"""Section-domain scoping to prevent cross-section contamination (STEP 6).

Roofing prompts must only see roofing evidence; drainage only drainage, etc.
This module is a pure, deterministic keyword classifier — no embeddings, no
network — so the scoping decision is reproducible and unit-testable.

Usage::

    domain = classify_section("Roofing", template_id="E1")
    roofing_only = scope_chunks(domain, retrieved_chunks)

The classifier is intentionally conservative: a chunk with no clear domain
signal is treated as ``GENERAL`` and is admissible to any section (so we never
starve a section of context), but a chunk that clearly belongs to a *different*
domain is excluded.
"""

from __future__ import annotations

from app.extraction.citation_validator import normalize_for_match

GENERAL = "general"

# Domain → indicative terms. Order matters for deterministic tie-breaking
# (earlier domains win ties). Terms are matched as normalized substrings, so
# multi-word phrases are supported.
DOMAIN_LEXICON: dict[str, frozenset[str]] = {
    "roofing": frozenset({
        "roof", "roofing", "slate", "tile", "tiling", "ridge", "hip", "valley",
        "verge", "eaves", "soffit", "fascia", "felt", "flashing", "parapet",
        "covering", "rafter", "purlin", "truss", "flat roof", "pitched roof",
    }),
    "chimney": frozenset({
        "chimney", "chimney stack", "flaunching", "pot", "cowl", "flue",
        "breast", "corbel", "pointing to the stack",
    }),
    "rainwater": frozenset({
        "gutter", "guttering", "downpipe", "rainwater", "hopper", "rwp",
        "rainwater goods", "fall pipe",
    }),
    "drainage": frozenset({
        "drain", "drainage", "sewer", "manhole", "inspection chamber", "gully",
        "soil pipe", "foul", "surface water", "septic", "soakaway", "below ground",
    }),
    "walls": frozenset({
        "wall", "masonry", "brickwork", "blockwork", "render", "rendering",
        "pointing", "cavity", "spalling", "cracking", "lintel", "dpc",
        "damp proof course", "external wall", "load bearing",
    }),
    "dampness": frozenset({
        "damp", "dampness", "moisture", "rising damp", "penetrating damp",
        "condensation", "mould", "mold", "hygroscopic", "salts", "tide mark",
    }),
    "timber": frozenset({
        "timber", "woodworm", "beetle", "wet rot", "dry rot", "joist", "decay",
        "fungal", "infestation", "rot to the",
    }),
    "windows_doors": frozenset({
        "window", "windows", "door", "doors", "glazing", "double glazing",
        "frame", "casement", "sash", "joinery", "fenestration", "sill",
    }),
    "ceilings_floors": frozenset({
        "ceiling", "floor", "flooring", "floorboard", "screed", "plaster",
        "lath", "cornice", "skirting", "subfloor",
    }),
    "electrical": frozenset({
        "electric", "electrical", "wiring", "consumer unit", "fuse box",
        "rcd", "socket", "circuit", "earthing", "eicr", "fixed wiring",
        "distribution board",
    }),
    "heating": frozenset({
        "boiler", "heating", "radiator", "central heating", "flue", "gas",
        "thermostat", "hot water", "cylinder", "underfloor heating", "combi",
    }),
    "plumbing": frozenset({
        "plumbing", "pipework", "water supply", "stopcock", "waste pipe",
        "mains water", "lead pipe", "tank", "overflow", "sanitary",
    }),
    "insulation_energy": frozenset({
        "insulation", "epc", "energy performance", "sap", "u-value",
        "thermal", "loft insulation", "cavity insulation", "energy efficiency",
    }),
    "grounds": frozenset({
        "garden", "boundary", "fence", "fencing", "patio", "driveway", "path",
        "tree", "hedge", "retaining wall", "outbuilding", "grounds",
    }),
    "services_other": frozenset({
        "ventilation", "extractor", "smoke alarm", "carbon monoxide",
        "asbestos", "fire", "security",
    }),
}

