1
0
mirror of https://github.com/krateng/maloja.git synced 2023-08-10 21:12:55 +03:00
maloja/utilities.py
2019-02-03 16:52:37 +01:00

306 lines
7.5 KiB
Python

import re
import os
import hashlib
from threading import Thread
### TSV files
def parseTSV(filename,*args):
f = open(filename)
result = []
for l in [l for l in f if (not l.startswith("#")) and (not l.strip()=="")]:
l = l.replace("\n","").split("#")[0]
data = list(filter(None,l.split("\t"))) # Multiple tabs are okay, we don't accept empty fields unless trailing
entry = [] * len(args)
for i in range(len(args)):
if args[i]=="list":
try:
entry.append(data[i].split(""))
except:
entry.append([])
elif args[i]=="string":
try:
entry.append(data[i])
except:
entry.append("")
elif args[i]=="int":
try:
entry.append(int(data[i]))
except:
entry.append(0)
elif args[i]=="bool":
try:
entry.append((data[i].lower() in ["true","yes","1","y"]))
except:
entry.append(False)
result.append(entry)
f.close()
return result
def checksumTSV(folder):
sums = ""
for f in os.listdir(folder + "/"):
if (f.endswith(".tsv")):
f = open(folder + "/" + f,"rb")
sums += hashlib.md5(f.read()).hexdigest() + "\n"
f.close()
return sums
# returns whether checksums match and sets the checksum to invalid if they don't (or sets the new one if no previous one exists)
def combineChecksums(filename,checksums):
import os
if os.path.exists(filename + ".rulestate"):
f = open(filename + ".rulestate","r")
oldchecksums = f.read()
f.close()
if oldchecksums == checksums:
# the new checksum given by the calling db server represents the rule state that all current unsaved scrobbles were created under
# if this is the same as the existing one, we're all good
return True
elif (oldchecksums != "INVALID"):
#if not, the file is not consistent to any single rule state (some scrobbles were created with an old ruleset, some not)
f = open(filename + ".rulestate","w")
f.write("INVALID") # this will never match any sha256sum
f.close()
return False
else:
#if the file already says invalid, no need to open it and rewrite
return False
else:
f = open(filename + ".rulestate","w")
f.write(checksums)
f.close()
return True
# checks ALL files for their rule state. if they are all the same as the current loaded one, the entire database can be assumed to be consistent with the current ruleset
# in any other case, get out
def consistentRulestate(folder,checksums):
result = []
for scrobblefile in os.listdir(folder + "/"):
if (scrobblefile.endswith(".tsv")):
try:
f = open(folder + "/" + scrobblefile + ".rulestate","r")
if f.read() != checksums:
return False
except:
return False
finally:
f.close()
return True
def parseAllTSV(path,*args):
result = []
for f in os.listdir(path + "/"):
if (f.endswith(".tsv")):
result += parseTSV(path + "/" + f,*args)
return result
def createTSV(filename):
if not os.path.exists(filename):
open(filename,"w").close()
def addEntry(filename,a):
createTSV(filename)
line = "\t".join(a)
with open(filename,"a") as f:
f.write(line + "\n")
def addEntries(filename,al):
with open(filename,"a") as f:
for a in al:
line = "\t".join(a)
f.write(line + "\n")
### Logging
def log(msg):
import inspect
module = inspect.getmodule(inspect.stack()[1][0]).__name__
if module == "__main__": module = "mainserver"
print("[" + module + "] " + msg)
# best function ever
### Media info
def apirequest(artists=None,artist=None,title=None):
import urllib.parse, urllib.request
import json
with open("apikey","r") as keyfile:
apikey = keyfile.read().replace("\n","")
sites = [
{
"name":"lastfm",
"artisturl":"https://ws.audioscrobbler.com/2.0/?method=artist.getinfo&artist={artist}&api_key=" + apikey + "&format=json",
