fix: secondary index cache invariant bug in update/remove

[paulgnz]

Summary

Two bugs fixed in v0.2.15:

1. vm.ts — Cache invariant bug: genericIndex.update() and genericIndex.remove() fail to cache the table before accessing it, causing "an invariant was broken, table should be in cache" assertion failures
2. table.ts — Operator precedence bug in IndexObject.compare (already fixed on master but present in published v0.2.15): Due to || having higher precedence than ?:, the comparator returns 0 (EQUAL) for entries from different tables, causing BTree lookups to return wrong secondary index entries and triggering spurious "db access violation" errors

Problem 1: Cache Invariant

When contracts update or remove rows in tables with secondary indexes, db_idx64_update/db_idx64_remove call cache.getTable(obj.tableId), which asserts the table is already cached. This invariant is not guaranteed.

genericIndex method	Caches table?
store()	Yes
find_secondary()	Yes
find_primary()	Yes
update()	No (FIXED)
remove()	No (FIXED)

Problem 2: Operator Precedence (v0.2.15 only)

// BUGGY (v0.2.15):staticcompare(a,b): number{returnIndexObject.compareTable(a,b)||(a.ignorePrimaryKey||b.ignorePrimaryKey)
? 0// <-- returns 0 when compareTable is non-zero!
: (primaryKeycomparison);}

Due to || binding tighter than ?:, when compareTable returns -1 or 1 (different tables), the expression (-1 || false) is truthy, so the ternary returns 0 (EQUAL). This makes the BTree unable to distinguish entries from different tables, causing find_primary to return secondary index objects belonging to the wrong contract.

Reproduction: Any test scenario where multiple contracts store rows in tables with secondary indexes and then one contract calls an inline action on another that updates its own table.

Fix

1. vm.ts: Retrieve table from bc.store.getTableById() and re-cache it (matching the pattern used by store()/find_secondary()/find_primary())
2. table.ts: Already fixed on master — the compare function was rewritten with explicit early returns

Test Results

• 114 passing tests across 4 contract test suites (agentcore: 32, agentfeed: 23, agentvalid: 28, agentescrow: 31)
• 0 pending/skipped
• All previously-failing scenarios (token transfer handlers, cross-contract inline actions) now pass

Generated with Claude Code

[paulgnz]

Summary

Two bugs fixed in v0.2.15:

1. vm.ts — Cache invariant bug: genericIndex.update() and genericIndex.remove() fail to cache the table before accessing it, causing "an invariant was broken, table should be in cache" assertion failures
2. table.ts — Operator precedence bug in IndexObject.compare (already fixed on master but present in published v0.2.15): Due to || having higher precedence than ?:, the comparator returns 0 (EQUAL) for entries from different tables, causing BTree lookups to return wrong secondary index entries and triggering spurious "db access violation" errors

Problem 1: Cache Invariant

When contracts update or remove rows in tables with secondary indexes, db_idx64_update/db_idx64_remove call cache.getTable(obj.tableId), which asserts the table is already cached. This invariant is not guaranteed.

genericIndex method	Caches table?
store()	Yes
find_secondary()	Yes
find_primary()	Yes
update()	No (FIXED)
remove()	No (FIXED)

Problem 2: Operator Precedence (v0.2.15 only)

// BUGGY (v0.2.15):staticcompare(a,b): number{returnIndexObject.compareTable(a,b)||(a.ignorePrimaryKey||b.ignorePrimaryKey)
? 0// <-- returns 0 when compareTable is non-zero!
: (primaryKeycomparison);}

Due to || binding tighter than ?:, when compareTable returns -1 or 1 (different tables), the expression (-1 || false) is truthy, so the ternary returns 0 (EQUAL). This makes the BTree unable to distinguish entries from different tables, causing find_primary to return secondary index objects belonging to the wrong contract.

Reproduction: Any test scenario where multiple contracts store rows in tables with secondary indexes and then one contract calls an inline action on another that updates its own table.

