diff --git a/plugins/talks.py b/plugins/talks.py
deleted file mode 100644
index 01bfa60..0000000
--- a/plugins/talks.py
+++ /dev/null
@@ -1,521 +0,0 @@
-# -*- coding: utf-8 -*-
-'''
-Manage talks scheduling in a semantic way
-'''
-
-
-from __future__ import print_function
-import os
-import io
-from functools import wraps
-import logging
-import re
-import datetime
-import shutil
-from copy import copy
-import locale
-from contextlib import contextmanager
-from babel.dates import format_date, format_datetime, format_time
-
-from markdown import markdown
-from docutils import nodes
-from docutils.parsers.rst import directives, Directive
-import six
-
-from pelican import signals, generators
-import jinja2
-
-try:
- import ics
-except ImportError:
- ICS_ENABLED = False
-else:
- ICS_ENABLED = True
-import unidecode
-import dateutil
-
-pelican = None # This will be set during register()
-
-
-def memoize(function):
- '''decorators to cache'''
- memo = {}
-
- @wraps(function)
- def wrapper(*args):
- if args in memo:
- return memo[args]
- else:
- rv = function(*args)
- memo[args] = rv
- return rv
- return wrapper
-
-
-@contextmanager
-def setlocale(name):
- saved = locale.setlocale(locale.LC_ALL)
- try:
- yield locale.setlocale(locale.LC_ALL, name)
- finally:
- locale.setlocale(locale.LC_ALL, saved)
-
-
-@memoize
-def get_talk_names():
- names = [name for name in os.listdir(pelican.settings['TALKS_PATH'])
- if not name.startswith('_') and
- get_talk_data(name) is not None
- ]
- names.sort()
- return names
-
-
-def all_talks():
- return [get_talk_data(tn) for tn in get_talk_names()]
-
-
-def unique_attr(iterable, attr):
- return {x[attr] for x in iterable
- if attr in x}
-
-
-@memoize
-def get_global_data():
- fname = os.path.join(pelican.settings['TALKS_PATH'], 'meta.yaml')
- if not os.path.isfile(fname):
- return None
- with io.open(fname, encoding='utf8') as buf:
- try:
- data = yaml.load(buf)
- except Exception:
- logging.exception("Syntax error reading %s; skipping", fname)
- return None
- if data is None:
- return None
- if 'startdate' not in data:
- logging.error("Missing startdate in global data")
- data['startdate'] = datetime.datetime.now()
- if 'rooms' not in data:
- data['rooms'] = {}
- return data
-
-
-def _get_time_shift(timestring):
- ''' Il problema che abbiamo è che vogliamo dire che le 2 di notte del sabato sono in realtà "parte" del
- venerdì. Per farlo accettiamo orari che superano le 24, ad esempio 25.30 vuol dire 1.30.
-
- Questa funzione ritorna una timedelta in base alla stringa passata
- '''
- timeparts = re.findall(r'\d+', timestring)
- if not timeparts or len(timeparts) > 2:
- raise ValueError("Malformed time %s" % timestring)
- timeparts += [0,0] # "padding" per essere sicuro ci siano anche [1] e [2]
- duration = datetime.timedelta(hours=int(timeparts[0]),
- minutes=int(timeparts[1]),
- seconds=int(timeparts[2]))
- if duration.total_seconds() > 3600 * 31 or duration.total_seconds() < 0:
- raise ValueError("Sforamento eccessivo: %d" % duration.hours)
- return duration
-
-
-
-@memoize
-def get_talk_data(talkname):
- fname = os.path.join(pelican.settings['TALKS_PATH'], talkname, 'meta.yaml')
- if not os.path.isfile(fname):
- return None
- with io.open(fname, encoding='utf8') as buf:
- try:
- data = yaml.load(buf)
- except Exception:
- logging.exception("Syntax error reading %s; skipping", fname)
- return None
- if data is None:
- return None
- try:
- gridstep = pelican.settings['TALKS_GRID_STEP']
- data.setdefault('nooverlap', [])
- if 'title' not in data:
- logging.warn("Talk <{}> has no `title` field".format(talkname))
- data['title'] = six.text_type(talkname)
- else:
- data['title'] = six.text_type(data['title'])
- if 'text' not in data:
- logging.warn("Talk <{}> has no `text` field".format(talkname))
