`TRANSMUTE()`: Fix input type size not being properly checked.

Fortune for naka's current commit: Future curse − 末凶

include/macros.h

@@ -18,11 +18,13 @@
 
 #define _cold __attribute__((cold))
 #define _hot __attribute__((hot))
+#define _dead __attribute__((unused))
 
 #ifndef __cplusplus
 #define noreturn __attribute__((noreturn))
-#define deprecated __attribute__((deprecated))
+#define _deprecated __attribute__((deprecated))
 #define deprecated_message(str) __attribute__((deprecated(str)))
+#define _deprecated_msg(str) deprecated_message(str)
 #endif
 
 #define noinline __attribute__((noinline))
@@ -42,6 +44,31 @@
 #define restrict __restrict__
 #endif
 
+//  Function conventions
+
+#define _cconv__mixin _mixin
+#define _cconv__pure _pure
+#define _cconv__readonly _readonly
+#define _cconv__cold _cold
+#define _cconv__dead _dead
+#define _cconv__unused _dead
+#define _cconv__hot _hot
+#define _cconv__entry __attribute__((constructor))
+#define _cconv__exit __attribute__((destructor))
+
+// Allocates
+#define _cconv__alloc __attribute__((malloc))
+// Infallible alloc
+#define _cconv__ialloc __attribute__((malloc)) __attribute__((returns_nonnull))
+#define _cconv__nonnull __attribute__((returns_nonnull))
+
+#define _callcv(name) _cconv__ ## name
+#define _callconv(name) _callcv(name)
+
+#define _callat_entry _callcv(entry)
+#define _callat_exit  _callcv(exit)
+#define _callat(when) _callat_ ## when
+
 // Argument attribute
 
 // Used like: `int *pOUT output, const int *pIN input`
@@ -108,6 +135,8 @@ _mixin void _drain_val(void* x, ...) { IGNORE(x); } // This compiles to no-op on
 // Allocation macros
 
 #define box(t) aligned_alloc(_Alignof(t), sizeof(t))
+#define box_value(v) ({ let _box = box(var(v)); *_box = (v); _box; })
+#define unbox_value(v) ({ let _v = (v); let _res = *_v; free(_v); _res; })
 #define stackalloc(t) __builtin_alloca_with_align(sizeof(t), _Alignof(t))
 
 // Function macros
@@ -177,6 +206,15 @@ static_assert_eq(bswap(128lu), 9223372036854775808lu, "bswap128 (lu) failed (1)"
 static_assert_eq(128lu, bswap(9223372036854775808lu), "bswap128 (lu) failed (2)");
 static_assert_eq(bswap(bswap(128lu)), 128, "bswap128 (lu) failed (3)");
 
+// Transmute
+
+#define TRANSMUTE(val, type) ({ union _trans { var(val) input; type output; }; \
+				_Static_assert(sizeof(var(val)) == sizeof(type), "Cannot transmute values of different sizes"); \
+				_Static_assert(sizeof(union _trans) == sizeof(type), "Cannot transmute values of different sizes"); \
+				/* XXX: Do we need do check this for equality? Can we re-word it or do we even need it at all? _Static_assert(_Alignof(union _trans) == _Alignof(type), "Cannot transmute values of different alignments");*/ \
+				union _trans _trans__value = { .input = (val), }; \
+				_trans__value.output; })
+
 //  Trace message output
 
 #include "trace.h"
@@ -234,4 +272,9 @@ static_assert_eq(bswap(bswap(128lu)), 128, "bswap128 (lu) failed (3)");
 #define VERSION_REV(ver) _AS_u32(_AS_u32(ver) & 0xffu)
 _Static_assert( (VERSION_COMP(VERSION(1,2,3,4), VER_COMP_MIN) >> VER_COMP_MIN) == 2u, "invalid native version spec");
 
