Phase 1.final of the flat-memory comptime VM — wire the host through it,
reach corpus parity, and gate it behind a build flag — plus the first
Phase 3 (compiler-API) step. Default OFF; legacy interpreter unchanged.
Host wiring + hardening:
- Machine accessors return error.OutOfBounds (no debug panic) on bad
addresses; Frame.get/set bounds-check and bail (no panic) on a malformed
operand ref (e.g. a ret Ref.none from an unresolved name).
- tryEval routed at both comptime call sites in emit_llvm — the const-init
fold and the #run side-effect path — with per-eval legacy fallback;
yields .void_val for void/noreturn entries. Both sites sx_trace_clear()
before the legacy fallback so a partial VM run that pushed trace frames
doesn't double-push on re-run.
VM coverage (all corpus const-inits except the inline-asm global):
- Implicit context materialized from the __sx_default_context global; the
full allocator protocol runs on the VM (context.allocator.alloc ->
call_indirect -> CAllocator thunk -> libc_malloc -> native flat malloc).
- Native libc memory builtins (malloc/calloc/free/memcpy/memmove/memset)
on flat memory; f32 stored/loaded as the 4-byte single; signed sub-64-bit
loads sign-extended; global_get (lazy + memoized); func_ref/call_indirect
(func-ref encoded fid+1, 0 reserved for null); string/slice fat-pointer
field access; is_comptime; the failable/error cluster (error_set tuples,
trace_frame + native sx_trace_push/clear -> raise/catch/or + return traces).
Build flag + Phase 3 seed:
- -Dcomptime-flat (build_opts module) OR SX_COMPTIME_FLAT env enables the VM;
zig build test -Dcomptime-flat runs the full corpus on the VM (688/0).
- intern/text_of serviced natively on flat memory via Vm.callCompilerFn
(compiler_welded boundary) — the seed the rest of the compiler-API grows on.
Parity 688/688 gate ON and OFF. Unit tests added throughout. The
lowering-time #insert wiring was explored and reverted (lowering-time IR can
be malformed; full malformed-IR hardening is a prerequisite, deferred).
Phase 1 of the flat-memory comptime VM (current/PLAN-COMPILER-VM.md),
built standalone + unit-tested with the legacy interpreter still live and
the corpus untouched (688 green).
src/ir/comptime_vm.zig:
- Machine: one linear byte memory (comptime stack+heap) with a bump/stack
allocator (mark/reset), scalar readWord/writeWord (1/2/4/8 LE) + byte
views; addr 0 reserved as null_addr. Frame: a Ref-indexed register file
(Reg = raw u64: immediate scalar bits OR an Addr). Target-aware layout
comes from the type table, so cross-compilation stays correct.
- Vm executor over the SAME SSA IR, mirroring the legacy interp's scalar
semantics (i64 wrapping/signed, f64). Ported: constants, arithmetic,
comparison, logical, conversions, control flow (br/cond_br/ret + block
params); structs (alloca/load/store/struct_init/get/gep at target
offsets); tuples; arrays (index_get/gep, length); slices+strings as
{ptr,len} fat pointers (const_string, data_ptr, subslice,
array_to_slice, str_eq/ne, index-through-slice); optionals (pointer and
{T,i1} shapes); payloadless enums; deref/addr_of; direct + recursive
call over the shared flat memory (depth-guarded). The value model: a
word for scalars/pointers, by-address for aggregates (a struct's value
IS its Addr). Any unported op bails loudly (error.Unsupported + detail).
- Reg<->Value boundary bridge (valueToReg / regToValue) + tryEval, the
hybrid-wiring entry point: run a comptime fn on the VM, return a legacy
Value or null to fall back. Transitional, for the legacy interop edge.
Registered in the ir.zig barrel.
The byte-weld (sx structs whose layout was validated to mirror the
compiler's Zig records) plus the serialization/marshaling bridge was the
wrong direction: it bolted a parallel layout regime and hand-built
byte-copies onto a comptime value model that fundamentally isn't bytes.
Strip the struct-weld machinery:
- compiler_lib.zig loses the type registry (weldStruct / bound_types /
BoundType / FieldLayout / findType / SxField / LayoutMismatch /
validateStructLayout); it is now just the intern/text_of function
host-call bridge (kept as the Phase-3 compiler-call seed).
- nominal.zig loses validateWeldedStruct / weldedFieldOrderStr + the
sd.abi == .zig validation call.
- Remove the struct-weld unit tests and examples 0625/0627 (welded
structs) + 1183/1186 (weld-layout diagnostics).
- The #library / abi / extern syntax stays.
