New Gedcom reader

svn: r67
This commit is contained in:
Don Allingham 2001-05-30 12:42:08 +00:00
parent 4d7122c6e1
commit 64fcaaa9e9

View File

@ -0,0 +1,686 @@
#
# Gramps - a GTK+/GNOME based genealogy program
#
# Copyright (C) 2000 Donald N. Allingham
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 2 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program; if not, write to the Free Software
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
#
"Import from GEDCOM (0.1.X)"
from RelLib import *
import latin_ansel
import latin_utf8
import intl
_ = intl.gettext
import os
import re
import string
import const
import utils
import shutil
from gtk import *
from gnome.ui import *
from libglade import *
topDialog = None
db = None
callback = None
glade_file = None
clear_data = 0
InvalidGedcom = "Invalid GEDCOM file"
ged2rel = {}
for val in const.personalConstantEvents.keys():
key = const.personalConstantEvents[val]
if key != "":
ged2rel[key] = val
ged2fam = {}
for val in const.familyConstantEvents.keys():
key = const.familyConstantEvents[val]
if key != "":
ged2fam[key] = val
def find_file(fullname,altpath):
if os.path.isfile(fullname):
return fullname
other = altpath + os.sep + os.path.basename(fullname)
if os.path.isfile(other):
return other
else:
return ""
#-------------------------------------------------------------------------
#
#
#
#-------------------------------------------------------------------------
def importData(database, filename):
global callback
global topDialog
global glade_file
global statusWindow
headRegexp = re.compile(r"\s*0\s+HEAD")
charRegexp = re.compile(r"\s*1\s+CHAR\s+(\S+)\s*$")
sourceRegexp = re.compile(r"\s*1\s+SOUR\s+(\S?[\s\S]*\S)\s*$")
srcrefRegexp = re.compile(r"\s*2\s+SOUR\s+@(.+)@")
indiRegexp = re.compile(r"\s*0\s+@(.+)@\s+INDI")
familyRegexp = re.compile(r"\s*0\s+@(.+)@\s+FAM")
srcRegexp = re.compile(r"\s*0\s+@(.+)@\s+SOUR")
topRegexp = re.compile(r"\s*0\s+")
changeRegexp = re.compile(r"\s*1\s+CHAN")
geventRegexp = re.compile(r"\s*1\s+EVEN")
numRegexp = re.compile(r"\s*[2-9]\s")
eventRegexp = re.compile(r"\s*1\s+([\S]+)\s*(.*)?$")
titleRegexp = re.compile(r"\s*1\s+TITL\s*(.*)?$")
objeRegexp = re.compile(r"\s*1\s+OBJE")
title2Regexp = re.compile(r"\s*2\s+TITL\s*(.*)?$")
fileRegexp = re.compile(r"\s*2\s+FILE\s*(.*)?$")
uidRegexp = re.compile(r"\s*1\s+_UID\s*(.*)?$")
authorRegexp = re.compile(r"\s*1\s+AUTH\s*(.*)?$")
pubRegexp = re.compile(r"\s*1\s+PUBL\s*(.*)?$")
callnoRegexp = re.compile(r"\s*1\s+CALN\s*(.*)?$")
prefixRegexp = re.compile(r"\s*2\s+NPFX\s*(.*)?$")
suffixRegexp = re.compile(r"\s*2\s+NSFX\s*(.*)?$")
birthRegexp = re.compile(r"\s*1\s+BIRT\s*(.*)?$")
noterefRegexp= re.compile(r"\s*1\s+NOTE\s+@(.+)@")
noteactRegexp= re.compile(r"\s*1\s+NOTE\s+(.+)*")
refnRegexp = re.compile(r"\s*1\s+REFN")
noteRegexp = re.compile(r"\s*0\s+@(.+)@\s+NOTE\s*(.*)?$")
concRegexp = re.compile(r"\s*\d\s+CONC\s(.*)?$")
contRegexp = re.compile(r"\s*\d\s+CONT\s(.*)?$")
deathRegexp = re.compile(r"\s*1\s+DEAT\s*(.*)?$")
divorceRegexp= re.compile(r"\s*1\s+DIV\s*(.*)?$")
marriedRegexp= re.compile(r"\s*1\s+MAR\s*(.*)?$")
genderRegexp = re.compile(r"\s*1\s+SEX\s+(\S)?")
