"""Event & bubble contracts (Python side). Mirrors aria-worker/src/events.js.""" import json import re # Internal event/bubble ids must never leak into visible text. _ID_LEAK = re.compile(r'["\'„“”]?\b(?:evt|sys|bub|ac)_[A-Za-z0-9]+\b["\'„“”]?') # Split a multi-sentence reply into separate messenger bubbles (chat style). # Don't split right after a few common Polish abbreviations. _ABBR = ("np", "itd", "itp", "tzn", "tj", "m.in", "dr", "mgr", "godz", "ok") _SENT_SPLIT = re.compile(r'(?<=[.!?…])\s+') def _split_sentences(text): text = (text or "").strip() if not text: return [] out, buf = [], "" for part in _SENT_SPLIT.split(text): cand = (buf + " " + part).strip() if buf else part low = buf.rstrip(".").split()[-1].lower() if buf else "" if buf and low in _ABBR: # don't break after "np." etc. buf = cand continue if buf: out.append(buf) buf = part if buf: out.append(buf) return [s for s in (x.strip() for x in out) if s] EVENT_TYPES = ("message", "image", "file", "system_idle") BUBBLE_TYPES = ("response", "follow_up", "proactive", "reaction") ALLOWED_REACTIONS = ("❤️", "😆", "😮", "🥹", "😢", "😡") BUBBLES_MAX = 6 BUBBLE_CHARS = 1200 def _messenger_style(text): """Mimic how people text: lowercase start, no trailing full stop. Keeps '...', '?', '!', and all-caps tokens like 'XD'/'OK' intact.""" text = text.strip() if not text: return text # strip any leaked internal id (e.g. evt_xxx), then tidy spacing text = _ID_LEAK.sub("", text) text = re.sub(r"\s{2,}", " ", text).strip() if not text: return text # drop a single trailing period (but keep an ellipsis) if text.endswith(".") and not text.endswith(".."): text = text[:-1].rstrip() # lowercase the first letter unless the first word is an all-caps token first = text.split(" ", 1)[0] if not (len(first) >= 2 and first.isupper()): text = text[:1].lower() + text[1:] return text def sanitize_bubbles(raw, valid_event_ids): """Coerce whatever the model produced into a safe list of bubble dicts.""" arr = raw if isinstance(raw, str): arr = _extract_json_array(raw) if not isinstance(arr, list): arr = [] out = [] for b in arr: if len(out) >= BUBBLES_MAX: break if not isinstance(b, dict): continue btype = b.get("type") if b.get("type") in BUBBLE_TYPES else "response" target = b.get("target_event_id") or b.get("target") if target not in valid_event_ids: target = None # quote / reply-to: which earlier message this bubble answers (shown as a quote) reply_to = b.get("reply_to") or b.get("quote") if reply_to not in valid_event_ids: reply_to = None # Models vary on the field name — accept bubble/content/text/message/emoji. raw_text = (b.get("bubble") or b.get("content") or b.get("text") or b.get("message") or b.get("emoji") or "") # Reaction: keep the raw emoji (no messenger styling); must be allowed + targeted. if btype == "reaction": emoji = str(raw_text).strip() if emoji in ALLOWED_REACTIONS and target: out.append({"bubble": emoji, "target_event_id": target, "type": "reaction"}) continue # split a multi-sentence reply into separate short bubbles (chat style) for i, piece in enumerate(_split_sentences(str(raw_text)[:BUBBLE_CHARS])): if len(out) >= BUBBLES_MAX: break t = _messenger_style(piece) if not t: continue out.append({"bubble": t, "target_event_id": target, "reply_to": reply_to if i == 0 else None, # quote only on the first bubble "type": btype if i == 0 else "follow_up"}) return out def _extract_json_array(s: str): """Best-effort: pull the first JSON array out of a model response.""" s = s.strip() if s.startswith("```"): s = s.strip("`") if s.lower().startswith("json"): s = s[4:] try: return json.loads(s) except Exception: pass start, depth, in_str, esc = -1, 0, False, False for i, ch in enumerate(s): if start == -1: if ch == "[": start, depth = i, 1 continue if in_str: if esc: esc = False elif ch == "\\": esc = True elif ch == '"': in_str = False else: if ch == '"': in_str = True elif ch == "[": depth += 1 elif ch == "]": depth -= 1 if depth == 0: try: return json.loads(s[start:i + 1]) except Exception: return [] return []