MDEV-40672: Pluggable Aggregate Function

[drrtuy]

Add pluggable aggregate function support

What

Implement MDEV-40672 by extending function plugins to provide aggregate functions through the standard Item_sum lifecycle. Support grouped aggregation, DISTINCT, window functions, native and pluggable data types, and safe plugin lifetime management.

Key changes

• Extend Plugin_function descriptors to distinguish scalar and aggregate functions.
• Add Item_sum_plugin as the base class for plugin-provided aggregates.
• Support DISTINCT argument replay and pluggable result types such as UUID.
• Enable plugin aggregates as window functions, including moving-frame removal.
• Add test aggregates and MTR coverage for grouping, windows, prepared statements, plugin unloading, invalid descriptors, and type preservation.

How to test

Run:

/git/BuildOf_mdb-13/mysql-test/mtr function_plugin function_plugin_extra

Both tests pass.

[Copilot]

Pull request overview

Adds pluggable aggregate functions to MariaDB’s standard aggregation and window-function infrastructure.

Changes:

• Introduces Item_sum_plugin with plugin lifetime and data-type handling.
• Adds parser, DISTINCT, and window-function support.
• Adds test aggregates and MTR coverage.

Reviewed changes

Copilot reviewed 17 out of 17 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
sql/sql_yacc.yy	Parses and constructs plugin aggregates.
sql/sql_window.cc	Handles plugin window aggregates.
sql/sql_schema.h	Exposes plugin references during lookup.
sql/sql_schema.cc	Implements plugin-aware schema lookup.
sql/item_sum.h	Defines plugin aggregate interfaces.
sql/item_sum.cc	Implements lifecycle and DISTINCT replay.
sql/item_create.h	Adds aggregate builder marker.
sql/item_create.cc	Retains plugins during item creation.
plugin/func_test/plugin.cc	Adds test aggregate plugins.
plugin/func_test/mysql-test/func_test/function_plugin.test	Adds primary functional tests.
plugin/func_test/mysql-test/func_test/function_plugin.result	Records primary test results.
plugin/func_test/mysql-test/func_test/function_plugin_negative.test	Adds failure-path tests.
plugin/func_test/mysql-test/func_test/function_plugin_negative.result	Records negative test results.
plugin/func_test/mysql-test/func_test/function_plugin_extra.test	Adds extended lifecycle/type tests.
plugin/func_test/mysql-test/func_test/function_plugin_extra.result	Records extended test results.
include/mysql/plugin_function.h	Documents the aggregate plugin contract.
include/mysql/plugin_function.h.pp	Updates the preprocessed API header.

Suppressed comments (3)

sql/sql_yacc.yy:11375

• The descriptor contract is only checked when the returned object already casts to Item_sum. An aggregate-marked builder that accidentally returns a scalar skips this block and is accepted even though its arguments were parsed as an aggregate; conversely, a scalar-marked builder returning Item_sum_plugin is accepted as an aggregate. Validate the builder's declared aggregate kind against the returned item (when non-null), and require every aggregate descriptor to return Item_sum_plugin.

            Item_sum *sum_item= item ? dynamic_cast<Item_sum *>(item) : NULL;
            if (function_plugin && sum_item)

sql/item_sum.cc:582

• This null check does not make the allocation recoverable: ordinary throwing new never returns null (and typically terminates in no-exception builds), so allocation failure cannot reach the caller's ER_OUT_OF_RESOURCES path. Use a non-throwing allocation here.

  m_plugin_lifetime=
    new Item_sum_plugin_lifetime(static_cast<plugin_ref>(plugin));
  return !m_plugin_lifetime;

sql/item_sum.cc:905

• Routing plugin aggregates through this DISTINCT branch leaves their add() failures unhandled for the common in-memory tree replay path. unique_walk_function() discards the boolean returned by item_sum->add(), and endup() also ignores tree->walk()'s status, so a plugin that reports an allocation or evaluation failure can continue replaying and return a partial result. Propagate the callback failure and stop/fail the walk, as the new on-disk replay loop already does.

  if (item_sum->sum_func() == Item_sum::COUNT_FUNC ||
      item_sum->sum_func() == Item_sum::COUNT_DISTINCT_FUNC ||
      item_sum->sum_func() == Item_sum::PLUGIN_SUM_FUNC)

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[drrtuy]

Known issue. Similar situation is filed as MDEV-20842.