ed7665f8ae
Migrate lowerAssignment's `.field_access` target onto the shared
`fieldLvaluePtr` resolver, deleting its duplicated union / promoted /
tuple / vector / struct walk. All three lvalue field-store sites —
single-assign, address-of (lowerExprAsPtr), and multi-assign
(lowerMultiAssign) — now resolve through the one resolver, removing the
issue-0083 two-resolver divergence.
Fold vector-lane resolution into `fieldLvaluePtr` (reusing
vectorLaneIndex) so the single resolver covers struct fields, union
direct fields, promoted anonymous-struct union members, tuple elements,
and vector lanes — null only on a genuine miss, which every caller turns
into the read path's `emitFieldError` diagnostic.
`fieldLvaluePtr` now types every field GEP `*field_ty` (the convention
the single-assign path always used), not the bare field value type:
emitStore unwraps one pointer level to find the stored value's type.
The earlier lowerExprAsPtr / lowerMultiAssign walks typed the GEP with
the bare field type, so a field whose own type is a pointer-to-aggregate
(`*Pair`, a two-pointer struct) made emitStore unwrap to the aggregate
and coerceArg's closure auto-promotion store a 16-byte `{ptr,null}`
struct over the 8-byte slot, clobbering the neighbouring field.
Consolidating onto the one `*field_ty` resolver preserves single-assign
and fixes that pre-existing multi-assign / address-of clobber.
The types.zig `.unresolved` tripwire is untouched; no `.s64` / `.void` /
`.unresolved` default remains.
Regression: examples/0167-types-ptr-to-aggregate-field-store.sx (a
`*Pair` field stored via all three lvalue sites leaves the neighbour
intact) + a lowering unit test asserting the `*field_ty` GEP convention.
87 lines
5.5 KiB
Markdown
87 lines
5.5 KiB
Markdown
# 0094 — assigning to a missing struct field panics with "unresolved type reached LLVM emission"
|
|
|
|
> **RESOLVED** (F0.10). **Root cause:** the lvalue field lookup never diagnosed a
|
|
> missing field. In `Lowering.lowerAssignment`'s `.field_access` target path
|
|
> (`src/ir/lower.zig`), `field_ty` started as `.unresolved`; when no struct field
|
|
> matched, the code still built `ptrTo(field_ty)` / `structGepTyped` and stored —
|
|
> so a pointer-to-`.unresolved` reached LLVM emission and tripped the
|
|
> `src/backend/llvm/types.zig` tripwire. The nested lvalue-pointer path
|
|
> (`Lowering.lowerExprAsPtr`'s `.field_access` fallback) had the sibling defect:
|
|
> on a miss it returned `structGepTyped(obj_ptr, 0, .s64, obj_ty)` — a silent
|
|
> field-0/`.s64` default.
|
|
>
|
|
> **Fix (`src/ir/lower.zig`):** all three lvalue field-store sites — single
|
|
> assignment, address-of, and multi-target assignment — route field resolution
|
|
> through one shared helper, `fieldLvaluePtr(obj_ptr, obj_ty, field)`, which
|
|
> resolves struct fields, union/tagged-union direct fields, promoted
|
|
> anonymous-struct union members, tuple elements, and vector lanes (reusing
|
|
> `vectorLaneIndex`), and returns `null` (no field 0 / `.unresolved` /`.s64`
|
|
> default) when nothing matches. Each caller emits the read path's
|
|
> field-not-found diagnostic (`emitFieldError`) on a `null` result:
|
|
> 1. `lowerAssignment` `.field_access` target — delegates to `fieldLvaluePtr`;
|
|
> its own duplicated union / promoted / tuple / vector / struct walk is
|
|
> deleted (issue-0083 two-resolver divergence removed).
|
|
> 2. `lowerExprAsPtr` `.field_access` — delegates to `fieldLvaluePtr`, so the
|
|
> address-of path resolves promoted union members (`@v.x`) — not only direct
|
|
> union fields — and a genuine miss errors. The `.s64` sentinel is gone.
|
|
> 3. `lowerMultiAssign` `.field_access` target — replaced its struct-only loop
|
|
> (which defaulted `field_idx 0` / `field_ty .unresolved` on a miss, silently
|
|
> storing into field 0 — `p.q, y = 2, 3` printed `x=2 y=3`) with the shared
|
|
> `fieldLvaluePtr`; a missing field now errors, and a valid promoted-union /
|
|
> tuple member at a non-zero offset stores into its own slot, not field 0.
