0bc8005b99
In type position (T) is a 1-tuple (specs.md:843), so ?(?i64) is optional(tuple(?i64)); assigning a bare ?i64 had coerceToType classify .none and pass the value through, then optionalWrap built a corrupt insertvalue that aborted the LLVM verifier. After coercing toward an optional's child, verify the coerced type equals the child type (stmt.zig decl-init + coerce.zig .optional_wrap); on mismatch emit a located diagnostic (tuple-specific note only when the child is a tuple). formatTypeName now renders tuples as (x: i64, y: i64). Regressions: optionals/0911 (nested optional via alias, round-trip), diagnostics/1195 (the mismatch diagnostic). Updated diagnostics/1101 + protocols/0414 goldens for the improved tuple type-name rendering. Verified by 3 adversarial reviews. Filed adjacent bug 0171 (?any child not canonicalized).
48 lines
2.2 KiB
Markdown
48 lines
2.2 KiB
Markdown
# 0171 — `?any` optional child is a non-canonical `any` TypeId (box-into-any rule misses, value silently discarded)
|
|
|
|
## Symptom
|
|
|
|
An optional whose child is `any` (`?any`) is broken. Baseline (before the issue
|
|
0165 fix) silently DISCARDED the boxed value: `x : ?any = 42; v := x!` yields an
|
|
empty box `any{}`, not `42` — the payload is lowered as a zero-size `{}`. After
|
|
the 0165 fix the same code now produces a clean type-mismatch diagnostic
|
|
(`cannot assign a value of type 'i64' to optional '?any': its payload type is
|
|
'any'`), which is strictly better than silent corruption but still means `?any`
|
|
does not work.
|
|
|
|
## Root cause (from adversarial review of issue 0165)
|
|
|
|
The box-into-`any` coercion rule (`src/ir/conversions.zig` ~line 57) keys on the
|
|
BUILTIN `.any` enum TypeId. But an optional's child `any` is a SEPARATELY
|
|
interned TypeId (observed `@enumFromInt(246)`, type-name `"any"`) that is NOT
|
|
identity-equal to the builtin `.any`. So `classify(i64, child_any)` falls through
|
|
to `.none`, returns the value unchanged (`i64`), and the wrap is invalid. The
|
|
`any` type is not being canonicalized to the builtin TypeId when it appears as an
|
|
optional child.
|
|
|
|
## Reproduction
|
|
|
|
```sx
|
|
#import "modules/std.sx";
|
|
main :: () {
|
|
x : ?any = 42;
|
|
v := x!;
|
|
print("{}\n", v); // expected: a boxed 42; baseline yields empty any{}
|
|
}
|
|
```
|
|
|
|
## Investigation prompt
|
|
|
|
Canonicalize `any` as an optional child (and likely any other compound position)
|
|
to the builtin `.any` TypeId at type-resolution/interning time, so the
|
|
box-into-any rule in `src/ir/conversions.zig` classifies correctly and `?any`
|
|
round-trips. Find where the optional child type is resolved/interned
|
|
(`src/ir/types.zig` `optionalOf` / the type resolver) and ensure an `any` child
|
|
maps to the canonical builtin TypeId rather than a fresh interned copy.
|
|
Alternatively, make the box-into-any classifier compare by type-KIND
|
|
(`info == .any`) rather than TypeId identity — but canonicalization is the more
|
|
robust fix (it also fixes `==`, `size_of`, and any other identity check on the
|
|
`any` child). Verify the repro round-trips a boxed value; add an
|
|
`examples/types/01xx-optional-any.sx` regression. Low priority — `?any` is used
|
|
nowhere in `library/` or `examples/`.
|