- data['text'] = ''
- else:
- data['text'] = six.text_type(data['text'])
- if 'duration' not in data:
- logging.info("Talk <{}> has no `duration` field (50min used)"
- .format(talkname))
- data['duration'] = 50
- data['duration'] = int(data['duration'])
- if data['duration'] < gridstep:
- logging.info("Talk <{}> lasts only {} minutes; changing to {}"
- .format(talkname, data['duration'], gridstep))
- data['duration'] = gridstep
- if 'links' not in data or not data['links']:
- data['links'] = []
- if 'contacts' not in data or not data['contacts']:
- data['contacts'] = []
- if 'needs' not in data or not data['needs']:
- data['needs'] = []
- if 'room' not in data:
- logging.warn("Talk <{}> has no `room` field".format(talkname))
- else:
- if data['room'] in get_global_data()['rooms']:
- data['room'] = get_global_data()['rooms'][data['room']]
- if 'time' not in data or 'day' not in data:
- logging.warn("Talk <{}> has no `time` or `day`".format(talkname))
- if 'time' in data:
- del data['time']
- if 'day' in data:
- del data['day']
- else:
- data['day'] = get_global_data()['startdate'] + \
- datetime.timedelta(days=data['day'])
- try:
- shift = _get_time_shift(str(data['time']))
- except ValueError:
- logging.error("Talk <%s> has malformed `time`", talkname)
-
- data['delta'] = shift
- data['time'] = datetime.datetime.combine(
- data['day'], datetime.time(0,0,0))
- data['time'] += shift
- data['time'] = data['time'].replace(tzinfo=dateutil.tz.gettz('Europe/Rome'))
-
-
- data['id'] = talkname
- resdir = os.path.join(pelican.settings['TALKS_PATH'], talkname,
- pelican.settings['TALKS_ATTACHMENT_PATH'])
- if os.path.isdir(resdir) and os.listdir(resdir):
- data['resources'] = resdir
- return data
- except Exception:
- logging.exception("Error on talk %s", talkname)
- raise
-
-
-def overlap(interval_a, interval_b):
- '''how many minutes do they overlap?'''
- return max(0, min(interval_a[1], interval_b[1]) -
- max(interval_a[0], interval_b[0]))
-
-
-def get_talk_overlaps(name):
- data = get_talk_data(name)
- overlapping_talks = set()
- if 'time' not in data:
- return overlapping_talks
- start = int(data['time'].strftime('%s'))
- end = start + data['duration'] * 60
- for other in get_talk_names():
- if other == name:
- continue
- if 'time' not in get_talk_data(other):
- continue
- other_start = int(get_talk_data(other)['time'].strftime('%s'))
- other_end = other_start + get_talk_data(other)['duration'] * 60
-
- minutes = overlap((start, end), (other_start, other_end))
- if minutes > 0:
- overlapping_talks.add(other)
- return overlapping_talks
-
-
-@memoize
-def check_overlaps():
- for t in get_talk_names():
- over = get_talk_overlaps(t)
- noover = get_talk_data(t)['nooverlap']
- contacts = set(get_talk_data(t)['contacts'])
- for overlapping in over:
- if overlapping in noover or \
- set(get_talk_data(overlapping)['contacts']).\
- intersection(contacts):
- logging.warning('Talk %s overlaps with %s' % (t, overlapping))
-
-
-@memoize
-def jinja_env():
- env = jinja2.Environment(
- loader=jinja2.FileSystemLoader(os.path.join(
- pelican.settings['TALKS_PATH'],
- '_templates')),
- autoescape=True,
- )
- env.filters['markdown'] = lambda text: jinja2.Markup(markdown(text))
- env.filters['dateformat'] = format_date
- env.filters['datetimeformat'] = format_datetime
- env.filters['timeformat'] = format_time
- return env
-
-
-class TalkListDirective(Directive):
- final_argument_whitespace = True
- has_content = True
- option_spec = {
- 'lang': directives.unchanged
- }
-
- def run(self):
- lang = self.options.get('lang', 'C')
- tmpl = jinja_env().get_template('talk.html')
-
- def _sort_date(name):
- '''
- This function is a helper to sort talks by start date
-
- When no date is available, put at the beginning
- '''
