fix: violating strict aliasing rules in most areas in the new code.

Yes, I am aware there are plenty of violations in `conf.c`, but I'll
likely fix/rewrite those when I will use it. Since there are some other
changes I think I'll want to make.

makefile

@@ -14,7 +14,7 @@ MARCH   ?= $(shell uname -m)
 KERNEL  ?= $(shell uname -s | tr '[:upper:]' '[:lower:]')
 
 # compilation flags
-CFLAGS  += -c -std=gnu99 -Wall -Wextra -Wpedantic -MMD -MP
+CFLAGS  += -c -std=gnu99 -Wall -Wextra -Wpedantic -MMD -MP -Wno-pointer-arith
 LDFLAGS += -flto
 
 # architecture/OS detection
@@ -38,14 +38,14 @@ endif
 ifeq ($(DEBUG),1)
 PROF = dbg
 CFLAGS  += -UNDEBUG -Og -g -Wextra -Wpedantic
-CFLAGS  += $(if $(filter 1,$(ISWIN)),,-fsanitize=address,undefined)
-LDFLAGS += $(if $(filter 1,$(ISWIN)),,-fsanitize=address,undefined)
+CFLAGS  += $(if $(filter 1,$(ISWIN)),,-fsanitize=address,undefined) -ftrapv
+LDFLAGS += $(if $(filter 1,$(ISWIN)),,-fsanitize=address,undefined) -ftrapv
 # |--profile: testing
 else ifeq ($(DEBUG),test)
 PROF = test
 CFLAGS  += -UNDEBUG -O2 -g
-CFLAGS  += $(if $(filter 1,$(ISWIN)),,-fsanitize=address)
-LDFLAGS += $(if $(filter 1,$(ISWIN)),,-fsanitize=address)
+CFLAGS  += $(if $(filter 1,$(ISWIN)),,-fsanitize=address) -ftrapv
+LDFLAGS += $(if $(filter 1,$(ISWIN)),,-fsanitize=address) -ftrapv
 else
 PROF = rel
 CFLAGS += -DNDEBUG -O2

src/dat/mcx.c

@@ -11,14 +11,13 @@
 
 #include "../util/intdef.h"
 
-#define TABLE  0x2000 // table byte size
+#define TABLE  0x800  // table (total) element count
 #define SECTOR 0x1000 // sector size
 #define CHUNKS 0x400  // amount of chunks in a file
 
 /* Moves chunks `src_s` to `src_e` (inclusive) from `src`, back onto `dst`. */
-static void mvchunks(u8 *restrict buf, u8 *src, u8 *dst, int src_s, int src_e) {
+static void mvchunks(u32 *restrict table, void *src, void *dst, int src_s, int src_e) {
 	assert(src > dst);
-	u32 *table = (u32 *)buf;
 	size_t len = src - dst; // acquire the amount of bytes that we shall move
 	assert(!(len % SECTOR));
 
@@ -34,9 +33,8 @@ static void mvchunks(u8 *restrict buf, u8 *src, u8 *dst, int src_s, int src_e) {
 /* Deletes chunk `sidx` by moving chunks up to `eidx` back over `sidx` in `buf`.
  * `rmb` is an optional additional offset that can be applied, and signifies bytes already removed.
  * Returns the bytes removed by this function. */
-static size_t delchunk(u8 *restrict buf, size_t rmb, int sidx, int eidx) {
+static size_t delchunk(void *restrict buf, u32 *restrict table, size_t rmb, int sidx, int eidx) {
 	// load the table data
-	u32 *table = (u32 *)buf;
 	size_t slen, bidx, blen;
 	slen = be32toh(table[sidx]) & 0xFF;          // acquire the sector length of the chunk
 	bidx = (be32toh(table[sidx]) >> 8) * SECTOR; // acquire and compute the byte offset the chunk starts at
@@ -47,24 +45,29 @@ static size_t delchunk(u8 *restrict buf, size_t rmb, int sidx, int eidx) {
 	table[sidx + CHUNKS] = htobe32(time(NULL)); // assign the current time to the timestamp, for correctness  NOTE: might need to zero-out instead
 
