1
0
mirror of https://github.com/krateng/maloja.git synced 2023-08-10 21:12:55 +03:00
maloja/htmlmodules.py
2019-03-14 11:07:20 +01:00

277 lines
8.3 KiB
Python

from htmlgenerators import *
import database
from utilities import getArtistImage, getTrackImage
from malojatime import *
import urllib
#def getpictures(ls,result,tracks=False):
# from utilities import getArtistsInfo, getTracksInfo
# if tracks:
# for element in getTracksInfo(ls):
# result.append(element.get("image"))
# else:
# for element in getArtistsInfo(ls):
# result.append(element.get("image"))
# artist=None,track=None,since=None,to=None,within=None,associated=False,max_=None,pictures=False
def module_scrobblelist(max_=None,pictures=False,shortTimeDesc=False,earlystop=False,**kwargs):
kwargs_filter = pickKeys(kwargs,"artist","track","associated")
kwargs_time = pickKeys(kwargs,"since","to","within")
# if earlystop, we don't care about the actual amount and only request as many from the db
# without, we request everything and filter on site
maxkey = {"max_":max_} if earlystop else {}
scrobbles = database.get_scrobbles(**kwargs_time,**kwargs_filter,**maxkey)
if pictures:
scrobbleswithpictures = scrobbles if max_ is None else scrobbles[:max_]
#scrobbleimages = [e.get("image") for e in getTracksInfo(scrobbleswithpictures)] #will still work with scrobble objects as they are a technically a subset of track objects
#scrobbleimages = ["/image?title=" + urllib.parse.quote(t["title"]) + "&" + "&".join(["artist=" + urllib.parse.quote(a) for a in t["artists"]]) for t in scrobbleswithpictures]
scrobbleimages = [getTrackImage(t["artists"],t["title"],fast=True) for t in scrobbleswithpictures]
representative = scrobbles[0] if len(scrobbles) is not 0 else None
# build list
i = 0
html = "<table class='list'>"
for s in scrobbles:
html += "<tr>"
html += "<td class='time'>" + time_desc(s["time"],short=shortTimeDesc) + "</td>"
if pictures:
html += """<td class='icon'><div style="background-image:url('""" + scrobbleimages[i] + """')"></div></td>"""
html += "<td class='artists'>" + artistLinks(s["artists"]) + "</td>"
html += "<td class='title'>" + trackLink({"artists":s["artists"],"title":s["title"]}) + "</td>"
# Alternative way: Do it in one cell
#html += "<td class='title'><span>" + artistLinks(s["artists"]) + "</span> — " + trackLink({"artists":s["artists"],"title":s["title"]}) + "</td>"
html += "</tr>"
i += 1
if max_ is not None and i>=max_:
break
html += "</table>"
return (html,len(scrobbles),representative)
def module_pulse(max_=None,**kwargs):
kwargs_filter = pickKeys(kwargs,"artist","track","associated")
kwargs_time = pickKeys(kwargs,"since","to","within","step","stepn","trail")
ranges = database.get_pulse(**kwargs_time,**kwargs_filter)
if max_ is not None: ranges = ranges[:max_]
# if time range not explicitly specified, only show from first appearance
# if "since" not in kwargs:
# while ranges[0]["scrobbles"] == 0:
# del ranges[0]
maxbar = max([t["scrobbles"] for t in ranges])
maxbar = max(maxbar,1)
#build list
html = "<table class='list'>"
for t in ranges:
fromstr = "/".join([str(e) for e in t["from"]])
tostr = "/".join([str(e) for e in t["to"]])
html += "<tr>"
html += "<td>" + range_desc(t["from"],t["to"],short=True) + "</td>"
html += "<td class='amount'>" + scrobblesLink({"since":fromstr,"to":tostr},amount=t["scrobbles"],**kwargs_filter) + "</td>"
html += "<td class='bar'>" + scrobblesLink({"since":fromstr,"to":tostr},percent=t["scrobbles"]*100/maxbar,**kwargs_filter) + "</td>"
html += "</tr>"
html += "</table>"
return html
def module_trackcharts(max_=None,**kwargs):
kwargs_filter = pickKeys(kwargs,"artist","associated")
kwargs_time = pickKeys(kwargs,"since","to","within")
tracks = database.get_charts_tracks(**kwargs_filter,**kwargs_time)
if tracks != []:
maxbar = tracks[0]["scrobbles"]
representative = tracks[0]["track"]
else:
representative = None
i = 0
html = "<table class='list'>"
for e in tracks:
i += 1
if max_ is not None and i>max_:
break
html += "<tr>"
if i == 1 or e["scrobbles"] < prev["scrobbles"]:
html += "<td class='rank'>#" + str(i) + "</td>"
else:
html += "<td class='rank'></td>"
