#ifndef PORTABLE_H
#define PORTABLE_H
/*
 * Architecture portability layer.
 * Target: Raspberry Pi 5 (BCM2712, Cortex-A76, aarch64 LP64, 8GB RAM)
 *         and x86_64 dev host. Both are little-endian; PS/RTP/ISUP wire
 *         formats are BIG-ENDIAN, so every multi-byte field on the wire is
 *         built via explicit shift writers (never by casting a char* to a
 *         wider type, which is undefined on strict-alignment ABIs and
 *         endian-wrong on any host).
 */
#include <stdint.h>
#include <stddef.h>
#include <pthread.h>
#include <unistd.h>
#include <limits.h>
#include <errno.h>

/* char is UNSIGNED by default on the ARM AAPCS (aarch64). Never rely on
 * signedness of plain char for byte values: always use uint8_t or int. */

/* Fixed-width, ABI-stable types. On LP64 'long' is 8 bytes while on ILP32
 * it is 4; we therefore avoid long/int for cross-arch fields. */
typedef int32_t  hk_i32;
typedef uint32_t hk_u32;
typedef int64_t  hk_i64;
typedef uint64_t hk_u64;

/* Big-endian byte writers (wire order). Unaligned-safe via memcpy, correct
 * on x86_64 and aarch64 alike. */
static inline void be_write16(uint8_t *p, uint16_t v){
    p[0]=(uint8_t)(v>>8); p[1]=(uint8_t)v;
}
static inline void be_write24(uint8_t *p, uint32_t v){
    p[0]=(uint8_t)(v>>16); p[1]=(uint8_t)(v>>8); p[2]=(uint8_t)v;
}
static inline void be_write32(uint8_t *p, uint32_t v){
    p[0]=(uint8_t)(v>>24); p[1]=(uint8_t)(v>>16); p[2]=(uint8_t)(v>>8); p[3]=(uint8_t)v;
}
static inline void be_write48(uint8_t *p, uint64_t v){
    for(int i=0;i<6;i++) p[i]=(uint8_t)(v>>((5-i)*8));
}
/* Big-endian readers */
static inline uint16_t be_read16(const uint8_t *p){ return (uint16_t)((p[0]<<8)|p[1]); }
static inline uint32_t be_read32(const uint8_t *p){
    return ((uint32_t)p[0]<<24)|((uint32_t)p[1]<<16)|((uint32_t)p[2]<<8)|(uint32_t)p[3];
}

/* Monotonic milliseconds (immune to wall-clock jumps / NTP on Pi). */
int64_t portable_now_ms(void);
/* Portable sleep (nanosleep; usleep is deprecated on newer glibc). */
void portable_sleep_ms(int ms);
/* Configure a thread with an explicit stack size (Pi5 default 8MB is fine,
 * but big PS buffers are heap-allocated so worker threads get a modest
 * 256KB stack -> predictable on 8GB board and embedded builds). */
int portable_thread_create(pthread_t *t, void*(*fn)(void*), void*arg, size_t stack_bytes);

#endif /* PORTABLE_H */