Fix

1. vm.ts: Retrieve table from bc.store.getTableById() and re-cache it (matching the pattern used by store()/find_secondary()/find_primary())
2. table.ts: Already fixed on master — the compare function was rewritten with explicit early returns

Test Results

• 114 passing tests across 4 contract test suites (agentcore: 32, agentfeed: 23, agentvalid: 28, agentescrow: 31)
• 0 pending/skipped
• All previously-failing scenarios (token transfer handlers, cross-contract inline actions) now pass

Generated with Claude Code

[paulgnz]

Summary

Two bugs fixed in v0.2.15:

1. vm.ts — Cache invariant bug: genericIndex.update() and genericIndex.remove() fail to cache the table before accessing it, causing "an invariant was broken, table should be in cache" assertion failures
2. table.ts — Operator precedence bug in IndexObject.compare (already fixed on master but present in published v0.2.15): Due to || having higher precedence than ?:, the comparator returns 0 (EQUAL) for entries from different tables, causing BTree lookups to return wrong secondary index entries and triggering spurious "db access violation" errors

Problem 1: Cache Invariant

When contracts update or remove rows in tables with secondary indexes, db_idx64_update/db_idx64_remove call cache.getTable(obj.tableId), which asserts the table is already cached. This invariant is not guaranteed.

genericIndex method	Caches table?
store()	Yes
find_secondary()	Yes
find_primary()	Yes
update()	No (FIXED)
remove()	No (FIXED)

Problem 2: Operator Precedence (v0.2.15 only)

// BUGGY (v0.2.15):staticcompare(a,b): number{returnIndexObject.compareTable(a,b)||(a.ignorePrimaryKey||b.ignorePrimaryKey)
? 0// <-- returns 0 when compareTable is non-zero!
: (primaryKeycomparison);}

Due to || binding tighter than ?:, when compareTable returns -1 or 1 (different tables), the expression (-1 || false) is truthy, so the ternary returns 0 (EQUAL). This makes the BTree unable to distinguish entries from different tables, causing find_primary to return secondary index objects belonging to the wrong contract.

Reproduction: Any test scenario where multiple contracts store rows in tables with secondary indexes and then one contract calls an inline action on another that updates its own table.

Fix

1. vm.ts: Retrieve table from bc.store.getTableById() and re-cache it (matching the pattern used by store()/find_secondary()/find_primary())
2. table.ts: Already fixed on master — the compare function was rewritten with explicit early returns

Test Results

• 114 passing tests across 4 contract test suites (agentcore: 32, agentfeed: 23, agentvalid: 28, agentescrow: 31)
• 0 pending/skipped
• All previously-failing scenarios (token transfer handlers, cross-contract inline actions) now pass

Generated with Claude Code

[paulgnz]

Summary

Two bugs fixed in v0.2.15:

1. vm.ts — Cache invariant bug: genericIndex.update() and genericIndex.remove() fail to cache the table before accessing it, causing "an invariant was broken, table should be in cache" assertion failures
2. table.ts — Operator precedence bug in IndexObject.compare (already fixed on master but present in published v0.2.15): Due to || having higher precedence than ?:, the comparator returns 0 (EQUAL) for entries from different tables, causing BTree lookups to return wrong secondary index entries and triggering spurious "db access violation" errors

Problem 1: Cache Invariant

When contracts update or remove rows in tables with secondary indexes, db_idx64_update/db_idx64_remove call cache.getTable(obj.tableId), which asserts the table is already cached. This invariant is not guaranteed.

genericIndex method	Caches table?
store()	Yes
find_secondary()	Yes
find_primary()	Yes
update()	No (FIXED)
remove()	No (FIXED)

Problem 2: Operator Precedence (v0.2.15 only)

// BUGGY (v0.2.15):staticcompare(a,b): number{returnIndexObject.compareTable(a,b)||(a.ignorePrimaryKey||b.ignorePrimaryKey)
? 0// <-- returns 0 when compareTable is non-zero!
: (primaryKeycomparison);}

Due to || binding tighter than ?:, when compareTable returns -1 or 1 (different tables), the expression (-1 || false) is truthy, so the ternary returns 0 (EQUAL). This makes the BTree unable to distinguish entries from different tables, causing find_primary to return secondary index objects belonging to the wrong contract.

Reproduction: Any test scenario where multiple contracts store rows in tables with secondary indexes and then one contract calls an inline action on another that updates its own table.

