Repository navigation
Conversation
… permissive and unique literal mappers A tsgo function mapper is never nil, so instantiateType walks the type and getObjectTypeInstantiation looks a mapped type instance up under its own key. The pair list that stands in for those mappers was empty when every type parameter mapped to itself, and an empty list is the identity mapper, which instantiate treats as a no-op. A mapped type instance first stored under another alias's key then kept that identity on one side of a conditional type's extends check while the inferred type held the instance under its own key. The check went structural and entered the same conditional type again with the same arguments until the depth limit. The inference context's mappers already returned a mapper that maps nothing for such a type. The helper now serves the restrictive, permissive and unique literal mappers and the outer return mapper too.
…lias could contain no type variables isNonGenericTopLevelType: tsgo returns such a type unchanged from instantiateType, whatever its members. Here the flag was folded from the members, so the type was walked under a mapper, a mapped type instance in it was looked up under its own key, and the result was a second identity of the alias's type. A conditional type's extends check then compared the two identities structurally and entered itself again until the depth limit. Also one constructor for a mapper that tsgo has as a function.
|
Updated 4:55 AM PT - Oct 25th, 2026
✅ @robobun, your commit b24b15b07011c8df914f22bfd50596c079a7471d passed in 🧪 To try this PR locally: bunx bun-pr 46868That installs a local version of the PR into your bun-46868 --bun |
|
No actionable comments were generated in the recent review.
 |
| Layer / File(s) | Summary |
|---|---|
Record non-generic top-level alias provenance src/sema/types.rs, src/sema/check/alias.rs, src/sema/check/{context.rs, expr.rs, mapped.rs, unions.rs, task_tests.rs} |
Provenance now carries the non-generic top-level alias flag. Alias construction sets the flag, while other provenance construction sites initialize it to false. Union and intersection records with the flag clear COULD_CONTAIN_TYPE_VARIABLES. |
Select mappers and validate checker behavior src/sema/check/{instantiate.rs, infer.rs, call.rs}, test/cli/check/check.test.ts |
Mapper creation uses a shared helper for empty parameter mappings. The tests cover recursive mapped types and an aliased distributive conditional. |
Possibly related issues
- Issue 46793 — The recursive
ColumnDefregression test covers a relatedbun checkTS2589 pattern. The issue also reports abun lint --type-awarestack overflow, which this change does not address.
Suggested reviewers: jarred-sumner
Priority:
Merge Risk: 
b24b1
-
No identified issue blocks merging after normal checks.
Pre-merge checks: 
-
Check name Status Explanation Title check The title accurately identifies the main changes: walking types under restrictive and permissive mappers and handling non-generic top-level aliases. It is long and somewhat awkward, but it is specific… Description check The description clearly explains the problem, the fix, verification steps, test coverage, performance results, limitations, and related issue context. It provides the required information even though … Linked Issues check Check skipped because no linked issues were found for this pull request. Out of Scope Changes check Check skipped because no linked issues were found for this pull request. Full details: Title check
-
Explanation
The title accurately identifies the main changes: walking types under restrictive and permissive mappers and handling non-generic top-level aliases. It is long and somewhat awkward, but it is specific and related to the changeset.
Full details: Description check
-
Explanation
The description clearly explains the problem, the fix, verification steps, test coverage, performance results, limitations, and related issue context. It provides the required information even though it uses different headings from the repository template.
-
- Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.
Comment @coderabbitai help to get the list of available commands.
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
There was a problem hiding this comment.
Nothing blocking. The comments below are optional suggestions. There is no need to push a fix for them before merging.
Beyond the inline findings, I also checked the new Provenance::is_non_generic_top_level field end to end: every ProvenanceKey constructor (alias.rs, unions.rs, expr.rs, mapped.rs, context.rs, task_tests.rs, the types.rs unit test) sets it, and is_default, is, to_owned, follow_struct!, clone_in_struct! and visit_type_as_content hashing all include it, so interning identity stays consistent. The has_other_instantiation reorder in src/sema/check/alias.rs is equivalent to the old expression (is_non_generic_top_level_alias is a pure scope lookup, so evaluating it earlier has no side effect).
Extended reasoning...
The change touches the bun check type checker internals in src/sema/ (type interning keys, mapper construction for restrictive/permissive/inference instantiation) plus a new test block in test/cli/check/check.test.ts; it touches no security-sensitive surface. The two inline findings (shared sentinel mapper interacting with active_mappers dedup, and the untested call.rs caller) are what keep this from an approve; the ruled-out note above records the struct-field audit and refactor-equivalence check that were done in addition.
