#.-*- coding: utf-8 -*-
import troggle.settings as settings
import troggle.expo.models as models
import csv
import re
import datetime
import os
# Julian: the below code was causing errors and it seems like a duplication of the above. Hope I haven't broken anything by commenting it. -Aaron
#
# if name in expoers2008:
# print "2008:", name
# expomissing.discard(name) # I got an error which I think was caused by this -- python complained that a set changed size during iteration.
# yo = models.Expedition.objects.filter(year = "2008")[0]
# pyo = models.PersonExpedition(person = pObject, expedition = yo, is_guest=is_guest)
# pyo.save()
def parseMugShotAndBlurb(firstname, lastname, person, header, pObject):
#create mugshot Photo instance
mugShotPath = os.path.join(settings.EXPOWEB, "folk", person[header["Mugshot"]])
if mugShotPath[-3:]=='jpg': #if person just has an image, add it
mugShotObj = models.Photo(
caption="Mugshot for "+firstname+" "+lastname,
is_mugshot=True,
file=mugShotPath,
)
mugShotObj.save()
mugShotObj.contains_person.add(pObject)
mugShotObj.save()
elif mugShotPath[-3:]=='htm': #if person has an html page, find the image(s) and add it. Also, add the text from the html page to the "blurb" field in his model instance.
personPageOld=open(mugShotPath,'r').read()
pObject.blurb=re.search('
.*
", "", name)
mname = re.match("(\w+)(?:\s((?:van |ten )?\w+))?(?:\s\(([^)]*)\))?", name)
if mname.group(3):
nickname = mname.group(3)
else:
nickname = ""
firstname, lastname = mname.group(1), mname.group(2) or ""
print firstname, lastname, "NNN", nickname
#assert lastname == person[header[""]], person
href = firstname.lower()
if lastname:
href += "_" + lastname.lower()
pObject = models.Person(first_name = firstname,
last_name = lastname, href=href,
is_vfho = person[header["VfHO member"]],
)
is_guest = person[header["Guest"]] == "1" # this is really a per-expo catagory; not a permanent state
pObject.save()
parseMugShotAndBlurb(firstname, lastname, person, header, pObject)
for year, attended in zip(headers, person)[5:]:
yo = models.Expedition.objects.filter(year = year)[0]
if attended == "1" or attended == "-1":
pyo = models.PersonExpedition(person = pObject, expedition = yo, nickname=nickname, is_guest=is_guest)
pyo.save()
# error
elif (firstname, lastname) == ("Mike", "Richardson") and year == "2001":
print "Mike Richardson(2001) error"
pyo = models.PersonExpedition(person = pObject, expedition = yo, nickname=nickname, is_guest=is_guest)
pyo.save()
# this fills in those people for whom 2008 was their first expo
for name in expomissing:
firstname, lastname = name.split()
is_guest = name in ["Eeva Makiranta", "Keith Curtis"]
print "2008:", name
pObject = models.Person(first_name = firstname,
last_name = lastname,
is_vfho = False,
mug_shot = "")
pObject.save()
yo = models.Expedition.objects.filter(year = "2008")[0]
pyo = models.PersonExpedition(person = pObject, expedition = yo, nickname="", is_guest=is_guest)
pyo.save()
# expedition name lookup cached for speed (it's a very big list)
Gpersonexpeditionnamelookup = { }
def GetPersonExpeditionNameLookup(expedition):
global Gpersonexpeditionnamelookup
res = Gpersonexpeditionnamelookup.get(expedition.name)
if res:
return res
res = {}
duplicates = set()
personexpeditions = models.PersonExpedition.objects.filter(expedition=expedition)
for personexpedition in personexpeditions:
possnames = [ ]
f = personexpedition.person.first_name.lower()
l = personexpedition.person.last_name.lower()
if l:
possnames.append(f + " " + l)
possnames.append(f + " " + l[0])
possnames.append(f + l[0])
possnames.append(f[0] + " " + l)
possnames.append(f)
if personexpedition.nickname:
possnames.append(personexpedition.nickname.lower())
for possname in possnames:
if possname in res:
duplicates.add(possname)
else:
res[possname] = personexpedition
for possname in duplicates:
del res[possname]
Gpersonexpeditionnamelookup[expedition.name] = res
return res