 	// move the succeeding chunks over the deleted chunk
-	u8 *dst = buf + bidx - rmb;
-	u8 *src = buf + bidx + blen;
+	void *dst = buf + bidx - rmb;
+	void *src = buf + bidx + blen;
 	mvchunks(buf, src, dst, sidx, eidx - 1);
 	return blen;
 }
 
-/* Just call `delchunk` with the parameters and some defaults.
+/* Call `delchunk` with the parameters and some defaults. Ensuring the table is copied correctly as well.
  * This is done instead of `delchunk` being globally linked, because
  * `delchunk` requests more specific parameters, which is confusing outside this module. */
-size_t mcx_delchunk(u8 *restrict buf, int chunk) {
-	return delchunk(buf, 0, chunk, CHUNKS);
+size_t mcx_delchunk(void *restrict buf, int chunk) {
+	u32 table[TABLE];
+	memcpy(table, buf, sizeof(table));
+	size_t res = delchunk(buf, table, 0, chunk, CHUNKS);
+	memcpy(buf, table, sizeof(table));
+	return res;
 }
 
-size_t mcx_delchunk_range(u8 *restrict buf, int start, int end) {
+size_t mcx_delchunk_range(void *restrict buf, int start, int end) {
 	assert(start < end && end < CHUNKS);
-	u32 *table = (u32 *)buf;
-	u8 *dst = buf + (be32toh(table[start]) >> 8) * SECTOR;
-	u8 *src = buf + (be32toh(table[end]) >> 8) * SECTOR;
+	u32 table[TABLE];
+	memcpy(table, buf, sizeof(table));
+	void *dst = buf + (be32toh(table[start]) >> 8) * SECTOR;
+	void *src = buf + (be32toh(table[end]) >> 8) * SECTOR;
 	src += (be32toh(table[end]) & 0xFF) * SECTOR;
 
 	// zeroes-out the chunk data within this range. (and set the timestamp)
@@ -77,6 +80,7 @@ size_t mcx_delchunk_range(u8 *restrict buf, int start, int end) {
 	// move the remaining chunks down
 	if (end < (CHUNKS - 1))
 		mvchunks(buf, src, dst, end, (CHUNKS - 1));
+	memcpy(buf, table, sizeof(table));
 	return src - dst;
 }
 
@@ -89,24 +93,29 @@ static int cmp_chunkids(const void *restrict x, const void *restrict y) {
 
 /* Sorts the chunks marked for deletion from smallest to greatest index.
  * Then performs the deletion in this order. Making sure to only update the chunks up to the next. */
-size_t mcx_delchunk_bulk(u8 *restrict buf, const u16 *restrict chunks, int chunkc) {
+size_t mcx_delchunk_bulk(void *restrict buf, const u16 *restrict chunks, int chunkc) {
 	// ensure the chunks ids we're working on are sorted from least to greatest
 	u16 chunkids[chunkc + 1];
 	memcpy(chunkids, chunks, chunkc);
 	qsort(chunkids, chunkc, sizeof(int), cmp_chunkids);
 	chunkids[chunkc] = CHUNKS; // set the spare chunk to the max chunks, so the rest of the chunks are moved
 
+	u32 table[TABLE];
+	memcpy(table, buf, sizeof(table));
+
 	size_t rmb = 0;
 	for (int i = 0; i < chunkc; i++)
-		rmb += delchunk(buf, rmb, chunkids[i], chunkids[i + 1]);
+		rmb += delchunk(buf, table, rmb, chunkids[i], chunkids[i + 1]);
+
+	memcpy(buf, table, sizeof(table));
 	return rmb;
 }
 
