w6c+wwstage: chained-base array-field address via dotbaseaddr — close the family (#253)

cg_dotbase_addr / dotbaseaddr rejected a non-ident inner, so a chained
base (`o.p.m[i]` / `o.i.m[i]` / `o.a.b.m[i]`) fell to cgexpr(base) which
auto-derefs the array field's first 8 bytes AS a pointer -> garbage base
-> segfault (base64 fillobuf `s.enc.encmap[...]` blocker). Extend the one
helper per stage to accept a chained inner: a new cg_dotchain_addr /
dotchainaddr recovers the container base via the dot-chain spine (recurse
to &x, deref when x is a *struct, sum field offsets), keeping the same
no-AX/no-stack spill contract. dotbaseaddr then takes the pointer VALUE of
inner when viaptr, else its ADDRESS, and adds the field offset. One fix
closes every op (index r/w, addr-of, slice, compound) since all route
through the helper. Symmetric cs==ww byte-id.

test/949: +22 rows. Chained-PTR (rd/wr/addr/slice x2/compound), deeper
(value+ptr leaf links, triple-pointer exercising the internal deref),
non-u8 esz stride (i32 addr+slice), and single-level controls — all
byte-id. The chained VALUE-container arm (`o.i.m`) is run-only (byteid=0):
it needs a value nested-struct instance, which trips THREE orthogonal
pre-existing cs!=ww emission divergences (bare-let zero-init policy,
global DATAW byte count, i32 element-load opcode in the index fallback)
unrelated to #253. Run correctness proves the segfault is gone for that
cell; byte-id there awaits the separate wwstage value-nested-struct fix.
This commit is contained in:
2026-06-02 03:44:49 +09:00
parent 7e3271bf01
commit 585ec50676
5 changed files with 713 additions and 107 deletions

View File

@@ -1,27 +1,34 @@
/*
* 949_dotbase_addr_slice_run — runtime + byte-id net for #252, the
* addr-of + slice SIBLING of #135 (949_dotbase_arr_run covers the
* read/write index path). Taking `&x.o[i]` (address-of an element) or
* slicing `x.o[lo:hi]` / `x.o[lo:]` of a struct's `[N]T`-typed FIELD
* computed the field's VALUE as the base address instead of its
* ADDRESS: cgen emitted `MOVL off(BP),AX` (load the field's first
* 8 bytes as a pointer) where it must emit `LEAQ off(BP),AX` (the
* field's address) -> garbage pointer -> SEGFAULT. The index
* read/write path was fixed in #135; the addr-of N_INDEX "complex
* base" arm and the N_SLICE base arm still fell to the generic
* cgexpr(base) auto-deref. cs==ww BOTH stages broken identically
* 949_dotbase_addr_slice_run — runtime + byte-id net for the array-
* field-base-address family: #252 (the addr-of + slice SIBLING of #135)
* and #253 (the CHAINED-base close-out). 949_dotbase_arr_run covers the
* single-level read/write index path.
*
* #252: taking `&x.o[i]` (address-of an element) or slicing
* `x.o[lo:hi]` / `x.o[lo:]` of a struct's `[N]T`-typed FIELD computed
* the field's VALUE as the base address instead of its ADDRESS: cgen
* emitted `MOVL off(BP),AX` (load the field's first 8 bytes as a
* pointer) where it must emit `LEAQ off(BP),AX` (the field's address)
* -> garbage pointer -> SEGFAULT. The index read/write path was fixed
* in #135; the addr-of N_INDEX "complex base" arm and the N_SLICE base
* arm still fell to the generic cgexpr(base) auto-deref.
*
* #253: the same family with a CHAINED base — the inner is itself an
* N_DOT (`o.p.m[i]` / `o.i.m[i]` / `o.a.b.m[i]`), not a bare ident.
