baibaichen opened a new pull request, #11975:
URL: https://github.com/apache/gluten/pull/11975
## What changes are proposed in this pull request?
This PR establishes an ANSI SQL mode CI baseline for the Velox backend,
providing expression-level offload tracking to identify the full scope of ANSI
implementation work.
**Commit 1 — Offload tracking framework**
- Add `GlutenExpressionOffloadTracker` trait: per-expression
OFFLOAD/FALLBACK/FAIL tracking with structured JSON output
- Refactor `GlutenTestsTrait` for expression-level exception handling via
DataFrame → Velox path
- Mixin tracker to 22 ANSI-related test suites (spark40 + spark41)
- Enable `GlutenCastWithAnsiOnSuite` in VeloxTestSettings
- Add `dev/verify-ansi-expressions.sh` for local matrix testing
**Commit 2 — ANSI CI workflow with AI-powered analysis**
- Add GitHub Actions workflow (`velox_backend_ansi.yml`) with two trigger
modes:
- `/ansi-test` (PR comment): full build + test + analyze
- `/ansi-analyze` (PR comment): analyze-only using artifacts from latest
run
- Manual `workflow_dispatch` with mode selector
- Add `analyze-ansi.py`: parses tracker JSON, produces four-color
classification report (🟢 Passed, 🟡 Failed, 🔴 Fallback, 🟠 Failed+Fallback),
optionally calls AI for root-cause analysis
- Add `shared.md`: single-source-of-truth analysis prompt consumed by both
CI script and local AI agent
**Commit 3 — SKILL entry point**
- Add `.github/skills/ansi-analysis/SKILL.md` for local AI-assisted analysis
(Claude Code / Copilot)
> **Purpose:** Establish ANSI mode CI baseline and identify the full scope
of ANSI implementation work. Test failures are expected — the tracker
classifies them to guide prioritization.
Related issue: #10134
Supersedes: #11961
## How was this patch tested?
- Triggered full ANSI CI workflow (`/ansi-test`) on
[baibaichen/gluten#6](https://github.com/baibaichen/gluten/pull/6)
- Triggered analyze-only mode (`/ansi-analyze`) to verify artifact reuse and
report generation
- Local `dev/verify-ansi-expressions.sh all spark41` run: 498 tests, 43250
expression records (87.3% pass, 12.5% fallback, 0.1% fail)
- Existing non-ANSI tests unaffected (tracker is only active when
`SPARK_ANSI_SQL_MODE=true`)
## Was this patch authored or co-authored using generative AI tooling?
Generated-by: Claude Code (Claude Opus 4)
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
