// selfhost/cmd/wcc/cgenutil.ww — split out of cgen.ww.
//
// General helpers used across cgenexpr / cgenstmt / cgendecl:
//   - pushargsrev: per-call arg pushing
//   - type predicates: isstr*/isslice*/istagged*/nodeis* families
//   - field ops:    fieldloadop, fieldstoreop
//   - index helpers: elemsizeof, elemsizeofc
//   - slot sizing:  structlookup, primsize, slotsize, fieldsize,
//                   registerstruct, collectstructs
//   - rhs helpers:  taggedvariantindex
//
// Bundler pulls this in transitively via cgen.ww; consumers don't
// need to `use cgenutil;` directly.

package wcc;

import os;
import ast;
import tok;
import typ;
import sym;
import strconv;

// ---- variadic-call helpers (Hare-style `T...` param) -----------------

// slicewrap — synthesise an N_TSLICE node wrapping the given element
// type AST. Used by the Hare-style variadic path so the local entry
// for the param (callee side) and the call-site slice descriptor
// (caller side) both advertise their effective type as []ELEM —
// every isslicetype / nodeisslice check then succeeds naturally.
fn slicewrap(c: *cgen, elem: *node) *node = {
	let s: *node = newnode(nkind.N_TSLICE, "", 0, 0);
	s.lhs = elem;
	return s;
};

// findvariadicparam — walk a param-list head and return the variadic
// param node (the one with op == TK_ELLIPSIS) plus the count of
// non-variadic params before it. Returns nil/0 when no variadic.
// nfixed_out cannot be nil.
fn findvariadicparam(ps: *node, nfixed_out: *i32) *node = {
	*nfixed_out = 0;
	let p: *node = ps;
	for (p != nil) {
		if (p.kind == nkind.N_PARAM) {
			if (p.op == tkind.TK_ELLIPSIS) {
				return p;
			};
			*nfixed_out += 1;
		};
		p = p.next;
	};
	return nil;
};

// callee_variadic_param — convenience wrapper: looks up the callee
// by name and finds its variadic param + nfixed. Returns nil if the
// callee isn't registered or has no variadic param.
//
// N_DOT routes through fnparamslookupmod with the module hint
// (callee.lhs.str) — bare fnparamslookup walks same-module-first
// (#4d) which is wrong for a cross-module N_DOT call into a module
// whose same-leaf fn has divergent variadic-vs-non-variadic shape.
// #4d explicitly deferred this re-routing; surfaced by #16 when
// strings.contains gained a variadic shape and a caller's
// bytes.contains call site picked strings.contains' variadic
// params for arg-prep while emitting CALL bytes.contains.
fn callee_variadic_param(c: *cgen, callee: *node, nfixed_out: *i32) *node = {
	*nfixed_out = 0;
	if (callee == nil) { return nil; };
	let ps: *node = nil;
	if (callee.kind == nkind.N_IDENT) {
		if (callee.str.len == 0) { return nil; };
		ps = fnparamslookup(c, callee.str);
	} else { if (callee.kind == nkind.N_DOT) {
		if (callee.str.len == 0) { return nil; };
		let cmod: str;
		cmod.ptr = nil; cmod.len = 0;
		if (callee.lhs != nil) {
			if (callee.lhs.kind == nkind.N_IDENT) {
				cmod = callee.lhs.str;
			};
		};
		ps = fnparamslookupmod(c, callee.str, cmod);
	}; };
	return findvariadicparam(ps, nfixed_out);
};

// mkvarargname — fresh local-slot name "<prefix><seq>". Used for
// the per-variadic-call scratch buffers (`@vararg_d_N` for the
// element-data buffer, `@vararg_sl_N` for the 24B slice descriptor).
// N is recorded on the N_CALL node at first emit so re-entry into
// cgcall picks the same names regardless of walk order.
fn mkvarargname(c: *cgen, prefix: str, seq: i32) str = {
	let buf: [128]u8;
	let i: i32 = 0;
	let j: i32 = 0;
	for (j < prefix.len) {
		buf[i] = prefix[j];
		i += 1; j += 1;
	};
	let ns: str = strconv.i64tos(seq: i64, strconv.base.DEC);
	let n: i32 = ns.len;
	let dk: i32 = 0;
	for (dk < n) { buf[i + dk] = ns.ptr[dk]; dk += 1; };
	let total: i32 = i + n;
	let p: []u8 = alloc([], (total: u64) + 1u64)!;
	let k: i32 = 0;
	for (k < total) { p[k] = buf[k]; k += 1; };
	p[total] = 0u8;
	let r: str;
	r.ptr = p.ptr;
	r.len = total;
	return r;
};

// ---- expression cgen -------------------------------------------------

// pushargsrev — recursively walks the arg list, evaluates rightmost
// first, and pushes. str args take two slots (ptr in AX, len in BX);
// the order on the stack so a left-to-right pop into argregs lands
// (ptr, len) correctly is: PUSHQ BX (top), PUSHQ AX (above) — the
// pop sequence then yields AX, then BX.
//
// `param` is the corresponding declared parameter for `arg` (N_PARAM
// node from the callee's signature) or nil. When param's type is a
// tagged union and `arg`'s surface type is a concrete variant of it,
// we materialise (tag, value-words, pad) for the parameter slot before
// pushing — mirrors cmd/w6c/cgen.c's call-arg widening.
fn pushargsrev(c: *cgen, arg: *node, param: *node) i32 = {
	if (arg == nil) { return 0; };
	let nextparam: *node = nil;
	if (param != nil) { nextparam = param.next; };
	let rest: i32 = pushargsrev(c, arg.next, nextparam);
	// Implicit widening from a concrete variant to a tagged-union
	// parameter slot. Skips when the arg is already a tagged local
	// (line 121's slice-or-tagged shortcut handles that).
	let widensz: i32 = 0;
	let widentag: i32 = 0;
	if (param != nil) {
		if (param.kind == nkind.N_PARAM) {
			// Hare-style variadic `T...`: effective param type is
			// []T (slice). The arg here is the synthesised slice
			// descriptor (or a forwarded `xs...` slice), not a
			// value of T being widened into a tagged slot — skip
			// the widening detection so the slice-ident fast path
			// at the bottom of pushargsrev gets the push.
			if (param.op == tkind.TK_ELLIPSIS) {
				widensz = 0;
			} else {
				let ptype: *node = param.lhs;
				if (istaggedtype(c, ptype)) {
					let aistagged: bool = false;
					if (arg.kind == nkind.N_IDENT) {
						let lc: *local = localfindnode(c, arg.str);
						if (lc != nil) {
							aistagged = istaggedtype(c, lc.tnode);
						};
					};
					// #21: a CALL returning a tagged-union must
					// skip widening — cgexpr leaves AX=tag,
					// DX=word0, CX=word1, R8=word2 per the
					// tagged-return ABI; the widening branch would
					// treat AX as a concrete payload and silently
					// drop DX/CX/R8. Restrict to the matching-slot
					// case (mirrors cstage type_eq at
					// cmd/w6c/cgen.c:4216-4221); tagged-source
					// widening into a wider slot is out of scope.
					if (taggedcallslot(c, arg) == slotsize(c, ptype)) {
						aistagged = true;
					};
					// #12: N_INDEX of a sum-typed slice element —
					// cgindex emits the same AX/DX/CX/R8 tagged ABI.
					// Without this gate the widening scalar branch
					// hardcodes the param's first-variant tag and
					// the callee reads a fixed arm on garbage.
					// #60: arg.type_ is the checker-stamped element
					// tinfo (check.ww indexresult); istaggedtype/
					// slotsize read .type_, so feed the N_INDEX node
					// directly. cstage reads the element via
					// base->type->sub (cmd/w6c/cgen.c:3518).
					if (arg.kind == nkind.N_INDEX) {
						if (istaggedtype(c, arg)) {
							if (slotsize(c, arg) == slotsize(c, ptype)) {
								aistagged = true;
							};
						};
					};
					if (!aistagged) {
						widensz = slotsize(c, ptype);
						let tagged: *node = resolvetagged(c, ptype);
						let t: i32 = taggedvariantindex(c, tagged, arg);
						if (t < 0) { t = 0; };
						widentag = t;
					};
				};
			};
		};
	};
	if (widensz == 8) {
		// Nullable fold: pointer value IS the discriminator. No
		// separate tag word.
		cgexpr(c, arg);
		emitline("\tPUSHQ\tAX\n");
		return rest + 1;
	};
	if (widensz > 0) {
		// Struct-payload widening into a tagged-union param uses
		// @tagscr (zero + cgwidentaggedstore writes fields + tag,
		// then push slot words high → low). Scalar / str go via
		// the direct push fast path below — keeps wwstage's asm
		// byte-identical to cstage for selfhost source.
		let pname: str = rhsstructpayload(c, arg);
		if (pname.len > 0) {
			let ptype: *node = param.lhs;
			let scroff: i32 = localadd(c, "@tagscr", widensz, nil);
			emitline("\tXORQ\tAX, AX\n");
			let zz: i32 = 0;
			for (zz < widensz) {
				emitline("\tMOVQ\tAX, ");
				emitoff((scroff + zz): i64);
				emitline("(BP)\n");
				zz += 8;
			};
			cgwidentaggedstore(c, ptype.type_: *tinfo, arg, "BP", scroff, widensz);
			let pp: i32 = widensz - 8;
			for (pp >= 0) {
				emitline("\tMOVQ\t");
				emitoff((scroff + pp): i64);
				emitline("(BP), AX\n");
				emitline("\tPUSHQ\tAX\n");
				pp -= 8;
			};
			return rest + widensz / 8;
		};
		cgexpr(c, arg);
		if (nodeisslice(c, arg)) {
			// Slice payload (24B): cgexpr leaves (AX=ptr, BX=len,
			// CX=cap). Slot layout: [+0]=tag, [+8]=ptr, [+16]=len,
			// [+24]=cap. Push high→low so pop drains tag first.
			// Requires widensz >= 32; a smaller slot would mean the
			// destination union doesn't list slice as a variant
			// (caller should have flagged a type error).
			emitline("\tPUSHQ\tCX\n");
			emitline("\tPUSHQ\tBX\n");
			emitline("\tPUSHQ\tAX\n");
			emitline("\tMOVQ\t$");
			emitint(widentag: i64);
			emitline(", AX\n");
			emitline("\tPUSHQ\tAX\n");
		} else { if (nodeisstr(c, arg)) {
			// str IS []u8: slot 32 [+0]=tag,[+8]=ptr,[+16]=len,
			// [+24]=cap — same shape as the slice arm above. Push
			// cap, len, ptr, tag high→low so pop drains tag first
			// into arg-reg[0] (#1/Phase 3).
			emitline("\tPUSHQ\tCX\n");
			emitline("\tPUSHQ\tBX\n");
			emitline("\tPUSHQ\tAX\n");
			emitline("\tMOVQ\t$");
			emitint(widentag: i64);
			emitline(", AX\n");
			emitline("\tPUSHQ\tAX\n");
		} else {
			// Scalar variant: single value word at +8. Pad a zero
			// high word when slot is 24B (some other variant of
			// the union is 16B-shaped).
			let pp: i32 = widensz - 8;
			for (pp > 8) {
				emitline("\tXORQ\tDX, DX\n");
				emitline("\tPUSHQ\tDX\n");
				pp -= 8;
			};
			emitline("\tPUSHQ\tAX\n");
			emitline("\tMOVQ\t$");
			emitint(widentag: i64);
			emitline(", AX\n");
			emitline("\tPUSHQ\tAX\n");
		};};
		return rest + widensz / 8;
	};
	// nkind.N_SLICE expression as arg: `buf[lo:hi]` builds a slice header
	// on the stack matching C cgen's sequence — push base, push hi,
	// compute lo, pop into BX/CX, derive len/ptr, push (cap, len, ptr).
	if (arg.kind == nkind.N_SLICE) {
		let base: *node = arg.lhs;
		let lo: *node = arg.rhs;
		let hi: *node = arg.cond;
		let baselocal: *local = nil;
		let globaltn: *node = nil;
		let globalname: str;
		globalname.ptr = nil; globalname.len = 0;
		if (base != nil) {
			if (base.kind == nkind.N_IDENT) {
				let bn: str = base.str;
				baselocal = localfindnode(c, bn);
				if (baselocal == nil) {
					let gt: *node = letvartnode(c, bn);
					if (gt != nil) {
						globaltn = gt;
						globalname = bn;
					};
				};
			};
		};
		// esz from the type table for an N_IDENT base (#76; mirrors
		// the cgindex idiom). Non-ident base stays esz=1 -> ptr
		// unscaled, matching cstage's base->kind==N_IDENT gate.
		let esz: i32 = 1;
		if (baselocal != nil) {
			esz = elemsizeofc(c, baselocal.tnode);
		} else { if (globaltn != nil) {
			esz = elemsizeofc(c, globaltn);
		};};
		// base address → push
		if (baselocal != nil) {
			let tn: *node = baselocal.tnode;
			if (tn != nil) {
				if (tn.kind == nkind.N_TARRAY) {
					emitline("\tLEAQ\t");
					emitoff(baselocal.off: i64);
					emitline("(BP), AX\n");
				} else {
					emitline("\tMOVQ\t");
					emitoff(baselocal.off: i64);
					emitline("(BP), AX\n");
				};
			} else {
				emitline("\tMOVQ\t");
				emitoff(baselocal.off: i64);
				emitline("(BP), AX\n");
			};
		} else { if (globaltn != nil) {
			if (globaltn.kind == nkind.N_TARRAY) {
				emitline("\tLEAQ\t");
				emitsymname(c, globalname);
				emitline("(SB), AX\n");
			} else {
				emitline("\tMOVQ\t");
				emitsymname(c, globalname);
				emitline("(SB), AX\n");
			};
		} else {
			cgexpr(c, base);
		};};
		emitline("\tPUSHQ\tAX\n");
		// hi (default base length) → push
		if (hi != nil) {
			cgexpr(c, hi);
		} else { if (baselocal != nil) {
			let tn: *node = baselocal.tnode;
			if (tn != nil) {
				if (tn.kind == nkind.N_TARRAY) {
					let lenn: *node = tn.rhs;
					if (lenn != nil) {
						if (lenn.kind == nkind.N_INTLIT) {
							emitline("\tMOVQ\t$");
							emituint(lenn.uval);
							emitline(", AX\n");
						};
					};
				} else { if (tn.kind == nkind.N_TSLICE) {
					emitline("\tMOVQ\t");
					emitoff((baselocal.off + 8): i64);
					emitline("(BP), AX\n");
				} else { if (tn.kind == nkind.N_TNAME) {
					if (streq(tn.str, "str")) {
						emitline("\tMOVQ\t");
						emitoff((baselocal.off + 8): i64);
						emitline("(BP), AX\n");
					};
				};};};
			};
		} else { if (globaltn != nil) {
			if (globaltn.kind == nkind.N_TARRAY) {
				let lenn: *node = globaltn.rhs;
				if (lenn != nil) {
					if (lenn.kind == nkind.N_INTLIT) {
						emitline("\tMOVQ\t$");
						emituint(lenn.uval);
						emitline(", AX\n");
					};
				};
			} else { if (globaltn.kind == nkind.N_TSLICE) {
				emitline("\tLEAQ\t");
				emitsymname(c, globalname);
				emitline("(SB), CX\n");
				emitline("\tMOVQ\t8(CX), AX\n");
			};};
		} else {
			emitline("\tMOVQ\t$0, AX\n");
		};};};
		emitline("\tPUSHQ\tAX\n");
		// lo (default 0) → AX
		if (lo != nil) { cgexpr(c, lo); }
		else { emitline("\tMOVQ\t$0, AX\n"); };
		emitline("\tPOPQ\tBX\n");           // hi
		emitline("\tPOPQ\tCX\n");           // base
		emitline("\tMOVQ\tBX, DX\n");       // DX = hi
		emitline("\tSUBQ\tAX, DX\n");       // DX = hi - lo = len
		// ptr = base + lo*esz (#76; ensure.ha:30 membsz-unit).
		// BX=lo*esz; AX=lo PRESERVED for cap. BX (dead hi) reloaded
		// by cgbasecap below.
		if (esz > 1) {
			emitline("\tMOVQ\t$");
			emitint(esz: i64);
			emitline(", BX\n");
			emitline("\tIMULQ\tAX, BX\n");
			emitline("\tADDQ\tBX, CX\n");
		} else {
			emitline("\tADDQ\tAX, CX\n");   // CX = base + lo = ptr
		};
		// cap = base_cap - lo (#20); AX=lo, BX free.
		if (cgbasecap(c, base, "BX")) {
			emitline("\tSUBQ\tAX, BX\n");
			emitline("\tPUSHQ\tBX\n");  // cap
		} else {
			emitline("\tPUSHQ\tDX\n");  // cap = len
		};
		emitline("\tPUSHQ\tDX\n");          // len
		emitline("\tPUSHQ\tCX\n");          // ptr (top)
		return rest + 3;
	};
	// Slice/tagged ident args: emit per-register MOVQ+PUSHQ pairs in
	// reverse order (cap/v1, len/v0, ptr/tag) so a left-to-right pop
	// into argregs lands the canonical (ptr/tag, len/v0, cap/v1).
	// For tagged ident with a >24B slot (slice-payload variant),
	// push a fourth word from off+24.
	if (arg.kind == nkind.N_IDENT) {
		let nm: str = arg.str;
		let lc: *local = localfindnode(c, nm);
		if (lc != nil) {
			let off: i32 = lc.off;
			if (isslicetype(c, lc.tnode) || istaggedtype(c, lc.tnode)) {
				let nwords: i32 = 3;
				if (istaggedtype(c, lc.tnode)) {
					let ssz: i32 = slotsize(c, lc.tnode);
					nwords = ssz / 8;
				};
				let w: i32 = nwords - 1;
				for (w >= 0) {
					emitline("\tMOVQ\t");
					emitoff((off + w*8): i64);
					emitline("(BP), AX\n");
					emitline("\tPUSHQ\tAX\n");
					w -= 1;
				};
				return rest + nwords;
			};
			// By-value struct ident: load qword(s) from the slot
			// and push high → low so left-to-right pop on the
			// callee side lands word 0 / word 1 into the SysV arg
			// register pair. Mirrors cstage cgen.c §4240 (call
			// site) so the wwstage prologue's new struct spill arm
			// (cgendecl.ww structparamsize branch) sees the same
			// reg layout. Pre-#11 the call-site fell through to
			// `cgexpr(c, arg)` + scalar PUSHQ AX — only the first
			// 8B word made it across, and the callee's second-arg
			// slots picked up the wrong neighbour's value.
			let stsz: i32 = structparamsize(c, lc.tnode);
			if (stsz > 0) {
				if (stsz > 8) {
					emitline("\tMOVQ\t");
					emitoff((off + 8): i64);
					emitline("(BP), AX\n");
					emitline("\tPUSHQ\tAX\n");
				};
				emitline("\tMOVQ\t");
				emitoff(off: i64);
				emitline("(BP), AX\n");
				emitline("\tPUSHQ\tAX\n");
				let nw: i32 = 1;
				if (stsz > 8) { nw = 2; };
				return rest + nw;
			};
		};
	};
	// Float arg: cgexpr leaves the value in X0. Push 8 bytes from
	// X0 via SUBQ+MOVSD so cgcall's pop side can drain into the
	// XMM stream (X0..X7). f32 still occupies 8B on the stack —
	// the MOVSS load on the pop side touches only the low 4.
	let fk: i32 = 0;
	if (arg != nil) {
		let at: *tinfo = arg.type_: *tinfo;
		if (typeisf32(at)) { fk = 1; }
		else { if (typeisfloat(at)) { fk = 2; }; };
	};
	if (fk != 0) {
		cgexpr(c, arg);
		let mov: str = "MOVSD";
		if (fk == 1) { mov = "MOVSS"; };
		emitline("\tSUBQ\t$8, SP\n");
		emitline("\t");
		emitline(mov);
		emitline("\tX0, (SP)\n");
		return rest + 1;
	};
	cgexpr(c, arg);
	// #163: tuple ARG (param twin of #164's return). cgexpr left the
	// tuple in the return-ABI cursor (AX/DX/CX/R8 + X0/X1); restage it
	// into @tupargscr by SysV class (tupstore, the #164 helper) and push
	// the slot words high->low so the pop drains slot+0 first into the
	// SysV ARG cursor. The frame slot decouples the return-class regs
	// from the overlapping arg-class regs. rettupleof scopes to an
	// N_CALL producer (tuple idents/literals as values are a separate
	// unimplemented gap; the SEND never pushes stale regs, rule 7).
	let tuparg: *node = rettupleof(c, arg);
	if (tuparg != nil) {
		let gptot: i32 = 0;
		let sstot: i32 = 0;
		let tsz: i32 = 0;
		let p: *node = tuparg.list;
		for (p != nil) {
			let et: *node = p.lhs;
			let wide: bool = isstrtype(c, et) || isslicetype(c, et);
			if (isfloattype(c, et)) { sstot += 1; }
			else { gptot += tupebytes(wide); };
			tsz += slotsize(c, et);
			p = p.next;
		};
		// The producing call already satisfied #164's return caps;
		// guard anyway (tupstore indexes [AX,DX,CX,R8] / [X0,X1]).
		if (gptot > 4 || sstot > 2) {
			let msg: str = "tuple arg exceeds return-cursor ABI capacity; see #163/#164\n";
			os.write(2, msg.ptr, msg.len: u64);
			os.exit(1);
		};
		let scr: i32 = localadd(c, "@tupargscr", tsz, nil);
		let gpcur: i32 = 0;
		let ssecur: i32 = 0;
		let eoff: i32 = 0;
		p = tuparg.list;
		for (p != nil) {
			let et: *node = p.lhs;
			let wide: bool = isstrtype(c, et) || isslicetype(c, et);
			tupstore(c, gpcur, ssecur, scr + eoff, wide, et);
			if (isfloattype(c, et)) { ssecur += 1; }
			else { gpcur += tupebytes(wide); };
			eoff += slotsize(c, et);
			p = p.next;
		};
		let w: i32 = tsz - 8;
		for (w >= 0) {
			emitline("\tMOVQ\t");
			emitoff((scr + w): i64);
			emitline("(BP), AX\n");
			emitline("\tPUSHQ\tAX\n");
			w -= 8;
		};
		return rest + tsz / 8;
	};
	if (nodeisslice(c, arg)) {
		emitline("\tPUSHQ\tCX\n");
		emitline("\tPUSHQ\tBX\n");
		emitline("\tPUSHQ\tAX\n");
		return rest + 3;
	};
	if (nodeisstr(c, arg)) {
		// str IS []u8: cgexpr left (AX=ptr, BX=len, CX=cap). Push
		// the triple, same as the slice arm above (#1/Phase 3).
		emitline("\tPUSHQ\tCX\n");
		emitline("\tPUSHQ\tBX\n");
		emitline("\tPUSHQ\tAX\n");
		return rest + 3;
	};
	// #21: CALL returning a tagged-union — the aistagged guard
	// above kept us out of the widening path. Push the tagged-
	// return ABI registers (AX=tag, DX=word0, CX=word1, R8=word2)
	// high → low so the left-to-right POPQ into argregs drains the
	// tag first. Mirrors cstage at cmd/w6c/cgen.c:4373-4387.
	let tcs: i32 = taggedcallslot(c, arg);
	if (tcs > 0) {
		if (tcs > 24) { emitline("\tPUSHQ\tR8\n"); };
		if (tcs > 16) { emitline("\tPUSHQ\tCX\n"); };
		if (tcs > 8)  { emitline("\tPUSHQ\tDX\n"); };
		emitline("\tPUSHQ\tAX\n");
		return rest + tcs / 8;
	};
	// #12: N_INDEX of a sum-typed slice element. cgindex above left
	// the tagged-CALL ABI in AX/DX/CX/R8; the bare PUSHQ AX below
	// would only carry the tag word and drop the payload. #60:
	// arg.type_ is the element tinfo (istaggedtype/slotsize read
	// .type_) — feed the N_INDEX node directly, dropping the
	// indexvaluetnode walk.
	if (arg.kind == nkind.N_INDEX) {
		if (istaggedtype(c, arg)) {
			let isz: i32 = slotsize(c, arg);
			if (isz > 24) { emitline("\tPUSHQ\tR8\n"); };
			if (isz > 16) { emitline("\tPUSHQ\tCX\n"); };
			if (isz > 8)  { emitline("\tPUSHQ\tDX\n"); };
			emitline("\tPUSHQ\tAX\n");
			return rest + isz / 8;
		};
	};
	emitline("\tPUSHQ\tAX\n");
	return rest + 1;
};

