Release 0.2.1

[lemenkov]

Prepares 0.2.1. Per AGENTS.md: branch only — not merged, not tagged.

What ships

One change since 0.2.0 touches the wheel: the dialect asks the driver for LOB values rather than LOB objects (#15). seerdb 3.0 returns a LOB object from a SELECT of a CLOB / BLOB by default, and this dialect is built on the values — SQLAlchemy's Text and LargeBinary result processors, and every comparison the compliance suite makes, expect str / bytes.

Of the 11 commits since 0.2.0, that is the only one in sqlalchemy_seerdb/ (+28 lines). The rest are CI workflows, AGENTS.md / README.md, and test/ — none of it in the wheel.

Why patch, not minor

By this repository's own rule, a release is scored against the public surface: the URL scheme, the connect arguments it derives, the SQLAlchemy types it maps, and requirements.py. None of those moved. A user's SELECT of a CLOB still returns str, exactly as in 0.2.0 — the change only keeps that true against a newer driver, which is the definition of a patch.

Driver floor

Stays seerdb>=2.5.0. fetch_lobs is feature-detected, not required, so this release does not depend on an unreleased driver — which the release procedure forbids. Removing that probe once the floor reaches 3.0 is tracked as #16.

Validation

Compliance suite on 11g (oracle+seerdb://…/?service_name=XE), run both ways:

driver	probe	result
released seerdb 2.5.0 (from PyPI, clean venv)	False	52 failed, 947 passed, 350 skipped
driver checkout carrying the LOB flip	True	52 failed, 945 passed, 349 skipped

The 52 are this bed's own pre-existing failures, identical on 0.2.0 against driver master — the same baseline either way, so neither the change nor the probe's branch alters the outcome. The released-driver run is the one that matters for shipping, and it exercises the probe's False path, which is the path every current user gets.

Known failures: the 52 above are environmental on this local bed and not tied to a dialect ticket; CI's prepared schema is the reference.

🤖 Generated with Claude Code