenerate/render.py
$ python -m pytest tests/test_autogen_render.py -q
FAILED tests/test_autogen_render.py::AutogenRenderTest::test_render_table_w_quoted_name_schema
AssertionError:
  "...schema='database.schema')"
  != "...schema=quoted_name('database.schema', False))"
1 failed, 178 passed, 2 skipped

and with render.py restored, 179 passed, 2 skipped. So it fails before and passes after rather than only covering the happy path — the rendered output genuinely loses the quote=False marker without the change, which is the reported bug.

Not checked: PostgreSQL/MySQL/Oracle behaviour (I ran the default sqlite configuration), and I did not review the render.py changes for design intent — this is a platform and regression-coverage report, not a design review.

[MohammedAlkindi]

Checked this out and ran it on Windows 11 (10.0.26200), Python 3.13.13, against main at c116cbc0.

Test evidence:

pytest tests/test_autogen_render.py	passed	skipped
main	178	2
this PR	179	2

The +1 is the PR's own test, and no existing test changes state.

Fail-before confirmed, which is the part worth having: with alembic/autogenerate/render.py reverted to main and the new test kept, test_render_table_w_quoted_name_schema fails. So the test genuinely discriminates the fix rather than just documenting current behaviour.

The diagnosis looks right to me. quoted_name is a str subclass, so %r collapses it to a plain string literal and the quote= flag simply disappears from the generated migration — the rendered file then re-renders with default quoting rules, which is exactly the #1526 symptom. Routing through _render_schema_ident and emitting the explicit wrapper preserves the caller's intent.

Coverage check: I grepped render.py for any remaining schema render that still goes through %r, and found none — _add_table, _drop_table, _render_modify_table and the index path all go through the new helper now. The change looks complete rather than partial.

One thing I checked rather than flagged. _render_schema_ident returns str | None, and _render_modify_table interpolates it with %s without an if op.schema: guard, unlike _add_table and _drop_table. That looked like it might emit a literal schema=None, but it's fine and unchanged: %r of None and %s of None both produce None, so a schema-less batch op renders schema=None exactly as it did before. Noting it only so the asymmetry between guarded and unguarded call sites doesn't read as an oversight later.

I ran only test_autogen_render.py rather than the full suite, since that is the file this touches — worth saying explicitly so the numbers above aren't read as a whole-suite result.

---

Disclosure: I used an AI assistant while reviewing. The before/after runs, the fail-before check, and the render.py grep were executed by me on the machine described.