Repository navigation
M4 type-checking rules: hole inference and the missing diagnostics #99
Description
Part of #77 (Phase 4). The M4 cases that need rules, not codes. 61 M4 cases total; 44 of them become measurable the moment #81 lands, leaving these.
| Code | Cases | Status |
|---|---|---|
TypesDoNotUnify |
41 | Needs the mapping only — this is the bulk of L4 |
HoleInferredType |
10 | No rule exists; needs hole tracking through inference |
CannotApplyExpressionOfTypeOnType |
2 | No rule exists |
ExpectedType |
2 | No rule exists; needs the purs semantics decision (#81) |
AmbiguousTypeVariables |
1 | AmbiguousConstraint exists; needs the counterpart chosen |
InfiniteType |
2 | OccursCheck exists; mapping only |
IntOutOfRange |
1 | IntegerOutOfRange exists; mapping only |
EscapedSkolem |
2 | Already mapped; verify the two cases agree |
Hole tracking
HoleInferredType is 10 cases and is the one genuine feature here. A hole must survive unification as a named constraint on an inference variable, so the reported type is the type the hole was solved to, not a fresh variable. That is a real change to the inference core, and it is FE-17's "holes" item.
Acceptance
All 61 M4 cases agree on the L4 scoreboard, and FE-08/FE-09/FE-10/FE-12 each state a concrete remaining gap rather than "reach the complete L4/L6 gate".
Activity
biuld commentedon Oct 22, 2026
L4 is 65/65, including HoleInferredType 10/10 and all original mapped type codes. The relevant feature rows now name concrete remaining passing-program gaps.
Verified by PR #164 (41a16e2): formatting, default Clippy, full workspace (2331 passed, zero failed), five separately executed frontend boards, all 22 official purs differential tests, and the complete serial runtime board (297/340; 25 FFI exclusions; every agreeing exit zero). Required Wasmtime CI execution check 119961920075 completed successfully. The three losses versus the older board reproduce on the unmodified starting HEAD; BigFunction remains an explicit incomplete compiler verdict.
Evidence: Phase 4 acceptance review.