// taggedcallslot — if `n` is an N_CALL whose callee returns a tagged
// type, returns the slot size in bytes; else 0. Used by pushargsrev's
// aistagged guard and natural-push arm, and by cgcall's pop sizer, to
// route a tagged-return call result through the AX/DX/CX/R8 high→low
// push convention rather than the concrete-variant widening path
// (which drops DX/CX/R8). See task #21.
export fn taggedcallslot(c: *cgen, n: *node) i32 = {
	if (n == nil) { return 0; };
	if (n.kind != nkind.N_CALL) { return 0; };
	let callee: *node = n.lhs;
	if (callee == nil) { return 0; };
	if (callee.kind != nkind.N_IDENT) { return 0; };
	let rtyp: *node = fnretlookup(c, callee.str);
	if (!istaggedtype(c, rtyp)) { return 0; };
	return slotsize(c, rtyp);
};

fn nodeisslice(c: *cgen, n: *node) bool = {
	if (n == nil) { return false; };
	let k: nkind = n.kind;
	if (k == nkind.N_IDENT) {
		let nm: str = n.str;
		let lc: *local = localfindnode(c, nm);
		if (lc != nil) { return isslicetype(c, lc.tnode); };
		return false;
	};
	if (k == nkind.N_SLICE) { return true; };
	if (k == nkind.N_CAST) { return isslicetype(c, n.rhs); };
	// #24: N_CALL returning a slice — cgexpr leaves (AX=ptr,
	// BX=len, CX=cap); pushargsrev's slice arm pushes CX/BX/AX
	// and cgcall pops 3 words. Without this arm the natural-push
	// fallthrough emits one PUSHQ AX (loses .len/.cap) and the pop
	// side under-drains by 2 words, leaving R8/R9 unset for the
	// receiver. Mirrors nodeisstr's N_CALL arm just below.
	// N_DOT (cross-module callee, #34): route through fnretlookupmod
	// so a same-leaf caller-module fn with diverging return shape
	// doesn't shadow the explicit `mod.f()` qualifier — surfaced by
	// strings.slice returning `frombytes(utf8.slice(...))`
	// where strings.slice itself returns str.
	if (k == nkind.N_CALL) {
		let callee: *node = n.lhs;
		if (callee != nil) {
			if (callee.kind == nkind.N_IDENT) {
				let rtyp: *node = fnretlookupmod(c, callee.str, c.curmod);
				return isslicetype(c, rtyp);
			};
			if (callee.kind == nkind.N_DOT) {
				let cmod: str;
				cmod.ptr = nil; cmod.len = 0;
				if (callee.lhs != nil) {
					if (callee.lhs.kind == nkind.N_IDENT) {
						cmod = callee.lhs.str;
					};
				};
				let rtyp: *node = fnretlookupmod(c, callee.str, cmod);
				return isslicetype(c, rtyp);
			};
		};
		return false;
	};
	// N_DOT: read the checker-stamped n.type_. Struct field, nested
	// dot, value-struct hops, and pseudo-fields (.ptr/.len/.cap) all
	// resolve to the right tinfo via check.ww:1947-1978 (pseudo-field
	// + struct-field stamps). Cstage cgen.c:182-184 node_isslice =
	// type_isslice(n->type) — same shape. Collapsed per A.6.3h (#56).
	if (k == nkind.N_DOT) {
		return typeisslice(n.type_: *tinfo);
	};
	return false;
};

// nodeisstr — best-effort surface check: does this expression
// evaluate to a str value? Used to drive the call-arg push convention
// (str args take two slots: ptr + len).
//
// TODO(#11): every consumer of "is-str" here reconstructs the answer
// from raw N_kind because wwstage has no typed AST. Each new expression
// shape needs an explicit arm or it silently falls through to false,
// which downstream drops the second slot (BX/len) at the call site.
// A typed AST check (cstage reads n->type) would replace this whole
// function. Covered arms below: N_STRLIT, N_IDENT (local/let-typed),
// N_CALL (return type), N_INDEX (element type of [N]T / []T / *T base),
// N_DOT (reads checker-stamped n.type_ — #56 A.6.3h), N_CAST.
// Not covered (separate bugs / out of scope):
//   - N_UN(TK_STAR) of `*str` — cgun itself emits only `MOVQ (AX), AX`
//     and never loads .len into BX; fixing the recognizer alone won't
//     help. Tracked alongside the broader cgun-load-shape gap.
fn nodeisstr(c: *cgen, n: *node) bool = {
	if (n == nil) { return false; };
	let k: nkind = n.kind;
	if (k == nkind.N_STRLIT) { return true; };
	if (k == nkind.N_IDENT) {
		let nm: str = n.str;
		let lc: *local = localfindnode(c, nm);
		if (lc != nil) {
			// Use isstrtype so `!str` aliases (parserr = !str) and
			// `type foo = str;` chains resolve through. The bare
			// `streq("str", ...)` test missed them and dropped the
			// MOVQ BX,CX shuffle on returns of str-aliased locals.
			if (isstrtype(c, lc.tnode)) { return true; };
		};
		return false;
	};
	if (k == nkind.N_CALL) {
		let callee: *node = n.lhs;
		if (callee != nil) {
			if (callee.kind == nkind.N_IDENT) {
				let rtyp: *node = fnretlookupmod(c, callee.str, c.curmod);
				return isstrtype(c, rtyp);
			};
			// #34: cross-module N_DOT — route through fnretlookupmod
			// so a same-leaf caller-module fn (different return shape)
			// doesn't shadow the explicit qualifier.
			if (callee.kind == nkind.N_DOT) {
				let cmod: str;
				cmod.ptr = nil; cmod.len = 0;
				if (callee.lhs != nil) {
					if (callee.lhs.kind == nkind.N_IDENT) {
						cmod = callee.lhs.str;
					};
				};
				let rtyp: *node = fnretlookupmod(c, callee.str, cmod);
				return isstrtype(c, rtyp);
			};
		};
		return false;
	};
	// N_INDEX: `arr[i]` whose base is an indexable type carrying a
	// str element. cgindex correctly loads (AX=ptr, BX=len) for a
	// 16B element; without this arm pushargsrev only pushes AX and
	// the call-arg pop reads .len from stack residue. Mirror of
	// cstage's node_isstr → type_isstr(n->type), where n->type is
	// the resolved element type after check.
	if (k == nkind.N_INDEX) {
		let base: *node = n.lhs;
		if (base != nil) {
			if (base.kind == nkind.N_IDENT) {
				let bt: *node = nil;
				let lc: *local = localfindnode(c, base.str);
				if (lc != nil) { bt = lc.tnode; }
				else { bt = letvartnode(c, base.str); };
				if (bt != nil) {
					let elem: *node = nil;
					let bk: nkind = bt.kind;
					if (bk == nkind.N_TARRAY) { elem = bt.lhs; };
					if (bk == nkind.N_TSLICE) { elem = bt.lhs; };
					if (bk == nkind.N_TPTR)   { elem = bt.lhs; };
					if (elem != nil) {
						return isstrtype(c, elem);
					};
				};
			};
			// N_INDEX through a struct field: e.g. cmd.argsptr[i]
			// where argsptr: *str. cgindex correctly loads the
			// (ptr, len) pair off the stamped element size; without
			// this arm pushargsrev would only push AX and lose .len.
			if (base.kind == nkind.N_DOT) {
				let fld: str = base.str;
				if (streq(fld, "ptr")) { return false; };
				if (streq(fld, "len")) { return false; };
				if (streq(fld, "cap")) { return false; };
				let inner: *node = base.lhs;
				if (inner != nil) {
					if (inner.kind == nkind.N_IDENT) {
						let lc: *local = localfindnode(c, inner.str);
						if (lc != nil) {
							let tn: *node = lc.tnode;
							let sname: str;
							sname.ptr = nil; sname.len = 0;
							if (tn != nil) {
								if (tn.kind == nkind.N_TNAME) { sname = tn.str; };
								if (tn.kind == nkind.N_TPTR) {
									let pinner: *node = tn.lhs;
									if (pinner != nil) {
										if (pinner.kind == nkind.N_TNAME) {
											sname = pinner.str;
										};
									};
								};
							};
							if (sname.len > 0) {
								let si: *structinfo = structlookup(c, sname);
								if (si != nil) {
									let fi: *fieldinfo = si.fields;
									for (fi != nil) {
										if (streq(fi.fname, fld)) {
											let ft: *node = fi.tnode;
											if (ft != nil) {
												let elem: *node = nil;
												let fk: nkind = ft.kind;
												if (fk == nkind.N_TPTR)   { elem = ft.lhs; };
												if (fk == nkind.N_TSLICE) { elem = ft.lhs; };
												if (fk == nkind.N_TARRAY) { elem = ft.lhs; };
												if (elem != nil) {
													return isstrtype(c, elem);
												};
											};
										};
										fi = fi.finext;
									};
								};
							};
						};
					};
				};
			};
		};
		return false;
	};
	// N_DOT: read the checker-stamped n.type_. Struct field, nested
	// dot, value-struct hops, and pseudo-fields (.ptr/.len/.cap) all
	// resolve to the right tinfo via check.ww:1947-1978. Cstage
	// cgen.c:168-170 node_isstr = type_isstr(n->type) — same shape.
	// Collapsed per A.6.3h (#56).
	if (k == nkind.N_DOT) {
		return typeisstr(n.type_: *tinfo);
	};
	if (k == nkind.N_CAST) {
		return isstrtype(c, n.rhs);
	};
	return false;
};

// typeis8byteprimitive — does this type take exactly one 8-byte
// slot rather than a wider aggregate? One-liner via typeis8byteprim
// (cstage N_LET sz==8 ladder SSoT). t.type_ is stamped at check.ww
// L426-436 for every type-AST kind callers reach. Collapsed onto the
// tinfo helper per A.6.3b (#46).
fn typeis8byteprimitive(c: *cgen, t: *node) bool = {
	if (t == nil) { return false; };
	return typeis8byteprim(t.type_: *tinfo);
};

// elemissignedc — given an indexable type-AST (`*T`, `[]T`, `[N]T`),
// is its element a signed narrow primitive? Used by cgindex to pick
// MOVSXD vs MOVL at esz=4 (and MOVSBQ/MOVSWQ at esz=1/2). Mirrors
// cstage's `signed_elem` (cmd/w6c/cgen.c idx_eff path). Reads through
// the stamped tinfo so alias/enum recursion lives in lib/ww/typ.ww.
fn elemissignedc(c: *cgen, t: *node) bool = {
	if (t == nil) { return false; };
	let ti: *tinfo = t.type_: *tinfo;
	if (ti == nil) { return false; };
	// #65 Phase-N step-2 cleanup: peel TY_NAMED before the .sub read.
	// #64 now flows per-decl NAMED wrappers, so a NAMED-of-(`*T`/`[]T`/
	// `[N]T`) reaching here would read NAMED.sub (nil) instead of the
	// element. Mirrors cstage idx_eff's type_unwrap (cmd/w6c/cgen.c:790)
	// before the eff->sub read (:3518-3520). Byte-id-neutral: every
	// aliased indexable in-tree has a u8 element (typeissigned=false
	// either way). Transitive peel matches the #63 idiom.
	for (ti != nil && ti.kind == tykind.TY_NAMED) { ti = ti.under; };
	if (ti == nil) { return false; };
	return typeissigned(ti.sub);
};

// elemisfloatc — given an indexable type-AST (`*T`, `[]T`, `[N]T`), is
// its element an f32/f64? Used by cgindex to route the element load to
// MOVSS/MOVSD into X0 instead of the integer loadopsz into AX (#119 —
// the array-element twin of the scalar-float global load at cgen.c:
// 2014). Reads through the stamped tinfo, peeling TY_NAMED before the
// .sub read exactly as elemissignedc does (#64/#65). Float-ness comes
// from the SAME tinfo the esz already reads — never a fresh node-stamp
// (the unstamped-base trap that broke the exprfloatkind collapse, #121).
fn elemisfloatc(c: *cgen, t: *node) bool = {
	if (t == nil) { return false; };
	let ti: *tinfo = t.type_: *tinfo;
	if (ti == nil) { return false; };
	for (ti != nil && ti.kind == tykind.TY_NAMED) { ti = ti.under; };
	if (ti == nil) { return false; };
	return typeisfloat(ti.sub);
};

// elemisf32c — narrower elemisfloatc: true only when the element is f32,
// so cgindex picks MOVSS over MOVSD at the #119 element load.
fn elemisf32c(c: *cgen, t: *node) bool = {
	if (t == nil) { return false; };
	let ti: *tinfo = t.type_: *tinfo;
	if (ti == nil) { return false; };
	for (ti != nil && ti.kind == tykind.TY_NAMED) { ti = ti.under; };
	if (ti == nil) { return false; };
	return typeisf32(ti.sub);
};

// elemisarrayc — given an indexable type-AST (`*T` / `[]T` / `[N]T`), is
// its element itself an array (`[N][M]T` → element `[M]T`)? cgindex then
// leaves the sub-array's ADDRESS in the result reg rather than
// dereferencing — a nested index adds its offset and only the final
// scalar element dereferences (#156, sister of #135 N_DOT-base-on-
// array-field). Node-based with the `*[N]T` drill-through, mirroring
// elemsizeof (:920-948) so elem-is-array aligns with the esz this same
// tnode feeds. cstage twin: idx_eff(bt)->sub unwrapped == TY_ARRAY
// (cmd/w6c/cgen.c). `c` kept for signature symmetry with elemisfloatc.
fn elemisarrayc(c: *cgen, t: *node) bool = {
	if (t == nil) { return false; };
	let k: nkind = t.kind;
	let elem: *node = nil;
	if (k == nkind.N_TPTR)   { elem = t.lhs; };
	if (k == nkind.N_TSLICE) { elem = t.lhs; };
	if (k == nkind.N_TARRAY) { elem = t.lhs; };
	if (elem == nil) { return false; };
	if (k == nkind.N_TPTR) {
		if (elem.kind == nkind.N_TARRAY) {
			if (elem.lhs != nil) { elem = elem.lhs; };
		};
	};
	return elem.kind == nkind.N_TARRAY;
};

// tinfoisarray — TY_ARRAY (NAMED-aware), the tinfo-keyed companion to
// elemisarrayc for cgindex's N_DOT/N_INDEX base branches, where the
// element type comes from n.type_ (stamped tinfo) not a tnode. Same
// role as typeisslice/typeisstr in lib/ww/typ.ww; kept cgen-local to
// avoid widening the frontend surface for one #156 read-half check.
fn tinfoisarray(t: *tinfo) bool = {
	let u: *tinfo = t;
	for (u != nil && u.kind == tykind.TY_NAMED) { u = u.under; };
	if (u == nil) { return false; };
	return u.kind == tykind.TY_ARRAY;
};

// fieldissignedc — does this field/element type-AST need sign-
// extension on a sub-word load? One-liner via typeissigned (cstage
// cgen.c:240 `fld_issigned` SSoT). t.type_ is stamped at check.ww
// L426-436 for every type-AST kind we see here (TNAME / TPTR /
// TBANG / TENUM / TARRAY / TSLICE — see resolvewalk).
fn fieldissignedc(c: *cgen, t: *node) bool = {
	if (t == nil) { return false; };
	return typeissigned(t.type_: *tinfo);
};

