gramps/gramps2/src/PeopleModel.py
Martin Hawlisch 971a8a29ec * src/PageView.py (button_press): Catch problem with no loaded database
* src/PersonView.py (button_press): Catch problem with no loaded database
* src/GrampsDbBase.py (_get_from_handle): Catch problem with no loaded database
* src/MapView.py: Updates
* src/EditPerson.py: Change complete flag into marker combo
* src/GrampsDbBase.py: Add map for custom marker values
* src/PeopleModel.py: Add support for colored markers
* src/PersonView.py: Add support for colored markers
* src/RelLib.py: Change complete flag into marker combo
* src/Utils.py: Add marker types
* src/edit_person.glade: Change complete flag into marker combo


svn: r5198
2005-09-16 15:25:27 +00:00

489 lines
17 KiB
Python

#
# Gramps - a GTK+/GNOME based genealogy program
#
# Copyright (C) 2000-2005 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
#
# $Id$
#-------------------------------------------------------------------------
#
# Standard python modules
#
#-------------------------------------------------------------------------
from gettext import gettext as _
import time
import locale
import cgi
import sets
import sys
import traceback
#-------------------------------------------------------------------------
#
# GTK modules
#
#-------------------------------------------------------------------------
import gobject
import gtk
import pango
#-------------------------------------------------------------------------
#
# GRAMPS modules
#
#-------------------------------------------------------------------------
from RelLib import *
import NameDisplay
import DateHandler
import ToolTips
#-------------------------------------------------------------------------
#
# Localized constants
#
#-------------------------------------------------------------------------
_date_format = locale.nl_langinfo(locale.D_T_FMT)
_codeset = locale.nl_langinfo(locale.CODESET)
#-------------------------------------------------------------------------
#
# constants
#
#-------------------------------------------------------------------------
_ID_COL = 1
_GENDER_COL= 2
_NAME_COL = 3
_DEATH_COL = 6
_BIRTH_COL = 7
_EVENT_COL = 8
_FAMILY_COL= 9
_CHANGE_COL= 20
_MARKER_COL=21
#-------------------------------------------------------------------------
#
# python 2.3 has a bug in the unicode sorting using locale.strcoll. Seems
# to have a buffer overrun. We can convince it to do the right thing by
# forcing the string to be nul terminated, sorting, then stripping off the
# nul.
#
#-------------------------------------------------------------------------
if sys.version_info[0:2] == (2,3):
def locale_sort(mylist):
mylist = map(lambda x: x + "\x00", mylist)
mylist.sort(locale.strcoll)
return map(lambda x: x[:-1], mylist)
else:
def locale_sort(mylist):
mylist.sort(locale.strcoll)
return mylist
#-------------------------------------------------------------------------
#
# PeopleModel
#
#-------------------------------------------------------------------------
class PeopleModel(gtk.GenericTreeModel):
def __init__(self,db,data_filter=None,invert_result=False):
gtk.GenericTreeModel.__init__(self)
self.db = db
self.visible = {}
self.top_visible = {}
self.invert_result = invert_result
self.sortnames = {}
self.marker_color_column = 11
self.tooltip_column = 12
self.rebuild_data(data_filter)
def rebuild_data(self,data_filter=None,skip=None):
self.calculate_data(data_filter,skip)
self.assign_data()
def calculate_data(self,data_filter=None,skip=None):
if data_filter:
self.data_filter = data_filter
self.temp_top_path2iter = []
self.temp_iter2path = {}
self.temp_path2iter = {}
self.temp_sname_sub = {}
if not self.db.is_open():
return
if data_filter:
keys = data_filter.apply(self.db)
if self.invert_result:
handle_list = self.db.get_person_handles(sort_handles=False)
#TODO: Could be optimized by using a cursor
keys = [k for k in handle_list if k not in keys]
del handle_list
else:
keys = self.db.get_person_handles(sort_handles=False)
flist = sets.Set(keys)
if skip and skip in flist:
flist.remove(skip)
self.sortnames = {}
cursor = self.db.get_person_cursor()
node = cursor.next()
ngn = NameDisplay.displayer.name_grouping_name
nsn = NameDisplay.displayer.sorted_name
while node:
if node[0] in flist:
primary_name = node[1][_NAME_COL]
surname = unicode(ngn(self.db,primary_name))
self.sortnames[node[0]] = unicode(nsn(primary_name))
if self.temp_sname_sub.has_key(surname):
self.temp_sname_sub[surname].append(node[0])
else:
self.temp_sname_sub[surname] = [node[0]]
node = cursor.next()
cursor.close()
self.temp_top_path2iter = locale_sort(self.temp_sname_sub.keys())
for name in self.temp_top_path2iter:
self.build_sub_entry(name)
def build_sub_entry(self,name):
slist = map(lambda x: (self.sortnames[x],x),self.temp_sname_sub[name])
slist.sort(self.byname)
entries = map(lambda x: x[1], slist)
val = 0
for person_handle in entries:
tpl = (name,val)
self.temp_iter2path[person_handle] = tpl
self.temp_path2iter[tpl] = person_handle
val += 1
def assign_data(self):
self.top_path2iter = self.temp_top_path2iter
self.iter2path = self.temp_iter2path
self.path2iter = self.temp_path2iter
self.sname_sub = self.temp_sname_sub
def byname(self,f,s):
return locale.strcoll(f[0],s[0])
def on_get_flags(self):
'''returns the GtkTreeModelFlags for this particular type of model'''
return gtk.TREE_MODEL_ITERS_PERSIST
def on_get_n_columns(self):
return len(COLUMN_DEFS)
def on_get_path(self, node):
'''returns the tree path (a tuple of indices at the various
levels) for a particular node.'''
