wwstage: remap error tag on try-propagate across reordered unions (#173)

wwstage cgtryprop returned the operand union's RAW tag when propagating a
`c(s)?` error, while cstage (cmd/w6c/cgen.c:6161-6184) remaps it to the
enclosing return union's variant ordering. When the operand and return
unions differ in variant order, wwstage propagated the WRONG error variant
at runtime — gate-blind: byte-id (990-997) and cstage==wwstage asm both pass
because the bootstrap only ever tries same-order unions, while the
differing-order case is silently wrong.

Port cstage's remap loop into cgtryprop (iserror-only): for each error
variant whose return-union index differs, emit the CMPQ/JNE/MOVQ/JMP that
rewrites the tag in AX; the error payload words (DX/CX/R8) are untouched and
ride the RET. Mirror cstage's emission exactly — lazy tryprop_ret allocation
on the first remap, j==i skip, j<0 fallback, no dead label when empty,
identical label strings and operand order — so same-order emits zero extra
instructions (byte-id preserved) and differing-order is now byte-identical
cstage==wwstage.

Scope: error-variant remap only. wwstage's hardcoded success-tag=0 and
iserror-only error detection (vs cstage's cg_tagged_success_tag +
cg_variant_is_error legacy fallback) diverge for non-idx-0-success or
unmarked unions — also gate-blind, also latent — filed separately as #216.

test/wcc/925_tryprop_tag_remap_run: 5 rows (differing-order for both error
variants, success unwrap, same-order byte-id witness, and a multi-word !str
payload row asserting the payload bytes survive the remap), each with a
cstage==wwstage .s byte-id check.
This commit is contained in:
2026-05-29 19:43:48 +09:00
parent f6ac7fb2f8
commit f9f83faca7
5 changed files with 493 additions and 9 deletions

View File

@@ -248,6 +248,7 @@ TESTS = $(BIN)/test_smoke $(BIN)/test_lex $(BIN)/test_parse $(BIN)/test_check \
$(BIN)/test_signed_data_emit_run \
$(BIN)/test_tagged_call_arg \
$(BIN)/test_tagged_call_arg_run \
$(BIN)/test_tryprop_tag_remap_run \
$(BIN)/test_sret_struct_return \
$(BIN)/test_sret_struct_return_run \
$(BIN)/test_sret_narrow_field \
@@ -822,6 +823,12 @@ $(BIN)/test_tagged_call_arg_run: test/wcc/924_tagged_call_arg_run.c \
$(LIB)/libwwrt.a | $(BIN)
$(CC) $(CFLAGS) -o $@ $<
$(BIN)/test_tryprop_tag_remap_run: test/wcc/925_tryprop_tag_remap_run.c \
$(BIN)/ww $(BIN)/w6c $(BIN)/w6a $(BIN)/w6l \
$(BIN)/ww_ww $(BIN)/w6c_ww $(BIN)/w6a_ww $(BIN)/w6l_ww \
$(LIB)/libwwrt.a | $(BIN)
$(CC) $(CFLAGS) -o $@ $<
$(BIN)/test_sret_struct_return: test/wcc/721_sret_struct_return.c \
$(BIN)/w6c $(BIN)/w6c_ww | $(BIN)
$(CC) $(CFLAGS) -o $@ $<

