[felipepessoto]

What changes were proposed in this pull request?

Removes 4 entries from the Delta-on-Gluten baseline (known-failures.txt) that now pass after #12051.

#12051 ("[GLUTEN-10134][VL] Preserve store assignment cast modes") makes CastTransformer forward the per-expression EvalMode (LEGACY / ANSI / TRY) to Velox through Substrait, instead of letting Velox derive cast behaviour from the session-level spark.sql.ansi.enabled. Previously EvalMode.LEGACY and EvalMode.ANSI serialized to identical Substrait nodes, so a LEGACY cast wrongly inherited ANSI overflow semantics whenever the session flag was on.

The Delta implicit-casting tests below run with storeAssignmentPolicy: LEGACY and ansiEnabled: true, i.e. Delta builds a Cast node with evalMode = LEGACY while the session flag is true. They expect the cast to return NULL rather than throw, so they now pass and must be dropped from the baseline:

org.apache.spark.sql.delta.ImplicitMergeCastingSuite#MERGE overflow in WHEN MATCHED THEN UPDATE SET t.value = s.value sourceType: BIGINT, targetType: DECIMAL(7,2) followAnsiEnabled: false, ansiEnabled: true, storeAssignmentPolicy: LEGACY
org.apache.spark.sql.delta.ImplicitMergeCastingSuite#MERGE overflow in WHEN MATCHED THEN UPDATE SET t.value = s.value sourceType: STRING, targetType: INT followAnsiEnabled: false, ansiEnabled: true, storeAssignmentPolicy: LEGACY
org.apache.spark.sql.delta.ImplicitStreamingMergeCastingSuite#Streaming MERGE overflow sourceType: BIGINT, targetType: DECIMAL(7,2) followAnsiEnabled: false, ansiEnabled: true, storeAssignmentPolicy: LEGACY
org.apache.spark.sql.delta.ImplicitStreamingMergeCastingSuite#Streaming MERGE overflow sourceType: STRING, targetType: INT followAnsiEnabled: false, ansiEnabled: true, storeAssignmentPolicy: LEGACY

This is required, not cosmetic: the gate runs with --fail-on-fixed true, so a baseline entry that starts passing turns the Delta workflow red until it is removed.

How was this patch tested?

Evidence comes from the scheduled Delta Spark UT (Gluten) run for commit 8439151, which contains the fix (95f280f is an ancestor of the run's head SHA):

https://github.com/apache/gluten/actions/runs/32101727188

• Shard 0 failed with NOW-PASSING (remove from baseline) for the two ImplicitStreamingMergeCastingSuite tests.
• Shard 3 failed with NOW-PASSING (remove from baseline) for the two ImplicitMergeCastingSuite tests.
• The delta-spark-aggregate job reports Now-passing (global) = 4 (exactly the lines removed here), alongside Regressions = 0, Stale = 0, and Skipped this run = 0.

Since the run reports no regressions and no stale entries, these 4 deletions are the only baseline change needed; the remaining 60 Implicit*MergeCastingSuite entries (other type/config combinations that still fail) are untouched.

Diff is 4 deletions and 0 additions; the file's sorted order and header are preserved.

Was this patch authored or co-authored using generative AI tooling?

Generated-by: GitHub Copilot CLI

Related issue: #10134

[Copilot]

Pull request overview

Removes four Delta-on-Gluten baseline entries from known-failures.txt that are now passing after #12051 preserved per-expression cast EvalMode semantics (avoiding incorrect ANSI overflow behavior when storeAssignmentPolicy=LEGACY with spark.sql.ansi.enabled=true).

Changes:

• Drop 2 ImplicitMergeCastingSuite baseline lines for storeAssignmentPolicy: LEGACY + ansiEnabled: true overflow cases.
• Drop 2 ImplicitStreamingMergeCastingSuite baseline lines for the same configuration/cases.
• Keep the remaining baseline intact (ordering and surrounding entries preserved).

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[ReemaAlzaid]

Thanks @felipepessoto! Confirmed these 4 tests are fixed by #12051 (LEGACY casts under ansiEnabled=true that previously ran as ANSI casts and threw on overflow)

cc:@philo-he

[philo-he]

Thank you!