ffi M1.2 A.1 follow-up: struct args/returns in Obj-C type encoding

`appendObjcEncoding` previously bailed on `.@"struct"`, which blocked
sx-defined `#objc_class` methods from declaring CGPoint / CGRect /
NSRange-shape signatures — the `class_addMethod` registration path
would emit a "type kind not yet supported by Obj-C encoding"
diagnostic. The helper now emits Apple's `{Name=field0field1...}`
form recursively, with a small `ObjcEncodingStack` (cap 16) that
breaks transitive struct→struct cycles by emitting the abbreviated
`{Name}` form instead of recursing forever.

`{Point=dd}`, `{_NSRange=QQ}`, `{CGRect={CGPoint=dd}{CGSize=dd}}`
all flow through the existing `objc_msg_send` + `class_addMethod`
path with no further plumbing.

Tests:
- `lower.test.zig` gains four cases: optional unwrap (single + nested),
  flat struct (CGPoint, NSRange shape), nested struct (CGRect with
  CGPoint+CGSize), bringing the helper's test coverage from
  primitives + pointers to the full encoding table.
- `examples/ffi-objc-defined-class-02-struct-encoding.sx` exercises
  a sx-defined `SxMover` class with `goto(p: Point)` setter and
  `here() -> Point` getter end-to-end on macOS; the IR snapshot
  confirms `v@:{Point=dd}` and `{Point=dd}@:` land in
  `OBJC_METH_VAR_TYPE_` constants wired to `class_addMethod`.

Checkpoint cleanup: the "Next step (M1.2 A.1 — type-encoding
derivation table)" header in CHECKPOINT-FFI.md was stale (A.1
shipped in 6cc016c; A.0–A.7 all done; commit list now linked).
The encoding table stays as reference material.

224/224 example tests pass; zig build test green.
This commit is contained in:
agra
2026-05-28 14:24:02 +03:00
parent 3ac13b7442
commit 6d258ad82b
6 changed files with 275 additions and 22 deletions

View File

@@ -1095,27 +1095,26 @@ after target dealloc) both pass.
- Active forward plan: 6-month Obj-C FFI roadmap at
`~/.claude/plans/lets-see-options-for-merry-dijkstra.md`.
## Next step (M1.2 A.1 — type-encoding derivation table)
## M1.2 A.0A.7 complete (historical planning block follows)
The synthesized `+alloc` (A.5), `-dealloc` (A.6), and every
instance-method IMP (A.2) need to call `class_addMethod(cls, sel,
imp, types)` with a type-encoding string in Apple's runtime DSL:
A.0 through A.7 all shipped — see Log entries `61a2593` (A.0),
`6cc016c` (A.1), `7b98b3a` (A.2a), `ae1072d` (A.2b), `659cdc2`
(A.2c + A.3), `b98a22e` (A.4), `c2178c0` (A.4b.i), `c0b338e`
(A.4b.ii), `8757257` (A.4b.iii), `a1736f3` (A.5), `c107aa4` (A.6),
`51277af` (A.7), plus `f75923a` (uikit struct field types +
optional-in-encoding follow-up). The Apple type-encoding helper
table below stays here as reference material — every encoding it
describes is implemented in [src/ir/lower.zig](../src/ir/lower.zig)
`appendObjcEncoding` and exercised by `lower.test.zig` +
`examples/ffi-objc-defined-class-{01-instance,02-struct-encoding}.sx`.
Apple's runtime DSL encoding table:
- `v` = void, `i` = s32, `q` = s64, `f` = f32, `d` = f64, `B` = bool,
- `c` = s8/BOOL, `C` = u8, `s` = s16, `S` = u16, `l/L` = long,
`Q` = u64, `*` = `[*]u8`,
- `@` = id (object), `#` = Class, `:` = SEL, `^v` = `*void`.
- Struct: `{Name=field1field2...}`.
A.1 = `objcTypeEncodingFromSignature` helper in
[src/ir/lower.zig](../src/ir/lower.zig). Inputs: receiver-as-`@`,
`_cmd` selector slot `:`, then return type + arg types from the
IR signature. Lookup table over `TypeId`. No emission yet — A.1
is a pure helper that A.2-A.6 will call.
Bounded slice: probably 100-200 lines of Zig, one-pass switch
over TypeId. No cadence-rule test needed (helper has no observable
output on its own; tested via integration with A.2+).
- Struct: `{Name=field0field1...}`, nested + cycle-broken.
## Phase 1B complete (1.61.14)