View File

@@ -18377,9 +18377,8 @@ fn cgtagvariantidx(c: *cgen, tagged: *node, vt: *node) i32 = {
};
// cgtryprop — `e?` propagates the error variant up the stack.
// Legacy semantics only (success tag = 0). No tag remap; the
// selfhost code that uses ? today has the same variant order in
// operand and enclosing fn.
// Success tag = 0 (#216 tracks the legacy/flag-aware success-tag
// divergence — out of scope here, success check stays `CMPQ $0`).
fn cgtryprop(c: *cgen, n: *node) void = {
cgexpr(c, n.lhs);
// AX = tag. If non-zero, this is an error; pop frame and RET.
@@ -18388,6 +18387,56 @@ fn cgtryprop(c: *cgen, n: *node) void = {
emitline("\tJE\t");
emitline(cl);
emitline("\n");
// #173: remap the operand union's error-variant tag to the
// enclosing fn return union's variant order before propagating.
// When the `?` operand and the enclosing fn return differ in
// variant order, the raw operand tag names the WRONG variant in
// the return union. Mirrors cstage cmd/w6c/cgen.c:6161-6184.
// Payload words DX/CX/R8 ride the RET untouched; only AX (the
// tag) is rewritten. Same-order unions map every error variant to
// itself → zero instructions, byte-id with the pre-#173 emit.
let u: *tinfo = nil;
if (n.lhs != nil) { u = n.lhs.type_: *tinfo; };
for (u != nil && u.kind == tykind.TY_NAMED) { u = u.under; };
let r: *tinfo = nil;
if (c.fnret != nil) { r = c.fnret.type_: *tinfo; };
for (r != nil && r.kind == tykind.TY_NAMED) { r = r.under; };
if (u != nil && r != nil && r.kind == tykind.TY_TAGGED && u.params != nil) {
let propret: str;
propret.ptr = nil; propret.len = 0;
let haveret: bool = false;
let p: *tparam = u.params;
let i: i32 = 0;
for (p != nil) {
if (p.iserror) {
let j: i32 = flatvariantidxt(r, p.type_);
if (j < 0) { j = 0; };
if (j != i) {
let skip: str = mklabel(c, "tryprop_skip");
emitline("\tCMPQ\t$");
emitint(i: i64);
emitline(", AX\n");
emitline("\tJNE\t");
emitline(skip);
emitline("\n");
emitline("\tMOVQ\t$");
emitint(j: i64);
emitline(", AX\n");
if (!haveret) {
propret = mklabel(c, "tryprop_ret");
haveret = true;
};
emitline("\tJMP\t");
emitline(propret);
emitline("\n");
emitlabel(skip);
};
};
p = p.tnext;
i += 1;
};
if (haveret) { emitlabel(propret); };
};
emitline("\tMOVQ\tBP, SP\n\tPOPQ\tBP\n\tRET\n");
emitlabel(cl);
// Success: unwrap value. Tag-only result was AX; the rest of

View File

@@ -156,9 +156,8 @@ fn cgtagvariantidx(c: *cgen, tagged: *node, vt: *node) i32 = {
};
// cgtryprop — `e?` propagates the error variant up the stack.
// Legacy semantics only (success tag = 0). No tag remap; the
// selfhost code that uses ? today has the same variant order in
// operand and enclosing fn.
// Success tag = 0 (#216 tracks the legacy/flag-aware success-tag
// divergence — out of scope here, success check stays `CMPQ $0`).
fn cgtryprop(c: *cgen, n: *node) void = {
cgexpr(c, n.lhs);
// AX = tag. If non-zero, this is an error; pop frame and RET.
@@ -167,6 +166,56 @@ fn cgtryprop(c: *cgen, n: *node) void = {
emitline("\tJE\t");
emitline(cl);
emitline("\n");
// #173: remap the operand union's error-variant tag to the
// enclosing fn return union's variant order before propagating.
// When the `?` operand and the enclosing fn return differ in
// variant order, the raw operand tag names the WRONG variant in
// the return union. Mirrors cstage cmd/w6c/cgen.c:6161-6184.
// Payload words DX/CX/R8 ride the RET untouched; only AX (the
// tag) is rewritten. Same-order unions map every error variant to
// itself → zero instructions, byte-id with the pre-#173 emit.
let u: *tinfo = nil;
if (n.lhs != nil) { u = n.lhs.type_: *tinfo; };
for (u != nil && u.kind == tykind.TY_NAMED) { u = u.under; };
let r: *tinfo = nil;
if (c.fnret != nil) { r = c.fnret.type_: *tinfo; };
for (r != nil && r.kind == tykind.TY_NAMED) { r = r.under; };
if (u != nil && r != nil && r.kind == tykind.TY_TAGGED && u.params != nil) {
let propret: str;
propret.ptr = nil; propret.len = 0;
let haveret: bool = false;
let p: *tparam = u.params;
let i: i32 = 0;
for (p != nil) {
if (p.iserror) {
let j: i32 = flatvariantidxt(r, p.type_);
if (j < 0) { j = 0; };
if (j != i) {
let skip: str = mklabel(c, "tryprop_skip");
emitline("\tCMPQ\t$");
emitint(i: i64);
emitline(", AX\n");
emitline("\tJNE\t");
emitline(skip);
emitline("\n");
emitline("\tMOVQ\t$");
emitint(j: i64);
emitline(", AX\n");
if (!haveret) {
propret = mklabel(c, "tryprop_ret");
haveret = true;
};
emitline("\tJMP\t");
emitline(propret);
emitline("\n");
emitlabel(skip);
};
};
p = p.tnext;
i += 1;
};
if (haveret) { emitlabel(propret); };
};
emitline("\tMOVQ\tBP, SP\n\tPOPQ\tBP\n\tRET\n");
emitlabel(cl);
// Success: unwrap value. Tag-only result was AX; the rest of

