diff --git a/Makefile b/Makefile index 13da284d..11cc6329 100644 --- a/Makefile +++ b/Makefile @@ -249,6 +249,7 @@ TESTS = $(BIN)/test_smoke $(BIN)/test_lex $(BIN)/test_parse $(BIN)/test_check \ $(BIN)/test_tagged_subset_reject \ $(BIN)/test_callarg_typecheck \ $(BIN)/test_idxarg_run \ + $(BIN)/test_chainidx_run \ $(BIN)/test_arr_ptr_global \ $(BIN)/test_def_arr_infer_len \ $(BIN)/test_def_arr_len \ @@ -644,6 +645,16 @@ $(BIN)/test_idxarg_run: test/wcc/989_idxarg_run.c \ $(LIB)/libwwrt.a | $(BIN) $(CC) $(CFLAGS) -o $@ $< +# 989_chainidx_run (F7-c3, #22): a chained index `m[i][k]` whose element is +# a str/slice must load the full 24B/16B header. Builds+runs each fixture on +# BOTH the cstage `ww` and wwstage `ww_ww` drivers (rule-10). +$(BIN)/test_chainidx_run: test/wcc/989_chainidx_run.c \ + $(BIN)/ww $(BIN)/ww_ww \ + $(BIN)/w6c $(BIN)/w6a $(BIN)/w6l \ + $(BIN)/w6c_ww $(BIN)/w6a_ww $(BIN)/w6l_ww \ + $(LIB)/libwwrt.a | $(BIN) + $(CC) $(CFLAGS) -o $@ $< + $(BIN)/test_let_global: test/wcc/630_let_global.c $(BIN)/ww $(BIN)/w6c \ $(BIN)/w6a $(BIN)/w6l $(LIB)/libwwrt.a | $(BIN) $(CC) $(CFLAGS) -o $@ $< diff --git a/selfhost/cmd/w6c/main.combined.ww b/selfhost/cmd/w6c/main.combined.ww index 5232d761..fa95c4ff 100644 --- a/selfhost/cmd/w6c/main.combined.ww +++ b/selfhost/cmd/w6c/main.combined.ww @@ -24573,13 +24573,23 @@ fn cgindex(c: *cgen, n: *node) void = { // stamped outer element tinfo (indexresult over the inner // index's value type). cstage reads base->type->sub->size // for esz (cmd/w6c/cgen.c:2070-2071). Drops the - // indexvaluetnode walk. Kept as its OWN arm (it does not set - // elemisstr/elemisslice) so the chained-index byte-id is - // preserved verbatim; the catch-all below would re-stamp it. + // indexvaluetnode walk. + // #22 (F7-c3): also stamp elemisstr/elemisslice off the same + // chased element tinfo. Without them a chained index whose + // element is a str/slice (`m[i][k]` over [N][M]str) loaded + // only the ptr word — the 24B/16B header (len/cap) was + // dropped (stale BX/CX) → garbage .len downstream. cstage's + // idx_eff path classifies the element uniformly via + // type_isstr/type_isslice (cmd/w6c/cgen.c); align ww UP by + // reading the SAME stamp the esz read above uses. CLASS-N: + // the corpus has no chained str/slice element, so the prior + // byte-id is preserved (this only fires on the missed shape). let et: *tinfo = n.type_: *tinfo; if (et != nil) { esz = et.size: i32; signed_elem = typeissigned(et); + elemisstr = typeisstr(et); + elemisslice = typeisslice(et); float_elem = typeisfloat(et); f32_elem = typeisf32(et); elem_isarray = tinfoisarray(et); diff --git a/selfhost/cmd/wcc/cgenexpr.ww b/selfhost/cmd/wcc/cgenexpr.ww index 48c12be4..6eaf7d2e 100644 --- a/selfhost/cmd/wcc/cgenexpr.ww +++ b/selfhost/cmd/wcc/cgenexpr.ww @@ -1875,13 +1875,23 @@ fn cgindex(c: *cgen, n: *node) void = { // stamped outer element tinfo (indexresult over the inner // index's value type). cstage reads base->type->sub->size // for esz (cmd/w6c/cgen.c:2070-2071). Drops the - // indexvaluetnode walk. Kept as its OWN arm (it does not set - // elemisstr/elemisslice) so the chained-index byte-id is - // preserved verbatim; the catch-all below would re-stamp it. + // indexvaluetnode walk. + // #22 (F7-c3): also stamp elemisstr/elemisslice off the same + // chased element tinfo. Without them a chained index whose + // element is a str/slice (`m[i][k]` over [N][M]str) loaded + // only the ptr word — the 24B/16B header (len/cap) was + // dropped (stale BX/CX) → garbage .len downstream. cstage's + // idx_eff path classifies the element uniformly via + // type_isstr/type_isslice (cmd/w6c/cgen.c); align ww UP by + // reading the SAME stamp the esz read above uses. CLASS-N: + // the corpus has no chained str/slice element, so the prior + // byte-id is preserved (this only fires on the missed shape). let et: *tinfo = n.type_: *tinfo; if (et != nil) { esz = et.size: i32; signed_elem = typeissigned(et); + elemisstr = typeisstr(et); + elemisslice = typeisslice(et); float_elem = typeisfloat(et); f32_elem = typeisf32(et); elem_isarray = tinfoisarray(et); diff --git a/selfhost/cmd/wwdump/main.combined.ww b/selfhost/cmd/wwdump/main.combined.ww index 80db6e15..e150f371 100644 --- a/selfhost/cmd/wwdump/main.combined.ww +++ b/selfhost/cmd/wwdump/main.combined.ww @@ -24573,13 +24573,23 @@ fn cgindex(c: *cgen, n: *node) void = { // stamped outer element tinfo (indexresult over the inner // index's value type). cstage reads base->type->sub->size // for esz (cmd/w6c/cgen.c:2070-2071). Drops the - // indexvaluetnode walk. Kept as its OWN arm (it does not set - // elemisstr/elemisslice) so the chained-index byte-id is - // preserved verbatim; the catch-all below would re-stamp it. + // indexvaluetnode walk. + // #22 (F7-c3): also stamp elemisstr/elemisslice off the same + // chased element tinfo. Without them a chained index whose + // element is a str/slice (`m[i][k]` over [N][M]str) loaded + // only the ptr word — the 24B/16B header (len/cap) was + // dropped (stale BX/CX) → garbage .len downstream. cstage's + // idx_eff path classifies the element uniformly via + // type_isstr/type_isslice (cmd/w6c/cgen.c); align ww UP by + // reading the SAME stamp the esz read above uses. CLASS-N: + // the corpus has no chained str/slice element, so the prior + // byte-id is preserved (this only fires on the missed shape). let et: *tinfo = n.type_: *tinfo; if (et != nil) { esz = et.size: i32; signed_elem = typeissigned(et); + elemisstr = typeisstr(et); + elemisslice = typeisslice(et); float_elem = typeisfloat(et); f32_elem = typeisf32(et); elem_isarray = tinfoisarray(et); diff --git a/test/wcc/989_chainidx_run.c b/test/wcc/989_chainidx_run.c new file mode 100644 index 00000000..d3959fb5 --- /dev/null +++ b/test/wcc/989_chainidx_run.c @@ -0,0 +1,196 @@ +/* + * 989_chainidx_run — F7-c3 (#22): a CHAINED index `m[i][k]` whose element + * is a str/slice must load the full 24B/16B header, both stages. + * + * THE BUG (cat-A silent miscompile, gate-blind): cgindex + * (selfhost/cmd/wcc/cgenexpr.ww) has a dedicated arm for a chained index + * (the outer index's base is itself an N_INDEX). That arm read the + * checker-stamped element tinfo for esz/signedness/float-ness but NOT for + * elemisstr/elemisslice — so a chained index over [N][M]str / [N][M][]T + * loaded only the .ptr word and left .len/.cap as stale BX/CX. cstage's + * idx_eff path classifies the element uniformly (type_isstr/type_isslice), + * so it loaded the full header and ran correct — the cat-A divergence. + * 990-997 stay green because the bootstrap corpus never chains an index + * to a str/slice element; only a runtime row catches it. THE FIX: the + * chained-index arm stamps elemisstr/elemisslice off the SAME element + * tinfo it already reads for esz, aligning wwstage UP. + * + * NB: the [N][M] arrays are built by per-element store — the nested array + * literal `[[..],[..]]` is independently #270-1c-blocked (orthogonal). + * + * Rows (each builds+runs on cstage `ww` and, when present, wwstage `ww_ww`; + * rule-10 — both stages must agree AND hit want_exit): + * row | shape | want + * -------------------+------------------------------------+------ + * chain_str_read | let s = m[1][1]; len(s) m:[2][2]str | 5 [#22 bug] + * chain_str_call | take(m[1][1]) m:[2][2]str | 5 [#22+#46: + * | needs the c2 push AND this c3 read] + * chain_slice_read | let xs = m[1][1]; len(xs) m:[2][2][]int | 7 [#22 bug] + * chain_scalar_read | m[1][1] m:[2][2]int | 42 (control: + * | the scalar chained index the arm already handled — + * | c3 must not regress its byte-id) + */ +#include +#include +#include +#include +#include +#include + +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_exit; +}; + +static const struct row rows[] = { + /* (1) #22 — chained str element