fix: reject implicit ?T -> bool coercion instead of silent false (issue 0169)
The Optional->Concrete unwrap classify rule treated ?i64 -> bool as unwrap+narrow (both builtin), silently yielding false for every optional (present or null). specs.md defines no implicit optional->bool conversion. Reject it: conversions.zig adds an optional_to_bool_reject plan (dst == bool, child != bool); coerce.zig emits a located diagnostic suggesting '!= null'. Covers arg/field-init/return via the shared coerceMode. The if-opt presence test (issue 0164) is a separate path, untouched. Regression: examples/diagnostics/1199-diagnostics-optional-to-bool.sx + conversions.test.zig unit test. Verified by 3 adversarial reviews, suite 789/0. Filed adjacent issue 0179 (whole implicit ?T->concrete unwrap family silently miscompiles a null optional; design-touching).
This commit is contained in:
19
examples/diagnostics/1199-diagnostics-optional-to-bool.sx
Normal file
19
examples/diagnostics/1199-diagnostics-optional-to-bool.sx
Normal file
@@ -0,0 +1,19 @@
|
||||
// Passing an optional (`?T`) where a `bool` is expected is a hard type error,
|
||||
// not a silent miscompile. Regression (issue 0169): the coercion classifier's
|
||||
// "optional → concrete (unwrap+narrow)" rule used to accept `?i64 → bool`
|
||||
// (both child and dst are builtin), unwrapping the payload and narrowing it to
|
||||
// `i1` — which silently produced `false` for EVERY optional, present or null
|
||||
// alike. There is no implicit optional→bool coercion in the language (only
|
||||
// `T → ?T` wrapping and flow-sensitive narrowing); a bool position wants an
|
||||
// explicit presence test. The classifier now rejects `?T → bool` (payload not
|
||||
// itself a bool) at the coercion site (exit 1) with a `!= null` fix-it. The
|
||||
// `if opt {}` presence test (issue 0164) is a separate path and still works.
|
||||
|
||||
#import "modules/std.sx";
|
||||
|
||||
takes_bool :: (b: bool) { if b { print("true\n"); } else { print("false\n"); } }
|
||||
|
||||
main :: () {
|
||||
a : ?i64 = 42;
|
||||
takes_bool(a); // <- '?i64' cannot be used where 'bool' is expected
|
||||
}
|
||||
@@ -0,0 +1 @@
|
||||
1
|
||||
@@ -0,0 +1,5 @@
|
||||
error: cannot use a value of type '?i64' where 'bool' is expected; test presence explicitly with '!= null'
|
||||
--> examples/diagnostics/1199-diagnostics-optional-to-bool.sx:18:3
|
||||
|
|
||||
18 | takes_bool(a); // <- '?i64' cannot be used where 'bool' is expected
|
||||
| ^^^^^^^^^^^^^
|
||||
@@ -0,0 +1 @@
|
||||
|
||||
@@ -1,5 +1,22 @@
|
||||
# 0169 — optional passed where `bool` is expected silently coerces to `false` (always)
|
||||
|
||||
> **RESOLVED** (scoped to `?T → bool`, `T ≠ bool`). The "Optional → Concrete
|
||||
> unwrap" classify rule treated `?i64 → bool` as unwrap+narrow (both i64 and bool
|
||||
> are builtin), silently yielding `false`. specs.md defines no implicit
|
||||
> optional→bool conversion, so the fix REJECTS it: `src/ir/conversions.zig` adds
|
||||
> an `optional_to_bool_reject` plan (when `dst == bool` and `child ≠ bool`);
|
||||
> `src/ir/lower/coerce.zig` emits a located diagnostic ("…use '!= null'…")
|
||||
> instead of a constant false. Covers arg / field-init / return (all share
|
||||
> `coerceMode`). `if opt` presence-test (issue 0164) is a separate path,
|
||||
> untouched. Verified by 3 adversarial reviews; suite 789/0. Regression:
|
||||
> `examples/diagnostics/1199-diagnostics-optional-to-bool.sx` + a
|
||||
> `conversions.test.zig` unit test. NOTE: review surfaced a larger sibling
|
||||
> surface — the whole implicit `?T → concrete` unwrap family (incl. the
|
||||
> `?bool → bool` cell still allowed here) silently miscompiles a NULL optional to
|
||||
> garbage — filed as **0179** (design-touching: needs the flow-narrowing
|
||||
> decision).
