"""Span attributes demo — entities carry per-span attribute groups (sentiment,
role, assertion, impact), single- or multi-label, scoped with applies_to.

MOCK until GLiNER 2.5 ships. Response contract:
  {"spans": [{"start", "end", "text", "label", "score",
              "attributes": {"<group>": [{"label": str, "score": float}]}}]}
  Multi-label groups return several entries in the group list.
"""

import json
import os

MOCK = os.environ.get("GLINER_MOCK", "1") == "1"
MODEL_ID = os.environ.get("MODEL_ID", "fastino/gliner2.5")
MODEL_URL = os.environ.get("MODEL_URL", f"https://huggingface.co/{MODEL_ID}")

DEMO = {
    "title": "Sentiment and role, <em>per span</em>",
    "subtitle": "Document classification labels the whole review. GLiNER 2.5 attaches <strong>attributes to every entity</strong> it extracts.",
    "model_name": "GLiNER 2.5",
    "model_variant": "0.3B parameters · f32 · CPU",
    "code": (
        'schema = (model.create_schema()\n'
        '    .entities({"product": "A product mentioned in the text",\n'
        '               "person": "A person by name",\n'
        '               "condition": "A medical condition"})\n'
        '    .entity_attributes({\n'
        '        "sentiment": AttributeGroup(\n'
        '            labels=["positive", "neutral", "negative"],\n'
        '            applies_to=["product"], qualify_labels=True),\n'
        '        "role": AttributeGroup(\n'
        '            labels=["executive", "employee", "customer", "analyst"],\n'
        '            applies_to=["person"]),\n'
        '        "assertion": AttributeGroup(\n'
        '            labels=["present", "absent", "possible", "historical"],\n'
        '            applies_to=["condition"]),\n'
        '        "impact": AttributeGroup(\n'
        '            labels=["blocks_work", "data_loss", "security_risk"],\n'
        '            multi_label=True, threshold=0.40)}))\n'
        'model.extract(text, schema)'
    ),
    "examples": [
        {"chip": "Mixed review",
         "text": "The screen is gorgeous and the keyboard feels great, but the battery is disappointing and the fan noise is unacceptable.",
         "fixture": "review.json"},
        {"chip": "Org announcement",
         "text": "CEO Maya Chen announced that CFO Daniel Okafor will lead the acquisition, while analyst Priya Nair briefed reporters.",
         "fixture": "org.json"},
        {"chip": "Clinical note",
         "text": "Patient denies chest pain. No evidence of pneumonia. Possible mild anemia; history of hypertension noted.",
         "fixture": "clinical.json"},
        {"chip": "Support ticket",
         "text": "The export button silently deletes rows, which blocked our quarterly report and risks losing audited data.",
         "fixture": "support.json"},
    ],
}

_HERE = os.path.dirname(os.path.abspath(__file__))


def warmup():
    if MOCK:
        for ex in DEMO["examples"]:
            _fixture(ex["fixture"], ex["text"])
    else:
        _load_real()


def infer(data):
    text = (data.get("text") or "").replace("\r", "")
    if not text.strip():
        return {"spans": []}
    if MOCK:
        return _infer_mock(text)
    return _infer_real(text)


def _fixture(name, expect_text=None):
    with open(os.path.join(_HERE, "fixtures", name)) as f:
        out = json.load(f)
    if expect_text is not None:
        for sp in out.get("spans", []):
            assert expect_text[sp["start"]:sp["end"]] == sp["text"], \
                f"{name}: bad offsets for {sp['text']!r}"
    return out


def _infer_mock(text):
    for ex in DEMO["examples"]:
        if ex["text"] == text:
            return _fixture(ex["fixture"])
    return {"spans": []}


_model = None


def _load_real():
    global _model
    from gliner import GLiNER
    _model = GLiNER.from_pretrained(MODEL_ID, token=os.environ.get("HF_TOKEN"))


def _infer_real(text):
    raise NotImplementedError("wire up GLiNER 2.5 here, keep the response shape")