 /* Sum together the 4th byte in each location integer to compute the sector size of all chunks.
  * Multiplying by `SECTOR`, and adding the size of the table itself. */
-size_t mcx_calcsize(const u8 *restrict buf) {
+size_t mcx_calcsize(const void *restrict buf) {
 	size_t size = 0;
 	for (uint i = 0; i < CHUNKS; i++)
-		size += *(buf + (i * 4) + 3);
-	return (size * CHUNKS) + TABLE;
+		size += *(u8 *)(buf + (i * 4) + 3);
+	return (size * CHUNKS) + (TABLE * 4);
 }

src/dat/mcx.h

@@ -18,20 +18,20 @@ struct mcx_chunk {
  * The chunk's location data shall become `0`, and timestamp data the current time.
  * All succeeding chunks shall be moved back, freeing space.
  * Returns the amount of bytes removed. */
-size_t mcx_delchunk(u8 *restrict buf, int chunk) NONNULL((1));
+size_t mcx_delchunk(void *restrict buf, int chunk) NONNULL((1));
 
 /* Deletes the range defined by `start`—`end` (inclusive) of chunks out of `buf`.
  * The chunk's location data shall become `0`, and timestamp data the current time.
  * All succeeding chunks shall be moved back, freeing space.
  * Returns the amount of bytes removed */
-size_t mcx_delchunk_range(u8 *restrict buf, int start, int end) NONNULL((1));
+size_t mcx_delchunk_range(void *restrict buf, int start, int end) NONNULL((1));
 
 /* Deletes a `chunkc` chunks from `chunks` out of `buf`.
  * If the `chunks` indices are known to be sequential, i.e. have a constant difference of `1`, `mcx_delchunk_range` should be preferred.
  * The chunk's location data shall become `0`, and timestamp data the current time.
  * All succeeding chunks shall be moved back, freeing space.
  * Returns the amount of bytes removed */
-size_t mcx_delchunk_bulk(u8 *restrict buf, const u16 *restrict chunks, int chunkc) NONNULL((1, 2));
+size_t mcx_delchunk_bulk(void *restrict buf, const u16 *restrict chunks, int chunkc) NONNULL((1, 2));
 
 /* Computes the byte size of the `*.mcX` file in `buf` and returns it. */
-size_t mcx_calcsize(const u8 *restrict buf) NONNULL((1)) PURE;
+size_t mcx_calcsize(const void *restrict buf) NONNULL((1)) PURE;

src/dat/nbt.c

@@ -12,28 +12,52 @@
 
 #define MAX_DEPTH 512
 
+/* Extracts a big endian 16 bit integer from address `buf`, converts it to host byte size if needed and returns. */
+static inline u16 buftoh16(const void *restrict buf) {
+	u16 i;
+	memcpy(&i, buf, sizeof(i));
+	return be16toh(i);
+}
+
+/* Extracts a big endian 32 bit integer from address `buf`, converts it to host byte size if needed and returns. */
+static inline u32 buftoh32(const void *restrict buf) {
+	u32 i;
+	memcpy(&i, buf, sizeof(i));
+	return be32toh(i);
+}
+
+/* Extracts a big endian 64 bit integer from address `buf`, converts it to host byte size if needed and returns. */
+static inline u64 buftoh64(const void *restrict buf) {
+	u64 i;
+	memcpy(&i, buf, sizeof(i));
+	return be64toh(i);
+}
+
 /* Processes the incoming array data in `buf`. Which contains `nmem` items of `size`.
  * The data shall be converted to little-endian on little-endian systems
  * Outputs the allocated data to `out`, returns where the next pointer would be. */
-static const u8 *procarr(const u8 *restrict buf, i32 nmem, uint size, struct nbt_array *restrict *restrict out) {
-	size_t len = nmem * size;
-	*out = malloc(sizeof(struct nbt_array) + len);
-	if (!*out) return buf + len;
+static const void *procarr(const void *restrict buf, i32 nmemb, uint size, struct nbt_array *restrict out) {
+	size_t len = nmemb * size;
+	*out = (struct nbt_array){
+		out->nmemb = nmemb,
+		out->dat = malloc(len),
+	};
+	if (!out->dat)
+		return buf + len;
 