Fix

1. vm.ts: Retrieve table from bc.store.getTableById() and re-cache it (matching the pattern used by store()/find_secondary()/find_primary())
2. table.ts: Already fixed on master — the compare function was rewritten with explicit early returns

Test Results

• 114 passing tests across 4 contract test suites (agentcore: 32, agentfeed: 23, agentvalid: 28, agentescrow: 31)
• 0 pending/skipped
• All previously-failing scenarios (token transfer handlers, cross-contract inline actions) now pass

Generated with Claude Code

[paulgnz]

Summary

Two bugs fixed in v0.2.15:

1. vm.ts — Cache invariant bug: genericIndex.update() and genericIndex.remove() fail to cache the table before accessing it, causing "an invariant was broken, table should be in cache" assertion failures
2. table.ts — Operator precedence bug in IndexObject.compare (already fixed on master but present in published v0.2.15): Due to || having higher precedence than ?:, the comparator returns 0 (EQUAL) for entries from different tables, causing BTree lookups to return wrong secondary index entries and triggering spurious "db access violation" errors

Problem 1: Cache Invariant

When contracts update or remove rows in tables with secondary indexes, db_idx64_update/db_idx64_remove call cache.getTable(obj.tableId), which asserts the table is already cached. This invariant is not guaranteed.

genericIndex method	Caches table?
store()	Yes
find_secondary()	Yes
find_primary()	Yes
update()	No (FIXED)
remove()	No (FIXED)

Problem 2: Operator Precedence (v0.2.15 only)

// BUGGY (v0.2.15):staticcompare(a,b): number{returnIndexObject.compareTable(a,b)||(a.ignorePrimaryKey||b.ignorePrimaryKey)
? 0// <-- returns 0 when compareTable is non-zero!
: (primaryKeycomparison);}

Due to || binding tighter than ?:, when compareTable returns -1 or 1 (different tables), the expression (-1 || false) is truthy, so the ternary returns 0 (EQUAL). This makes the BTree unable to distinguish entries from different tables, causing find_primary to return secondary index objects belonging to the wrong contract.

Reproduction: Any test scenario where multiple contracts store rows in tables with secondary indexes and then one contract calls an inline action on another that updates its own table.

Fix

1. vm.ts: Retrieve table from bc.store.getTableById() and re-cache it (matching the pattern used by store()/find_secondary()/find_primary())
2. table.ts: Already fixed on master — the compare function was rewritten with explicit early returns

Test Results

• 114 passing tests across 4 contract test suites (agentcore: 32, agentfeed: 23, agentvalid: 28, agentescrow: 31)
• 0 pending/skipped
• All previously-failing scenarios (token transfer handlers, cross-contract inline actions) now pass

Generated with Claude Code

[paulgnz]

Summary

Two bugs fixed in v0.2.15:

1. vm.ts — Cache invariant bug: genericIndex.update() and genericIndex.remove() fail to cache the table before accessing it, causing "an invariant was broken, table should be in cache" assertion failures
2. table.ts — Operator precedence bug in IndexObject.compare (already fixed on master but present in published v0.2.15): Due to || having higher precedence than ?:, the comparator returns 0 (EQUAL) for entries from different tables, causing BTree lookups to return wrong secondary index entries and triggering spurious "db access violation" errors

Problem 1: Cache Invariant

When contracts update or remove rows in tables with secondary indexes, db_idx64_update/db_idx64_remove call cache.getTable(obj.tableId), which asserts the table is already cached. This invariant is not guaranteed.

genericIndex method	Caches table?
store()	Yes
find_secondary()	Yes
find_primary()	Yes
update()	No (FIXED)
remove()	No (FIXED)

Problem 2: Operator Precedence (v0.2.15 only)

// BUGGY (v0.2.15):staticcompare(a,b): number{returnIndexObject.compareTable(a,b)||(a.ignorePrimaryKey||b.ignorePrimaryKey)
? 0// <-- returns 0 when compareTable is non-zero!
: (primaryKeycomparison);}

Due to || binding tighter than ?:, when compareTable returns -1 or 1 (different tables), the expression (-1 || false) is truthy, so the ternary returns 0 (EQUAL). This makes the BTree unable to distinguish entries from different tables, causing find_primary to return secondary index objects belonging to the wrong contract.

Reproduction: Any test scenario where multiple contracts store rows in tables with secondary indexes and then one contract calls an inline action on another that updates its own table.