// fieldloadop — pick the load instruction for a non-str struct
// field by its declared size + signedness. Mirrors cstage's
// fldloadop: MOVZBQ/MOVSBQ for 1B, MOVZWQ/MOVSWQ for 2B,
// MOVL/MOVSXD for 4B, MOVQ for 8B. f might be nil for fields
// outside our struct registry.
fn fieldloadop(c: *cgen, f: *fieldinfo) str = {
	if (f == nil) { return "MOVQ"; };
	let sz: i32 = f.fsz;
	let sigd: bool = fieldissignedc(c, f.tnode);
	if (sz == 1) { if (sigd) { return "MOVSBQ"; }; return "MOVZBQ"; };
	if (sz == 2) { if (sigd) { return "MOVSWQ"; }; return "MOVZWQ"; };
	if (sz == 4) { if (sigd) { return "MOVSXD"; }; return "MOVL"; };
	return "MOVQ";
};

// fieldstoreop — pick the store instruction for a non-str struct
// field by its declared size. MOVB for 1, MOVW for 2, MOVL for 4,
// MOVQ for 8. c kept in the signature for symmetry with fieldloadop.
fn fieldstoreop(c: *cgen, f: *fieldinfo) str = {
	if (f == nil) { return "MOVQ"; };
	let sz: i32 = f.fsz;
	if (sz == 1) { return "MOVB"; };
	if (sz == 2) { return "MOVW"; };
	if (sz == 4) { return "MOVL"; };
	return "MOVQ";
};

// tnodeloadop / tnodestoreop — same dispatch as fieldloadop /
// fieldstoreop but keyed on a raw type-AST node (tuple element type,
// pointer-target, slice-element, etc.) rather than a struct fieldinfo.
// Used at the index / tuple / pointer-deref sites where there's no
// fieldinfo entry but the type-node + size are both known.
fn tnodeloadop(c: *cgen, t: *node, sz: i32) str = {
	let sigd: bool = fieldissignedc(c, t);
	if (sz == 1) { if (sigd) { return "MOVSBQ"; }; return "MOVZBQ"; };
	if (sz == 2) { if (sigd) { return "MOVSWQ"; }; return "MOVZWQ"; };
	if (sz == 4) { if (sigd) { return "MOVSXD"; }; return "MOVL"; };
	return "MOVQ";
};

fn tnodestoreop(c: *cgen, t: *node, sz: i32) str = {
	if (sz == 1) { return "MOVB"; };
	if (sz == 2) { return "MOVW"; };
	if (sz == 4) { return "MOVL"; };
	return "MOVQ";
};

// loadopsz — load op when the (size, signedness) pair has already
// been resolved upstream and the type-node isn't carried through.
// cgindex precomputes `signed_elem` via elemissignedc; cgforrange
// precomputes `bind_signed[b]` via paramissigned. Same dispatch as
// tnodeloadop's tail; only the keying differs.
fn loadopsz(sigd: bool, sz: i32) str = {
	if (sz == 1) { if (sigd) { return "MOVSBQ"; }; return "MOVZBQ"; };
	if (sz == 2) { if (sigd) { return "MOVSWQ"; }; return "MOVZWQ"; };
	if (sz == 4) { if (sigd) { return "MOVSXD"; }; return "MOVL"; };
	return "MOVQ";
};

// localloadop — read instruction for a scalar local/let load. Same
// dispatch as fieldloadop, but keyed on the value's own tnode.type_.
// Lets the caller emit MOVSXD/MOVSWQ/MOVSBQ on a signed-narrow slot
// instead of a raw MOVQ, so a slot that was last written by a narrow
// deref-store (`*p: *i32 = v` lowers to MOVL, only 4B) reads back as
// a properly-sign-extended i64. The natural N_ASSIGN / N_LET paths
// store the rhs as a sign-extended 8B word, so MOVQ accidentally
// works; deref-stores are the only path that touches fewer bytes
// than MOVQ reads. Mirror of cstage's localloadop in cmd/w6c/cgen.c
// — tinfo.size carries the same numeric width cstage's `t->size`
// reports, with TBANG / TENUM / TNAME-alias chains pre-folded by
// tinfofornode (check.ww:1102-1153 TNAME, 1154-1161 TBANG,
// 1196-1208 TENUM).
export fn localloadop(c: *cgen, tnode: *node) str = {
	if (tnode == nil) { return "MOVQ"; };
	let ti: *tinfo = tnode.type_: *tinfo;
	if (ti == nil) { return "MOVQ"; };
	let sz: i32 = ti.size: i32;
	if (sz != 1) { if (sz != 2) { if (sz != 4) { return "MOVQ"; }; }; };
	let sigd: bool = typeissigned(ti);
	return loadopsz(sigd, sz);
};

// elemsizeof — given the type node of an indexable (`*T`, `[]T`,
// `[N]T`, `str`), return the byte size of one element (1 for u8/i8/
// bool/str-byte, 8 otherwise — same shape as C cgen's esz fallback).
// For aliased element types (e.g. `[N]formattable`), callers that
// need the resolved slot size should use elemsizeofc(c, t) which
// follows aliases via slotsize.
fn elemsizeof(t: *node) i32 = {
	if (t == nil) { return 1; };
	let k: nkind = t.kind;
	let elem: *node = nil;
	if (k == nkind.N_TPTR)   { elem = t.lhs; };
	if (k == nkind.N_TSLICE) { elem = t.lhs; };
	if (k == nkind.N_TARRAY) { elem = t.lhs; };
	if (k == nkind.N_TNAME) {
		let nm: str = t.str;
		// str's element is u8 (F1: tystr.sub = tyu8), named directly
		// rather than read off str.sub because elemsizeof gets a raw
		// N_TNAME at the ident-base index path with no checker-stamped
		// tinfo — str.sub lives on .type_.sub, unstamped at this site
		// (cf. cgforrange's `if (sti != nil)` guard). This IS the
		// str.sub-equivalent; the structural collapse onto str.sub is
		// blocked on tinfo-stamping here, not intent (#24). cstage twin
		// reads eff->sub->size (cmd/w6c/cgen.c N_INDEX).
		if (streq(nm, "str")) { return primtypesize("u8"): i32; };
		// Indexing a primitive name (rare): element size = the prim.
		let ps: i32 = primsize(nm);
		if (ps > 0) { return ps; };
		return 1;
	};
	if (elem == nil) { return 1; };
	// `*[N]T`: drill through the pointer into the array's element so
	// indexing scales by T's width, not the whole-array byte size.
	// FOOTGUN (#156): this drill ALSO fires for a bare 2D `[N][M]T`
	// (elem = the inner `[M]T`), so elemsizeof of a 2D array bottoms
	// out at the SCALAR T size, NOT the `[M]T` sub-array stride. 2D
	// double-index (cgindex) needs the sub-array stride — call
	// elemsizeofc, the 2D-correct entry, which recovers slotsize([M]T)
	// when elemsizeof returns 8. Never call elemsizeof for a 2D stride.
	if (elem.kind == nkind.N_TARRAY) {
		if (elem.lhs != nil) { elem = elem.lhs; };
	};
	// `*[]T`: stride is the slice header (24B). Hare-faithful — a
	// pointer-to-slice is a 1D array of slices, not of T. Mirrors the
	// cstage check.c default `*U → U` path for U=[]T (slice element).
	if (elem.kind == nkind.N_TSLICE) { return tyslicesize(): i32; };
	if (elem.kind == nkind.N_TNAME) {
		let nm: str = elem.str;
		// str element is 16B (ptr+len). primsize returns 0 for it.
		if (streq(nm, "str")) { return primtypesize("str"): i32; };
		let ps: i32 = primsize(nm);
		if (ps > 0) { return ps; };
	};
	return 8;
};

// elemsizeofc — like elemsizeof but resolves aliased element types
// (struct / tagged / `type foo = bar;`) via slotsize. Used where
// cgindex / cgassign need a correct stride for `[N]Alias` arrays
// whose Alias resolves to a tagged union (e.g. `[N]formattable`).
fn elemsizeofc(c: *cgen, t: *node) i32 = {
	if (t == nil) { return 1; };
	let direct: i32 = elemsizeof(t);
	if (direct != 8) { return direct; };
	let k: nkind = t.kind;
	let elem: *node = nil;
	if (k == nkind.N_TPTR)   { elem = t.lhs; };
	if (k == nkind.N_TSLICE) { elem = t.lhs; };
	if (k == nkind.N_TARRAY) { elem = t.lhs; };
	if (elem == nil) { return direct; };
	if (elem.kind == nkind.N_TNAME) {
		let ps: i32 = primsize(elem.str);
		if (ps > 0) { return ps; };
	};
	return slotsize(c, elem);
};

// nodeisunsigned — best-effort cgen-time inference from the AST. We
// walk surface nodes (N_DOT now reads n.type_ — #55 A.6.3g):
//   nkind.N_INTLIT      — never marked unsigned (no tsuffix plumbing yet)
//   nkind.N_IDENT       — look up the local's declared type
//   nkind.N_DOT         — read the checker-stamped n.type_ (#55 A.6.3g)
//   nkind.N_BIN / nkind.N_UN  — recurse: unsigned if either operand is unsigned
//   nkind.N_CAST        — use the cast target type
//
// Conservative: if we can't tell, return false (signed). The cost of
// being wrong here is byte-different asm vs C, not bad runtime.
fn nodeisunsigned(c: *cgen, n: *node) bool = {
	if (n == nil) { return false; };
	let k: nkind = n.kind;
	if (k == nkind.N_IDENT) {
		let nm: str = n.str;
		let lc: *local = localfindnode(c, nm);
		if (lc != nil) {
			if (lc.tnode == nil) { return false; };
			return typeisunsigned(lc.tnode.type_: *tinfo);
		};
		return false;
	};
	if (k == nkind.N_DOT) {
		return typeisunsigned(n.type_: *tinfo);
	};
	if (k == nkind.N_CAST) {
		if (n.rhs == nil) { return false; };
		return typeisunsigned(n.rhs.type_: *tinfo);
	};
	if (k == nkind.N_BIN) {
		if (nodeisunsigned(c, n.lhs)) { return true; };
		return nodeisunsigned(c, n.rhs);
	};
	if (k == nkind.N_UN) { return nodeisunsigned(c, n.lhs); };
	// nkind.N_INDEX: `p[i]` is unsigned iff its element type is
	// unsigned. Read the checker-stamped result type directly,
	// mirroring the N_DOT arm above and cstage cgen.c:2541
	// (`type_isunsigned(n->lhs->type)` on operand's stamped tinfo).
	// Replaces the prior structural base-walk that only fired for
	// N_IDENT base — fell through to `return false` for N_DOT base
	// (e.g. `d.digits[nd]` where d is a *struct), making
	// `d.digits[nd] >= 5u8` pick signed JGE instead of unsigned JAE.
	// Embodies the #121 principle (collapse structural onto stamp).
	// #134.
	if (k == nkind.N_INDEX) {
		return typeisunsigned(n.type_: *tinfo);
	};
	// nkind.N_CALL: a call returning an unsigned type (e.g. `fn f() u64`)
	// is unsigned. Read the checker-stamped result type directly — the
	// N_CALL twin of the #134 N_INDEX arm above. cstage reads the same
	// stamp via `type_isunsigned(n->lhs->type)`, stamped at check.c N_CALL
	// `n->type = u->ret`; without this arm the wwstage fell through to
	// `return false`, picking signed IDIV/SAR over unsigned DIV/SHR on a
	// call-result div/mod/shift operand. #168.
	if (k == nkind.N_CALL) {
		return typeisunsigned(n.type_: *tinfo);
	};
	return false;
};

// nodeprimwidth — primitive byte width of an expression, or 0 if not
// statically determinable. Mirrors nodeisunsigned's structural walk.
// Used by cgun TK_TILDE to clamp narrow unsigned ~ results to type
// width (NOTQ inverts the full 64-bit register).
fn nodeprimwidth(c: *cgen, n: *node) i32 = {
	if (n == nil) { return 0; };
	let k: nkind = n.kind;
	if (k == nkind.N_IDENT) {
		let lc: *local = localfindnode(c, n.str);
		if (lc != nil) {
			let tn: *node = lc.tnode;
			if (tn != nil) {
				if (tn.kind == nkind.N_TNAME) { return primsize(tn.str); };
			};
		};
		return 0;
	};
	if (k == nkind.N_CAST) {
		let tn: *node = n.rhs;
		if (tn != nil) {
			if (tn.kind == nkind.N_TNAME) { return primsize(tn.str); };
		};
		return 0;
	};
	if (k == nkind.N_UN) { return nodeprimwidth(c, n.lhs); };
	return 0;
};

// ---- type-driven slot sizing ----------------------------------------

// structnaturalsize — type-natural size of `si`, i.e. max(foff +
// fsz) across declared fields, UNROUNDED. This is the memory-copy
// extent: cstage copies exactly these bytes for the >24B sret
// write-through (cgen.c:8150 `int sz; for fields end=foff+fsz`) and
// struct-to-struct moves, so a trailing narrow field (bool@32 in a
// 33B struct padded to 40) keeps its MOVB tail rather than widening
// to a slot-overrunning MOVQ. #33 fixed cstage to use this; ww
// mirrors it. The ≤24B register RECV/RETURN ABI wants a DIFFERENT
// number — see structabisize.
//
// NOTE: si.totsize is yet a THIRD metric — the slot-padded size
// (rounded up to 8 for stack-slot use; see registerstruct's tail
// `if ((off & 7) != 0) ...`). Frame allocation and [N]foo stride
// want that slot number.
fn structnaturalsize(si: *structinfo) i32 = {
	if (si == nil) { return 0; };
	let n: i32 = 0;
	let fi: *fieldinfo = si.fields;
	for (fi != nil) {
		let end: i32 = fi.foff + fi.fsz;
		if (end > n) { n = end; };
		fi = fi.finext;
	};
	return n;
};

// structabisize — the ≤24B register-return ABI size of `si`: the
// natural extent rounded up to the struct's maxalign. SSoT-equal to
// cstage's `lu->size` (check.c:760 `(off+maxalign-1)&~(maxalign-1)`).
// Distinct from structnaturalsize because the register RECV/RETURN
// ABI packs the value into AX/DX/CX at 8-byte granularity: cstage
// writes/reads the tail at maxalign width (cgen.c:7720 `sz=lu->size`,
// :8230 `sz=rt->size`), so a maxalign==8 struct with a sub-8 tail
// (struct{i64,i32}, natural 12) round-trips as MOVQ+MOVQ (16), not
// MOVQ+MOVL (12). Used ONLY at those register-ABI sites; memory
// copies (sret >24B, struct ident-copy) and field-offset math stay
// on structnaturalsize. #169.
//
// maxalign comes from each field's TRUE alignment (tinfo.align), not
// the slot-padded fsz: a [N]u8 / sub-struct field has slot ≥8 but
// align 1, so an fsz ladder would over-round. Mirrors cstage's
// maxalign = max(ft->align) (check.c:708).
fn structabisize(si: *structinfo) i32 = {
	if (si == nil) { return 0; };
	let n: i32 = 0;
	let maxaln: i32 = 1;
	let fi: *fieldinfo = si.fields;
	for (fi != nil) {
		let end: i32 = fi.foff + fi.fsz;
		if (end > n) { n = end; };
		if (fi.tnode != nil) {
			let ti: *tinfo = fi.tnode.type_: *tinfo;
			for (ti != nil && ti.kind == tykind.TY_NAMED) {
				ti = ti.under;
			};
			if (ti != nil) {
				let aln: i32 = ti.align: i32;
				if (aln > maxaln) { maxaln = aln; };
			};
		};
		fi = fi.finext;
	};
	return (n + maxaln - 1) & ~(maxaln - 1);
};

// sretretsize — if `t` ultimately denotes a plain TY_STRUCT > 24B,
// return its natural size; else 0. Tagged unions, tuples, str,
// slices, scalars route through their existing register-return ABIs
// (AX/DX/CX/[R8]) regardless of size. Task #23 mirrors cstage's
// cg_sret_retsize predicate. Resolves N_TNAME → struct via structlookup
// and unwraps one leading N_TBANG so `type box = !big;` still
// triggers sret on the underlying big.
//
// Chain-of-aliases (#22): `type a = struct{...}; type b = a;` registers
// `b → a` in c.aliases (target node = N_TNAME "a"), not `b → struct`.
// When structlookup(c, "b") misses, fall through to aliaslookup and
// recurse on the alias target — mirrors slotsize's N_TNAME arm
// (cgenutil.ww:1955) and the cstage while-loop in cg_sret_retsize.

// structlookupchain — resolve TNAME `tn` to its registered struct,
// chasing alias-of-alias (#22). Returns nil if the chain doesn't
// bottom out at a struct. Mirrors cstage's transitive
// `while (t->kind == TY_NAMED) t = t->under` peel; consumed by
// cgdot / cgassign at every "field-walk on a struct-typed local"
// site so a transitively-aliased struct name resolves to its
// fieldinfo list regardless of chain depth.
export fn structlookupchain(c: *cgen, tn: *node) *structinfo = {
	if (tn == nil) { return nil; };
	if (tn.kind != nkind.N_TNAME) { return nil; };
	let si: *structinfo = structlookup(c, tn.str);
	if (si != nil) { return si; };
	let cur: *node = tn;
	for (cur != nil && cur.kind == nkind.N_TNAME && si == nil) {
		let aliased: *node = aliaslookup(c, cur.str);
		if (aliased == nil) { cur = nil; }
		else {
			if (aliased.kind == nkind.N_TNAME) {
				si = structlookup(c, aliased.str);
				cur = aliased;
			} else { cur = nil; };
		};
	};
	return si;
};

export fn sretretsize(c: *cgen, t: *node) i32 = {
	if (t == nil) { return 0; };
	let r: *node = t;
	if (r.kind == nkind.N_TBANG) {
		r = r.lhs;
		if (r == nil) { return 0; };
	};
	if (r.kind != nkind.N_TNAME) { return 0; };
	// Primitives / aliased-to-primitives are never sret.
	if (primsize(r.str) > 0) { return 0; };
	if (streq(r.str, "str")) { return 0; };
	let si: *structinfo = structlookup(c, r.str);
	if (si == nil) {
		if (c != nil) {
			let aliased: *node = aliaslookup(c, r.str);
			if (aliased != nil) {
				return sretretsize(c, aliased);
			};
		};
		return 0;
	};
	let n: i32 = structnaturalsize(si);
	if (n <= 24) { return 0; };
	return n;
};

// callsretsize — if N_CALL `n`'s callee returns a plain TY_STRUCT
// > 24B, return its natural size; else 0. Wraps sretretsize over the
// callee's resolved return type, used by cglet / cgassign receive
// sites and cgcall to detect sret at the receive / emit boundaries.
export fn callsretsize(c: *cgen, n: *node) i32 = {
	if (n == nil) { return 0; };
	if (n.kind != nkind.N_CALL) { return 0; };
	let callee: *node = n.lhs;
	if (callee == nil) { return 0; };
	let cn: str;
	cn.ptr = nil; cn.len = 0;
	let cmod: str;
	cmod.ptr = nil; cmod.len = 0;
	if (callee.kind == nkind.N_IDENT) {
		cn = callee.str;
		cmod = c.curmod;
	};
	if (callee.kind == nkind.N_DOT) {
		cn = callee.str;
		if (callee.lhs != nil) {
			if (callee.lhs.kind == nkind.N_IDENT) {
				cmod = callee.lhs.str;
			};
		};
	};
	if (cn.len == 0) { return 0; };
	let rtyp: *node = fnretlookupmod(c, cn, cmod);
	return sretretsize(c, rtyp);
};

fn structlookup(c: *cgen, name: str) *structinfo = {
	// Same-module first, then any. Trio-leaf graduation mirroring
	// aliaslookup (#27), fnret/fnparamslookupmod (#28/#31), and
	// enumlookup (#4a): without the prefer pass a bare-leaf struct
	// name in module M can collapse onto another module's same-leaf
	// struct prepended earlier in c.structs, silently picking the
	// wrong totsize / field offsets.
	let s: *structinfo = c.structs;
	for (s != nil) {
		if (streq(s.sname, name)) {
			if (streq(s.smod, c.curmod)) { return s; };
		};
		s = s.sinext;
	};
	s = c.structs;
	for (s != nil) {
		let sn: str = s.sname;
		if (streq(sn, name)) { return s; };
		s = s.sinext;
	};
	// Module-qualified form embedded in name (`pkg.S`): scope the
	// leaf to its originating module. The `smod == pkg` guard
	// prevents same-leaf structs in two modules from collapsing.
	let i: i32 = name.len - 1;
	for (i >= 0) {
		if (name[i] == 46u8) {	// '.'
			let pkg: str;
			pkg.ptr = name.ptr;
			pkg.len = i;
			let leaf: str;
			leaf.ptr = name.ptr + ((i + 1): u64);
			leaf.len = name.len - (i + 1);
			let b: *structinfo = c.structs;
			for (b != nil) {
				if (streq(b.sname, leaf)) {
					if (streq(b.smod, pkg)) {
						return b;
					};
				};
				b = b.sinext;
			};
			return nil;
		};
		i -= 1;
	};
	return nil;
};