- d = get_talk_data(name)
- room = d.get('room', '')
- time = d.get('time', datetime.datetime(1, 1, 1).replace(
- tzinfo=dateutil.tz.gettz('Europe/Rome')))
- title = d.get('title', '')
- return (time, room, title)
-
- return [
- nodes.raw('', tmpl.render(lang=lang, **get_talk_data(n)),
- format='html')
- for n in sorted(get_talk_names(),
- key=_sort_date)
- ]
-
-
-class TalkDirective(Directive):
- required_arguments = 1
- final_argument_whitespace = True
- has_content = True
- option_spec = {
- 'lang': directives.unchanged
- }
-
- def run(self):
- lang = self.options.get('lang', 'C')
- tmpl = jinja_env().get_template('talk.html')
- data = get_talk_data(self.arguments[0])
- if data is None:
- return []
- return [
- nodes.raw('', tmpl.render(lang=lang, **data),
- format='html')
- ]
-
-
-def _delta_to_position(delta):
- gridstep = pelican.settings['TALKS_GRID_STEP']
- sec = delta.total_seconds() // gridstep * gridstep
- return int('%2d%02d' % (sec // 3600, (sec % 3600) // 60))
-
-
-def _delta_inc_position(delta, i):
- gridstep = pelican.settings['TALKS_GRID_STEP']
- delta = delta + datetime.timedelta(minutes=i*gridstep)
- sec = delta.total_seconds() // gridstep * gridstep
- return int('%2d%02d' % (sec // 3600, (sec % 3600) // 60))
-
-def _approx_timestr(timestr):
- gridstep = pelican.settings['TALKS_GRID_STEP']
- t = str(timestr)
- minutes = int(t[-2:])
- hours = t[:-2]
- minutes = minutes // gridstep * gridstep
- return int('%s%02d' % (hours, minutes))
-
-
-class TalkGridDirective(Directive):
- '''A complete grid'''
- final_argument_whitespace = True
- has_content = True
- option_spec = {
- 'lang': directives.unchanged
- }
-
- def run(self):
- try:
- lang = self.options.get('lang', 'C')
- tmpl = jinja_env().get_template('grid.html')
- output = []
- days = unique_attr(all_talks(), 'day')
- gridstep = pelican.settings['TALKS_GRID_STEP']
- for day in sorted(days):
- talks = {talk['id'] for talk in all_talks()
- if talk.get('day', None) == day
- and 'time' in talk
- and 'room' in talk}
- if not talks:
- continue
- talks = [get_talk_data(t) for t in talks]
- rooms = set()
- for t in talks:
- if type(t['room']) is list:
- for r in t['room']:
- rooms.add(r)
- else:
- rooms.add(t['room'])
- # TODO: ordina in base a qualcosa nel meta.yaml globale
- rooms = list(sorted(rooms))
-
- # room=* is not a real room.
- # Remove it unless that day only has special rooms
- if '*' in rooms and len(rooms) > 1:
- del rooms[rooms.index('*')]
- mintimedelta = min({talk['delta'] for talk in talks})
- maxtimedelta = max({talk['delta'] + datetime.timedelta(minutes=talk['duration']) for talk in talks})
- mintime = _delta_to_position(mintimedelta)
- maxtime = _delta_to_position(maxtimedelta)
- times = {}
-
- t = mintimedelta
- while t <= maxtimedelta:
- times[_delta_to_position(t)] = [None] * len(rooms)
- t += datetime.timedelta(minutes=gridstep)
- for talk in sorted(talks, key=lambda x: x['delta']):
- talktime = _delta_to_position(talk['delta'])
- position = _approx_timestr(talktime)
- assert position in times, 'pos=%d,time=%d' % (position, talktime)
- if talk['room'] == '*':
- roomnums = range(len(rooms))
- elif type(talk['room']) is list:
- roomnums = [rooms.index(r) for r in talk['room']]
- else:
- roomnums = [rooms.index(talk['room'])]
- for roomnum in roomnums:
- if times[position][roomnum] is not None:
- logging.error("Talk %s and %s overlap! (room %s)",
- times[position][roomnum]['id'],
- talk['id'],
- rooms[roomnum]
- )
- continue
- times[position][roomnum] = copy(talk)
- times[position][roomnum]['skip'] = False
- for i in range(1, talk['duration'] // gridstep):
- p = _approx_timestr(_delta_inc_position(talk['delta'], i))
- times[p][roomnum] = copy(talk)
- times[p][roomnum]['skip'] = True
-
-
- render = tmpl.render(times=times,