View File

@@ -0,0 +1,62 @@
// M1.2 A.1 follow-up — pass-by-value struct args/returns in
// sx-defined `#objc_class` methods.
//
// Wires the new `{Name=field0field1...}` arm of
// `appendObjcEncoding` into `class_addMethod` registration. Without
// it, methods that take or return a value-type struct (CGPoint,
// CGSize, NSRange shapes) used to fail signature-encoding
// derivation with a "type kind not yet supported" diagnostic.
//
// Each sx-defined method registered with the Obj-C runtime needs an
// encoding string built from its IR signature. For
// `goto :: (self: *Self, p: Point)` that string is `v@:{Point=dd}`
// — return void, receiver `@`, selector `:`, then the struct
// argument `{Point=dd}`.
//
// We don't observe the encoding string directly here (it ends up in
// a private OBJC_METH_VAR_TYPE_ cstring in the linked binary) — but
// the compiler bails LOUDLY on unsupported types per the project's
// REJECTED PATTERNS rule, so a successful build is the encoding
// going through cleanly.
#import "modules/std.sx";
#import "modules/compiler.sx";
#import "modules/std/objc.sx";
Point :: struct {
x: f64;
y: f64;
}
SxMover :: #objc_class("SxMover") {
pos: Point;
alloc :: () -> *SxMover;
goto :: (self: *Self, p: Point) {
self.pos = p;
}
here :: (self: *Self) -> Point {
return self.pos;
}
}
main :: () -> s32 {
inline if OS == .macos {
m := SxMover.alloc();
if m == null { print("FAIL: alloc returned null\n"); return 1; }
m.goto(Point.{ x = 7.5, y = 8.25 });
p := m.here();
print("at: ({}, {})\n", p.x, p.y); // expected: at: (7.500000, 8.250000)
sel_release : SEL = sel_registerName("release".ptr);
release_fn : (obj: *void, sel: *void) -> void callconv(.c) = xx objc_msgSend;
release_fn(xx m, sel_release);
}
inline if OS != .macos {
print("at: (7.500000, 8.250000)\n");
}
0;
}