// #223: same-module-ONLY struct lookup. structlookup's any-module
// fallback returns a foreign same-leaf struct; the cgdot alias-peel
// needs to break ONLY on a struct that THIS module defines (a genuine
// struct-value receiver), not on a foreign struct that merely shares a
// leaf with a same-module alias (io.stream alias vs memio.stream
// struct). Returns the struct only when it lives in c.curmod.
fn structsamemod(c: *cgen, name: str) *structinfo = {
	let s: *structinfo = c.structs;
	for (s != nil) {
		if (streq(s.sname, name)) {
			if (streq(s.smod, c.curmod)) { return s; };
		};
		s = s.sinext;
	};
	return nil;
};

// primsize — size in bytes of a primitive type name (or 0 if not
// recognised as a primitive — the caller falls back to other paths).
// fldnumidx — parse a tuple field name like "0" / "1" / "12" into an
// index, or -1 if not all-digits. Used by cgdot to dispatch
// `t.0` / `t.1` against an nkind.N_TTUPLE local without pulling in strconv.
fn fldnumidx(s: str) i32 = {
	if (s.len == 0) { return -1; };
	let r: i32 = 0;
	let i: i32 = 0;
	for (i < s.len) {
		let b: u8 = s[i];
		if (b < 48u8) { return -1; };
		if (b > 57u8) { return -1; };
		r = r * 10 + ((b - 48u8): i32);
		i += 1;
	};
	return r;
};

fn primsize(name: str) i32 = {
	if (streq(name, "u8"))   { return 1; };
	if (streq(name, "i8"))   { return 1; };
	if (streq(name, "bool")) { return 1; };
	if (streq(name, "u16"))  { return 2; };
	if (streq(name, "i16"))  { return 2; };
	if (streq(name, "u32"))  { return 4; };
	if (streq(name, "i32"))  { return 4; };
	if (streq(name, "f32"))  { return 4; };
	if (streq(name, "u64"))  { return 8; };
	if (streq(name, "i64"))  { return 8; };
	if (streq(name, "uint")) { return 8; };
	if (streq(name, "int"))  { return 8; };
	if (streq(name, "uintptr")) { return 8; };
	if (streq(name, "size")) { return 8; };
	if (streq(name, "f64"))  { return 8; };
	if (streq(name, "rune")) { return 4; };
	if (streq(name, "void")) { return 0; };
	return 0;
};

// typenodeprimresolved — walk N_TBANG / N_TENUM / N_TNAME alias
// chains to the underlying primitive, returning its byte size and
// signedness. Sets *sz_out = 0 when the type doesn't reduce to a
// width-known primitive (composite, unresolved name, default-storage
// enum, etc.). Mirrors cstage's `type_isint(t) ? t->size : 0` /
// `type_isunsigned` recursion through TY_NAMED and TY_ENUM. Used by
// cgcast's identity-width identity-sign clamp-skip predicate (#33).
export fn typenodeprimresolved(c: *cgen, t: *node,
    sz_out: *i32, unsigned_out: *bool) void = {
	*sz_out = 0;
	*unsigned_out = false;
	let cur: *node = t;
	for (cur != nil) {
		let k: nkind = cur.kind;
		if (k == nkind.N_TBANG) { cur = cur.lhs; }
		else { if (k == nkind.N_TENUM) { cur = cur.lhs; }
		else { if (k == nkind.N_TNAME) {
			let nm: str = cur.str;
			// bool is excluded from the int-prim contract: cstage's
			// `type_isint(TY_BOOL)` is false, so its identity check
			// leaves src_w=0 on a bool source. Match that here so a
			// `let y: i8 = b: i8;` (bool b) doesn't fire identity in
			// wwstage and skip the MOVSBQ that cstage emits. Other
			// call sites (slot sizing, etc.) still want
			// primsize("bool")=1, so the exclusion stays local. The
			// dedicated `is_bool` path in cgcast owns bool→bool's
			// ANDQ $255 on both stages.
			if (streq(nm, "bool")) { return; };
			let ps: i32 = primsize(nm);
			if (ps > 0) {
				*sz_out = ps;
				*unsigned_out = typeisunsigned(cur.type_: *tinfo);
				return;
			};
			let al: *node = aliaslookup(c, nm);
			if (al == nil) { return; };
			cur = al;
		}
		else { return; }; }; };
	};
};

// exprprimresolved — best-effort static (primsize, signedness) for an
// expression. Used by cgcast (#33) to derive the source-side primitive
// width and signedness so the identity-width identity-sign clamp-skip
// predicate fires. Sets *sz_out = 0 when the type can't be derived
// (untyped literal, call result with no return-type lookup, etc.);
// caller treats sz=0 as "not identity", which conservatively keeps
// the clamp. Mirror of cstage's `n->lhs->type` lookup with the same
// TY_NAMED / TY_ENUM recursion through type_isint / type_isunsigned.
export fn exprprimresolved(c: *cgen, n: *node,
    sz_out: *i32, unsigned_out: *bool) void = {
	*sz_out = 0;
	*unsigned_out = false;
	if (n == nil) { return; };
	let k: nkind = n.kind;
	if (k == nkind.N_INTLIT) {
		// Typed-int literal: `7u32` has tsuffix = "u32". Mirrors
		// cstage's `cexpr` which assigns `lookup_builtin(tsuffix)`
		// as the node's type — without this, wwstage misses the
		// suffix and emits a defensive clamp where cstage skips,
		// breaking byte-id on rows like `let y: mymode = 7u32:
		// mymode;` (mymode = enum u32).
		let s: str = n.tsuffix;
		if (s.len > 0) {
			let ps: i32 = primsize(s);
			if (ps > 0) {
				*sz_out = ps;
				*unsigned_out = typeisunsigned(n.type_: *tinfo);
			};
		};
		return;
	};
	if (k == nkind.N_IDENT) {
		let lc: *local = localfindnode(c, n.str);
		if (lc != nil) {
			typenodeprimresolved(c, lc.tnode,
			    sz_out, unsigned_out);
		};
		return;
	};
	if (k == nkind.N_CAST) {
		typenodeprimresolved(c, n.rhs, sz_out, unsigned_out);
		return;
	};
	if (k == nkind.N_UN)   {
		exprprimresolved(c, n.lhs, sz_out, unsigned_out);
		return;
	};
	if (k == nkind.N_DOT)  {
		// #59: read the checker-stamped tinfo instead of re-deriving the
		// field type via dotfieldtnode's structinfo walk. Mirrors cstage
		// castsrcprim N_DOT (cmd/w6c/cgen.c:323-344): the base must
		// resolve to a struct (or ptr-to-struct) before the field type
		// counts. That guard excludes pseudo-fields .len/.cap/.ptr (the
		// checker stamps them i32/*T at check.ww:1987-2004) and tuple
		// positionals, keeping them at sz=0 — asymmetry there breaks 995
		// byte-id (cgen.c:286-290). The field's own width/sign is the
		// N_DOT's stamped type_ (check.ww:2012). One TY_NAMED peel, then
		// typeisint ? size : 0; bool falls out because typeisint(bool) is
		// false — the same exclusion the old streq("bool") arm encoded.
		let bu: *tinfo = nil;
		if (n.lhs != nil) { bu = n.lhs.type_: *tinfo; };
		if (bu != nil && bu.kind == tykind.TY_NAMED) { bu = bu.under; };
		if (bu != nil && bu.kind == tykind.TY_PTR)   { bu = bu.sub;   };
		if (bu != nil && bu.kind == tykind.TY_NAMED) { bu = bu.under; };
		if (bu != nil && bu.kind == tykind.TY_STRUCT) {
			let u: *tinfo = n.type_: *tinfo;
			if (u != nil && u.kind == tykind.TY_NAMED) { u = u.under; };
			if (typeisint(u)) {
				*sz_out = u.size: i32;
				*unsigned_out = typeisunsigned(u);
			};
		};
		return;
	};
};

// variantnamematch — tagged-union variant names are compared as if
// they'd been alias-resolved. Pattern names can be module-qualified
// (`strconv.invalid` from a `case let e: strconv.invalid =>`),
// while the variant's declared name inside its own module is bare
// (`invalid`). With no checker the cgen can't follow imports, so we
// accept exact match plus suffix-after-`.` on either side. Mirrors
// the C cgen's type_eq, which goes through resolved Type pointers.
fn variantnamematch(vname: str, pname: str) bool = {
	if (streq(vname, pname)) { return true; };
	// `pname` is qualified, `vname` is bare: drop module prefix.
	let i: i32 = 0;
	for (i < pname.len) {
		if (pname[i] == '.': u8) {
			let tail: str;
			tail.ptr = pname.ptr + i + 1;
			tail.len = pname.len - i - 1;
			if (streq(tail, vname)) { return true; };
		};
		i += 1;
	};
	// `vname` is qualified, `pname` is bare: same trick in reverse.
	let j: i32 = 0;
	for (j < vname.len) {
		if (vname[j] == '.': u8) {
			let tail: str;
			tail.ptr = vname.ptr + j + 1;
			tail.len = vname.len - j - 1;
			if (streq(tail, pname)) { return true; };
		};
		j += 1;
	};
	return false;
};

// inferletcalltype — for an annotation-less `let x = expr;`, return
// a usable tnode for cgen's struct-aware paths. Today: `let x =
// f()?` infers x's type from the success variant of f's tagged
// return; without this, x has tnode = nil and `x.field` falls into
// the SB-symbol fallback (linker reports `undefined reference to
// <fieldname>`). We don't infer for plain `let x = f()` yet —
// non-tagged returns don't carry their type back the same way.
fn inferletcalltype(c: *cgen, rhs: *node) *node = {
	if (rhs == nil) { return nil; };
	// `?` (N_TRYPROP) and `!` (N_TRYUNW) both unwrap a tagged
	// return to its success variant; the rhs we want the type of
	// is the inner call expression.
	let unwrap: bool = false;
	let call: *node = rhs;
	if (rhs.kind == nkind.N_TRYPROP) { call = rhs.lhs; unwrap = true; };
	if (rhs.kind == nkind.N_TRYUNW)  { call = rhs.lhs; unwrap = true; };
	if (call == nil) { return nil; };
	if (call.kind != nkind.N_CALL) { return nil; };
	let callee: *node = call.lhs;
	if (callee == nil) { return nil; };
	let cname: str;
	cname.ptr = nil; cname.len = 0;
	let cmod: str;
	cmod.ptr = nil; cmod.len = 0;
	if (callee.kind == nkind.N_IDENT) {
		cname = callee.str;
		cmod = c.curmod;
	};
	if (callee.kind == nkind.N_DOT) {
		cname = callee.str;
		if (callee.lhs != nil) {
			if (callee.lhs.kind == nkind.N_IDENT) {
				cmod = callee.lhs.str;
			};
		};
	};
	if (cname.len == 0) { return nil; };
	let rtyp: *node = fnretlookupmod(c, cname, cmod);
	if (rtyp == nil) { return nil; };
	if (unwrap) {
		// Strip error variants — success type is the first
		// variant of the tagged return.
		if (rtyp.kind != nkind.N_TTAGGED) { return nil; };
		return rtyp.list;
	};
	// Plain call: declared return type is the local's type.
	return rtyp;
};

// letslotsize — slot size for a `let` binding. Like slotsize, but
// detects `[_]T = arrlit;` (the type-AST has rhs == nil as the
// length-inferred sentinel) and computes count × element-size from
// the initialiser. Called from cglet at emit time so the frame
// grows monotonically per first-use (#15).
//
// `let x = f();` (no annotation): infer from `f`'s declared return
// type so a 24B tagged-union return reserves all three spill slots,
// not the default 8B. Without this, the AX:DX:CX spill in cglet's
// tagged-init branch writes past the local and tramples the next
// slot.
export fn letslotsize(c: *cgen, n: *node) i32 = {
	// `[_]T = arrlit;` — inferred-length array. slotsize would
	// return elem_size * 1 (treating missing length as 1); intercept
	// and compute the real count first.
	if (n.lhs != nil) {
		if (n.lhs.kind == nkind.N_TARRAY) {
			if (n.lhs.rhs == nil) {
				if (n.rhs != nil) {
					if (n.rhs.kind == nkind.N_ARRLIT) {
						let elemn: *node = n.lhs.lhs;
						let esz: i32 = 8;
						if (elemn != nil) {
							if (elemn.kind == nkind.N_TNAME) {
								// Composite primitive: `str` is 16B
								// (ptr+len) — primsize returns 0 for
								// it, so it'd slot 8B without this.
								if (streq(elemn.str, "str")) {
									esz = primtypesize("str"): i32;
								} else {
									let ps: i32 = primsize(elemn.str);
									if (ps > 0) { esz = ps; };
								};
							};
						};
						let cnt: i32 = 0;
						let e: *node = n.rhs.list;
						for (e != nil) {
							let adv: bool = true;
							if (e.kind == nkind.N_FIELD) {
								if (streq(e.str, "...")) {
									e = nil;
									adv = false;
								};
							};
							if (adv) {
								cnt += 1;
								e = e.next;
							};
						};
						return esz * cnt;
					};
				};
			};
		};
	};
	if (n.lhs != nil) { return slotsize(c, n.lhs); };
	// Annotation-less init: defer to the call's return type if we
	// can infer it. Tagged-union returns need 24B; everything else
	// matches slotsize on the inferred type.
	let inferred: *node = inferletcalltype(c, n.rhs);
	if (inferred != nil) { return slotsize(c, inferred); };
	return 8;
};

// #48 A.6.3d: AST walker retired. resolvewalk (check.ww:426-436) stamps
// `n.type_` on every N_T* kind via tinfofornode, which folds TBANG
// (inner unchanged, check.ww:1154-1161), TNAME alias chains
// (resolvealias, check.ww:1128-1153), TARRAY/TPTR/TSLICE/TCHAN/TFN/
// TENUM/TTUPLE/TSTRUCT/TTAGGED with size + slot-padded slotsize.
//
// cstage SSoT is distributed — there is no single slot_size(Type*).
// Tagged slot follows cstage cmd/wcc/check.c:348 + :998 (tag (8) +
// max variant payload rounded to 8); the wwstage TTAGGED arm at
// check.ww:1296-1346 mirrors that layout. cgen.c:4850 / :5193 use
// the same `(su->kind == TY_TAGGED) ? su->size : 16` pattern for the
// match-spill slot. Pointer-and-narrow → 8 is the local-frame
// convention encoded at every cstage `localoff(…, 8, …)` callsite
// (cmd/w6c/cgen.c throughout); wwstage encodes the same pad-to-8 at
// this read site so `[N]i32` stride stays 4 (natural) — moving it
// into ti.slotsize would lift array stride to 8/elem.
//
// `c: *cgen` retained unused for callsite stability (localloadop
// precedent, 68219a1).
fn slotsize(c: *cgen, typn: *node) i32 = {
	if (typn == nil) { return 8; };
	let ti: *tinfo = typn.type_: *tinfo;
	if (ti == nil) { return 8; };
	// #63 Phase-N step 1: peel TY_NAMED before this structural query.
	// #64 builds per-decl NAMED wrappers (tinfofornode), so the peel
	// now fires on aliased operands; byte-id holds because it collapses
	// NAMED to the alias-invariant underlying this read consumes.
	for (ti != nil && ti.kind == tykind.TY_NAMED) { ti = ti.under; };
	if (ti == nil) { return 8; };
	let kk: tykind = ti.kind;
	if (kk == tykind.TY_VOID) { return 0; };
	if (kk == tykind.TY_PTR    || kk == tykind.TY_SLICE  ||
	    kk == tykind.TY_CHAN   || kk == tykind.TY_FN     ||
	    kk == tykind.TY_STR    || kk == tykind.TY_TAGGED) {
		return ti.size: i32;
	};
	if (kk == tykind.TY_STRUCT || kk == tykind.TY_TUPLE ||
	    kk == tykind.TY_ARRAY) {
		return ti.slotsize: i32;
	};
	return 8;
};

// fieldsize — slot-padded byte width of a struct field's type-AST,
// consumed by registerstruct's alignment + offset math (≥8→8 /
// ≥4→4 / ≥2→2 ladder at L1875-1877) and by the `*p OP=` deref-
// compound at cgenexpr.ww:3594. Mirror of check.ww:1053
// `fieldslotsize` (the same dispatch on the populated tinfo);
// slotsize-template precedent at a828c03. cstage SSoT is
// `f->type->size` (cmd/w6c/cgen.c:1386, :1656, :2515, :2535);
// wwstage routes through ti.slotsize for composites since the
// stack-slot pad rules live on tinfo (#48 verdict), and through
// ti.size for the kinds whose natural size already equals their
// in-struct width.
//
// `c: *cgen` retained unused for callsite stability (slotsize /
// localloadop precedent, a828c03 / 68219a1).
fn fieldsize(c: *cgen, tnode: *node) i32 = {
	if (tnode == nil) { return 8; };
	let ti: *tinfo = tnode.type_: *tinfo;
	if (ti == nil) { return 8; };
	// #63 Phase-N step 1: peel TY_NAMED before this structural query.
	// #64 builds per-decl NAMED wrappers (tinfofornode), so the peel
	// now fires on aliased operands; byte-id holds because it collapses
	// NAMED to the alias-invariant underlying this read consumes.
	for (ti != nil && ti.kind == tykind.TY_NAMED) { ti = ti.under; };
	if (ti == nil) { return 8; };
	let k: tykind = ti.kind;
	if (k == tykind.TY_STRUCT) { return ti.slotsize: i32; };
	if (k == tykind.TY_ARRAY)  { return ti.slotsize: i32; };
	if (k == tykind.TY_TAGGED) { return ti.size: i32; };
	if (k == tykind.TY_SLICE)  { return ti.size: i32; };
	if (k == tykind.TY_PTR || k == tykind.TY_FN ||
	    k == tykind.TY_CHAN) { return 8; };
	if (k == tykind.TY_STR)    { return ti.size: i32; };
	// Primitives + TY_ENUM keep natural width inside structs
	// (cstage parity: cgen.c reads f->type->size directly). TY_TUPLE
	// flows here too — pre-collapse fallback was 8, the populated
	// ti.size carries the natural sum; ken-thompson 2026-05-23 review:
	// keep the corrected behavior, no fixtures in selfhost exercise
	// a tuple-typed struct field today (995 byte-id is the gate).
	if (ti.size > 0u64) { return ti.size: i32; };
	return 8;
};

fn registerstruct(c: *cgen, name: str, srcmod: str, tstruct: *node) void = {
	let si: *structinfo = alloc(structinfo{
		sname = name,
		smod = srcmod,
	})!;
	let head: *fieldinfo = nil;
	let tail: *fieldinfo = nil;
	let off: i32 = 0;
	let f: *node = tstruct.list;
	for (f != nil) {
		if (f.kind == nkind.N_TFIELD) {
			let sz: i32 = fieldsize(c, f.lhs);
			// Align to 8 for any field >= 4 bytes (matches our other
			// cgen choices). i8/u8/bool may sit on odd byte offsets;
			// the C cgen does similar best-effort packing.
			let aln: i32 = 1;
			if (sz >= 8) { aln = 8; }
			else { if (sz >= 4) { aln = 4; }
			else { if (sz >= 2) { aln = 2; }; }; };
			if ((off & (aln - 1)) != 0) {
				off = (off + aln - 1) & ~(aln - 1);
			};
			let fi: *fieldinfo = alloc(fieldinfo{
				fname = f.str,
				foff  = off,
				fsz   = sz,
				tnode = f.lhs,
			})!;
			if (head == nil) { head = fi; tail = fi; }
			else { tail.finext = fi; tail = fi; };
			off += sz;
		};
		f = f.next;
	};
	// Round total to 8 for stack-slot use.
	if ((off & 7) != 0) { off = (off + 7) & ~7; };
	si.fields = head;
	si.totsize = off;
	si.sinext = c.structs;
	c.structs = si;
};

fn collectstructs(c: *cgen, file: *node) void = {
	c.structs = nil;
	if (file == nil) { return; };
	let d: *node = file.list;
	for (d != nil) {
		if (d.kind == nkind.N_TYPEDECL) {
			let body: *node = d.lhs;
			// #9: an error-struct (`type X = !struct{...}`) carries an
			// N_TBANG-wrapped body; peel it so X registers like any
			// struct. cstage is tinfo-based and needs no table, but
			// wwstage's name-keyed widen dispatch (cgreturn needswiden
			// → cgwidentaggedstore) resolves struct layout through
			// c.structs; an unregistered error-struct made the
			// struct-variant-of-large-union return silently drop its
			// construction (cs!=ww). The strategic fix is #222's sret
			// cutover, which deletes this name-keyed dispatch outright;
			// until then the table must be complete (errors.opaque_ is
			// the first such type). #10 tracks retiring the table.
			if (body != nil && body.kind == nkind.N_TBANG) {
				body = body.lhs;
			};
			if (body != nil) {
				if (body.kind == nkind.N_TSTRUCT) {
					registerstruct(c, d.str, d.nmod, body);
				};
			};
		};
		d = d.next;
	};
};

