WIP: MDEV-38975 Phase 1 — Promote wide VARCHAR to BLOB for HEAP internal temp tables

[arcivanov]

Summary

Depends on: branch MDEV-38975 (HEAP BLOB/TEXT support) — this PR is a single commit on top of that branch and requires it to be merged first.

This change has the exact same semantics and side effects as the existing CONVERT_IF_BIGGER_TO_BLOB (512 chars) promotion, just at a lower HEAP-specific threshold of 32 bytes (not characters). The threshold is in bytes because HEAP waste is in bytes: the same VARCHAR(100) wastes 100 bytes in latin1, 300 in utf8mb3, and 400 in utf8mb4. A 32-byte floor ensures promotion helps at any encoding.

Why: HEAP uses fixed-width rows, so every VARCHAR allocates its full declared octet_length per row regardless of actual data. For I_S tables like COLUMNS (12,560 bytes/row baseline), promoting VARCHARs wider than 32 bytes to BLOBs collapses the primary record to 312 bytes and stores actual data in continuation chains, reducing per-row memory by ~75%.

Promotion paths:

• I_S temp tables via add_schema_fields()
• Expression-path temp tables via creating_heap_tmp_table flag in varstring_type_handler()

Side effects (identical to CONVERT_IF_BIGGER_TO_BLOB):

• CREATE TABLE ... AS SELECT from promoted temp tables produces BLOB columns where VARCHAR was expected. This is a pre-existing limitation of CONVERT_IF_BIGGER_TO_BLOB type leakage into permanent tables, not new behavior introduced here. Fixing this across all thresholds has a large blast radius and should be a separate MDEV.
• Result set metadata shows MYSQL_TYPE_BLOB instead of MYSQL_TYPE_VARCHAR for promoted columns.
• Sysschema DESC output shows text/tinytext/mediumtext/longtext instead of varchar(N) for promoted columns (cosmetic, result files re-recorded).
• innodb.innodb-mdev-7408 test adapted to use explicit CREATE TABLE + INSERT ... SELECT instead of CREATE TABLE ... AS SELECT to avoid InnoDB stopword type validation failure.

Test plan

• All sysschema tests pass (89/89)
• innodb.innodb-mdev-7408 passes
• heap.* suite passes (12/12)
• innodb_fts.* suite passes (98/98)
• Subselect tests pass (subselect_mat, subselect_sj_mat, sj_mat_debug, etc.)
• Full CI run

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
10 out of 17 committers have signed the CLA.

✅ ayurchen
✅ dbart
✅ janlindstrom
✅ sjaakola
✅ ParadoxV5
✅ arcivanov
✅ gkodinov
✅ hadeer-r
✅ mariadb-TafzeelShams
✅ nadaelsayed11
❌ Alexey Botchkov
❌ dr-m
❌ vuvova
❌ mariadb-poojalamba
❌ midenok
❌ plampio
❌ Thirunarayanan

---

Alexey Botchkov seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}