Origin: test-analyzer agent finding C1 / silent-failure-hunter M1 / code-reviewer S4 on PR #631.

The pattern

_resolve_base_type_name in vera/wasm/inference.py:~1340 is a recursive alias-chain follower with no cycle guard:

def _resolve_base_type_name(self, name: str) -> str:
    if name not in self._type_aliases:
        return name
    alias = self._type_aliases[name]
    if isinstance(alias, ast.RefinementType):
        if isinstance(alias.base_type, ast.NamedType):
            return self._resolve_base_type_name(alias.base_type.name)
    if isinstance(alias, ast.NamedType):
        return self._resolve_base_type_name(alias.name)
    return name

The deleted helper _resolve_type_name_to_wasm_canonical (removed in #630) had explicit cycle protection via a seen set. Its replacement _canonical_named_type (the new central walker) also has cycle protection. But _resolve_base_type_name continues to recurse without one, and is called from ~9 sites across inference.py plus vera/wasm/data.py and vera/wasm/context.py.

Reachability

Empirically verified during PR #631 review: a cyclic type alias (type A = B; type B = A;) crashes the codegen with RecursionError at vera/codegen/core.py:_type_expr_to_wasm_type before reaching _resolve_base_type_name. So the immediate vulnerability is masked by a different recursive resolver crashing first.

This means: _resolve_base_type_name is currently dead-code-safe with respect to cycles. But:

• If the type checker ever lets a cycle through and the upstream _type_expr_to_wasm_type is fixed first, this becomes the next crash site.
• The inconsistency itself is a code-smell — the new walker has cycle protection, the old helper doesn't, and they sit side by side.

Recommended fix

Either:

1. Add a cycle guard to _resolve_base_type_name for defence-in-depth (5-line change, low risk):

   def _resolve_base_type_name(self, name: str, seen: set[str] | None = None) -> str:
       if seen is None:
           seen = set()
       if name in seen or name not in self._type_aliases:
           return name
       seen.add(name)
       # ... existing logic, threading `seen` through recursive calls

2. Migrate remaining callers onto _canonical_named_type — strictly aligned with the Cross-cutting: centralise canonical Vera-type-name resolution to close the #602 bug class structurally #630 goal but a bigger surgery (the call sites use string→string, the new walker uses TypeExpr→NamedType). Probably too large for a single follow-up.

Option 1 is the cheap defensive move. File only if there's appetite.

Verification needed

Before either fix, confirm whether the type checker rejects cyclic aliases. If yes, this whole concern is purely belt-and-braces. If no, the latent bug is real and Option 1 should land.

Pairs with

• Cross-cutting: centralise canonical Vera-type-name resolution to close the #602 bug class structurally #630 (canonicalisation centralisation — landed; this is the residual cleanup).
• Cross-cutting: convert 'translate returns None → silent skip' failures into loud diagnostics #626 (broader silent-skip pattern).