// isstrtype — alias-aware. Reads the stamped tinfo so `str`,
// `type alias = str`, `!str`, and chained aliases all route to the
// str-shaped slot. Cite cstage cgen.c:159 `type_isstr` SSoT.
// Collapsed onto typeisstr per A.6.3b (#46); the prior AST walker is
// reconstituted by typeisstr's TY_NAMED chase + tinfofornode's
// TBANG-unwrap (check.ww:1145).
fn isstrtype(c: *cgen, t: *node) bool = {
	if (t == nil) { return false; };
	return typeisstr(t.type_: *tinfo);
};

fn isslicetype(c: *cgen, t: *node) bool = {
	if (t == nil) { return false; };
	return typeisslice(t.type_: *tinfo);
};

// resolvetagged — return the underlying N_TTAGGED node for `t`, or nil
// if `t` doesn't ultimately denote a tagged union. Follows N_TNAME
// aliases (via resolvetype) and unwraps one leading N_TBANG so
// `type error = !(invalid | overflow);` resolves to its inner
// `(invalid | overflow)` node. Use at sites that read variant lists
// or detect nullable folding off a scrutinee — cgmatch, cgtypetest,
// cgtypeassert — so aliased `!(A|B)` shapes still dispatch.
export fn resolvetagged(c: *cgen, t: *node) *node = {
	let r: *node = resolvetype(c, t);
	if (r == nil) { return nil; };
	if (r.kind == nkind.N_TBANG) {
		let inner: *node = r.lhs;
		if (inner == nil) { return nil; };
		r = resolvetype(c, inner);
		if (r == nil) { return nil; };
	};
	if (r.kind == nkind.N_TTAGGED) { return r; };
	return nil;
};

// matchscrutt — resolve a non-ident match scrutinee node to its tagged
// type (or nil if unresolvable). Used by cgmatch to size the
// @match_spill slot at first use (#15 first-use+fail-loud convergence).
// IDENT scrutinees use a different lookup path (read off the local
// directly, no spill) so this returns nil for them too.
fn matchscrutt(c: *cgen, scrut: *node) *node = {
	if (scrut == nil) { return nil; };
	let k: nkind = scrut.kind;
	if (k == nkind.N_IDENT) { return nil; };
	if (k == nkind.N_CALL) {
		let callee: *node = scrut.lhs;
		if (callee != nil) {
			let cnm: str;
			cnm.ptr = nil; cnm.len = 0;
			let cmod: str;
			cmod.ptr = nil; cmod.len = 0;
			if (callee.kind == nkind.N_IDENT) { cnm = callee.str; };
			if (callee.kind == nkind.N_DOT) {
				cnm = callee.str;
				// Same-module-first disambiguation: a leaf collision
				// on `next` (utf8.next + caller-side next) otherwise
				// returns the last-declared (caller) rtype and the
				// 4-arm match collapses arms 2+ to tag 0. Task #31.
				if (callee.lhs != nil) {
					if (callee.lhs.kind == nkind.N_IDENT) {
						cmod = callee.lhs.str;
					};
				};
			};
			if (cnm.len > 0) {
				let rtyp: *node = fnretlookupmod(c, cnm, cmod);
				if (rtyp != nil) { return resolvetagged(c, rtyp); };
			};
		};
		return nil;
	};
	if (k == nkind.N_INDEX) {
		let ibase: *node = scrut.lhs;
		if (ibase == nil) { return nil; };
		if (ibase.kind != nkind.N_IDENT) { return nil; };
		let bl: *local = localfindnode(c, ibase.str);
		let btn: *node = nil;
		if (bl != nil) { btn = bl.tnode; }
		else           { btn = letvartnode(c, ibase.str); };
		if (btn == nil) { return nil; };
		let bk: nkind = btn.kind;
		let etn: *node = nil;
		if (bk == nkind.N_TARRAY) { etn = btn.lhs; };
		if (bk == nkind.N_TSLICE) { etn = btn.lhs; };
		if (bk == nkind.N_TPTR)   { etn = btn.lhs; };
		if (etn == nil) { return nil; };
		return resolvetagged(c, etn);
	};
	if (k == nkind.N_DOT) {
		// #67: read the field's stamped tinfo off the N_DOT node
		// (post-#66 N_DOT carries the field type) instead of the
		// dotfieldtnode AST walk. cgmatch's gate reads scrutt.type_
		// via istaggedtype, so the resolved N_TTAGGED node the walk
		// produced is no longer the carrier; the istaggedtype guard
		// preserves the nil-for-non-tagged contract the sibling
		// N_CALL/N_INDEX branches get from resolvetagged.
		if (!istaggedtype(c, scrut)) { return nil; };
		return scrut;
	};
	return nil;
};

// matchspillsz — slot size for the @match_spill scratch a non-ident
// scrutinee lands in. Mirrors cstage's `slot_size = (su->kind ==
// TY_TAGGED) ? su->size : 16` (cmd/w6c/cgen.c cgmatch). 16 default
// when the scrutinee type can't be resolved keeps the historical
// alloc for non-tagged / unresolved cases. Called by cgmatch at first
// use; #15 first-use+fail-loud pins this size per fn.
fn matchspillsz(c: *cgen, scrutt: *node) i32 = {
	if (scrutt == nil) { return 16; };
	let sz: i32 = slotsize(c, scrutt);
	if (sz <= 0) { return 16; };
	return sz;
};

// structparamsize — bytes occupied by a user-defined by-value struct
// param if it fits in 1-2 SysV integer eightbytes (cstage cgen.c
// struct_arg_size mirror; gates on size <= 16). Returns 0 for non-
// struct types or oversized structs so callers can fall through to
// other dispatch arms. Pre-#11 the wwstage prologue had no struct
// branch — user-defined struct params dropped through to the 8B
// scalar catch-all, the second-half value registers (DX/CX) were
// never spilled, and field reads from the under-allocated slot
// trailed into the saved-BP word.
fn structparamsize(c: *cgen, t: *node) i32 = {
	if (c == nil) { return 0; };
	let r: *node = resolvetype(c, t);
	if (r == nil) { return 0; };
	if (r.kind != nkind.N_TNAME) { return 0; };
	let nm: str = r.str;
	if (streq(nm, "str")) { return 0; };
	if (primsize(nm) > 0) { return 0; };
	let si: *structinfo = structlookup(c, nm);
	if (si == nil) { return 0; };
	if (si.totsize <= 0) { return 0; };
	if (si.totsize > 16) { return 0; };
	return si.totsize;
};

// structfloatclass — SysV per-eightbyte classification for the #165
// float-bearing-struct param case (param twin of #171's struct return;
// classifies per-eightbyte, not #163's per-element). Returns 0 when the
// struct does NOT qualify — the caller keeps the all-GP transport, which
// is correct + byte-identical there — for: not a <=16B struct; an all-
// integer layout (no float to route); an f32 field; >1 float packed in
// one eightbyte; a float straddling the 8-byte SysV eightbyte boundary;
// or an aggregate field (SysV would recurse, out of scope). Otherwise a
// packed result whose low bits hold the eightbyte count nb (1|2) and bit
// (4+e) marks eightbyte e SSE-class (a lone f64). Qualifies iff every
// eightbyte is pure-INT or a lone f64 AND at least one is f64. f32 /
// sub-eightbyte packing deferred (#165b). Mirrors cstage
// struct_float_class (cmd/w6c/cgen.c).
fn structfloatclass(c: *cgen, t: *node) i32 = {
	if (c == nil) { return 0; };
	let r: *node = resolvetype(c, t);
	if (r == nil) { return 0; };
	if (r.kind != nkind.N_TNAME) { return 0; };
	let nm: str = r.str;
	if (streq(nm, "str")) { return 0; };
	if (primsize(nm) > 0) { return 0; };
	let si: *structinfo = structlookup(c, nm);
	if (si == nil) { return 0; };
	if (si.totsize <= 0) { return 0; };
	if (si.totsize > 16) { return 0; };
	// SysV classifies aggregates in 8-byte eightbytes; 8 is the
	// eightbyte stride, not a type footprint.
	let nb: i32 = 1;
	if (si.totsize > 8) { nb = 2; };
	let nflt0: i32 = 0; let nflt1: i32 = 0;
	let nint0: i32 = 0; let nint1: i32 = 0;
	let fi: *fieldinfo = si.fields;
	for (fi != nil) {
		let foff: i32 = fi.foff;
		let fsz: i32 = fi.fsz;
		let e: i32 = foff / 8;
		if (e < 0) { return 0; };
		if (e >= nb) { return 0; };
		if (isfloattype(c, fi.tnode)) {
			if (isf32type(c, fi.tnode)) { return 0; };
			if ((foff & 7) != 0) { return 0; };
			if (fsz != 8) { return 0; };
			if (e == 0) { nflt0 += 1; } else { nflt1 += 1; };
		} else {
			if (isslicetype(c, fi.tnode)) { return 0; };
			if (isstrtype(c, fi.tnode)) { return 0; };
			if (istaggedtype(c, fi.tnode)) { return 0; };
			if (structparamsize(c, fi.tnode) > 0) { return 0; };
			// Alias-aware array/tuple reject, mirroring cstage's
			// NAMED-peeled TY_ARRAY/TY_TUPLE (cgen.c struct_float_class).
			// A direct-AST-kind N_TARRAY test misses an aliased array
			// and every tuple field; the fsz>8 guard below also lets a
			// <=8B one slip, so such a struct would wrongly SSE-route on
			// this stage but stay GP on cstage (a #165b leak).
			let rf: *node = resolvetype(c, fi.tnode);
			if (rf != nil) {
				if (rf.kind == nkind.N_TARRAY) { return 0; };
				if (rf.kind == nkind.N_TTUPLE) { return 0; };
			};
			if (fsz > 8) { return 0; };
			if ((foff + fsz - 1) / 8 != e) { return 0; };
			if (e == 0) { nint0 += 1; } else { nint1 += 1; };
		};
		fi = fi.finext;
	};
	let enc: i32 = nb;
	let hasfloat: bool = false;
	if (nflt0 == 1 && nint0 == 0) { enc += 16; hasfloat = true; }
	else { if (nflt0 != 0) { return 0; }; };
	if (nb == 2) {
		if (nflt1 == 1 && nint1 == 0) { enc += 32; hasfloat = true; }
		else { if (nflt1 != 0) { return 0; }; };
	};
	if (!hasfloat) { return 0; };
	return enc;
};

// istaggedtype — alias-aware. Reads stamped tinfo so `T`,
// `type alias = (A|B)`, `type error = !(invalid|overflow)` all
// resolve to TY_TAGGED — tinfofornode handles the N_TBANG unwrap
// (check.ww:1145) so we don't re-walk it here. Cite cstage cgen.c
// (`type_chase_named` + TY_TAGGED). Collapsed per A.6.3b (#46).
fn istaggedtype(c: *cgen, t: *node) bool = {
	if (t == nil) { return false; };
	return typeistagged(t.type_: *tinfo);
};

// isfloattype — f32 / f64 / untyped_float (alias-aware). Cite cstage
// cgen.c:117 `cg_isfloat`. Dispatches MOVSS/MOVSD-shaped paths across
// cglet, cgident, cgassign, cgbin, cgcast, cgcall, cgreturn, fn-
// prologue. Collapsed per A.6.3b (#46).
export fn isfloattype(c: *cgen, t: *node) bool = {
	if (t == nil) { return false; };
	return typeisfloat(t.type_: *tinfo);
};

// isf32type — narrower: true only for f32 (after alias chase). Cite
// cstage cgen.c:188 `type_isf32`. Picks MOVSS vs MOVSD and the SS-
// variant arithmetic / cast opcodes. Collapsed per A.6.3b (#46).
export fn isf32type(c: *cgen, t: *node) bool = {
	if (t == nil) { return false; };
	return typeisf32(t.type_: *tinfo);
};

// isnullabletype — `(*T | void)` one-word fold per Hare's
// `(*T | null)` semantics. Cite cstage cgen.c:396 `type_isnullable`;
// the .nullable flag lands on tinfo at check.ww:1309-1318 when the
// two-variant shape matches. Collapsed per A.6.3b (#46).
export fn isnullabletype(t: *node) bool = {
	if (t == nil) { return false; };
	return typeisnullable(t.type_: *tinfo);
};

// nullableptrtag — 0-based index of the *T variant in a nullable
// union. Mirror of cstage cgen.c:404-416 `nullable_ptr_tag`: linear
// scan ti.params, strip TY_NAMED on each variant, return idx of first
// TY_PTR. Phase 1 (26724fe) populated the chain in tinfofornode's
// TTAGGED arm so this walk could retire the AST-keyed predecessor.
export fn nullableptrtag(t: *node) i32 = {
	if (t == nil) { return 0; };
	let ti: *tinfo = t.type_: *tinfo;
	if (ti == nil) { return 0; };
	// #63 Phase-N step 1: peel TY_NAMED before this structural query.
	// #64 builds per-decl NAMED wrappers (tinfofornode), so the peel
	// now fires on aliased operands; byte-id holds because it collapses
	// NAMED to the alias-invariant underlying this read consumes.
	for (ti != nil && ti.kind == tykind.TY_NAMED) { ti = ti.under; };
	if (ti == nil) { return 0; };
	if (ti.kind != tykind.TY_TAGGED) { return 0; };
	let p: *tparam = ti.params;
	let i: i32 = 0;
	for (p != nil) {
		let vt: *tinfo = p.type_;
		if (vt != nil) {
			if (vt.kind == tykind.TY_NAMED) { vt = vt.under; };
			if (vt != nil) {
				if (vt.kind == tykind.TY_PTR) { return i; };
			};
		};
		p = p.tnext;
		i += 1;
	};
	return 0;
};

// voidvariantindex — find the 0-based index of the `void` variant in a
// tagged-union type expr, -1 if absent. Used by cgreturn to map bare
// `return;` in a tagged-union-returning fn to the void variant's tag.
fn voidvariantindex(tagged: *node) i32 = {
	if (tagged == nil) { return -1; };
	if (tagged.kind != nkind.N_TTAGGED) { return -1; };
	let v: *node = tagged.list;
	let idx: i32 = 0;
	for (v != nil) {
		if (v.kind == nkind.N_TNAME) {
			if (streq(v.str, "void")) { return idx; };
		};
		v = v.next;
		idx += 1;
	};
	return -1;
};

// taggedvariantindex — given the tagged-union type expr and the
// returned value's surface type, find the matching variant's 0-based
// index. Compare by exact type name first; if no match, fall back to
// "any str-shape variant matches an str-typed value".
fn taggedvariantindex(c: *cgen, tagged: *node, rhs: *node) i32 = {
	if (tagged == nil) { return -1; };
	return taggedvariantindext(c, tagged.type_: *tinfo, rhs);
};

// taggedvariantindext — tinfo-keyed core of taggedvariantindex. Given
// the dst tagged tinfo `du` (NAMED-peeled internally, gated TY_TAGGED)
// and the source value node, returns the 0-based variant index. #68: the
// tagged-store machinery reads tinfo directly (no type node), so the
// node-keyed taggedvariantindex delegates here off `tagged.type_`.
//
// #66 Phase-N step 3: match the value's stamped type against the variant
// types by typeeq (flatvariantidxt), replacing the rhstargetname surface-
// name compare. Untyped/loose values (whose .type_ is untyped_* and can't
// typeeq a concrete variant) return -1 there and drop to the str/slice
// shape scan below — ww has no type_assignable to mirror cg_variant_match's
// untyped-src arm.
fn taggedvariantindext(c: *cgen, du: *tinfo, rhs: *node) i32 = {
	if (du == nil)  { return -1; };
	if (rhs == nil) { return -1; };
	let ti: *tinfo = du;
	for (ti != nil && ti.kind == tykind.TY_NAMED) { ti = ti.under; };
	if (ti == nil) { return -1; };
	if (ti.kind != tykind.TY_TAGGED) { return -1; };
	let r: i32 = flatvariantidxt(ti, rhs.type_: *tinfo);
	if (r >= 0) { return r; };
	// Shape fallback: classify rhs as (str, slice, scalar/other) and
	// pick the first variant of matching shape. Stands in for cstage's
	// type_assignable on an untyped src — the typeeq pass above can't
	// match untyped_* against a concrete variant, and the slice axis
	// keeps a (u8 | []u8) widen off the leading scalar variant (task
	// #19). tinfo.params is already spread-flattened (#61a — `...inner`
	// inlined in declaration order), so the old N_TTAGGED.list spread-
	// walk collapses to a flat scan over p.type_.
	let wantstr: bool = nodeisstr(c, rhs);
	let wantslice: bool = nodeisslice(c, rhs);
	let p: *tparam = ti.params;
	let idx: i32 = 0;
	for (p != nil) {
		let vt: *tinfo = p.type_;
		let visstr: bool = typeisstr(vt);
		let visslice: bool = typeisslice(vt);
		if (visstr == wantstr && visslice == wantslice) { return idx; };
		p = p.tnext;
		idx += 1;
	};
	return -1;
};

// flatvariantidx — flat 0-based index of the variant whose type matches
// the pattern node `pat`, by typeeq on the stamped tinfos. Reads the
// pre-flattened variant chain off tinfo.params (#61a — `...inner`
// spreads already inlined in declaration order); peels TY_NAMED then
// gates TY_TAGGED.
//
// #66 Phase-N step 3 (THE FLIP, user-ruled B-full): match by
// typeeq(p.type_, pat.type_) — nominal identity carried by the per-decl
// TY_NAMED ptr — instead of the surface-name compare. So `type linerr =
// !str` ≠ str and a cross-module `a.T` ≠ `b.T` are now distinguished.
// Mirrors cstage cg_variant_match (cmd/w6c/cgen.c:451): both-NAMED →
// ptr-id (typeeq, typ.ww:514), one-NAMED → kind mismatch → false. ww has
// no type_assignable, so the untyped/loose arm (cg_variant_match's first
// branch) lives in the caller's str/slice shape fallback, not here.
fn flatvariantidx(c: *cgen, tagged: *node, pat: *node) i32 = {
	if (tagged == nil) { return -1; };
	if (pat == nil) { return -1; };
	return flatvariantidxt(tagged.type_: *tinfo, pat.type_: *tinfo);
};