+// Misc.
+
+#define PTR_ASSIGN(ptr, val) ( (ptr) ? (*(ptr) = (val), (ptr)) : (ptr) )
+#define PTR_ASSIGNv(ptr, val) ({ let _ptr = (ptr); let _val = (val); PTR_ASSIGN(_ptr, _val); })
+
 #endif /* _MACROS_H */

include/str.h

@@ -0,0 +1,31 @@
+//! Better string handling
+#ifndef _STR_H
+#define _STR_H
+
+#include <macros.h>
+#include <ints.h>
+
+#define STRF_OWNED	AS(1ul << 5, int)
+#define STRF_DERIVED 	AS(1ul << 6, int)
+enum str_ownership {
+	STR_NULL = 0,
+
+	STR_OWNED = STRF_OWNED,
+	STR_OWNED_STATIC,
+	STR_OWNED_MALLOC,
+	STR_OWNED_STACK,
+
+	STR_DERIVED = STRF_DERIVED,
+	STR_DERIVED_STATIC,
+	STR_DERIVED_MALLOC,
+	STR_DERIVED_STACK,
+};
+
+typedef struct string str_t;
+
+str_t* str_alloc(usize cap) _callconv(alloc);
+char* str_new(const char* pIN cstr) _callconv(alloc);
+
+void str_free(str_t* restrict str);
+
+#endif /* _STR_H */

src/display.c

@@ -11,8 +11,101 @@
 #define FLAGSET(f, flag) (( (f) & (flag) ) == (flag))
 #define DFLAGSET(f, flagname) FLAGSET(f, DISPF_SHOW_ ## flagname)
 
-//TODO: This probably won't work. Write wrappers for strncpy(), strncat() that return number of bytes written isntead of useless pointer that we already have.
-static usize _display_get_fmt(dispflags_t flags, bool matched, usize _len, char str[pOUT _len])
+#define CMP(x,cmp,y)  ( (x) cmp (y) ? (x) : (y) )
+#define CMPv(x,cmp,y) ({ let _x = (x); let _y = (y); CMP(_x, cmp, _y); })
+#define MIN(x, y) CMP((x), <, (y))
+#define MINv(x, y) CMPv((x), <, (y))
+#define MAX(x, y) CMP((x), >, (y))
+#define MAXv(x, y) CMPv((x), >, (y))
+
+// Returns number of bytes written to `dest`. Otherwise same as `strncpy()`.
+static usize strncpy_n(char*pOUT dest, const char* restrict src, usize n)
+{
+	usize i;
+	for(i=0;i<n;i++)
+		if(! (*dest++ = *src++)) return i;
+	dest[i] = 0;
+	return i;
+}
+
+// Will not write past `dn` bytes of `dest`, will not read past `sn` bytes of `src`. Otherwise, same as `strncpy_n()`.
+static usize s_strncpy_n(usize sn, usize dn; char dest[static pOUT dn], usize dn, const char src[static restrict sn], usize sn)
+{
+	usize i;
+	usize n = MIN(sn,dn); 
+	for(i=0;i<n;i++)
+		if(! (*dest++ = *src++)) return i;
+	if(i<dn) dest[i] = 0;
+	return i;
+}
+
+// Will not write past `dn` bytes of `dest`, will not read past `sn` bytes of `src`. Otherwise, same as `strncat_n()`.
+static inline usize s_strncat_n(usize sn, usize dn; char dest[static pOUT dn], usize dn, const char src[static restrict sn], usize sn)
+{
+	usize dl = MINv(strlen(dest), dn);
+	return s_strncpy_n(dest + dl, dn - dl, src, sn);
+}
+
+// Returns number of bytes written to (dest+strlen(dest)). Otherwise same as `strncat()`.
+static inline usize strncat_n(char*pOUT dest, const char* restrict src, usize n)
+{
+	usize dl = strlen(dest);
+	return strncpy_n(dest+dl, src, n);
+}
+
+//TODO: This isn't working at all...
+#ifdef _DEBUG_USE_UNSAFE_DISPLAY_FMT_ARGS
+_dead
+#endif
+static usize _display_get_fmt_n(dispflags_t flags, bool matched, usize _n, char str[pOUT _n])
+{
+	register usize n = _n, fw=0;
+	const char* fmts[] = { 
+				DFLAGSET(flags, FAILURES) 
+				? "failure: " 
+				: "",  
+
+				"match: ", 
+				"%lu: ", 
+				SLICE_FORMAT, 
+				" (len %lu)", 
+				(const char*)NULL,
+	};
+	usize   w = s_strncpy_n(str, n, fmts[!!matched], n); // SAFETY: Will never read past `fmts[x]` nul-terminator, so setting n as both dest and src length here is safe.
+//TODO: Fix this... It isn't working...
+#define ADDSTR(s) do { \
+		debug_assert((s)); \
+		w = MINv(s_strncat_n(str, n, (s), n), n); \
+		/* Remove written from `n` */ \
+		n -= w; \
+		/* Check if we have written into the limit (if so, w will be equal to n above, since w is high-bounded to n by `MINx()` above. So if n is 0, we have hit the limit. */ \
+		ifU(!n) { \
+			/*XXX: Should this be `str[w-1]?` If this assert fails then yes it should. */ \
+			debug_assert(w < _n); \
+			str[w] = 0; \
+			return fw + w; \
+		} \
+		/* Increase `str` pointer by written */ \
+		str += w; \
+		/* Increase return value by written. */ \
+		fw += w; \
+	} while(0)
+
+	if(DFLAGSET(flags, NUMBER)) ADDSTR(fmts[2]);
+	if(DFLAGSET(flags, SLICE)) ADDSTR(fmts[3]);
+	if(DFLAGSET(flags, LENGTH)) ADDSTR(fmts[4]);
+	ADDSTR("\n");
+	TRACE("Output string written: %lu, (rlen %lu)", fw, n);
+#undef ADDSTR
+	return fw;
+}
+
+
+//XXX: This works, but seems inherantly unsafe. Rewrite to use the above `strn*_n()` functions for better length tracking.
+#ifndef _DEBUG_USE_UNSAFE_DISPLAY_FMT_ARGS
+_dead deprecated_message("this is unsafe, use `_display_get_fmt_n`") 
+#endif
+static usize _display_get_fmt0(dispflags_t flags, bool matched, usize _len, char str[pOUT _len])
 {
 	register usize len = _len;
 	static const char* fmts[] = { "match: ", "%lu: ", SLICE_FORMAT, " (len %lu)", "failure: ", (const char*)NULL };
@@ -30,6 +123,12 @@ static usize _display_get_fmt(dispflags_t flags, bool matched, usize _len, char
 #undef ADDSTR
 }
 
