/*
 * 732_def_const_fold — top-level `def` rhs const-fold extended to
 * sibling/imported def references, casts, and arithmetic (PROJECT #88).
 *
 * Predecessor 631_def_neg_global lifted the LITERAL fold (INTLIT/RUNELIT/
 * TRUE/FALSE/NIL + unary +/-/~) into the shared fold_int_literal /
 * foldintliteral helper; it explicitly left N_CAST, sibling-ident, and
 * N_BIN OUT OF SCOPE (see 631's header, "negative-u32-cast"). #88 brings
 * them in: a Hare-faithful `def SCHAR_MAX = types::I8_MAX;` or
 * `def MASK: i32 = (1 << 7) - 1;` was previously inexpressible — the rhs
 * folded to nothing, no DATA row was emitted, and any reference failed
 * to link.
 *
 * The fix folds at CHECK time (both stages' pass-2 N_DEF arm) via a new
 * eval_def_const / evaldefconst that reuses the shared fold_binop arith
 * core (factored out of eval_enum_value / enumvalfold) and the checker's
 * EXISTING scope lookup (scope_lookup_prefer for a sibling ref,
 * scope_lookup_in_module for a `mod.NAME` qualified ref). On success the
 * rhs is stamped to an N_INTLIT in place, so cgen's literal-only
 * emit_defs / emitdefconstants lays down the DATA row with no codegen
 * change. The fold is GATED on the plain literal fold missing first, so
 * every pre-#88 def keeps its node and the emitted bytes stay
 * byte-identical (the bootstrap has zero def-ref-def, so 990-997 never
 * touch this latent path — hence this targeted fixture).
 *
 * Coverage:
 *   1. EXEC (same-module, via the `ww` driver, + `ww_ww` if built):
 *      sibling-ref, sibling+arith, def->def->def chain, shift+arith,
 *      and a widening cast — each read back at a use site to prove the
 *      DATA row links and carries the right value.
 *   2. BYTE-ID: cstage w6c vs wwstage w6c_ww `.s` for every exec row.
 *   3. CROSS-MODULE (#88 a2): a combined multi-`package` source (the
 *      driver's internal concat form, fed straight to w6c like
 *      728_match_4arm_cross_module) where `def K = a.J + 1;` resolves
 *      the imported `a.J`. Assert the folded DATA row value AND byte-id.
 *   4. FAIL-LOUD (rule 7): a same-module cycle (def A=B; def B=A), a
 *      cross-module cycle, and a narrowing cast whose value overflows
 *      the target must each fail the build on BOTH stages (never a
 *      silent missing row / silent truncation).
 */
#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;
}

/* ---- 1. same-module EXEC rows (compile+link+run) -------------------- */

struct row { const char *label; const char *src; int want; };

static const struct row exec_rows[] = {
	/* Bare sibling reference: `def B = A;`. Pre-#88 the rhs (N_IDENT)
	 * folded to nothing. */
	{ "sibling-ref",
	  "def A: i32 = 7;\n"
	  "def B: i32 = A;\n"
	  "fn main() i32 = {\n"
	  "\tlet x: i32 = B;\n"
	  "\tif (x == 7) { return 42; };\n"
	  "\treturn 1;\n"
	  "};\n",
	  42 },

	/* Sibling + arithmetic: the headline `def B = A + 1;`. */
	{ "sibling-add",
	  "def A: i32 = 5;\n"
	  "def B: i32 = A + 1;\n"
	  "fn main() i32 = {\n"
	  "\tlet x: i32 = B;\n"
	  "\tif (x == 6) { return 42; };\n"
	  "\treturn 1;\n"
	  "};\n",
	  42 },

	/* def -> def -> def chain: C resolves through B through A. Pins
	 * the recursive resolve (and that the depth guard doesn't trip on
	 * a legitimate short chain). */
	{ "def-chain",
	  "def A: i32 = 5;\n"
	  "def B: i32 = A + 1;\n"
	  "def C: i32 = B * A;\n"
	  "fn main() i32 = {\n"
	  "\tlet x: i32 = C;\n"
	  "\tif (x == 30) { return 42; };\n"
	  "\treturn 1;\n"
	  "};\n",
	  42 },

	/* Pure arithmetic with a shift: `(1 << 7) - 1` == 127. Exercises
	 * the fold_binop shift + subtract path off any sibling ref. */
	{ "shift-arith",
	  "def MASK: i32 = (1 << 7) - 1;\n"
	  "fn main() i32 = {\n"
	  "\tlet x: i32 = MASK;\n"
	  "\tif (x == 127) { return 42; };\n"
	  "\treturn 1;\n"
	  "};\n",
	  42 },

	/* Widening cast over a sibling ref: `A: i64` where A: i32 = 5.
	 * The N_CAST is stripped (i32->i64 widens, value fits), so W folds
	 * to 5 in an 8-byte slot. */
	{ "widening-cast",
	  "def A: i32 = 5;\n"
	  "def W: i64 = A: i64;\n"
	  "fn main() i32 = {\n"
	  "\tlet x: i64 = W;\n"
	  "\tif (x == 5i64) { return 42; };\n"
	  "\treturn 1;\n"
	  "};\n",
	  42 },
};

static int
run_driver(const char *driver, const struct row *r, int i)
{
	char tmpdir[64], src[128], outbin[128], rmcmd[160], cmd[1024];
	snprintf(tmpdir, sizeof tmpdir, "/tmp/dcf_%d_d_%d", getpid(), i);
	mkdir(tmpdir, 0755);
	snprintf(src, sizeof src, "%s/dcf_%d_%d.ww", tmpdir, getpid(), i);
	snprintf(outbin, sizeof outbin, "%s/dcf_%d_%d", tmpdir, getpid(), i);
	snprintf(rmcmd, sizeof rmcmd, "rm -rf %s", tmpdir);

	FILE *f = fopen(src, "wb");
	if (!f) { runwait(rmcmd); return -1; }
	fputs(r->src, f);
	fclose(f);

	/* timeout 180 per repo convention (990-997); test/run does not bound
	 * individual binaries, so an unguarded hang would stall make test. */
	snprintf(cmd, sizeof cmd, "timeout 180 %s build -o %s %s 2>/dev/null",
	    driver, outbin, src);
	if (runwait(cmd) != 0) {
		fprintf(stderr, "row[%s]: build via %s failed\n",
		    r->label, driver);
		runwait(rmcmd);
		return -1;
	}

	int got = runwait(outbin);

	runwait(rmcmd);
	return got;
}

/* ---- shared .s emit + slurp ----------------------------------------- */

static int
emit_s(const char *w6c, const char *src, char *out_s, size_t cap)
{
	char cmd[1024];
	snprintf(cmd, sizeof cmd, "timeout 180 %s -o %s %s 2>/dev/null",
	    w6c, out_s, src);
	return runwait(cmd);
}

static int
slurp(const char *path, char *buf, size_t cap)
{
	FILE *f = fopen(path, "rb");
	if (!f) return -1;
	size_t n = fread(buf, 1, cap - 1, f);
	fclose(f);
	buf[n] = '\0';
	return (int)n;
}

/* ---- 2. byte-identity of cstage vs wwstage .s ----------------------- */

static int
asm_byte_identical(const char *bin, const char *src, const char *label, int i)
{
	char wwsrc[64], cs[64], ws[64];
	snprintf(wwsrc, sizeof wwsrc, "/tmp/dcf_bi_%d_%d.ww", getpid(), i);
	snprintf(cs, sizeof cs, "/tmp/dcf_bi_%d_%d_c.s", getpid(), i);
	snprintf(ws, sizeof ws, "/tmp/dcf_bi_%d_%d_w.s", getpid(), i);

	FILE *f = fopen(wwsrc, "wb");
	if (!f) return -1;
	fputs(src, f);
	fclose(f);

	char w6c[640], w6c_ww[640];
	snprintf(w6c, sizeof w6c, "%s/w6c", bin);
	snprintf(w6c_ww, sizeof w6c_ww, "%s/w6c_ww", bin);

	if (emit_s(w6c, wwsrc, cs, sizeof cs) != 0) {
		fprintf(stderr, "row[%s]: w6c errored\n", label);
		unlink(wwsrc);
		return -1;
	}
	if (emit_s(w6c_ww, wwsrc, ws, sizeof ws) != 0) {
		fprintf(stderr, "row[%s]: w6c_ww errored\n", label);
		unlink(wwsrc); unlink(cs);
		return -1;
	}

	FILE *fc = fopen(cs, "rb");
	FILE *fw = fopen(ws, "rb");
	int rc = 0;
	if (!fc || !fw) {
		rc = -1;
	} else {
		for (;;) {
			int a = fgetc(fc);
			int b = fgetc(fw);
			if (a != b) { rc = -1; break; }
			if (a == EOF) break;
		}
	}
	if (fc) fclose(fc);
	if (fw) fclose(fw);
	if (rc != 0)
		fprintf(stderr, "row[%s]: cstage vs wwstage asm differs\n",
		    label);
	unlink(wwsrc); unlink(cs); unlink(ws);
	return rc;
}

/* ---- 3. cross-module fold: a combined multi-package source ----------
 * Fed straight to w6c (the driver's internal concat form — `ww build`
 * can't expand an already-concatenated source). `def K = a.J + 1;`
 * resolves the imported `a.J` (41) through scope_lookup_in_module and
 * folds to 42, whose DATA row first byte is '*' (0x2a). */
static const char xmod_src[] =
	"package a;\n"
	"export def J: i32 = 41;\n"
	"package b;\n"
	"import a;\n"
	"def K: i32 = a.J + 1;\n"
	"export fn main() i32 = {\n"
	"\tlet k: i32 = K;\n"
	"\tif (k == 42) { return 42; };\n"
	"\treturn 1;\n"
	"};\n";

/* Assert the folded DATA row for symbol `b.K` is present with value 42
 * (first byte '*'). Returns 0 on success. */
static int
xmod_fold_present(const char *w6c, const char *stage)
{
	char src[64], s[64], buf[1 << 15];
	snprintf(src, sizeof src, "/tmp/dcf_xm_%d_%s.ww", getpid(), stage);
	snprintf(s, sizeof s, "/tmp/dcf_xm_%d_%s.s", getpid(), stage);

	FILE *f = fopen(src, "wb");
	if (!f) return -1;
	fputs(xmod_src, f);
	fclose(f);

	if (emit_s(w6c, src, s, sizeof s) != 0) {
		fprintf(stderr, "xmod[%s]: w6c emit failed\n", stage);
		unlink(src);
		return -1;
	}
	int rc = 0;
	if (slurp(s, buf, sizeof buf) < 0) {
		rc = -1;
	} else if (strstr(buf, "b.K(SB),\"*") == NULL) {
		fprintf(stderr,
		    "xmod[%s]: no folded `DATA b.K(SB),\"*` (value 42) row\n",
		    stage);
		rc = -1;
	}
	unlink(src); unlink(s);
	return rc;
}

/* ---- 4. fail-loud rows: must fail the build on BOTH stages ---------- */

struct failrow { const char *label; const char *src; };

static const struct failrow fail_rows[] = {
	/* Same-module cycle: the depth guard must trip and fail loud, not
	 * hang and not silently drop the row. */
	{ "same-module-cycle",
	  "def A: i32 = B;\n"
	  "def B: i32 = A;\n"
	  "export fn main() i32 = { return A; };\n" },

	/* Cross-module cycle (combined source): a.J -> b.K -> a.J. */
	{ "cross-module-cycle",
	  "package a;\n"
	  "import b;\n"
	  "export def J: i32 = b.K;\n"
	  "package b;\n"
	  "import a;\n"
	  "export def K: i32 = a.J;\n"
	  "export fn main() i32 = { return 0; };\n" },

	/* Narrowing cast that loses the value: 0x1FF (511) does not fit a
	 * u8, so the cast must fail loud rather than silently truncate. */
	{ "narrowing-cast",
	  "def Z: u8 = 0x1FF: u8;\n"
	  "export fn main() i32 = { let z: u8 = Z; return z: i32; };\n" },
};

/* Compile via w6c only (no link/run); success means the build FAILED as
 * required (nonzero w6c exit). */
static int
compile_fails(const char *w6c, const struct failrow *r, const char *stage, int i)
{
	char src[64], s[64], cmd[1024];
	snprintf(src, sizeof src, "/tmp/dcf_fl_%d_%d.ww", getpid(), i);
	snprintf(s, sizeof s, "/tmp/dcf_fl_%d_%d.s", getpid(), i);

	FILE *f = fopen(src, "wb");
	if (!f) return -1;
	fputs(r->src, f);
	fclose(f);

	/* A regressed depth guard would HANG the cycle rows rather than fail
	 * loud (rule 7). timeout's 124 is distinct from a clean guard failure
	 * -- fold it back to a TEST failure so a hang can't pass as fail-loud. */
	snprintf(cmd, sizeof cmd, "timeout 180 %s -o %s %s 2>/dev/null",
	    w6c, s, src);
	int rc = runwait(cmd);
	unlink(src); unlink(s);
	if (rc == 0) {
		fprintf(stderr,
		    "failrow[%s][%s]: w6c exited 0 (expected loud failure)\n",
		    r->label, stage);
		return -1;
	}
	if (rc == 124) {
		fprintf(stderr,
		    "failrow[%s][%s]: w6c timed out (depth guard hung?)\n",
		    r->label, stage);
		return -1;
	}
	return 0;
}

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], w6c[1024], w6c_ww[1024];
	snprintf(cdrv, sizeof cdrv, "%s/ww", bin);
	snprintf(wdrv, sizeof wdrv, "%s/ww_ww", bin);
	snprintf(w6c, sizeof w6c, "%s/w6c", bin);
	snprintf(w6c_ww, sizeof w6c_ww, "%s/w6c_ww", bin);

	int have_ww = (access(w6c_ww, X_OK) == 0);
	int total = 0, fail = 0;

	/* 1. same-module exec via cstage `ww` (+ wwstage `ww_ww` if built) */
	struct { const char *name; const char *path; int gated; }
	drivers[] = {
		{ "cstage",  cdrv, 0 },
		{ "wwstage", wdrv, 1 },
		{ NULL, NULL, 0 },
	};
	int nexec = (int)(sizeof exec_rows / sizeof exec_rows[0]);
	for (int d = 0; drivers[d].name; d++) {
		if (drivers[d].gated && access(drivers[d].path, X_OK) != 0) {
			fprintf(stderr, "def_const_fold: skip %s (no %s)\n",
			    drivers[d].name, drivers[d].path);
			continue;
		}
		for (int i = 0; i < nexec; i++) {
			int got = run_driver(drivers[d].path, &exec_rows[i], i);
			total++;
			if (got != exec_rows[i].want) {
				fprintf(stderr,
				    "def_const_fold[%s][%s]: exit=%d want=%d\n",
				    drivers[d].name, exec_rows[i].label,
				    got, exec_rows[i].want);
				fail++;
			}
		}
	}

	/* 2. byte-id on every exec row (only when wwstage is built) */
	if (have_ww) {
		for (int i = 0; i < nexec; i++) {
			total++;
			if (asm_byte_identical(bin, exec_rows[i].src,
			    exec_rows[i].label, i) != 0)
				fail++;
		}
	}

	/* 3. cross-module fold: DATA row value on cstage, + byte-id */
	total++;
	if (xmod_fold_present(w6c, "cstage") != 0) fail++;
	if (have_ww) {
		total++;
		if (xmod_fold_present(w6c_ww, "wwstage") != 0) fail++;
		total++;
		if (asm_byte_identical(bin, xmod_src, "xmod-fold", 900) != 0)
			fail++;
	}

	/* 4. fail-loud on BOTH stages */
	int nfail = (int)(sizeof fail_rows / sizeof fail_rows[0]);
	for (int i = 0; i < nfail; i++) {
		total++;
		if (compile_fails(w6c, &fail_rows[i], "cstage", i) != 0) fail++;
		if (have_ww) {
			total++;
			if (compile_fails(w6c_ww, &fail_rows[i], "wwstage",
			    1000 + i) != 0)
				fail++;
		}
	}

	if (fail) {
		fprintf(stderr,
		    "def_const_fold: %d/%d fixtures failed\n", fail, total);
		return 1;
	}
	printf("def_const_fold: %d/%d ok\n", total, total);
	return 0;
}