"""Writing style analyser — extracts a WritingStyleProfile from surveyor documents.

Two analysis modes:
* ``analyze_writing_style`` — lightweight (1 200 tokens), used for per-tenant
  profiles built from the user's own uploaded documents.
* ``build_reference_style_profile`` — thorough (4 000 tokens), used once at
  startup to build the Behrang reference profile from the KB corpus.  Extracts
  actual example paragraphs for few-shot style injection in generation prompts.
"""

import json
import logging

from app.models.schemas import WritingStyleProfile

logger = logging.getLogger(__name__)

MAX_SAMPLE_TOKENS = 1200
MAX_REFERENCE_SAMPLE_TOKENS = 4000

# ---------------------------------------------------------------------------
# Lightweight per-tenant style analyser
# ---------------------------------------------------------------------------

_STYLE_SYSTEM_PROMPT = """\
You are an expert writing-style analyst specialising in professional UK property \
survey and technical reports written in British English. Analyse the sample text \
provided and return a single JSON object describing the author's writing style. \
All values in the JSON must themselves use British English spellings \
(e.g. "colour" not "color", "analyse" not "analyze", "organise" not "organize"). \
Output ONLY the JSON object, no markdown fences, no commentary.\
"""

_STYLE_USER_TEMPLATE = """\
Analyse the writing style of the following sample text and return a JSON \
object with exactly these keys:

{{
  "tone": "<formal | semi-formal | technical | casual>",
  "formality_level": "<professional | academic | conversational>",
  "avg_sentence_complexity": "<simple | moderate | complex>",
  "vocabulary_level": "<basic | intermediate | technical | specialist>",
  "common_phrases": ["<phrase 1>", "<phrase 2>", "<phrase 3>"],
  "structural_patterns": ["<pattern 1>", "<pattern 2>"],
  "writing_style_summary": "<One sentence description of the overall style>",
  "example_paragraphs": [
    "<Copy verbatim a SHORT paragraph (40–80 words) from the sample that best shows how this surveyor describes a building condition or defect. Choose factual prose, not headings or fragments. If no suitable paragraph exists in the sample, return an empty list.>",
    "<Optionally a second verbatim paragraph (40–80 words) showing how this surveyor phrases a recommendation or observation. Verbatim or empty list — do not paraphrase.>"
  ]
}}

IMPORTANT for example_paragraphs: copy text VERBATIM from the sample — do not paraphrase, \
do not invent, do not blend multiple sentences from different places. Each entry must be a \
contiguous extract that exists word-for-word in the sample. If the sample is too short or \
fragmentary to extract a clean paragraph, return an empty list.

SAMPLE TEXT:
{sample_text}
"""

# ---------------------------------------------------------------------------
# Thorough reference style analyser (used for KB / Behrang's reports)
# ---------------------------------------------------------------------------

_REFERENCE_STYLE_SYSTEM_PROMPT = """\
You are an expert writing-style analyst specialising in professional UK RICS \
Building Survey reports. You have been given extracts from a specific surveyor's \
completed reports. Your task is to capture their EXACT writing voice so that an \
AI can reproduce it faithfully when generating new survey sections.

Focus on:
- Characteristic sentence openings and transitions ("It is noted that…", "At the time of inspection…")
- How defects, condition ratings, and recommendations are phrased
- Preferred technical vocabulary and professional hedging phrases
- Paragraph rhythm, typical sentence length, and use of passive vs active voice
- How the surveyor opens and closes a section

All values must use British English spellings throughout.
Output ONLY the JSON object — no markdown, no commentary.\
"""

_REFERENCE_STYLE_USER_TEMPLATE = """\
Analyse the following extracts from a professional RICS Building Surveyor's \
completed reports. Capture their writing style precisely.

Return a JSON object with exactly these keys:

{{
  "tone": "<formal | semi-formal | technical | casual>",
  "formality_level": "<professional | academic | conversational>",
  "avg_sentence_complexity": "<simple | moderate | complex>",
  "vocabulary_level": "<basic | intermediate | technical | specialist>",
  "common_phrases": [
    "<characteristic phrase 1>",
    "<phrase 2>",
    "<phrase 3>",
    "<phrase 4>",
    "<phrase 5>"
  ],
  "structural_patterns": [
    "<writing pattern 1 — e.g. how sentences begin>",
    "<pattern 2 — e.g. how defects are described>",
    "<pattern 3 — e.g. how recommendations are worded>"
  ],
  "writing_style_summary": "<Two sentences describing this surveyor's exact voice and approach>",
  "example_paragraphs": [
    "<Copy verbatim a SHORT paragraph (40–80 words) from the sample that best shows how this surveyor describes a building condition or defect>",
    "<Copy verbatim a SECOND paragraph from a different section — ideally a recommendation or next-step statement>",
    "<Copy verbatim a THIRD paragraph showing how they introduce or conclude a section>"
  ]
}}