+#ifdef _DEBUG_USE_UNSAFE_DISPLAY_FMT_ARGS
+#define _display_get_fmt(...) _display_get_fmt0(__VA_ARGS__)
+#else
+#define _display_get_fmt(...) _display_get_fmt_n(__VA_ARGS__)
+#endif
+
 static void _display_normal(FILE* output, dispin_t*pIN input, dispflags_t flags)
 {
 	bool matched = input->matched;
@@ -49,9 +148,14 @@ static void _display_normal(FILE* output, dispin_t*pIN input, dispflags_t flags)
 #endif
 	INFO("Mapped region slice for %lu, pre-transform (at %p): " SLICE_FORMAT ", len: %lu", i+1, base, SLICE_FORMAT_ARGS(&cslice), cslice.len);
 
-	//Honour `flags`
+	// --- get format string from `flags` ---
 	char fmt[200]; //TODO: Find the realistic max for this.
-	fmt[_display_get_fmt(flags, matched, 100, fmt)] = 0;
+	usize fmtw = _display_get_fmt(flags, matched, sizeof(fmt)-1, fmt);
+	assert(fmtw+1 < sizeof(fmt));
+	fmt[fmtw+1] = 0;
+	// ---
+	TRACE("Format string get: `%s' (%lu)", fmt, fmtw);
+
 	fprintf(output, fmt, i+1, pos.start, pos.end, pos.end-pos.start);
 } 
 

