comptime VM: Phase 3 — type_kind + type_field_value readers (read side complete)

The last two read-only readers the metatype's type_info(T) needs, each backed by
a TypeTable query both the legacy handler and the VM call (no drift):

  type_kind(t: TypeId) -> i64            (kindCode; stable discriminant, total — never bails)
  type_field_value(t: TypeId, idx) -> i64 (memberValue; enum explicit value or ordinal)

kindCode codes (compiler-owned, stable): 0 other / 1 struct / 2 enum /
3 tagged_union / 4 tuple / 5 union / 6 array / 7 vector / 8 error_set.

With these, the READ side is complete: find_type + type_kind + type_field_count +
type_field_{name,type} + type_nominal_name + type_field_value cover everything
reflectTypeInfo reads — a comptime sx fn can fully reflect a struct/enum/tuple
into data with no #builtin.

Example 0630 reflects Color / WindowFlags(flags) / Point. VM unit test added.

Revised forward direction: the write side will be ONE register_type(info) fn that
branches on the kind in the compiler (subsuming define's per-kind dispatch), not a
per-kind register_struct.

Parity 691/691 (gate OFF and -Dcomptime-flat).
This commit is contained in:
agra
2026-06-18 09:47:23 +03:00
parent d23e208430
commit 27bc301651
10 changed files with 252 additions and 29 deletions