* `cg_dotbase_addr` / `dotbaseaddr` rejected a non-ident inner, so the
* caller fell to cgexpr(base) which auto-derefs the array field's first
* 8 bytes AS a pointer -> garbage base -> SEGFAULT (base64 fillobuf
* `s.enc.encmap[...]` blocker). The fix recovers the container base via
* the dot-chain spine (`cg_dotchain_addr` / `dotchainaddr`): the pointer
* VALUE of inner when inner is a *struct, else the ADDRESS of inner,
* then adds the field offset. One helper extension per stage closes the
* whole family — every op (index r/w, addr-of, slice, compound) routes
* through the same helper. cs==ww BOTH stages broken identically
* pre-fix (gate-blind, pure correctness — not a byte-id divergence).
*
* Fix wires the same `cg_dotbase_addr` (cstage) / `dotbaseaddr`
* (wwstage) helper #135 introduced into those two base-address paths
* (guarded `if(!cg_dotbase_addr(...)) cgexpr(base)`). For the slice
* path the element stride (esz) and default-hi length are extended to
* an N_DOT array-field base too (read from the field's element tinfo /
* array length via the type table — rule-13), so non-u8 element slices
* scale correctly and `s.obuf[lo:]` gets the array's element count.
*
* Rows (cstage `ww build` + run for exit code; w6c vs w6c_ww `.s` cmp
* for rule-10 byte-id):
* Byte-id rows (cstage `ww build` + run for exit code; w6c vs w6c_ww
* `.s` cmp for rule-10 byte-id):
* #252 (bare-ident base):
* - addr_local_u8 &x.o[1] on a local value-struct, *p read → 66
* - addr_ptr_u8 &x.o[2] via a *struct param, *p read → 77
* - addr_i32 &x.o[2] on [4]i32 field, *p read (esz=4) → 88
@@ -31,10 +38,40 @@
* - slice_i32_expl [4]i32 field x.o[1:3], s[1] read (esz=4) → 88
* - slice_i32_dflt [4]i32 field x.o[1:], s[2] read (esz=4) → 55
* - control_bare bare-local [4]u8 &a[1] write + a[1:4] read → 44
* #253 (chained base — *struct-field-pointer / deeper / non-u8):
* - chain_ptr_rd o.p.m[1] read (p:*inner field) → 66
* - chain_ptr_wr o.p.m[2] write, read back via a.m[2] → 77
* - chain_ptr_addr &o.p.m[2] then *q read → 55
* - chain_ptr_sl_e o.p.m[1:4] explicit hi, s[0] → 66
* - chain_ptr_sl_d o.p.m[1:] default hi, s[0] → 66
* - chain_ptr_comp o.p.m[1] += v compound → 66
* - chain_deep_lf o.a.b.m[1] read (a value, b:*inner leaf) → 66
* - chain_triple o.p.q.m[1] read (two ptr links: dotchain → 66
* internal deref, pointer-only structs)
* - chain_tri_comp o.p.q.m[1] += v through the triple chain → 66
* - chain_i32_addr &o.p.m[2] on [4]i32 (esz=4 stride) → 88
* - chain_i32_slice o.p.m[1:3] on [4]i32, s[1] (esz=4) → 88
* - ctrl_ptr_rd p.m[1] read via *e param (control) → 66
* - ctrl_local_rd x.m[1] read, value-local field (control) → 66
*
* The bare-local control asserts the N_IDENT base paths (untouched by
* this fix) still emit correct code; the non-u8 rows assert the esz
* stride extension is wired (not silently esz=1).
* Run-only rows (byteid=0): the chained VALUE-container arm (`o.i.m`,
* inner is a value nested struct). These exercise the same fixed helper
* and run correctly, but a value nested-struct instance trips THREE
* orthogonal pre-existing cs!=ww divergences unrelated to #253 — bare-
* let zero-init policy (wwstage emits an extra `MOVQ $0,off(BP)`),
* global DATAW byte count (wwstage over-emits), and the i32 element-
* LOAD opcode in the index fallback (cstage MOVSXD vs wwstage MOVL) —
* so the rule-10 byte-id gate can't apply here until those are fixed
* (filed: wwstage value-nested-struct emission divergence). Run
* correctness alone proves the #253 segfault is gone for this cell.