- rooms=rooms,
- mintime=mintime, maxtime=maxtime,
- timestep=gridstep,
- lang=lang,
- )
- output.append(nodes.raw(
- '', u'
%s
' % format_date(day, format='full',
- locale=lang),
- format='html'))
- output.append(nodes.raw('', render, format='html'))
- except:
- logging.exception("Error on talk grid")
- import traceback
- traceback.print_exc()
- return []
- return output
-
-
-def talks_to_ics():
- c = ics.Calendar()
- c.creator = u'pelican'
- for t in all_talks():
- e = talk_to_ics(t)
- if e is not None:
- c.events.add(e)
- return six.text_type(c)
-
-
-def talk_to_ics(talk):
- def _decode(s):
- if six.PY2:
- return unidecode.unidecode(s)
- else:
- return s
-
- if 'time' not in talk or 'duration' not in talk:
- return None
- e = ics.Event(
- uid="%s@%d.hackmeeting.org" % (talk['id'],
- get_global_data()['startdate'].year),
- name=_decode(talk['title']),
- begin=talk['time'],
- duration=datetime.timedelta(minutes=talk['duration']),
- transparent=True,
- )
- # ics.py has some problems with unicode
- # unidecode replaces letters with their most similar ASCII counterparts
- # (ie: accents get stripped)
- if 'text' in talk:
- e.description = _decode(talk['text'])
- e.url = pelican.settings['SCHEDULEURL'] + '#talk-' + talk['id']
- if 'room' in talk:
- e.location = talk['room']
-
- return e
-
-
-class TalksGenerator(generators.Generator):
- def __init__(self, *args, **kwargs):
- self.talks = []
- super(TalksGenerator, self).__init__(*args, **kwargs)
-
- def generate_context(self):
- self.talks = {n: get_talk_data(n) for n in get_talk_names()}
- self._update_context(('talks',))
- check_overlaps()
-
- def generate_output(self, writer=None):
- for talkname in sorted(self.talks):
- if 'resources' in self.talks[talkname]:
- outdir = os.path.join(self.output_path,
- pelican.settings['TALKS_PATH'], talkname,
- pelican.settings['TALKS_ATTACHMENT_PATH']
- )
- if os.path.isdir(outdir):
- shutil.rmtree(outdir)
- shutil.copytree(self.talks[talkname]['resources'], outdir)
- if ICS_ENABLED:
- with io.open(os.path.join(self.output_path,
- pelican.settings.get('TALKS_ICS')),
- 'w', encoding='utf8') as buf:
- buf.write(talks_to_ics())
- else:
- logging.warning('module `ics` not found. '
- 'ICS calendar will not be generated')
-
-
-def add_talks_option_defaults(pelican):
- pelican.settings.setdefault('TALKS_PATH', 'talks')
- pelican.settings.setdefault('TALKS_ATTACHMENT_PATH', 'res')
- pelican.settings.setdefault('TALKS_ICS', 'schedule.ics')
- pelican.settings.setdefault('TALKS_GRID_STEP', 30)
-
-
-def get_generators(gen):
- return TalksGenerator
-
-
-def pelican_init(pelicanobj):
- global pelican
- pelican = pelicanobj
-
-
-try:
- import yaml
-except ImportError:
- print('ERROR: yaml not found. Talks plugins will be disabled')
-
- def register():
- pass
-else:
-
- def register():
- signals.initialized.connect(pelican_init)
- signals.get_generators.connect(get_generators)
- signals.initialized.connect(add_talks_option_defaults)
- directives.register_directive('talklist', TalkListDirective)
- directives.register_directive('talk', TalkDirective)
- directives.register_directive('talkgrid', TalkGridDirective)
diff --git a/plugins/talks/__init__.py b/plugins/talks/__init__.py
new file mode 100644
index 0000000..64e7f13
--- /dev/null
+++ b/plugins/talks/__init__.py
@@ -0,0 +1,3 @@
+from .talks import *
+
+# flake8: noqa
diff --git a/plugins/talks/style.css b/plugins/talks/style.css
new file mode 100644
index 0000000..bca9fb5
--- /dev/null
+++ b/plugins/talks/style.css
@@ -0,0 +1,104 @@
+
+.talk-resources {
+ word-break: break-all;
+}
+
+.talk-grid {
+ table-layout: auto;
+ min-width: 100%;
+ border-collapse: separate;
+ text-align: center;
+}
+
+.talk-grid > thead th:first-child {
+ max-width: 5em;
+}
+
+.talk-grid > thead th {
+ text-align: center;