try:
return (self.top_path2iter.index(node),)
except:
(surname,index) = self.iter2path[node]
return (self.top_path2iter.index(surname),index)
def is_visable(self,handle):
return self.iter2path.has_key(handle)
def on_get_column_type(self,index):
# return column data-type, from table
return COLUMN_DEFS[index][COLUMN_DEF_TYPE]
def on_get_iter(self, path):
try:
if len(path)==1: # Top Level
return self.top_path2iter[path[0]]
else: # Sublevel
surname = self.top_path2iter[path[0]]
return self.path2iter[(surname,path[1])]
except:
return None
def on_get_value(self,node,col):
# test for header or data row-type
if self.sname_sub.has_key(node):
# Header rows dont get the background color set
if col==self.marker_color_column:
return None
# test for 'header' column being empty (most are)
if not COLUMN_DEFS[col][COLUMN_DEF_HEADER]:
return u''
# return values for 'header' row, calling a function
# according to column_defs table
val = COLUMN_DEFS[col][COLUMN_DEF_HEADER](self,node)
return val
else:
# return values for 'data' row, calling a function
# according to column_defs table
try:
return COLUMN_DEFS[col][COLUMN_DEF_LIST](self,self.db.person_map[str(node)],node)
except:
print "".join(traceback.format_exception(*sys.exc_info()))
return u'error'
def reset_visible(self):
pass
def set_visible(self,node,val):
pass
def on_iter_next(self, node):
'''returns the next node at this level of the tree'''
try:
path = self.top_path2iter.index(node)
if path+1 == len(self.top_path2iter):
return None
return self.top_path2iter[path+1]
except:
(surname,val) = self.iter2path[node]
return self.path2iter.get((surname,val+1))
def on_iter_children(self,node):
"""Return the first child of the node"""
if node == None:
return self.top_path2iter[0]
else:
return self.path2iter.get((node,0))
def on_iter_has_child(self, node):
'''returns true if this node has children'''
if node == None:
return len(self.sname_sub)
if self.sname_sub.has_key(node) and len(self.sname_sub[node]) > 0:
return True
return False
def on_iter_n_children(self,node):
if node == None:
return len(self.sname_sub)
try:
return len(self.sname_sub[node])
except:
return 0
def on_iter_nth_child(self,node,n):
try:
if node == None:
return self.top_path2iter[n]
try:
return self.path2iter[(node,n)]
except:
return None
except IndexError:
return None
def on_iter_parent(self, node):
'''returns the parent of this node'''
path = self.iter2path.get(node)
if path:
return path[0]
return None
def column_sort_name(self,data,node):
return data[_NAME_COL].get_sort_name()
def column_spouse(self,data,node):
spouses_names = u""
handle = data[0]
for family_handle in data[_FAMILY_COL]:
family = self.db.get_family_from_handle(family_handle)
for spouse_id in [family.get_father_handle(), family.get_mother_handle()]:
if not spouse_id:
continue
if spouse_id == handle:
continue
spouse = self.db.get_person_from_handle(spouse_id)
if len(spouses_names) > 0:
spouses_names += ", "
spouses_names += NameDisplay.displayer.display(spouse)
return spouses_names
def column_name(self,data,node):
return NameDisplay.displayer.sorted_name(data[_NAME_COL])
def column_id(self,data,node):
return data[_ID_COL]
def column_change(self,data,node):
return unicode(time.strftime(_date_format,
time.localtime(data[_CHANGE_COL])),
_codeset)
def column_gender(self,data,node):
return _GENDER[data[_GENDER_COL]]
def column_birth_day(self,data,node):
if data[_BIRTH_COL]:
birth = self.db.get_event_from_handle(data[_BIRTH_COL].ref)
date_str = DateHandler.get_date(birth)
if date_str != "":
return cgi.escape(date_str)
for event_ref in data[_EVENT_COL]:
event = self.db.get_event_from_handle(event_ref.ref)
etype = event.get_type()[0]
date_str = DateHandler.get_date(event)
if (etype in [Event.BAPTISM, Event.CHRISTEN]
and date_str != ""):
return "<i>" + cgi.escape(date_str) + "</i>"
return u""
def column_death_day(self,data,node):
if data[_DEATH_COL]:
death = self.db.get_event_from_handle(data[_DEATH_COL].ref)
date_str = DateHandler.get_date(death)
