MarseyWorld/drama/routes/search.py

324 lines
8.3 KiB
Python
Raw Normal View History

2021-07-22 19:19:49 +00:00
from drama.helpers.wrappers import *
2021-07-21 01:12:26 +00:00
import re
from sqlalchemy import *
from flask import *
2021-07-22 19:19:49 +00:00
from drama.classes.domains import reasons as REASONS
from drama.__main__ import app, cache
2021-07-21 01:12:26 +00:00
import random
query_regex=re.compile("(\w+):(\S+)")
valid_params=[
'author',
'domain',
'over18'
]
def searchparse(text):
#takes test in filter:term format and returns data
criteria = {x[0]:x[1] for x in query_regex.findall(text)}
for x in criteria:
if x in valid_params:
text = text.replace(f"{x}:{criteria[x]}", "")
text=text.strip()
if text:
criteria['q']=text
return criteria
@cache.memoize(300)
def searchlisting(criteria, v=None, page=1, t="None", sort="top", b=None):
posts = g.db.query(Submission).options(
lazyload('*')
).join(
Submission.submission_aux,
).join(
Submission.author
)
if 'q' in criteria:
words=criteria['q'].split()
words=[SubmissionAux.title.ilike('%'+x+'%') for x in words]
words=tuple(words)
posts=posts.filter(*words)
if 'over18' in criteria:
posts = posts.filter(Submission.over_18==True)
if 'author' in criteria:
posts=posts.filter(
Submission.author_id==get_user(criteria['author']).id,
User.is_private==False,
User.is_deleted==False
)
if 'domain' in criteria:
domain=criteria['domain']
posts=posts.filter(
or_(
SubmissionAux.url.ilike("https://"+domain+'/%'),
SubmissionAux.url.ilike("https://"+domain+'/%'),
SubmissionAux.url.ilike("https://"+domain),
SubmissionAux.url.ilike("https://"+domain),
SubmissionAux.url.ilike("https://www."+domain+'/%'),
SubmissionAux.url.ilike("https://www."+domain+'/%'),
SubmissionAux.url.ilike("https://www."+domain),
SubmissionAux.url.ilike("https://www."+domain),
SubmissionAux.url.ilike("https://old." + domain + '/%'),
SubmissionAux.url.ilike("https://old." + domain + '/%'),
SubmissionAux.url.ilike("https://old." + domain),
SubmissionAux.url.ilike("https://old." + domain)
)
)
if not(v and v.admin_level >= 3):
posts = posts.filter(
Submission.deleted_utc == 0,
Submission.is_banned == False,
)
if v and v.admin_level >= 4:
pass
elif v:
blocking = g.db.query(
UserBlock.target_id).filter_by(
user_id=v.id).subquery()
blocked = g.db.query(
UserBlock.user_id).filter_by(
target_id=v.id).subquery()
posts = posts.filter(
Submission.author_id.notin_(blocking),
Submission.author_id.notin_(blocked),
)
if t:
now = int(time.time())
if t == 'hour':
cutoff = now - 3600
elif t == 'day':
cutoff = now - 86400
elif t == 'week':
cutoff = now - 604800
elif t == 'month':
cutoff = now - 2592000
elif t == 'year':
cutoff = now - 31536000
else:
cutoff = 0
posts = posts.filter(Submission.created_utc >= cutoff)
posts=posts.options(
contains_eager(Submission.submission_aux),
contains_eager(Submission.author),
)
if sort == "new":
posts = posts.order_by(Submission.created_utc.desc()).all()
elif sort == "old":
posts = posts.order_by(Submission.created_utc.asc()).all()
elif sort == "controversial":
posts = sorted(posts.all(), key=lambda x: x.score_disputed, reverse=True)
elif sort == "top":
posts = posts.order_by(Submission.score.desc()).all()
elif sort == "bottom":
posts = posts.order_by(Submission.score.asc()).all()
elif sort == "comments":
posts = posts.order_by(Submission.comment_count.desc()).all()
elif sort == "random":
posts = posts.all()
posts = random.sample(posts, k=len(posts))
else:
abort(400)
total = len(posts)
firstrange = 25 * (page - 1)
secondrange = firstrange+26
posts = posts[firstrange:secondrange]
return total, [x.id for x in posts]
@cache.memoize(300)
def searchcommentlisting(criteria, v=None, page=1, t="None", sort="top"):
comments = g.db.query(Comment).options(lazyload('*')).filter(Comment.parent_submission != None).join(Comment.comment_aux)
if 'q' in criteria:
words=criteria['q'].split()
words=[CommentAux.body.ilike('%'+x+'%') for x in words]
words=tuple(words)
comments=comments.filter(*words)
if not(v and v.admin_level >= 3):
