ted_render.py

from __future__ import annotations

import re

from ted.ted_history import has_pronoun_reference
from ted.ted_support import guard_names


def rewrite_history_context(question: str, condensed_history: str) -> str:
    if not condensed_history:
        return ""
    return condensed_history if has_pronoun_reference(question) else ""


def rewrite_has_unsupported_fact(answer: str, retrieved_names: list[str]) -> bool:
    text = answer.lower()
    if re.search(r"\bdr\.?\s+[a-zà-ÿ-]+", text) or "docteur" in text or "vétérinaire" in text or "veterinaire" in text:
        return True
    return bool(guard_names(answer, retrieved_names)) if retrieved_names else False


def select_nodes_named_in_answer(answer: str, nodes: list, source_limit: int = 3):
    if not nodes:
        return [], None
    focus_node = nodes[0]
    focus_count = 0
    answer_lower = answer.lower()
    mentioned_nodes = []
    for nd in nodes:
        name = str(nd.metadata.get("nom", "")).strip()
        if not name:
            continue
        pattern = rf"(?<!\w){re.escape(name.lower())}(?!\w)"
        occurrences = re.findall(pattern, answer_lower)
        if occurrences:
            first_match = re.search(pattern, answer_lower)
            mentioned_nodes.append((first_match.start(), nd))
            if len(occurrences) > focus_count:
                focus_count = len(occurrences)
                focus_node = nd

    if mentioned_nodes:
        selected = [nd for _, nd in sorted(mentioned_nodes, key=lambda x: x[0])][:source_limit]
    else:
        selected = nodes[:source_limit]
    return selected, focus_node