This PR addresses part of issue #901.
What type of PR is this?
Description
Correct the explicit DecimalParameter type to use SQL's DECIMAL(precision,scale) order, and honor an explicitly declared zero scale. Correct the parameter descriptions and document the compatibility change for callers who followed the old reversed descriptions. The public signature, validation rules, automatic sizing and value payload are unchanged. Explicit zero precision is now sent as declared instead of silently using automatic sizing.
This is a partial fix: it does not change the issue's broader under-declared-scale/truncation semantics or claim that its entire conformance test now passes. The adjacent automatic-sizing work in PR #952 is untouched; its full patch applies cleanly alongside this change.
AI-assisted implementation, independently reviewed. @leninworld personally certified DCO 1.1 and confirmed the commit sign-off.
How is this tested?
- Unit tests
- E2E Tests
- Manually
Linux aarch64 Docker, Python 3.13.16, Poetry 2.5.1, official optional kernel wheel 1.1.0:
- Before the fix: two new regression cases fail (
DECIMAL(2,10) and DECIMAL(3,0)); the automatic-sizing control passes.
- After: all 59 parameter tests pass. The ordinary unit suite with the kernel wheel installed reports 1,250 passed, 4 skipped, 351 subtests passed. Its one deselected real-kernel test passes in a separate fresh invocation as required by the project.
- Source and touched-test Black checks pass; mypy reports no issues in 67 source files; source distribution and wheel builds pass.
- The real local binding API produces
DECIMAL(10,2) and DECIMAL(10,0) for explicit sizes, and preserves automatic DECIMAL(5,2) sizing. Named and positional bindings preserve their string payloads.
poetry run python -m pytest tests/unit/test_parameters.py -q
poetry run python -m pytest tests/unit -m 'not realkernel' -q -rs
poetry run python -m pytest tests/unit/test_session.py -m realkernel -v -rs
poetry run black --check src tests/unit/test_parameters.py
poetry run mypy --install-types --non-interactive src
poetry build
The four skips are existing upstream markers: two legacy interface cases, an outdated fetch benchmark, and a Thrift native-binding case. Eleven fixture/FutureWarning reports remain. No test, skip marker or warning filter was weakened.
Not run: live Databricks/e2e tests, kernel/SEA SQL-server behavior, other operating systems or the hosted CI matrix. No server-acceptance or performance claim.
Related Tickets & Documents
Refs #901 (explicit precision/scale and zero-scale handling only).
