branch-4.1 [opt](fe) Bound not-null inference cost #63318

[englefly]

pick #63318

[hello-stephen]

Thank you for your contribution to Apache Doris.
Don't know what should be done next? See How to process your PR.

Please clearly describe your PR:

1. What problem was fixed (it's best to include specific error reporting information). How it was fixed.
2. Which behaviors were modified. What was the previous behavior, what is it now, why was it modified, and what possible impacts might there be.
3. What features were added. Why was this function added?
4. Which code was refactored and why was this part of the code refactored?
5. Which functions were optimized and what is the difference before and after the optimization?

[englefly]

run buildall

[morrySnow]

run buildall

[hello-stephen]

FE Regression Coverage Report

Increment line coverage 61.47% (67/109) 🎉
Increment coverage report
Complete coverage report

[englefly]

/review

[github-actions]

I found one optimizer regression in the new not-null inference budget logic. Existing inline review context was empty, and there were no additional user focus points.

Critical checkpoints:

• AGENTS.md Nereids expression checkpoint: not applicable; this PR does not add or change overloaded checkLegalityBeforeTypeCoercion methods.
• Correctness: needs attention because cheap predicates can now be skipped based on unrelated predicate iteration order.
• Tests: I attempted to run the focused FE tests with ./run-fe-ut.sh --run org.apache.doris.nereids.rules.rewrite.InferAggNotNullTest,org.apache.doris.nereids.rules.rewrite.InferJoinNotNullTest,org.apache.doris.nereids.rules.rewrite.InferFilterNotNullTest,org.apache.doris.nereids.rules.rewrite.EliminateNotNullTest, but the runner failed during generated-code setup because thirdparty/installed/bin/protoc is missing.

User focus response: no additional user-provided review focus was supplied.

[github-actions]

This global slot budget makes inference depend on the iteration order of the predicate collection. If an earlier cheap predicate uses 32 input slots, a later simple predicate like x = 1 reaches this branch and is skipped only because x would be the 33rd accumulated slot, so InferFilterNotNull will not infer x IS NOT NULL and EliminateNotNull will not remove an explicit redundant x IS NOT NULL. Several callers pass sets, so two equivalent predicate sets can get different not-null inference depending on iteration order. The budget should be applied per predicate, or the cumulative cap should not prevent independently cheap predicates from being considered.