typeRegexp = re.compile(r"\s*2\s+TYPE\s*(.*)?$")
placeRegexp = re.compile(r"\s*2\s+PLAC\s*(.*)?$")
pageRegexp = re.compile(r"\s*3\s+PAGE\s*(.*)?$")
dateRegexp = re.compile(r"\s*2\s+DATE\s*(.*)?$")
nameRegexp = re.compile(r"\s*1\s+NAME\s+([\S\s]*\S)?\s*/([^/]+)?/\s*,?\s*([\S]+)?")
famsRegexp = re.compile(r"\s*1\s+FAMS\s+@(.*)@")
famcRegexp = re.compile(r"\s*1\s+FAMC\s+@(.*)@")
fatherRegexp = re.compile(r"\s*1\s+HUSB\s+@(.*)@")
motherRegexp = re.compile(r"\s*1\s+WIFE\s+@(.*)@")
childRegexp = re.compile(r"\s*1\s+CHIL\s+@(.*)@")
noteId = ""
inlineNote = 0
inlocalNote = 0
user2note = {}
person2note = {}
personActive = 0
familyActive = 0
sourceActive = 0
noteActive = 0
allLines = []
values = {}
index = 0
currentLine = 0
fmap = {}
pmap = {}
smap = {}
familyTree = 0
in_change = 0
photo = None
encoding = 0
# add some checking here
gedcom = open(filename,"r")
if clear_data == 1:
database.new()
statusTop = GladeXML(glade_file,"status")
statusWindow = statusTop.get_widget("status")
progressWindow = statusTop.get_widget("progress")
allLines = gedcom.readlines()
gedcom.close()
# use search here instead of match, since PAF5 seems to like to
# insert garbage as the first couple of characters in the file
regex_match = headRegexp.search(allLines[0])
if regex_match == None:
raise InvalidGedcom
total = len(allLines)
value = 0
for index in range(1,51):
value = value + total/50
values[index] = value
index = 1
value = values[1]
for line in allLines:
line = string.replace(line, '\r', "")
if encoding == 1:
line = latin_ansel.ansel_to_latin(line)
elif encoding == 2:
line = latin_utf8.utf8_to_latin(line)
if currentLine == value and index <= 50:
index = index + 1
if index <= 50:
value = values[index]
progressWindow.set_percentage(float(currentLine)/float(total))
while events_pending():
mainiteration()
currentLine = currentLine + 1
regex_match = charRegexp.match(line)
if regex_match:
id = regex_match.groups()
if id[0] == "ANSEL":
encoding = 1
elif id[0] == "UNICODE" or id[0] == "UTF-8" or id[0] == "UTF8":
encoding = 2
continue
regex_match = changeRegexp.match(line)
if regex_match:
in_change = 1
continue
if in_change:
if numRegexp.match(line):
in_change = 0
continue
regex_match = indiRegexp.match(line)
if regex_match:
id = regex_match.groups()
person = database.findPerson(id[0],pmap)
personActive = 1
familyActive = 0
sourceActive = 0
noteActive = 0
continue
regex_match = familyRegexp.match(line)
if regex_match:
id = regex_match.groups()
family = database.findFamily(id[0],fmap)
personActive = 0
familyActive = 1
sourceActive = 0
noteActive = 0
continue
regex_match = srcRegexp.match(line)
if regex_match:
id = regex_match.groups()
source = database.findSource(id[0],smap)
personActive = 0
familyActive = 0
sourceActive = 1
noteActive = 0
continue
regex_match = noteRegexp.match(line)
if regex_match:
matches = regex_match.groups()
noteId = matches[0]
noteActive = 1
sourceActive = 0
familyActive = 0
personActive = 0
if matches[1] == None:
user2note[noteId] = ""
else:
user2note[noteId] = matches[1]
continue
regex_match = sourceRegexp.match(line)
if regex_match:
matches = regex_match.groups()
if matches[0] == "FTW":
familyTree = 1
continue
regex_match = topRegexp.match(line)
if regex_match:
personActive = 0
familyActive = 0
noteActive = 0
continue
if familyActive == 1:
regex_match = srcrefRegexp.match(line)
if regex_match:
matches = regex_match.groups()
source = database.findSource(matches[0],smap)
mySource = Source()
mySource.setBase(source)
event.setSource(mySource)
continue
regex_match = pageRegexp.match(line)
if regex_match:
matches = regex_match.groups()
mySource.setPage(matches[0])
continue
regex_match = fatherRegexp.match(line)
if regex_match:
matches = regex_match.groups()
father = database.findPerson(matches[0],pmap)
family.setFather(father)
continue
regex_match = motherRegexp.match(line)
if regex_match:
matches = regex_match.groups()
mother = database.findPerson(matches[0],pmap)
family.setMother(mother)
continue