View File

@@ -18377,9 +18377,8 @@ fn cgtagvariantidx(c: *cgen, tagged: *node, vt: *node) i32 = {
};
// cgtryprop — `e?` propagates the error variant up the stack.
// Legacy semantics only (success tag = 0). No tag remap; the
// selfhost code that uses ? today has the same variant order in
// operand and enclosing fn.
// Success tag = 0 (#216 tracks the legacy/flag-aware success-tag
// divergence — out of scope here, success check stays `CMPQ $0`).
fn cgtryprop(c: *cgen, n: *node) void = {
cgexpr(c, n.lhs);
// AX = tag. If non-zero, this is an error; pop frame and RET.
@@ -18388,6 +18387,56 @@ fn cgtryprop(c: *cgen, n: *node) void = {
emitline("\tJE\t");
emitline(cl);
emitline("\n");
// #173: remap the operand union's error-variant tag to the
// enclosing fn return union's variant order before propagating.
// When the `?` operand and the enclosing fn return differ in
// variant order, the raw operand tag names the WRONG variant in
// the return union. Mirrors cstage cmd/w6c/cgen.c:6161-6184.
// Payload words DX/CX/R8 ride the RET untouched; only AX (the
// tag) is rewritten. Same-order unions map every error variant to
// itself → zero instructions, byte-id with the pre-#173 emit.
let u: *tinfo = nil;
if (n.lhs != nil) { u = n.lhs.type_: *tinfo; };
for (u != nil && u.kind == tykind.TY_NAMED) { u = u.under; };
let r: *tinfo = nil;
if (c.fnret != nil) { r = c.fnret.type_: *tinfo; };
for (r != nil && r.kind == tykind.TY_NAMED) { r = r.under; };
if (u != nil && r != nil && r.kind == tykind.TY_TAGGED && u.params != nil) {
let propret: str;
propret.ptr = nil; propret.len = 0;
let haveret: bool = false;
let p: *tparam = u.params;
let i: i32 = 0;
for (p != nil) {
if (p.iserror) {
let j: i32 = flatvariantidxt(r, p.type_);
if (j < 0) { j = 0; };
if (j != i) {
let skip: str = mklabel(c, "tryprop_skip");
emitline("\tCMPQ\t$");
emitint(i: i64);
emitline(", AX\n");
emitline("\tJNE\t");
emitline(skip);
emitline("\n");
emitline("\tMOVQ\t$");
emitint(j: i64);
emitline(", AX\n");
if (!haveret) {
propret = mklabel(c, "tryprop_ret");
haveret = true;
};
emitline("\tJMP\t");
emitline(propret);
emitline("\n");
emitlabel(skip);
};
};
p = p.tnext;
i += 1;
};
if (haveret) { emitlabel(propret); };
};
emitline("\tMOVQ\tBP, SP\n\tPOPQ\tBP\n\tRET\n");
emitlabel(cl);
// Success: unwrap value. Tag-only result was AX; the rest of