if date_str != "":
return cgi.escape(date_str)
for event_ref in data[_EVENT_COL]:
event = self.db.get_event_from_handle(event_ref.ref)
etype = event.get_type()[0]
date_str = DateHandler.get_date(event)
if (etype in [Event.BURIAL, Event.CREMATION]
and date_str != ""):
return "<i>" + cgi.escape(date_str) + "</i>"
return u""
def column_cause_of_death(self,data,node):
if data[_DEATH_COL]:
return self.db.get_event_from_handle(data[_DEATH_COL].ref).get_cause()
else:
return u""
def column_birth_place(self,data,node):
if data[_BIRTH_COL]:
event = self.db.get_event_from_handle(data[_BIRTH_COL].ref)
if event:
place_handle = event.get_place_handle()
if place_handle:
place_title = self.db.get_place_from_handle(place_handle).get_title()
if place_title != "":
return cgi.escape(place_title)
for event_ref in data[_EVENT_COL]:
etype = event.get_type()[0]
if etype in [Event.BAPTISM, Event.CHRISTEN]:
event = self.db.get_event_from_handle(event_ref.ref)
place_handle = event.get_place_handle()
if place_handle:
place_title = self.db.get_place_from_handle(place_handle).get_title()
if place_title != "":
return "<i>" + cgi.escape(place_title) + "</i>"
return u""
def column_death_place(self,data,node):
if data[_DEATH_COL]:
event = self.db.get_event_from_handle(data[_DEATH_COL].ref)
if event:
place_handle = event.get_place_handle()
if place_handle:
place_title = self.db.get_place_from_handle(place_handle).get_title()
if place_title != "":
return cgi.escape(place_title)
for event_ref in data[_EVENT_COL]:
etype = event.get_type()[0]
if etype in [Event.BURIAL, Event.CREMATION]:
event = self.db.get_event_from_handle(event_ref.ref)
place_handle = event.get_place_handle()
if place_handle:
place_title = self.db.get_place_from_handle(place_handle).get_title()
if place_title != "":
return "<i>" + cgi.escape(place_title) + "</i>"
return u""
def column_marker_text(self,data,node):
try:
if data[_MARKER_COL]:
if data[_MARKER_COL][0] == PrimaryObject.MARKER_CUSTOM:
return data[_MARKER_COL][1]
elif data[_MARKER_COL][0] in Utils.marker_types:
return Utils.marker_types[data[_MARKER_COL][0]]
except IndexError:
return ""
return ""
def column_marker_color(self,data,node):
try:
if data[_MARKER_COL]:
if data[_MARKER_COL][0] == PrimaryObject.MARKER_COMPLETE:
return u"#46a046" # green
if data[_MARKER_COL][0] == PrimaryObject.MARKER_TODO:
return u"#df421e" # red
if data[_MARKER_COL][0] == PrimaryObject.MARKER_CUSTOM:
return u"#eed680" # blue
except IndexError:
pass
return None
def column_tooltip(self,data,node):
return ToolTips.TipFromFunction(self.db, lambda: self.db.get_person_from_handle(data[0]))
def column_int_id(self,data,node):
return node
def column_header(self,node):
return node
def column_header_view(self,node):
return True
_GENDER = [ _(u'female'), _(u'male'), _(u'unknown') ]
# table of column definitions
# (unless this is declared after the PeopleModel class, an error is thrown)
COLUMN_DEFS = [
# data column (method) header column (method) column data type
(PeopleModel.column_name, PeopleModel.column_header, str),
(PeopleModel.column_id, None, str),
(PeopleModel.column_gender, None, str),
(PeopleModel.column_birth_day, None, str),
(PeopleModel.column_birth_place,None, str),
(PeopleModel.column_death_day, None, str),
(PeopleModel.column_death_place,None, str),
(PeopleModel.column_spouse, None, str),
(PeopleModel.column_change, None, str),
(PeopleModel.column_cause_of_death, None, str),
(PeopleModel.column_marker_text, None, str),
(PeopleModel.column_marker_color, None, str),
# the order of the above columns must match PeopleView.column_names
# these columns are hidden, and must always be last in the list
(PeopleModel.column_tooltip, None, object),
(PeopleModel.column_sort_name, None, str),
(PeopleModel.column_int_id, None, str),
]
# dynamic calculation of column indices, for use by various Views
COLUMN_INT_ID = len(COLUMN_DEFS) - 1
# indices into main column definition table
COLUMN_DEF_LIST = 0
COLUMN_DEF_HEADER = 1
COLUMN_DEF_TYPE = 2