+}
+
+.talk-grid tr {
+ height: 1.5em;
+}
+
+
+.rooms-4 .talk {
+ width: 25%;
+}
+
+.rooms-3 .talk {
+ width: 33%;
+}
+
+.rooms-2 .talk {
+ width: 50%;
+}
+
+.rooms-1 .talk {
+ width: 100%;
+}
+
+td.talk {
+ border: 1px solid #444;
+ padding: 4px;
+ vertical-align: middle;
+}
+
+td.talk > a {
+ text-decoration: none;
+ border: none;
+}
+
+.talk-grid tr {
+ line-height: 1em;
+}
+
+.talk-title a,
+.talk-title a:hover,
+.talk-title a:focus {
+ border-bottom: none;
+}
+
+.talk-description strong {
+ background: inherit;
+ color: inherit;
+}
+
+/* tag speciali nei talk {{{ */
+/* generati guardando i tag di glypicon come glyphicon-pushpin e copiandone il campo content */
+td.talk::before {
+ font-family: 'Glyphicons Halflings';
+ float: right;
+}
+td.tag-presentazione_libro::before {
+ content: "\e043";
+}
+td.tag-percorso_base::before {
+ content: "\e146";
+}
+/* tag speciali nei talk }}} */
+
+/* END TALK }}} */
+
+/* Pagine speciali */
+.body-info .entry-content > ul {
+ list-style: none;
+}
+
+/*media query min dal piccolo, max dal grande*/
+
+@media all and (max-width: 770px) {
+ .talk-grid {
+ font-size: 0.8em;
+ }
+ .talk-grid td {
+ hyphens: auto;
+ }
+}
+@media all and (max-width: 450px) {
+ .talk-grid {
+ font-size: 0.5em;
+ }
+}
+
diff --git a/plugins/talks/talks.py b/plugins/talks/talks.py
new file mode 100644
index 0000000..7bcb85d
--- /dev/null
+++ b/plugins/talks/talks.py
@@ -0,0 +1,559 @@
+# -*- coding: utf-8 -*-
+"""
+Manage talks scheduling in a semantic way
+"""
+
+
+from __future__ import print_function
+import os
+import io
+from functools import wraps
+import logging
+import re
+import datetime
+import shutil
+from copy import copy
+import locale
+from contextlib import contextmanager
+import inspect
+
+from babel.dates import format_date, format_datetime, format_time
+from markdown import markdown
+from docutils import nodes
+from docutils.parsers.rst import directives, Directive
+import six
+
+from pelican import signals, generators
+import jinja2
+
+try:
+ import ics
+except ImportError:
+ ICS_ENABLED = False
+else:
+ ICS_ENABLED = True
+import unidecode
+import dateutil
+
+pelican = None # This will be set during register()
+
+
+def memoize(function):
+ """decorators to cache"""
+ memo = {}
+
+ @wraps(function)
+ def wrapper(*args):
+ if args in memo:
+ return memo[args]
+ else:
+ rv = function(*args)
+ memo[args] = rv
+ return rv
+
+ return wrapper
+
+
+@contextmanager
+def setlocale(name):
+ saved = locale.setlocale(locale.LC_ALL)
+ try:
+ yield locale.setlocale(locale.LC_ALL, name)
+ finally:
+ locale.setlocale(locale.LC_ALL, saved)
+
+
+@memoize
+def get_talk_names():
+ names = [
+ name
+ for name in os.listdir(pelican.settings["TALKS_PATH"])
+ if not name.startswith("_") and get_talk_data(name) is not None
+ ]
+ names.sort()
+ return names
+
+
+def all_talks():
+ return [get_talk_data(tn) for tn in get_talk_names()]
+
+
+def unique_attr(iterable, attr):
+ return {x[attr] for x in iterable if attr in x}
+
+
+@memoize
+def get_global_data():
+ fname = os.path.join(pelican.settings["TALKS_PATH"], "meta.yaml")
+ if not os.path.isfile(fname):
+ return None
+ with io.open(fname, encoding="utf8") as buf:
+ try:
+ data = yaml.load(buf)
+ except Exception:
+ logging.exception("Syntax error reading %s; skipping", fname)
+ return None
+ if data is None:
+ return None
+ if "startdate" not in data:
+ logging.error("Missing startdate in global data")
+ data["startdate"] = datetime.datetime.now()
+ if "rooms" not in data:
+ data["rooms"] = {}
+ return data
+
+
+def _get_time_shift(timestring):
+ """ Il problema che abbiamo è che vogliamo dire che le 2 di notte del sabato sono in realtà "parte" del
+ venerdì. Per farlo accettiamo orari che superano le 24, ad esempio 25.30 vuol dire 1.30.