View File

@@ -0,0 +1,330 @@
/*
* 925_tryprop_tag_remap_run — the `?` try operator must remap the
* operand union's error-variant tag into the ENCLOSING fn return
* union's variant order before propagating (project #173).
*
* Gate-blind hazard: when the `?` operand union and the enclosing fn
* return union differ in variant ORDER, the raw operand tag names the
* WRONG variant in the return union. Pre-fix wwstage cgtryprop did NO
* remap (selfhost/cmd/wcc/cgenexpr.ww) and propagated the raw tag, so
* the caller saw the wrong error variant at runtime — while the
* byte-id (990-997) and cs==ww asm gates both passed, because the
* bootstrap only ever tries SAME-order unions. cstage already remapped
* (cmd/w6c/cgen.c:6161-6184); the fix ports that loop into wwstage.
*
* Two assertions per row:
* - RUNTIME: build with both `ww` (cstage) and `ww_ww` (wwstage),
* run, compare the exit code (= which variant the caller matched).
* This is what was wrong pre-fix.
* - ASM BYTE-ID: compile the same source through `w6c` and `w6c_ww`
* and require byte-identical .s. Proves rule-10 (symmetric stages)
* holds AFTER the fix on differing-order too, and that same-order
* emits no remap (zero delta).
*
* Rows:
* 1. diff_order_e1 — g→e1 on a branched callee; f's union swaps the
* error order; caller asserts the e1 arm. Pre-fix wwstage = 2.
* 2. diff_order_e2 — g→e2; caller asserts the e2 arm. Proves BOTH
* error variants remap (not just the first).
* 3. diff_order_success — g→value; caller asserts the i32 success
* unwrap survives unperturbed by the remap.
* 4. same_order — f's union same order as g; the only shape the
* bootstrap exercises; byte-id witness that the remap is inert.
* 5. multiword — differing-order propagation of a MULTI-WORD `!str`
* error (payload in DX/CX/R8). Caller asserts the PAYLOAD BYTES
* ("AB") survive, not just the tag — proves the remap touches
* only AX and the payload regs ride the RET.
*/
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <sys/stat.h>
#include <sys/wait.h>
static int
runwait(const char *cmd)
{
int rc = system(cmd);
if (rc == -1) return -1;
if (WIFEXITED(rc)) return WEXITSTATUS(rc);
return -1;
}
struct row { const char *label; const char *src; int want; };
static const struct row rows[] = {
{ "diff_order_e1",
"type e1 = !void;\n"
"type e2 = !void;\n"
"fn g(which: i32) (i32 | e1 | e2) = {\n"
" if (which == 1) { return e1{}; };\n"
" if (which == 2) { return e2{}; };\n"
" return 100;\n"
"};\n"
"fn f(which: i32) (i32 | e2 | e1) = {\n"
" let v: i32 = g(which)?;\n"
" return v + 1;\n"
"};\n"
"export fn main() i32 = {\n"
" match (f(1)) {\n"
" case let v: i32 => return 50;\n"
" case e2 => return 2;\n"
" case e1 => return 1;\n"
" };\n"
" return 99;\n"
"};\n",
1 },
{ "diff_order_e2",
"type e1 = !void;\n"
"type e2 = !void;\n"
"fn g(which: i32) (i32 | e1 | e2) = {\n"