Record the new direction: a bytecode VM over flat, byte-addressable
memory so comptime values are native bytes (no weld/validation/marshal),
target-aware (preserves cross-compilation) and sandboxed. See
current/PLAN-COMPILER-VM.md (Phase 0 strip -> Phase 1 flat-memory value
model -> Phase 2 bytecode -> Phase 3 compiler-API on flat memory).
design/comptime-compiler-api.md gets a SUPERSEDED banner. Also drop the
"~500 lines / split the step" rule from CLAUDE.md.
Replace the explored byte-layout-override engine (offset-ordered LLVM structs /
weld plans / byte-blobs — all unnecessary) with a much simpler design: a welded
`struct abi(.zig) extern compiler { … }` is a bodied header declaring its fields
in the bound compiler type's MEMORY order. The compiler reflects the real Zig
type (field names via @typeInfo, offsets via @offsetOf, size via @sizeOf —
nothing hand-maintained) and validates the header matches, with loud diagnostics.
On pass it is an ordinary struct whose natural layout already equals the Zig
layout — no reorder, no padding, no index/remap tables, no special LLVM path — so
@ptrCast'ing it to the compiler's own type and dereferencing is byte-identical.
When types.zig shifts, the header stops matching and the developer gets a specific
message to fix it.
- compiler_lib.zig: weldStruct reflects field names and bakes bound_types fields
in ascending-offset (memory) order; deleted computeWeldPlan/WeldPlan/WeldElement.
- nominal.zig validateWeldedStruct: precise diagnostics — field-not-found,
wrong-field-order (+ expected memory order), type-layout (size) mismatch,
total-size mismatch.
- Examples: 0627 (StructInfo in memory order, byte-identical, usable),
1186 (source-order StructInfo -> wrong-field-order diagnostic); 1183 refreshed.
- Design doc + checkpoint updated.
`zig build test -Dname=examples/0625-foo.sx[,examples/0626-bar.sx]` runs ONLY the
named example(s) — full repo-relative .sx paths, comma-separated (a leading `./`
is tolerated). Empty = run everything (unchanged default).
Why: a full `-Dupdate-goldens` re-runs and rewrites all ~690 snapshots, so one
flaky/host-divergent example (AOT links, cross-arch `target` examples) can clobber
a good snapshot. `-Dname` regenerates only the named example(s) and touches
nothing else. It also busts the cached test-run result — the corpus enumerates
.sx/expected files at runtime, so a bare snapshot edit alone is otherwise served
from cache.
- build.zig: new `name` option threaded onto corpus_paths.
- corpus_run.test.zig: `nameMatchesFilter` + a per-example skip in the run loop.
- CLAUDE.md: document the targeted-regen workflow under Snapshot integrity.
Add the COMPILER-API stream to CLAUDE.md's session-start router and a
`## ⏯ Resume` block to CHECKPOINT-COMPILER-API.md (next action = sub-step 2.2,
read order, build/verify, and the cross-arch snapshot-regen gotcha).
Introduce the welded comptime `compiler` library (`#library "compiler"` +
`abi(.zig) extern compiler`), per design/comptime-compiler-api.md, and unify
`callconv(...)` into the new `abi(...)` annotation.
abi(...) replaces callconv(...):
- New ABI enum { default, c, zig, pure }; `abi(.c|.zig|.pure)` parses in the
postfix slot before extern/export (and standalone). `kw_callconv` -> `kw_abi`.
- Migrated 52 sx files, the call-convention-mismatch diagnostic, and docs
(readme/specs) from `callconv(.c)` to `abi(.c)`.
Phase 1 — welded compiler library (parse -> registry -> validation -> bridge):
- `abi(.zig) extern compiler` parses on fn decls (carries abi/extern_lib) and
struct decls (StructDecl.abi/extern_lib).
- `#library "compiler"` is the comptime-only internal surface — never dlopen'd.
- src/ir/compiler_lib.zig: the binding registry (the safety boundary). `Field`
welded to StructInfo.Field with layout baked from the real Zig type
(@offsetOf/@sizeOf); `findType`/`findFn`. Welded structs are layout-validated
at registration (field set + total size) as a header checked against the impl.
- Host-call bridge: a `fn abi(.zig) extern compiler` dispatches under the
comptime interp to its registered Zig handler (intern/text_of round-trip),
never dlsym. IR Function.compiler_welded; validated in declareFunction.
- Comptime-only enforcement: a runtime call to a welded fn is a clean
build-gating error (emitCall), not an undefined-symbol link failure.