+
+ Questa funzione ritorna una timedelta in base alla stringa passata
+ """
+ timeparts = re.findall(r"\d+", timestring)
+ if not timeparts or len(timeparts) > 2:
+ raise ValueError("Malformed time %s" % timestring)
+ timeparts += [0, 0] # "padding" per essere sicuro ci siano anche [1] e [2]
+ duration = datetime.timedelta(
+ hours=int(timeparts[0]), minutes=int(timeparts[1]), seconds=int(timeparts[2])
+ )
+ if duration.total_seconds() > 3600 * 31 or duration.total_seconds() < 0:
+ raise ValueError("Sforamento eccessivo: %d" % duration.hours)
+ return duration
+
+
+@memoize
+def get_talk_data(talkname):
+ fname = os.path.join(pelican.settings["TALKS_PATH"], talkname, "meta.yaml")
+ if not os.path.isfile(fname):
+ return None
+ with io.open(fname, encoding="utf8") as buf:
+ try:
+ data = yaml.load(buf)
+ except Exception:
+ logging.exception("Syntax error reading %s; skipping", fname)
+ return None
+ if data is None:
+ return None
+ try:
+ gridstep = pelican.settings["TALKS_GRID_STEP"]
+ data.setdefault("nooverlap", [])
+ if "title" not in data:
+ logging.warn("Talk <{}> has no `title` field".format(talkname))
+ data["title"] = six.text_type(talkname)
+ else:
+ data["title"] = six.text_type(data["title"])
+ if "text" not in data:
+ logging.warn("Talk <{}> has no `text` field".format(talkname))
+ data["text"] = ""
+ else:
+ data["text"] = six.text_type(data["text"])
+ if "duration" not in data:
+ logging.info(
+ "Talk <{}> has no `duration` field (50min used)".format(talkname)
+ )
+ data["duration"] = 50
+ data["duration"] = int(data["duration"])
+ if data["duration"] < gridstep:
+ logging.info(
+ "Talk <{}> lasts only {} minutes; changing to {}".format(
+ talkname, data["duration"], gridstep
+ )
+ )
+ data["duration"] = gridstep
+ if "links" not in data or not data["links"]:
+ data["links"] = []
+ if "contacts" not in data or not data["contacts"]:
+ data["contacts"] = []
+ if "needs" not in data or not data["needs"]:
+ data["needs"] = []
+ if "room" not in data:
+ logging.warn("Talk <{}> has no `room` field".format(talkname))
+ else:
+ if data["room"] in get_global_data()["rooms"]:
+ data["room"] = get_global_data()["rooms"][data["room"]]
+ if "time" not in data or "day" not in data:
+ logging.warn("Talk <{}> has no `time` or `day`".format(talkname))
+ if "time" in data:
+ del data["time"]
+ if "day" in data:
+ del data["day"]
+ else:
+ data["day"] = get_global_data()["startdate"] + datetime.timedelta(
+ days=data["day"]
+ )
+ try:
+ shift = _get_time_shift(str(data["time"]))
+ except ValueError:
+ logging.error("Talk <%s> has malformed `time`", talkname)
+
+ data["delta"] = shift
+ data["time"] = datetime.datetime.combine(
+ data["day"], datetime.time(0, 0, 0)
+ )
+ data["time"] += shift
+ data["time"] = data["time"].replace(tzinfo=dateutil.tz.gettz("Europe/Rome"))
+
+ data["id"] = talkname
+ resdir = os.path.join(
+ pelican.settings["TALKS_PATH"],
+ talkname,
+ pelican.settings["TALKS_ATTACHMENT_PATH"],
+ )
+ if os.path.isdir(resdir) and os.listdir(resdir):
+ data["resources"] = resdir
+ return data
+ except Exception:
+ logging.exception("Error on talk %s", talkname)
+ raise
+
+
+def overlap(interval_a, interval_b):
+ """how many minutes do they overlap?"""
+ return max(0, min(interval_a[1], interval_b[1]) - max(interval_a[0], interval_b[0]))
+
+
+def get_talk_overlaps(name):
+ data = get_talk_data(name)
+ overlapping_talks = set()
+ if "time" not in data:
+ return overlapping_talks
+ start = int(data["time"].strftime("%s"))
+ end = start + data["duration"] * 60
+ for other in get_talk_names():
+ if other == name:
+ continue
+ if "time" not in get_talk_data(other):
+ continue
+ other_start = int(get_talk_data(other)["time"].strftime("%s"))
+ other_end = other_start + get_talk_data(other)["duration"] * 60
+
+ minutes = overlap((start, end), (other_start, other_end))
