"""Full-text search with a narrow fallback for stop-word abbreviations."""

import re

from django.db import connections
from watson.backends import PostgresSearchBackend
from watson.models import SearchEntry
from watson.search import search as watson_search


class AbbreviationPostgresSearchBackend(PostgresSearchBackend):
    def do_search(self, engine_slug, queryset, search_text):
        # English stemming drops short names such as IT as stop words. Only
        # empty, short single-token tsqueries need a literal title fallback.
        # Normal queries retain Watson's GIN-backed full-text search unchanged.
        if 2 <= len(search_text) <= 8 and search_text.isalnum():
            with connections[queryset.db].cursor() as cursor:
                cursor.execute(
                    "SELECT numnode(to_tsquery(%s::regconfig, %s)) = 0",
                    (self.search_config, self.escape_postgres_query(search_text)),
                )
                is_stop_word = cursor.fetchone()[0]
            if is_stop_word:
                # Match an entire token, not 'it' inside 'Security'. Do not scan
                # descriptions or extracted document contents for stop words.
                pattern = rf"(^|[^[:alnum:]_]){re.escape(search_text)}([^[:alnum:]_]|$)"
                return queryset.filter(title__iregex=pattern)
        return super().do_search(engine_slug, queryset, search_text)


def search_with_abbreviations(search_text):
    backend_name = None
    if connections[SearchEntry.objects.db].vendor == "postgresql":
        backend_name = "search.backends.AbbreviationPostgresSearchBackend"
    # Watson still applies its engine/model/live-queryset filters. Candidates
    # also go through the API's normal object ACL, counting and pagination.
    return watson_search(search_text, backend_name=backend_name)