from __future__ import annotations

import re

from .models import Phrase, WordTiming

_SENTENCE_END = re.compile(r'[.!?][\"”’\']?$')
_CLAUSE_END = re.compile(r'[,;:][\"”’\']?$')


def build_phrases(
    words: list[WordTiming],
    *,
    max_words: int = 11,
    max_chars: int = 74,
    hard_pause: float = 0.70,
    clause_pause: float = 0.28,
) -> list[Phrase]:
    """Group measured words into complete, readable conceptual phrases.

    The algorithm uses punctuation, measured pauses, and size limits. It never
    changes the word order or inserts words not present in the aligned audio.
    """
    phrases: list[Phrase] = []
    current: list[int] = []

    def close_phrase() -> None:
        nonlocal current
        if not current:
            return
        selected = [words[i] for i in current]
        phrase_id = len(phrases)
        for item in selected:
            item.phrase_id = phrase_id
        phrases.append(Phrase(
            id=phrase_id,
            word_indices=current.copy(),
            text=" ".join(item.word for item in selected),
            start=selected[0].start,
            end=selected[-1].end,
        ))
        current = []

    for i, word in enumerate(words):
        current.append(i)
        text = " ".join(words[j].word for j in current)
        gap = words[i + 1].start - word.end if i + 1 < len(words) else 999.0

        sentence_end = bool(_SENTENCE_END.search(word.word))
        clause_end = bool(_CLAUSE_END.search(word.word))
        size_limit = len(current) >= max_words or len(text) >= max_chars
        pause_break = gap >= hard_pause
        clause_break = clause_end and gap >= clause_pause and len(current) >= 4

        if sentence_end or size_limit or pause_break or clause_break:
            close_phrase()

    close_phrase()
    return phrases