types: make ownership return-alias inference terminate on recursive fns

[medvednikov]

Summary

vlib/v/types/checker_ownership_alias_test.v (// vtest vflags: -d ownership) never finished compiling, so ./v test vlib/v/types/ hung. The time went into ownership return-alias inference over the v.types sources, not into the test itself.

Return-alias inference composes a callee's returned parameter path behind the argument projection. With f(p.next) and a callee that returns .name, the path becomes .next.name. Through recursion that projection is prepended again every fixed-point round (.next.next.name, ...). Subsumption only merges a path that extends a shallower one, so the set never converged. ownership_default_clone_missing_method_inner recurses through four fields (base_type, elem_type, key_type, value_type), so the path set grew exponentially. A plain linked-list function (return f(n.next) / return n.name) also looped.

Such paths are now widened to an explicit prefix source (OwnershipReturnParamDescendant.source_is_prefix): the result aliases storage at or below the argument projection, but not a known exact path.

• Each entry records via, the functions that composed its path. A path is widened only when it is composed again in a function already in its via, i.e. when it went around a call cycle; direct recursion is the one-function case. Every other composition extends via by a new function, so exact paths stay bounded by simple call chains, and acyclic chains stay exact at any depth, even when they repeat a projection (outer(p) returning inner(p.left), where inner(p) returns p.left.name, stays .left.left.name).
• Composing a prefix keeps it a prefix, and an exact source never subsumes a prefix source.
• Call-site ownership marking does not transfer ownership from a prefix source, so the target stays unowned. At worst the returned storage leaks; it is never dropped twice or with the wrong type.
• ownership_call_result_sources reports the flag. The array.map lowering treats such a result as external when the covering path, or any path below it, is.

checker_ownership_alias_test.v only needed -d ownership to see one pure helper, ownership_alias_chain_borrows_indexed_storage. With the hang gone, that define selects the ownership compiler, which rejects the types module itself with ~5.6k move errors. So the helper and its sentinel const move into an always compiled checker_ownership_alias.v, together with the new path-composition helpers, and the test drops its vflags line.

Fixes #28923

Related: #28678 describes the same hang. module types tests built with -d ownership now finish, but they then stop at the move errors above, so that issue is left open.

Validation

All runs use V_MACOS_V3_NO_FALLBACK=1.

• New vlib/v/tests/ownership_recursive_return_param_alias_test.v covers direct and mutual recursion under -d ownership. It passes, and still hangs on master after 60s.
• New unit tests in checker_ownership_alias_test.v:
  • exact vs. prefix path composition, including the acyclic outer/inner chain;
  • fixed-point simulations: mutual recursion through two fields each converges in a few rounds, and an acyclic chain that ends in a recursive function has only the recursion widened.
• A -d ownership build of a module types test finishes in ~20s (4 return fixed-point rounds); it never finished on master.
• -ownership checks of small programs:
  • an acyclic outer/inner chain that returns an owned string from a nested literal argument still reports use of moved value for the result, with type string;
  • a recursive getter no longer marks its string result as an owned Holder.
• ./v -silent test vlib/v/types/: 31/31 passed. It never finished on master.
• ./v -silent test vlib/v/transform/: 17/17 passed
• ./v -silent test vlib/v/gen/c/: 27/27 passed (checked before the provenance commit, which does not touch cgen or the transform)
• ./v -silent vlib/v/compiler_errors_test.v: 1721 passed, 0 failed
• The 15 vlib tests whose vflags use ownership or autofree, excluding the valgrind ones and including the new test: 15/15 passed
• vlib/v/compiler_tests/ownership/ownership_test.v builds vlib/v/v.v with -d ownership, which doesn't build on master or here. A copy run against ownership compilers built from master and from this branch fails the same 5 of 28 test fns on both. This was checked on the first commit, not re-run after the follow-up.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	🔄 Running since 2026-09-25T02:47:18.315428Z	cc2ed26	New commits

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: fc7dd4728f

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e544101094

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".