-	memcpy((*out)->dat, buf, len);
-	(*out)->len = nmem;
+	memcpy(out->dat, buf, len);
 	buf += len;
 
 	/* Only include this code for little-endian systems. Since only they require this logic.
 	 * Producing optimised code for other platforms. */
 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
 	if (size == 1) return buf;
-	size_t i = 0;
-	while (i < len) {
+	i32 i = 0;
+	while (i < nmemb) {
 		switch (size) {
-		case 2:  *(u16 *)((*out)->dat + i) = be16toh(*(u16 *)((*out)->dat + i)); break;
-		case 4:  *(u32 *)((*out)->dat + i) = be32toh(*(u32 *)((*out)->dat + i)); break;
-		case 8:  *(u64 *)((*out)->dat + i) = be64toh(*(u64 *)((*out)->dat + i)); break;
+		case 2:  ((u16 *)out->dat)[i] = be16toh(((u16 *)out->dat)[i]); break;
+		case 4:  ((u32 *)out->dat)[i] = be16toh(((u32 *)out->dat)[i]); break;
+		case 8:  ((u64 *)out->dat)[i] = be16toh(((u64 *)out->dat)[i]); break;
 		default: __builtin_unreachable(); // this should be impossible
 		}
 		i += size;
@@ -43,12 +67,10 @@ static const u8 *procarr(const u8 *restrict buf, i32 nmem, uint size, struct nbt
 }
 
 /* calls `procarr` for the simple types available. */
-static const u8 *proclist(const u8 *restrict buf, struct nbt_array *restrict *restrict out) {
+static const void *proclist(const void *restrict buf, struct nbt_array *restrict out) {
 	uint size;
 
-	*out = NULL;
-
-	switch (*buf) {
+	switch (*(u8 *)buf) {
 	case NBT_I8:  size = 1; break;
 	case NBT_I16: size = 2; break;
 	case NBT_I32: // fall through
@@ -59,39 +81,41 @@ static const u8 *proclist(const u8 *restrict buf, struct nbt_array *restrict *re
 	}
 
 	buf++;
-	i32 len = (i32)be32toh(*(u32 *)buf);
+	i32 len;
+	memcpy(&len, buf, 4);
+	len = be32toh(len);
 	buf += 4;
 	return procarr(buf, len, size, out);
 }
 
-const u8 *nbt_proctag(const u8 *restrict buf, u16 slen, void *restrict out) {
-	const u8 *ptr, *tmp;
+const void *nbt_proctag(const void *restrict buf, u16 slen, void *restrict out) {
+	const void *ptr, *tmp;
 	ptr = buf + 3 + slen;
 
 	i32 nmem;
 	uint size;
 
-	switch (*buf) {
-	case NBT_I8:  *(u8 *)out = *ptr; return ptr + 1;
-	case NBT_I16: *(u16 *)out = be16toh(*(u16 *)ptr); return ptr + 2;
+	switch (*(u8 *)buf) {
+	case NBT_I8:  *(u8 *)out = *(u8 *)ptr; return ptr + 1;
+	case NBT_I16: *(u16 *)out = buftoh16(ptr); return ptr + 2;
 	case NBT_I32: // fall through
-	case NBT_F32: *(u32 *)out = be16toh(*(u32 *)ptr); return ptr + 4;
+	case NBT_F32: *(u32 *)out = buftoh32(ptr); return ptr + 4;
 	case NBT_I64: // fall through
-	case NBT_F64: *(u64 *)out = be16toh(*(u64 *)ptr); return ptr + 8;
+	case NBT_F64: *(u64 *)out = buftoh64(ptr); return ptr + 8;
 
