1
0
mirror of https://github.com/searxng/searxng.git synced 2024-11-09 06:30:10 +01:00
searxng/searx/engines/pdbe.py
Markus Heiser 3d96a9839a [format.python] initial formatting of the python code
This patch was generated by black [1]::

    make format.python

[1] https://github.com/psf/black

Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
2021-12-27 09:26:22 +01:00

123 lines
3.7 KiB
Python

# SPDX-License-Identifier: AGPL-3.0-or-later
"""
PDBe (Protein Data Bank in Europe)
"""
from json import loads
from flask_babel import gettext
# about
about = {
"website": 'https://www.ebi.ac.uk/pdbe',
"wikidata_id": 'Q55823905',
"official_api_documentation": 'https://www.ebi.ac.uk/pdbe/api/doc/search.html',
"use_official_api": True,
"require_api_key": False,
"results": 'JSON',
}
categories = ['science']
hide_obsolete = False
# status codes of unpublished entries
pdb_unpublished_codes = ['HPUB', 'HOLD', 'PROC', 'WAIT', 'AUTH', 'AUCO', 'REPL', 'POLC', 'REFI', 'TRSF', 'WDRN']
# url for api query
pdbe_solr_url = 'https://www.ebi.ac.uk/pdbe/search/pdb/select?'
# base url for results
pdbe_entry_url = 'https://www.ebi.ac.uk/pdbe/entry/pdb/{pdb_id}'
# link to preview image of structure
pdbe_preview_url = 'https://www.ebi.ac.uk/pdbe/static/entry/{pdb_id}_deposited_chain_front_image-200x200.png'
def request(query, params):
params['url'] = pdbe_solr_url
params['method'] = 'POST'
params['data'] = {'q': query, 'wt': "json"} # request response in parsable format
return params
def construct_body(result):
# set title
title = result['title']
# construct content body
content = """{title} - {authors} {journal} ({volume}) {page} ({year})"""
# replace placeholders with actual content
try:
if result['journal']:
content = content.format(
title=result['citation_title'],
authors=result['entry_author_list'][0],
journal=result['journal'],
volume=result['journal_volume'],
page=result['journal_page'],
year=result['citation_year'],
)
else:
content = content.format(
title=result['citation_title'],
authors=result['entry_author_list'][0],
journal='',
volume='',
page='',
year=result['release_year'],
)
img_src = pdbe_preview_url.format(pdb_id=result['pdb_id'])
except (KeyError):
content = None
img_src = None
# construct url for preview image
try:
img_src = pdbe_preview_url.format(pdb_id=result['pdb_id'])
except (KeyError):
img_src = None
return [title, content, img_src]
def response(resp):
results = []
json = loads(resp.text)['response']['docs']
# parse results
for result in json:
# catch obsolete entries and mark them accordingly
if result['status'] in pdb_unpublished_codes:
continue
if hide_obsolete:
continue
if result['status'] == 'OBS':
# expand title to add some sort of warning message
title = gettext('{title} (OBSOLETE)').format(title=result['title'])
try:
superseded_url = pdbe_entry_url.format(pdb_id=result['superseded_by'])
except:
continue
# since we can't construct a proper body from the response, we'll make up our own
msg_superseded = gettext("This entry has been superseded by")
content = '{msg_superseded}: {url} ({pdb_id})'.format(
msg_superseded=msg_superseded, url=superseded_url, pdb_id=result['superseded_by']
)
# obsoleted entries don't have preview images
img_src = None
else:
title, content, img_src = construct_body(result)
results.append(
{
'url': pdbe_entry_url.format(pdb_id=result['pdb_id']),
'title': title,
'content': content,
'img_src': img_src,
}
)
return results