// flatvariantidxt — tinfo-keyed core of flatvariantidx: flat 0-based
// index of the variant whose type typeeq's `want`, over the pre-
// flattened tinfo.params chain (#61a). Peels TY_NAMED then gates
// TY_TAGGED. #68: the tagged-store machinery reads tinfo directly and
// has no type node to hand the node-keyed flatvariantidx, so the typeeq
// core lives here; flatvariantidx + taggedvariantindext + cgwidentagremap
// all funnel through it. Mirrors cstage cg_tag_for_variant
// (cmd/w6c/cgen.c:503) + cg_variant_match's both-NAMED ptr-id / typeeq
// arm (:451).
fn flatvariantidxt(tagged: *tinfo, want: *tinfo) i32 = {
	if (want == nil) { return -1; };
	let ti: *tinfo = tagged;
	for (ti != nil && ti.kind == tykind.TY_NAMED) { ti = ti.under; };
	if (ti == nil) { return -1; };
	if (ti.kind != tykind.TY_TAGGED) { return -1; };
	// Pass 1: exact match (NAMED-vs-NAMED typeeq, tagged-vs-tagged, bare
	// typeeq). Exact matches take precedence and need no guard — distinct
	// variants don't exact-match the same source.
	let p: *tparam = ti.params;
	let idx: i32 = 0;
	for (p != nil) {
		if (cgvariantmatch(p.type_, want)) { return idx; };
		p = p.tnext;
		idx += 1;
	};
	// Pass 2 (#15): no exact variant matched — structurally match a BARE
	// source against a NAMED-alias variant (bare *vtable into the
	// `stream` (= *vtable) variant of `(file | stream)`). The bare side
	// has no nominal identity, so structure is the only discriminator;
	// without this the widen found no variant and defaulted to tag 0,
	// miscompiling emitbytes' io.write(&cgoutstream.vt). Exact-first
	// (pass 1) keeps a bare `i64` into `(i64 | oserror)` binding the
	// exact `i64`. drew's proviso: guard the structural fallback like the
	// #218 nested-widen site — a bare source matching >=2 NAMED variants
	// needs nominal layout to disambiguate, so hard-error.
	if (want.kind != tykind.TY_NAMED) {
		let q: *tparam = ti.params;
		let qi: i32 = 0;
		let found: i32 = -1;
		let n: i32 = 0;
		for (q != nil) {
			// One-level NAMED unwrap: a chained ptr-alias variant
			// (type a=*X; type b=a) isn't reached here, so it would
			// silently mis-tag — unexercised (zero in corpus), see
			// task #17.
			let pu: *tinfo = q.type_;
			if (pu != nil && pu.kind == tykind.TY_NAMED
			    && pu.under != nil && typeeq(pu.under, want)) {
				if (found < 0) { found = qi; };
				n += 1;
			};
			q = q.tnext;
			qi += 1;
		};
		if (n >= 2) {
			let msg: str = "flatvariantidxt: bare source structurally matches >=2 NAMED variants — ambiguous without nominal layout (#15/#218/#199b/#10)\n";
			os.write(2, msg.ptr, msg.len: u64);
			os.exit(1);
		};
		return found;
	};
	return -1;
};

// cgvariantmatch — does a source value of type `want` tag as variant
// `vt` in a tagged-union dispatch? Mirrors cstage cg_variant_match
// (cmd/w6c/cgen.c):
//   - both NAMED → nominal ptr-id (typeeq line 545 = same ptr only)
//   - exactly one NAMED → #218 nominal lost: fall back to structural
//     equality of the two unwrapped tagged unions, so an outer widen of
//     a NAMED multi-variant union into an enclosing union computes its
//     tag (project tinfo_lossy_nominal). Sound only while the model is
//     nominal-lossy; the collision guard in cgwidentaggedstorebp enforces
//     the invariant for when #199b/B-full lands true nominal layout.
//   - neither NAMED → structural typeeq
// The untyped/loose arm (cstage's type_assignable) is NOT mirrored here —
// ww has no type_assignable, so it lives in taggedvariantindext's str/
// slice shape fallback (the existing documented divergence).
fn cgvariantmatch(vt: *tinfo, want: *tinfo) bool = {
	if (vt == nil)   { return false; };
	if (want == nil) { return false; };
	if (vt.kind == tykind.TY_NAMED && want.kind == tykind.TY_NAMED) {
		return typeeq(vt, want);
	};
	if (vt.kind == tykind.TY_NAMED || want.kind == tykind.TY_NAMED) {
		let vu: *tinfo = vt;
		for (vu != nil && vu.kind == tykind.TY_NAMED) { vu = vu.under; };
		let wu: *tinfo = want;
		for (wu != nil && wu.kind == tykind.TY_NAMED) { wu = wu.under; };
		if (vu != nil && wu != nil
		    && vu.kind == tykind.TY_TAGGED
		    && wu.kind == tykind.TY_TAGGED) {
			return typeeq(vu, wu);
		};
		return false;
	};
	return typeeq(vt, want);
};

// cgvariantstructmatch — structural equality ignoring nominal identity
// (peel NAMED, then typeeq). #218 collision guard: counts how many dst
// variants share the source's *shape*; ≥2 means the structural fallback
// could not disambiguate them once nominal identity is lost. Mirrors
// cstage cg_variant_struct_match (cmd/w6c/cgen.c).
fn cgvariantstructmatch(vt: *tinfo, want: *tinfo) bool = {
	let vu: *tinfo = vt;
	for (vu != nil && vu.kind == tykind.TY_NAMED) { vu = vu.under; };
	let wu: *tinfo = want;
	for (wu != nil && wu.kind == tykind.TY_NAMED) { wu = wu.under; };
	if (vu == nil) { return false; };
	if (wu == nil) { return false; };
	return typeeq(vu, wu);
};

// flatslicevariantidx — flat 0-based index of a slice-shape variant in
// `tagged`. Prefers the variant whose element typeeq's the pattern
// element `elem`; falls back to the first slice-shape slot when no exact
// element match is found (the untyped/loose arm — ww has no
// type_assignable). Reads the flattened tinfo.params chain (#61a); peels
// TY_NAMED then gates TY_TAGGED. The slice axis exists because a scalar-
// vs-`[]T` distinction has no surface name to key on (task #19).
//
// #66 Phase-N step 3: element compare flips from surface-name to
// typeeq(p.type_.sub, elem.type_). Mirrors cstage cg_tag_for_variant
// over Type->params. Returns -1 when no slice variant exists.
fn flatslicevariantidx(c: *cgen, tagged: *node, elem: *node) i32 = {
	if (tagged == nil) { return -1; };
	let ti: *tinfo = tagged.type_: *tinfo;
	for (ti != nil && ti.kind == tykind.TY_NAMED) { ti = ti.under; };
	if (ti == nil) { return -1; };
	if (ti.kind != tykind.TY_TAGGED) { return -1; };
	let want: *tinfo = nil;
	if (elem != nil) { want = elem.type_: *tinfo; };
	let fallback: i32 = -1;
	let p: *tparam = ti.params;
	let idx: i32 = 0;
	for (p != nil) {
		let vt: *tinfo = p.type_;
		if (vt != nil) {
			if (typeisslice(vt)) {
				if (fallback < 0) { fallback = idx; };
				if (want != nil) {
					let su: *tinfo = vt;
					for (su != nil && su.kind == tykind.TY_NAMED) { su = su.under; };
					if (su != nil) {
						if (typeeq(su.sub, want)) { return idx; };
					};
				};
			};
		};
		p = p.tnext;
		idx += 1;
	};
	return fallback;
};

// cgwidentagremap — when widening from one tagged union to a wider one,
// rewrite the source's variant tag at slot_off+0 to use the destination's
// variant indices. No-op when src and dst index orders coincide.
//
// #68: both `du` (dst) and `su` (src) are now the tagged tinfos — peel
// TY_NAMED then walk su.params, mapping each source variant to its dst
// index by typeeq (flatvariantidxt). Mirrors cg_widen_tag_remap
// (cmd/w6c/cgen.c:1177) over su->params + cg_tag_for_variant (:503).
fn cgwidentagremap(c: *cgen, du: *tinfo, su: *tinfo, slot_off: i32) void = {
	let dt: *tinfo = du;
	for (dt != nil && dt.kind == tykind.TY_NAMED) { dt = dt.under; };
	if (dt == nil) { return; };
	if (dt.kind != tykind.TY_TAGGED) { return; };
	let st: *tinfo = su;
	for (st != nil && st.kind == tykind.TY_NAMED) { st = st.under; };
	if (st == nil) { return; };
	if (st.kind != tykind.TY_TAGGED) { return; };
	let identity: bool = true;
	let p: *tparam = st.params;
	let idx: i32 = 0;
	for (p != nil) {
		let di: i32 = flatvariantidxt(dt, p.type_);
		if (di < 0) { di = 0; };
		if (di != idx) { identity = false; p = nil; }
		else           { p = p.tnext; idx += 1; };
	};
	if (identity) { return; };
	let done: str = mklabel(c, "remap_done");
	emitline("\tMOVQ\t");
	emitoff(slot_off: i64);
	emitline("(BP), AX\n");
	p = st.params;
	idx = 0;
	for (p != nil) {
		let next: str = mklabel(c, "remap_next");
		let di: i32 = flatvariantidxt(dt, p.type_);
		if (di < 0) { di = 0; };
		emitline("\tCMPQ\t$");
		emitint(idx: i64);
		emitline(", AX\n");
		emitline("\tJNE\t");
		emitline(next);
		emitline("\n");
		emitline("\tMOVQ\t$");
		emitint(di: i64);
		emitline(", AX\n");
		emitline("\tMOVQ\tAX, ");
		emitoff(slot_off: i64);
		emitline("(BP)\n");
		emitline("\tJMP\t");
		emitline(done);
		emitline("\n");
		emitlabel(next);
		p = p.tnext;
		idx += 1;
	};
	emitlabel(done);
	return;
};

// rhsisstructpayload — is `src` a struct value (literal or local ident
// of a struct type)? Returns the struct name, or empty str. Only true
// when the name is registered in c.structs — `!void` / `!i32` aliases
// share the N_STRUCTLIT / N_TNAME shape but aren't structs, and must
// fall through to the scalar/str/tagged-source paths instead.
fn rhsstructpayload(c: *cgen, src: *node) str = {
	let empty: str;
	empty.ptr = nil; empty.len = 0;
	if (src == nil) { return empty; };
	if (src.kind == nkind.N_STRUCTLIT) {
		let trefn: *node = src.lhs;
		if (trefn != nil) {
			let nm: str;
			nm.ptr = nil; nm.len = 0;
			if (trefn.kind == nkind.N_IDENT) { nm = trefn.str; };
			if (trefn.kind == nkind.N_TNAME) { nm = trefn.str; };
			if (nm.len > 0) {
				if (structlookup(c, nm) != nil) { return nm; };
			};
		};
		return empty;
	};
	if (src.kind == nkind.N_IDENT) {
		let lc: *local = localfindnode(c, src.str);
		if (lc != nil) {
			let tn: *node = lc.tnode;
			if (tn != nil) {
				if (tn.kind == nkind.N_TNAME) {
					if (structlookup(c, tn.str) != nil) {
						return tn.str;
					};
				};
			};
		};
	};
	return empty;
};

// rhstaggedsource — return the tagged-type node for `src` when src is a
// tagged-typed local ident; nil otherwise. The slot-copy path uses this
// to walk variants for tag remap.
fn rhstaggedident(c: *cgen, src: *node) *node = {
	if (src == nil) { return nil; };
	if (src.kind != nkind.N_IDENT) { return nil; };
	let lc: *local = localfindnode(c, src.str);
	if (lc == nil) { return nil; };
	let tn: *node = lc.tnode;
	if (!istaggedtype(c, tn)) { return nil; };
	return resolvetagged(c, tn);
};

// rhstaggedabicall — does `src` produce a tagged value via the AX/DX/CX
// return ABI? True for N_CALL of a tagged-returning fn, N_INDEX of a
// tagged-element base, and N_DOT of a tagged-typed struct field (after
// #28's cgdot fix loads AX/DX/CX/R8 from the field's slot). Used to
// decide whether cgexpr/spill works for the tagged-source branch of
// cgwidentaggedstore.
fn rhstaggedabicall(c: *cgen, src: *node) bool = {
	if (src == nil) { return false; };
	if (src.kind == nkind.N_CALL) {
		let callee: *node = src.lhs;
		if (callee != nil) {
			let calleename: str;
			calleename.ptr = nil; calleename.len = 0;
			let cmod: str;
			cmod.ptr = nil; cmod.len = 0;
			if (callee.kind == nkind.N_IDENT) {
				calleename = callee.str;
				cmod = c.curmod;
			};
			if (callee.kind == nkind.N_DOT) {
				calleename = callee.str;
				if (callee.lhs != nil) {
					if (callee.lhs.kind == nkind.N_IDENT) {
						cmod = callee.lhs.str;
					};
				};
			};
			if (calleename.len > 0) {
				let rtyp: *node = fnretlookupmod(c, calleename, cmod);
				if (rtyp != nil) {
					if (istaggedtype(c, rtyp)) { return true; };
				};
			};
		};
		return false;
	};
	if (src.kind == nkind.N_INDEX) {
		let base: *node = src.lhs;
		if (base != nil) {
			if (base.kind == nkind.N_IDENT) {
				let bl: *local = localfindnode(c, base.str);
				if (bl != nil) {
					let btn: *node = bl.tnode;
					if (btn != nil) {
						let bk: nkind = btn.kind;
						let elemt: *node = nil;
						if (bk == nkind.N_TARRAY) { elemt = btn.lhs; };
						if (bk == nkind.N_TSLICE) { elemt = btn.lhs; };
						if (bk == nkind.N_TPTR)   { elemt = btn.lhs; };
						if (elemt != nil) {
							if (istaggedtype(c, elemt)) {
								return true;
							};
						};
					};
				};
			};
		};
	};
	// N_DOT of a tagged-typed struct field — cgdot loads
	// AX=tag, DX=word0, CX=word1[, R8=word2], so downstream
	// spill matches the call/index shapes. #58 A.6.3i-phase-2:
	// read the checker-stamped n.type_ (check.ww N_DOT struct-field
	// stamp) instead of re-deriving via dotfieldtnode — matches
	// cstage cg_widen_tagged_store reading src->type directly
	// (cmd/w6c/cgen.c:1302-1305). typeistagged(nil) is false.
	if (src.kind == nkind.N_DOT) {
		if (typeistagged(src.type_: *tinfo)) { return true; };
	};
	return false;
};

// cgloadtaggedfield — load a tagged-union slot at `basereg`+foff
// into the tagged-return ABI registers (AX=tag, DX=word0, CX=word1,
// R8=word2). Slot sizes: 16B = (tag, word0), 24B = + word1, 32B
// = + word2 (slice variant). Mirrors the cstage tagged-field load
// in cmd/w6c/cgen.c (N_DOT TY_STRUCT/TY_PTR branches).
//
// Load order is fixed regardless of basereg: tag, word0, word2,
// word1. CX (word1 target) goes LAST because basereg may itself
// be CX — top-level globals address via LEAQ name(SB), CX — and
// overwriting it earlier would trash the base address for the
// remaining loads. For BP / BX bases the order is harmless.
// Callers must guarantee basereg is one of "BP", "BX", "CX"; the
// only register loaded into that is NOT a target is BX, so AX-
// or DX-rooted callers must spill first.
fn cgloadtaggedfield(c: *cgen, basereg: str, foff: i32, slot_sz: i32) void = {
	// tag → AX
	emitline("\tMOVQ\t");
	emitdispreg(foff: i64, basereg);
	emitline(", AX\n");
	// word0 → DX
	emitline("\tMOVQ\t");
	emitdispreg((foff + 8): i64, basereg);
	emitline(", DX\n");
	// word2 → R8 (slice variant: slot = 8 tag + 24 payload = 32).
	if (slot_sz > 24) {
		emitline("\tMOVQ\t");
		emitdispreg((foff + 24): i64, basereg);
		emitline(", R8\n");
	};
	// word1 → CX (load LAST; conflicts with CX-base globals).
	if (slot_sz > 16) {
		emitline("\tMOVQ\t");
		emitdispreg((foff + 16): i64, basereg);
		emitline(", CX\n");
	};
};

// cgwidentaggedstore — write tagged-union slot bytes for `src` into
// the slot at `basereg`+slot_off, sized to slot_sz. Mirrors
// cg_widen_tagged_store in cmd/w6c/cgen.c.
//
// `basereg` selects the addressing root:
//   - "BP": function-frame slot (let / assign / return / structlit /
//     array-elem scratch). Body writes straight to slot_off(BP).
//   - else (e.g. "BX" for *struct field, top-level struct LEAQ
//     base): pointer-rooted dst. cgexpr inside trashes every GPR,
//     so we route through a fresh BP-rooted scratch slot, spill
//     basereg before the body, reload after, then word-copy
//     scratch → (basereg, slot_off).
//
// Branches by source shape:
//   - nullable dst (8B slot): cgexpr → AX → slot+0.
//   - tagged src ident: copy slot words, zero-pad, tag-remap.
//   - tagged src via AX/DX/CX ABI (call / tagged-arr index): cgexpr,
//     spill words; no remap (callee already speaks dst tag order — or
//     it doesn't, in which case the source is the wider one and remap
//     would need a reversed direction we don't currently emit).
//   - struct src (literal or ident): zero slot, write fields at +8+foff,
//     tag last.
//   - str src: tag@+0, ptr@+8, len@+16, cap@+24 (str IS []u8, #1/Phase 3).
//   - scalar src: tag@+0, value@+8.
fn cgwidentaggedstore(c: *cgen, dst: *tinfo, src: *node,
    basereg: str, slot_off: i32, slot_sz: i32) void = {
	if (streq(basereg, "BP")) {
		cgwidentaggedstorebp(c, dst, src, slot_off, slot_sz);
		return;
	};
	// Pointer-rooted dst: spill basereg (cgexpr will trash it),
	// materialise into a BP-rooted scratch via the BP path, then
	// reload basereg and word-copy scratch → caller's slot.
	let bspill: i32 = localadd(c, "@tagbase", 8, nil);
	emitline("\tMOVQ\t");
	emitline(basereg);
	emitline(", ");
	emitoff(bspill: i64);
	emitline("(BP)\n");
	// Shared scratch sized at first use per #15/#26c. A sibling
	// site (cgreturn, pushargsrev, cgindex) hitting @tagscr later
	// with a larger size fatals (rule 7) — pinned offset can't
	// grow in place.
	let scr: i32 = localadd(c, "@tagscr", slot_sz, nil);
	emitline("\tXORQ\tAX, AX\n");
	let z: i32 = 0;
	for (z < slot_sz) {
		emitline("\tMOVQ\tAX, ");
		emitoff((scr + z): i64);
		emitline("(BP)\n");
		z += 8;
	};
	cgwidentaggedstorebp(c, dst, src, scr, slot_sz);
	emitline("\tMOVQ\t");
	emitoff(bspill: i64);
	emitline("(BP), ");
	emitline(basereg);
	emitline("\n");
	let k: i32 = 0;
	for (k < slot_sz) {
		emitline("\tMOVQ\t");
		emitoff((scr + k): i64);
		emitline("(BP), AX\n");
		emitline("\tMOVQ\tAX, ");
		emitdispreg((slot_off + k): i64, basereg);
		emitline("\n");
		k += 8;
	};
};

