fix(ci): Disable sparse-checkout cone mode for Claude context files (f…

…acebookincubator#17713)

Summary:
The `CI Failure Comment` workflow has been failing on every run since git 2.54 landed on `ubuntu-latest` runners (around May 20). The `Checkout for Claude context and prompt` step passes file paths to sparse-checkout:

```yaml
sparse-checkout: |
  CLAUDE.md
  .claude/CLAUDE.md
  .claude/skills/ci-failure-analysis/SKILL.md
```

`actions/checkout` enables cone mode by default, and git 2.54 hard-rejects non-directory entries in cone mode:

```
fatal: '.claude/CLAUDE.md' is not a directory; to treat it as a directory anyway, rerun with --skip-checks
```

Earlier git versions on the runner (e.g. 2.53.0 on May 15 runs) accepted file paths in cone mode, which is why the same YAML used to work.

This change sets `sparse-checkout-cone-mode: false` on the affected step so file paths are accepted. The earlier `.github/scripts` checkout stays in cone mode — it is a directory and works fine.

Pull Request resolved: facebookincubator#17713

Reviewed By: srsuryadev

Differential Revision: D107456949

Pulled By: kgpai

fbshipit-source-id: 62aded58697523d51a0cd9ddaea501d3327a3fb1

Fix iceberg min max statistics for decimal type when encoded as int32

Signed-off-by: Hazmi <[email protected]>

Alchemy-item: (ID = 1203) Fix iceberg min max statistics for decimal type when encoded as int32 commit 1/1 - 0ac9930

[OAP] fix iceberg reader to use split reader

Signed-off-by: Yuan <[email protected]>
Signed-off-by: Mohammad Linjawi <[email protected]>
(cherry picked from commit 4cd49ab)
(cherry picked from commit 4fc58be)

Alchemy-item: (ID = 1537) [OAP] fix to use spliter iceberg reader commit 1/1 - 064cc0e

feat: Support casting from large negative-exponent string to decimal (f…

…acebookincubator#17665)

Summary:
The PR adds support for casting from large negative-exponent string to decimal zero, to align with Spark and Presto.

E.g., the PR adds support for `CAST('6E-120' AS DECIMAL(38, 0)) == 0`, instead of returning an error.

This is the step 2 in facebookincubator#17593, following the previous fix facebookincubator#17594.

Pull Request resolved: facebookincubator#17665

Reviewed By: apurva-meta

Differential Revision: D107421146

Pulled By: bikramSingh91

fbshipit-source-id: 11dc31328bed487036abb00f75f9c4fd3b281e76

Fix iceberg min max statistics for decimal type when encoded as int32

Signed-off-by: Hazmi <[email protected]>

Alchemy-item: (ID = 1203) Fix iceberg min max statistics for decimal type when encoded as int32 commit 1/1 - 0ac9930

[OAP] fix iceberg reader to use split reader

Signed-off-by: Yuan <[email protected]>
Signed-off-by: Mohammad Linjawi <[email protected]>
(cherry picked from commit 4cd49ab)
(cherry picked from commit 4fc58be)

Alchemy-item: (ID = 1537) [OAP] fix to use spliter iceberg reader commit 1/1 - 064cc0e

feat(cudf): Add Iceberg support to velox-cudf (facebookincubator#17054)

Summary:
## Description

This PR implements GPU-accelerated Iceberg support in Velox-cuDF.  Non-exhaustive list of implemented features:

- Adds a new set of `CudfIcebergXX` classes derived from `CudfHiveXX` bases. Same class setup as in `velox/connectors/hive`
- Similar to the base `CudfHiveConnector`, `CudfIcebergConnector` only supports Parquet data files.
- Delete files may be in any format supported by the Velox `RowReader`s.
- The workhorse class: `CudfIcebergSplitReader` derives from the `CudfSplitReader` and adds deletion and schema evolution (contributed by ducndh) support to the read `cudf::table`s.
- GPU accelerated roaring bitmaps are applied using `cudf::roaring_bitmap` (V3).
- Uses Velox's `PositionalDeleteFileReader` to fetch delete positions (V2).
- Adds a new `CudfEqualityDeleteFileReader` for GPU accelerated equality delete support (V2).
- Lots and lots of new tests.
- Reading files using multiple splits using the `splitOffset` field is not yet supported.

## Checklist
- [x] All velox-cudf tests are passing
- [x] I am familiar with the contributing guide

Pull Request resolved: facebookincubator#17054

Reviewed By: kevinwilfong

Differential Revision: D106419556

Pulled By: kKPulla

fbshipit-source-id: 32c06cd381c3ffd59e01f47c9933e24350ab46f5

perf: Remove TraceContext from DWIO I/O hot paths (facebookincubator#…

…17693)

Summary:
Pull Request resolved: facebookincubator#17693

CONTEXT: Strobelight profiling shows `TraceContext::TraceContext` consuming significant CPU on the I/O read path. Each constructor call allocates a `std::string` label, reads `steady_clock::now()`, pushes to `TraceHistory`, and updates the thread-local registry — all on every batch read, column load, and I/O operation.

WHAT: Remove all `process::TraceContext` usage from the DWIO I/O hot paths:
- `ColumnLoader::loadInternal` (called per lazy column load)
- `TransformColumnLoader::loadInternal`
- `DeltaUpdateColumnLoader::loadInternal`
- `SelectiveStructColumnReaderBase::next` (called per batch)
- `DirectInputStream::loadSync` (called per I/O read)
- `CacheInputStream::loadSync` (called per cache load)
- `CachedBufferedInput` and `DirectBufferedInput` read-ahead lambdas
- `CacheInputStream` error-path `TraceContext::statusLine()` call

Reviewed By: tanjialiang

Differential Revision: D107201661

fbshipit-source-id: 2ab58bb26c2616aef95e04745920ba92c45b8b5e

Fix iceberg min max statistics for decimal type when encoded as int32

Signed-off-by: Hazmi <[email protected]>

Alchemy-item: (ID = 1203) Fix iceberg min max statistics for decimal type when encoded as int32 commit 1/1 - 0ac9930

Fix iceberg min max statistics for decimal type when encoded as int32

Signed-off-by: Hazmi <[email protected]>

Alchemy-item: (ID = 1203) Fix iceberg min max statistics for decimal type when encoded as int32 commit 1/1 - 0ac9930