Summary

Closes #584.
Closes #583.
Closes #568.

All three are codegen bugs in the same class — type checker accepts code that the WASM codegen handles incorrectly, with no Vera-native diagnostic in between. #584 and #583 surfaced via a sandboxed Claude.ai instance writing a cellular-automaton program; #568 was a pre-existing leading-colon round-trip bug. Bundled together because all three are small, surgical fixes (#568 is two lines of WAT plus a comment).

A fourth fix — a CLI display polish for String-returning main — was added after testing the alias work surfaced it. Details in its own section below.

#584 — user @Unit-returning fn in non-tail position

Block-statement sequencing in vera/wasm/context.py::_is_void_expr recognised IO.* qualified calls, UnitLit, effect-op FnCalls, and compound expressions as void, but missed FnCall to a user-declared @Unit fn. The surrounding sequencer fell through to "produces a value", emitted a stray drop, and WASM validation failed with expected a type but nothing on stack. The natural render(grid); IO.sleep(120); recurse(...) shape lands directly on this whenever render is a user helper.

Fix (5 lines logic + comment):

• vera/codegen/functions.py — flip the _fn_ret_types filter from if ret_wt and ret_wt != "unsupported" (which excluded Unit returns because None is falsy) to if ret_wt != "unsupported" so Unit-returning fns are registered with None value. Already done this way in vera/codegen/closures.py; this brings the two builders into agreement.
• vera/wasm/context.py — add a FnCall clause to _is_void_expr that returns True when the call target is in _fn_ret_types with value None.

#583 — type aliases over Array<T>

_is_pair_type_name in vera/wasm/inference.py was a @staticmethod doing string-pattern matching against unresolved alias names, so type Row = Array<Bool> left "Row" unrecognised as a pair type. Two failure modes from the issue body:

• SlotRef @Row.0 emitted only local.get ptr not (ptr, len) — parameter use yielded invalid WAT at instantiate.
• _slot_name_to_wasm_type("Row") returned None for a let binding, triggering silent E602 skip of the enclosing function.

Fix:

• vera/wasm/inference.py — convert _is_pair_type_name from @staticmethod to instance method that resolves aliases via _resolve_base_type_name first. All existing callers already passed names through self._is_pair_type_name(...) so call sites need no change.
• vera/wasm/data.py::_translate_array_lit — resolve element-type aliases before the pair-layout helpers (so [@Row.0, @Row.0] lays out elements at the correct stride, not the 4-byte i32 path).
• vera/wasm/inference.py::_infer_index_element_type_expr — follow type aliases on the collection type so @Row.0[1] resolves the element type correctly. New helper _alias_array_element handles the common case of a non-generic alias pointing at a concrete Array<T>.

Aliases now work in parameter, let-binding, indexing, and array-literal-element positions.

#568 — url_parse / url_join leading-colon round-trip

url_parse(":foo") returned Ok with empty scheme; url_join then emitted bare "foo", losing the leading colon. Two viable fixes were noted in the issue body — RFC 3986-strict rejection vs. round-trip preservation flag. Going with RFC 3986 §3.1-aligned rejection (the cleaner option per maintainer call).

Fix (2 lines WAT + comment):

• vera/wasm/calls_encoding.py::_translate_url_parse — after the colon-scan loop, branch to $err_up if colon_pos == 0. Returns Err("missing scheme") (the existing error message — empty scheme is also a missing scheme). No changes to url_join; the s_len > 0 gate stays since post-fix it can never see an empty scheme.

We don't yet enforce the full ALPHA / [ALPHA / DIGIT / "+" / "-" / "."]* scheme grammar from RFC 3986 — that's a wider RFC-conformance follow-up — but the empty-scheme case alone is the only one that lost its leading colon in the round-trip; full grammar enforcement isn't needed to close #568.

CLI display polish — vera run on String-returning main decodes the pair

Spotted while testing the #583 alias fixes against host-handle types: a public main(@Unit -> @String) returning "hello" printed something like 147492 instead of hello. The (ptr, len) i32_pair return was being collapsed to its first half — the heap pointer — by the value extractor in execute(). Distinct from the codegen bugs above (the WAT was already correct; only the CLI display was lying), but small enough to bundle.

Fix:

• vera/codegen/api.py — new CompileResult.fn_string_returns: set[str] populated by compile_program from each FnDecl's return type. Alias-resolved via a new _return_type_is_string helper in vera/codegen/core.py so type Greeting = String participates (cooperates with Type aliases over Array<T> break WASM codegen (parameter case: invalid WAT; let-binding case: silent E602 skip) #583's alias work).
• vera/codegen/api.py::execute — when raw_result is a 2-tuple AND fn_name is in fn_string_returns, read memory[ptr:ptr+len] and decode as UTF-8. Stored in ExecuteResult.value (type widened from int | float | None → int | float | str | None). CLI's print(value) then renders the actual string.

