refactor: Change HANA trigger to be statement based

[stefanrudi]

Refactor: Convert HANA Triggers to Statement-Level

♻️ Refactor: Migrated HANA database triggers from row-level (FOR EACH ROW) to statement-level (FOR EACH STATEMENT) execution, replacing :old/:new row references with transition table aliases (ot/nt via REFERENCING OLD TABLE old_tab / REFERENCING NEW TABLE new_tab).

Changes

This refactor eliminates the dependency on a dummy table (CHANGE_TRACKING_DUMMY) and rewrites all trigger SQL generation to use set-based INSERT ... SELECT patterns instead of per-row procedural logic.

• index.cds: Removed the CHANGE_TRACKING_DUMMY entity definition that was previously required as a workaround for HANA HDI triggers.
• lib/TriggerCQN2SQL.js: Updated trigger row reference regex to support nt/ot table aliases; removed the custom from() override that handled the now-deleted dummy table.
• lib/hana/triggers.js: Switched all three trigger types (INSERT, UPDATE, DELETE) from REFERENCING NEW/OLD ROW to REFERENCING NEW/OLD TABLE ... FOR EACH STATEMENT. Added getFromClause() helper to generate the appropriate FROM clause with transition table aliases. Updated all row references from new/old to nt/ot. Rewrote buildInsertSQL to embed all column expressions directly in the UNION ALL sub-selects, removing the outer wrapping SELECT.
• lib/hana/composition.js: Updated all composition parent context builders to use transition table aliases (nt/ot) instead of :new/:old row references. Replaced procedural DECLARE/IF/END IF SQL blocks with set-based INSERT ... SELECT DISTINCT patterns. Removed buildParentLookupSQL (row-level helper). Added parentLookupExpr as a correlated subquery for parent ID resolution.
• lib/hana/sql-expressions.js: Replaced all :old.column / :new.column references with ot.column / nt.column. Updated nullSafeChanged, getValueExpr, getWhereCondition, buildAssocLookup, buildObjectIDExpr, and buildGrandParentObjectIDExpr accordingly. Changed SELECT.one.from(...) to SELECT.from(...) throughout.
• lib/hana/register.js: Removed the HDI compiler hook logic that injected the dummy CSV data and patched the CSN definition for CHANGE_TRACKING_DUMMY.
• tests/build/hana-build.test.js: Removed test assertion for CHANGE_TRACKING_DUMMY CSV generation.

• 🔄 Regenerate and Update Summary

---

📬 Subscribe to the Hyperspace PR Bot DL to get the latest announcements and pilot features!

PR Bot Information

Version: 1.19.3 | 📖 Documentation | 🚨 Create Incident | 💬 Feedback

• Output Template: Default Template
• Summary Prompt: Default Prompt
• Correlation ID: a4f96c20-27a6-11f1-82b7-8bfcfe8a6519
• Event Trigger: pull_request.opened
• LLM: anthropic--claude-4.6-sonnet

[hyperspace-insights]

The refactor successfully migrates HANA triggers from row-level (FOR EACH ROW) to statement-level (FOR EACH STATEMENT) semantics, eliminating the CHANGE_TRACKING_DUMMY table workaround and replacing procedural variable-based logic with set-based INSERT … SELECT DISTINCT patterns. The overall approach is sound, but there are four issues worth addressing: a dead variable left from the old implementation (entityKey in generateDeleteTrigger), another dead code block (grandParentWhereSubquery in buildGrandParentObjectIDExpr), a correctness problem with SELECT DISTINCT in buildCompositionOfOneParentContext where the correlated parentKeyExpr subquery prevents effective deduplication across multiple child rows belonging to the same parent, and a potential SQL corruption risk in TriggerCQN2SQL where subquery-valued val strings (e.g. (SELECT … WHERE pk = nt.ID)) fall through to super.val() and get quoted as string literals instead of being emitted verbatim.

PR Bot Information

Version: 1.19.3 | 📖 Documentation | 🚨 Create Incident | 💬 Feedback

• LLM: anthropic--claude-4.6-sonnet
• Correlation ID: a4f96c20-27a6-11f1-82b7-8bfcfe8a6519
• Event Trigger: pull_request.opened