Phase 2.1 — byte-layout weld foundation:
- Decision: full byte-layout weld (sx struct laid out byte-identically to the
bound Zig type). Registered StructInfo (first non-natural / Zig-reordered
layout). `computeWeldPlan` — pure offset-ordered element plan + padding +
sx-field->LLVM-element remap; unit-tested. Emit/interp wiring is the next
sub-step (2.2+, see current/CHECKPOINT-COMPILER-API.md).
Examples: 0625/0626 (welded struct + fn round-trip), 1183/1184/1185
(layout-mismatch, unexported-fn, runtime-call diagnostics).
#library already lexes/parses (library_decl node); extern/export are
keywords. Phase 1 new work pinned to concrete sites: parser (extern(.zig)
postfix at the #builtin/#compiler positions), AST binding field,
compiler_hooks.zig as the registry, types/llvm layout emission, host_ffi
comptime bridge. First testable sub-step: extern(.zig) <lib> parses on a
fn decl.
Unified sx<->compiler binding that subsumes the metatype declare/define
primitives AND the #compiler struct attribute. A named 'compiler' library
exposes the compiler's real types (layout-welded via extern(.zig), offsets
queried from the Zig type at compiler-build time + a build-time equality
assertion) and functions (comptime-only, host-call bridged). declare/
define/type_info become sx library code over register_*/find_type; the
projected meta.sx TypeInfo + hand marshaling are deleted; BuildOptions
migrates onto it and #compiler is removed. Includes the safety boundary
(curated export list, guarded mutators, comptime-only), the honest limit
(the ordering law stays, but stops leaking as 'weird stages' — dissolving
the 0141 class), a phased suite-green build order, and the open risks
(union(enum) welding, optional fields, LLVM offset emission).
Wired a minimal deferral (eval at a new Pass 1c' after the CAllocator
thunks exist) — the List repro STILL bailed with struct_get, and it
destabilized examples/0620. So deferring past the thunks isn't the cause
of the wrong IR; the field-access lowering only emits struct_gep at
body-lowering/emit time. No single pass slot satisfies both 'body lowers
correctly' and 'layout ready before use'. Pivot to Direction 1 (robust
*Struct field-access lowering). Experiment reverted; tree clean.
Instrumentation shows List.append lowers list.len/list.cap to struct_gep
(correct) at #run/emit time but struct_get (wrong, value access on a *T
receiver) at scanDecls/metatype time — same source, different IR. The
function IS lowered both ways, just to wrong IR at scanDecls due to
incomplete generic-instantiation context. So an interp-side lazy-lower
hook can't fix it (IR is wrong before the interp runs); the fix is either
robust field-access lowering or deferring the comptime type-construction
eval to a complete-world pass (like #run). Supersedes the two-layer framing.
File the last METATYPE deferred enhancement: List(T).append at comptime
bails ('struct_get: base has no fields') in a type-construction ::.
Standalone repro + two-layer root cause (null comptime allocator at
scanDecls; *T slot_ptr struct_get) + investigation prompt. Non-blocking:
array-literal locals already build variant lists (examples/0620/0624).
Checkpoint + Known issues reference 0141.
Investigated the last deferred enhancement. List(T).append at comptime
fails in two independent layers (both reproduce with plain List(i64);
List works via #run because that evaluates at emit time, after lowering):
1. null comptime allocator — defaultContextValue looks up the
CAllocator->Allocator thunks by name, but they aren't lowered at
scanDecls time. Fixable by forcing getOrCreateThunks before the interp
runs in runComptimeTypeFunc (tried, works for this layer).
2. struct_get through a *T slot_ptr chain (the *List receiver) — the
deep part; comptime pointer/struct/slot resolution, its own session.
Speculative fixes reverted (no end-to-end win without layer 2).
A generic ($T) -> Type type-fn comptime-evaluated only its return
EXPRESSION, so a local declared before the return ('vs := …; return
make_enum(…, vs)') was unresolved. Now a body with a prelude (statements
before the return) has its full body evaluated: createComptimeFunction-
WithPrelude lowers the pre-return statements into the comptime function's
scope before the return expr, so the locals resolve.
- comptime.zig: createComptimeFunctionWithPrelude (prelude stmts +
expr); evalComptimeTypeBody (extract prelude + return expr, scan the
whole body for declare() forward types); runComptimeTypeFunc factored
out of evalComptimeType (shared bail/declare-never-defined handling).
- generic.zig: route a type-fn body WITH a prelude through
evalComptimeTypeBody; no-prelude bodies stay on evalComptimeType (zero
change for RecvResult/TryResult etc.).
Non-generic builders (whole body already evaluated) and the List-growth
path are unaffected. Suite green (684).