View File

@@ -26,30 +26,33 @@ with ONE welded mechanism. Branch: `reify` (off `master`). Update after every st
> breaks cross-compilation — host vs target layout — and loses the sandbox. A
> flat-memory VM keeps both while getting native bytes + speed.)
>
> **Next action (2026-06-18):** **Phase 3 is UNDER WAY.** The VM now hosts five read-only
> reflection readers, all on the same plain-`u32`-handle shape (a `TypeId` is a `u32`, like
> `StringId`), so the calls stay clean scalar host-calls — handle in, scalar out, no
> marshaling: `find_type(name) -> TypeId`, `type_field_count(t) -> i64`,
> `type_nominal_name(t) -> StringId`, `type_field_name(t, idx) -> StringId`,
> `type_field_type(t, idx) -> TypeId`. Each is backed by a `TypeTable` query
> (`findByName`/`memberCount`/`nominalName`/`memberName`/`memberType`) that BOTH the legacy
> handler and the VM call, so the two paths can't drift. Examples `0628` (`find_type` +
> `type_field_count`) and `0629` (field reflection over `Pair { lo: Point; hi: Point }`),
> both VM-HANDLED natively. Parity **690/690** (gate ON and OFF), VM unit tests added.
> Phase 1.final op-porting was already complete (the VM covers
> scalars/control-flow/aggregates/strings/optionals/enums, calls+recursion, the implicit
> context + full allocator protocol, globals, failables + return traces); both comptime
> call sites route through the VM with legacy fallback.
> **Forward (P3.3):** `register_struct` (the first MUTATING fn — mints a `TypeId`; resolve
> the mutable-table / host-ABI-vs-target-ABI boundary deliberately), plus any remaining
> read-only readers the metatype needs (kind query, `field_value_int`). Re-expressing
> `declare`/`define`/`type_info` as sx (the metatype, which runs at LOWERING time) needs the
> VM hardened against malformed lowering-time IR first — keep it on the legacy path until
> then. Phase 2 (bytecode) is the orthogonal speed work. **Decisions recorded:** `find_type`
> returns a non-optional `TypeId` using the `unresolved` (0) sentinel, NOT `?Type`; reader
> names use the `type_*` family to avoid colliding with the std metatype builtins
> (`field_name`/`type_name` in core.sx) — see `PLAN-COMPILER-VM.md` Phase 3 progress note.
> Build/verify: `zig build && zig build test` (690, gate OFF). Run the corpus ON the VM:
> **Next action (2026-06-18):** **Phase 3 is UNDER WAY — the READ side is COMPLETE.** The VM
> hosts seven read-only reflection readers, all on the same plain-`u32`-handle shape (a
> `TypeId` is a `u32`, like `StringId`), so the calls stay clean scalar host-calls — handle
> in, scalar out, no marshaling: `find_type(name) -> TypeId`, `type_kind(t) -> i64`,
> `type_field_count(t) -> i64`, `type_nominal_name(t) -> StringId`,
> `type_field_name(t, idx) -> StringId`, `type_field_type(t, idx) -> TypeId`,
> `type_field_value(t, idx) -> i64`. Each is backed by a `TypeTable` query
> (`findByName`/`kindCode`/`memberCount`/`nominalName`/`memberName`/`memberType`/`memberValue`)
> that BOTH the legacy handler and the VM call, so the two paths can't drift. Together they
> cover everything `reflectTypeInfo` reads. Examples `0628``0630`, all VM-HANDLED natively.
> Parity **691/691** (gate ON and OFF), VM unit tests added. Phase 1.final op-porting was
> already complete (the VM covers scalars/control-flow/aggregates/strings/optionals/enums,
> calls+recursion, the implicit context + full allocator protocol, globals, failables +
> return traces); both comptime call sites route through the VM with legacy fallback.
> **Forward (P3.3)revised direction:** the WRITE side is ONE `register_type(info)` fn that
> takes a type-info value and **branches on the kind in the compiler** (subsuming
> `define`'s per-kind dispatch), NOT a per-kind `register_struct`. Open design points when
> reached: the flat-memory shape of `info`, the mutable-table / host-ABI-vs-target-ABI
> boundary, pointer-escape/lifetime. Re-expressing `declare`/`define`/`type_info` as sx (the
> metatype, which runs at LOWERING time) needs the VM hardened against malformed
> lowering-time IR first — keep it on the legacy path until then. Phase 2 (bytecode) is the
> orthogonal speed work. **Decisions recorded:** `find_type` returns a non-optional `TypeId`
> using the `unresolved` (0) sentinel, NOT `?Type`; reader names use the `type_*` family to
> avoid colliding with the std metatype builtins (`field_name`/`type_name` in core.sx); the
> write side is a single kind-branching `register_type` — see `PLAN-COMPILER-VM.md` Phase 3
> progress note.
> Build/verify: `zig build && zig build test` (691, gate OFF). Run the corpus ON the VM:
> `zig build test -Dcomptime-flat` (the build flag) OR env `SX_COMPTIME_FLAT=1`. Coverage
> trace: `SX_COMPTIME_FLAT_TRACE=1`.
@@ -322,6 +325,20 @@ when reached (sentinels or accessor fns; see the design doc Risks).
`List` growth; orthogonal, see `current/CHECKPOINT-METATYPE.md`.)
## Log
- **Phase 3 P3.2b (VM plan) — kind + enum-value readers: `type_kind` + `type_field_value`; READ side complete (2026-06-18).**
The last two read-only readers the metatype's `type_info(T)` needs (added to
`compiler_lib.bound_fns` AND `Vm.callCompilerFn`, each backed by a `TypeTable` query both
call): `type_kind(t) -> i64` (`kindCode` — a stable, compiler-owned discriminant: 0 other ·
1 struct · 2 enum · 3 tagged_union · 4 tuple · 5 union · 6 array · 7 vector · 8 error_set;
TOTAL, never bails) and `type_field_value(t, idx) -> i64` (`memberValue` — an enum variant's
explicit value or ordinal; mirrors the `field_value_int` builtin; loud-bail for non-enum /
out-of-range). Example `0630-comptime-compiler-type-kind` reflects `Color` / `WindowFlags`
(flags) / `Point`. **The READ side is now COMPLETE** — `find_type` + `type_kind` +
`type_field_count` + `type_field_name`/`type_field_type`/`type_nominal_name` +
`type_field_value` cover everything `reflectTypeInfo` reads. VM unit test added. **Parity
691/691** (gate ON and OFF). **Revised forward direction (per the user):** the WRITE side is
ONE `register_type(info)` fn that branches on the kind IN THE COMPILER (subsuming `define`'s
per-kind dispatch), not a per-kind `register_struct`.
- **Phase 3 P3.2 (VM plan) — field-level reflection readers: `type_nominal_name` + `type_field_name` + `type_field_type` (2026-06-18).**
Three more `compiler`-library readers on the same `TypeId`-handle shape (added to
`compiler_lib.bound_fns` AND `Vm.callCompilerFn`), each backed by a new `TypeTable` query

