Fix issue #89: assign() and intersection_for() not recognized as builtins

[particlesector]

Summary

Closes #89.

Root-caused: neither assign() nor intersection_for() was ever added to Parser::kBuiltinNodeNames, the table the parser uses to recognize builtin module names by text (since builtin names aren't reserved lexer keywords, matching real OpenSCAD). A statement-position call to either name therefore fell through to parseModuleCall(), which looks up a user-defined module of that name — since neither test file defines one, every top-level statement in both assign-tests.scad and intersection_for-tests.scad silently resolved to nothing. This explains "zero valid combined geometry" for every root in both files, matching the issue's own observation that these two are unlike the harness-bug files it also investigated. It is not a MeshEvaluator/PrimitiveGen tessellation bug.

• assign(x = ..., ...) { ... } — the deprecated statement form of let() — is grammatically and semantically identical to let() (block-scoped bindings visible only to its children), so it's routed straight to the existing Parser::parseLetNode(). No new AST node or evaluator code needed.
• intersection_for(i = ...) { ... } shares for()'s entire grammar, differing only in how the loop's results are combined. Added a ForNode::isIntersection flag, threaded through a new Parser::parseFor(bool isIntersection = false) parameter (default preserves every existing for() call site), and read by CsgEvaluator::evalFor() to pick CsgBoolean::Op::Intersection instead of Op::Union when combining the flattened per-iteration children.

Verification

This session's network egress is scoped to this repo only, so the full CMake/vcpkg/Manifold build isn't available here. Followed the same approach as a prior pass (v3.10, see docs/roadmap.md): compiled the actual Manifold-free src/lang/src/csg sources directly with g++ against real glm and a from-source Catch2 (extras/catch_amalgamated.*), then ran the real test suites:

• tests/test_parser.cpp, tests/test_csg_evaluator.cpp, tests/test_lexer.cpp, tests/test_interpreter.cpp, tests/test_source_loader.cpp
• 580 test cases / 3266 assertions, all passing — including 7 new regression tests added by this change (parser-shape tests for both constructs, plus evaluator tests covering intersection_for's intersect-not-union behavior, its empty-range/single-iteration edge cases, and assign()'s let()-equivalent block scoping).

Also hand-verified with corpus-shaped snippets (a rotate-and-loop intersection_for, a nested assign() shadowing an outer variable) fed through the real CsgEvaluator directly, confirming each now produces non-empty geometry where it previously produced none.

Exact volumetric correctness against a live OpenSCAD oracle (the v3.9-style sym_diff_volume check) is still unverified — no oracle was available in this session — but the root cause is confirmed and fixed, answering the specific question issue #89 raised.

Files changed

• src/lang/Token.h — new IntersectionFor/Assign token tags
• src/lang/Parser.cpp / Parser.h — recognize both builtins; parseFor() gains isIntersection
• src/lang/AST.h — ForNode::isIntersection field
• src/csg/CsgEvaluator.cpp — evalFor() combines with Intersection when the flag is set
• tests/test_parser.cpp, tests/test_csg_evaluator.cpp — new regression tests
• docs/roadmap.md — v3.13 entry documenting the root cause and fix

Test plan

• New parser tests: intersection_for parses to ForNode{isIntersection=true}; assign() parses to LetNode
• New evaluator tests: intersection_for combines via Intersection, handles empty-range/single-iteration edge cases; assign() bindings scope correctly and don't leak
• Full existing test suite (580 cases / 3266 assertions) still passes — no regressions to plain for()/let()/intersection()

---

Generated by Claude Code