" if (which == 1) { return e1{}; };\n"
" if (which == 2) { return e2{}; };\n"
" return 100;\n"
"};\n"
"fn f(which: i32) (i32 | e2 | e1) = {\n"
" let v: i32 = g(which)?;\n"
" return v + 1;\n"
"};\n"
"export fn main() i32 = {\n"
" match (f(2)) {\n"
" case let v: i32 => return 50;\n"
" case e2 => return 2;\n"
" case e1 => return 1;\n"
" };\n"
" return 99;\n"
"};\n",
2 },
{ "diff_order_success",
"type e1 = !void;\n"
"type e2 = !void;\n"
"fn g(which: i32) (i32 | e1 | e2) = {\n"
" if (which == 1) { return e1{}; };\n"
" if (which == 2) { return e2{}; };\n"
" return 100;\n"
"};\n"
"fn f(which: i32) (i32 | e2 | e1) = {\n"
" let v: i32 = g(which)?;\n"
" return v + 1;\n"
"};\n"
"export fn main() i32 = {\n"
" match (f(0)) {\n"
" case let v: i32 => { if (v != 101) { return 40; }; return 7; };\n"
" case e2 => return 2;\n"
" case e1 => return 1;\n"
" };\n"
" return 99;\n"
"};\n",
7 },
{ "same_order",
"type e1 = !void;\n"
"type e2 = !void;\n"
"fn g(which: i32) (i32 | e1 | e2) = {\n"
" if (which == 1) { return e1{}; };\n"
" if (which == 2) { return e2{}; };\n"
" return 100;\n"
"};\n"
"fn f(which: i32) (i32 | e1 | e2) = {\n"
" let v: i32 = g(which)?;\n"
" return v + 1;\n"
"};\n"
"export fn main() i32 = {\n"
" match (f(1)) {\n"
" case let v: i32 => return 50;\n"
" case e1 => return 1;\n"
" case e2 => return 2;\n"
" };\n"
" return 99;\n"
"};\n",
1 },
{ "multiword",
"type e1 = !void;\n"
"type emsg = !str;\n"
"fn g(which: i32) (i32 | e1 | emsg) = {\n"
" if (which == 1) { return e1{}; };\n"
" if (which == 2) { let m: emsg = \"AB\": emsg; return m; };\n"
" return 100;\n"
"};\n"
"fn f(which: i32) (i32 | emsg | e1) = {\n"
" let v: i32 = g(which)?;\n"
" return v + 1;\n"
"};\n"
"export fn main() i32 = {\n"
" match (f(2)) {\n"
" case let v: i32 => return 50;\n"
" case let s: emsg => {\n"
" if (s.len != 2) { return 30; };\n"
" if (s[0] != 65u8) { return 31; };\n"
" if (s[1] != 66u8) { return 32; };\n"
" return 7;\n"
" };\n"
" case e1 => return 1;\n"
" };\n"
" return 99;\n"
"};\n",
7 },
};
/* Write r->src to <dir>/<base>.ww; returns 0 on success. */
static int
write_src(const char *dir, const char *base, const struct row *r, char *out,
size_t outsz)
{
snprintf(out, outsz, "%s/%s.ww", dir, base);
FILE *f = fopen(out, "wb");
if (!f) return -1;
fputs(r->src, f);
fclose(f);
return 0;
}
/* Build src with `driver build` inside its own subdir, run the binary,
* return its exit code (or -1 on a build/exec failure). */
static int
build_run(const char *driver, const char *src, const char *workdir)
{
char cmd[8192];
snprintf(cmd, sizeof cmd, "cd %s && %s build %s > /dev/null 2>&1",
workdir, driver, src);
if (runwait(cmd) != 0) return -1;
const char *base = strrchr(src, '/');
base = base ? base + 1 : src;
char outbin[1024];
snprintf(outbin, sizeof outbin, "%s/%s", workdir, base);
char *dot = strrchr(outbin, '.');