|
||||
|
||||
|
||||
## Symptom
|
||||
|
||||
Passing an optional (`?T`) to a `bool` parameter (or any bool-typed position:
|
||||
|
||||
@@ -0,0 +1,71 @@
|
||||
# 0179 — implicit `?T → <concrete>` unwrap silently miscompiles a NULL optional to garbage (whole family)
|
||||
|
||||
## Symptom
|
||||
|
||||
Passing an optional `?T` where a concrete `T`/other builtin is expected (function
|
||||
arg, field initializer, `-> T` return) — WITHOUT any explicit `!`/`??`/binding —
|
||||
compiles silently and unconditionally unwraps the payload. For a PRESENT optional
|
||||
it yields the value; for a NULL optional it yields the zero/garbage payload with
|
||||
NO diagnostic. Silent miscompile across the whole `?T → concrete` family. (Issue
|
||||
0169 fixed only the `?T → bool` cell where `child != bool`; the rest of the
|
||||
family — and the `?bool → bool` cell — remain.)
|
||||
|
||||
Per specs.md §Optional Types, the ONLY legal ways to extract `T` from `?T` are
|
||||
`!` (force unwrap), `??` (coalesce), `if v := opt` / while-binding, pattern
|
||||
match, and flow-sensitive narrowing after a `!= null` guard. There is NO implicit
|
||||
unwrap-at-a-value-position. The `T → ?T` direction is the only implicit optional
|
||||
conversion sanctioned.
|
||||
|
||||
## Reproduction
|
||||
|
||||
```sx
|
||||
#import "modules/std.sx";
|
||||
takes_i32 :: (x: i32) { print("got {}\n", x); }
|
||||
main :: () {
|
||||
n : ?i64 = null;
|
||||
takes_i32(n); // prints "got 0", exit 0 — silent miscompile (no diagnostic)
|
||||
}
|
||||
```
|
||||
|
||||
Confirmed silently wrong for the NULL case across: `?i64 → i32`, `?i64 → f64`,
|
||||
`?f64 → i64`, `?i64 → u8`, `?i32 → i64` (widen), `?i64 → i64` (same type), and
|
||||
`?bool → bool` (yields `false`). Present optionals unwrap the payload; null
|
||||
optionals yield `0`/garbage. (Found during adversarial review of issue 0169.)
|
||||
|
||||
## Root cause
|
||||
|
||||
`src/ir/conversions.zig` (`CoercionResolver.classify`, the "Optional → Concrete
|
||||
unwrap" rule, ~line 114): `if (child_ty == dst_ty or (dst_ty.isBuiltin() and
|
||||
child_ty.isBuiltin()))` classifies any `?builtin → builtin` (and `?T → T`) as
|
||||
`.optional_unwrap`. The emitter `emitOptionalUnwrap`
|
||||
(`src/backend/llvm/ops.zig` ~2212) does an UNCONDITIONAL `ExtractValue` of the
|
||||
payload field — it never reads the has_value flag — so a null optional yields its
|
||||
zero payload. (The comptime VM's `optHas` path traps, so runtime and comptime
|
||||
diverge.)
|
||||
|
||||
## Investigation prompt — NOTE: this is design-touching, resolve the semantics first
|
||||
|
||||
Per spec, implicit `?T → concrete` should be REJECTED entirely (require explicit
|
||||
`!`/`??`/binding). Broadening issue 0169's `optional_to_bool_reject` to all
|
||||
`?T → concrete` is the mechanical change (in `classify`), BUT it interacts with
|
||||
flow-sensitive narrowing:
|
||||
|
||||
1. **Flow narrowing is not actually implemented as type refinement.** The
|
||||
non-binding `if x != null { use(x) }` spec example (specs.md §Flow-Sensitive
|
||||
Narrowing) currently "works" only because the broken unconditional unwrap
|
||||
fires everywhere, guard or not — there is no real narrowed type. The genuine
|
||||
narrowing mechanism is the BINDING form `if v := opt {}`
|
||||
(`src/ir/lower/control_flow.zig`), which emits an explicit `optional_unwrap`
|
||||
into a fresh `inner_ty` scope binding and does NOT route through `classify`.
|
||||
So a broad reject would reject the non-binding `if x != null { use(x) }`
|
||||
pattern. Decide: (a) implement real flow-sensitive type refinement so `x` is
|
||||
genuinely `T` inside the guarded branch (then the broad reject is safe), or
|
||||
(b) require the binding form / `!` / `??` and update specs.md §Flow-Sensitive
|
||||
Narrowing accordingly.
|
||||
2. **Close the `?bool → bool` hole** — issue 0169's `child_ty != .bool` carve-out
|
||||
still silently unwraps a null `?bool` to `false`; per spec there's no implicit
|
||||
`?bool → bool` either.