READ. m[1][1] = "ddddd", len 5. The + * chained-index arm dropped the .len load → garbage. */ + { "chain_str_read", + "package main;\n" + "export fn main() int = {\n" + " let m: [2][2]str;\n" + " m[0][0] = \"aa\"; m[0][1] = \"bbb\";\n" + " m[1][0] = \"c\"; m[1][1] = \"ddddd\";\n" + " let s: str = m[1][1];\n" + " return len(s): int;\n" + "};\n", + 5 }, + + /* (2) #22+#46 — chained str element as a CALL ARG. Needs both the c2 + * push-side recognizer (nodeisstr N_INDEX arm) AND this c3 read-side + * header load; pinned here where both halves are present. len 5. */ + { "chain_str_call", + "package main;\n" + "fn take(s: str) int = { return len(s): int; };\n" + "export fn main() int = {\n" + " let m: [2][2]str;\n" + " m[0][0] = \"aa\"; m[0][1] = \"bbb\";\n" + " m[1][0] = \"c\"; m[1][1] = \"ddddd\";\n" + " return take(m[1][1]);\n" + "};\n", + 5 }, + + /* (3) #22 — chained slice element READ. m[1][1] = b (len 7). The arm + * dropped the .len/.cap load for the 24B slice header. */ + { "chain_slice_read", + "package main;\n" + "export fn main() int = {\n" + " let a: []int = [1, 2];\n" + " let b: []int = [9, 9, 9, 9, 9, 9, 9];\n" + " let m: [2][2][]int;\n" + " m[0][0] = a; m[0][1] = a;\n" + " m[1][0] = a; m[1][1] = b;\n" + " let xs: []int = m[1][1];\n" + " return len(xs): int;\n" + "};\n", + 7 }, + + /* (4) control — chained SCALAR element. The arm already handled the + * scalar esz; c3 must not regress its byte-id. m[1][1] == 42. */ + { "chain_scalar_read", + "package main;\n" + "export fn main() int = {\n" + " let m: [2][2]int;\n" + " m[0][0] = 1; m[0][1] = 2;\n" + " m[1][0] = 3; m[1][1] = 42;\n" + " return m[1][1]: int;\n" + "};\n", + 42 }, +}; + +/* run_build — build+run `src` via `driver`; returns the binary's exit + * code, or -1 on a build failure. */ +static int +run_build(const char *driver, const struct row *r, int i) +{ + char src[64], tmpdir[64], cmd[1024]; + snprintf(src, sizeof src, "/tmp/chainidx_%d_%d.ww", getpid(), i); + snprintf(tmpdir, sizeof tmpdir, "/tmp/chainidx_%d_d_%d", getpid(), i); + + FILE *f = fopen(src, "wb"); + if (!f) return -2; + fputs(r->src, f); + fclose(f); + + mkdir(tmpdir, 0755); + snprintf(cmd, sizeof cmd, "cd %s && %s build %s 2>/dev/null", + tmpdir, driver, src); + int brc = runwait(cmd); + + const char *base = strrchr(src, '/'); + base = base ? base + 1 : src; + char outbin[128]; + snprintf(outbin, sizeof outbin, "%s/%s", tmpdir, base); + char *dot = strrchr(outbin, '.'); + if (dot && strcmp(dot, ".ww") == 0) *dot = '\0'; + + int got = -1; + if (brc == 0) got = runwait(outbin); + + unlink(src); unlink(outbin); rmdir(tmpdir); + return brc == 0 ? got : -1; +} + +int +main(void) +{ + const char *bin = getenv("BIN"); + if (!bin) bin = "out/bin"; + char absbin[1024]; + if (bin[0] != '/') { + char cwd[1024]; + if (getcwd(cwd, sizeof cwd) == NULL) return 1; + snprintf(absbin, sizeof absbin, "%s/%s", cwd, bin); + bin = absbin; + } + + char cdrv[1024], wdrv[1024]; + snprintf(cdrv, sizeof cdrv, "%s/ww", bin); + snprintf(wdrv, sizeof wdrv, "%s/ww_ww", bin); + + struct { const char *name; const char *drv; int gated; } + drivers[] = { + { "cstage", cdrv, 0 }, + { "wwstage", wdrv, 1 }, + { NULL, NULL, 0 }, + }; + + int n = (int)(sizeof rows / sizeof rows[0]); + int total = 0, fail = 0; + + for (int d = 0; drivers[d].name; d++) { + if (drivers[d].gated && access(drivers[d].drv, X_OK) != 0) { + fprintf(stderr, "chainidx_run: skip %s (no %s)\n", + drivers[d].name, drivers[d].drv); + continue; + } + for (int i = 0; i < n; i++) { + total++; + int got = run_build(drivers[d].drv, &rows[i], i); + if (got != rows[i].want_exit) { + fprintf(stderr, "chainidx_run[%s][%s]: exit=%d " + "want=%d\n", drivers[d].name, rows[i].label, + got, rows[i].want_exit); + fail++; + } + } + } + + if (fail) { + fprintf(stderr, "chainidx_run: %d/%d fixtures failed\n", + fail, total); + return 1; + } + printf("chainidx_run: %d/%d ok\n", total, total); + return 0; +}