parallel-work-plan.md

Parallel Work Plan: Qlpack End-to-End Execution

Track A: Fix Stratification (Evaluation Unblocking)

Goal: Get the 56,105 engine rules from vendor/codeql/cpp to evaluate successfully.

Deliverables:

1. Engine evaluates 56K rules against basic.c without stratification error
2. Class characteristic predicates (Function#char, Element#char) produce correct tuples
3. Regression test: qlpack_probe.rs evaluation phase succeeds

Tasks:

1. Add a "skip-cycles" mode to the engine stratifier that drops rules involved in negation cycles
2. Evaluate the remaining rules and verify core predicates work
3. Implement proper fix: inline double-negation patterns in MIR (forall/implies → direct AntiJoin without auxiliary predicate)
4. Re-run full evaluation and measure how many predicates produce results

Testable Milestones:

• Engine evaluates with cycle-skipping, >90% of rules execute
• Function#char relation has 3 rows when run against basic.c
• Proper fix: 0 false negation cycles on vendor/codeql

---

Track B: Fix Inheritance Dispatch + Closure Operators (MIR)

Goal: Member calls on subclass-typed variables resolve to parent class methods. Transitive closure works.

Deliverables:

1. f.getName() where f : SubClass finds ParentClass#getName
2. edges+(a, b) generates recursive transitive closure predicate
3. x in [1, 2, 3] creates proper disjunction
4. Regression tests for all three

Tasks:

1. Dispatch predicates: After lowering all files, build a map of method_name → [Class#method_name]. For each method name, generate a dispatch predicate: method_name(this, args...) :- Class1#method_name(this, args...); ...
2. Closure operators: When ClosureOp::Plus is present on a call, generate a recursive predicate: _tc_P(a, b) :- P(a, b). _tc_P(a, b) :- P(a, c), _tc_P(c, b).
3. SetLiteral fix: Generate disjunction for [a, b, c].

Testable Milestones:

• E2E test: from SubClass f select f.parentMethod() returns correct results
• E2E test: edges+(a, b) returns transitive closure
• E2E test: x in [1, 2, 3] matches all three values
• Existing 34 E2E tests still pass

---

Track C: Schema Alignment + String Builtins (Engine/Database)

Goal: Our C++ extractor schema matches what vendor library classes expect. String methods work.

Deliverables:

1. Audit of which database tables vendor library classes reference vs what our extractor produces
2. Engine built-in predicates for string methods (toString, regexpMatch, matches)
3. Mapping layer between our schema column names and CodeQL schema column names

Tasks:

1. Schema audit: Compare semmlecode.cpp.dbscheme tables with our cpp_schema(). List missing tables and column mismatches.
2. String builtins: Add engine support for string.toString(), int.toString(), string.length(), string.matches(pattern).
3. Table name mapping: If our extractor uses different table/column names than the vendor schema, add a translation layer.

Testable Milestones:

• Document listing all table mismatches between schemas
• Engine test: "hello".length() = 5 evaluates correctly
• Engine test: 42.toString() = "42" evaluates correctly
• Schema coverage report: X/Y tables from vendor schema are populated by our extractor