Implement ODBC catalog metadata

[Shfdis]

Summary

• implement SQLTables special enumeration and metadata-ID semantics
• resolve SQLColumns, SQLPrimaryKeys, and SQLStatistics relative to the current catalog
• report complete column descriptors, composite primary keys, and index statistics
• preserve null versus empty metadata arguments and escaped SQL LIKE matching
• add native regression coverage for catalog paths, descriptors, validation, and statistics

[Shfdis]

@codex review

[robot-vibe-db]

AI Review Summary

Verdict: ✅ No critical issues found

Critical issues

No critical issues found.

Other findings

• Major | High: catalog metadata traversal no longer prunes to the pattern's literal prefix — every SQLTables/SQLColumns/SQLPrimaryKeys/SQLStatistics call recursively lists the entire database tree, even for exact-name lookups — odbc/src/statement_metadata.cpp:659
• Minor | Medium: SQLTables special-value handling (SQL_ALL_CATALOGS / SQL_ALL_SCHEMAS / SQL_ALL_TABLE_TYPES) runs even when SQL_ATTR_METADATA_ID = SQL_TRUE, where those values are supposed to be treated as literal identifiers — odbc/src/statement_metadata.cpp:388
• Minor | Medium: with SQL_ATTR_METADATA_ID = SQL_FALSE, catalog arguments no longer support pattern matching in SQLColumns/SQLTables (e.g. catalog "%" now yields an empty result set instead of the current catalog); only the fully-empty SQLTables combination handles SQL_ALL_CATALOGS — odbc/src/statement_metadata.cpp:333
• Minor | Medium: table name patterns are matched against the absolute path, so patterns like "prefix%" or "%" miss tables located in subdirectories ("%" matches only the top level while a null argument enumerates everything) — odbc/src/statement_metadata.cpp:663
• Minor | Low: CARDINALITY is silently reported as NULL for tables with more rows than INT32_MAX (Int32OrNull), losing the statistic instead of reporting a value — odbc/src/statement_metadata.cpp:480
• Minor | Low: no test covers SQLStatistics with SQL_INDEX_ALL when both unique and non-unique indexes exist (ordering by NON_UNIQUE then INDEX_NAME), since the unique-index coverage only queries SQL_INDEX_UNIQUE — odbc/tests/integration/metadata_api_it.cpp:666
• Nit | Low: the SQLTables result-set sort comparator allocates temporary strings (GetTableType/GetMetadataTableName) on every comparison — odbc/src/statement_metadata.cpp:426
• Nit | Low: a trailing lone backslash in a LIKE pattern is treated as a literal backslash rather than an invalid/malformed escape — odbc/src/utils/sql_like.h:30

---

This review was generated automatically. Critical issues require attention; other findings are advisory.
If this comment was useful, please give it a 👍 — it helps us improve the review bot.

[robot-vibe-db]

Full analysis log

Analysis performed by claude, z-ai/glm-5.3-flash.

[robot-vibe-db]

AI Review Summary

Verdict: ✅ No critical issues found

Critical issues

No critical issues found.

Other findings

• Major | High: SQLColumns passes catalogPatternsAllowed=false, so catalog search patterns (e.g. "%" or "loca%") are compared literally against the single catalog and return an empty result set instead of matching it — per ODBC, CatalogName is a pattern argument when SQL_ATTR_METADATA_ID=SQL_FALSE — odbc/src/statement_metadata.cpp:352
• Minor | Medium: SQLStatistics rejects SQL_INDEX_HASH with HY100; ODBC defines SQL_INDEX_HASH as a valid Unique value, and a driver without hash-index support should return an empty result set rather than an error — odbc/src/statement_metadata.cpp:468
• Minor | Medium: SQLSpecialColumns was not migrated to the new null/empty metadata-argument semantics: null pointers are silently converted to empty strings, so with SQL_ATTR_METADATA_ID=SQL_TRUE a null catalog/schema/table is not rejected with HY009 like the other catalog APIs, and an empty TableName now yields an empty result set instead of a list of candidates — odbc/src/statement_metadata.cpp:535
• Minor | Medium: the SQL_ALL_TABLE_TYPES special enumeration is not gated on patternsAllowed; with SQL_ATTR_METADATA_ID=SQL_TRUE (where "%" is documented as a literal identifier and the new HY009 rules apply) passing TableType="%" still returns the type list, inconsistent with the strict semantics introduced in this PR — odbc/src/statement_metadata.cpp:420
• Minor | Medium: no integration coverage for catalog arguments in SQLColumns (e.g. catalog "%" with MetadataId=SQL_FALSE); such a test would have caught the catalog-pattern regression — odbc/tests/integration/metadata_api_it.cpp:568
• Minor | Low: GetBufferLength falls back to GetColumnSize for character types, reporting the character count instead of the byte length (spec: BUFFER_LENGTH is the size in bytes, including the terminator, needed to transfer the column) — odbc/src/statement_metadata.cpp:233
• Minor | Low: null and empty-string TableType are both treated as "match all types" (MatchesTableTypeFilter with an empty filter returns true); per ODBC an empty list of table types is distinct from a null pointer — odbc/src/statement_metadata.cpp:443
• Nit | High: directory pruning builds a temporary fullPath + "/" string for every visited node; comparing prefix bounds directly (e.g. via std::string_view::starts_with on the parent path plus the separator length) avoids the allocation on large catalogs — odbc/src/statement_metadata.cpp:682

---

This review was generated automatically. Critical issues require attention; other findings are advisory.
If this comment was useful, please give it a 👍 — it helps us improve the review bot.

[robot-vibe-db]

Full analysis log

Analysis performed by claude, z-ai/glm-5.3-flash.