// cgwidentaggedstorebp — BP-rooted body. Called via cgwidentaggedstore
// for the natural "BP" case and via the wrapper's scratch path for
// pointer-rooted dst. Direct callers exist only in case of future
// inlined uses inside this file; new code should call the wrapper.
fn cgwidentaggedstorebp(c: *cgen, dst: *tinfo, src: *node, slot_off: i32, slot_sz: i32) void = {
	// #68: dst is the tagged tinfo. Peel TY_NAMED → du and gate
	// TY_TAGGED, mirroring cstage cg_widen_tagged_store's
	// `du = (dst->kind==TY_NAMED)?dst->under:dst` + TY_TAGGED guard
	// (cmd/w6c/cgen.c:1273). resolvetagged's N_TBANG unwrap is already
	// handled upstream by tinfofornode (check.ww:1203-1210).
	let dt: *tinfo = dst;
	for (dt != nil && dt.kind == tykind.TY_NAMED) { dt = dt.under; };
	if (dt == nil) { return; };
	if (dt.kind != tykind.TY_TAGGED) { return; };
	// Nullable fold: one 8B word holding the pointer (or 0 for void).
	if (dt.nullable != 0) {
		cgexpr(c, src);
		emitline("\tMOVQ\tAX, ");
		emitoff(slot_off: i64);
		emitline("(BP)\n");
		return;
	};
	// `expr: TaggedAlias` where the cast's destination IS the union
	// itself is a widening, not a re-interpret. cgexpr on a CAST
	// produces the inner's register shape (str: AX=ptr, BX=len), not
	// the tagged AX/DX/CX triple — so peel to the inner and route
	// through the matching concrete-variant branch below. A cast to
	// a concrete variant (`7: i32`) is left intact so the existing
	// scalar / str / slice branches pick the right variant tag.
	if (src != nil) {
		if (src.kind == nkind.N_CAST) {
			if (src.lhs != nil) {
				let inner: *node = src.lhs;
				let inneristagged: bool = false;
				if (inner.kind == nkind.N_IDENT) {
					let lc: *local = localfindnode(c, inner.str);
					if (lc != nil) {
						inneristagged = istaggedtype(c, lc.tnode);
					};
				};
				if (rhstaggedabicall(c, inner)) {
					inneristagged = true;
				};
				// Cast's destination = the dst tagged union
				// itself? The rhs of N_CAST holds the target
				// type. #68: compare on the stamped tinfos —
				// `castu == dt` (peeled-underlying ptr-id) ||
				// (castu tagged && typeeq(castt, dst)) — mirroring
				// cstage cg_widen_tagged_store's
				// `cast_is_widen = (castu==du) || (castu->kind==
				// TY_TAGGED && type_eq(castt, dst))`
				// (cmd/w6c/cgen.c:1295-1296). Replaces the prior
				// surface-name streq; same-alias TNAMEs share one
				// NAMED tinfo (check.ww:1160-1173) so typeeq hits
				// the a==b fast path.
				let castisdst: bool = false;
				let castrhs: *node = src.rhs;
				if (castrhs != nil) {
					let castt: *tinfo = castrhs.type_: *tinfo;
					let castu: *tinfo = castt;
					for (castu != nil && castu.kind == tykind.TY_NAMED) {
						castu = castu.under;
					};
					if (castu != nil) {
						if (castu == dt) {
							castisdst = true;
						} else { if (castu.kind == tykind.TY_TAGGED) {
							if (typeeq(castt, dst)) {
								castisdst = true;
							};
						}; };
					};
				};
				if (castisdst && !inneristagged) {
					src = inner;
				};
			};
		};
	};
	// #218: is the source itself a single NESTED variant of dt (its
	// whole tagged type matches one dt variant), rather than a flattened
	// SUBSET whose members spread into dt? If so, the inner tagged value
	// is the payload: store it at slot_off+8 with the outer tag at
	// slot_off+0, mirroring the scalar/struct/str single-variant arms —
	// NOT a copy-to-+0 + sub-variant remap. flatvariantidxt's structural
	// fallback (cgvariantmatch) recovers the index after the nominal-
	// lossy collapse. Gated on a tagged source so scalar/str/struct
	// sources keep their existing arms. Mirrors cstage
	// cg_widen_tagged_store's nested arm (cmd/w6c/cgen.c).
	let srctagged: bool = (rhstaggedident(c, src) != nil)
	    || rhstaggedabicall(c, src);
	if (srctagged) {
		let nested: i32 = flatvariantidxt(dt, src.type_: *tinfo);
		if (nested >= 0) {
			// drew collision guard: the structural fallback over-
			// matches if ≥2 nominally-distinct dt variants share the
			// source's shape. Unreachable under today's nominal-lossy
			// model, but INVERTS when #199b/B-full lands the nominal
			// layer — hard-error NOW so a future collision STOPS the
			// compiler instead of silently mis-tagging.
			let nmatch: i32 = 0;
			let gp: *tparam = dt.params;
			for (gp != nil) {
				if (cgvariantstructmatch(gp.type_,
				    src.type_: *tinfo)) {
					nmatch += 1;
				};
				gp = gp.tnext;
			};
			if (nmatch >= 2) {
				let msg: str = "cgwidentaggedstore: structural fallback cannot disambiguate nominally-distinct same-shape variants without nominal layout (#218/#199b/B-full)\n";
				os.write(2, msg.ptr, msg.len: u64);
				os.exit(1);
			};
			let su: *tinfo = src.type_: *tinfo;
			for (su != nil && su.kind == tykind.TY_NAMED) {
				su = su.under;
			};
			let ssz: i32 = su.size: i32;
			emitline("\tXORQ\tAX, AX\n");
			let zk: i32 = 0;
			for (zk < slot_sz) {
				emitline("\tMOVQ\tAX, ");
				emitoff((slot_off + zk): i64);
				emitline("(BP)\n");
				zk += 8;
			};
			if (src.kind == nkind.N_IDENT) {
				let lc: *local = localfindnode(c, src.str);
				let soff: i32 = lc.off;
				let ck: i32 = 0;
				for (ck < ssz) {
					emitline("\tMOVQ\t");
					emitoff((soff + ck): i64);
					emitline("(BP), AX\n");
					emitline("\tMOVQ\tAX, ");
					emitoff((slot_off + 8 + ck): i64);
					emitline("(BP)\n");
					ck += 8;
				};
			} else {
				cgexpr(c, src);
				emitline("\tMOVQ\tAX, ");
				emitoff((slot_off + 8): i64);
				emitline("(BP)\n");
				if (ssz > 8) {
					emitline("\tMOVQ\tDX, ");
					emitoff((slot_off + 16): i64);
					emitline("(BP)\n");
				};
				if (ssz > 16) {
					emitline("\tMOVQ\tCX, ");
					emitoff((slot_off + 24): i64);
					emitline("(BP)\n");
				};
				if (ssz > 24) {
					emitline("\tMOVQ\tR8, ");
					emitoff((slot_off + 32): i64);
					emitline("(BP)\n");
				};
			};
			emitline("\tMOVQ\t$");
			emitint(nested: i64);
			emitline(", ");
			emitoff(slot_off: i64);
			emitline("(BP)\n");
			return;
		};
	};
	// Tagged source ident: byte-copy slot words then tag-remap.
	// rhstaggedident gates "src is a tagged-typed local ident"; the
	// remap reads the source tagged tinfo off the local's tnode (#68).
	let st: *node = rhstaggedident(c, src);
	if (st != nil) {
		let lc: *local = localfindnode(c, src.str);
		let ssz: i32 = slotsize(c, lc.tnode);
		let soff: i32 = lc.off;
		let k: i32 = 0;
		for (k < ssz) {
			emitline("\tMOVQ\t");
			emitoff((soff + k): i64);
			emitline("(BP), AX\n");
			emitline("\tMOVQ\tAX, ");
			emitoff((slot_off + k): i64);
			emitline("(BP)\n");
			k += 8;
		};
		if (ssz < slot_sz) {
			emitline("\tXORQ\tAX, AX\n");
			let p: i32 = ssz;
			for (p < slot_sz) {
				emitline("\tMOVQ\tAX, ");
				emitoff((slot_off + p): i64);
				emitline("(BP)\n");
				p += 8;
			};
		};
		cgwidentagremap(c, dt, lc.tnode.type_: *tinfo, slot_off);
		return;
	};
	// Tagged source via AX/DX/CX/R8 register ABI (N_CALL, N_INDEX
	// of tagged element). R8 carries the 4th word for slice-payload
	// variants (slot 32B).
	if (rhstaggedabicall(c, src)) {
		cgexpr(c, src);
		emitline("\tMOVQ\tAX, ");
		emitoff(slot_off: i64);
		emitline("(BP)\n");
		if (slot_sz > 8) {
			emitline("\tMOVQ\tDX, ");
			emitoff((slot_off + 8): i64);
			emitline("(BP)\n");
		};
		if (slot_sz > 16) {
			emitline("\tMOVQ\tCX, ");
			emitoff((slot_off + 16): i64);
			emitline("(BP)\n");
		};
		if (slot_sz > 24) {
			emitline("\tMOVQ\tR8, ");
			emitoff((slot_off + 24): i64);
			emitline("(BP)\n");
		};
		return;
	};
	// Struct payload (literal or ident).
	let sname: str = rhsstructpayload(c, src);
	if (sname.len > 0) {
		let si: *structinfo = structlookup(c, sname);
		if (si != nil) {
			emitline("\tXORQ\tAX, AX\n");
			let zoff: i32 = 0;
			for (zoff < slot_sz) {
				emitline("\tMOVQ\tAX, ");
				emitoff((slot_off + zoff): i64);
				emitline("(BP)\n");
				zoff += 8;
			};
			let tag: i32 = taggedvariantindext(c, dt, src);
			if (tag < 0) { tag = 0; };
			if (src.kind == nkind.N_STRUCTLIT) {
				let fnode: *node = src.list;
				for (fnode != nil) {
					if (fnode.kind == nkind.N_FIELD) {
						let fname: str = fnode.str;
						let fi: *fieldinfo = si.fields;
						for (fi != nil) {
							if (streq(fi.fname, fname)) {
								cgexpr(c, fnode.lhs);
								if (isfloattype(c, fi.tnode)) {
									let mov: str = "MOVSD";
									if (isf32type(c, fi.tnode)) {
										mov = "MOVSS";
									};
									emitline("\t");
									emitline(mov);
									emitline("\tX0, ");
									emitoff((slot_off + 8 + fi.foff): i64);
									emitline("(BP)\n");
								} else { if (isstrtype(c, fi.tnode)) {
									// str IS []u8: 3-word field
									// (ptr,len,cap) from cgexpr's
									// AX/BX/CX (#1/Phase 3).
									emitline("\tMOVQ\tAX, ");
									emitoff((slot_off + 8 + fi.foff): i64);
									emitline("(BP)\n");
									emitline("\tMOVQ\tBX, ");
									emitoff((slot_off + 8 + fi.foff + 8): i64);
									emitline("(BP)\n");
									emitline("\tMOVQ\tCX, ");
									emitoff((slot_off + 8 + fi.foff + 16): i64);
									emitline("(BP)\n");
								} else {
									let sop: str = fieldstoreop(c, fi);
									emitline("\t");
									emitline(sop);
									emitline("\tAX, ");
									emitoff((slot_off + 8 + fi.foff): i64);
									emitline("(BP)\n");
								}; };
								fi = nil;
							} else {
								fi = fi.finext;
							};
						};
					};
					fnode = fnode.next;
				};
			} else {
				// Struct ident source: byte-copy struct words to slot+8+k.
				let lc: *local = localfindnode(c, src.str);
				let soff: i32 = 0;
				if (lc != nil) { soff = lc.off; };
				let stotal: i32 = si.totsize;
				let ki: i32 = 0;
				for (ki + 8 <= stotal) {
					emitline("\tMOVQ\t");
					emitoff((soff + ki): i64);
					emitline("(BP), AX\n");
					emitline("\tMOVQ\tAX, ");
					emitoff((slot_off + 8 + ki): i64);
					emitline("(BP)\n");
					ki += 8;
				};
				if (ki < stotal) {
					let tail: i32 = stotal - ki;
					let lop: str = "MOVQ";
					if (tail == 4) { lop = "MOVL"; }
					else { if (tail == 1) { lop = "MOVB"; }; };
					emitline("\t");
					emitline(lop);
					emitline("\t");
					emitoff((soff + ki): i64);
					emitline("(BP), AX\n");
					emitline("\t");
					emitline(lop);
					emitline("\tAX, ");
					emitoff((slot_off + 8 + ki): i64);
					emitline("(BP)\n");
				};
			};
			emitline("\tMOVQ\t$");
			emitint(tag: i64);
			emitline(", ");
			emitoff(slot_off: i64);
			emitline("(BP)\n");
			return;
		};
	};
	// str IS []u8 — same 32B payload as a slice: cgexpr leaves
	// (AX=ptr, BX=len, CX=cap); slot layout [+0]=tag, [+8]=ptr,
	// [+16]=len, [+24]=cap. str folds onto the slice arm (#1/Phase 3
	// collapse; cite cstage cg_widen_tagged_store).
	if (nodeisslice(c, src) || nodeisstr(c, src)) {
		cgexpr(c, src);
		emitline("\tMOVQ\tAX, ");
		emitoff((slot_off + 8): i64);
		emitline("(BP)\n");
		emitline("\tMOVQ\tBX, ");
		emitoff((slot_off + 16): i64);
		emitline("(BP)\n");
		emitline("\tMOVQ\tCX, ");
		emitoff((slot_off + 24): i64);
		emitline("(BP)\n");
		let tag: i32 = taggedvariantindext(c, dt, src);
		if (tag < 0) { tag = 0; };
		emitline("\tMOVQ\t$");
		emitint(tag: i64);
		emitline(", ");
		emitoff(slot_off: i64);
		emitline("(BP)\n");
		return;
	};
	// Float arm: cgexpr on an f64/f32 source leaves the bit pattern in
	// X0 only — the AX-store fallback below would silently write whatever
	// was loaded into AX before the SSE conversion. Literal `1.0` works
	// by coincidence (TK_FLOAT lowering loads the f64 bit pattern into AX
	// before MOVSD'ing into X0); every runtime f64 shape (cast, call,
	// unary, ident, struct-field load) needs the explicit MOVSD path.
	// Mirror of cstage cg_widen_tagged_store's float arm. Classify off
	// the checker stamp (src.type_) — the SSoT cstage reads via
	// node_isfloat / type_isf32 — and resolve the variant tag by name
	// directly: rhstargetname has no N_FLOATLIT / N_CALL / N_DOT branch
	// and would fall through to the str-shape fallback that picks tag 0
	// for an `(i64 | f64)` union. The armed asserttyped bail (check.ww)
	// guarantees src carries a non-nil stamp, so the sibling-evidence
	// loud-abort that used to pin "the float arm requires a stamped
	// value" is dead and removed.
	let fkind: i32 = 0;
	if (src != nil) {
		let srct: *tinfo = src.type_: *tinfo;
		if (typeisf32(srct)) { fkind = 1; }
		else { if (typeisfloat(srct)) { fkind = 2; }; };
	};
	if (fkind != 0) {
		let fmov: str = "MOVSD";
		if (fkind == 1) { fmov = "MOVSS"; };
		cgexpr(c, src);
		emitline("\t");
		emitline(fmov);
		emitline("\tX0, ");
		emitoff((slot_off + 8): i64);
		emitline("(BP)\n");
		// #66 Phase-N step 3: the float arm has no pattern node to ride
		// the typeeq flatvariantidx path, so pick the variant by float
		// kind (f32 vs f64) over tinfo.params — a shape classification
		// like the slice axis, not nominal identity.
		let wantf32: bool = (fkind == 1);
		let ftag: i32 = -1;
		let fti: *tinfo = dt;
		for (fti != nil && fti.kind == tykind.TY_NAMED) { fti = fti.under; };
		if (fti != nil) { if (fti.kind == tykind.TY_TAGGED) {
			let fp: *tparam = fti.params;
			let fidx: i32 = 0;
			for (fp != nil) {
				let fvt: *tinfo = fp.type_;
				for (fvt != nil && fvt.kind == tykind.TY_NAMED) { fvt = fvt.under; };
				if (fvt != nil) {
					if (typeisfloat(fvt)) {
						if (typeisf32(fvt) == wantf32) { ftag = fidx; break; };
					};
				};
				fp = fp.tnext;
				fidx += 1;
			};
		}; };
		if (ftag < 0) { ftag = 0; };
		emitline("\tMOVQ\t$");
		emitint(ftag: i64);
		emitline(", ");
		emitoff(slot_off: i64);
		emitline("(BP)\n");
		return;
	};
	// Scalar payload.
	cgexpr(c, src);
	emitline("\tMOVQ\tAX, ");
	emitoff((slot_off + 8): i64);
	emitline("(BP)\n");
	let tag: i32 = taggedvariantindext(c, dt, src);
	if (tag < 0) { tag = 0; };
	emitline("\tMOVQ\t$");
	emitint(tag: i64);
	emitline(", ");
	emitoff(slot_off: i64);
	emitline("(BP)\n");
	return;
};