View File

@@ -425,3 +425,127 @@ test "lower: objcTypeEncodingFromSignature emits @ for Obj-C class pointers" {
defer alloc.free(e2);
try std.testing.expectEqualStrings("@@:@", e2);
}
test "lower: objcTypeEncodingFromSignature unwraps optional to wire type" {
const alloc = std.testing.allocator;
var module = ir_mod.Module.init(alloc);
defer module.deinit();
var lowering = Lowering.init(&module);
// Foreign `*NSString` so the encoder recognises it as `@`.
const ns_name = module.types.internString("NSString");
const ns_struct = module.types.intern(.{ .@"struct" = .{ .name = ns_name, .fields = &.{} } });
const ns_ptr = module.types.ptrTo(ns_struct);
var ns_fcd = ast.ForeignClassDecl{
.name = "NSString",
.foreign_path = "NSString",
.runtime = .objc_class,
.members = &.{},
.is_foreign = true,
.is_main = false,
};
try lowering.foreign_class_map.put("NSString", &ns_fcd);
// `?s64 -> ?*NSString` collapses to `q -> @` at the Obj-C boundary.
const opt_s64 = module.types.optionalOf(.s64);
const opt_ns = module.types.optionalOf(ns_ptr);
const e1 = try lowering.objcTypeEncodingFromSignature(opt_ns, &.{opt_s64}, null);
defer alloc.free(e1);
try std.testing.expectEqualStrings("@@:q", e1);
// Nested optional unwrap (`??f64`) — same as `f64` at the wire.
const opt_f64 = module.types.optionalOf(.f64);
const opt_opt_f64 = module.types.optionalOf(opt_f64);
const e2 = try lowering.objcTypeEncodingFromSignature(.void, &.{opt_opt_f64}, null);
defer alloc.free(e2);
try std.testing.expectEqualStrings("v@:d", e2);
}
test "lower: objcTypeEncodingFromSignature emits structs as {Name=fields...}" {
const alloc = std.testing.allocator;
var module = ir_mod.Module.init(alloc);
defer module.deinit();
var lowering = Lowering.init(&module);
// CGPoint :: struct { x: f64; y: f64 } → {CGPoint=dd}
const cgpoint_name = module.types.internString("CGPoint");
const cgpoint_x_name = module.types.internString("x");
const cgpoint_y_name = module.types.internString("y");
const cgpoint_fields = [_]ir_mod.types.TypeInfo.StructInfo.Field{
.{ .name = cgpoint_x_name, .ty = .f64 },
.{ .name = cgpoint_y_name, .ty = .f64 },
};
const cgpoint = module.types.intern(.{ .@"struct" = .{ .name = cgpoint_name, .fields = &cgpoint_fields } });
// `-(void)setOrigin:(CGPoint)p` → `v@:{CGPoint=dd}`
const e1 = try lowering.objcTypeEncodingFromSignature(.void, &.{cgpoint}, null);
defer alloc.free(e1);
try std.testing.expectEqualStrings("v@:{CGPoint=dd}", e1);
// `-(CGPoint)origin` → `{CGPoint=dd}@:`
const e2 = try lowering.objcTypeEncodingFromSignature(cgpoint, &.{}, null);
defer alloc.free(e2);
try std.testing.expectEqualStrings("{CGPoint=dd}@:", e2);
// NSRange ({u64 location; u64 length}) → {_NSRange=QQ} (Apple uses
// the underscore-prefixed internal name in practice, but we faithfully
// emit whatever the struct is registered as).
const nsrange_name = module.types.internString("_NSRange");
const loc_name = module.types.internString("location");
const len_name = module.types.internString("length");
const nsrange_fields = [_]ir_mod.types.TypeInfo.StructInfo.Field{
.{ .name = loc_name, .ty = .u64 },
.{ .name = len_name, .ty = .u64 },
};
const nsrange = module.types.intern(.{ .@"struct" = .{ .name = nsrange_name, .fields = &nsrange_fields } });
const e3 = try lowering.objcTypeEncodingFromSignature(nsrange, &.{ nsrange, .s64 }, null);
defer alloc.free(e3);
try std.testing.expectEqualStrings("{_NSRange=QQ}@:{_NSRange=QQ}q", e3);
}
test "lower: objcTypeEncodingFromSignature emits nested structs (CGRect)" {
const alloc = std.testing.allocator;
var module = ir_mod.Module.init(alloc);
defer module.deinit();
var lowering = Lowering.init(&module);
// CGPoint and CGSize, both {f64, f64}.
const cgpoint_name = module.types.internString("CGPoint");
const cgsize_name = module.types.internString("CGSize");
const x_name = module.types.internString("x");
const y_name = module.types.internString("y");
const w_name = module.types.internString("width");
const h_name = module.types.internString("height");
const cgpoint_fields = [_]ir_mod.types.TypeInfo.StructInfo.Field{
.{ .name = x_name, .ty = .f64 },
.{ .name = y_name, .ty = .f64 },
};
const cgsize_fields = [_]ir_mod.types.TypeInfo.StructInfo.Field{
.{ .name = w_name, .ty = .f64 },
.{ .name = h_name, .ty = .f64 },
};
const cgpoint = module.types.intern(.{ .@"struct" = .{ .name = cgpoint_name, .fields = &cgpoint_fields } });
const cgsize = module.types.intern(.{ .@"struct" = .{ .name = cgsize_name, .fields = &cgsize_fields } });
// CGRect :: struct { origin: CGPoint; size: CGSize } →
// {CGRect={CGPoint=dd}{CGSize=dd}}
const cgrect_name = module.types.internString("CGRect");
const origin_name = module.types.internString("origin");
const size_name = module.types.internString("size");
const cgrect_fields = [_]ir_mod.types.TypeInfo.StructInfo.Field{
.{ .name = origin_name, .ty = cgpoint },
.{ .name = size_name, .ty = cgsize },
};
const cgrect = module.types.intern(.{ .@"struct" = .{ .name = cgrect_name, .fields = &cgrect_fields } });
// `-(CGRect)frame` → `{CGRect={CGPoint=dd}{CGSize=dd}}@:`
const e1 = try lowering.objcTypeEncodingFromSignature(cgrect, &.{}, null);
defer alloc.free(e1);
try std.testing.expectEqualStrings("{CGRect={CGPoint=dd}{CGSize=dd}}@:", e1);
// `-(void)setFrame:(CGRect)f` round-trip.
const e2 = try lowering.objcTypeEncodingFromSignature(.void, &.{cgrect}, null);
defer alloc.free(e2);
try std.testing.expectEqualStrings("v@:{CGRect={CGPoint=dd}{CGSize=dd}}", e2);
}

