toolchain: banner purge + WHY-only comment sweep (rule 8)

selfhost/, cmd/, internal/ join the tree-wide sweep: every section banner dies (91 selfhost + the cmd C-style dividers -> 0); narration and stale contracts deleted (pre-#22 bundler notes, retired single-PT_LOAD and no-archive claims, superseded ABI tables); every ref/harec/qbe cite, task cite, encoding/ELF contract, and rule-10 twin pointer kept; lost lifetime/rationale lines restored where the sweep over-cut (elf_globals ownership, kwtab linear-scan). Comment- only proven: all five wwstage tool binaries byte-identical across the sweep; test-commit, test-byteid (161+1399, 0 pinned-divergent), and test-bootstrap (fixed point + 991-995 byte-id) all exit 0. The read-through banked 66 latent-bug leads (checkpoint).
2026-08-08 23:14:03 +09:00
parent 83f5956df2
commit 62b9d20383
60 changed files with 232 additions and 1045 deletions
--- a/selfhost/cmd/wcc/cgen.ww
+++ b/selfhost/cmd/wcc/cgen.ww
@@ -1,26 +1,4 @@
-// selfhost/cmd/wcc/cgen.ww — port of cmd/w6c/cgen.c.
-//
-// Status: GROWING. Each subsystem we add is verified by `wwdump_ww -c`
-// producing byte-identical output to C-side `w6c` for the same source,
-// then by assembling + linking + running the result.
-//
-// Current coverage:
-//   - decls:  nkind.N_FILE, nkind.N_FNDECL (params, frame for locals, prologue
-//             + dual-epilogue suppression; FFI body-less fn skipped)
-//   - stmts:  nkind.N_BLOCK, nkind.N_RETURN, nkind.N_EXPRSTMT, nkind.N_LET (no init),
-//             nkind.N_LET (int-literal / ident / call / nkind.N_BIN init),
-//             nkind.N_IF (with optional else), nkind.N_FOR (cond-only and full
-//             init/cond/post), nkind.N_BREAK, nkind.N_CONTINUE
-//   - exprs:  nkind.N_INTLIT, nkind.N_IDENT (local/param), nkind.N_BIN with full op
-//             coverage (+/-/*/// %, &/|/^, <</>>, comparisons with
-//             signed-vs-unsigned dispatch, &&/||), nkind.N_UN (- ! ~ &amp; *),
-//             nkind.N_CALL (recursive R-to-L push, pop into argregs L-to-R),
-//             nkind.N_ASSIGN to local idents (plain and compound +=/-=)
-//
-// Type info is shallow — frame slots are 8 bytes per local, all loads
-// /stores are MOVQ. Programs that mix i8/i32/i64 locals work but spill
-// 8 bytes per local. Float, str, slice, struct, match, defer, alloc,
-// tagged-union return — none of those are wired yet.
+// Port of cmd/w6c/cgen.c.

 package wcc;

@@ -38,12 +16,8 @@ import cgenexpr;
 import cgenstmt;
 import cgendecl;

-// ---- typedef alias registry -----------------------------------------
-//
-// `type error = str;` makes `error` a struct-shape alias. We track
-// alias→target so isstrtype / isslicetype / structlookup can
-// resolve through the chain. Only direct nkind.N_TNAME aliases are mapped;
-// `type p = struct {...}` is handled by collectstructs.
+// Only direct nkind.N_TNAME aliases are mapped; `type p = struct {...}`
+// is handled by collectstructs.

 type aliasent = struct {
 	aname:   str,
@@ -156,12 +130,9 @@ fn aliassamemod(c: *cgen, name: str) *syntax.node = {
 	return nil;
 };

-// ---- enum registry --------------------------------------------------
-//
-// Mirrors cmd/wcc/check.c's enum resolution at collect time: walk
-// every `type Foo = enum [storage] { ... }`, pre-compute each
-// member's u64 value (supporting auto-increment and sibling refs),
-// and stash them so cgdot can fold `Foo.MEMBER` → MOVQ $value, AX.
+// Enum member values are pre-computed at collect time (auto-increment
+// + sibling refs) so cgdot can fold `Foo.MEMBER` → MOVQ $value, AX.
+// Mirrors cmd/wcc/check.c's enum resolution.

 // foldintliteral — fold the literal subset usable for top-level
 // constant slots: int/rune literal, true/false/nil, and a unary
@@ -376,13 +347,6 @@ fn resolvetype(c: *cgen, t: *syntax.node) *syntax.node = {
 	return cur;
 };

-// ---- struct registry ------------------------------------------------
-//
-// Per-file map from struct name → list of fields with computed offsets
-// and sizes. Built when cgfile walks nkind.N_TYPEDECL with nkind.N_TSTRUCT lhs.
-// nkind.N_DOT and nkind.N_ASSIGN consult this to resolve `s.field` for struct or
-// *struct bases.
-
 type fieldinfo = struct {
 	fname: str,
 	foff:  i32,
@@ -399,8 +363,6 @@ type structinfo = struct {
 	sinext:   *structinfo,
 };

-// ---- locals / frame --------------------------------------------------
-
 type local = struct {
 	name:  str,
 	off:   i32,
@@ -764,8 +726,6 @@ fn localfind(c: *cgen, name: str) i32 = {
 	return 0;
 };

-// ---- emit helpers ---------------------------------------------------
-
 // Cgfn defers its prologue (TEXT / SUBQ) until after the body so the
 // frame size reflects every emit-time localadd — the scanlocals pre-
 // pass that previously pre-computed it was dropped per #15/#26c. The
@@ -942,10 +902,6 @@ fn mkscratchname(c: *cgen, prefix: str) str = {
 	return r;
 };