html += "<td class='artists'>" + artistLinks(e["track"]["artists"]) + "</td>"
html += "<td class='title'>" + trackLink(e["track"]) + "</td>"
html += "<td class='amount'>" + scrobblesTrackLink(e["track"],kwargs_time,amount=e["scrobbles"]) + "</td>"
html += "<td class='bar'>" + scrobblesTrackLink(e["track"],kwargs_time,percent=e["scrobbles"]*100/maxbar) + "</td>"
html += "</tr>"
prev = e
html += "</table>"
return (html,representative)
def module_artistcharts(max_=None,**kwargs):
kwargs_filter = pickKeys(kwargs,"associated") #not used right now
kwargs_time = pickKeys(kwargs,"since","to","within")
artists = database.get_charts_artists(**kwargs_filter,**kwargs_time)
if artists != []:
maxbar = artists[0]["scrobbles"]
representative = artists[0]["artist"]
else:
representative = None
i = 0
html = "<table class='list'>"
for e in artists:
i += 1
if max_ is not None and i>max_:
break
html += "<tr>"
if i == 1 or e["scrobbles"] < prev["scrobbles"]:
html += "<td class='rank'>#" + str(i) + "</td>"
else:
html += "<td class='rank'></td>"
html += "<td class='artist'>" + artistLink(e["artist"])
if (e["counting"] != []):
html += " <span class='extra'>incl. " + ", ".join([artistLink(a) for a in e["counting"]]) + "</span>"
html += "</td>"
html += "<td class='amount'>" + scrobblesArtistLink(e["artist"],kwargs_time,amount=e["scrobbles"],associated=True) + "</td>"
html += "<td class='bar'>" + scrobblesArtistLink(e["artist"],kwargs_time,percent=e["scrobbles"]*100/maxbar,associated=True) + "</td>"
html += "</tr>"
prev = e
html += "</table>"
return (html, representative)
def module_artistcharts_tiles(**kwargs):
kwargs_filter = pickKeys(kwargs,"associated") #not used right now
kwargs_time = pickKeys(kwargs,"since","to","within")
artists = database.get_charts_artists(**kwargs_filter,**kwargs_time)[:14]
while len(artists)<14: artists.append(None)
i = 1
bigpart = [0,1,2,6,15]
smallpart = [0,1,2,4,6,9,12,15]
rnk = (0,0) #temporary store so entries with the same scrobble amount get the same rank
html = """<table class="tiles_top"><tr>"""
for e in artists:
if i in bigpart:
n = bigpart.index(i)
html += """<td><table class="tiles_""" + str(n) + """x""" + str(n) + """ tiles_sub">"""
if i in smallpart:
html += "<tr>"
if e is not None:
rank = i if e["scrobbles"] != rnk[1] else rnk[0]
rnk = (rank,e["scrobbles"])
rank = "#" + str(rank)
#image = "/image?artist=" + urllib.parse.quote(e["artist"])
image = getArtistImage(e["artist"],fast=True)
link = artistLink(e["artist"])
else:
rank = ""
image = ""
link = ""
html += """<td style="background-image:url('""" + image + """')"><span class="stats">""" + rank + "</span> <span>" + link + "</span></td>"
i += 1
if i in smallpart:
html += "</tr>"
if i in bigpart:
html += "</table></td>"
html += """</tr></table>"""
return html
def module_trackcharts_tiles(**kwargs):
kwargs_filter = pickKeys(kwargs,"artist","associated")
kwargs_time = pickKeys(kwargs,"since","to","within")
tracks = database.get_charts_tracks(**kwargs_filter,**kwargs_time)[:14]
while len(tracks)<14: tracks.append(None) #{"track":{"title":"","artists":[]}}
i = 1
bigpart = [0,1,2,6,15]
smallpart = [0,1,2,4,6,9,12,15]
rnk = (0,0) #temporary store so entries with the same scrobble amount get the same rank
html = """<table class="tiles_top"><tr>"""
for e in tracks:
if i in bigpart:
n = bigpart.index(i)
html += """<td><table class="tiles_""" + str(n) + """x""" + str(n) + """ tiles_sub">"""
if i in smallpart:
html += "<tr>"
if e is not None:
rank = i if e["scrobbles"] != rnk[1] else rnk[0]
rnk = (rank,e["scrobbles"])
rank = "#" + str(rank)
#image = "/image?title=" + urllib.parse.quote(e["track"]["title"]) + "&" + "&".join(["artist=" + urllib.parse.quote(a) for a in e["track"]["artists"]])
image = getTrackImage(e["track"]["artists"],e["track"]["title"],fast=True)
link = trackLink(e["track"])
else:
rank = ""
image = ""
link = ""
html += """<td style="background-image:url('""" + image + """')"><span class="stats">""" + rank + "</span> <span>" + link + "</span></td>"
i += 1
if i in smallpart:
html += "</tr>"
if i in bigpart:
html += "</table></td>"
html += """</tr></table>"""
return html