if (dot && strcmp(dot, ".ww") == 0) *dot = '\0';
return runwait(outbin);
}
static int
files_equal(const char *a, const char *b)
{
FILE *fa = fopen(a, "rb");
FILE *fb = fopen(b, "rb");
if (!fa || !fb) { if (fa) fclose(fa); if (fb) fclose(fb); return -1; }
int ca, cb, eq = 1;
do {
ca = fgetc(fa);
cb = fgetc(fb);
if (ca != cb) { eq = 0; break; }
} while (ca != EOF);
fclose(fa);
fclose(fb);
return eq;
}
int
main(void)
{
const char *bin = getenv("BIN");
if (!bin) bin = "out/bin";
char absbin[512];
if (bin[0] != '/') {
char cwd[256];
if (getcwd(cwd, sizeof cwd) == NULL) return 1;
snprintf(absbin, sizeof absbin, "%s/%s", cwd, bin);
bin = absbin;
}
char cdrv[640], wdrv[640], cw6[640], ww6[640];
snprintf(cdrv, sizeof cdrv, "%s/ww", bin);
snprintf(wdrv, sizeof wdrv, "%s/ww_ww", bin);
snprintf(cw6, sizeof cw6, "%s/w6c", bin);
snprintf(ww6, sizeof ww6, "%s/w6c_ww", bin);
int have_ww = (access(wdrv, X_OK) == 0);
int have_w6cww = (access(ww6, X_OK) == 0);
int n = (int)(sizeof rows / sizeof rows[0]);
int total = 0, fail = 0;
for (int i = 0; i < n; i++) {
const struct row *r = &rows[i];
char dir[] = "/tmp/tprm.XXXXXX";
if (mkdtemp(dir) == NULL) {
fprintf(stderr, "row[%s]: mkdtemp failed\n", r->label);
fail++; total++;
continue;
}
char src[1024];
if (write_src(dir, "p", r, src, sizeof src) != 0) {
fprintf(stderr, "row[%s]: write src failed\n", r->label);
fail++; total++;
goto cleanup;
}
/* ASM byte-id: w6c vs w6c_ww .s must be identical. */
if (have_w6cww) {
char css[1024], wss[1024], cmd[8192];
snprintf(css, sizeof css, "%s/cs.s", dir);
snprintf(wss, sizeof wss, "%s/ww.s", dir);
snprintf(cmd, sizeof cmd, "%s -o %s %s > /dev/null 2>&1",
cw6, css, src);
int rc1 = runwait(cmd);
snprintf(cmd, sizeof cmd, "%s -o %s %s > /dev/null 2>&1",
ww6, wss, src);
int rc2 = runwait(cmd);
total++;
if (rc1 != 0 || rc2 != 0) {
fprintf(stderr,
"row[%s]: w6c/w6c_ww emit failed (%d/%d)\n",
r->label, rc1, rc2);
fail++;
} else if (files_equal(css, wss) != 1) {
fprintf(stderr,
"row[%s]: cs.s != ww.s (rule-10 break)\n",
r->label);
fail++;
}
}
/* RUNTIME: cstage. */
{
char wk[1024];
snprintf(wk, sizeof wk, "%s/cs", dir);
mkdir(wk, 0755);
int got = build_run(cdrv, src, wk);
total++;
if (got != r->want) {
fprintf(stderr,
"row[%s][cstage]: exit=%d want=%d\n",
r->label, got, r->want);
fail++;
}
}
/* RUNTIME: wwstage (the side #173 fixes). */
if (have_ww) {
char wk[1024];
snprintf(wk, sizeof wk, "%s/ww", dir);
mkdir(wk, 0755);
int got = build_run(wdrv, src, wk);
total++;
if (got != r->want) {
fprintf(stderr,
"row[%s][wwstage]: exit=%d want=%d\n",
r->label, got, r->want);
fail++;
}
}
cleanup:
{
char rm[1100];
snprintf(rm, sizeof rm, "rm -rf %s", dir);
runwait(rm);
}
}
if (fail) {
fprintf(stderr,
"tryprop_tag_remap_run: %d/%d checks failed\n", fail, total);
return 1;
}
printf("tryprop_tag_remap_run: %d/%d ok\n", total, total);
return 0;
}