TypeInfo gains a `tuple(TupleInfo) variant (TupleInfo{elements: []Type},
positional/unnamed) — completing the reflect/construct triad with enum
and struct.
- meta.sx: TupleInfo + `tuple TypeInfo variant.
- interp: reflectTypeInfo builds .tuple (tag 2) as bare type_tag elements
(no name pairs); defineType dispatches tag 2 -> defineTuple, which
decodes []Type and completes the declare slot as a structural .tuple
via replaceKeyedInfo (kind change). Tuples are structural so the
declared name is vestigial, but the slot is still completed in place so
define returns the handle (consistent with enum/struct).
- call.zig: the lower-time type_info guard now admits .tuple.
define(declare("P"), .tuple(.{elements=.[i64,f64]})) builds a tuple, and
define(declare("T"), type_info((i64,bool,f64))) round-trips one. Suite
green (683).
TypeInfo gains a `struct(StructInfo) variant (StructField{name,type});
the metatype system now reflects AND constructs structs, not just enums.
- meta.sx: StructField / StructInfo / `struct TypeInfo variant.
- interp: reflectTypeInfo builds .struct (tag 1) for a source @"struct";
define dispatches on the TypeInfo tag (defineType) -> defineEnum (0) /
defineStruct (1). defineStruct mirrors defineEnum (dup-field-name check
included) but completes the declare slot AS a struct via replaceKeyedInfo
(a kind change re-keys the intern map; updatePreservingKey asserts no
key change, true only for the enum path).
- call.zig: the lower-time type_info guard now admits @"struct".
define(declare("P"), .struct(.{ fields = .[ … ] })) builds a struct, and
define(declare("C"), type_info(SrcStruct)) round-trips one. Suite green
(682); enum path (0619) unchanged.
Constructed-type companion to examples/1178 (source form): a declare/
define enum whose variant references itself BY VALUE is rejected by the
same checkInfiniteSize guard ('infinitely sized'). Pins the use-before-
define corner of the validation story — by-value self-reference is the
one self-ref shape that isn't legal; *L (pointer) is fine (see 0618).
No compiler change (locks existing behavior).
A bare declare("X") with no define left a zero-field nominal slot that
panicked at codegen (verifySizes: llvm_size != ir_size). evalComptimeType
now detects a zero-variant tagged_union result and emits a clean
build-gating diagnostic naming the type — a zero-variant enum is never a
legitimate construction result (defineEnum rejects empty variant lists
too). Self-reference (a declared slot completed by define) is unaffected.
Two same-named variants in a constructed enum silently succeeded —
construction (.a) and matching would ambiguously pick one. defineEnum
now bails when a variant name repeats, naming it. The name is dynamic so
it sets last_bail_detail directly (bailDetail takes a comptime string);
evalComptimeType renders it as a build-gating diagnostic.
make_enum from dirs[0..2] — mints Axis from a comptime SUBSLICE of a
local EnumVariant array. Locks the interp subslice-over-non-string-
aggregate fix (d22037c); previously bailed.
`arr[lo..hi]` at comptime bailed for any non-string base — the interp's
.subslice op only handled string-backed values. Worse, the open-ended
`hi` came from a .length op that misread a 2-element array as a {ptr,len}
fat pointer (returning the 2nd element, not the count), so even lo/hi
weren't valid ints.
Fix, interp-only (runtime already handles arrays via LLVMTypeOf):
- Thread the base operand's IR type onto the Subslice op (base_ty); the
interp uses it to tell a bare array (elements = aggregate fields) from a
{data,len} slice (elements in the data field) — indistinguishable by
Value shape alone.
- Fold an open-ended slice's hi to the array's static length for fixed
arrays at lower time (runtime emitLength folds the same constant, so the
IR result is unchanged — no snapshot churn — but the comptime interp no
longer hits the ambiguous .length op).
- subsliceElements() resolves the element list (array/slice, inline or
slot_ptr-backed) and subslice returns a proper {data,len} slice value.
Suite green (678), no .ir changes.
make_enum(name, variants: []EnumVariant) -> Type mints a nominal enum
from a variant list passed as a VALUE, not a hardcoded literal — the
open-ended form the channel-result constructors are special cases of.
Pure sx over declare/define; no compiler machinery.
Because variants is an ordinary comptime value, a non-generic builder
can ASSEMBLE it in a local before minting. examples/0620: build_level
fills a local array, then make_enum mints Level from it — exercising
define decoding a value-arg SLICE (decodeVariantElements' slice branch),
vs. the inline .[ … ] array the 0614-0618 examples pass directly.
