There are a number of query plan constructs that are using temoprary (work) tables:

• "temporary_table" (GROUP-BY temp. table)
• "materialized" (derived table)
• "materialized" (non-merged semi-join materialization)
• "materialized" (merged semi-join materialization)
• "duplicates_removal" - temporary table used by Duplicate Weedout.
• subqueries: [ query_block: { ... }

  ] - subquery materialization

• ??? - subquery cache

It would be nice to be able to track all of these in a uniform way.
(Is TMP_TABLE_PARAM a good place to put the pointer to tracker helper object in?)

Overview of how this can be achieved:

Create a Work_table_tracker class.

• A Work_table_tracker object will store information about how the temp. table was used.
  = its storage engine
  = its #rows
  = its size-on-storage-media
• It will have setter-like methods for reporting information about the temp table.
• (TODO: how does this work with parts of the query plan that are re-executed multiple times?)

Some data structure will keep a pointer to Work_table_tracker so that free_tmp_table() can call the reporting methods.

"Explain data structures" also need pointers to Work_table_tracker objects so that they can print them in the ANALYZE FORMAT=JSON output.