Problem
The parser only handles ?T when T is a single-token type (identifier or primitive). It rejects:
device: ?[]const u8 // optional slice (struct field)
fn get(): ?[]const u8 { ... } // optional slice return
tail: ?*?*RcuCallback // chained optional/pointer
runtime_suspend: ?fn(d: *D) -> i32 // optional fn-pointer (likely fails too)Today: error[EP0001]: Expected type name at the type position after ?.
?u32, ?Foo, ?*T (single pointer) all work; the failure is when ? precedes a complex type form.
Affected kernel files (sample)
- kernel/src/perf/rcu_tree.home:88 (
tail: ?*?*RcuCallback) - kernel/src/iot/voice_languages.home:387 (
device: ?[]const u8and similar fields) - kernel/src/iot/voice_routines.home:337 (
?[]const u8return type) - kernel/src/power/device_pm.home (multiple
Option<fn(...) -> i32>— distinct issue but related)
Expected behavior
?T should accept any type expression for T, identical to what struct-field type or variable-annotation accepts:
?T(named type)?u32(primitive)?*T(pointer)?*const T(const pointer)?*?*T(chained — optional pointer to optional pointer)?[]T/?[]const T(slice / const slice)?[N]T(fixed array)?fn(...) Ret(optional function pointer)??T(double optional, not unusual in some contexts)
Implementation hint
Where ? is consumed as an optional-type prefix, recursively call the full parseTypeAnnotation (or whatever entry point handles compound types) instead of the narrow type-name reader. The recursion handles all the nested forms naturally.
Tests
- Positive: each of the 7 forms above
- Regression:
?u32,?Foo,?*Tkeep working
Acceptance
- The 3 listed kernel files clear
home-debug astpast the optional-type point. - Existing test suite green.