No compiler change (locks existing capability). Suite green (678).
Move the issue 0140 repro into the feature suite as a regression test.
Asserts the build-gating diagnostic 'comptime type construction failed:
comptime define(): enum has no variants' at the construction site, exit
1 — locking out the prior 'unresolved type reached LLVM emission' panic.
evalComptimeType did `interp.call(...) catch return null`, dropping the
interp's last_bail_detail; callers poisoned to .unresolved with no
diagnostic, so the sentinel reached LLVM emission and panicked
("unresolved type reached LLVM emission"), or hid behind a downstream
cascade.
Clear last_bail_detail before the call; on the catch emit a build-gating
.err at the construction expr's span ("comptime type construction
failed: {detail}", mirroring the #run surfacing in emit_llvm.zig), then
return null to keep the .unresolved poison — now gated by a real message
so no unresolved type reaches emission unannounced.
Empty-variant define now prints 'comptime define(): enum has no
variants' and exits 1 (no panic); make_enum-style computed-slice
failures show their root reason at the construction site.
A failing declare/define (e.g. empty variant list) bails correctly in
the interp, but evalComptimeType swallows last_bail_detail via
`catch return null`; the decl poisons to .unresolved with no diagnostic
and reaches LLVM emission -> panic ("unresolved type reached LLVM
emission"), or hides behind a misleading downstream cascade.
Pre-existing (plain define path), surfaced while starting the make_enum
step. Blocks make_enum's computed (pointer-backed) []EnumVariant slice
decode. Repro + investigation prompt filed; CHECKPOINT-METATYPE marked
BLOCKED. Session paused pending fix per CLAUDE.md IMPASSABLE rule.
type_info reflects an enum / tagged-union INTO a TypeInfo value — the
inverse of define's decode — so define(declare(n), type_info(T)) mints
a byte-identical copy with NO literal variant list.
- inst.zig: new BuiltinId.type_info (comptime-only, like declare/define).
- lower/call.zig: replace the 'not yet implemented' bail. Resolve $T at
lower time, reject non-enum/non-tagged-union loudly with a good span,
emit callBuiltin(.type_info, [const_type], TypeInfo).
- interp.zig: reflectTypeInfo builds the exact nested-aggregate Value
defineEnum decodes — variant {name,payload}, slice {data,len}, EnumInfo
{variants}, TypeInfo {tag0, EnumInfo}. tagged_union reflects field.ty
(tagless already void); payloadless `enum` reflects void per variant.
- emit: unchanged — type_info is always comptime-evaluated, the existing
comptime-only else arm (shared with declare/define) never fires.
0619 turns green: a source enum (circle:f64 / rect:i64 / empty) reflected
and reconstructed, constructs and matches like the original.
type_info($T) is still unimplemented, so the round-trip
define(declare("ShapeCopy"), type_info(Shape)) bails with
"type_info is not yet implemented" plus the downstream
enum-inference cascade. Snapshot pins that current behavior;
the next commit implements type_info and turns this green.
A nominal aggregate that contains itself (or a mutual peer) BY VALUE has no
finite layout and infinite-recursed typeSizeBytes into a stack overflow —
for SOURCE enums/structs as well as comptime-constructed types.
New `checkInfiniteSize` pass (lower/decl.zig, Pass 1g — after type
registration, before body lowering): walks the by-VALUE containment graph
(pointer/slice/optional payloads break the cycle, so `*Self` stays valid);
on a back-edge it emits a loud diagnostic — "type 'X' is infinitely sized
(it contains itself by value); use a pointer ('*X') to break the cycle" —
and poisons the offending field to `.unresolved` so sizing can't recurse
before the build halts on the error. Covers source + declare/define types,
direct + mutual recursion.
examples/1178 locks the diagnostic; issue 0139 marked RESOLVED. This also
completes METATYPE PLAN F5's by-value-self-reference rejection. Full suite
green (675).
Discovered while testing metatype self-reference: a by-VALUE self-ref
(`payload = List`, not `*List`) infinite-loops typeSizeBytes → segfault
instead of a loud "infinite size" diagnostic. PRE-EXISTING — a hand-written
source enum `enum { node: Bad; leaf }` crashes identically, so it's a
general type-system gap (the comptime F5 by-value-rejection inherits the
fix). Filed per the IMPASSABLE rule; metatype checkpoint notes it.
examples/0618 mints a recursive `List` enum (`cons: *List; nil`) via
declare("List")/define, builds a 3-node list, matches the pointer payload
directly and via deref, and counts it recursively. Locks the self-reference
capability. Full suite green (674).
