Cx Dev Log — 2026-03-18

#cx #programming

The biggest piece of work happening right now in Cx is Semantic Parity Phase 2: migrating the interpreter from raw AST execution to running off SemanticProgram. No commits landed today, but there are 782 lines of uncommitted changes across 9 files in the wonderful-saha worktree. The interpreter migration is being written. Here is where things stand.

The semantic interpreter exists

The core of the work is in src/runtime/runtime.rs, where a full semantic IR interpreter has been written. New methods -- eval_semantic_expr, run_semantic_stmt, and call_semantic_func -- implement expression evaluation and statement execution against SemanticExpr and SemanticStmt types instead of raw AST nodes.

The RunTime struct now has a semantic_funcs registry (HashMap) alongside the existing function registry. This is a dual-registry setup, and that is deliberate.

In src/main.rs, run_with_interpreter now accepts SemanticProgram instead of Program. The pre-pass that registers structs, functions, and impl blocks has been rewritten to work with SemanticStmt variants. Function registration bridges both the semantic and legacy registries so the old copy/bleedback fallback behavior still works during the transition.

This is not a clean swap. It is a staged migration. Both the old AST function registry and the new semantic function registry run at the same time. The old paths remain as fallback, which means the 45-test matrix does not have to pass through the new interpreter all at once. Work can land incrementally.

Control flow gets first-class IR nodes

As part of this migration, IfElse and WhileIn have been promoted to first-class nodes in both the AST and the semantic IR.

In src/frontend/ast.rs, new statement variants for IfElse and WhileIn were added along with a WhileInChain struct for cursor iteration chains. The corresponding SemanticStmt::IfElse, SemanticStmt::WhileIn, and SemanticWhileInChain variants were added in src/frontend/semantic_types.rs.

Previously these control flow constructs were handled through more general paths. Giving them dedicated variants in both layers makes the semantic IR complete for control flow. That is a prerequisite for the interpreter to run entirely off the semantic IR without falling back to AST interpretation for certain constructs.

Groundwork for generics in the semantic layer

A smaller but notable change: the analyzer's semantic_type_from_decl now takes a generic params slice, and analyze_function gains a type parameters argument. Duplicate StructDef handling was removed. This is not the generics work itself, but it prepares the analyzer for generics-aware semantic analysis. It came along naturally while reworking the semantic layer for the interpreter migration.

Branch state

Submain still has not been merged into main. This was flagged yesterday and remains the case. Main is still missing structs, operators, and Semantic Parity Phase 1 from the March 17 work. The wonderful-saha worktree branches from submain, so until submain merges, none of this Phase 2 work can reach main either.

The branch situation is straightforward but it needs to be resolved. Two feature branches are stacked (submain, then wonderful-saha on top of it) and main is behind both.

What is next

The immediate path forward has two threads:

Finish the semantic interpreter migration in wonderful-saha. The interpreter is largely written, but the bridge between old and new runtime paths still needs work. Once the full test matrix passes through SemanticProgram, the raw AST interpretation path can be removed.
Merge submain into main. This unblocks everything downstream. Until it happens, the real development state of the language lives on feature branches rather than main.

On the backend side, Phase 6 (function call lowering) is next on the roadmap. No work started on it today, but it can proceed in parallel once the branch state is cleaned up.

Follow the Cx language project: