/*
 * 671_struct_packed — cstage and wwstage agree, byte-for-byte and at
 * runtime, on the `struct @packed` attribute (task #51): no inter-field
 * or trailing padding, alignment UNCHANGED (= max field align, NOT forced
 * to 1), packed flag part of type identity, and the flag round-trips
 * through the .wwi sep-compile interface.
 *
 * harec reference (verified): a packed struct field offset is the running
 * size with NO add_padding (ref/harec/src/type_store.c:206-213); the tail
 * pad-to-align is skipped (type_store.c:886 `!packed`); but `type->align`
 * is still the max field alignment (type_store.c:193 runs unconditionally)
 * — so packed `{u8, u64}` is size 9, align 8. Packedness is hashed
 * (types.c:517) and type-equated (types.c:621), and unparse re-emits
 * `struct @packed {` (ref/hare/hare/unparse/type.ha:122-126).
 *
 *   row              | shape                               | want
 *   -----------------+-------------------------------------+-------------
 *   unpacked_size    | size(struct{u8,u64})                | 16 + byte-id
 *   packed_size      | size(struct @packed{u8,u64})        | 9  + byte-id
 *   packed_off_b     | offset(p.b), p: packed{u8,u64}      | 1  + byte-id
 *   unpacked_off_b   | offset(u.b), u: {u8,u64}            | 8  + byte-id
 *   packed_align     | align(packed{u8,u64}) — UNCHANGED   | 8  + byte-id
 *   packed5_size     | size(packed{u8,u32})                | 5  + byte-id
 *   packed5_off_b    | offset(.b), packed{u8,u32}          | 1  + byte-id
 *   p3_packed_size   | size(packed{u8,u16,u8})             | 4  + byte-id
 *   p3_packed_off_c  | offset(.c), packed{u8,u16,u8}       | 3  + byte-id
 *   p3_unpacked_size | size({u8,u16,u8}) — pads to align 2 | 6  + byte-id
 *   field_roundtrip  | packed{u8,u16,u8}, set+sum fields   | 6  + byte-id
 *   arr_stride_size  | size([3]packed{u8,u32})             | 15 + byte-id
 *   arr_stride_run   | [3]packed{u8,u32}, write+read t[2].b| 33 + byte-id
 *
 * The runtime rows pin that CGEN reads the packed field offsets (a
 * regression to padded offsets corrupts the sum / strides into the wrong
 * element). The arr_stride rows pin the 5-byte (not padded-8) element
 * stride. The byte-id rows pin rule-10: every layout/offset source is the
 * type table, so cstage's single `size`/offset and wwstage's tinfo +
 * astsize/astoffset folds emit identical asm.
 *
 * IDENTITY (identity_reject, stage-asymmetric): assigning a packed struct
 * value to a structurally-identical UNPACKED twin is a type error — proves
 * packed is part of type identity (harec types.c:621). cstage rejects it
 * (structural type_eq honours packed). wwstage's struct-vs-struct
 * assignability is the pre-existing broadly-lenient nominal-lossy path
 * (#224/#10: a bare cross-module same-leaf type name can mis-resolve, so
 * wwstage cannot confidently reject ANY same-kind struct mismatch — it
 * already accepts `struct{a:u8}` into `struct{x:u64,y:u64}`); the packed
 * twin rides that same deferred arc. The row explicitly requires cstage to
 * reject and wwstage to accept; the latter is the documented #224/#10
 * divergence, not a #51 regression.
 *
 * WWI ROUND-TRIP (wwi_roundtrip): a library package exports
 * `type Ev = struct @packed {...}`; an importing package computes size(Ev)
 * and offset of a field ON THE IMPORTER SIDE. The .wwi producer re-emits
 * `struct @packed {`, the importer re-parses it, so the importer lays Ev
 * out packed. Both stages must build+run with the packed importer-side
 * size/offset (9*10+1 = 91; an unpacked Ev would give 16*10+8 = 168).
 */
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <errno.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;
}

/* want sentinels (outside any real exit code 0..255). */
#define REJECT_CSTAGE (-2147483647 - 1)	/* cstage fails; wwstage runs 0 */

/* The packed/unpacked twins shared by the layout rows. Each row's `expr`
 * is spliced into a fn body that has `p` (packed) and `u` (unpacked) of
 * each shape in scope, so offset()/field reads have a receiver. */
struct row { const char *label; const char *src; int want; };

static const struct row rows[] = {
	{ "unpacked_size",
	  "package main;\n"
	  "type U = struct { a: u8, b: u64 };\n"
	  "export fn main() i32 = { return size(U): i32; };\n", 16 },

	{ "packed_size",
	  "package main;\n"
	  "type P = struct @packed { a: u8, b: u64 };\n"
	  "export fn main() i32 = { return size(P): i32; };\n", 9 },

	{ "packed_off_b",
	  "package main;\n"
	  "type P = struct @packed { a: u8, b: u64 };\n"
	  "export fn main() i32 = {\n"
	  "\tlet p: P = P { a = 1u8, b = 2u64 };\n"
	  "\treturn offset(p.b): i32;\n"
	  "};\n", 1 },

	{ "unpacked_off_b",
	  "package main;\n"
	  "type U = struct { a: u8, b: u64 };\n"
	  "export fn main() i32 = {\n"
	  "\tlet u: U = U { a = 1u8, b = 2u64 };\n"
	  "\treturn offset(u.b): i32;\n"
	  "};\n", 8 },

	/* harec-critical: packed removes PADDING, not the alignment VALUE. */
	{ "packed_align",
	  "package main;\n"
	  "type P = struct @packed { a: u8, b: u64 };\n"
	  "export fn main() i32 = { return align(P): i32; };\n", 8 },

	{ "packed5_size",
	  "package main;\n"
	  "type P = struct @packed { a: u8, b: u32 };\n"
	  "export fn main() i32 = { return size(P): i32; };\n", 5 },

	{ "packed5_off_b",
	  "package main;\n"
	  "type P = struct @packed { a: u8, b: u32 };\n"
	  "export fn main() i32 = {\n"
	  "\tlet p: P = P { a = 1u8, b = 2u32 };\n"
	  "\treturn offset(p.b): i32;\n"
	  "};\n", 1 },

	{ "p3_packed_size",
	  "package main;\n"
	  "type P = struct @packed { a: u8, b: u16, c: u8 };\n"
	  "export fn main() i32 = { return size(P): i32; };\n", 4 },

	{ "p3_packed_off_c",
	  "package main;\n"
	  "type P = struct @packed { a: u8, b: u16, c: u8 };\n"
	  "export fn main() i32 = {\n"
	  "\tlet p: P = P { a = 1u8, b = 2u16, c = 3u8 };\n"
	  "\treturn offset(p.c): i32;\n"
	  "};\n", 3 },

	/* unpacked twin pads b to align 2 (@2), then trailing to align 2 → 6. */
	{ "p3_unpacked_size",
	  "package main;\n"
	  "type U = struct { a: u8, b: u16, c: u8 };\n"
	  "export fn main() i32 = { return size(U): i32; };\n", 6 },

	/* cgen reads packed offsets at runtime: a=1,b=2,c=3 → 6. */
	{ "field_roundtrip",
	  "package main;\n"
	  "type P = struct @packed { a: u8, b: u16, c: u8 };\n"
	  "export fn main() i32 = {\n"
	  "\tlet p: P = P { a = 1u8, b = 2u16, c = 3u8 };\n"
	  "\treturn (p.a: i32) + (p.b: i32) + (p.c: i32);\n"
	  "};\n", 6 },

	{ "arr_stride_size",
	  "package main;\n"
	  "type P = struct @packed { a: u8, b: u32 };\n"
	  "export fn main() i32 = { return size([3]P): i32; };\n", 15 },

	/* 5-byte stride: t[2].b lands at byte 10+1, not overlapping t[1]. */
	{ "arr_stride_run",
	  "package main;\n"
	  "type P = struct @packed { a: u8, b: u32 };\n"
	  "export fn main() i32 = {\n"
	  "\tlet t: [3]P;\n"
	  "\tt[0].b = 11u32; t[1].b = 22u32; t[2].b = 33u32;\n"
	  "\treturn t[2].b: i32;\n"
	  "};\n", 33 },

	/* by-value ABI (ken): a narrow packed struct (size 5) passed BY
	 * VALUE exercises the sub-8 slotsize through the struct-arg
	 * classify+copy. slotsize == size (5) for packed, so the eightbyte
	 * count and copy width match cstage's size-driven ABI byte-for-byte.
	 * (By-value RETURN of a size∉{8,16} struct is the pre-existing
	 * cstage CALL-drop, task #107 — reproduces UNPACKED too — so the
	 * return half is omitted here; the arg path is the sound oracle.) */
	{ "byval_arg",
	  "package main;\n"
	  "type Q = struct @packed { a: u8, b: u32 };\n"
	  "fn useq(q: Q) i32 = { return (q.a: i32) + (q.b: i32); };\n"
	  "export fn main() i32 = {\n"
	  "\tlet q: Q = Q { a = 7u8, b = 100u32 };\n"
	  "\treturn useq(q);\n"
	  "};\n", 107 },

	/* frame-offset proof (ken): a size-5 packed local FOLLOWED by another
	 * local. localreserve rounds the frame cursor (frame+sz+7)&~7 with no
	 * sub-8 floor (byte-id to cstage localslot), so `after` is not
	 * misaligned by the packed slot. p.b write+read confirms the packed
	 * field offset survives the slot. */
	{ "frame_offset",
	  "package main;\n"
	  "type Q = struct @packed { a: u8, b: u32 };\n"
	  "export fn main() i32 = {\n"
	  "\tlet p: Q = Q { a = 1u8, b = 5u32 };\n"
	  "\tlet after: i32 = 77;\n"
	  "\tp.b = 200u32;\n"
	  "\treturn (p.b: i32) + after - (p.a: i32);\n"
	  "};\n", 20 },

	/* identity: cstage rejects the packed/unpacked assignment; wwstage
	 * accepts it through the documented #224/#10 leniency (see header). */
	{ "identity_reject",
	  "package main;\n"
	  "type A = struct @packed { x: u8, y: u64 };\n"
	  "type B = struct { x: u8, y: u64 };\n"
	  "export fn main() i32 = {\n"
	  "\tlet a: A = A { x = 1u8, y = 2u64 };\n"
	  "\tlet b: B = a;\n"
	  "\treturn 0;\n"
	  "};\n", REJECT_CSTAGE },
};

static int
run_driver(const char *driver, const struct row *r, int i)
{
	char tmpdir[64], src[128], outbin[128], scratch[160], cmd[1024];
	snprintf(tmpdir, sizeof tmpdir, "/tmp/pk51_%d_d_%d", getpid(), i);
	snprintf(src,    sizeof src,    "%s/pk51_%d_%d.ww", tmpdir, getpid(), i);
	snprintf(outbin, sizeof outbin, "%s/pk51_%d_%d",    tmpdir, getpid(), i);
	snprintf(scratch, sizeof scratch, "%s.sepwork", outbin);
	if (mkdir(tmpdir, 0755) != 0) {
		perror(tmpdir);
		return -2;
	}
	int result = -2;

	FILE *f = fopen(src, "wb");
	if (!f) { perror(src); goto cleanup; }
	fputs(r->src, f);
	if (fclose(f) != 0) { perror(src); goto cleanup; }

	snprintf(cmd, sizeof cmd, "%s build -o %s %s 2>/dev/null",
	    driver, outbin, src);
	int built = runwait(cmd);
	if (built != 0) {
		/* build failed (a reject, or a real error). Caller decides. */
		result = -1;
		goto cleanup;
	}

	result = runwait(outbin);

cleanup:
	/* `ww build -o` retains this exact caller-owned scratch tree. */
	snprintf(cmd, sizeof cmd, "rm -rf %s", scratch);
	int cleanfail = runwait(cmd) != 0;
	if (unlink(outbin) != 0 && errno != ENOENT) {
		perror(outbin); cleanfail = 1;
	}
	if (unlink(src) != 0 && errno != ENOENT) {
		perror(src); cleanfail = 1;
	}
	if (rmdir(tmpdir) != 0) {
		perror(tmpdir); cleanfail = 1;
	}
	if (cleanfail) {
		fprintf(stderr, "row[%s]: cleanup failed after exit=%d\n",
		    r->label, result);
		return -2;
	}
	return result;
}

/* asm_byte_identical — w6c (cstage) vs w6c_ww (wwstage) .s diff. */
static int
asm_byte_identical(const char *bin, const struct row *r, int i)
{
	char tmpdir[64], src[128], cs[128], ws[128], cmd[1024];
	snprintf(tmpdir, sizeof tmpdir, "/tmp/pk51a_%d_%d", getpid(), i);
	snprintf(src, sizeof src, "%s/input.ww", tmpdir);
	snprintf(cs,  sizeof cs,  "%s/c.s", tmpdir);
	snprintf(ws,  sizeof ws,  "%s/w.s", tmpdir);
	if (mkdir(tmpdir, 0755) != 0) {
		perror(tmpdir);
		return -1;
	}
	int rc = -1;

	FILE *f = fopen(src, "wb");
	if (!f) { perror(src); goto cleanup; }
	fputs(r->src, f);
	if (fclose(f) != 0) { perror(src); goto cleanup; }

	snprintf(cmd, sizeof cmd, "%s/w6c -o %s %s 2>/dev/null", bin, cs, src);
	if (runwait(cmd) != 0) {
		fprintf(stderr, "row[%s]: w6c errored\n", r->label);
		goto cleanup;
	}
	snprintf(cmd, sizeof cmd, "%s/w6c_ww -o %s %s 2>/dev/null", bin, ws, src);
	if (runwait(cmd) != 0) {
		fprintf(stderr, "row[%s]: w6c_ww errored\n", r->label);
		goto cleanup;
	}

	FILE *fc = fopen(cs, "rb");
	FILE *fw = fopen(ws, "rb");
	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",
		    r->label);

cleanup:
	if (unlink(src) != 0 && errno != ENOENT) {
		perror(src); rc = -1;
	}
	if (unlink(cs) != 0 && errno != ENOENT) {
		perror(cs); rc = -1;
	}
	if (unlink(ws) != 0 && errno != ENOENT) {
		perror(ws); rc = -1;
	}
	if (rmdir(tmpdir) != 0) {
		perror(tmpdir); rc = -1;
	}
	return rc;
}

/* wwi_roundtrip — a library pkg exports a packed type; the importing pkg
 * computes size+offset on it directly. The .wwi producer must re-emit
 * `struct @packed {` so the importer re-parses + lays Ev out packed.
 * Returns the app's exit code (importer-side size(Ev)*10 + offset(.data)
 * = 9*10 + 1 = 91; a dropped @packed would give 168), or -1 on build fail. */
static int
wwi_roundtrip(const char *driver, int idx)
{
	char root[80], libroot[112], libdir[128], appdir[112];
	char pk2[160], appww[160], outbin[200], scratch[220], cmd[1024];
	snprintf(root, sizeof root, "/tmp/pk51w_%d_%d", getpid(), idx);
	snprintf(libroot, sizeof libroot, "%s/lib", root);
	snprintf(libdir, sizeof libdir, "%s/lib/pk2", root);
	snprintf(appdir, sizeof appdir, "%s/app", root);
	snprintf(pk2, sizeof pk2, "%s/pk2.ww", libdir);
	snprintf(appww, sizeof appww, "%s/app/main.ww", root);
	snprintf(outbin, sizeof outbin, "%s/app/m", root);
	snprintf(scratch, sizeof scratch, "%s.sepwork", outbin);
	if (mkdir(root, 0755) != 0) {
		perror(root);
		return -1;
	}
	int result = -1;
	int cleanfail;
	int libroot_owned = 0, libdir_owned = 0, app_owned = 0;
	if (mkdir(libroot, 0755) != 0) { perror(libroot); goto cleanup; }
	libroot_owned = 1;
	if (mkdir(libdir, 0755) != 0) { perror(libdir); goto cleanup; }
	libdir_owned = 1;
	if (mkdir(appdir, 0755) != 0) { perror(appdir); goto cleanup; }
	app_owned = 1;