-	case NBT_STR:     nmem = be16toh(*(u16 *)ptr), size = 1, ptr += 2; break;
-	case NBT_ARR_I8:  nmem = be32toh(*(u32 *)ptr), size = 1, ptr += 4; break;
-	case NBT_ARR_I32: nmem = be32toh(*(u32 *)ptr), size = 4, ptr += 4; break;
-	case NBT_ARR_I64: nmem = be32toh(*(u32 *)ptr), size = 8, ptr += 4; break;
+	case NBT_STR:     nmem = buftoh16(ptr), size = 1, ptr += 2; break;
+	case NBT_ARR_I8:  nmem = buftoh32(ptr), size = 1, ptr += 4; break;
+	case NBT_ARR_I32: nmem = buftoh32(ptr), size = 4, ptr += 4; break;
+	case NBT_ARR_I64: nmem = buftoh32(ptr), size = 8, ptr += 4; break;
 
 	case NBT_LIST:
-		return proclist(ptr, (struct nbt_array **)out);
+		return proclist(ptr, (struct nbt_array *)out);
 		return tmp;
 
 	default: return NULL;
 	}
 
-	return procarr(ptr, nmem, size, (struct nbt_array **)out);
+	return procarr(ptr, nmem, size, (struct nbt_array *)out);
 }
 
 
@@ -99,22 +123,22 @@ const u8 *nbt_proctag(const u8 *restrict buf, u16 slen, void *restrict out) {
  * `ptr` is assumed to be the start of the `NBT_LIST` data, e.i. The list's ID, followed by the list's length.
  * If `ID` is `NBT_I8`, `NBT_I16`, `NBT_I32`, `NBT_I64`, `NBT_F32`, or `NBT_F64`, the entire list length is computed and returned.
  * For other types this won't be possible, and thus will add `1` to `dpt`, and write the list data to `lens` and `tags` at this new `dpt`. */
-static const u8 *nexttag_list(const u8 *restrict ptr, uint *restrict const dpt, i32 *restrict const lens, u8 *restrict const tags) {
-	const u8 *tag = ptr;
+static const void *nexttag_list(const void *restrict ptr, uint *restrict const dpt, i32 *restrict const lens, u8 *restrict const tags) {
+	const void *tag = ptr;
 	ptr++;
-	switch (*tag) {
+	switch (*(u8 *)tag) {
 	case NBT_END: break;
-	case NBT_I8:  ptr += (i32)be32toh(*(u32 *)ptr) * 1; break;
-	case NBT_I16: ptr += (i32)be32toh(*(u32 *)ptr) * 2; break;
+	case NBT_I8:  ptr += (i32)buftoh32(ptr) * 1; break;
+	case NBT_I16: ptr += (i32)buftoh32(ptr) * 2; break;
 	case NBT_I32: // fall through
-	case NBT_F32: ptr += (i32)be32toh(*(u32 *)ptr) * 4; break;
+	case NBT_F32: ptr += (i32)buftoh32(ptr) * 4; break;
 	case NBT_I64: // fall through
-	case NBT_F64: ptr += (i32)be32toh(*(u32 *)ptr) * 8; break;
+	case NBT_F64: ptr += (i32)buftoh32(ptr) * 8; break;
 	default:
 		// TODO: handle out of bounds... Might not be required if we use flexible array member
 		(*dpt)++;
-		tags[*dpt] = *tag;
-		lens[*dpt] = (i32)be32toh(*(u32 *)ptr);
+		tags[*dpt] = *(u8 *)tag;
+		lens[*dpt] = (i32)buftoh32(ptr);
 		break;
 	}
 	ptr += 4;
@@ -127,16 +151,16 @@ static const u8 *nexttag_list(const u8 *restrict ptr, uint *restrict const dpt,
  * - `lens` shall contain `MAX_DEPTH` of items representing the list length, if the current item is non-zero we shall assume we're in a list.
  *     Where the value is decremented until we reach `0`.
  * - `tags` shall contain `MAX_DEPTH` of items representing the list's stored type. */
-static const u8 *nexttag(const u8 *restrict tag, uint *restrict const dpt, i32 *restrict const lens, u8 *restrict const tags) {
+static const void *nexttag(const void *restrict tag, uint *restrict const dpt, i32 *restrict const lens, u8 *restrict const tags) {
 	u8 type;
-	const u8 *ptr = tag;
+	const void *ptr = tag;
 	if (lens[*dpt]) {
 		type = tags[*dpt];
 		lens[*dpt]--;
 		*dpt -= !lens[*dpt];
 	} else {
-		type = *tag;
-		ptr += be16toh(*(u16 *)(tag + 1)) + 3;
+		type = *(u8 *)tag;
+		ptr += buftoh16(tag + 1) + 3;
 	}
 