+ if minutes > 0:
+ overlapping_talks.add(other)
+ return overlapping_talks
+
+
+@memoize
+def check_overlaps():
+ for t in get_talk_names():
+ over = get_talk_overlaps(t)
+ noover = get_talk_data(t)["nooverlap"]
+ contacts = set(get_talk_data(t)["contacts"])
+ for overlapping in over:
+ if overlapping in noover or set(
+ get_talk_data(overlapping)["contacts"]
+ ).intersection(contacts):
+ logging.warning("Talk %s overlaps with %s" % (t, overlapping))
+
+
+@memoize
+def jinja_env():
+ env = jinja2.Environment(
+ loader=jinja2.FileSystemLoader(
+ os.path.join(pelican.settings["TALKS_PATH"], "_templates")
+ ),
+ autoescape=True,
+ )
+ env.filters["markdown"] = lambda text: jinja2.Markup(markdown(text))
+ env.filters["dateformat"] = format_date
+ env.filters["datetimeformat"] = format_datetime
+ env.filters["timeformat"] = format_time
+ return env
+
+
+@memoize
+def get_css():
+ plugindir = os.path.dirname(
+ os.path.abspath(inspect.getfile(inspect.currentframe()))
+ )
+ with open(os.path.join(plugindir, "style.css")) as buf:
+ return buf.read()
+
+
+class TalkListDirective(Directive):
+ final_argument_whitespace = True
+ has_content = True
+ option_spec = {"lang": directives.unchanged}
+
+ def run(self):
+ lang = self.options.get("lang", "C")
+ tmpl = jinja_env().get_template("talk.html")
+
+ def _sort_date(name):
+ """
+ This function is a helper to sort talks by start date
+
+ When no date is available, put at the beginning
+ """
+ d = get_talk_data(name)
+ room = d.get("room", "")
+ time = d.get(
+ "time",
+ datetime.datetime(1, 1, 1).replace(
+ tzinfo=dateutil.tz.gettz("Europe/Rome")
+ ),
+ )
+ title = d.get("title", "")
+ return (time, room, title)
+
+ return [
+ nodes.raw("", tmpl.render(lang=lang, **get_talk_data(n)), format="html")
+ for n in sorted(get_talk_names(), key=_sort_date)
+ ]
+
+
+class TalkDirective(Directive):
+ required_arguments = 1
+ final_argument_whitespace = True
+ has_content = True
+ option_spec = {"lang": directives.unchanged}
+
+ def run(self):
+ lang = self.options.get("lang", "C")
+ tmpl = jinja_env().get_template("talk.html")
+ data = get_talk_data(self.arguments[0])
+ if data is None:
+ return []
+ return [nodes.raw("", tmpl.render(lang=lang, **data), format="html")]
+
+
+def _delta_to_position(delta):
+ gridstep = pelican.settings["TALKS_GRID_STEP"]
+ sec = delta.total_seconds() // gridstep * gridstep
+ return int("%2d%02d" % (sec // 3600, (sec % 3600) // 60))
+
+
+def _delta_inc_position(delta, i):
+ gridstep = pelican.settings["TALKS_GRID_STEP"]
+ delta = delta + datetime.timedelta(minutes=i * gridstep)
+ sec = delta.total_seconds() // gridstep * gridstep
+ return int("%2d%02d" % (sec // 3600, (sec % 3600) // 60))
+
+
+def _approx_timestr(timestr):
+ gridstep = pelican.settings["TALKS_GRID_STEP"]
+ t = str(timestr)
+ minutes = int(t[-2:])
+ hours = t[:-2]
+ minutes = minutes // gridstep * gridstep
+ return int("%s%02d" % (hours, minutes))
+
+
+class TalkGridDirective(Directive):
+ """A complete grid"""
+
+ final_argument_whitespace = True
+ has_content = True
+ option_spec = {"lang": directives.unchanged}
+
+ def run(self):
+ try:
+ lang = self.options.get("lang", "C")
+ tmpl = jinja_env().get_template("grid.html")
+ output = []
+ days = unique_attr(all_talks(), "day")
+ gridstep = pelican.settings["TALKS_GRID_STEP"]
+ for day in sorted(days):
+ talks = {
+ talk["id"]
+ for talk in all_talks()
+ if talk.get("day", None) == day
+ and "time" in talk
+ and "room" in talk
+ }
+ if not talks:
+ continue
+ talks = [get_talk_data(t) for t in talks]
+ rooms = set()
+ for t in talks:
+ if type(t["room"]) is list:
+ for r in t["room"]:
+ rooms.add(r)
+ else:
+ rooms.add(t["room"])
+ # TODO: ordina in base a qualcosa nel meta.yaml globale
+ rooms = list(sorted(rooms))
+
+ # room=* is not a real room.