View File

@@ -289,11 +289,29 @@ host through it:
exist in `core.sx`). Example `0629` reflects `Pair { lo: Point; hi: Point }` — reads each
field name and the nominal name of a field's type, all folded at `#run`, all VM-HANDLED
natively. Parity **690/690** (gate ON and OFF); VM unit test added.
- **Next (P3.3):** `register_struct` (the first MUTATING fn — mints a `TypeId`; resolve the
mutable-table / host-ABI-vs-target-ABI boundary deliberately, per the open questions),
plus any remaining read-only readers needed to re-express the metatype (kind query,
`field_value_int` for enums). Re-expressing `declare`/`define`/`type_info` as sx (the
metatype, which runs at LOWERING time) still needs the VM hardened against malformed
- **(P3.2b) Kind + enum-value readers — `type_kind` + `type_field_value` (DONE).** The last
two read-only readers the metatype's `type_info(T)` needs, completing the READ side: a
comptime sx fn can now fully reflect a struct/enum/tagged-union/tuple into data with no
`#builtin`. `type_kind(t: TypeId) -> i64` (`TypeTable.kindCode` — a stable, compiler-owned
discriminant: 0 other · 1 struct · 2 enum · 3 tagged_union · 4 tuple · 5 union · 6 array ·
7 vector · 8 error_set; TOTAL — never bails, an unnamed/non-aggregate type reads `other`)
and `type_field_value(t: TypeId, idx: i64) -> i64` (`TypeTable.memberValue` — an enum
variant's explicit value or ordinal; mirrors the `field_value_int` builtin; loud-bail for
a non-enum / out-of-range idx). Example `0630` reflects `Color`/`WindowFlags`(flags)/`Point`.
Parity **691/691** (gate ON and OFF); VM unit test added.
- **READ side now complete:** `find_type` + `type_kind` + `type_field_count` +
`type_field_name` + `type_field_type` + `type_nominal_name` + `type_field_value` cover
everything `reflectTypeInfo` reads.
- **Next (P3.3) — ONE `register_type(info)` write fn (revised direction, 2026-06-18):** per
the user, the mutating side is NOT per-kind (`register_struct`/`register_enum`/…) but a
SINGLE function that takes a type-info value and **branches on the kind in the compiler**,
minting the right `TypeInfo`. This subsumes `define`'s `defineStruct`/`defineEnum`/
`defineTuple` dispatch into one host-side switch. Open design points to resolve when
reached: the flat-memory shape of the `info` argument the sx side passes (a tagged
`{ kind, payload }` over the readers' handle types), the mutable-table / host-ABI-vs-
target-ABI boundary, and pointer-escape/lifetime (escaping field arrays copied into
compiler-owned memory at the boundary). Re-expressing `declare`/`define`/`type_info` as sx
(the metatype, which runs at LOWERING time) still needs the VM hardened against malformed
lowering-time IR first — keep that on the legacy path until then (see the resume note in
CHECKPOINT-COMPILER-API.md).

View File

@@ -0,0 +1,47 @@
// Comptime compiler API — type-kind + enum-value reflection readers (Phase 3).
//
// Completes the READ side the metatype needs to re-express `type_info(T)` as sx:
//
// type_kind(t) → a stable kind discriminant, to branch on the shape
// (0 other · 1 struct · 2 enum · 3 tagged_union ·
// 4 tuple · 5 union · 6 array · 7 vector · 8 error_set)
// type_field_value(t, i) → enum variant i's integer value (explicit or ordinal)
//
// Together with find_type / type_field_count / type_field_name / type_field_type
// / type_nominal_name (examples 06280629), a comptime sx function can now fully
// reflect a struct/enum/tuple into data — no `#builtin` needed. All folded at
// `#run`, all serviced natively by the flat-memory VM.
#import "modules/std.sx";
compiler :: #library "compiler";
StringId :: u32;
TypeId :: u32;
intern :: (s: string) -> StringId abi(.zig) extern compiler;
text_of :: (id: StringId) -> string abi(.zig) extern compiler;
find_type :: (name: StringId) -> TypeId abi(.zig) extern compiler;
type_kind :: (t: TypeId) -> i64 abi(.zig) extern compiler;
type_field_name :: (t: TypeId, idx: i64) -> StringId abi(.zig) extern compiler;
type_field_value :: (t: TypeId, idx: i64) -> i64 abi(.zig) extern compiler;
Color :: enum { red; green; blue; }
WindowFlags :: enum flags u32 { vsync :: 64; resizable :: 4; hidden :: 128; }
Point :: struct { x: i64; y: i64; }
color_kind :: #run type_kind(find_type(intern("Color"))); // 2 = enum
point_kind :: #run type_kind(find_type(intern("Point"))); // 1 = struct
// Plain enum → ordinal values.
green_val :: #run type_field_value(find_type(intern("Color")), 1); // 1
// Flags enum → explicit values, read by name + value.
vsync_name :: #run text_of(type_field_name(find_type(intern("WindowFlags")), 0)); // "vsync"
vsync_val :: #run type_field_value(find_type(intern("WindowFlags")), 0); // 64
main :: () {
print("Color kind = {}, Point kind = {}\n", color_kind, point_kind);
print("Color.green = {}\n", green_val);
print("WindowFlags.{} = {}\n", vsync_name, vsync_val);
}

