LuciferYang opened a new issue, #13088:
URL: https://github.com/apache/gluten/issues/13088

   ### Description
   
   After the Spark 3.3 removal (#12807) and the follow-up cleanup (#13003), two 
Spark 3.2/3.3 references remain that I did not want to change unilaterally, 
because each needs a decision from the code's owner rather than a mechanical 
edit.
   
   ### 1. `Makefile`: dead `-Pspark-3.3` / `-Pspark-3.2` / `-Pspark-3.2-las` 
targets
   
   The bolt `Makefile` (added with #12454, not wired into any CI workflow) 
still drives several targets off profiles that no longer exist:
   
   - `jar` (`:158`) builds with `-Pspark-3.3`
   - `jar-skip-check` (`:164`), `test` (`:211`), `clean` (`:224`) build with 
`-Pspark-3.2`
   - `spark32-las` (`:169-170`) builds with `-Pspark-3.2-las`
   - `jar_spark33` (`:192-196`) builds with `-Pspark-3.3`; `jar_spark33` is 
also in `.PHONY` (`:83`)
   - `fast-jar` / `zip` / `fast-zip` and the 
`jar`/`jar-skip-check`/`spark32-las` output names are hardcoded 
`gluten-spark3.2_2.12-...` (already inconsistent, since `jar` builds 3.3 but 
names the artifact 3.2)
   
   Because Maven only warns on an unknown `-P`, these fall through to the 
root-pom default properties (`sparkbundle.version=3.5` / `sparkshim=spark35` 
but `spark.version=4.0.2`) with `-Denforcer.skip=true` suppressing the guard, 
so `make jar` today produces a nonsense build rather than an error.
   
   `jar_spark34` / `jar_spark35` / `test_spark35` are the working targets. The 
fix is presumably to repoint the defaults to a supported version (3.5 matches 
the repo default and bolt builds on it) and drop the 3.2/3.3-only targets, but 
the default target and the output-naming scheme are bolt's call, so I am 
leaving it here rather than guessing. cc the bolt owners.
   
   ### 2. `cpp-ch/local-engine/tests/json/native_write_plan_1_spark33.json`
   
   This fixture, and its sibling `native_write_plan_1_spark35.json`, appear 
unreferenced: the only `INCBIN` loads in `gtest_write_pipeline.cpp` are 
`native_write_plan.json` and `native_write_one_partition.json`, and a tree-wide 
grep for `native_write_plan_1_spark` finds no loader. If they are genuinely 
dead, the `spark33` one is a 3.3 leftover and the `spark35` one is orphaned 
too. I cannot build or run the ClickHouse tests locally, so I would rather a CH 
maintainer confirm these are unused before they are deleted. cc the ClickHouse 
owners.
   
   ### Handled separately
   
   The `tools/workload/benchmark_velox/initialize.ipynb` reference to 
`spark-3.3.1` (it provisioned a Spark that Gluten can no longer run on) is a 
straightforward version bump and is fixed in the PR that references this issue, 
together with a stale `Spark 3.3/3.4` comment in `WriteFilesExecTransformer`.
   


-- 
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]

Reply via email to