SURVEYOR'S REPORT EXTRACTS:
{sample_text}
"""

_MOCK_PROFILE = WritingStyleProfile(
    tone="formal",
    formality_level="professional",
    avg_sentence_complexity="moderate",
    vocabulary_level="technical",
    common_phrases=[
        "The property",
        "It is noted that",
        "In respect of",
        "At the time of inspection",
        "We would recommend",
    ],
    structural_patterns=[
        "Topic sentence followed by supporting detail and recommendation",
        "Passive voice preferred for observations; active for recommendations",
        "Condition stated first, cause second, implication third",
    ],
    writing_style_summary=(
        "Professional RICS surveyor style with measured, factual language and passive voice. "
        "Observations are precise and conservative; recommendations are proportionate."
    ),
    example_paragraphs=[],
)


def _truncate_to_tokens(texts: list[str], max_tokens: int) -> str:
    """Concatenate texts up to ``max_tokens`` using exact tiktoken counts."""
    import tiktoken

    enc = tiktoken.get_encoding("cl100k_base")
    parts: list[str] = []
    total = 0
    for text in texts:
        token_ids = enc.encode(text)
        count = len(token_ids)
        if total + count > max_tokens:
            remaining = max(0, max_tokens - total)
            if remaining > 0:
                parts.append(enc.decode(token_ids[:remaining]))
            break
        parts.append(text)
        total += count
    return "\n\n".join(parts)


async def analyze_writing_style_async(
    sample_texts: list[str],
    openai_api_key: str = "",
    chat_model: str = "gpt-4o-mini",
) -> WritingStyleProfile:
    """Async style analysis using throttled OpenAI chat (async pipeline)."""
    if not sample_texts:
        return _MOCK_PROFILE
    if not (openai_api_key or "").strip():
        return _MOCK_PROFILE

    sample = _truncate_to_tokens(sample_texts, MAX_SAMPLE_TOKENS)
    user_prompt = _STYLE_USER_TEMPLATE.format(sample_text=sample)
    try:
        from app.llm.openai_chat import chat_completions_create

        raw = await chat_completions_create(
            messages=[
                {"role": "system", "content": _STYLE_SYSTEM_PROMPT},
                {"role": "user", "content": user_prompt},
            ],
            model=chat_model,
            max_tokens=800,
            temperature=0.0,
            response_format={"type": "json_object"},
            phase="style_analysis",
        )
        data = json.loads(raw or "{}")
        data.setdefault("example_paragraphs", [])
        profile = WritingStyleProfile(**data)
        verified: list[str] = []
        for p in profile.example_paragraphs or []:
            if isinstance(p, str) and p.strip() and p.strip()[:80] in sample:
                verified.append(p.strip())
        if verified:
            profile = WritingStyleProfile(
                **{**profile.model_dump(), "example_paragraphs": verified}
            )
        return profile
    except Exception as exc:  # noqa: BLE001
        logger.warning("Async style analysis failed (%s) — mock profile", exc)
        return _MOCK_PROFILE


def _call_openai_style(
    system: str,
    user: str,
    api_key: str,
    model: str,
    max_tokens: int = 600,
) -> WritingStyleProfile:
    """Shared OpenAI call for both style analyser variants."""
    from openai import OpenAI

    client = OpenAI(api_key=api_key)
    response = client.chat.completions.create(
        model=model,
        messages=[
            {"role": "system", "content": system},
            {"role": "user", "content": user},
        ],
        max_tokens=max_tokens,
        temperature=0.0,
        response_format={"type": "json_object"},
    )
    raw = (response.choices[0].message.content or "").strip()
    data = json.loads(raw)
    # example_paragraphs may not be returned by lightweight prompt — default to []
    data.setdefault("example_paragraphs", [])
    profile = WritingStyleProfile(**data)
    return profile


def analyze_writing_style(
    sample_texts: list[str],
    openai_api_key: str = "",
    chat_model: str = "gpt-4o-mini",
) -> WritingStyleProfile:
    """Lightweight style analysis from the user's own uploaded documents.

    Uses up to 1 200 tokens of sample text. Falls back to the mock profile
    when ``openai_api_key`` is empty or the call fails.

    Args:
        sample_texts: Ordered list of text chunks sampled from user documents.
        openai_api_key: OpenAI API key; empty string triggers mock fallback.
        chat_model: Chat model name used for analysis.

