xbrl_processing/parser.py

# xbrl_processing/parser.py
from .arelle_loader import load_model
from .fact_extractor import get_fact

from .taxonomy_map import (
    REVISION_TYPE,
    AUDITOR_DESCRIPTION,
    MAIN_ACTIVITY,
    MATERIAL_ERROR_CORRECTION,
    GOING_CONCERN,
    ACCOUNTING_CLASS,
    ACCOUNTING_CLASS_UPGRADE,
)

def _get_ixbrl_non_numeric(model, names):
    """
    Extract text from ix:nonNumeric facts in XHTML/iXBRL.
    """
    if not hasattr(model, "ixFacts"):
        return None

    for ix in model.ixFacts:
        local = getattr(ix.qname, "localName", None)
        if not local:
            continue

        if local in names:
            text = ix.value
            if text:
                return text.strip()

    return None

def _find_first(model, names):
    # Try normal XBRL facts first
    for name in names:
        val = get_fact(model, name)
        if val:
            return val

    # Fallback: look in iXBRL nonNumeric tags
    ix_val = _get_ixbrl_non_numeric(model, names)
    if ix_val:
        return ix_val

    return None


def _clean_activity(text: str) -> str:
    if not text:
        return text

    # Remove soft hyphens and weird invisible characters
    text = text.replace("\u00AD", "")   # soft hyphen
    text = text.replace("\u2011", "-")  # non-breaking hyphen

    # Normalize whitespace
    t = " ".join(text.split())

    # Case 1 — prefix at the beginning
    prefix = "Selskabets væsentligste aktiviteter"
    if t.startswith(prefix):
        t = t[len(prefix):].lstrip(" .")

    # Case 2 — prefix concatenated inside the string ("aktiviteterSelskabets")
    t = t.replace(prefix + " ", "")
    t = t.replace(prefix, "")

    # Final trim
    return t.strip()

def _normalize_revisionstype(value: str) -> str:
    """
    Normalize various wording forms of revision types into:
    - 'Revision'
    - 'Udvidet Gennemgang'
    - 'Assistance'
    - 'Ingen bistand'
    - 'Review'  (rare but supported)
    """
    if not value:
        return None

    v = value.lower().strip()

    # -------------------------
    # 1) INGEN BISTAND
    # -------------------------
    if any(x in v for x in [
        "ingen bistand",
        "ingen",                       # generic
        "uden revisor",
        "ikke revideret",
        "uden revision",
        "uden gennemgang",
        "uden erklæring"
    ]):
        return "Ingen bistand"

    # -------------------------
    # 2) ASSISTANCE (uden sikkerhed)
    # -------------------------
    if any(x in v for x in [
        "andre erklæringer uden sikkerhed",
        "uden sikkerhed",
        "assistance",
        "revisorassistance",
        "udarbejdet med bistand",
        "udarbejdet uden",  # e.g. "udarbejdet uden revisor"
        "kompilering"
    ]):
        return "Assistance"

    # -------------------------
    # 3) UDV. GENNEMGANG
    # -------------------------
    if any(x in v for x in [
        "erklæring om udvidet gennemgang",
        "udvidet gennemgang",
        "udv. gennemgang",
        "gennemgangserklæring",
        "gennemgang"  # careful but safe: most Danish uses are "udvidet gennemgang"
    ]):
        return "Udvidet Gennemgang"

    # -------------------------
    # 4) REVISION
    # -------------------------
    if any(x in v for x in [
        "revisionspåtegning",
        "revision",
        "revideret regnskab",
        "aflagt med revision",
        "revisionsfirma",
        "audited financial statements",
    ]):
        return "Revision"

    # -------------------------
    # 5) REVIEW (rare, English)
    # -------------------------
    if "review" in v:
        return "Review"

    # -------------------------
    # Fallback — unknown label
    # -------------------------
    return value

def _normalize_revisortype(value: str) -> str:
    """
    Normalize wording of 'Revisortype' into:
    - 'Statsautoriseret revisor'
    - 'Registreret revisor'
    - 'Ingen revisor'
    - 'Andet'
    """
    if not value:
        return None

    v = value.lower().strip()

    # -------------------------
    # 1) INGEN REVISOR
    # -------------------------
    if any(x in v for x in [
        "ingen", 
        "uden revisor",
        "ikke revideret",
        "ikke valgt revisor",
        "fravalgt revision",
        "fravalg af revisor",
        "uden revision"
    ]):
        return "Ingen revisor"

    # -------------------------
    # 2) STATSautorisERet REVIsor
    # -------------------------
    if any(x in v for x in [
        "statsautoriseret",
        "state-authorised",
        "state authorised",
        "statsaut."
    ]):
        return "Statsautoriseret revisor"

    # -------------------------
    # 3) REGISTRERET REVISOR
    # -------------------------
    if any(x in v for x in [
        "registreret revisor",
        "reg. revisor",
        "registreret"
    ]):
        return "Registreret revisor"

    # -------------------------
    # 4) ANDRE – side cases
    # -------------------------
    # Some reports include the firm name but no type:
    # e.g. "Deloitte", "PwC", "BDO"
    if any(x in v for x in [
        "deloitte", "pwc", "bdo", "ey", "grant thornton", 
        "martinsen", "kpmg"
    ]):
        # Usually a statsaut. firm but we cannot assume.
        return "Andet"

    # -------------------------
    # Fallback
    # -------------------------
    return "Andet"

def _extract_periods_from_dcca_tags(model):
    """
    Extract CY/PY using Danish GAAP period tags.
    Same logic as in financial_parser.
    """
    cy_start = None
    cy_end = None
    py_start = None
    py_end = None

    for fact in model.facts:
        name = fact.qname.localName

        if name == "ReportingPeriodStartDate":
            cy_start = fact.value
        elif name == "ReportingPeriodEndDate":
            cy_end = fact.value
        elif name == "PrecedingReportingPeriodStartDate":
            py_start = fact.value
        elif name == "PredingReportingPeriodEndDate":  # DCCA typo
            py_end = fact.value

    return {
        "CY": {"start": cy_start, "end": cy_end},
        "PY": {"start": py_start, "end": py_end},
    }

def extract_xbrl_data(filepath: str) -> dict:
    """
    Parse XBRL/iXBRL file with Arelle and extract general qualitative facts.
    No ML, no SBERT — pure taxonomy-based extraction.
    """
    try:
        model = load_model(filepath)

        data = {
            # Revision info
            "Revisionstype": _normalize_revisionstype(_find_first(model, REVISION_TYPE)),
            "Revisortype": _normalize_revisortype(_find_first(model, AUDITOR_DESCRIPTION)),

            # Company activity description
            "Væsentlig aktivitet": _clean_activity(_find_first(model, MAIN_ACTIVITY)),

            # Corrections of material errors
            "Korrektion af væsentlig fejl": _find_first(model, MATERIAL_ERROR_CORRECTION),

            # Going concern
            "Going concern usikkerhed": _find_first(model, GOING_CONCERN),

            # Accounting class
            "Anvendt regnskabsklasse": _find_first(model, ACCOUNTING_CLASS),

            # Optional use of higher accounting class
            "Tilvalg af højere regnskabsklasse": _find_first(model, ACCOUNTING_CLASS_UPGRADE),
        }
        data["Years"] = _extract_periods_from_dcca_tags(model)

        return data

    except Exception as e:
        print("[Fejl] XBRL parsing:", e)
        return {"Fejl": str(e)}