comments = comments.filter(
Comment.deleted_utc == 0,
Comment.is_banned == False)
if t:
now = int(time.time())
if t == 'hour':
cutoff = now - 3600
elif t == 'day':
cutoff = now - 86400
elif t == 'week':
cutoff = now - 604800
elif t == 'month':
cutoff = now - 2592000
elif t == 'year':
cutoff = now - 31536000
else:
cutoff = 0
comments = comments.filter(Comment.created_utc >= cutoff)
comments=comments.options(contains_eager(Comment.comment_aux))
if sort == "new":
comments = comments.order_by(Comment.created_utc.desc()).all()
elif sort == "old":
comments = comments.order_by(Comment.created_utc.asc()).all()
elif sort == "controversial":
comments = sorted(comments.all(), key=lambda x: x.score_disputed, reverse=True)
elif sort == "top":
comments = comments.order_by(Comment.score.desc()).all()
elif sort == "bottom":
comments = comments.order_by(Comment.score.asc()).all()
total = len(list(comments))
firstrange = 25 * (page - 1)
secondrange = firstrange+26
comments = comments[firstrange:secondrange]
return total, [x.id for x in comments]
@app.route("/search/posts", methods=["GET"])
@app.route("/api/v1/search", methods=["GET"])
@app.route("/api/vue/search")
@auth_desired
@api("read")
def searchposts(v, search_type="posts"):
if v and v.is_banned and not v.unban_utc: return render_template("seized.html")
query = request.args.get("q", '').strip()
page = max(1, int(request.args.get("page", 1)))
sort = request.args.get("sort", "top").lower()
t = request.args.get('t', 'all').lower()
criteria=searchparse(query)
total, ids = searchlisting(criteria, v=v, page=page, t=t, sort=sort)
next_exists = (len(ids) == 26)
ids = ids[0:25]
posts = get_posts(ids, v=v)
if v and v.admin_level>3 and "domain" in criteria:
domain=criteria['domain']
domain_obj=get_domain(domain)
else:
domain=None
domain_obj=None
return {"html":lambda:render_template("search.html",
v=v,
query=query,
total=total,
page=page,
listing=posts,
sort=sort,
t=t,
next_exists=next_exists,
domain=domain,
domain_obj=domain_obj,
reasons=REASONS
),
"api":lambda:jsonify({"data":[x.json for x in posts]})
}
@app.route("/search/comments", methods=["GET"])
@app.route("/api/v1/search/comments", methods=["GET"])
@app.route("/api/vue/search/comments")
@auth_desired
@api("read")
def searchcomments(v):
if v and v.is_banned and not v.unban_utc: return render_template("seized.html")
query = request.args.get("q", '').strip()
try: page = max(1, int(request.args.get("page", 1)))
except: page = 1
sort = request.args.get("sort", "top").lower()
t = request.args.get('t', 'all').lower()
criteria=searchparse(query)
total, ids = searchcommentlisting(criteria, v=v, page=page, t=t, sort=sort)
next_exists = (len(ids) == 26)
ids = ids[0:25]
comments = get_comments(ids, v=v)
return {"html":lambda:render_template("search_comments.html",
v=v,
query=query,
total=total,
page=page,
comments=comments,
sort=sort,
t=t,
next_exists=next_exists,
),
"api":lambda:jsonify({"data":[x.json for x in comments]})
}
@app.route("/search/users", methods=["GET"])
@app.route("/api/v1/search/users", methods=["GET"])
@app.route("/api/vue/search/users")
@auth_desired
@api("read")
def searchusers(v, search_type="posts"):
if v and v.is_banned and not v.unban_utc: return render_template("seized.html")
query = request.args.get("q", '').strip()
page = max(1, int(request.args.get("page", 1)))
sort = request.args.get("sort", "top").lower()
t = request.args.get('t', 'all').lower()
term=query.lstrip('@')
term=term.replace('\\','')
term=term.replace('_','\_')
now=int(time.time())
users=g.db.query(User).filter(User.username.ilike(f'%{term}%'))
users=users.order_by(User.username.ilike(term).desc(), User.stored_subscriber_count.desc())
total=users.count()
users=[x for x in users.offset(25 * (page-1)).limit(26)]
next_exists=(len(users)==26)
users=users[0:25]
return {"html":lambda:render_template("search_users.html",
v=v,
query=query,
total=total,
page=page,
users=users,
sort=sort,
t=t,
next_exists=next_exists
),
"api":lambda:jsonify({"data":[x.json for x in users]})
}