# Map common RICS section titles / template hints to a canonical domain.
_SECTION_ALIASES: dict[str, str] = {
    "roof": "roofing", "roof coverings": "roofing", "main roof": "roofing",
    "chimney stacks": "chimney", "chimneys": "chimney",
    "rainwater pipes and gutters": "rainwater", "gutters": "rainwater",
    "drainage": "drainage",
    "main walls": "walls", "external walls": "walls", "walls": "walls",
    "dampness": "dampness", "damp": "dampness",
    "windows": "windows_doors", "doors": "windows_doors",
    "ceilings": "ceilings_floors", "floors": "ceilings_floors",
    "electricity": "electrical", "electrical": "electrical",
    "heating": "heating", "gas": "heating",
    "water": "plumbing", "plumbing": "plumbing",
    "insulation": "insulation_energy", "energy efficiency": "insulation_energy",
    "grounds": "grounds", "gardens": "grounds", "boundaries": "grounds",
}


def _score_domains(text: str) -> dict[str, int]:
    norm = normalize_for_match(text)
    if not norm:
        return {}
    scores: dict[str, int] = {}
    for domain, terms in DOMAIN_LEXICON.items():
        hits = 0
        for term in terms:
            # Count occurrences; multi-word terms matched as substrings.
            if " " in term:
                hits += norm.count(term)
            else:
                # word-ish boundary check to avoid 'gas' in 'gasket' etc.
                hits += _count_word(norm, term)
        if hits:
            scores[domain] = hits
    return scores


def _count_word(haystack: str, word: str) -> int:
    count = 0
    start = 0
    n = len(word)
    while True:
        idx = haystack.find(word, start)
        if idx == -1:
            break
        before = haystack[idx - 1] if idx > 0 else " "
        after = haystack[idx + n] if idx + n < len(haystack) else " "
        if not before.isalnum() and not after.isalnum():
            count += 1
        start = idx + n
    return count


def classify_text(text: str) -> str:
    """Return the dominant domain for ``text`` or :data:`GENERAL`."""
    scores = _score_domains(text)
    if not scores:
        return GENERAL
    # Highest score wins; ties broken by lexicon insertion order (stable).
    best = max(DOMAIN_LEXICON.keys(), key=lambda d: scores.get(d, 0))
    return best if scores.get(best, 0) > 0 else GENERAL


def classify_section(section_name: str | None, template_id: str | None = None) -> str:
    """Resolve a section title/template into a canonical domain.

    Falls back to keyword classification of the section name, then
    :data:`GENERAL`.
    """
    name = normalize_for_match(section_name or "")
    if name in _SECTION_ALIASES:
        return _SECTION_ALIASES[name]
    for alias, domain in _SECTION_ALIASES.items():
        if alias in name:
            return domain
    guessed = classify_text(section_name or "")
    return guessed


def scope_chunks(
    section_domain: str,
    chunks: list[object],
    *,
    strict: bool = True,
) -> list[object]:
    """Filter retrieved chunks to those admissible for ``section_domain``.

    A chunk is admissible when its dominant domain equals ``section_domain`` or
    is :data:`GENERAL`. Chunks clearly belonging to a *different* domain are
    excluded. If filtering would remove everything (e.g. weak classification),
    the original list is returned to avoid starving extraction — unless
    ``strict`` is False, in which case the filtered (possibly empty) list is
    always returned.

    ``chunks`` are SearchResult-like objects exposing ``.text``.
    """
    if section_domain == GENERAL or not chunks:
        return chunks
    kept: list[object] = []
    for c in chunks:
        text = getattr(c, "text", "") or ""
        cdom = classify_text(text)
        if cdom == section_domain or cdom == GENERAL:
            kept.append(c)
    if not kept and strict:
        # Never starve the extractor on a misclassification; better to extract
        # from broader context than to silently drop the whole section.
        return chunks
    return kept