regex_match = childRegexp.match(line)
if regex_match :
matches = regex_match.groups()
child = database.findPerson(matches[0],pmap)
family.addChild(child)
continue
regex_match = marriedRegexp.match(line)
if regex_match :
matches = regex_match.groups()
event = Event()
event.setName("Marriage")
family.setMarriage(event)
continue
regex_match = divorceRegexp.match(line)
if regex_match :
matches = regex_match.groups()
event = Event()
event.setName("Divorce")
family.setDivorce(event)
continue
regex_match = placeRegexp.match(line)
if regex_match :
matches = regex_match.groups()
event.setPlace(matches[0])
continue
regex_match = geventRegexp.match(line)
if regex_match :
matches = regex_match.groups()
event = Event()
family.addEvent(event)
continue
regex_match = dateRegexp.match(line)
if regex_match :
matches = regex_match.groups()
try:
event.setDate(matches[0])
except:
pass
continue
regex_match = typeRegexp.match(line)
if regex_match :
matches = regex_match.groups()
event.setName(matches[0])
continue
elif personActive == 1:
regex_match = nameRegexp.match(line)
if regex_match :
matches = regex_match.groups()
name = Name()
if matches[0] :
name.setFirstName(matches[0])
if matches[1] :
name.setSurname(matches[1])
if matches[2] :
name.setSuffix(matches[2])
person.setPrimaryName(name)
continue
regex_match = titleRegexp.match(line)
if regex_match:
continue
regex_match = prefixRegexp.match(line)
if regex_match:
matches = regex_match.groups()
name.setTitle(matches[0])
continue
regex_match = uidRegexp.match(line)
if regex_match:
matches = regex_match.groups()
person.setPafUid(matches[0])
continue
regex_match = suffixRegexp.match(line)
if regex_match:
matches = regex_match.groups()
name.setSuffix(matches[0])
continue
if inlocalNote == 1:
regex_match = concRegexp.match(line)
if regex_match :
matches = regex_match.groups()
user2note[noteId] = user2note[noteId] + matches[0]
continue
regex_match = contRegexp.match(line)
if regex_match :
matches = regex_match.groups()
user2note[noteId] = user2note[noteId] + "\n" + matches[0]
continue
inlocalNote = 0
person.setNote(user2note[noteId])
regex_match = noterefRegexp.match(line)
if regex_match :
matches = regex_match.groups()
person2note[person] = matches[0]
continue
regex_match = noteactRegexp.match(line)
if regex_match :
inlocalNote = 1
matches = regex_match.groups()
noteId = "local%d" % inlineNote
inlineNote = inlineNote + 1
if matches[0] == None:
user2note[noteId] = ""
else:
user2note[noteId] = matches[0]
continue
regex_match = srcrefRegexp.match(line)
if regex_match:
matches = regex_match.groups()
source = database.findSource(matches[0],smap)
mySource = Source()
mySource.setBase(source)
event.setSource(mySource)
continue
regex_match = pageRegexp.match(line)
if regex_match:
matches = regex_match.groups()
mySource.setPage(matches[0])
continue
regex_match = genderRegexp.match(line)
if regex_match :
matches = regex_match.groups()
if matches[0] == "M":
person.setGender(Person.male)
else:
person.setGender(Person.female)
continue
regex_match = famcRegexp.match(line)
if regex_match :
matches = regex_match.groups()
new_family = database.findFamily(matches[0],fmap)
person.setMainFamily(new_family)
continue
regex_match = famsRegexp.match(line)
if regex_match :
matches = regex_match.groups()
new_family = database.findFamily(matches[0],fmap)
person.addFamily(new_family)
continue
regex_match = birthRegexp.match(line)
if regex_match :
matches = regex_match.groups()
event = Event()
# check to see if the person already had a birthdate set.
# if he/she does, then add the birthdate as an alternate.
# this assumes that the first birthdate in the GEDCOM file
# is the one that should be considered the primary
lbirth = person.getBirth()
if lbirth.getDate() != "" or lbirth.getPlace() != "":
person.addEvent(event)
event.setName("Alternate Birth")
else:
person.setBirth(event)
continue
regex_match = deathRegexp.match(line)
if regex_match :
matches = regex_match.groups()
event = Event()