Part of #46793
Problem
bun checkreportsTS2589: Type instantiation is excessively deep and possibly infinitewhere tsc accepts: a TanStack TableColumnDefwhoseTDatais a mapped type instance, such as a supabase-js row, or a top-level alias around one (type Row = Item & { id: string }).bun lint --type-awaresays "The type checker ran out of stack".derive_instantiation_of_type_parameters(src/sema/check/instantiate.rs:543) returns the identity mapper, a no-op, when every type parameter maps to itself. tsgo's function mappers always walk. And a union or intersection named by a non-generic top-level alias keepsCOULD_CONTAIN_TYPE_VARIABLESfrom its members. tsgo'sisNonGenericTopLevelTypemakes it false.UnionToIntersection'sextendscheck compares a mapped type instance under one alias key with the same instance under its own key. The relation goes structural and enters the same conditional type again until the depth limit.Fix
function_mapper: an empty pair list becomes a mapper that maps nothing if the type hasHAS_OTHER_INSTANTIATION, soinstantiatewalks and finds the instance under its own key. The helper is the onebun check: fix more differences fromtsc#44685 added for the inference mappers. The outer return mapper in call.rs uses it too.Provenance::is_non_generic_top_level: such a union or intersection losesCOULD_CONTAIN_TYPE_VARIABLESwhen it is created, soinstantiatereturns it unchanged (checker.go:22117).test/cli/check/check.test.ts, "walks a type with no type parameters under the restrictive and permissive mappers" (five programs) and "computes the constraint of a distributive conditional type behind an alias". Also all 13101 conformance cases, byte for byte, in both modes.Background
(type parameter, type)pairs. The empty list is tsgo's nil mapper:instantiatereturns the type unchanged.Copy<Item>first reached throughtype S = Copy<Item>is stored underS's key. Rust marks a type that contains such an instance withHAS_OTHER_INSTANTIATION.is_assignable_restrictiveonly: the permissive check (c.ts) keeps the stale instance. Considered one identity at creation: tsgo printsCopy<Item> | Copy<Item>forS | Copy<Item>.Downsides
is_non_generic_top_level_aliasper aliased union creation (it ran there already). Instantiations computed on bun check: TS2589 on a typed $fetch call (nitropack route types) at ~60% of the instantiations at which tsc 7.0.2 reports it #45691 and bun check: TS2589 when anew PrismaClient({ ... })value is assigned toPrismaClientwith a schema of ~500 models (tsc 7.0.2 accepts) #46139 are unchanged (1,475,298 and 1,923,505). Wall clock, 5 interleaved runs: 1,316 to 1,288 ms and 2,978 to 2,960 ms, noise about 8%.bun-sematext: 9,039,060 to 9,039,156 bytes (size, +96).perf,valgrindandbloatyare not installed here, so instructions and allocations are not measured.Notes
Readonly<{ id: number }>asTDataof a published@tanstack/react-table9.2.3ColumnDeffails on 1.4.3 too.(x: infer R extends unknown) => anymakes the pair list non-empty and the 23-line file passes on main.d.tsis an infinite type over the same instance. It still reports TS2589 at the position tsc reports. Across the 13101 conformance cases the walking mapper is returned 3 times, for one type, and the walk returns it unchanged. The changed-type path is covered by the new tests and by a local corpus of 1010 reductions: 112 outputs differ from main, all of them TS2589 removals, except one spread input that still needs bun check: infer from a context sensitive function whose parameter type has an instantiated type argument #46564.Bar<U>in the second test: its restrictive instantiation is now another instance without the alias, as in tsgo, soconstraint_of_distributive_conditionalcomputes a constraint and a false TS2322 goes away.bun lint --type-awarethe 5,000,000 instantiation limit is hit inside a rule query that has no node to report at, so the driver prints "ran out of stack". Limit hits are not cached, so every later rule query recomputes the cycle. That reporting is unchanged here.check.test.ts"reports TS2589 for a conditional type that recurs through a type node that isn't deferred" reports 5 of 7 TS2589 on the debug build of main (7 on release and tsc), and the Eden Treaty test needstest/node_modules.no test proof · iteration 0 · platform-specific test(s) that do not run on this machine, deferring to CI, which covers all platforms: test/cli/check/check.test.ts