* - chain_val_rd o.i.m[1] read (i value nested) → 66
* - chain_val_addr &o.i.m[2] then *q read → 55
* - chain_val_slice o.i.m[1:4], s[0] → 66
* - chain_deep_val o.a.b.m[1] read, full value chain → 66
*
* The bare-local control asserts the N_IDENT base paths still emit
* correct code; the non-u8 rows assert the esz stride extension is
* wired (not silently esz=1).
*/
#include <stdio.h>
#include <stdlib.h>
@@ -52,7 +89,7 @@ runwait(const char *cmd)
return -1;
}
struct row { const char *label; const char *src; int want_exit; };
struct row { const char *label; const char *src; int want_exit; int byteid; };
static const struct row rows[] = {
{ "addr_local_u8",
@@ -63,7 +100,7 @@ static const struct row rows[] = {
" x.o[1] = 66u8;\n"
" let p: *u8 = &x.o[1];\n"
" return (*p): i32;\n"
"};\n", 66 },
"};\n", 66, 1 },
{ "addr_ptr_u8",
"package main;\n"
"type e = struct { o: [4]u8 };\n"
@@ -72,7 +109,7 @@ static const struct row rows[] = {
" let x: e;\n"
" x.o[2] = 77u8;\n"
" return rd(&x): i32;\n"
"};\n", 77 },
"};\n", 77, 1 },
{ "addr_i32",
"package main;\n"
"type e = struct { o: [4]i32 };\n"
@@ -81,7 +118,7 @@ static const struct row rows[] = {
" x.o[2] = 88;\n"
" let p: *i32 = &x.o[2];\n"
" return *p;\n"
"};\n", 88 },
"};\n", 88, 1 },
{ "slice_u8_expl",
"package main;\n"
"type e = struct { o: [4]u8 };\n"
@@ -90,7 +127,7 @@ static const struct row rows[] = {
" x.o[1] = 66u8;\n"
" let s: []u8 = x.o[1:4];\n"
" return s[0]: i32;\n"
"};\n", 66 },
"};\n", 66, 1 },
{ "slice_u8_dflthi",
"package main;\n"
"type e = struct { o: [4]u8 };\n"
@@ -99,7 +136,7 @@ static const struct row rows[] = {
" x.o[1] = 66u8;\n"
" let s: []u8 = x.o[1:];\n"
" return s[0]: i32;\n"
"};\n", 66 },
"};\n", 66, 1 },
{ "slice_ptr_u8",
"package main;\n"
"type e = struct { o: [4]u8 };\n"
@@ -108,7 +145,7 @@ static const struct row rows[] = {
" let x: e;\n"
" x.o[1] = 66u8;\n"
" return sl(&x): i32;\n"
"};\n", 66 },
"};\n", 66, 1 },
{ "slice_i32_expl",
"package main;\n"
"type e = struct { o: [4]i32 };\n"
@@ -118,7 +155,7 @@ static const struct row rows[] = {
" x.o[2] = 88;\n"
" let s: []i32 = x.o[1:3];\n"
" return s[1];\n"
"};\n", 88 },
"};\n", 88, 1 },
{ "slice_i32_dflt",
"package main;\n"
"type e = struct { o: [4]i32 };\n"
@@ -127,7 +164,7 @@ static const struct row rows[] = {
" x.o[3] = 55;\n"
" let s: []i32 = x.o[1:];\n"
" return s[2];\n"
"};\n", 55 },
"};\n", 55, 1 },
{ "control_bare",
"package main;\n"
"export fn main() i32 = {\n"
@@ -137,8 +174,176 @@ static const struct row rows[] = {
" *p = 33u8;\n"
" let s: []u8 = a[1:4];\n"
" return s[1]: i32;\n"
"};\n", 44 },
{ NULL, NULL, 0 }