View File

@@ -0,0 +1 @@
0

View File

@@ -0,0 +1 @@

View File

@@ -0,0 +1,3 @@
Color kind = 2, Point kind = 1
Color.green = 1
WindowFlags.vsync = 64

View File

@@ -52,6 +52,8 @@ pub const bound_fns = [_]BoundFn{
.{ .sx_name = "type_nominal_name", .handler = handleTypeNominalName },
.{ .sx_name = "type_field_name", .handler = handleTypeFieldName },
.{ .sx_name = "type_field_type", .handler = handleTypeFieldType },
.{ .sx_name = "type_kind", .handler = handleTypeKind },
.{ .sx_name = "type_field_value", .handler = handleTypeFieldValue },
};
/// Look up a compiler function by its sx name. Returns null when the name is not
@@ -151,3 +153,20 @@ fn handleTypeFieldType(interp: *Interpreter, args: []const Value) InterpError!Va
const mty = interp.module.types.memberType(tid, args[1].int) orelse return error.TypeError;
return Value{ .int = mty.index() };
}
/// `type_kind(t: TypeId) -> i64` — the stable kind discriminant (see
/// `TypeTable.kindCode`). Total: an unnamed/non-aggregate type reads `other` (0).
fn handleTypeKind(interp: *Interpreter, args: []const Value) InterpError!Value {
if (args.len != 1) return error.TypeError;
const tid: types.TypeId = @enumFromInt(try handleArg(args, 0));
return Value{ .int = interp.module.types.kindCode(tid) };
}
/// `type_field_value(t: TypeId, idx: i64) -> i64` — enum variant `idx`'s integer
/// value (explicit or ordinal). Loud error for a non-enum or out-of-range idx.
fn handleTypeFieldValue(interp: *Interpreter, args: []const Value) InterpError!Value {
if (args.len != 2 or args[1] != .int) return error.TypeError;
const tid: types.TypeId = @enumFromInt(try handleArg(args, 0));
const v = interp.module.types.memberValue(tid, args[1].int) orelse return error.TypeError;
return Value{ .int = v };
}

View File

@@ -907,6 +907,70 @@ test "comptime_vm exec: compiler-fn type_field_name/type/nominal_name (native re
try std.testing.expectEqual(@as(i64, @intFromEnum(point_name)), toI64(try v.run(module.getFunction(main2), &.{})));
}
test "comptime_vm exec: compiler-fn type_kind + type_field_value (native reflection)" {
const alloc = std.testing.allocator;
var module = Module.init(alloc);
defer module.deinit();
// A struct and an enum with explicit values.
const pfields = [_]types.TypeInfo.StructInfo.Field{
.{ .name = module.types.internString("x"), .ty = .i64 },
.{ .name = module.types.internString("y"), .ty = .i64 },
};
const point = module.types.intern(.{ .@"struct" = .{ .name = module.types.internString("Point"), .fields = &pfields } });
const variants = [_]types.StringId{ module.types.internString("ok"), module.types.internString("missing") };
const evals = [_]i64{ 200, 404 };
const status = module.types.intern(.{ .@"enum" = .{
.name = module.types.internString("Status"),
.variants = &variants,
.explicit_values = &evals,
} });
// extern type_kind(t: u32) -> i64 [compiler] (FuncId 0)
const kp = [_]Function.Param{param(.u32)};
var kb = Fb.init(alloc, &kp, .i64);
kb.func.is_extern = true;
kb.func.compiler_welded = true;
kb.func.name = module.types.internString("type_kind");
const kind_id = module.addFunction(kb.func);
// extern type_field_value(t: u32, idx: i64) -> i64 [compiler] (FuncId 1)
const vp = [_]Function.Param{ param(.u32), param(.i64) };
var vb = Fb.init(alloc, &vp, .i64);
vb.func.is_extern = true;
vb.func.compiler_welded = true;
vb.func.name = module.types.internString("type_field_value");
const val_id = module.addFunction(vb.func);
var v = vm.Vm.init(alloc);
v.table = &module.types;
v.module = &module;
defer v.deinit();
// type_kind(Point) → 1 (struct); type_kind(Status) → 2 (enum).
inline for (.{ .{ point, 1 }, .{ status, 2 } }) |case| {
var fb = Fb.init(alloc, &.{}, .i64);
const b0 = fb.block(&.{});
const t = fb.add(b0, inst(.{ .const_int = @intFromEnum(case[0]) }, .u32));
const kargs = [_]Ref{ref(t)};
const k = fb.add(b0, inst(.{ .call = .{ .callee = kind_id, .args = &kargs } }, .i64));
_ = fb.add(b0, inst(.{ .ret = .{ .operand = ref(k) } }, .void));
const mid = module.addFunction(fb.func);
try std.testing.expectEqual(@as(i64, case[1]), toI64(try v.run(module.getFunction(mid), &.{})));
}
// type_field_value(Status, 1) → 404 (explicit value).
var fb = Fb.init(alloc, &.{}, .i64);
const b0 = fb.block(&.{});
const t = fb.add(b0, inst(.{ .const_int = @intFromEnum(status) }, .u32));
const one = fb.add(b0, inst(.{ .const_int = 1 }, .i64));
const vargs = [_]Ref{ ref(t), ref(one) };
const val = fb.add(b0, inst(.{ .call = .{ .callee = val_id, .args = &vargs } }, .i64));
_ = fb.add(b0, inst(.{ .ret = .{ .operand = ref(val) } }, .void));
const mid = module.addFunction(fb.func);
try std.testing.expectEqual(@as(i64, 404), toI64(try v.run(module.getFunction(mid), &.{})));
}
test "comptime_vm exec: func_ref + call_indirect dispatch" {
const alloc = std.testing.allocator;
var module = Module.init(alloc);