"trackurl":"https://ws.audioscrobbler.com/2.0/?method=track.getinfo&track={title}&artist={artist}&api_key=" + apikey + "&format=json",
"result_artist_imgurl":lambda data:data["artist"]["image"][2]["#text"],
"result_track_imgurl":lambda data:data["track"]["album"]["image"][2]["#text"]
#"result_artist_desc":lambda data:data["artist"]["bio"]["summary"],
#"result_track_desc":lambda data:None
}
]
# TRACKS
if title is not None:
for s in sites:
try:
artiststr = urllib.parse.quote(", ".join(artists))
titlestr = urllib.parse.quote(title)
response = urllib.request.urlopen(s["trackurl"].format(artist=artiststr,title=titlestr))
data = json.loads(response.read())
if s["result_track_imgurl"](data) != "":
return {"image":s["result_track_imgurl"](data)}
except:
pass
if len(artists) == 1:
#return {"image":apirequest(artist=artists[0])["image"]}
return {"image":None}
# try the same track with every single artist
for a in artists:
rec = apirequest(artists=[a],title=title)
if rec["image"] is not None:
return rec
return {"image":None}
# ARTISTS
else:
for s in sites:
try:
response = urllib.request.urlopen(s["artisturl"].format(artist=urllib.parse.quote(artist)))
data = json.loads(response.read())
if s["result_artist_imgurl"](data) != "":
return {"image":s["result_artist_imgurl"](data)}
except:
pass
return {"image":""}
# I think I've only just understood modules
cachedTracks = {}
cachedArtists = {}
def getTrackInfo(artists,title):
obj = (frozenset(artists),title)
filename = "-".join([re.sub("[^a-zA-Z0-9]","",artist) for artist in artists]) + "_" + re.sub("[^a-zA-Z0-9]","",title)
if filename == "": filename = str(hash(obj))
filepath = "images/tracks/" + filename
# check if custom image exists
if os.path.exists(filepath + ".png"):
imgurl = "/" + filepath + ".png"
return {"image":imgurl}
elif os.path.exists(filepath + ".jpg"):
imgurl = "/" + filepath + ".jpg"
return {"image":imgurl}
elif os.path.exists(filepath + ".jpeg"):
imgurl = "/" + filepath + ".jpeg"
return {"image":imgurl}
try:
return {"image":cachedTracks[(frozenset(artists),title)]}
except:
pass
result = apirequest(artists=artists,title=title)
if result.get("image") is not None:
cachedTracks[(frozenset(artists),title)] = result["image"]
return result
else:
result = getArtistInfo(artist=artists[0])
cachedTracks[(frozenset(artists),title)] = result["image"]
return result
def getArtistInfo(artist):
obj = artist
filename = re.sub("[^a-zA-Z0-9]","",artist)
if filename == "": filename = str(hash(obj))
filepath = "images/artists/" + filename
#filepath_cache = "info/artists_cache/" + filename
# check if custom image exists
if os.path.exists(filepath + ".png"):
imgurl = "/" + filepath + ".png"
return {"image":imgurl}
elif os.path.exists(filepath + ".jpg"):
imgurl = "/" + filepath + ".jpg"
return {"image":imgurl}
elif os.path.exists(filepath + ".jpeg"):
imgurl = "/" + filepath + ".jpeg"
return {"image":imgurl}
try:
return {"image":cachedArtists[artist]}
except:
pass
result = apirequest(artist=artist)
cachedArtists[artist] = result["image"]
return result
def getTracksInfo(trackobjectlist):
threads = []
for track in trackobjectlist:
t = Thread(target=getTrackInfo,args=(track["artists"],track["title"],))
t.start()
threads.append(t)
for t in threads:
t.join()
return [getTrackInfo(t["artists"],t["title"]) for t in trackobjectlist]
def getArtistsInfo(artistlist):
threads = []
for artist in artistlist:
t = Thread(target=getArtistInfo,args=(artist,))
t.start()
threads.append(t)
for t in threads:
t.join()
# async calls only cached results, now we need to get them
return [getArtistInfo(a) for a in artistlist]