lang: fix struct-field null/undef over-store (issue 0154)

Assigning null/--- to a struct field picked up a leaked enclosing
target_type (the function's return type, set for the whole body), so
constNull/constUndef built a whole-struct-typed value. The oversized
store overran the field's slot and clobbered the saved frame pointer,
so the function returned to 0x0. Surfaced building a by-value-returned
struct whose array field precedes a pointer field (Scheduler.init()).

Fix: add null_literal/undef_literal to the needs_target switch in
lowerAssignment so the field's own type is used. Regression:
examples/types/0193-types-sret-array-before-pointer.sx.
This commit is contained in:
agra
2026-06-21 18:43:33 +03:00
parent bdf83db4c8
commit b1e06f21e3
6 changed files with 95 additions and 1 deletions

View File

@@ -0,0 +1,28 @@
// Assigning `null` / `---` to a struct field must store a FIELD-typed value,
// not pick up an enclosing `target_type` (e.g. the function's return type while
// lowering its body) — which would build a whole-struct-typed null and emit an
// oversized store that overruns the field's slot.
//
// Regression (issue 0154): `s.p = null` inside a struct-returning fn used to
// store a 32-byte struct `zeroinitializer` through the pointer field's GEP,
// clobbering the saved x29/x30 so the fn `ret`'d to 0x0. The array-before-
// pointer order puts the pointer at a non-zero offset, pushing the over-store
// off the end of the alloca. Fixed by adding null/undef literals to
// `needs_target` in lowerAssignment.
#import "modules/std.sx";
S :: struct {
arr: [2]u64; // fixed-array field FIRST
p: *i64; // pointer field AFTER the array
n: i64;
}
mk :: () -> S {
s : S = ---;
s.p = null; // must store a *i64 null, not a whole-struct null
s.n = 0;
return s;
}
main :: () -> i64 {
s := mk();
print("n {}\n", s.n); // n 0
return 0;
}

View File

@@ -0,0 +1 @@
n 0