|
||||
|
||||
Whichever path, the NULL case must never silently yield garbage. Verify the
|
||||
present cases still work via the legal explicit forms; add regressions. This
|
||||
likely warrants its own focused session given the flow-narrowing decision.
|
||||
@@ -48,6 +48,13 @@ test "conversions: classify covers the built-in coercion ladder" {
|
||||
try std.testing.expectEqual(Plan.optional_unwrap, cr.classify(opt_i64, .i64));
|
||||
try std.testing.expectEqual(Plan.void_to_optional, cr.classify(.void, opt_i64));
|
||||
|
||||
// `?T → bool` is NOT an unwrap-then-narrow presence test (issue 0169):
|
||||
// it must reject, never silently produce `false`. But `?bool → bool`
|
||||
// is a genuine unwrap of a bool payload.
|
||||
try std.testing.expectEqual(Plan.optional_to_bool_reject, cr.classify(opt_i64, .bool));
|
||||
const opt_bool = tt.optionalOf(.bool);
|
||||
try std.testing.expectEqual(Plan.optional_unwrap, cr.classify(opt_bool, .bool));
|
||||
|
||||
// Tuple → tuple, same arity.
|
||||
const t_ss = tt.intern(.{ .tuple = .{ .fields = &[_]TypeId{ .i64, .i64 }, .names = null } });
|
||||
const t_ii = tt.intern(.{ .tuple = .{ .fields = &[_]TypeId{ .i32, .i32 }, .names = null } });
|
||||
|
||||
@@ -34,6 +34,7 @@ pub const CoercionResolver = struct {
|
||||
closure_to_fn_reject, // closure value → bare fn-ptr (diagnostic, returns operand)
|
||||
tuple_elementwise, // (A,B) → (C,D), same arity
|
||||
optional_unwrap, // ?T → concrete (narrowing)
|
||||
optional_to_bool_reject, // ?T → bool (no presence-test coercion; diagnostic)
|
||||
void_to_optional, // void (null literal) → ?T
|
||||
optional_wrap, // concrete → ?T
|
||||
erase_protocol, // concrete → protocol value
|
||||
@@ -99,6 +100,17 @@ pub const CoercionResolver = struct {
|
||||
const src_info = self.l.module.types.get(src_ty);
|
||||
if (src_info == .optional) {
|
||||
const child_ty = src_info.optional.child;
|
||||
// `?T → bool` is NOT a presence test. The unwrap-then-narrow
|
||||
// ladder below would extract the payload and narrow it to `i1`,
|
||||
// which silently yields `false` for every optional (issue 0169).
|
||||
// There is no implicit optional→bool coercion in the language
|
||||
// (only `T → ?T` wrapping and flow-sensitive narrowing); a bool
|
||||
// position wants an explicit presence test. Reject loudly unless
|
||||
// the payload is itself a bool (`?bool → bool` is a genuine
|
||||
// unwrap of a bool payload, handled by the same arm below).
|
||||
if (dst_ty == .bool and child_ty != .bool) {
|
||||
return .optional_to_bool_reject;
|
||||
}
|
||||
if (child_ty == dst_ty or (dst_ty.isBuiltin() and child_ty.isBuiltin())) {
|
||||
return .optional_unwrap;
|
||||
}
|
||||
|
||||
@@ -647,6 +647,16 @@ pub fn coerceMode(self: *Lowering, val: Ref, src_ty: TypeId, dst_ty: TypeId, mod
|
||||
const unwrapped = self.builder.emit(.{ .optional_unwrap = .{ .operand = val } }, child_ty);
|
||||
return self.coerceMode(unwrapped, child_ty, dst_ty, mode);
|
||||
},
|
||||
// Optional → bool: there is no implicit presence-test coercion. The
|
||||
// old unwrap-then-narrow ladder silently produced `false` for every
|
||||
// optional (issue 0169). Reject with a fix-it pointing at `!= null`.
|
||||
.optional_to_bool_reject => {
|
||||
if (self.diagnostics) |d| {
|
||||
const cs = self.builder.current_span;
|
||||
d.addFmt(.err, ast.Span{ .start = cs.start, .end = cs.end }, "cannot use a value of type '{s}' where 'bool' is expected; test presence explicitly with '!= null'", .{self.formatTypeName(src_ty)});
|
||||
}
|
||||
return val;
|
||||
},
|
||||
// string → cstring: ONLY a string LITERAL coerces implicitly — its
|
||||
// bytes are a terminated constant (Odin's literal blessing). Any
|
||||
// other string may be an unterminated view, so it must materialize
|
||||
|
||||
Reference in New Issue
Block a user