 	switch (type) {
@@ -147,10 +171,10 @@ static const u8 *nexttag(const u8 *restrict tag, uint *restrict const dpt, i32 *
 	case NBT_I64: // fall through
 	case NBT_F64: ptr += 8; break;
 
-	case NBT_ARR_I8:  ptr += 4 + (i32)be32toh(*(u32 *)ptr) * 1; break;
-	case NBT_ARR_I32: ptr += 4 + (i32)be32toh(*(u32 *)ptr) * 4; break;
-	case NBT_ARR_I64: ptr += 4 + (i32)be32toh(*(u32 *)ptr) * 8; break;
-	case NBT_STR:     ptr += 2 + (u16)be16toh(*(u16 *)ptr) * 1; break;
+	case NBT_ARR_I8:  ptr += 4 + (i32)buftoh32(ptr) * 1; break;
+	case NBT_ARR_I32: ptr += 4 + (i32)buftoh32(ptr) * 4; break;
+	case NBT_ARR_I64: ptr += 4 + (i32)buftoh32(ptr) * 8; break;
+	case NBT_STR:     ptr += 2 + (u16)buftoh16(ptr) * 1; break;
 
 	case NBT_END:      (*dpt)--; break;
 	case NBT_COMPOUND: (*dpt)++; break;
@@ -169,8 +193,8 @@ static const u8 *nexttag(const u8 *restrict tag, uint *restrict const dpt, i32 *
  * - compound:list:int32
  * - string
  */
-const u8 *nbt_nexttag(const u8 *restrict buf) {
-	const u8 *tag;
+const void *nbt_nexttag(const void *restrict buf) {
+	const void *tag;
 	u8 tags[MAX_DEPTH] = {0};
 	i32 lens[MAX_DEPTH] = {0};
 	uint dpt = 0;

src/dat/nbt.h

@@ -37,8 +37,8 @@ enum nbt_tagid {
 };
 
 struct nbt_array {
-	i32 len;
-	u8 dat[];
+	i32 nmemb;
+	void *dat;
 };
 
 
@@ -49,8 +49,8 @@ struct nbt_array {
  * if `buf` points to `NBT_I8`, `NBT_I16`, `NBT_I32`, `NBT_I64`, `NBT_F32`, or `NBT_F64`, `*out` is assumed
  * to have the available byte width for one of these types. In the case of `NBT_ARR*` and `NBT_LIST`
  * it must point to a `struct nbt_array*`. Where in the case of `NBT_LIST`, it must be one of the previous static-width types. */
-const u8 *nbt_proctag(const u8 *restrict buf, u16 slen, void *restrict out) NONNULL((1, 3));
+const void *nbt_proctag(const void *restrict buf, u16 slen, void *restrict out) NONNULL((1, 3));
 
 /* searches for the end of a named tag without processing data, the final pointer is returned.
  * `NULL` is returned upon failure, the otherwise returned pointer is not guaranteed to be valid. */
-const u8 *nbt_nexttag(const u8 *restrict buf) NONNULL((1)) PURE;
+const void *nbt_nexttag(const void *restrict buf) NONNULL((1)) PURE;