// Spine-walk a chained N_DOT (n) inward to a root ident, summing field
// offsets through value-struct intermediates. Optional slice/str leaf
// pseudo-field (.ptr / .len / .cap) on the last segment is folded into
// *outslicedelta (0/8/16); otherwise *outleaftype is the leaf *tinfo
// and *outslicedelta stays -1. Returns true on success; on false the
// caller falls through to other branches.
//
// Mirrors cmd/w6c/cgen.c's N_DOT chained walker; both stages must agree
// on the same shapes so the bootstrap fixed-point holds. The chain
// depth is capped at 16 — deeper chains are vanishingly rare and fall
// through.
//
// On success the caller emits one load/store at root_base + *outtotaloff
// (+ slicedelta for pseudo leaf). Root resolves as: local frame slot
// (*outisglobal false, base = *outrootoff(BP)) or top-level let
// (*outisglobal true, base reached via LEAQ *outrootname(SB), CX).
//
// Numeric out-params are i32 — offsets fit naturally and the post-#19
// localloadop sign-extends i32 deref-stored slots on read, so negative
// frame offsets round-trip intact.
// #71 (A.6.3j): the spine offset-sum + leaf-type now read off
// tinfo.fields (natural layout) instead of the structinfo/fieldinfo
// walk. Mirror of cstage cmd/w6c/cgen.c:3156-3216 which walks
// `cur->lhs->type` fields. Root resolution (the local/ptr/global split
// and the rootoff/ptrroot/isglobal flags) stays on localfindnode/
// letvarstructinfo unchanged, so the firing set + addressing mode are
// byte-identical to the structinfo era; only the layout SOURCE moves.
// The slot-padded foff and the natural tfield.offset coincide for every
// shape the byte-id gate exercises (cstage already reads the natural
// offset), so this is offset-preserving. Leaf out-param is the field's
// stamped *tinfo (was *fieldinfo); the str/slice pseudo-leaf leaves it
// nil and the callers gate on slicedelta>=0 first.
export fn dotchainresolve(c: *cgen, n: *node,
    outrootname: *str, outrootoff: *i32, outtotaloff: *i32,
    outleaftype: **tinfo, outslicedelta: *i32,
    outisglobal: *bool, outptrroot: *bool) bool = {
    *outrootname = "";
    *outrootoff = 0;
    *outisglobal = false;
    *outptrroot = false;
    *outtotaloff = 0;
    *outleaftype = nil;
    *outslicedelta = -1;
    if (n == nil) { return false; };
    if (n.kind != nkind.N_DOT) { return false; };
    let stk: [16]*node;
    let nsteps: i32 = 0;
    let cur: *node = n;
    for (cur != nil) {
        if (cur.kind != nkind.N_DOT) { break; };
        if (nsteps >= 16) { return false; };
        stk[nsteps] = cur;
        nsteps += 1;
        cur = cur.lhs;
    };
    if (nsteps < 2) { return false; };
    if (cur == nil) { return false; };
    if (cur.kind != nkind.N_IDENT) { return false; };
    *outrootname = cur.str;
    let resolved: bool = false;
    let lc: *local = localfindnode(c, cur.str);
    if (lc != nil) {
        if (lc.tnode != nil) {
            if (lc.tnode.kind == nkind.N_TNAME) {
                *outrootoff = lc.off;
                resolved = true;
            };
            // `*T` root (param/local): dereference at emit time;
            // pointee struct supplies the field layout. Callers
            // that opt in via *outptrroot emit a MOVQ load of the
            // slot before indexing.
            if (lc.tnode.kind == nkind.N_TPTR) {
                let pe: *node = lc.tnode.lhs;
                if (pe != nil) {
                    if (pe.kind == nkind.N_TNAME) {
                        *outrootoff = lc.off;
                        *outptrroot = true;
                        resolved = true;
                    };
                };
            };
        };
    };
    if (!resolved) {
        let gsi: *structinfo = letvarstructinfo(c, cur.str);
        if (gsi != nil) {
            *outisglobal = true;
            resolved = true;
        };
    };
    if (!resolved) { return false; };
    // Root struct layout = the stamped root-ident type_, peeled NAMED
    // (plus one TY_PTR hop for a `*struct` root). tfield.type_ then
    // supplies each nested struct directly, so no name re-lookup.
    let curstruct: *tinfo = cur.type_: *tinfo;
    for (curstruct != nil && curstruct.kind == tykind.TY_NAMED) {
        curstruct = curstruct.under;
    };
    if (*outptrroot) {
        if (curstruct == nil) { return false; };
        if (curstruct.kind != tykind.TY_PTR) { return false; };
        curstruct = curstruct.sub;
        for (curstruct != nil && curstruct.kind == tykind.TY_NAMED) {
            curstruct = curstruct.under;
        };
    };
    let i: i32 = nsteps - 1;
    for (i >= 0) {
        if (curstruct == nil) { return false; };
        if (curstruct.kind != tykind.TY_STRUCT) { return false; };
        if (stk[i] == nil) { return false; };
        let stepnm: str = stk[i].str;
        let tf: *tfield = curstruct.fields;
        let found: *tfield = nil;
        for (tf != nil) {
            if (streq(tf.name, stepnm)) { found = tf; break; };
            tf = tf.tnext;
        };
        if (found == nil) { return false; };
        let foff: i32 = found.offset: i32;
        if (i == 0) {
            *outtotaloff = *outtotaloff + foff;
            *outleaftype = found.type_;
            return true;
        };
        let ft: *tinfo = found.type_;
        for (ft != nil && ft.kind == tykind.TY_NAMED) { ft = ft.under; };
        if (ft == nil) { return false; };
        if (ft.kind == tykind.TY_STR) {
            // str IS []u8: .cap is the third header word, same as
            // the TY_SLICE leaf below — cstage treats str≡slice for
            // .ptr/.len/.cap (cmd/w6c/cgen.c:2478) (#1/Phase 3, #11).
            if (i != 1) { return false; };
            let pseudo: str = stk[0].str;
            let delta: i32 = -1;
            if (streq(pseudo, "ptr")) { delta = 0; }
            else { if (streq(pseudo, "len")) { delta = 8; }
            else { if (streq(pseudo, "cap")) { delta = 16; }; }; };
            if (delta < 0) { return false; };
            *outtotaloff = *outtotaloff + foff;
            *outslicedelta = delta;
            return true;
        };
        if (ft.kind == tykind.TY_SLICE) {
            if (i != 1) { return false; };
            let pseudo: str = stk[0].str;
            let delta: i32 = -1;
            if (streq(pseudo, "ptr")) { delta = 0; }
            else { if (streq(pseudo, "len")) { delta = 8; }
            else { if (streq(pseudo, "cap")) { delta = 16; }; }; };
            if (delta < 0) { return false; };
            *outtotaloff = *outtotaloff + foff;
            *outslicedelta = delta;
            return true;
        };
        if (ft.kind != tykind.TY_STRUCT) { return false; };
        *outtotaloff = *outtotaloff + foff;
        curstruct = ft;
        i -= 1;
    };
    return false;
};

// cgstructlitfill — fill a struct-typed slot from an N_STRUCTLIT
// value into one of three destination flavors. Mirror of cstage
// cgen.c's cg_structlit_fill. Used by cglet, cgreturn N_STRUCTLIT,
// cgassign N_IDENT-lhs N_STRUCTLIT (BP-rel) AND cgassign N_DOT-lhs
// N_STRUCTLIT (BP-rel / via *struct local / via struct global) at
// single-dot and chained-dot sites.
//
// Destination modes:
//   0 = DST_BP        — base = BP, no reload. Stores at disp+i(BP).
//                       srcoff/srcname unused.
//   1 = DST_PTR_LOCAL — base = BX, reloaded from srcoff(BP) before
//                       the ELLIPSIS zero-fill loop and before EVERY
//                       field store (cgexpr clobbers BX between
//                       fields). Stores at disp+i(BX). srcname
//                       unused.
//   2 = DST_GLOBAL    — base = BX, reloaded via `LEAQ srcname(SB),
//                       BX` with the same cadence as DST_PTR_LOCAL.
//                       srcoff unused.
//
// Param semantics (locked in here so the recursion contract is
// clear):
//   - `disp` is the per-recursion accumulator — grows by `fi.foff`
//     as we descend into a nested struct-typed structlit field.
//   - `srcoff` (DST_PTR_LOCAL) and `srcname` (DST_GLOBAL) are
//     *constant* across the whole call tree — they identify the
//     root dst, which doesn't change with depth.
//   - the ELLIPSIS zero-fill extent is read internally as
//     structabisize(si) — cstage's cg_structlit_fill computes
//     `sz = lu->size` (cgen.c:2085), the maxalign-rounded ABI size
//     (check.c:760 lu->size = (off+maxalign-1)&~(maxalign-1)). The
//     pre-#169 callers passed two different sizes (natural at DOT
//     sites, slot-padded at BP-rel sites); neither matched cstage
//     for maxalign<8 structs (the zero-fill ran MOVQ where cstage
//     ran MOVL — value-correct, asm-divergent).
//
// Why a helper? The inline field-walk previously did
// `cgexpr(field.lhs); store AX sized`. For struct-typed fields whose
// value is itself a nested N_STRUCTLIT, cgexpr has no whole-struct-
// in-register convention — it lands AX = first qword and the
// trailing bytes silently stay zero. #17 fixed the BP-rel sites;
// #18 extends the same recursion to the four cgassign N_DOT-lhs
// structlit walks (single-dot via_ptr/global/local + chained
// depth>=2).
//
// The non-BP modes emit a redundant BX reload at the start of each
// recursive nested zero-fill / each recursive scalar store — this is
// correctness-by-construction (BX is always freshly loaded right
// before use), and the redundancy only fires on the nested-STRUCTLIT
// shapes that didn't compile before. Byte-identity for the no-
// nested case (the only shape selfhost source uses today) is
// preserved because the existing inline code's reload-before-each-
// store pattern matches the helper's per-store reload exactly.
//
// Graduation note (task #13): the scalar store currently uses the
// explicit {1→MOVB, 4→MOVL, else MOVQ} dispatch to match cstage
// byte-identically — cstage hasn't yet learned MOVW for fsz==2. Once
// #13 aligns both stages, the dispatch can switch to fieldstoreop
// which already returns MOVW where appropriate.
fn cgstructlitfill(c: *cgen, si: *structinfo, lit: *node,
    mode: i32, srcoff: i32, srcname: str,
    disp: i32) void = {
    if (si == nil) { return; };
    let basereg: str = "BP";
    if (mode != 0) { basereg = "BX"; };
    let totsize: i32 = structabisize(si);
    if (lit.op == tkind.TK_ELLIPSIS) {
        // `..., ...` autofill — zero the entire slot first so
        // unmentioned fields read as 0. Sized stores: 8/4/1. For
        // non-BP modes, reload BX once before the loop (cgexpr-free
        // region between iterations, so one reload is enough).
        emitline("\tXORQ\tAX, AX\n");
        if (mode == 1) {
            emitline("\tMOVQ\t");
            emitoff(srcoff: i64);
            emitline("(BP), BX\n");
        };
        if (mode == 2) {
            emitline("\tLEAQ\t");
            emitsymname(c, srcname);
            emitline("(SB), BX\n");
        };
        let zi: i32 = 0;
        for (zi + 8 <= totsize) {
            emitline("\tMOVQ\tAX, ");
            if (mode == 0) {
                emitoff((disp + zi): i64);
                emitline("(BP)\n");
            } else {
                emitdispreg((disp + zi): i64, basereg);
                emitline("\n");
            };
            zi += 8;
        };
        for (zi + 4 <= totsize) {
            emitline("\tMOVL\tAX, ");
            if (mode == 0) {
                emitoff((disp + zi): i64);
                emitline("(BP)\n");
            } else {
                emitdispreg((disp + zi): i64, basereg);
                emitline("\n");
            };
            zi += 4;
        };
        for (zi < totsize) {
            emitline("\tMOVB\tAX, ");
            if (mode == 0) {
                emitoff((disp + zi): i64);
                emitline("(BP)\n");
            } else {
                emitdispreg((disp + zi): i64, basereg);
                emitline("\n");
            };
            zi += 1;
        };
    };
    let fieldnode: *node = lit.list;
    for (fieldnode != nil) {
        if (fieldnode.kind == nkind.N_FIELD) {
            let fname: str = fieldnode.str;
            let fi: *fieldinfo = si.fields;
            for (fi != nil) {
                let fn_: str = fi.fname;
                if (streq(fn_, fname)) {
                    // Tagged-union field: delegate to the shared
                    // widening writer (handles str/scalar/struct
                    // literal/ident payload + tagged-subset tag
                    // remap). For non-BP modes, reload BX first so
                    // the widener sees a valid base reg.
                    if (istaggedtype(c, fi.tnode)) {
                        if (mode == 1) {
                            emitline("\tMOVQ\t");
                            emitoff(srcoff: i64);
                            emitline("(BP), BX\n");
                        };
                        if (mode == 2) {
                            emitline("\tLEAQ\t");
                            emitsymname(c, srcname);
                            emitline("(SB), BX\n");
                        };
                        cgwidentaggedstore(c, fi.tnode.type_: *tinfo,
                            fieldnode.lhs, basereg,
                            disp + fi.foff, fi.fsz);
                        fi = nil;
                    } else {
                    // Nested struct-typed structlit value: look up
                    // the inner struct's metadata and recurse at the
                    // field's offset. Pre-#17/#18 the cgexpr-then-
                    // store below would land AX = first qword and
                    // the rest silently stayed zero.
                    let nested: bool = false;
                    if (fieldnode.lhs != nil) {
                        if (fieldnode.lhs.kind == nkind.N_STRUCTLIT) {
                            if (fi.tnode != nil) {
                                if (fi.tnode.kind == nkind.N_TNAME) {
                                    if (primsize(fi.tnode.str) == 0) {
                                        let isi: *structinfo = structlookup(c, fi.tnode.str);
                                        if (isi != nil) {
                                            cgstructlitfill(c, isi,
                                                fieldnode.lhs,
                                                mode, srcoff, srcname,
                                                disp + fi.foff);
                                            nested = true;
                                        };
                                    };
                                };
                            };
                        };
                    };
                    // Nested struct-typed CALL value (#20). cgexpr
                    // leaves AX=bytes[0..7], DX=bytes[8..15], CX=
                    // bytes[16..23] per #4's cgreturn ABI. Pre-#20
                    // the cgexpr-then-AX-store fallthrough below
                    // silently dropped past the first qword for any
                    // fsz > 8 (only AX got stored).
                    //
                    // Sized stores: MOVQ for full 8B chunks plus a
                    // sized tail (MOVL/MOVW/MOVB) by `tail = fsz%8`.
                    // Mirror of cstage cg_structlit_fill's #20 branch.
                    // MOVW-for-tail==2 only fires on shapes that
                    // didn't compile before, so no #13 byte-identity
                    // concern.
                    //
                    // Guard `fsz <= 24 && fsz%8 ∈ {0,1,2,4}` matches
                    // #4's cgreturn ABI: >24B falls through (sret
                    // deferred); fsz%8 ∈ {3,5,6,7} would need shift-
                    // store and is also unsupported by #4 — falls
                    // through to the existing AX-only wrongness
                    // (consistent, tracked as follow-up).
                    //
                    // INVARIANT: between cgexpr(N_CALL) and the
                    // AX/DX/CX stores below, NO instruction may touch
                    // AX/DX/CX. The BX reload is safe; any other
                    // emission added here will silently corrupt the
                    // return value.
                    let callwhole: bool = false;
                    if (!nested) {
                        if (fieldnode.lhs != nil) {
                            if (fieldnode.lhs.kind == nkind.N_CALL) {
                                if (fi.tnode != nil) {
                                    if (fi.tnode.kind == nkind.N_TNAME) {
                                        if (primsize(fi.tnode.str) == 0) {
                                            let csi: *structinfo = structlookup(c, fi.tnode.str);
                                            if (csi != nil) {
                                                // Inner struct's ABI size
                                                // (maxalign-rounded) — cstage
                                                // reads fl->type->size at the
                                                // nested-call branch
                                                // (cgen.c:2121); check.c:760
                                                // sets that to the
                                                // maxalign-rounded ABI extent
                                                // (NOT natural). fi.fsz is
                                                // wwstage's slot-padded
                                                // totsize (round-to-8); the
                                                // pre-#169 structnaturalsize
                                                // shorts struct{i64,i32}
                                                // (natural 12, ABI 16) to
                                                // MOVQ+MOVL where cstage
                                                // writes MOVQ+MOVQ.
                                                let cfsz: i32 = structabisize(csi);
                                                let crem: i32 = cfsz - (cfsz / 8) * 8;
                                                if (cfsz <= 24) {
                                                if (crem == 0 || crem == 1
                                                    || crem == 2 || crem == 4) {
                                                    cgexpr(c, fieldnode.lhs);
                                                    if (mode == 1) {
                                                        emitline("\tMOVQ\t");
                                                        emitoff(srcoff: i64);
                                                        emitline("(BP), BX\n");
                                                    };
                                                    if (mode == 2) {
                                                        emitline("\tLEAQ\t");
                                                        emitsymname(c, srcname);
                                                        emitline("(SB), BX\n");
                                                    };
                                                    let full: i32 = cfsz / 8;
                                                    let ci: i32 = 0;
                                                    for (ci < full) {
                                                        let r: str = "AX";
                                                        if (ci == 1) { r = "DX"; };
                                                        if (ci == 2) { r = "CX"; };
                                                        emitline("\tMOVQ\t");
                                                        emitline(r);
                                                        emitline(", ");
                                                        if (mode == 0) {
                                                            emitoff((disp + fi.foff + ci * 8): i64);
                                                            emitline("(BP)\n");
                                                        } else {
                                                            emitdispreg((disp + fi.foff + ci * 8): i64, basereg);
                                                            emitline("\n");
                                                        };
                                                        ci += 1;
                                                    };
                                                    if (crem > 0) {
                                                        let top: str = "MOVB";
                                                        if (crem == 4) { top = "MOVL"; };
                                                        if (crem == 2) { top = "MOVW"; };
                                                        let tr: str = "AX";
                                                        if (full == 1) { tr = "DX"; };
                                                        if (full == 2) { tr = "CX"; };
                                                        emitline("\t");
                                                        emitline(top);
                                                        emitline("\t");
                                                        emitline(tr);
                                                        emitline(", ");
                                                        if (mode == 0) {
                                                            emitoff((disp + fi.foff + full * 8): i64);
                                                            emitline("(BP)\n");
                                                        } else {
                                                            emitdispreg((disp + fi.foff + full * 8): i64, basereg);
                                                            emitline("\n");
                                                        };
                                                    };
                                                    callwhole = true;
                                                };
                                                };
                                            };
                                        };
                                    };
                                };
                            };
                        };
                    };
                    if (nested) {
                        fi = nil;
                    } else if (callwhole) {
                        fi = nil;
                    } else if (isstrtype(c, fi.tnode)) {
                        // str IS []u8: 3-word field (ptr,len,cap).
                        // cgexpr leaves AX/BX/CX; for non-BP modes
                        // the dst base goes in DX to dodge BX=len /
                        // CX=cap (the generic store reloads BX, which
                        // would clobber len) (#1/Phase 3).
                        cgexpr(c, fieldnode.lhs);
                        if (mode == 0) {
                            emitline("\tMOVQ\tAX, ");
                            emitoff((disp + fi.foff): i64);
                            emitline("(BP)\n");
                            emitline("\tMOVQ\tBX, ");
                            emitoff((disp + fi.foff + 8): i64);
                            emitline("(BP)\n");
                            emitline("\tMOVQ\tCX, ");
                            emitoff((disp + fi.foff + 16): i64);
                            emitline("(BP)\n");
                        } else {
                            if (mode == 1) {
                                emitline("\tMOVQ\t");
                                emitoff(srcoff: i64);
                                emitline("(BP), DX\n");
                            } else {
                                emitline("\tLEAQ\t");
                                emitsymname(c, srcname);
                                emitline("(SB), DX\n");
                            };
                            emitline("\tMOVQ\tAX, ");
                            emitdispreg((disp + fi.foff): i64, "DX");
                            emitline("\n");
                            emitline("\tMOVQ\tBX, ");
                            emitdispreg((disp + fi.foff + 8): i64, "DX");
                            emitline("\n");
                            emitline("\tMOVQ\tCX, ");
                            emitdispreg((disp + fi.foff + 16): i64, "DX");
                            emitline("\n");
                        };
                        fi = nil;
                    } else {
                        cgexpr(c, fieldnode.lhs);
                        // For non-BP modes, cgexpr just clobbered
                        // BX; reload it before the store.
                        if (mode == 1) {
                            emitline("\tMOVQ\t");
                            emitoff(srcoff: i64);
                            emitline("(BP), BX\n");
                        };
                        if (mode == 2) {
                            emitline("\tLEAQ\t");
                            emitsymname(c, srcname);
                            emitline("(SB), BX\n");
                        };
                        if (isfloattype(c, fi.tnode)) {
                            let mov: str = "MOVSD";
                            if (isf32type(c, fi.tnode)) { mov = "MOVSS"; };
                            emitline("\t");
                            emitline(mov);
                            emitline("\tX0, ");
                            if (mode == 0) {
                                emitoff((disp + fi.foff): i64);
                                emitline("(BP)\n");
                            } else {
                                emitdispreg((disp + fi.foff): i64, basereg);
                                emitline("\n");
                            };
                            fi = nil;
                        } else {
                            // Explicit {1→MOVB, 4→MOVL, else MOVQ}
                            // dispatch (not fieldstoreop) to match
                            // cstage byte-identically. wwstage's
                            // fieldstoreop would return MOVW for
                            // fsz==2 which cstage doesn't emit —
                            // tracked as task #13.
                            let fsz: i32 = fi.fsz;
                            let op: str = "MOVQ";
                            if (fsz == 1) { op = "MOVB"; };
                            if (fsz == 4) { op = "MOVL"; };
                            emitline("\t");
                            emitline(op);
                            emitline("\tAX, ");
                            if (mode == 0) {
                                emitoff((disp + fi.foff): i64);
                                emitline("(BP)\n");
                            } else {
                                emitdispreg((disp + fi.foff): i64, basereg);
                                emitline("\n");
                            };
                            fi = nil;
                        };
                    };
                    };
                } else {
                    fi = fi.finext;
                };
            };
        };
        fieldnode = fieldnode.next;
    };
};

// Thin wrapper preserving the BP-rel call shape used by cglet,
// cgreturn, and cgassign N_IDENT-lhs N_STRUCTLIT.
fn cgstructlitfillbp(c: *cgen, si: *structinfo, lit: *node, bpoff: i32) void = {
    if (si == nil) { return; };
    cgstructlitfill(c, si, lit, 0, 0, "", bpoff);
};