[
https://issues.apache.org/jira/browse/CALCITE-7757?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18111209#comment-18111209
]
Mihai Budiu commented on CALCITE-7757:
--------------------------------------
this is quite a list. You may want to file each one as a separate linked Jira
issue and solve them in separate PRs.
> RelMdFunctionalDependency can infer unsound dependencies and cause incorrect
> query results
> ------------------------------------------------------------------------------------------
>
> Key: CALCITE-7757
> URL: https://issues.apache.org/jira/browse/CALCITE-7757
> Project: Calcite
> Issue Type: Bug
> Components: core
> Affects Versions: 1.42.0
> Reporter: Darpan Lunagariya (e6data)
> Assignee: Darpan Lunagariya (e6data)
> Priority: Major
> Labels: pull-request-available
>
> h2. Summary
> {{RelMdFunctionalDependency}} currently infers functional dependencies that
> do not always hold under SQL null, grouping, ordinal, and value-equality
> semantics.
> {{AggregateRemoveDuplicateKeysRule}} consumes this metadata and can
> consequently remove a necessary grouping key, changing query results.
> h2. Correctness problems
> # *Outer-join equality dependencies* - {{LEFT}} and {{RIGHT}} joins must not
> infer bidirectional functional dependencies from equality predicates when
> null generation invalidates one direction.
> # *Dependencies from null-generated inputs* - Outer joins must not blindly
> preserve functional dependencies from the null-generated input because null
> padding can invalidate them.
> # *Incorrect Aggregate ordinal mapping* - Aggregate input group ordinals must
> be mapped to packed output positions before exposing their functional
> dependencies.
> # *Equality-derived dependencies for approximate numerics* - Equality
> predicates involving {{FLOAT}} or {{DOUBLE}} must not produce functional
> dependencies when SQL equality and grouping equality differ.
> # *Unsafe derived-expression dependencies* - Determinism alone is
> insufficient for removing derived grouping expressions; unsafe scalar and
> nested types must be rejected.
> # *Grouping-set aggregate dependencies* - Grouping columns must not be
> assumed to determine aggregate results when grouping sets can produce
> identical null-padded keys.
> # *Projected TableScan ordinals* - Table keys expressed in base-table
> ordinals must not be applied directly to projected or reordered {{TableScan}}
> output columns.
> # *Unsafe generic unary-node passthrough* - Unknown single-input relational
> nodes must not automatically inherit input functional dependencies because
> they may change schema, values, or cardinality.
> # *Join offsets with system fields* - Functional-dependency ordinals for join
> inputs must account for system fields prefixed to the output, including semi
> and anti joins.
> # *Equality-derived dependencies for collated values* - Equality using a
> custom collator can consider strings equal even when Enumerable grouping
> distinguishes their Java keys, making the inferred dependency unsafe.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