View File

@@ -1067,6 +1067,19 @@ pub const Vm = struct {
return self.failMsg("comptime type_field_type: out-of-range idx or member has no type");
return @as(Reg, mty.index());
}
if (std.mem.eql(u8, name, "type_kind")) {
if (args.len != 1) return self.failMsg("comptime type_kind: expected one TypeId arg");
const tid = try self.argTypeId(args, frame, 0);
return @as(Reg, @bitCast(table.kindCode(tid))); // total — never bails
}
if (std.mem.eql(u8, name, "type_field_value")) {
if (args.len != 2) return self.failMsg("comptime type_field_value: expected (TypeId, idx)");
const tid = try self.argTypeId(args, frame, 0);
const idx: i64 = @bitCast(frame.get(args[1].index()));
const v = table.memberValue(tid, idx) orelse
return self.failMsg("comptime type_field_value: non-enum or out-of-range idx");
return @as(Reg, @bitCast(v));
}
return null; // not a known compiler function → caller bails to legacy
}

View File

@@ -554,6 +554,46 @@ pub const TypeTable = struct {
};
}
/// Stable kind discriminant of a type, for comptime reflection branching.
/// TOTAL (never fails): an unnamed / non-aggregate type or an out-of-range id
/// is `other` (0). Codes are compiler-owned and stable — NOT tied to any sx
/// enum's declaration order; the sx side maps them. Backs the `type_kind`
/// reader. (A `tagged_union` is a payload-carrying enum; the sx metatype folds
/// codes 2 and 3 onto its single `.enum` TypeInfo variant.)
/// 0 other · 1 struct · 2 enum · 3 tagged_union · 4 tuple
/// 5 union · 6 array · 7 vector · 8 error_set
pub fn kindCode(self: *const TypeTable, id: TypeId) i64 {
if (id.index() >= self.infos.items.len) return 0;
return switch (self.get(id)) {
.@"struct" => 1,
.@"enum" => 2,
.tagged_union => 3,
.tuple => 4,
.@"union" => 5,
.array => 6,
.vector => 7,
.error_set => 8,
else => 0,
};
}
/// Integer value of enum variant `idx`: its explicit value when the enum
/// declares one (custom values or flags), else its ordinal. Null for a
/// non-enum type, a negative / out-of-range `idx`, or an out-of-range id.
/// Backs the `type_field_value` reader (mirrors the `field_value_int` builtin).
pub fn memberValue(self: *const TypeTable, id: TypeId, idx: i64) ?i64 {
if (idx < 0 or id.index() >= self.infos.items.len) return null;
const i: usize = @intCast(idx);
return switch (self.get(id)) {
.@"enum" => |e| blk: {
if (i >= e.variants.len) break :blk null;
if (e.explicit_values) |vals| if (i < vals.len) break :blk vals[i];
break :blk @intCast(i); // ordinal default
},
else => null,
};
}
/// Source-sensitive variant of `findByName`: asserts at most one named type
/// matches, then returns it (or null). Quarantines the global first-match
/// scan — new resolver code that must not silently pick a first-of-many