src/str.c

@@ -0,0 +1,197 @@
+
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+
+#include <str.h>
+
+struct string {
+	usize len, cap;
+	struct str_meta {	
+		enum str_ownership owned;
+		union _str_meta_inner {
+			// `STRF_OWNED` 
+			struct str_meta_owned {
+				_Atomic usize refs; // Number of derrivations from this 
+				void (*m_free)(void*); // How to free (when `_MALLOC`).
+				_Atomic bool orphan; // If a `derive` references this parent somehow, but this string has been `str_free()`d. Tell the derived string that sets the `refs` to 0 to free it.
+			} owned;
+			// `STRF_DERIVED`
+			struct str_meta_derived {
+				// String this is derived from, can be either owned or derived.
+				struct string* _shared parent;
+				// `str`'s offset from `parent->str`.
+				usize offset;
+			} derived;
+		} _inner;
+	} meta;
+	// The start of this string slice. Regardless of slicing
+	char* str;
+};
+
+//TODO: We can use offsetof() to reverse the result of `str_new()` from char* (str field) to str_t*.
+
+inline static isize _str_wrap_index(usize _max, isize wr)
+{
+	isize max = (isize)_max;
+	ifU(max<0) FATAL("max too large, must be <= isize::max (%lu)", SIZE_MAX>>1);
+	else if(!max) return 0;
+
+	while(wr<0) wr = max + wr;
+	wr = wr >=max ? wr % max : wr;
+	debug_assert(wr>0);
+	return wr;
+}
+
+// Create a cstring (nul-terminated) from `from`. This string is statically allocated and should not be modified.
+// `sl` - Max size of `from` buffer. If 0, then just free the allocated buffer and return empty string;
+// `from` - Non-nul-terminated string buffer to copy from. 
+// `len` - A pointer to a wrappable index for `from`. If `len` is negative, then offset from `sl`, if it's positive, offset from 0; if it's NULL, use `sl` itself. It is wrapped around `sl` in the case that the computed offset is >= sl.
+//
+// # Notes
+// Don't `free()` this returned pointer, this function handles that on call and on program exit.
+static const char* _str_raw_to_cstrsa(usize sl, const char from[restrict sl], isize*pINOUT len)
+{
+	static _Thread_local char* _ma_buf = NULL;
+
+	if(!sl) {
+		free(_ma_buf); //free(NULL) is allowed.
+		_ma_buf=NULL;
+		return "";
+	}
+
+	_callcv(exit) void _cleanup() { free(_ma_buf); _ma_buf = NULL; }
+
+	isize fp = (isize) (sl & (SIZE_MAX >> 1));
+	if(!len) len = &fp;
+	TRACE("copying wrapped %ld bytes from buffer %p[%lu]", *len, from, sl);
+	
+	
+	usize rl = (usize) (*len = _str_wrap_index(sl, *len));
+	
+	if(_ma_buf)	_ma_buf = realloc(_ma_buf, rl+1);
+	else		_ma_buf = malloc(rl+1);
+	memmove(_ma_buf, from, rl);
+	_ma_buf[rl] = 0;
+	return (const char*)_ma_buf;
+}
+
+inline static bool _str_meta_is_owned(enum str_ownership owned)
+{
+	return AS(owned, u32) & AS(STRF_OWNED, u32);
+}
+
+// Returns true for owned, false for derived
+inline static  bool _str_meta_parts_from(enum str_ownership o, union _str_meta_inner *pIN inner, struct str_meta_owned* *pOUT own, struct str_meta_derived* *pOUT derive)
+{
+	if(_str_meta_is_owned(o))
+		return (PTR_ASSIGN(own, &inner->owned), true);
+	else	return (PTR_ASSIGN(derive, &inner->derived), false);
+}
+
+inline static bool _str_is_owned(const str_t*pIN str)
+{
+	return _str_meta_is_owned(str->meta.owned);
+}
+
+inline static bool _str_is_owned_parts(str_t*pIN str, struct str_meta_owned* *pOUT own, struct str_meta_derived* *pOUT derive)
+{
+	return _str_meta_parts_from(str->meta.owned, &str->meta._inner, own, derive);