-// ---- string interning ------------------------------------------------
-//
-// streq is provided by sym.ww and reused here.
-
 // internstrlit — return a stable label for `bytes`. Dedups by content
 // so identical literals share storage.
 fn internstrlit(c: *cgen, bytes: str) str = {
@@ -1008,8 +964,6 @@ fn letscalarprim(nm: str) bool = {
 	return false;
 };

-// letfloatprim — float type-name keywords. f32 → 4B slot, f64 → 8B.
-// Returns the slot size or 0 if not a float type.
 fn letfloatprim(nm: str) i32 = {
 	if (syntax.streq(nm, "f32")) { return 4; };
 	if (syntax.streq(nm, "f64")) { return 8; };
@@ -1019,11 +973,6 @@ fn letfloatprim(nm: str) i32 = {
 // letemitsize — slot size in bytes for a top-level `let`, or 0 if
 // the type isn't yet supported as a writable global. Walks type
 // aliases so byte output matches C cgen, which resolves Type kinds.
-//   4       → f32    (literal init supported)
-//   8       → scalar or f64 (literal init supported)
-//   16      → str    (only zero-init / nil / "" supported)
-//   24      → slice  (only zero-init supported)
-//   varies  → struct (zero-init only; field reads/scalar-field writes)
 fn letemitsize(c: *cgen, d: *syntax.node) i32 = {
 	if (d == nil) { return 0; };
 	let t: *syntax.node = d.lhs;
@@ -2024,7 +1973,6 @@ fn emitarraylitbytes(c: *cgen, arrt: *syntax.tinfo, rhs: *syntax.node,
 	};

 	if (eu != nil && eu.kind == syntax.tykind.TY_STRUCT) {
-		// Validate: every element must be N_STRUCTLIT (after N_CAST).
 		let idx: i32 = 0;
 		let last_ev: *syntax.node = nil;
 		let e: *syntax.node = rhs.list;
@@ -2113,7 +2061,6 @@ fn emitarraylitbytes(c: *cgen, arrt: *syntax.tinfo, rhs: *syntax.node,

 	if (syntax.typeisfloat(au.sub)) {
 		let isf32: bool = syntax.typeisf32(au.sub);
-		// Validate.
 		let idx: i32 = 0;
 		let e: *syntax.node = rhs.list;
 		for (e != nil && idx < alen) {
@@ -2211,7 +2158,6 @@ fn emitarraylitbytes(c: *cgen, arrt: *syntax.tinfo, rhs: *syntax.node,
 	let e: *syntax.node = rhs.list;
 	let last: u64 = 0u64;
 	let repeat: bool = false;
-	// Validate first.
 	for (e != nil && idx < alen) {
 		if (e.kind == syntax.nkind.N_FIELD) {
 			if (syntax.streq(e.str, "...")) { repeat = true; break; };
@@ -2551,7 +2497,6 @@ fn emitslicedata(c: *cgen, name: str, module: str, slt: *syntax.tinfo,
 			os.write(2, m.ptr, m.len: u64);
 			os.exit(1);
 		};
-		// Writable backing data.
 		emitline("DATAW ");
 		emitfnname(c, name, module);
 		emitline(".d(SB),\"");
@@ -3494,12 +3439,9 @@ fn emitdatasection(c: *cgen) void = {
 	};
 };

-// ---- fn return-type map ---------------------------------------------
-//
-// Per-file: ident → ret-type-node. Used to decide whether to shuffle
-// (AX, DX) → (AX, BX) after a CALL — needed for str-returning fns so
-// the value flows through cgen as the canonical (AX, BX) str pair.
-
+// fnret decides whether to shuffle (AX, DX) → (AX, BX) after a CALL —
+// needed for str-returning fns so the value flows through cgen as the
+// canonical (AX, BX) str pair.
 type fnret = struct {
 	fname: str,
 	fmod:  str,
@@ -3670,12 +3612,8 @@ fn fnparamslookupmod(c: *cgen, name: str, mod: str) *syntax.node = {
 	return fnparamslookup(c, name);
 };

-// ---- def-constant registry ------------------------------------------
-//
 // `def NAME: T = LIT;` becomes a DATA symbol the C-side w6c emits; an
-// ident reference loads it via `MOVQ NAME(SB), AX`. We collect them at
-// file load and consult on nkind.N_IDENT lookup.
-
+// ident reference loads it via `MOVQ NAME(SB), AX`.
 type defent = struct {
 	dname: str,
 	dmod:  str,	// originating module (`// MODULE: foo`), or empty
@@ -3819,8 +3757,6 @@ fn defisaddressable(c: *cgen, opnd: *syntax.node) bool = {
 	return false;
 };

-// ---- module-private symbol map --------------------------------------
-//
 // Every non-FFI top-level fn decl lives in its module's namespace —
 // cgen mangles the leaf to `<module>.<name>` at the def site (TEXT)
 // and at every call/load site, so cross-module same-leaf fns (lib/os
@@ -4046,8 +3982,6 @@ fn emitfnname(c: *cgen, ident: str, hint: str) void = {
 	emitbytes( ident.ptr, ident.len: u64);
 };

-// ---- FFI map ---------------------------------------------------------
-
 fn fficollect(c: *cgen, file: *syntax.node) void = {
 	c.ffis = nil;
 	if (file == nil) { return; };
@@ -4085,8 +4019,6 @@ fn ffiresolve(c: *cgen, ident: str) str = {
 	return ident;
 };

-// ---- ABI argreg helpers ---------------------------------------------
-
 fn argregname(i: i32) str = {
 	if (i == 0) { return "DI"; };
 	if (i == 1) { return "SI"; };