"};\n", 44, 1 },
/* #253 chained-base rows. inner/outer types declared per-row so
* each source is self-contained. */
{ "chain_ptr_rd",
"package main;\n"
"type inner = struct { m: [4]u8 };\n"
"type outer = struct { p: *inner };\n"
"export fn main() i32 = {\n"
" let a: inner; a.m[1] = 66u8;\n"
" let o: outer; o.p = &a;\n"
" return o.p.m[1]: i32;\n"
"};\n", 66, 1 },
{ "chain_ptr_wr",
"package main;\n"
"type inner = struct { m: [4]u8 };\n"
"type outer = struct { p: *inner };\n"
"export fn main() i32 = {\n"
" let a: inner;\n"
" let o: outer; o.p = &a;\n"
" o.p.m[2] = 77u8;\n"
" return a.m[2]: i32;\n"
"};\n", 77, 1 },
{ "chain_ptr_addr",
"package main;\n"
"type inner = struct { m: [4]u8 };\n"
"type outer = struct { p: *inner };\n"
"export fn main() i32 = {\n"
" let a: inner; a.m[2] = 55u8;\n"
" let o: outer; o.p = &a;\n"
" let q: *u8 = &o.p.m[2];\n"
" return (*q): i32;\n"
"};\n", 55, 1 },
{ "chain_ptr_sl_e",
"package main;\n"
"type inner = struct { m: [4]u8 };\n"
"type outer = struct { p: *inner };\n"
"export fn main() i32 = {\n"
" let a: inner; a.m[1] = 66u8;\n"
" let o: outer; o.p = &a;\n"
" let s: []u8 = o.p.m[1:4];\n"
" return s[0]: i32;\n"
"};\n", 66, 1 },
{ "chain_ptr_sl_d",
"package main;\n"
"type inner = struct { m: [4]u8 };\n"
"type outer = struct { p: *inner };\n"
"export fn main() i32 = {\n"
" let a: inner; a.m[1] = 66u8;\n"
" let o: outer; o.p = &a;\n"
" let s: []u8 = o.p.m[1:];\n"
" return s[0]: i32;\n"
"};\n", 66, 1 },
{ "chain_ptr_comp",
"package main;\n"
"type inner = struct { m: [4]u8 };\n"
"type outer = struct { p: *inner };\n"
"export fn main() i32 = {\n"
" let a: inner; a.m[1] = 60u8;\n"
" let o: outer; o.p = &a;\n"
" o.p.m[1] += 6u8;\n"
" return o.p.m[1]: i32;\n"
"};\n", 66, 1 },
{ "chain_deep_lf",
"package main;\n"
"type inner = struct { m: [4]u8 };\n"
"type mid = struct { b: *inner };\n"
"type top = struct { a: mid };\n"
"export fn main() i32 = {\n"
" let z: inner; z.m[1] = 66u8;\n"
" let o: top; o.a.b = &z;\n"
" return o.a.b.m[1]: i32;\n"
"};\n", 66, 1 },
{ "chain_triple",
"package main;\n"
"type inner = struct { m: [4]u8 };\n"
"type amid = struct { q: *inner };\n"
"type otop = struct { p: *amid };\n"
"export fn main() i32 = {\n"
" let z: inner; z.m[1] = 66u8;\n"
" let aa: amid; aa.q = &z;\n"
" let o: otop; o.p = &aa;\n"
" return o.p.q.m[1]: i32;\n"
"};\n", 66, 1 },
{ "chain_tri_comp",
"package main;\n"
"type inner = struct { m: [4]u8 };\n"
"type amid = struct { q: *inner };\n"
"type otop = struct { p: *amid };\n"
"export fn main() i32 = {\n"
" let z: inner; z.m[1] = 60u8;\n"
" let aa: amid; aa.q = &z;\n"