|
|
>
|
|
> `fieldLvaluePtr` types every GEP `*field_ty` (a pointer to the field), the
|
|
> convention the single-assign path always used: `emitStore` reads the
|
|
> store-target pointer's IR type and unwraps exactly one `.pointer` level to
|
|
> find the stored value's type. The earlier `lowerExprAsPtr` / `lowerMultiAssign`
|
|
> walks typed the GEP with the *bare* field value type, so a field whose own
|
|
> type is a pointer-to-aggregate (`*Pair`, a two-pointer struct) made `emitStore`
|
|
> unwrap to the aggregate and `coerceArg`'s closure auto-promotion store a
|
|
> 16-byte `{ptr,null}` struct over the 8-byte slot — clobbering the neighbouring
|
|
> field. Consolidating all three sites onto the one `*field_ty` resolver
|
|
> preserves single-assign and fixes that pre-existing multi-assign / address-of
|
|
> clobber.
|
|
>
|
|
> All sites reuse `emitFieldError` (the exact facility the read path
|
|
> `lowerFieldAccessOnType` uses), so the read and write paths reject identically.
|
|
> The `src/backend/llvm/types.zig` tripwire is untouched — the fix is to never
|
|
> produce `.unresolved` for a missing-field store.
|
|
>
|
|
> **Regression tests:** `examples/1145-diagnostics-missing-struct-field-assign.sx`
|
|
> (negative — single-assign, address-of, AND multi-assign missing-field all error,
|
|
> exit 1), `examples/0165-types-nested-struct-field-assign.sx` (positive — nested
|
|
> struct field write + address-of a matched field), `examples/0166-types-union-promoted-member-lvalue.sx`
|
|
> (positive — promoted union member written and address-of'd, including a non-zero
|
|
> offset member), `examples/0167-types-ptr-to-aggregate-field-store.sx` (positive —
|
|
> a `*Pair` field stored via all three lvalue sites leaves the neighbour intact),
|
|
> and three lowering unit tests in `src/ir/lower.test.zig` (single- and
|
|
> multi-assign missing-field field-not-found, plus the `*field_ty` GEP convention).
|
|
|
|
## Symptom
|
|
Assigning to a nonexistent struct field (`p.q = ...`) panics during LLVM emission instead of reporting a source diagnostic.
|
|
|
|
Observed: the compiler reaches the `.unresolved` LLVM tripwire in `src/backend/llvm/types.zig:175` via `emitStore`.
|
|
Expected: a normal compile error like `field 'q' not found on type 'Point'`, matching the read-field diagnostic path.
|
|
|
|
## Reproduction
|
|
```sx
|
|
Point :: struct { x: s64; }
|
|
|
|
main :: () {
|
|
p := Point.{ x = 1 };
|
|
p.q = 2;
|
|
}
|
|
```
|
|
|
|
Running `./zig-out/bin/sx run repro.sx` currently panics with:
|
|
```text
|
|
panic: unresolved type reached LLVM emission — a type resolution failure was not diagnosed/aborted
|
|
```
|
|
|
|
## Investigation prompt
|
|
Fix issue 0094 in the sx compiler: assigning to a missing struct field (`p.q = 2`) panics with `.unresolved` reaching LLVM emission instead of emitting a field-not-found diagnostic.
|
|
|
|
Suspected area: `src/ir/lower.zig`, especially `Lowering.lowerAssignment`'s `.field_access` target path around the struct-field lookup (`field_ty` starts as `.unresolved`, no matched field diagnoses, then `ptrTo(field_ty)` is stored) and the related `Lowering.lowerExprAsPtr` field-access fallback that returns `structGepTyped(obj_ptr, 0, .s64, obj_ty)` on lookup failure. The fix should make failed lvalue field lookup loud, reusing `emitFieldError(obj_ty, field, span)` or equivalent, and should not use `.s64`, `.void`, or any real type as a sentinel.
|
|
|
|
Verification: run the repro and expect exit 1 with a source diagnostic `field 'q' not found on type 'Point'`; no LLVM panic. Then run `zig build`, `zig build test`, and `bash tests/run_examples.sh`.
|