+ # Remove it unless that day only has special rooms
+ if "*" in rooms and len(rooms) > 1:
+ del rooms[rooms.index("*")]
+ mintimedelta = min({talk["delta"] for talk in talks})
+ maxtimedelta = max(
+ {
+ talk["delta"] + datetime.timedelta(minutes=talk["duration"])
+ for talk in talks
+ }
+ )
+ mintime = _delta_to_position(mintimedelta)
+ maxtime = _delta_to_position(maxtimedelta)
+ times = {}
+
+ t = mintimedelta
+ while t <= maxtimedelta:
+ times[_delta_to_position(t)] = [None] * len(rooms)
+ t += datetime.timedelta(minutes=gridstep)
+ for talk in sorted(talks, key=lambda x: x["delta"]):
+ talktime = _delta_to_position(talk["delta"])
+ position = _approx_timestr(talktime)
+ assert position in times, "pos=%d,time=%d" % (position, talktime)
+ if talk["room"] == "*":
+ roomnums = range(len(rooms))
+ elif type(talk["room"]) is list:
+ roomnums = [rooms.index(r) for r in talk["room"]]
+ else:
+ roomnums = [rooms.index(talk["room"])]
+ for roomnum in roomnums:
+ if times[position][roomnum] is not None:
+ logging.error(
+ "Talk %s and %s overlap! (room %s)",
+ times[position][roomnum]["id"],
+ talk["id"],
+ rooms[roomnum],
+ )
+ continue
+ times[position][roomnum] = copy(talk)
+ times[position][roomnum]["skip"] = False
+ for i in range(1, talk["duration"] // gridstep):
+ p = _approx_timestr(_delta_inc_position(talk["delta"], i))
+ times[p][roomnum] = copy(talk)
+ times[p][roomnum]["skip"] = True
+
+ render = tmpl.render(
+ times=times,
+ rooms=rooms,
+ mintime=mintime,
+ maxtime=maxtime,
+ timestep=gridstep,
+ lang=lang,
+ )
+ output.append(
+ nodes.raw(
+ "",
+ u"%s
" % format_date(day, format="full", locale=lang),
+ format="html",
+ )
+ )
+ output.append(nodes.raw("", render, format="html"))
+ except:
+ logging.exception("Error on talk grid")
+ import traceback
+
+ traceback.print_exc()
+ return []
+ css = get_css()
+ if css:
+ output.insert(
+ 0,
+ nodes.raw("", '' % css, format="html"),
+ )
+ return output
+
+
+def talks_to_ics():
+ c = ics.Calendar()
+ c.creator = u"pelican"
+ for t in all_talks():
+ e = talk_to_ics(t)
+ if e is not None:
+ c.events.add(e)
+ return six.text_type(c)
+
+
+def talk_to_ics(talk):
+ def _decode(s):
+ if six.PY2:
+ return unidecode.unidecode(s)
+ else:
+ return s
+
+ if "time" not in talk or "duration" not in talk:
+ return None
+ e = ics.Event(
+ uid="%s@%d.hackmeeting.org" % (talk["id"], get_global_data()["startdate"].year),
+ name=_decode(talk["title"]),
+ begin=talk["time"],
+ duration=datetime.timedelta(minutes=talk["duration"]),
+ transparent=True,
+ )
+ # ics.py has some problems with unicode
+ # unidecode replaces letters with their most similar ASCII counterparts
+ # (ie: accents get stripped)
+ if "text" in talk:
+ e.description = _decode(talk["text"])
+ e.url = pelican.settings["SCHEDULEURL"] + "#talk-" + talk["id"]
+ if "room" in talk:
+ e.location = talk["room"]
+
+ return e
+
+
+class TalksGenerator(generators.Generator):
+ def __init__(self, *args, **kwargs):
+ self.talks = []
+ super(TalksGenerator, self).__init__(*args, **kwargs)
+
+ def generate_context(self):
+ self.talks = {n: get_talk_data(n) for n in get_talk_names()}
+ self._update_context(("talks",))
+ check_overlaps()
+
+ def generate_output(self, writer=None):
+ for talkname in sorted(self.talks):
+ if "resources" in self.talks[talkname]:
+ outdir = os.path.join(
+ self.output_path,
+ pelican.settings["TALKS_PATH"],
+ talkname,
+ pelican.settings["TALKS_ATTACHMENT_PATH"],
+ )
+ if os.path.isdir(outdir):
+ shutil.rmtree(outdir)
+ shutil.copytree(self.talks[talkname]["resources"], outdir)
+ if ICS_ENABLED:
+ with io.open(
+ os.path.join(self.output_path, pelican.settings.get("TALKS_ICS")),
+ "w",
+ encoding="utf8",
+ ) as buf:
+ buf.write(talks_to_ics())
+ else:
+ logging.warning(
+ "module `ics` not found. " "ICS calendar will not be generated"
+ )
+
+
+def add_talks_option_defaults(pelican):
+ pelican.settings.setdefault("TALKS_PATH", "talks")
+ pelican.settings.setdefault("TALKS_ATTACHMENT_PATH", "res")
+ pelican.settings.setdefault("TALKS_ICS", "schedule.ics")
+ pelican.settings.setdefault("TALKS_GRID_STEP", 30)
+
+
+def get_generators(gen):
+ return TalksGenerator
+
+
+def pelican_init(pelicanobj):
+ global pelican
+ pelican = pelicanobj
+
+
+try:
+ import yaml
+except ImportError:
+ print("ERROR: yaml not found. Talks plugins will be disabled")
+
+ def register():
+ pass
+
+
+else:
+
+ def register():
+ signals.initialized.connect(pelican_init)
+ signals.get_generators.connect(get_generators)
+ signals.initialized.connect(add_talks_option_defaults)
+ directives.register_directive("talklist", TalkListDirective)
+ directives.register_directive("talk", TalkDirective)
+ directives.register_directive("talkgrid", TalkGridDirective)