#!/usr/bin/env python3
"""firtrace.py — live capture of mask->FIR pipeline buffers during realtime
playback, with GUI-publish flag forced ON so bands[i] curves are copied into
the 0x540728+ slots inside the audio callback (decomp 529fe0:144-183).

Dumps full vector-object arrays for every ctx slot in [0x540500,0x540900):
reads {data_ptr, end_ptr, cap_ptr} (std::vector layout) when available,
falls back to fixed-length float view.

Usage: python3 scripts/firtrace.py [rpp] [nsnap]
"""
import json
import os
import struct
import subprocess
import sys
import time

import numpy as np

SNAPDIR = '/tmp/opencode/firtrace'


def find_host():
    import glob
    for p in glob.glob('/proc/[0-9]*'):
        pid = int(os.path.basename(p))
        try:
            cmd = open(f'/proc/{pid}/cmdline', 'rb').read().replace(b'\0', b' ').decode('utf8', 'replace')
            maps = open(f'/proc/{pid}/maps').read()
        except Exception:
            continue
        if 'soothe2' in maps and 'reaper' not in cmd:
            return pid
    return None


def mem_read(fd, addr, n):
    try:
        return os.pread(fd, n, addr)
    except OSError:
        return None


def mem_write(fd, addr, data):
    try:
        os.pwrite(fd, data, addr)
        return True
    except OSError:
        return False


def parse_snap(path):
    data = open(path, 'rb').read()
    regs = []
    i = 0
    while i + 16 <= len(data):
        lo, sz = struct.unpack_from('<QQ', data, i)
        regs.append((lo, data[i + 16:i + 16 + sz]))
        i += 16 + sz
    return regs


def readabs(regs, addr, n):
    for lo, body in regs:
        if lo <= addr < lo + len(body) and addr - lo + n <= len(body):
            return body[addr - lo:addr - lo + n]
    return None


def find_ctx(regs):
    sig = struct.pack('<I', 0x473b8000)
    cands = []
    for lo, body in regs:
        j = body.find(sig)
        while j >= 0:
            base = lo + j - 0x24
            b = readabs(regs, base + 0x540870, 4)
            if b and struct.unpack('<f', b)[0] > 100:
                cands.append(base)
            j = body.find(sig, j + 1)
    return sorted(set(cands))


SLOTS = list(range(0x540600, 0x540900, 8))
EXTRA = [0x540548, 0x540550, 0x540558, 0x540560, 0x540568, 0x540570,
         0x540578, 0x540580, 0x540588, 0x540590, 0x540598, 0x5405a0,
         0x5405a8, 0x5405b0, 0x5405b8, 0x5405c0, 0x5405c8, 0x5405d0,
         0x5405d8, 0x5405e0, 0x5405e8, 0x5405f0, 0x5405f8]


def dump_slot(regs, ctx, off, maxf=65536):
    """Read qword at ctx+off; if it looks like a heap array, dump floats."""
    b = readabs(regs, ctx + off, 8)
    if not b:
        return None
    ptr = struct.unpack('<Q', b)[0]
    if ptr < 0x10000 or ptr > 0x7fffffffffff:
        return None
    body = readabs(regs, ptr, min(maxf, 262144) * 4)
    if body is None or len(body) < 64:
        return None
    a = np.frombuffer(body, dtype='<f4').astype(np.float64)
    # trim trailing zeros beyond a floor of 2049 samples
    nz = np.nonzero(a != 0)[0]
    keep = max(2049, (nz[-1] + 1) if len(nz) else 0)
    return a[:min(len(a), ((keep + 63) // 64) * 64)]


def main():
    rpp = sys.argv[1] if len(sys.argv) > 1 else '/home/m/soothe-bt/dual_b1q_0.5.rpp'
    nsnap = int(sys.argv[2]) if len(sys.argv) > 2 else 4
    os.makedirs(SNAPDIR, exist_ok=True)
    subprocess.run('pkill -9 -x reaser 2>/dev/null; pkill -9 -x reaper 2>/dev/null; '
                   "pkill -9 -f '[y]abridge' 2>/dev/null; sleep 1", shell=True)
    proc = subprocess.Popen(
        ['/usr/bin/reaper', '-nosplash', '-ignoreerrors', rpp,
         '/home/m/re-tools/play_loop.lua'],
        stdout=open('/dev/null', 'w'), stderr=subprocess.STDOUT)
    t0 = time.time()
    host = None
    while time.time() - t0 < 60 and not host:
        host = find_host()
        time.sleep(0.2)
    if not host:
        print('NO HOST')
        return 1
    print('host', host, flush=True)
    time.sleep(12)

    fd = os.open(f'/proc/{host}/mem', os.O_RDONLY)
    # locate ctx via one full snapshot
    snap0 = f'{SNAPDIR}/probe.bin'
    snapshot(fd, host, snap0)
    regs = parse_snap(snap0)
    cands = find_ctx(regs)
    print('ctx candidates:', ['0x%x' % c for c in cands][:5], flush=True)
    if not cands:
        return 1
    ctx = cands[0]

    # force GUI publish flag (byte at ctx+0x2404bc)
    ok = mem_write(fd, ctx + 0x2404bc, b'\x01')
    print('GUI flag write:', ok, flush=True)

    store = {}
    meta = {}
    for k in range(nsnap):
        time.sleep(1.5)
        # re-arm flag (callback consumes it)
        mem_write(fd, ctx + 0x2404bc, b'\x01')
        time.sleep(0.05)
        p = f'{SNAPDIR}/s{k}.bin'
        snapshot(fd, host, p)
        rg = parse_snap(p)
        cs = find_ctx(rg)
        if not cs:
            continue
        c = cs[0]
        tag = f's{k}'
        scal = {}
        for a in range(0x540860, 0x5408a8, 4):
            bb = readabs(rg, c + a, 4)
            scal[hex(a)] = struct.unpack('<f', bb)[0] if bb else None
        store[f'{tag}_scalars'] = json.dumps(scal)
        for off in SLOTS + EXTRA:
            arr = dump_slot(rg, c, off)
            if arr is not None:
                store[f'{tag}_{hex(off)}'] = arr
                meta.setdefault(hex(off), []).append(len(arr))
        print(f'snap {k}: dumped {sum(1 for x in store if x.startswith(tag+"_") and x!=tag+"_scalars")} arrays',
              flush=True)
    os.close(fd)
    if proc.poll() is None:
        proc.kill()

    json.dump(meta, open(f'{SNAPDIR}/meta.json', 'w'), indent=1)
    np.savez_compressed(f'{SNAPDIR}/firtrace.npz', **store)
    print('saved', f'{SNAPDIR}/firtrace.npz')
    return 0


def snapshot(fd, host, path):
    out = open(path, 'wb')
    nreg = nbytes = 0
    for line in open(f'/proc/{host}/maps').read().splitlines():
        p = line.split()
        if len(p) < 2 or 'r' not in p[1]:
            continue
        lo, hi = (int(x, 16) for x in p[0].split('-'))
        a = lo
        while a < hi:
            n = min(hi - a, 8 * 1024 * 1024)
            d = mem_read(fd, a, n)
            if d:
                out.write(struct.pack('<QQ', a, len(d)))
                out.write(d)
                nreg += 1
                nbytes += len(d)
            a += n
    out.close()
    return nreg, nbytes


if __name__ == '__main__':
    sys.exit(main())