+}
+
+static union _str_meta_inner _str_default_meta(enum str_ownership oship, bool *pOUT owned)
+{
+	union _str_meta_inner output;
+	
+	struct str_meta_owned* o_own = NULL;
+	struct str_meta_derived* o_der = NULL;
+
+	// PTR_ASSIGNv always evaluates both args, so this is fine
+	PTR_ASSIGNv(owned,
+			_str_meta_parts_from(oship, &output, 
+				&o_own,
+				&o_der)
+	);
+	switch(oship)
+	{
+		case STR_OWNED_STACK:
+		case STR_OWNED_STATIC:
+			debug_assert(o_own);
+			*o_own = (struct str_meta_owned){ 
+				.refs = 0,
+				.m_free = NULL,	
+				.orphan = false,
+			};
+			break;
+		case STR_OWNED_MALLOC:
+			debug_assert(o_own);
+			*o_own = (struct str_meta_owned){
+				.refs = 0,
+				.m_free = &free,
+				.orphan = false,
+			};
+			break;
+
+		case STR_DERIVED_STATIC:
+		case STR_DERIVED_STACK:
+		case STR_DERIVED_MALLOC:
+			debug_assert(o_der);
+			*o_der = (struct str_meta_derived){
+				.parent = NULL,
+				.offset = 0,
+			};
+			break;
+		default:
+			ERROR("Cannot assign metadata from ownership type 0x%x, returning 0'd metadata", AS(oship, u32));	
+			memset(&output, 0, sizeof(output));
+	}
+	return output;
+}
+
+/// Allocates a bare new str_t. The metadata will be incomplete for `derived` string slices, and `str` will be `NULL`. For `owned` strings, `str` will be `calloc()`ed to `cap+1`.
+static str_t _str_alloc_bare(usize cap, enum str_ownership owned)
+{
+	bool is_owned;
+	let meta_inner = _str_default_meta(owned, &is_owned);
+	struct str_meta meta = {
+		.owned = owned,
+		._inner = meta_inner,
+	};
+	str_t str = {
+		.len = 0,
+		.cap = cap,
+		.meta = meta,
+	};
+	str.str = is_owned ? calloc(cap+1, 1) : NULL;
+	return str;
+}
+
+static void _str_force_free(struct str_meta_owned*pIN o_own, void* ptr)
+{
+	(o_own->m_free ?: &free)(ptr);
+}
+
+void str_free(str_t* restrict strp)
+{
+	str_t str = unbox_value(strp);
+	
+	struct str_meta_owned* o_own = NULL;
+	struct str_meta_derived* o_der = NULL;
+
+	if(_str_is_owned_parts(&str, &o_own, &o_der))
+	{
+		debug_assert(o_own);
+		if(o_own->refs > 0) o_own->orphan = true;
+		else if(str.meta.owned == STR_OWNED_MALLOC && str.str)
+			_str_force_free(o_own, str.str);
+	}
+	else {
+		debug_assert(o_der);
+		TODO(" \
+		//TODO: Find top-level parent (traverse list of potentially derived `o_der->parent`s. \
+		//TODO: While doing so, compute the full offset that this slice has from the top owned parent. \
+		//TODO: Then, decrement its `refs` \
+		//TODO: Then, if it is marked as `orphan`, and we have decremented its `refs` to 0, free its `ptr`. \
+		");
+	}
+}

src/tests/transmute.c

@@ -0,0 +1,23 @@
+
+#include <string.h>
+
+#include <ints.h>
+#include <macros.h>
+#include <tests.h>
+
+DEFTEST(transmute_copy)
+{
+	static const u64 EXPECTED = 18446743474098365025lu;
+	struct { char i[4]; i32 u; } input = { .i = {'a','b','c','d'}, .u = -140 };
+	let output = TRANSMUTE(input, u64);
+	TRACE("Transmute test, expected 0x%lx, got 0x%lx", EXPECTED, output);
+	//XXX: TODO: This test will fail on Big Endian machines.
+	TEST_ASSERT(output == EXPECTED);
+	TRACE("Alignment of output: %lu, alignment of input: %lu", _Alignof(output), _Alignof(input));
+	let input2 = TRANSMUTE(output, var(input));
+	TRACE("Transmute output passed, trying to transmute back");
+	TEST_ASSERT( memcmp(&input2, &input, sizeof(input)) == 0 );
+
+	return TEST_OK;
+}
+RUNTEST_DEBUG(transmute_copy);