Array<T> returns deliberately keep the bare-pointer fallback — their bytes-at-ptr aren't UTF-8 and rendering them meaningfully would need element-type-aware formatting (separate scope). A new test_array_return_unchanged test pins this asymmetry.

Test plan

• vera check + vera run clean on the issue-body reproducers (all five variants for User-defined Unit-returning function call in non-tail position breaks WASM (block statement-sequencing emits drop with empty stack) #584, four for Type aliases over Array<T> break WASM codegen (parameter case: invalid WAT; let-binding case: silent E602 skip) #583, leading-colon for url_parse / url_join drops leading colon — ":foo" round-trips as "foo" #568, plus String-returning main now prints text not a pointer).
• New conformance tests:
  • tests/conformance/ch07_unit_fn_nontail.vera (User-defined Unit-returning function call in non-tail position breaks WASM (block statement-sequencing emits drop with empty stack) #584; mid-block, leading, and if-arm-non-tail positions)
  • tests/conformance/ch02_type_alias_array.vera (Type aliases over Array<T> break WASM codegen (parameter case: invalid WAT; let-binding case: silent E602 skip) #583; param, let, index, nested-literal-element positions)
  • tests/conformance/ch09_url_parsing.vera extended with test_parse_empty_scheme (url_parse / url_join drops leading colon — ":foo" round-trips as "foo" #568)
• New unit tests in tests/test_codegen.py::TestStringArraySignatures:
  • test_string_return_execution rewritten — asserts value == "hello" (was: isinstance(value, int))
  • test_string_alias_return_execution — locks in alias cooperation
  • test_array_return_unchanged — pins the intentional Array-vs-String asymmetry
• All 84 conformance programs pass (was 82; +2 new files, 1 extension).
• All 33 examples pass check + verify.
• pytest tests/ — 3,714 passed, 14 skipped (no new failures).
• mypy vera/ clean.
• All pre-commit hooks pass on every commit (doc-counts, version-sync, limitations-sync, site assets all consistent).

Doc updates

• KNOWN_ISSUES.md — drop the Type aliases over Array<T> break WASM codegen (parameter case: invalid WAT; let-binding case: silent E602 skip) #583, User-defined Unit-returning function call in non-tail position breaks WASM (block statement-sequencing emits drop with empty stack) #584, and url_parse / url_join drops leading colon — ":foo" round-trips as "foo" #568 rows.
• SKILL.md — drop their workaround entries (the Type aliases over Array<T> break WASM codegen (parameter case: invalid WAT; let-binding case: silent E602 skip) #583 type-alias paragraph, the User-defined Unit-returning function call in non-tail position breaks WASM (block statement-sequencing emits drop with empty stack) #584 inline iteration-section footgun note, both rows from the Known Bugs and Workarounds table including url_parse / url_join drops leading colon — ":foo" round-trips as "foo" #568). Only GC-aware tail-call optimization for allocating functions #549 (TCO + GC) remains in the table.
• CHANGELOG.md — new [0.0.135] entry rolling up the [Unreleased] sandbox-install affordance from SKILL.md: tell sandbox-running agents they can install Vera #582 plus the four new fixes (CLI String display, url_parse / url_join drops leading colon — ":foo" round-trips as "foo" #568, User-defined Unit-returning function call in non-tail position breaks WASM (block statement-sequencing emits drop with empty stack) #584, Type aliases over Array<T> break WASM codegen (parameter case: invalid WAT; let-binding case: silent E602 skip) #583).
• Version files (vera/__init__.py, pyproject.toml, README.md, docs/index.html) bumped 0.0.134 → 0.0.135.
• Conformance count + test count + per-level table in TESTING.md, CLAUDE.md, SKILL.md, AGENTS.md, FAQ.md, ROADMAP.md updated 82 → 84 / 3,716 → 3,728.
• Site assets regenerated (docs/llms.txt + docs/llms-full.txt + docs/SKILL.md + docs/index.md).
• uv.lock regenerated to record vera 0.0.135 (the lint CI job runs uv lock --check which trips on a project version bump).
• HISTORY.md Stage 11 ledger row for v0.0.135.

🤖 Generated with Claude Code

Summary by CodeRabbit

• Bug Fixes

  • Fixed WASM array handling when element types use aliases
  • Corrected handling of Unit-returning functions in non-tail positions
  • Prevented empty-scheme URLs from parsing/round-tripping

• New Features

  • Runtime now decodes String results into host strings when returned from functions

• Documentation

  • Updated docs and guides to reflect an expanded conformance suite (84 programs) and project metrics
  • Updated Known Issues entry

• Tests

  • Added two conformance tests for array aliases and non-tail Unit returns

• Chores

  • Bumped project version to v0.0.135