Bug triage results: 2026-07-27

[andygrove]

Triage pass over the open requires-triage queue, per the project Bug Triage Guide.

• Date: 2026-07-27
• Total issues processed: 20 (16 triaged, 4 skipped, 0 failed)
• Type counts: 7 bugs, 9 enhancements
• Priority counts applied: priority:critical 2, priority:high 2, priority:medium 2, priority:low 1
• Guide: docs/source/contributor-guide/bug_triage.md

Labels have already been applied. A reviewer should spot-check the calls below and close this issue when satisfied; corrections should be made directly on the affected issue.

Bugs

priority:critical

• RANGE window frame boundary overflow diverges from Spark for DATE and DECIMAL ORDER BY (#5022)
  • Area labels: area:expressions
  • Rationale: Native and Spark produce different window frames on boundary overflow with no error — silent wrong results, which the guide defines as priority:critical.
• Object store cache key drops the ABFS container, so two containers in one storage account share a store instance (#4993)
  • Area labels: area:scan
  • Rationale: Reads from one ABFS container can silently return another container's data — a data-corruption / silent wrong-results case that the guide places at priority:critical.

priority:high

• Sporadic SIGSEGV in macOS [scans] workflow in CI (#5023)
  • Area labels: area:ci, area:scan
  • Rationale: JVM SIGSEGV in the native scan path — a crash on a supported code path, matching the guide's priority:high definition; kept off critical because it appears only sporadically in CI and has not been shown to produce wrong results.
• Native panic casting to negative-scale decimal when spark.sql.legacy.allowNegativeScaleOfDecimal=true (#5013)
  • Area labels: area:expressions
  • Rationale: Reproducible native panic on a supported cast path; priority:high per the crash rule. See Escalations below — the release build silently wraps instead of panicking, which is a critical-tier hazard once confirmed.

priority:medium

• Datetime rebase: track the documented scan limitation, and spark.comet.exceptionOnDatetimeRebase is dead code (#5010)
  • Area labels: area:scan, documentation, correctness
  • Rationale: Silent wrong results on legacy-calendar Parquet, but the reader has a documented workaround (spark.comet.scan.enabled=false) and the surface is narrow — pre-3.0 files or writes with datetimeRebaseModeInWrite=LEGACY. See Escalations below.
• Native Parquet scan ignores session-level datafusion.execution.parquet.* options (#4990)
  • Area labels: area:scan
  • Rationale: A documented config path is silently inert but the scan still returns correct data; broken feature with a workaround → priority:medium.

priority:low

• CI: non-gating label (e.g. auto-added performance) spawns a phantom all-skipped CI run that shadows real checks (#5007)
  • Area labels: area:ci
  • Rationale: CI/tooling issue — the real CI run still executes; the phantom run only cosmetically misleads the merge box → priority:low.

Enhancements

• Follow-up: docs and test-coverage hardening for native collect_list / array_agg (feat: native collect_list / array_agg aggregate #4720) (#5024)
  • Area labels: area:aggregation, area:expressions
  • Rationale: Deferred docs additions and test-coverage improvements from PR feat: native collect_list / array_agg aggregate #4720 review; nothing is broken, so this is enhancement work.
• Centralize data-type support predicates so new types don't have to be added in multiple places (#5021)
  • Area labels: none
  • Rationale: The reporter explicitly notes "This is a maintainability cleanup, not a correctness bug" — a refactor to consolidate duplicated type predicates.
• Allow individual cast conversions to be enabled/disabled (#5015)
  • Area labels: area:expressions
  • Rationale: New config to selectively disable individual (fromType, toType) cast pairs — feature addition to the Cast serde.
• perf: reuse Arrow IPC CompressionContext across shuffle blocks instead of per-batch allocation (#5008)
  • Area labels: area:shuffle, performance
  • Rationale: Shuffle-writer performance optimization — reuse an existing allocation across blocks; behavior is unchanged.
• Share Parquet footer and page index metadata across tasks in an executor (#4994)
  • Area labels: area:scan, performance
  • Rationale: New feature to share Parquet metadata caches across tasks in an executor; a performance improvement over parity with Spark.
• Support native Iceberg scans in Spark Structured Streaming (#4992)
  • Area labels: area:scan
  • Rationale: New functionality — allow the existing native Iceberg scan to be used in Structured Streaming queries that are currently skipped by CometExecRule.
• Row-level Parquet filter pushdown regresses uncorrelated-predicate scans (TPC-H Q6 +80%) (#4988)
  • Area labels: area:scan, performance
  • Rationale: Investigation into whether spark.comet.parquet.rowFilterPushdown.enabled can be turned on by default; the flag is off by default so no user is affected today — enhancement, not a user-visible regression.
• Upstream Spark-compatible approx_count_distinct (HyperLogLogPlusPlus) to the datafusion-spark crate (#4984)
  • Area labels: area:aggregation, area:expressions
  • Rationale: Refactor/upstreaming of an existing native aggregate to the upstream datafusion-spark crate; no behavior change.
• Implement TimeType support - Time extraction (from TimeType to Int/Decimal) (#4983)
  • Area labels: area:expressions
  • Rationale: New expression support for TimeType extraction — a subtask of the TimeType umbrella (Implement TimeType support #4288).

Escalations to consider

• Native panic casting to negative-scale decimal when spark.sql.legacy.allowNegativeScaleOfDecimal=true (#5013)
  • Guide trigger: "A priority:high crash is discovered to also produce wrong results silently in some cases → escalate to priority:critical." The reporter observed that debug builds panic but the release profile has overflow-checks = false, so a release build would silently wrap and produce wrong values. If confirmed on a release build this should move to priority:critical.
• Datetime rebase: track the documented scan limitation, and spark.comet.exceptionOnDatetimeRebase is dead code (#5010)
  • Guide trigger: correctness principle — "A bug that silently returns wrong results is worse than one that crashes." Left at priority:medium because the surface is narrow (legacy-calendar Parquet only) and a working workaround exists (spark.comet.scan.enabled=false), but a reviewer familiar with the deployed workloads may want to escalate to priority:critical.

Skipped — needs more info

• Bug triage results: 2026-07-20 (#4980)
  • Prior triage summary issue that was left carrying the auto-applied requires-triage label; not itself a bug or enhancement — reviewer should sanity-check the referenced results and close.
• Bug triage results: 2026-07-13 (#4905)
  • Prior triage summary issue that was left carrying the auto-applied requires-triage label; not itself a bug or enhancement — reviewer should sanity-check the referenced results and close.
• Bug triage results: 2026-07-06 (#4838)
  • Prior triage summary issue that was left carrying the auto-applied requires-triage label; not itself a bug or enhancement — reviewer should sanity-check the referenced results and close.
• Bug triage results: 2026-06-29 (#4751)
  • Prior triage summary issue that was left carrying the auto-applied requires-triage label; not itself a bug or enhancement — reviewer should sanity-check the referenced results and close.