" let o: otop; o.p = &aa;\n"
" o.p.q.m[1] += 6u8;\n"
" return o.p.q.m[1]: i32;\n"
"};\n", 66, 1 },
{ "chain_i32_addr",
"package main;\n"
"type inneri = struct { m: [4]i32 };\n"
"type outeri = struct { p: *inneri };\n"
"export fn main() i32 = {\n"
" let a: inneri; a.m[2] = 88;\n"
" let o: outeri; o.p = &a;\n"
" let q: *i32 = &o.p.m[2];\n"
" return *q;\n"
"};\n", 88, 1 },
{ "chain_i32_slice",
"package main;\n"
"type inneri = struct { m: [4]i32 };\n"
"type outeri = struct { p: *inneri };\n"
"export fn main() i32 = {\n"
" let a: inneri; a.m[1] = 99; a.m[2] = 88;\n"
" let o: outeri; o.p = &a;\n"
" let s: []i32 = o.p.m[1:3];\n"
" return s[1];\n"
"};\n", 88, 1 },
{ "ctrl_ptr_rd",
"package main;\n"
"type e = struct { m: [4]u8 };\n"
"fn rd(p: *e) u8 = { return p.m[1]; };\n"
"export fn main() i32 = {\n"
" let x: e; x.m[1] = 66u8;\n"
" return rd(&x): i32;\n"
"};\n", 66, 1 },
{ "ctrl_local_rd",
"package main;\n"
"type e = struct { m: [4]u8 };\n"
"export fn main() i32 = {\n"
" let x: e; x.m[1] = 66u8;\n"
" return x.m[1]: i32;\n"
"};\n", 66, 1 },
/* #253 chained VALUE-container arm (o.i.m). Run-only (byteid=0):
* a value nested-struct instance trips orthogonal pre-existing
* cs!=ww emission divergences (see header). The fixed helper runs
* these correctly — the segfault is gone. */
{ "chain_val_rd",
"package main;\n"
"type inner = struct { m: [4]u8 };\n"
"type outv = struct { i: inner };\n"
"export fn main() i32 = {\n"
" let o: outv; o.i.m[1] = 66u8;\n"
" return o.i.m[1]: i32;\n"
"};\n", 66, 0 },
{ "chain_val_addr",
"package main;\n"
"type inner = struct { m: [4]u8 };\n"
"type outv = struct { i: inner };\n"
"export fn main() i32 = {\n"
" let o: outv; o.i.m[2] = 55u8;\n"
" let q: *u8 = &o.i.m[2];\n"
" return (*q): i32;\n"
"};\n", 55, 0 },
{ "chain_val_slice",
"package main;\n"
"type inner = struct { m: [4]u8 };\n"
"type outv = struct { i: inner };\n"
"export fn main() i32 = {\n"
" let o: outv; o.i.m[1] = 66u8;\n"
" let s: []u8 = o.i.m[1:4];\n"
" return s[0]: i32;\n"
"};\n", 66, 0 },
{ "chain_deep_val",
"package main;\n"
"type inner = struct { m: [4]u8 };\n"
"type mid = struct { b: inner };\n"
"type top = struct { a: mid };\n"
"export fn main() i32 = {\n"
" let o: top; o.a.b.m[1] = 66u8;\n"
" return o.a.b.m[1]: i32;\n"
"};\n", 66, 0 },
{ NULL, NULL, 0, 0 }
};
static int
@@ -220,6 +425,13 @@ main(void)
}
unlink(outbin); rmdir(tmpdir);
if (!rows[i].byteid) {
/* Run-only row — byte-id blocked by an orthogonal
* pre-existing cs!=ww divergence (see header). */
unlink(src);
continue;
}
char cs_s[64], ws_s[64];
snprintf(cs_s, sizeof cs_s, "/tmp/wwdbs_%d_%d_cs.s",
getpid(), i);