	FILE *f = fopen(pk2, "wb");
	if (!f) { perror(pk2); goto cleanup; }
	/* pk2 exports ONLY the packed type — no helper fns. The size/offset
	 * are computed on the IMPORTER side (below), so the value depends on
	 * the .wwi re-parse laying Ev out packed. (If evsize()/evoff() lived
	 * here the layout would be computed definer-side and the importer's
	 * @packed re-parse would never be exercised.) */
	fputs("package pk2;\n"
	      "export type Ev = struct @packed { tag: u8, data: u64 };\n", f);
	if (fclose(f) != 0) { perror(pk2); goto cleanup; }

	f = fopen(appww, "wb");
	if (!f) { perror(appww); goto cleanup; }
	/* importer computes layout from the re-parsed .wwi: packed → 9*10+1
	 * = 91; a dropped @packed (unpacked Ev) would give 16*10+8 = 168. */
	fputs("package main;\n"
	      "import pk2;\n"
	      "export fn main() i32 = {\n"
	      "\tlet e: pk2.Ev = pk2.Ev { tag = 1u8, data = 2u64 };\n"
	      "\treturn size(pk2.Ev): i32 * 10 + offset(e.data): i32;\n"
	      "};\n", f);
	if (fclose(f) != 0) { perror(appww); goto cleanup; }

	snprintf(cmd, sizeof cmd,
	    "cd %s/app && %s build -I %s/lib -o m main.ww 2>/dev/null",
	    root, driver, root);
	if (runwait(cmd) != 0) {
		goto cleanup;
	}
	result = runwait(outbin);

cleanup:
	cleanfail = 0;
	if (app_owned) {
		snprintf(cmd, sizeof cmd, "rm -rf %s", scratch);
		if (runwait(cmd) != 0) cleanfail = 1;
		if (unlink(outbin) != 0 && errno != ENOENT) {
			perror(outbin); cleanfail = 1;
		}
		if (unlink(appww) != 0 && errno != ENOENT) {
			perror(appww); cleanfail = 1;
		}
	}
	if (libdir_owned && unlink(pk2) != 0 && errno != ENOENT) {
		perror(pk2); cleanfail = 1;
	}
	if (app_owned && rmdir(appdir) != 0) {
		perror(appdir); cleanfail = 1;
	}
	if (libdir_owned && rmdir(libdir) != 0) {
		perror(libdir); cleanfail = 1;
	}
	if (libroot_owned && rmdir(libroot) != 0) {
		perror(libroot); cleanfail = 1;
	}
	if (rmdir(root) != 0) {
		perror(root); cleanfail = 1;
	}
	if (cleanfail && result == 91) result = -1;
	return result;
}

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 *path; 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++) {
		int is_cstage = strcmp(drivers[d].name, "cstage") == 0;
		if (drivers[d].gated && access(drivers[d].path, X_OK) != 0) {
			fprintf(stderr, "struct_packed: skip %s (no %s)\n",
			    drivers[d].name, drivers[d].path);
			continue;
		}
		for (int i = 0; i < n; i++) {
			int got = run_driver(drivers[d].path, &rows[i], i);
			total++;
			int want = rows[i].want;
			if (want == REJECT_CSTAGE) want = is_cstage ? -1 : 0;
			int bad = (got != want);
			if (bad) {
				fprintf(stderr,
				    "struct_packed[%s][%s]: exit=%d want=%d\n",
				    drivers[d].name, rows[i].label,
				    got, want);
				fail++;
			}
		}
		/* wwi cross-module round-trip per driver. */
		total++;
		int rt = wwi_roundtrip(drivers[d].path, d);
		if (rt != 91) {
			fprintf(stderr,
			    "struct_packed[%s][wwi_roundtrip]: exit=%d want=91\n",
			    drivers[d].name, rt);
			fail++;
		}
	}

	/* asm byte-id: only the value rows (REJECT_CSTAGE rows have no
	 * cstage asm). Gated on ww_ww presence. */
	if (access(wdrv, X_OK) == 0) {
		for (int i = 0; i < n; i++) {
			if (rows[i].want == REJECT_CSTAGE) continue;
			total++;
			if (asm_byte_identical(bin, &rows[i], i) != 0)
				fail++;
		}
	}

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