Fix

1. vm.ts: Retrieve table from bc.store.getTableById() and re-cache it (matching the pattern used by store()/find_secondary()/find_primary())
2. table.ts: Already fixed on master — the compare function was rewritten with explicit early returns

Test Results

• 114 passing tests across 4 contract test suites (agentcore: 32, agentfeed: 23, agentvalid: 28, agentescrow: 31)
• 0 pending/skipped
• All previously-failing scenarios (token transfer handlers, cross-contract inline actions) now pass

Generated with Claude Code

[paulgnz]

Summary

Two bugs fixed in v0.2.15:

1. vm.ts — Cache invariant bug: genericIndex.update() and genericIndex.remove() fail to cache the table before accessing it, causing "an invariant was broken, table should be in cache" assertion failures
2. table.ts — Operator precedence bug in IndexObject.compare (already fixed on master but present in published v0.2.15): Due to || having higher precedence than ?:, the comparator returns 0 (EQUAL) for entries from different tables, causing BTree lookups to return wrong secondary index entries and triggering spurious "db access violation" errors

Problem 1: Cache Invariant

When contracts update or remove rows in tables with secondary indexes, db_idx64_update/db_idx64_remove call cache.getTable(obj.tableId), which asserts the table is already cached. This invariant is not guaranteed.

genericIndex method	Caches table?
store()	Yes
find_secondary()	Yes
find_primary()	Yes
update()	No (FIXED)
remove()	No (FIXED)

Problem 2: Operator Precedence (v0.2.15 only)

// BUGGY (v0.2.15):staticcompare(a,b): number{returnIndexObject.compareTable(a,b)||(a.ignorePrimaryKey||b.ignorePrimaryKey)
? 0// <-- returns 0 when compareTable is non-zero!
: (primaryKeycomparison);}

Due to || binding tighter than ?:, when compareTable returns -1 or 1 (different tables), the expression (-1 || false) is truthy, so the ternary returns 0 (EQUAL). This makes the BTree unable to distinguish entries from different tables, causing find_primary to return secondary index objects belonging to the wrong contract.

Reproduction: Any test scenario where multiple contracts store rows in tables with secondary indexes and then one contract calls an inline action on another that updates its own table.

Fix

1. vm.ts: Retrieve table from bc.store.getTableById() and re-cache it (matching the pattern used by store()/find_secondary()/find_primary())
2. table.ts: Already fixed on master — the compare function was rewritten with explicit early returns

Test Results

• 114 passing tests across 4 contract test suites (agentcore: 32, agentfeed: 23, agentvalid: 28, agentescrow: 31)
• 0 pending/skipped
• All previously-failing scenarios (token transfer handlers, cross-contract inline actions) now pass

Generated with Claude Code

[paulgnz]

Summary

Two bugs fixed in v0.2.15:

1. vm.ts — Cache invariant bug: genericIndex.update() and genericIndex.remove() fail to cache the table before accessing it, causing "an invariant was broken, table should be in cache" assertion failures
2. table.ts — Operator precedence bug in IndexObject.compare (already fixed on master but present in published v0.2.15): Due to || having higher precedence than ?:, the comparator returns 0 (EQUAL) for entries from different tables, causing BTree lookups to return wrong secondary index entries and triggering spurious "db access violation" errors

Problem 1: Cache Invariant

When contracts update or remove rows in tables with secondary indexes, db_idx64_update/db_idx64_remove call cache.getTable(obj.tableId), which asserts the table is already cached. This invariant is not guaranteed.

genericIndex method	Caches table?
store()	Yes
find_secondary()	Yes
find_primary()	Yes
update()	No (FIXED)
remove()	No (FIXED)

Problem 2: Operator Precedence (v0.2.15 only)

// BUGGY (v0.2.15):staticcompare(a,b): number{returnIndexObject.compareTable(a,b)||(a.ignorePrimaryKey||b.ignorePrimaryKey)
? 0// <-- returns 0 when compareTable is non-zero!
: (primaryKeycomparison);}

Due to || binding tighter than ?:, when compareTable returns -1 or 1 (different tables), the expression (-1 || false) is truthy, so the ternary returns 0 (EQUAL). This makes the BTree unable to distinguish entries from different tables, causing find_primary to return secondary index objects belonging to the wrong contract.

Reproduction: Any test scenario where multiple contracts store rows in tables with secondary indexes and then one contract calls an inline action on another that updates its own table.

Fix

1. vm.ts: Retrieve table from bc.store.getTableById() and re-cache it (matching the pattern used by store()/find_secondary()/find_primary())
2. table.ts: Already fixed on master — the compare function was rewritten with explicit early returns

Test Results

• 114 passing tests across 4 contract test suites (agentcore: 32, agentfeed: 23, agentvalid: 28, agentescrow: 31)
• 0 pending/skipped
• All previously-failing scenarios (token transfer handlers, cross-contract inline actions) now pass

Generated with Claude Code