forked from mirrors/bookwyrm
156 lines
4.9 KiB
Python
156 lines
4.9 KiB
Python
""" using a bookwyrm instance as a source of book data """
|
|
from dataclasses import asdict, dataclass
|
|
from functools import reduce
|
|
import operator
|
|
|
|
from django.contrib.postgres.search import SearchRank, SearchQuery
|
|
from django.db.models import OuterRef, Subquery, F, Q
|
|
|
|
from bookwyrm import models
|
|
from bookwyrm.settings import MEDIA_FULL_URL
|
|
|
|
|
|
# pylint: disable=arguments-differ
|
|
def search(query, min_confidence=0, filters=None, return_first=False):
|
|
"""search your local database"""
|
|
filters = filters or []
|
|
if not query:
|
|
return []
|
|
# first, try searching unqiue identifiers
|
|
results = search_identifiers(query, *filters, return_first=return_first)
|
|
if not results:
|
|
# then try searching title/author
|
|
results = search_title_author(
|
|
query, min_confidence, *filters, return_first=return_first
|
|
)
|
|
return results
|
|
|
|
|
|
def isbn_search(query):
|
|
"""search your local database"""
|
|
if not query:
|
|
return []
|
|
|
|
filters = [{f: query} for f in ["isbn_10", "isbn_13"]]
|
|
results = models.Edition.objects.filter(
|
|
reduce(operator.or_, (Q(**f) for f in filters))
|
|
).distinct()
|
|
|
|
# when there are multiple editions of the same work, pick the default.
|
|
# it would be odd for this to happen.
|
|
|
|
default_editions = models.Edition.objects.filter(
|
|
parent_work=OuterRef("parent_work")
|
|
).order_by("-edition_rank")
|
|
results = (
|
|
results.annotate(default_id=Subquery(default_editions.values("id")[:1])).filter(
|
|
default_id=F("id")
|
|
)
|
|
or results
|
|
)
|
|
return results
|
|
|
|
|
|
def format_search_result(search_result):
|
|
"""convert a book object into a search result object"""
|
|
cover = None
|
|
if search_result.cover:
|
|
cover = f"{MEDIA_FULL_URL}{search_result.cover}"
|
|
|
|
return SearchResult(
|
|
title=search_result.title,
|
|
key=search_result.remote_id,
|
|
author=search_result.author_text,
|
|
year=search_result.published_date.year
|
|
if search_result.published_date
|
|
else None,
|
|
cover=cover,
|
|
confidence=search_result.rank if hasattr(search_result, "rank") else 1,
|
|
connector="",
|
|
).json()
|
|
|
|
|
|
def search_identifiers(query, *filters, return_first=False):
|
|
"""tries remote_id, isbn; defined as dedupe fields on the model"""
|
|
# pylint: disable=W0212
|
|
or_filters = [
|
|
{f.name: query}
|
|
for f in models.Edition._meta.get_fields()
|
|
if hasattr(f, "deduplication_field") and f.deduplication_field
|
|
]
|
|
results = models.Edition.objects.filter(
|
|
*filters, reduce(operator.or_, (Q(**f) for f in or_filters))
|
|
).distinct()
|
|
if results.count() <= 1:
|
|
return results
|
|
|
|
# when there are multiple editions of the same work, pick the default.
|
|
# it would be odd for this to happen.
|
|
default_editions = models.Edition.objects.filter(
|
|
parent_work=OuterRef("parent_work")
|
|
).order_by("-edition_rank")
|
|
results = (
|
|
results.annotate(default_id=Subquery(default_editions.values("id")[:1])).filter(
|
|
default_id=F("id")
|
|
)
|
|
or results
|
|
)
|
|
if return_first:
|
|
return results.first()
|
|
return results
|
|
|
|
|
|
def search_title_author(query, min_confidence, *filters, return_first=False):
|
|
"""searches for title and author"""
|
|
query = SearchQuery(query, config="simple") | SearchQuery(query, config="english")
|
|
results = (
|
|
models.Edition.objects.filter(*filters, search_vector=query)
|
|
.annotate(rank=SearchRank(F("search_vector"), query))
|
|
.filter(rank__gt=min_confidence)
|
|
.order_by("-rank")
|
|
)
|
|
|
|
# when there are multiple editions of the same work, pick the closest
|
|
editions_of_work = results.values("parent_work__id").values_list("parent_work__id")
|
|
|
|
# filter out multiple editions of the same work
|
|
list_results = []
|
|
for work_id in set(editions_of_work):
|
|
editions = results.filter(parent_work=work_id)
|
|
default = editions.order_by("-edition_rank").first()
|
|
default_rank = default.rank if default else 0
|
|
# if mutliple books have the top rank, pick the default edition
|
|
if default_rank == editions.first().rank:
|
|
result = default
|
|
else:
|
|
result = editions.first()
|
|
if return_first:
|
|
return result
|
|
list_results.append(result)
|
|
return list_results
|
|
|
|
|
|
@dataclass
|
|
class SearchResult:
|
|
"""standardized search result object"""
|
|
|
|
title: str
|
|
key: str
|
|
connector: object
|
|
view_link: str = None
|
|
author: str = None
|
|
year: str = None
|
|
cover: str = None
|
|
confidence: int = 1
|
|
|
|
def __repr__(self):
|
|
# pylint: disable=consider-using-f-string
|
|
return "<SearchResult key={!r} title={!r} author={!r}>".format(
|
|
self.key, self.title, self.author
|
|
)
|
|
|
|
def json(self):
|
|
"""serialize a connector for json response"""
|
|
serialized = asdict(self)
|
|
del serialized["connector"]
|
|
return serialized
|