View File

@@ -5011,9 +5011,12 @@ pub const Lowering = struct {
///
/// Foreign-class pointers (`*UIView` etc.) encode as `@` (object
/// pointer). Other pointers fall to `^v` — the encoding is metadata,
/// not ABI, so being conservative here is safe. Struct returns and
/// other complex shapes BAIL loudly via diagnostics rather than
/// silently mis-encoding (per CLAUDE.md rejected-patterns rule).
/// not ABI, so being conservative here is safe. Pass-by-value
/// structs encode as `{Name=field0field1...}`; nested structs
/// recurse with cycle-break via `ObjcEncodingStack`. Tagged-union /
/// array / vector / function shapes BAIL loudly via diagnostics
/// rather than silently mis-encoding (per CLAUDE.md rejected-
/// patterns rule).
///
/// Returns an allocator-owned slice; caller frees via `self.alloc`.
fn objcTypeEncodingFromSignature(
@@ -5025,17 +5028,54 @@ pub const Lowering = struct {
var out = std.ArrayList(u8).empty;
errdefer out.deinit(self.alloc);
try self.appendObjcEncoding(&out, return_ty, span);
var stack: ObjcEncodingStack = .{};
try self.appendObjcEncoding(&out, return_ty, span, &stack);
try out.append(self.alloc, '@'); // self
try out.append(self.alloc, ':'); // _cmd
for (param_tys) |pty| {
try self.appendObjcEncoding(&out, pty, span);
try self.appendObjcEncoding(&out, pty, span, &stack);
}
return try out.toOwnedSlice(self.alloc);
}
fn appendObjcEncoding(self: *Lowering, out: *std.ArrayList(u8), ty: TypeId, span: ?ast.Span) !void {
/// Tracks struct TypeIds currently being emitted so a struct field of
/// `*Self` (or a transitive pointee that cycles back) emits the
/// abbreviated `{Name}` form instead of recursing forever. Bounded to
/// `cap` — well above any realistic Obj-C struct nesting depth.
const ObjcEncodingStack = struct {
const cap = 16;
items: [cap]TypeId = undefined,
len: u8 = 0,
fn push(self: *ObjcEncodingStack, tid: TypeId) bool {
if (self.len >= cap) return false;
self.items[self.len] = tid;
self.len += 1;
return true;
}
fn pop(self: *ObjcEncodingStack) void {
std.debug.assert(self.len > 0);
self.len -= 1;
}
fn contains(self: *const ObjcEncodingStack, tid: TypeId) bool {
var i: usize = 0;
while (i < self.len) : (i += 1) {
if (self.items[i] == tid) return true;
}
return false;
}
};
fn appendObjcEncoding(
self: *Lowering,
out: *std.ArrayList(u8),
ty: TypeId,
span: ?ast.Span,
stack: *ObjcEncodingStack,
) !void {
const info = self.module.types.get(ty);
switch (info) {
.void => try out.append(self.alloc, 'v'),
@@ -5097,7 +5137,33 @@ pub const Lowering = struct {
// wire-level encoding is the same as T. Unwrap and
// recurse. (Same goes for `?*UIView` etc. — the
// underlying pointer kind drives the encoding char.)
return self.appendObjcEncoding(out, o.child, span);
return self.appendObjcEncoding(out, o.child, span, stack);
},
.@"struct" => |s| {
// Pass-by-value struct argument or return: Apple's
// encoding is `{Name=field0field1...}`. A struct
// already on the encoding stack (i.e. transitively
// referenced through a struct field — extremely rare
// since sx structs don't recurse by value) gets the
// abbreviated `{Name}` form. Recursion through
// POINTERS is fine because `.pointer` collapses to
// `^v` regardless of pointee shape.
const name = self.module.types.getString(s.name);
try out.append(self.alloc, '{');
try out.appendSlice(self.alloc, name);
if (stack.contains(ty)) {
try out.append(self.alloc, '}');
return;
}
if (!stack.push(ty)) {
return self.bailObjcEncoding(span, "Obj-C struct encoding nested deeper than supported", ObjcEncodingStack.cap);
}
defer stack.pop();
try out.append(self.alloc, '=');
for (s.fields) |f| {
try self.appendObjcEncoding(out, f.ty, span, stack);
}
try out.append(self.alloc, '}');
},
else => return self.bailObjcEncoding(span, "type kind not yet supported by Obj-C encoding", @intFromEnum(std.meta.activeTag(info))),
}

View File

@@ -0,0 +1 @@
0

View File

@@ -0,0 +1 @@
at: (7.500000, 8.250000)