Revert "[SPARK-47895][SQL] group by alias should be idempotent"

[mihailotim-db]

What changes were proposed in this pull request?

This PR reverts #50461 because it introduces a correctness issue by replacing an Alias with incorrect literal. A followup will be made with a different way to fix this issue.

Why are the changes needed?

In the below example, alias abc is replaced with 2 resulting in 2 rows instead of correct 3.
Before erronous PR:

After erronous PR:

Does this PR introduce any user-facing change?

User now sees an error message instead of an incorrect result.

How was this patch tested?

Added a test case to check for this behavior in the future

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

No

[mihailom-db]

LGTM, but we still need to fix the problem of reanalysis for aliases

[dongjoon-hyun]

cc @cloud-fan

[cloud-fan]

This hack only works if the resolved integer literal is the root expression, which may not be true here.

Good catch!

[cloud-fan]

thanks, merging to master/4.0!

[dongjoon-hyun]

For the record, the JIRA ID seems to be reused mistakenly in the original PR and this revert unfortunately. SPARK-47895 was group by all ... instead of group by alias ....

$ git log --oneline | grep SPARK-47895
8718eba1f6f Revert "[SPARK-47895][SQL] group by alias should be idempotent"
02db87282da [SPARK-47895][SQL] group by alias should be idempotent
b5bb75ca240 [SPARK-47895][SQL] group by all should be idempotent

Could you confirm that, @mihailotim-db ?

[mihailotim-db]

Correct, but the root of the issue is the same so I guess that is why they reused the same ticket

[dongjoon-hyun]

Correct, but the root of the issue is the same so I guess that is why they reused the same ticket

Thank you for the comment. It seems that there is misunderstanding here if it's done intentionally.

To @mihailotim-db and @mihailom-db . I just want to give you the context in order to prevent future accident like the following.

02db87282da [SPARK-47895][SQL] group by alias should be idempotent
b5bb75ca240 [SPARK-47895][SQL] group by all should be idempotent

ASF JIRA IDs have Fixed Versions field. Even for the direct follow-up PRs, we cannot reuse JIRA IDs when the original patch is released already. It's because the original PR and the follow-up PR cannot have the same Fixed Versions field. Please make it sure that ASF JIRA ID is different from any internal customer ticket systems.

Since this reverting PR itself has no problem, let me post the same comment on the original PR, too.

• [SPARK-47895][SQL] group by alias should be idempotent #50461

[mihailotim-db]

@dongjoon-hyun Good to know, thanks!