chore: replace mypy with ty for type checking

[joshmarkovic]

Closes #712.

What

Replaces mypy with ty, completing the Astral toolchain consolidation from #707 (lint) and #711 (format). Scope and rules live in [tool.ty] in pyproject.toml; ty==0.0.64 in the dev group is the single source of the version, read by the pre-commit hook, make ty and CI alike.

• .pre-commit-config.yaml: two mirrors-mypy hooks become one local hook running uv run --frozen ty check.
• .github/workflows/type-check.yml: new job running the same command.
• Makefile: make mypy becomes make ty; lint and test updated.

Coverage is unchanged

The adapter package only, with unresolvable dbt.* imports ignored, which is what mypy's --ignore-missing-imports did. dbt is split across distributions: dbt-core and dbt-adapters own dbt/ and dbt/adapters/ in site-packages, this repo contributes dbt/adapters/sqlserver, and the editable install joins them at runtime via an import hook that no static checker can follow.

Why ty runs in Actions and not on pre-commit.ci

pre-commit.ci blocks network access while hooks run, and ty needs it to resolve this project's dependencies. Astral's own recommendation is ci: skip plus a separate CI invocation. pre-commit.ci is this repo's only lint runner today, so without the new job the swap would drop type checking from CI entirely. If a general lint workflow lands later, this job folds into it.

A local hook is used rather than astral-sh/ty-pre-commit so the version is declared once: the upstream hook hardcodes --ty-version in its entry, which would mean a rev pin and a dev-group pin drifting apart, the same dual-pin mypy had. Version bumps now arrive through the existing dependabot pip group as a pyproject.toml + uv.lock change. The hook needs uv on PATH, already a requirement via make dev.

Source changes

Fourteen touches to reach a clean baseline:

• Annotate Column: Type[SQLServerColumn], which __init__ overwrites with the SQLServerColumnNative subclass. The only real fix.
• Six suppressions on the behavior flags: dbt-common declares BehaviorFlag's optional keys with a NotRequired that falls back to Optional under a try/except ImportError shim, so source and docs_url read as required. Reproduced in isolation; the suppressions go stale (and ty says so) once that shim is dropped.
• Two unknown-argument suppressions on SQLServerIndexConfigChange, whose action field comes from the unresolvable RelationConfigChange base.
• Drop the # type: ignore on SQLServerRelation.type, which suppressed nothing and ty reports as unused.

Inline suppressions rather than disabling invalid-return-type, unknown-argument and missing-typed-dict-key project-wide, so those rules keep working everywhere else.

Two corrections to the issue

• Motivation 1 is already resolved: the pathspec override was removed in the dbt-core 1.12 upgrade (dd61d69), and pathspec now resolves to 1.0.4 on its own.
• Motivation 2 is overstated for this repo. Measured on dbt/adapters: mypy 2.1.0 at 2.7s cold and 0.19s warm cache, ty at 0.12s. Real but not the 28x in the issue. The durable wins are one toolchain and config in pyproject.toml.

Worth flagging: ty is still 0.0.x with no stability guarantees, hence the exact pin.

Testing

• pre-commit run -a passes, including ty.
• uv run --frozen ty check, the CI command, is clean.
• 499 unit tests pass.
• Injected a deliberate -> int: return "not an int" and confirmed the hook fails on it, so the config is not a no-op.
• Verified --frozen leaves uv.lock byte-identical, so a commit cannot silently relock.

Follow-up

With an extra search path pointing at site-packages, ty does resolve dbt.* and reports 9 findings mypy never saw: an AdapterPlugin protocol mismatch, from_dict incompatible across RelationConfigBase and DataClassDictMixin, mp_context=None against SpawnContext, a can_expand_to override incompatibility, and an unresolved attribute on self.connections. That search path is not committable, it hardcodes a Python version and breaks in CI, but the findings are real and deserve their own issue.

[axellpadilla]

@joshmarkovic taking into account the failure requieres changes, .gitignore: .mypy_cache/ dropped, ty writes no cache. could be kept, some could use the extension still.
Seems you got a good reason for the closed too #776 , ty not yet compatible with the app

[joshmarkovic]

@axellpadilla Done, kept .mypy_cache/ in .gitignore, so that file is out of the diff entirely. While I was in there I also rebased onto master and fixed the CI failure: astral-sh/setup-uv@v9 doesn't exist as a tag (they stopped publishing moving major tags after v7), so it's pinned to v9.0.0 now. uv run --frozen ty check, all pre-commit hooks and 499 unit tests pass locally on the rebased branch. Could you approve the workflow runs when you get a chance? The ty job's steps have never actually executed in CI, so I'd like to see it green.

[axellpadilla]

@joshmarkovic github actions is down but no worries, any issue follow up, need to work on top of this