[test](ci) Fresh Codex review prompt validation - #49

force_eager_agg_hint affects Nereids rewrite decisions, but it is not forwarded when a statement is proxied to another FE. getForwardVariables() only includes variables marked needForward, affectQueryResultInPlan, or affectQueryResultInExecution, and FEOpExecutor sends exactly that map to ConnectProcessor.proxyExecute. Since this annotation has none of those flags, the target FE sees the default empty hint map in EagerAggHints.currentHints() and can generate a different plan from the origin FE. Please mark this variable for forwarding/result-plan propagation and add a forward/restore assertion to SessionVariablesTest.

This new suite currently fails git diff --check: trailing whitespace is reported here and at several later lines in this file, and the generated .out has an extra blank line at EOF. Please strip the trailing spaces and regenerate or clean the output file so the repository whitespace check passes.

This order_qt block contains two statements: the aggregate query and then select * from pdagg_proj_t2. The generated -- !sum_to_2_side -- output is the four raw pdagg_proj_t2 rows, so the test is not checking the aggregate result it is named for. Please remove/split the debug select * and regenerate the .out so this eager-aggregation case is actually asserted.

This removes the old PushDownDistinctThroughJoin rewrite from the eager-aggregation stage, but the PR also deletes its unit/regression coverage and does not add equivalent coverage for the new context.getAggFunctions().isEmpty() path inside PushDownAggregation. The old regression also covered USE_CBO_RULE(push_down_distinct_through_join)/NO_USE_CBO_RULE(...), while PUSH_DOWN_DISTINCT_THROUGH_JOIN still remains in the hint blacklist. Please add replacement coverage for pure distinct-through-join and either preserve/redirect the old hint behavior or remove the stale hint surface intentionally.

needOutputCountForJoinChild() asks children of both inner and cross joins to produce a count slot, but computeJoinCount() only builds that slot for INNER_JOIN. For a multi-level cross join such as agg(sum(t1.v), sum(t2.v), sum(t3.v) group by t1.k) -> CrossJoin(CrossJoin(t1, t2), t3), the upper join needs the left subtree's row count to multiply sum(t3.v). The lower cross join registers no count slot here, so the upper project cannot recover that multiplicity and undercounts the right-side aggregate. Please handle CROSS_JOIN the same way as inner join when computing joinCnt, and add a multi-level cross-join regression.

This hint names PUSH_DOWN_AGG_THROUGH_JOIN_ONE_SIDE, but this PR removes that rewrite job from Rewriter; the active replacement is PUSH_DOWN_AGG_THROUGH_JOIN via PushDownAggregation. CostBasedRewriteJob only applies NO_USE_CBO_RULE when the current job's RuleType matches the hint name, so this hint is unused and does not disable the new eager-aggregation path. Please update the hint to the active rule or remove it and regenerate the expected output; the same stale hint appears again below.

This reset leaves the session in eager_aggregation_mode = -1, but the default in SessionVariable is 0; -1 explicitly disables eager aggregation. If the regression connection is reused after this suite, later cases inherit a non-default planner setting. Please reset this back to 0 (and regenerate output only if needed).

A rewritten FULL_OUTER_JOIN can be asked to publish a subtree row-count slot when it is the child of an upper bilateral join. One-side pushdown through full outer is still allowed when the aggregate references that side, but computeJoinCount() has no full-outer branch and falls through to Optional.empty(), so the upper join cannot multiply the opposite branch by the full-outer subtree cardinality. For example, agg(sum(t1.v), sum(t3.v)) -> InnerJoin(FullOuterJoin(t1,t2), t3) can leave sum(t3.v) unmultiplied by the rows produced by the full outer subtree. Please either compute the full-outer joinCnt from the available child counts or block count-dependent nested pushdown through full outer joins, and add a regression for this shape.