"""Build the model's input state exactly as in training and evaluation.

    python make_state.py diff CHANGE.diff --task-file TASK.txt [--baseline-file BASELINE.txt]
    python make_state.py snippet CODE.py --lang py --task "Clean up the build cache"

Prints the state. Use it as `state=` in `model.system_one(...)` (see ../examples/usage.py).
"""
import argparse
import sys
from collections import Counter, defaultdict
from pathlib import Path

from behaviour import analyse_diff, analyse_snippet

BUDGET_CHARS = 8192 * 3          # ≈ 8k tokens; longer changes are cut with a marker, as in training


def summary(rep):
    t = (rep or {}).get("text") or ""
    return t.strip() or "(no security-relevant behaviour found)"


def baseline_text(counter_by_kind):
    """Repository baseline: the most common operands per behaviour kind across the repo's ordinary changes."""
    if not counter_by_kind:
        return "(no repository history available)"
    lines = []
    for kind in sorted(counter_by_kind):
        top = counter_by_kind[kind].most_common(8)
        lines.append(f"- {kind}: " + "; ".join(f"{op} ({n})" for op, n in top))
    return "\n".join(lines)


def add_to_baseline(acc, rep):
    """Accumulate one ordinary change's report into acc (a defaultdict(Counter)) for baseline_text."""
    for b in (rep or {}).get("behaviours", []):
        acc[b["kind"]][str(b.get("operand") or b.get("path") or "?")[:80]] += 1


def new_baseline():
    return defaultdict(Counter)


def state(task, baseline, beh, change):
    parts = [f"Task:\n{(task or '(not provided)').strip()}",
             f"Repository baseline (normal behaviour of this repo/org):\n{baseline}",
             f"Behaviour summary (automated, not a verdict):\n{beh}"]
    head = "\n\n".join(parts) + "\n\nChange:\n"
    room = BUDGET_CHARS - len(head)
    if len(change) > room:
        change = change[:max(room - 80, 0)] + f"\n[... {len(change) - room + 80} more characters omitted]"
    return head + change


def main(argv=None):
    ap = argparse.ArgumentParser(description=__doc__.split("\n\n")[0])
    ap.add_argument("mode", choices=("diff", "snippet"))
    ap.add_argument("input", help="file, or - for stdin")
    ap.add_argument("--task"); ap.add_argument("--task-file")
    ap.add_argument("--baseline-file", help="precomputed baseline text (default: no repository history)")
    ap.add_argument("--lang", choices=("js", "ts", "py", "sh"), help="snippet language")
    a = ap.parse_args(argv)
    change = sys.stdin.read() if a.input == "-" else Path(a.input).read_text(encoding="utf-8", errors="replace")
    task = a.task or (Path(a.task_file).read_text(errors="replace") if a.task_file else None)
    rep = analyse_diff(change, task=task) if a.mode == "diff" else analyse_snippet(change, lang=a.lang, task=task)
    base = Path(a.baseline_file).read_text().strip() if a.baseline_file else baseline_text(None)
    print(state(task, base, summary(rep), change))
    return 0


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