# check to see if the person already had a birthdate set.
# if he/she does, then add the birthdate as an alternate.
# this assumes that the first birthdate in the GEDCOM file
# is the one that should be considered the primary
ldeath = person.getDeath()
if ldeath.getDate() != "" or ldeath.getPlace() != "":
person.addEvent(event)
event.setName("Alternate Death")
else:
person.setDeath(event)
continue
regex_match = refnRegexp.match(line)
if regex_match :
continue
regex_match = geventRegexp.match(line)
if regex_match :
matches = regex_match.groups()
event = Event()
person.addEvent(event)
continue
regex_match = eventRegexp.match(line)
if regex_match :
matches = regex_match.groups()
event = Event()
if ged2rel.has_key(matches[0]):
type = ged2rel[matches[0]]
elif ged2fam.has_key(matches[0]):
type = ged2fam[matches[0]]
else:
type = matches[0]
event.setName(type)
if matches[1] :
event.setDescription(matches[1])
person.addEvent(event)
continue
regex_match = typeRegexp.match(line)
if regex_match :
matches = regex_match.groups()
event.setName(matches[0])
continue
regex_match = placeRegexp.match(line)
if regex_match :
matches = regex_match.groups()
event.setPlace(matches[0])
continue
regex_match = dateRegexp.match(line)
if regex_match :
matches = regex_match.groups()
try:
event.setDate(matches[0])
except:
pass
continue
regex_match = fileRegexp.match(line)
if regex_match :
matches = regex_match.groups()
imagepath = find_file(matches[0],os.path.dirname(filename))
savepath = database.getSavePath()
if imagepath != "" and savepath != "":
id = person.getId()
for index in range(0,100):
base = "i%s_%d.jpg" % (id,index)
name = savepath + os.sep + base
if os.path.exists(name) == 0:
break
shutil.copy(imagepath,name)
photo = Photo()
continue
regex_match = title2Regexp.match(line)
if regex_match and photo != None:
matches = regex_match.groups()
photo.setDescripton(matches[0])
continue
if objeRegexp.match(line):
photo = None
continue
elif noteActive == 1:
regex_match = concRegexp.match(line)
if regex_match :
matches = regex_match.groups()
user2note[noteId] = user2note[noteId] + matches[0]
continue
regex_match = contRegexp.match(line)
if regex_match :
matches = regex_match.groups()
user2note[noteId] = user2note[noteId] + "\n" + matches[0]
continue
elif sourceActive == 1:
regex_match = titleRegexp.match(line)
if regex_match:
matches = regex_match.groups()
source.setTitle(matches[0])
continue
regex_match = authorRegexp.match(line)
if regex_match:
matches = regex_match.groups()
source.setAuthor(matches[0])
continue
regex_match = pubRegexp.match(line)
if regex_match:
matches = regex_match.groups()
source.setPubInfo(matches[0])
continue
regex_match = callnoRegexp.match(line)
if regex_match:
matches = regex_match.groups()
source.setCallNumber(matches[0])
continue
for person in person2note.keys():
name = person.getPrimaryName()
noteId = person2note[person]
person.setNote(user2note[noteId])
statusWindow.destroy()
callback(1)
#-------------------------------------------------------------------------
#
#
#
#-------------------------------------------------------------------------
def on_ok_clicked(obj):
global db
global topDialog
global clear_data
name = topDialog.get_widget("filename").get_text()
if name == "":
return
if topDialog.get_widget("new").get_active():
clear_data = 1
else:
clear_data = 0
utils.destroy_passed_object(obj)
try:
importData(db,name)
except InvalidGedcom:
GnomeErrorDialog(_("%s is not a valid GEDCOM file") % name)
statusWindow.destroy()
except IOError, val:
GnomeErrorDialog(_("Could not load %s") % name + "\n" + val[1])
except:
GnomeErrorDialog(_("Could not load %s") % name)
statusWindow.destroy()
#-------------------------------------------------------------------------
#
#
#
#-------------------------------------------------------------------------
def readData(database,active_person,cb):
global db
global topDialog
global callback
global glade_file
db = database
callback = cb
base = os.path.dirname(__file__)
glade_file = base + os.sep + "gedcomimport.glade"
dic = {
"destroy_passed_object" : utils.destroy_passed_object,
"on_ok_clicked" : on_ok_clicked
}
topDialog = GladeXML(glade_file,"gedcomImport")
topDialog.signal_autoconnect(dic)
topDialog.get_widget("gedcomImport").show()
#-------------------------------------------------------------------------
#
#
#
#-------------------------------------------------------------------------
def get_name():
return _("Import from GEDCOM (0.1.X)")