    Returns:
        :class:`~app.models.schemas.WritingStyleProfile` describing the style.
    """
    if not sample_texts:
        logger.debug("No sample texts — returning mock style profile")
        return _MOCK_PROFILE

    if not openai_api_key:
        logger.debug("No OpenAI key — returning mock style profile")
        return _MOCK_PROFILE

    sample = _truncate_to_tokens(sample_texts, MAX_SAMPLE_TOKENS)
    user_prompt = _STYLE_USER_TEMPLATE.format(sample_text=sample)

    try:
        # Bumped 500 → 800 to fit two verbatim 40–80-word example paragraphs in
        # the JSON output. Without this, the model returned the tone/formality
        # fields fine but produced an empty example_paragraphs list, and the
        # downstream LCEL/inspector prompts had no surveyor-voice anchor to
        # mirror — which is why generated reports never sounded "like the
        # uploaded ones".
        profile = _call_openai_style(
            system=_STYLE_SYSTEM_PROMPT,
            user=user_prompt,
            api_key=openai_api_key,
            model=chat_model,
            max_tokens=800,
        )

        # Defence-in-depth: drop any "example" the LLM hallucinated (i.e. a
        # paragraph that does NOT appear verbatim in the sample text). The
        # purpose of these examples is to mirror real surveyor voice; an
        # invented paragraph would teach the downstream LLM to copy a style
        # the surveyor never used. Strict substring containment is fine
        # because the prompt asks for verbatim copies; if the model paraphrases
        # we'd rather drop the example than poison style mirroring.
        verified: list[str] = []
        for p in profile.example_paragraphs or []:
            if isinstance(p, str) and p.strip() and p.strip()[:80] in sample:
                verified.append(p.strip())
        if len(verified) != len(profile.example_paragraphs or []):
            logger.debug(
                "Style example sanitisation: kept %d / %d (dropped non-verbatim entries)",
                len(verified), len(profile.example_paragraphs or []),
            )
        # Pydantic models are immutable by default; rebuild via dict.
        profile = WritingStyleProfile(**{**profile.model_dump(), "example_paragraphs": verified})

        logger.info(
            "Style analysis complete: tone=%s formality=%s examples=%d",
            profile.tone,
            profile.formality_level,
            len(profile.example_paragraphs),
        )
        return profile

    except json.JSONDecodeError as exc:
        logger.warning("Style analysis returned non-JSON — using mock: %s", exc)
        return _MOCK_PROFILE
    except Exception as exc:
        logger.warning("Style analysis failed (%s) — using mock profile", exc)
        return _MOCK_PROFILE


def build_reference_style_profile(
    sample_texts: list[str],
    openai_api_key: str = "",
    chat_model: str = "gpt-4o-mini",
) -> WritingStyleProfile:
    """Thorough style analysis from Behrang's reference reports (KB corpus).

    Uses up to 4 000 tokens of sample text and extracts actual example
    paragraphs for use as few-shot style guidance in generation prompts.

    Falls back to the mock profile when ``openai_api_key`` is empty.

    Args:
        sample_texts: Chunks retrieved from the KB tenant vector store.
        openai_api_key: OpenAI API key; empty string triggers mock fallback.
        chat_model: Chat model name used for analysis.

    Returns:
        Rich :class:`~app.models.schemas.WritingStyleProfile` with
        ``example_paragraphs`` populated.
    """
    if not sample_texts:
        logger.debug("No KB sample texts — returning mock style profile")
        return _MOCK_PROFILE

    if not openai_api_key:
        logger.debug("No OpenAI key — returning mock style profile for reference")
        return _MOCK_PROFILE

    sample = _truncate_to_tokens(sample_texts, MAX_REFERENCE_SAMPLE_TOKENS)
    user_prompt = _REFERENCE_STYLE_USER_TEMPLATE.format(sample_text=sample)

    try:
        profile = _call_openai_style(
            system=_REFERENCE_STYLE_SYSTEM_PROMPT,
            user=user_prompt,
            api_key=openai_api_key,
            model=chat_model,
            max_tokens=900,
        )
        logger.info(
            "Reference style analysis complete: tone=%s formality=%s examples=%d",
            profile.tone,
            profile.formality_level,
            len(profile.example_paragraphs),
        )
        return profile

    except json.JSONDecodeError as exc:
        logger.warning("Reference style analysis returned non-JSON — using mock: %s", exc)
        return _MOCK_PROFILE
    except Exception as exc:
        logger.warning("Reference style analysis failed (%s) — using mock profile", exc)
        return _MOCK_PROFILE