"""Anonimizzatore Gare d'Appalto — UI Gradio."""
import json
import gradio as gr
from config import (
MODES, DEFAULT_MIN_SCORE,
SEVERITY_ORDER, SEVERITY_LABELS, SEVERITY_COLORS,
SEVERITY_CRITICAL, SEVERITY_HIGH, SEVERITY_MEDIUM, SEVERITY_LOW,
get_severity,
)
from core.pipeline import detect, anonymize_from_entities, apply_custom_lines
from ui.renderers import (
render_highlighted_text, render_anonymized_text,
render_categorized_report, build_json_by_severity,
)
from utils import extract_pdf_text, extract_pdf_text_got_ocr, extract_pdf_text_smoldocling
from demo_text import DEMO_TEXT, DEMO_REGEX, DEMO_GLINER, DEMO_NER
_DEMO_TEXTS = {
"Bando completo": DEMO_TEXT,
"Layer 0 — Regex": DEMO_REGEX,
"Layer 1 — NER": DEMO_NER,
"Layer 2 — GLiNER": DEMO_GLINER,
}
_PDF_EXTRACTORS = {
"PyMuPDF (default)": extract_pdf_text,
"GOT-OCR 2.0 (~30s/pag, CPU)": extract_pdf_text_got_ocr,
"SmolDocling 256M (CPU)": extract_pdf_text_smoldocling,
}
_STATE_DEFAULT = {"orig_text": "", "mode": "", "anon_text": "", "entities": []}
_ALL_SEVERITIES = list(SEVERITY_ORDER)
_EMPTY_JSON = "{}"
# ---------------------------------------------------------------------------
# Helpers
# ---------------------------------------------------------------------------
def _active_set(active_severities: list[str]) -> set[str]:
return set(active_severities) if active_severities is not None else set(_ALL_SEVERITIES)
# ---------------------------------------------------------------------------
# Handlers
# ---------------------------------------------------------------------------
def _process(text: str, mode: str, min_score: float, use_regex: bool, active_severities: list[str]):
"""Esegue il pipeline completo e restituisce i 4 output Gradio.
Le entità rilevate vengono salvate nello stato in modo che variazioni
successive di modalità o gravità possano ricalcolare l'output senza
rieseguire il costoso step di rilevamento NLP.
"""
active = _active_set(active_severities)
entities = detect(text, min_score=min_score, use_regex=use_regex)
filtered = [e for e in entities if get_severity(e.entity_type) in active]
anon_text = anonymize_from_entities(text, filtered, mode)
new_state = {"orig_text": text, "mode": mode, "anon_text": anon_text, "entities": entities}
json_c, json_h, json_m, json_l = build_json_by_severity(text, entities, active)
return (
render_highlighted_text(text, entities, active),
render_anonymized_text(anon_text),
render_categorized_report(text, entities, active),
new_state,
json_c, json_h, json_m, json_l,
)
def handle_settings_change(state_data: dict, mode: str, c: bool, h: bool, m: bool, l: bool):
"""Ricalcola l'output quando l'utente cambia modalità o livelli di gravità.
Non riesegue il rilevamento NLP: usa le entità già salvate nello stato.
"""
entities = state_data.get("entities") or []
orig_text = state_data.get("orig_text") or ""
if not orig_text or not entities:
return (gr.update(), gr.update(), gr.update(), state_data,
gr.update(), gr.update(), gr.update(), gr.update())
active = _active_set([s for s, on in zip(SEVERITY_ORDER, [c, h, m, l]) if on])
filtered = [e for e in entities if get_severity(e.entity_type) in active]
anon_text = anonymize_from_entities(orig_text, filtered, mode)
new_state = {**state_data, "mode": mode, "anon_text": anon_text}
json_c, json_h, json_m, json_l = build_json_by_severity(orig_text, entities, active)
return (
render_highlighted_text(orig_text, entities, active),
render_anonymized_text(anon_text),
render_categorized_report(orig_text, entities, active),
new_state,
json_c, json_h, json_m, json_l,
)
def handle_text(text, mode, min_score, use_regex, active_severities):
_empty = (
'Inserisci del testo per iniziare.
'
)
if not text or not text.strip():
return (_empty, _empty, _empty, _STATE_DEFAULT.copy(),
_EMPTY_JSON, _EMPTY_JSON, _EMPTY_JSON, _EMPTY_JSON)
return _process(text, mode, min_score, use_regex, active_severities)
def handle_pdf(pdf_file, mode, min_score, use_regex, active_severities, extractor_label="PyMuPDF (default)"):
def _err(msg):
e = (
f'{msg}
'
)
return (e, e, e, _STATE_DEFAULT.copy(),
_EMPTY_JSON, _EMPTY_JSON, _EMPTY_JSON, _EMPTY_JSON)
if pdf_file is None:
return _err("Carica un PDF per iniziare.")
extractor = _PDF_EXTRACTORS.get(extractor_label, extract_pdf_text)
raw = extractor(pdf_file)
if not raw.strip():
return _err("⚠️ Impossibile estrarre testo dal PDF (scansionato?).")
return _process(raw, mode, min_score, use_regex, active_severities)
def handle_update_custom(selected_lines_json: str, state_data: dict):
"""Applica le righe selezionate (via JS al click) al testo anonimizzato."""
lines: list[str] = json.loads(selected_lines_json or "[]")
if not lines or not state_data.get("orig_text"):
return render_anonymized_text(state_data.get("anon_text", ""))
new_anon = apply_custom_lines(
state_data["anon_text"],
state_data["orig_text"],
lines,
state_data["mode"],
)
return render_anonymized_text(new_anon)
# ---------------------------------------------------------------------------
# JavaScript
# ---------------------------------------------------------------------------
_PAGE_JS = """
(function () {
/* ── Overlay fullscreen di loading ─────────────────────────────────── */
var ov = document.createElement('div');
ov.id = 'anon-fs-overlay';
ov.innerHTML = [
'',
'
',
'
Analisi in corso…
',
'